Sustaining Digital Therapeutic Relationships: The Evolution of Service Quality Cue Utilization from First to Repeat Consultations

Sustaining Digital Therapeutic Relationships: The Evolution of Service Quality Cue Utilization from First to Repeat Consultations | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Sustaining Digital Therapeutic Relationships: The Evolution of Service Quality Cue Utilization from First to Repeat Consultations Yumei Li This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9463453/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose Existing literature largely presupposes that as patients accumulate clinical encounters, their reliance on extrinsic quality cues—platform reputation chief among them—follows a trajectory of signal decay. Yet this paradigm prematurely generalizes from consumer goods contexts and overlooks boundary conditions in high-stakes credence services where diagnostic uncertainty persists despite repeated interaction. This study examines how physician quality cues differentially shape patient decisions at first versus repeat consultations, and identifies acquisition channel (online versus offline) as a critical moderator of signal evolution. Methodology Drawing on a time-lagged research design (T0→T1→T2), we analyze 1,812 asthma specialists from a leading Chinese online healthcare platform. Physician attributes measured at baseline (T0) predict first-time consultations (T1) and subsequent repeat consultations (T2). Zero-inflated negative binomial regression models distinguish between consideration-set formation (structural zeros) and volume conditional on selection. Findings Our findings contradict the prevailing signal decay thesis. Physician reputation maintains or intensifies its predictive power across consultation stages—a pattern consistent with signal verification rather than attenuation, and particularly pronounced among online-acquired patients. We identify a dual-mechanism architecture in which reputation operates as a threshold gatekeeper governing consideration-set formation, while process attributes such as response speed drive volume conditional on selection. Acquisition channel significantly moderates these dynamics. Offline-acquired patients exhibit stable reputation reliance and geographic continuity-seeking, preferring physicians in lower-tier cities. Conversely, online-acquired patients display spatial substitution and escalating reputation sensitivity post-verification, while the functional role of price transmutes from a quality-inference heuristic to a pure cost barrier as relationships mature. Implications Theoretically, this study establishes signal verification as a viable alternative mechanism to signal decay in platform-mediated professional services, demonstrating that institutional cues and accumulated direct experience operate as complementary governance mechanisms rather than substitutes. These findings suggest that platform interfaces should be adapted by patient origin: online-acquired patients benefit from prominent process-transparency and efficiency cues during trust formation, while offline-referred patients require geographic proximity signals and reputation confirmation to maintain therapeutic continuity. Information Retrieval and Management online healthcare signal verification trust transfer quality cues digital therapeutic relationships Introduction Chronic diseases—cardiovascular conditions, respiratory disorders, diabetes, and related non-communicable illnesses—represent a critical global public health challenge. In China, rising prevalence rates, accelerated aging demographics, and lifestyle shifts have exacerbated their socioeconomic burden [1] . Asthma, a chronic respiratory disease requiring long-term management, epitomizes this crisis: its complex etiology demands continuous medical supervision, yet healthcare resources remain scarce and geographically unevenly distributed [2] . Among chronic conditions, asthma presents an ideal empirical context for examining sustained digital therapeutic relationships: its relapsing-remitting nature necessitates iterative physician input; its symptom-based diagnostic criteria facilitate high-quality asynchronous online consultations; and its treatment protocols require long-term medication adherence monitoring, making continuity of care both clinically critical and technologically feasible through digital platforms. Traditional healthcare systems struggle to deliver accessible, continuous care, particularly for chronic conditions where sustained doctor-patient relationships correlate strongly with clinical outcomes and operational efficiency [3] . Digital healthcare platforms have emerged as transformative solutions, transcending geographical constraints to connect patients with physicians nationwide [4] . Platforms like Haodf.com enable asynchronous consultations, expanding patient choice while generating rich behavioral data. However, the very abundance of physician information (reputation scores, service pricing, response metrics) creates decision complexity for patients. Crucially, while initial consultation choices have been studied [5–7] , how patients sustain relationships through repeat consultations, and how their quality signal preferences evolve, remains poorly understood despite its centrality to chronic disease management. Existing research exhibits three critical limitations. First, studies predominantly examine static, cross-sectional consultation choice [3] , neglecting how patients’ information processing evolves across the relationship lifecycle. While psychological accounting theory suggests repeat consumers shift from risk-minimization to value-maximization [8] , emerging evidence suggests quality signals may undergo verification and reinforcement rather than simple decay. Patients who confirm platform reputation through direct experience may actually increase reliance on such signals, creating a "trust consolidation" effect that remains untested in digital healthcare contexts. Second, patient heterogeneity is oversimplified. Patients originate via distinct channels: online-acquired patients (initiating contact through direct platform search without prior interaction) rely on platform-mediated signals, while offline-acquired patients (transitioning from face-to-face encounters via the platform’s post-visit reporting service, importing pre-existing therapeutic trust) possess direct quality assessments [9] . However, conventional wisdom assumes offline patients rely less on platform signals, a presumption contradicted by preliminary evidence suggesting offline-acquired patients may transfer strong offline trust into sustained online engagement with high-reputation physicians. Third, methodological constraints plague existing studies. Most rely on single-period observations, rendering them unable to disentangle whether quality signals predict subsequent behavior or merely reflect concurrent demand. We address this by employing a time-lagged research design: physician service attributes measured in May 2017 predict first-time consultations in June 2017, which in turn predict repeat consultations from July to December 2017. This temporal sequencing mitigates reverse causality concerns while capturing relationship evolution. From a signal-theoretic perspective, physician reputation may function not merely as a volume determinant but as a threshold gatekeeper—creating structural barriers that exclude low-reputation physicians from patient consideration sets entirely. This suggests a dual-stage decision architecture wherein reputation governs the extensive margin (whether a physician is considered at all), while process quality shapes the intensive margin (consultation volume conditional on consideration). To address these theoretical and methodological limitations, this study employs a temporal sequencing design to investigate three specific research questions regarding (1) cue evolution across relationship stages, (2) channel-based heterogeneity in signal reliance, and (3) the dual mechanisms governing consideration-set formation versus volume generation. Specifically, we examine how physician service quality cues—reputation, price, information richness, and response speed—impact initial and repeat consultation volumes across patient acquisition channels. Our empirical strategy employs zero-inflated negative binomial regression (ZINB) to accommodate the over-dispersed and zero-heavy distribution of consultation counts across 1,812 asthma specialists, distinguishing between the structural probability of zero consultations and volume conditional on selection. We control for institutional and geographic heterogeneity (physician title, hospital tier, city hierarchy) that may confound signal effects, as physicians in tertiary hospitals or developed cities may possess both higher baseline reputation and differential patient access opportunities. This study contributes to platform-mediated service literature by: (1) challenging the signal decay paradigm through temporal analysis of cue evolution across first-time and repeat consultations; (2) delineating how patient acquisition channels moderate signal reliance in sustained digital therapeutic relationships; and (3) distinguishing between consideration-set formation (structural zeros governed by reputation thresholds) and volume drivers (conditional consultation behavior shaped by process quality) in digital healthcare choice. We guide platform designers and physicians in segment-specific engagement strategies—optimizing reputation and pricing for patient acquisition, while emphasizing response efficiency for chronic disease management retention. Research Hypotheses Theoretical Foundation: Signal Evolution in Platform-Mediated Healthcare The prevailing signal decay paradigm in consumer behavior literature posits an inverse relationship between experience accumulation and extrinsic cue reliance [10] . This theoretical tradition suggests that consumers progressively substitute direct experience for platform-mediated signals as cognitive investments yield quality information. However, this framework encounters boundaries in high-stakes, credence-dominant service contexts characterized by persistent outcome uncertainty. Signal Verification and Trust Consolidation. Contrary to the decay thesis, cognitive consistency theory suggests that verified expectations strengthen belief structures [11] . When initial service encounters confirm platform reputation signals, patients experience cognitive consonance that validates the platform's signaling infrastructure [12] . This generates trust consolidation—a self-reinforcing cycle wherein validated signals increase in perceived reliability over time. In healthcare, where diagnostic uncertainty persists despite accumulated interactions, reputation signals serve as risk anchors that maintain or increase salience as relationships deepen. Unlike consumer goods where direct experience can fully substitute for extrinsic cues, healthcare platforms create conditions under which verified reputation signals consolidate rather than decay [13] . Functional Transmutation of Economic Signals. Information economics distinguishes between the signaling function and allocative function of price [14] . Under high information asymmetry, price serves as a diagnostic attribute enabling quality inference [15] ; as consumers acquire direct experience, the information content of price marginalizes, and its role shifts toward pure cost assessment [16] . This transmutation implies that price loses predictive power for consumer choice as experience accumulates, regardless of whether price initially functioned as a quality signal or a budget constraint. Channel-Contingent Information Processing. Trust transfer theory posits that offline-acquired patients import pre-existing relational trust from face-to-face encounters, creating stable knowledge anchors that reduce sensitivity to experiential variation [17] . Conversely, online-acquired patients face pure platform uncertainty, necessitating active learning. Their cue reliance follows constructive trajectories: initial diversification across signals (risk-spreading) followed by concentration on verified signals (risk-reduction) as experience accumulates. Process Quality and Value Co-Creation. Service-dominant logic conceptualizes patients as value co-creators [18] . As relationships mature from diagnostic establishment to chronic management, the locus of value shifts from information acquisition (content) to relational continuity (process efficiency). This transition implies escalating salience of response speed as a coordination mechanism for sustained care. Geographic Proximity and Digital Substitution. Information economics suggests that digital platforms function as geographic proximity substitutes by aggregating quality signals that supersede spatial cues [19] . However, for patients maintaining offline-originated relationships, online consultations represent geographic continuity-seeking rather than spatial arbitrage. Hypotheses Development The theoretical tension between signal decay and signal verification in healthcare platforms centers on a single diagnostic question: do reputation signals function as temporary heuristics that patients discard upon acquiring direct experience, or as verified risk-reduction mechanisms that consolidate with repeated encounters? The persistence of diagnostic uncertainty in healthcare—where patients cannot fully assess service quality even after multiple consultations—suggests that signal verification may dominate over signal decay [20] . If verification prevails, reputation effects will demonstrate cross-stage stability or reinforcement trajectories as patients consolidate trust in validated signals.. H1: Physician reputation exhibits non-negative (stabilizing or reinforcing) effects on consultation volume across first-time to repeat consultation stages. Trust transfer theory suggests that offline-acquired patients possess imported trust that provides cognitive efficiency [21] , creating stable high baselines of signal reliance that resist experiential modification. Online-acquired patients, lacking such anchors, exhibit adaptive reliance patterns wherein sensitivity to platform signals intensifies post-verification. Specifically, online-acquired patients—lacking offline trust anchors—will exhibit lower reputation sensitivity during initial consultations, but this sensitivity will escalate post-verification as platform signals gain institutional credibility through direct experience. In contrast, offline-acquired patients, having imported face-to-face trust, will maintain stable high reputation reliance across both stages. This creates a channel-contingent moderation wherein the trajectory of reputation reliance diverges by acquisition origin. H2: Patient acquisition channel moderates the evolution of reputation reliance, such that online-acquired patients exhibit increasing reputation sensitivity across stages, while offline-acquired patients exhibit stable high sensitivity. Price functions as a market signal by conveying information under uncertainty [15] . However, the information value of any signal depreciates as receivers acquire private information through direct experience [15] . In initial consultations, patients may rely on price as a quality inference heuristic or as a cost barrier; in either case, the predictive validity of price for subsequent behavior diminishes as experience accumulation provides superior quality assessments. This information decay hypothesis posits that price loses its decisional weight—not necessarily reversing direction, but attenuating in magnitude—as patients transition from signal-dependent to experience-based evaluation. H3: The functional role of price transmutes from a quality-inference heuristic to a cost barrier as patient-physician relationships mature, with this transmutation manifesting as attenuation for online-acquired patients and directional reversal for offline-acquired patients. As relationships mature, service-dominant logic predicts a shift from transactional value (information exchange) to relational value (process efficiency). In initial consultations, high information asymmetry surrounding diagnosis creates substantial demand for detailed information disclosure; patients rely on extensive physician responses to assess service quality and reduce uncertainty. As clinical baselines establish and diagnostic uncertainty reduces, repeat consultations shift focus from information acquisition to therapeutic coordination. The marginal value of information volume thus declines once clinical baselines are established, because patients transition from evaluating diagnostic competence to managing ongoing care. Conversely, response speed gains importance as a coordination mechanism for chronic care management, reflecting an asymmetric escalation in process quality attributes. H4a: The influence of response speed on consultation volume will escalate across stages for offline-acquired patients, while attenuating for online-acquired patients. H4b: The influence of information volume on consultation volume attenuates across consultation stages. In platform-mediated professional services, physician reputation operates as a digital gatekeeper that determines consideration-set inclusion rather than merely shifting demand continuously. Drawing on endogenous consideration-set theory [22] , we argue that reputation signals serve as heuristic screening mechanisms: patients facing information overload use aggregate ratings to exclude low-reputation physicians entirely, creating structural barriers at the extensive margin (whether a physician is considered at all). Recent evidence from online labor and healthcare markets (2019–2023) confirms that platform ratings generate "winner-take-all" visibility dynamics, where sub-threshold providers face near-certain exclusion from patient choice sets [23] . Once a physician enters the consideration set, the decision calculus shifts from risk reduction to relational coordination. Here, process attributes—response speed and information richness—function as intensity drivers that shape consultation volume conditional on selection (the intensive margin). This hierarchical architecture, wherein aggregate signals govern entry and process attributes govern depth, aligns with recent platform economy research on two-stage consumer evaluation [24] . We therefore hypothesize: H5 Physician reputation primarily governs consideration-set formation (extensive margin), whereas process quality attributes drive consultation volume conditional on selection (intensive margin). Data and Methodology Temporal Research Design and Data Structure To establish predictive temporal precedence while mitigating reverse causality concerns, we employ a time-lagged research design comprising three observation stages: Stage 1 (T0: Baseline Period, 31 May 2017). We extracted comprehensive physician profiles preceding any observed consultation behavior. These time-invariant or pre-determined attributes serve as exogenous predictors: physician demographic characteristics (professional title, hospital tier, city level), platform tenure (registration duration), historical reputation accumulation (site-wide recommendation score aggregated through 31 May 2017), and service pricing structure (minimum posted fees). Crucially, these T0 measurements temporally precede the dependent variables, ensuring that physician attributes predict rather than reflect concurrent demand fluctuations. Stage 2 (T1: First-Consultation Window, 1–30 June 2017). We captured initial patient acquisition events, distinguishing between two distinct patient acquisition channels: (i) patients initiating contact via "paid online consultation services" without prior offline interaction (online-acquired), and (ii) patients transitioning from offline visits through the platform's "post-visit reporting service" (offline-acquired). These T1 outcomes represent relationship initiation, establishing the patient base available for subsequent tracking. Stage 3 (T2: Repeat-Consultation Window, 1 July–31 December, 2017). We tracked sustained engagement behaviors among patients acquired during T1, measuring cumulative repeat consultations within a six-month observation period. This duration accommodates asthma's clinical cyclicity (acute exacerbation, chronic persistence, and remission phases) while capturing meaningful relationship continuity. This T0→T1→T2 sequencing enables us to model how pre-existing physician characteristics (measured before any observed behavior) predict both initial patient attraction and subsequent relationship sustainability, approximating predictive causal precedence without experimental manipulation. Sample and Variables The analytic sample comprises 1,812 asthma specialists practicing across all 31 provincial-level jurisdictions of mainland China. The sample excludes physicians with missing profile data or zero platform tenure, and those without any consultation activity across the entire observation period. Our inclusion criteria require physicians to have complete T0 profile information and at least one measurable quality signal (Rating ≥ 3). Asthma constitutes an ideal empirical context: its chronic, relapsing-remitting nature necessitates iterative physician input and long-term medication management, making sustained doctor-patient relationships clinically essential rather than optional. We acknowledge potential selection constraints: Haodf.com physicians may not represent the full population of asthma specialists, and the 2017 data predates post-pandemic telemedicine expansion. However, the platform’s market position during the observation period and the national scope of our sample mitigate concerns about external validity for the studied context. The study examines four distinct consultation outcomes as dependent variables. Following the temporal structure, we operationalize four stage-specific dependent variables: OnFirstSale (T1), OffFirstSale (T1), OnRepeatSale (T2), and OffRepeatSale (T2), as defined in Table 1 . Repeat consultation is thus operationalized as any additional paid online interaction within six months of the index (first) consultation, capturing relationship sustainability rather than discrete transactional volume. Independent Variables. All physician quality signals—Rating, Price, ReplySpeed, and InfoAmount—are measured at T0 baseline, ensuring temporal precedence in predictive relationships. To enable direct comparison of effect sizes across metrics with disparate scales, all independent variables are standardized using z-score transformation (mean = 0, SD = 1) prior to model estimation (see Table 1 for operational definitions and descriptive statistics). Control Variables (T0-measured). Physician heterogeneity factors including Title, Hospital_3L, City, and Register are measured at baseline to control for institutional status, regional market conditions, and platform tenure effects (see Table 1 ). Table 1 Variables Definition Variables Variables Description OnFirstSale Count of first-time online consultations by patients acquired through digital channels (no prior offline contact). OnRepeatSale Cumulative repeat consultations within six months(T2) by online-acquired patients initially contacted during T1. OffFirstSale Count of first-time online consultations by patients transitioning from offline visits via post-visit reporting (established offline relationship migrating online). OffRepeatSale Cumulative repeat consultations within six months by offline-acquired patients who first connected online during T1. Rating Physician's platform-generated recommendation heat score (continuous, 1–5 scale), representing cumulative historical reputation preceding the observation window. Price The minimum service price across consultation modalities (text, voice). ReplySpeed Average response latency (days) between patient inquiry and physician reply, calculated from the 100 most recent consultations preceding the baseline observation (T0). InfoAmount Mean word count per physician response, operationalizing information richness (calculated from 100 most recent interactions). Title Professional hierarchy (Chief Physician; Associate Chief Physician; Attending/Resident ). Hospital_3L Binary indicator for Tier-3A hospital affiliation (highest accreditation tier). Register Platform tenure (years since registration as of 31 May 2017). City Urban hierarchy ranking (0 = most developed, 5 = least developed). Analytical Strategy: Zero-Inflated Negative Binomial Regression (ZINB) Model Selection. The dependent variables exhibit two statistical properties necessitating specialized handling. First, zero-inflation: a substantial proportion of physicians record zero consultations in specific categories, reflecting two distinct generative processes—(i) structural zeros, where physicians are excluded entirely from patient consideration sets due to low visibility or reputation (addressing RQ3 regarding consideration set thresholds), and (ii) random zeros, where patients evaluate but decline specific physicians [25] . Second, over-dispersion: variance substantially exceeds mean for all outcome variables, violating Poisson distribution assumptions. We employ Zero-Inflated Negative Binomial (ZINB) regression, which simultaneously models (a) the probability of structural zero occurrence via a logit component, and (b) consultation volume conditional on non-zero occurrence via a negative binomial count component. Estimation Strategy. To examine how cue effectiveness evolves across patient experience stages and acquisition channels, we implement a comprehensive analytical framework integrating stratified estimation, coefficient comparison tests, and interaction verification. Separate zero-inflated negative binomial (ZINB) models are estimated for each of the four outcome categories (OnFirstSale, OffFirstSale, OnRepeatSale, OffRepeatSale), enabling direct coefficient comparison across decision contexts. We then employ Seemingly Unrelated Estimation (SUEST) post-estimation procedures to statistically test whether reputation coefficients differ significantly between first-time versus repeat consultations and between online versus offline channels [26] . Pooled models with interaction terms (e.g., Rating × FirstPurchase, Rating × OnlineChannel) following reshape to long-format data, controlling for unobserved physician heterogeneity through clustering [27] . Descriptive Statistics Table 2 presents descriptive statistics for the full sample of 1,812 asthma specialists. The four consultation outcomes exhibit substantial dispersion, with standard deviations markedly exceeding means (e.g., OnFirstSale: mean = 12.16, SD = 42.73; OnRepeatSale: mean = 0.37, SD = 1.31), indicating right-skewed distributions with heavy tails. Notably, zero-inflation is pronounced in repeat consultation measures: 82.6% of physicians record zero online repeat sales (OnRepeatSale) and 77.8% record zero offline repeat sales (OffRepeatSale), while even first-time consultation variables exhibit substantial zero mass (OnFirstSale: 41.7% zeros; OffFirstSale: 43.7% zeros). Among physician quality signals, Rating concentrates at the upper bound (mean = 4.00 on a 3–5 scale, SD = 0.35), suggesting ceiling effects in platform reputation systems. Price exhibits wide heterogeneity (mean=¥27.13, SD = 41.60, range=¥5–¥900), reflecting diverse positioning strategies from accessible to premium services. ReplySpeed averages 0.34 days (SD = 0.37), indicating generally rapid physician responsiveness, while InfoAmount exhibits substantial variation (mean = 36.39 words, SD = 35.28). Control variables indicate that 35.4% hold Chief Physician titles, 34.9% hold Associate Chief titles, and 91.4% practice in Tier-3A hospitals, with platform tenure averaging 3.35 years (SD = 2.79). Table 2 Descriptive Statistics of Variables Variable N Mean SD Min Max OnFirstSale 1,812 12.156 42.729 0 593 OffFirstSale 1,812 16.793 36.282 0 370 OnRepeatSale 1,812 0.37 1.311 0 23 OffRepeatSale 1,812 0.7 2.314 0 41 Rating 1,812 4.001 0.352 3 5 price 1,812 27.134 41.596 5 900 lnInfoAmount 1,812 36.391 35.277 4.216 447.726 ReplySpeed 1,812 0.339 0.367 0 3 Chief 1,812 0.354 0.478 0 1 AssociateChief 1,812 0.349 0.477 0 1 Hospital3Level 1,812 0.914 0.28 0 1 City 1,812 1.666 1.44 0 6 Note: While Table 2 reports raw values for descriptive transparency, all continuous independent variables (Rating, Price, ReplySpeed, InfoAmount) are standardized (z-scored) in subsequent regression analyses to facilitate coefficient comparability and mitigate multicollinearity in interaction terms. The dependent variables remain in original count scale to preserve the zero-inflated, over-dispersed properties that motivate ZINB estimation. Results The Main Results We report ZINB estimates, which simultaneously model (a) the probability of structural zero occurrence (inflation component) and (b) consultation volume conditional on non-zero occurrence (count component). Table 3 presents ZINB estimates for the four stage-specific outcomes. Model 1 estimates the determinants of OnFirstSale (initial online consultations by online-acquired patients); Model 2 examines OffFirstSale (initial online consultations by patients transitioning from offline visits); Model 3 assesses OnRepeatSale (sustained online consultations by online-acquired patients within six months); and Model 4 corresponds to OffRepeatSale (sustained online consultations by offline-acquired patients). Consistent with the signal verification hypothesis, physician reputation (Rating) exhibits strong and persistent predictive power across all stages (β = 0.354–0.587, all p < 0.01). Notably, for online-acquired patients, the reputation coefficient for repeat consultations (β = 0.430, p < 0.01) is comparable to or slightly exceeds that for first-time consultations (β = 0.354, p < 0.01), suggesting that reputation signals retain or potentially increase salience as patient-physician relationships mature—a pattern consistent with trust verification rather than signal decay. For offline-acquired patients, reputation remains robustly significant across both stages (β = 0.587, p < 0.01 and 0.525, p < 0.01), indicating that established offline trust provides a durable foundation for online engagement. Price exhibits consistent negative effects on consultation volume, indicating straightforward price sensitivity or cost barrier effects in online healthcare. For online-acquired patients, higher prices significantly deter first-time consultations (β=-0.108, p < 0.05, Model 1), reflecting risk-aversion when quality is uncertain; this sensitivity dissipates for repeat consultations (β = 0.081, ns, Model 3) as patients establish quality assessments through direct experience. Strikingly, for offline-acquired patients, price turns strongly negative for repeat consultations (β=-0.229, p < 0.01, Model 4), suggesting that once offline trust is established, excessive pricing becomes a retention barrier—patients expect continued value from known physicians and are discouraged by high fees despite existing relationships. Process quality signals exhibit differential patterns by channel and stage. Response speed (ReplySpeed) exerts consistent negative effects across all specifications (all p < 0.01), indicating that faster response uniformly predicts higher consultation volume. Notably, the absolute magnitude of this effect varies: for online-acquired patients, the coefficient is strongest at the first consultation (β=-0.498, p < 0.01), suggesting immediate efficiency expectations among platform-native users; the effect persists for repeat consultations (β=-0.310, p < 0.01), though with somewhat reduced sensitivity. For offline-acquired patients, the speed effect intensifies from the initial (β=-0.228, p < 0.01) to the repeat consultation stage (β=-0.449, p < 0.01), suggesting that response efficiency becomes increasingly salient as these patients establish sustained online engagement patterns—potentially reflecting heightened expectations for digital convenience once offline familiarity is established. Information provision volume (InfoAmount) significantly predicts first-time consultations for both channels (β = 0.190, p < 0.01 for online, 0.112, p < 0.01 for offline), suggesting that information richness facilitates patient decision-making under uncertainty. However, this effect diminishes for online repeat consultations (β = 0.138, p < 0.05) and becomes non-significant for offline repeat consultations (β = 0.092, ns), consistent with the view that established relationships rely more on efficiency (speed) than on extensive information provision. Offline-acquired patients exhibit stronger reliance on reputation signals during initial online contact compared to online-acquired patients (OffFirstSale Rating: β = 0.587, p < 0.01 vs. OnFirstSale Rating: β = 0.354, p < 0.01), supporting the offline-to-online trust transfer hypothesis. These patients bring established face-to-face trust into the digital realm, seeking validation through high-reputation physicians. Conversely, online-acquired patients—lacking prior direct experience—distribute their attention across multiple cues, resulting in relatively lower marginal effects for individual signals. Geographic preferences diverge sharply by channel. City tier (City) exhibits no significant influence on either first-time or repeat consultations for online-acquired patients (β = 0.011 and − 0.032, both ns), suggesting that platform-mediated quality signals effectively supersede geographic proximity for online-native patients. In stark contrast, offline-acquired patients consistently prefer physicians in lower-tier cities (higher City values) for both initial (β = 0.197, p < 0.01, Model 2) and repeat consultations (β = 0.182, p < 0.01, Model 4), likely reflecting geographic proximity to their original offline treatment locations. The inflation component reveals reputation as a structural barrier to patient consideration. Across all four models, Rating carries a negative and highly significant coefficient in the zero-inflation equation (ranging from − 3.698, p < 0.01 to -5.973, p < 0.01), indicating that lower-reputation physicians face substantially elevated probabilities of being excluded entirely from patient choice sets (structural zeros) rather than merely being evaluated and rejected. This confirms RQ3: reputation operates not only as a volume determinant but as a threshold mechanism that determines whether physicians enter patient consideration sets at all. Professional titles (Chief, AssociateChief) significantly reduce the probability of structural zeros and increase expected volume conditional on non-zero outcomes. The effects of senior titles are generally stronger for repeat consultations (Chief β = 0.655, p < 0.01, Model 3; Chief β = 0.817, p < 0.01, Model 4) than for first-time consultations (Chief β = 0.215, p < 0.1, Model 1), suggesting that institutional credentials become more salient as patients deepen relationships and seek assurance for sustained engagements. SUEST coefficient comparison tests confirm that the difference in Chief Physician effects between repeat and first-time consultations is statistically significant (χ² = 7.83, p < 0.01), supporting the interpretation that institutional credentials gain salience as relationships mature. Table 3 Maximum Likelihood Parameter Estimates Model 1 (OnFirstSale) Model 2 (OffFirstSale) Model 3 (OnRepeatSale) Model 4 (OffRepeatSale) Variables Rating_std 0.354 *** (0.053) 0.587 *** (0.052) 0.430 *** (0.093) 0.525 *** (0.092) Price_std -0.108 ** (0.045) 0.052 (0.055) 0.081 (0.065) -0.229 *** (0.088) InfoAmount_std 0.190 *** (0.050) 0.112 *** (0.038) 0.138 ** (0.068) 0.092 (0.056) ReplySpeed_std -0.498 *** (0.044) -0.228 *** (0.037) -0.310 *** (0.080) -0.449 *** (0.079) Chief 0.215 * (0.130) 0.522 *** (0.111) 0.655 *** (0.189) 0.817 *** (0.178) Associate 0.263 ** (0.125) 0.443 *** (0.110) 0.331 * (0.189) 0.576 *** (0.180) Hospital3Level -0.443 ** (0.205) 0.128 (0.186) 0.228 (0.351) 0.376 (0.322) City 0.011 (0.041) 0.197 *** (0.038) -0.032 (0.059) 0.182 *** (0.059) _cons 3.270 *** (1.057) 1.896 ** (0.881) -0.139 (1.517) -0.819 (1.370) /lnalpha alpha 2.867 1.685 2.176 2.455 Zero inflation coef. Rating -5.531 *** (0.505) -3.698 *** (0.292) -4.138 *** (0.917) -5.973 *** (1.099) _cons 20.211 *** (1.875) 14.236 *** (1.125) 16.131 *** (3.429) 23.449 *** (4.129) Model summary obs 1812 1812 1812 1812 Nonzero obs 1057 946 316 396 Zero obs 755 866 1496 1416 LR chi2 163.11 231.65 80.93 108.19 Log likelihood -4882.327 -5055.638 -1150.085 -1526.670 Notes: * p < 0.1, ** p < 0.05, *** p < 0.01; Standard errors are in parentheses. Moderating Effects of ZINB Regression Results by Subgroups We now examine how the effectiveness of quality signals varies across purchase contexts and channels. Table 4 presents the results of Zero-Inflated Negative Binomial (ZINB) regression models examining the moderating effects of purchase type and channel choice on service sales. The analysis employs a subgroup design: Models 1 and 2 test the moderating effect of first-time purchase (vs. repeat purchase) within online and offline channels respectively; Models 3 and 4 test the moderating effect of online channel (vs. offline) within first-time and repeat purchase samples, respectively. Each model includes main effects of physician rating (Rating_std), service price (price_std), information amount (lnInfoAmount_std), response speed (ReplySpeed_std), and physician/hospital characteristics. The interaction terms capture how the effectiveness of physician reputation and service attributes varies across purchase contexts. The inflation part models the excess zero counts using physician rating as the inflation predictor. Consistent with the signal verification framework, physician reputation (Rating_std) exhibits strong and statistically significant positive effects on consultation volume across all four subgroups (β = 0.467–0.597, all p < 0.01). This persistence indicates that platform-generated reputation signals retain predictive power regardless of whether patients are acquired through digital channels or offline referrals, and whether they are engaging in first-time or repeat consultations. Notably, the point estimates suggest relatively higher reputation sensitivity in offline-acquired repeat consultations (β = 0.597, p < 0.01) and first-time consultations overall (β = 0.563, p < 0.01), compared to online-acquired repeat consultations (β = 0.467, p < 0.01). Process quality signals demonstrate differential baselines: response speed (ReplySpeed_std) consistently predicts higher volume across all contexts (β=-0.233 to -0.453, p < 0.01), while information amount (InfoAmount_std) shows modest positive effects (β = 0.086–0.121, p < 0.1 or p < 0.05) except in the offline channel where significance is marginal. The interaction terms reveal significant heterogeneity in signal effectiveness between first-time and repeat consultations, particularly within the online channel. In Model 1 (Online Channel), the negative and significant interaction between First Purchase and Rating_std (β=-0.290, p < 0.01) indicates that the reputation effect is substantially attenuated for first-time consultations (implied effect ≈ 0.307, p < 0.01) relative to repeat consultations (baseline β = 0.597, p < 0.01). This pattern is consistent with trust verification mechanisms wherein validated signals gain salience post-experience. Conversely, in Model 2 (Offline Channel), the interaction term is statistically indistinguishable from zero (β=-0.076, n.s.), indicating that offline-acquired patients maintain stable reputation responsiveness across both purchase stages. Price signaling exhibits contrasting stage dynamics by channel. In the online channel, the positive price effect observed in repeat consultations (baseline β = 0.205, p < 0.01) dissipates for first-time purchases, as evidenced by the negative interaction term (β=-0.212, p < 0.05), resulting in a near-zero net effect for initial consultations. This suggests that online-acquired patients may interpret price as a quality signal only after establishing initial contact. In contrast, offline-acquired patients show no significant price sensitivity in repeat consultations (baseline β = 0.026, n.s.), but display a positive price effect for first-time consultations (implied effect = 0.026 + 0.181 = 0.207, p < 0.05), consistent with price serving as a screening device when offline trust has not yet been established. Response speed priorities also diverge by stage and channel. In the online channel, the negative interaction between First Purchase and ReplySpeed_std (β=-0.195, p < 0.05) indicates that first-time patients exhibit heightened sensitivity to response latency (implied effect ≈ -0.501) compared to repeat patients (β=-0.306, p < 0.01). However, in the offline channel, the pattern reverses: the positive interaction term (β = 0.20, p < 0.05) suggests that repeat consultations demonstrate stronger speed sensitivity (baseline β=-0.442, p < 0.01) than first-time consultations (implied effect ≈ -0.235). Channel Differences within Purchase Stages (Models 3–4). When examining channel effects within purchase stages, Model 3 (First Purchase) reveals that the online channel significantly dampens reputation effectiveness relative to offline channels (interaction β=-0.245, p < 0.01), reducing the implied reputation effect for online first-time consultations to approximately 0.318 (compared to 0.563 for offline). This attenuation may reflect greater information diversification among online-acquired patients. For repeat consultations (Model 4), the channel interaction for reputation is negative but statistically insignificant (β=-0.090, n.s.), suggesting comparable reputation sensitivity across channels at the repeat stage. Price sensitivity in first-time consultations shows significant channel variation: the negative interaction between Online Channel and ln_price (β=-0.168, p < 0.05) superimposed on the positive offline baseline (β = 0.191, p < 0.01) results in a null net effect for online first-time purchases, while preserving the positive offline effect. Interestingly, for repeat consultations, the channel interaction for price is positive and significant (β = 0.259), transforming the offline null effect (β=-0.063, n.s.) into a positive online effect (implied ≈ 0.196), suggesting that price-quality inferences become salient for online-acquired patients only after initial experience accumulation. Zero-Inflation Component: Extensive Margin Evidence The inflation part of the ZINB models provides complementary evidence regarding the extensive margin of patient choice. Physician rating exhibits large negative coefficients across all specifications (β=-4.084 to -5.518, all p < 0.01), indicating that higher reputation significantly reduces the probability of zero consultations (i.e., the likelihood that a physician records no sales in a given period). This finding reinforces the main effects by demonstrating that reputation signals not only increase the volume of consultations conditional on purchase, but also expand the probability of any purchase occurring. The likelihood ratio tests (LR chi2) indicate that all four models provide statistically significant improvement over null models, with substantial log-likelihood values ranging from − 2686.475 to -9971.521. The dispersion parameters (alpha) range from 1.924 to 2.865, confirming overdispersion in the count data and validating the selection of ZINB over standard Poisson specifications. Table 4 Moderating Effects of ZINB Regression Results by Subgroups Variable Online Channel (Online = 1) Offline Channel (Online = 0) First Purchase (First = 1) Repeat Purchase (First = 0) Main Effects (Baseline Group) Rating_std 0.597 *** (0.075) 0.597 *** (0.073) 0.563 *** (0.059) 0.467 *** (0.087) price_std 0.205 *** (0.073) 0.026 (0.073) 0.191 *** (0.060) -0.063 (0.082) InfoAmount_std 0.121 * (0.066) 0.086 * (0.049) 0.103 ** (0.043) 0.094 * (0.055) ReplySpeed_std -0.306 *** (0.080) -0.442 *** (0.073) -0.233 *** (0.042) -0.453 *** (0.078) Chief 0.321 *** (0.108) 0.566 *** (0.096) 0.332 *** (0.089) 0.699 *** (0.133) Associate 0.245 ** (0.104) 0.447 *** (0.094) 0.305 *** (0.087) 0.422 *** (0.133) Hospital3Level -0.293 * (0.167) 0.221 (0.158) -0.164 (0.144) 0.301 (0.238) City 0.021 (0.034) 0.237 *** (0.033) 0.140 *** (0.030) 0.115 *** (0.044) _cons -0.522 (0.865) -1.337 * (0.749) 2.448 *** (0.707) -0.322 (1.022) Channel/Purchase Type Main Effects First/Online 3.750 *** (0.091) 3.190 *** (0.085) -0.145 ** (0.071) -0.635 *** (0.113) Interaction Effects (Moderating Effects) First/Online × Rating_std -0.290 *** (0.089) -0.076 (0.088) -0.245 *** (0.075) -0.090 (0.106) First/Online × price_std -0.212 ** (0.091) 0.181 ** (0.088) -0.168 ** (0.080) 0.259 ** (0.105) First/Online × InfoAmount_std 0.054 (0.082) 0.007 (0.063) 0.057 (0.064) 0.023 (0.088) First/Online × ReplySpeed_std -0.195 ** (0.091) 0.207 ** (0.083) -0.255 *** (0.059) 0.159 (0.111) Inflation Part (Zero-Inflation Model) Rating -5.518 *** (0.481) -4.084 *** (0.311) -4.733 *** (0.302) -5.019 *** (0.667) _cons 20.209 *** (1.783) 15.744 *** (1.186) 17.638 *** (1.133) 19.738 *** (2.501) Dispersion Parameters /lnalpha 1.053 *** (0.048) 0.655 *** (0.065) 0.909 *** (0.044) 0.848 *** (0.110) alpha 2.865 1.924 2.481 2.335 Model Summary Statistics Observations 3624 3624 3624 3624 Nonzero obs 1373 1342 2003 712 Zero obs 2251 2282 1621 2912 LR chi2 1369.45 1307.76 403.91 217.10 Log likelihood -6042.059 -6596.683 -9971.521 -2686.475 Notes: * p < 0.1, ** p < 0.05, *** p < 0.01; Standard errors are in parentheses. Models 1–2: Test the moderating effect of First Purchase (First) in online/offline channels; Models 3–4: Test the moderating effect of Online Channel (Online) in first/repeat purchases. Discussion Theoretical Implications: Dual-Signal Architecture and Channel Heterogeneity Our findings illuminate a dual-stage decision architecture in platform-mediated healthcare that extends conventional signaling theory. Consistent with H1, physician reputation and process quality operate as functionally distinct rather than competing mechanisms: reputation serves as a threshold gatekeeper determining consideration-set inclusion (evidenced by significant negative coefficients in the zero-inflation component), while response speed and information depth function as intensity drivers shaping volume conditional on physician entry. This complementarity suggests that in high-stakes credence services, patients rely on hierarchical evaluation schemas wherein platform-aggregated historical signals reduce search risk prior to engagement, and process attributes fine-tune ongoing relationship investment. Our findings extend recent evidence on digital healthcare quality assessment. Gong et al. [9] demonstrate comparable diagnostic quality between asynchronous webchats and in-person consultations for postpartum depression, suggesting that process quality attributes can substitute for physical presence in certain care contexts. Our results complement this by showing that response efficiency, rather than information depth, becomes the dominant process driver in sustained relationships—potentially because established patients have already resolved initial diagnostic uncertainty. This alignment between cross-sectional quality comparisons and our longitudinal signal evolution patterns strengthens the case for platform-mediated chronic disease management. H2 reveals that this dual-signal architecture is contingent upon patient knowledge endowments at platform entry. Offline-acquired patients—those transitioning from face-to-face encounters—exhibit stable high sensitivity to reputation signals and geographic continuity-seeking preferences. This pattern is consistent with trust transfer theory [21] : imported offline trust may serve as cognitive anchors that stabilize platform signal reliance, rendering reputation a confirmatory device rather than an exploratory heuristic. Conversely, online-acquired patients demonstrate spatial substitution effects, exhibiting reduced geographic constraints as platform-mediated quality signals effectively decouple their choices from physical proximity. They display constructive signal reliance, characterized by lower initial reputation sensitivity and heightened responsiveness to process quality, reflecting active knowledge construction in the absence of offline anchors. The geographic divergence we observe further nuances the digital disruption narrative. Online-acquired patients exhibit spatial substitution—null or negative associations with city tier—as platform signals effectively decouple quality assessment from physical proximity. In contrast, offline-acquired patients demonstrate geographic continuity-seeking, preferring physicians in lower-tier cities proximal to original treatment locations. This suggests that for relationship-maintenance scenarios, online platforms serve complementary infrastructure extending offline therapeutic relationships, rather than pure market mechanisms optimizing for quality independent of geography. While our confirmatory hypotheses address static signal-channel relationships, exploratory analyses of stage-wise variation reveal temporal dynamics that extend our theoretical framework. For online-acquired patients, reputation effects maintain or intensify across consultation stages, contradicting conventional signal decay predictions. This is consistent with a verification mechanism: initial consultations function as calibration opportunities where patients assess platform signal accuracy. When observed quality aligns with reputation indicators, cognitive consonance [11] reinforces trust in the platform's signaling infrastructure, creating self-reinforcing reliance cycles. This challenges the universality of signal decay and suggests that in contexts of persistent uncertainty, validated platform signals achieve institutional trust status that deepens with experience. Price signals exhibit functional transmutation across relationship stages. For online-acquired patients, price operates as a quality inference heuristic (non-negative initial associations) that attenuates or reverses as experience accumulates, transitioning to pure cost consideration. This aligns with means-end chain theory [28] , price shifts from diagnostic attribute to sacrifice attribute as private quality knowledge develops. Offline-acquired patients, possessing pre-validated quality assessments, demonstrate immediate cost sensitivity across stages, suggesting that trust transfer accelerates price function maturation. Process quality attributes demonstrate asymmetric escalation. Response speed gains increasing salience across stages—particularly for offline-acquired patients—supporting service-dominant logic [18] : as relationships mature from diagnostic to maintenance phases, coordination efficiency supersedes information acquisition as the primary value driver. Conversely, information volume exhibits satiation effects, particularly for online-acquired patients, suggesting that detailed explanations facilitate initial decision-making but yield diminishing returns once clinical baselines are established. Practical Implications Platform Design. Interfaces should implement adaptive signal architectures based on patient origin. For online acquisition, prioritize process quality transparency (response time guarantees, detailed service descriptions) to compensate for absent offline trust. For offline acquisition, emphasize geographic proximity indicators linking online physicians to original treatment locations, and prioritize reputation reinforcement over initial trust establishment. Implementing adaptive interfaces requires careful navigation of privacy constraints—channel identification may inadvertently reveal patient medical history—and physician workflow considerations, as response time guarantees necessitate compensation mechanisms or workload management tools. The cost-effectiveness of such segmentation should be evaluated against uniform interface designs. Physician Strategy. Practitioners must optimize both threshold and intensity mechanisms. Investment in platform reputation (review solicitation, recommendation score maintenance) functions as market entry tickets, while process efficiency (rapid response protocols) drives relationship sustainability. Physicians serving offline-acquired patients should prioritize responsiveness for retention, as this segment exhibits steeply escalating speed sensitivity. Those attracting online-acquired patients should initially emphasize information richness to facilitate risk reduction under uncertainty. Healthcare Policy. Digital platforms can function as spatial equalization mechanisms for online-acquired patients, enabling access to high-reputation specialists regardless of geographic hierarchy. However, the continuity-seeking behavior of offline-acquired patients indicates that local healthcare infrastructure remains critical for relationship initiation. Policymakers should view online platforms as complementary extensions of traditional care networks rather than substitute solutions, investing in hybrid care pathways that leverage digital efficiency while maintaining offline diagnostic capacity. Limitations and Future Research Several limitations warrant caution in interpreting our findings. First, data from 2017 may not fully capture post-pandemic telemedicine adoption patterns, where patient digital literacy and regulatory environments have shifted substantially. Second, as Haodf.com operates recommendation algorithms that may prioritize certain physicians, platform governance effects may confound pure signal interpretations. Third, our physician-level aggregation precludes observation of patient-level heterogeneity in disease severity and health literacy. Our physician-level aggregated data precludes observation of patient-level heterogeneity (disease severity, health literacy) and cross-physician learning effects. While temporal sequencing establishes precedence, we cannot fully exclude selection effects wherein unobserved physician characteristics simultaneously influence signal profiles and patient attraction. Future research employing patient-level panel data could disentangle self-selection from true learning effects. Our findings suggest signal verification as a potential boundary condition to the signal decay paradigm, particularly in high-stakes credence services where diagnostic uncertainty persists despite accumulated interactions. We caution that this pattern’s generalizability may be limited to similar professional service contexts. The signal verification mechanism we identify may operate more strongly in contexts of persistent outcome uncertainty (chronic diseases, complex diagnostics). Future research should test boundary conditions across acute versus chronic conditions, insurance-based versus out-of-pocket payment models, and cultural contexts varying in platform trust. Finally, our binary channel classification (online/offline) could be extended to hybrid trajectories—patients oscillating between digital and physical channels. Understanding temporal dynamics of trust transfer across these hybrid paths represents a critical avenue for research on healthcare system integration. Conclusion This study advances platform-mediated service theory by demonstrating that in professional service contexts, institutional trust signals and experiential learning operate as complementary governance mechanisms. Contrary to substitution hypotheses, our findings reveal that platform-verified quality cues gain predictive salience as patient-physician relationships mature, suggesting that patients use direct experience to validate rather than replace institutional signals. The identified acquisition channel boundary condition further indicates that online channels facilitate trust construction through process transparency, while offline channels require sustained relational continuity to maintain transferred trust—offering a parsimonious framework applicable to legal, financial, and educational platforms. These insights carry direct implications for dual-signal platform architecture: reputation investments should target consideration-set formation (search stage), while process quality investments should focus on relationship intensity (retention stage). Resource allocation must further differentiate by patient origin—prioritizing transparency cues for online-acquired users and geographic continuity for offline-acquired cohorts. For policymakers, the evidence suggests that digital health platforms reduce spatial barriers to specialist access but cannot substitute local infrastructure for relationship initiation; regulatory frameworks should thus incentivize platform-care network integration rather than standalone digital provision. Future research should examine hybrid care models combining synchronous telemedicine access with asynchronous follow-up, while addressing self-selection limitations through experimental designs that causally identify channel effects. Declarations This study was reviewed and granted an exemption by the Academic Committee of Inner Mongolia University. The research employs retrospective, de-identified administrative data obtained from Haodf.com under a data use agreement. All patient-identifying information was irreversibly removed by the platform prior to data provision; researchers accessed only physician-level aggregates and anonymized interaction metadata (response timestamps and word counts) with no ability to re-identify individual patients. The study involves no human subjects interaction, no intervention, and no collection of identifiable private information. All analyses adhered to the Personal Information Protection Law of the People's Republic of China and Haodf.com's data privacy policies. Competing interests: The authors declare no competing interests. Funding: Natural Science Foundation of Inner Mongolia (2022QN07006). Acknowledgments: The authors gratefully acknowledge Haodf.com for providing the data used in this study. This research was supported by the Inner Mongolia Natural Science Foundation (2022QN07006). Ethical Considerations : This study employed retrospective, aggregated administrative data obtained from Haodf.com via a data use agreement. The research constitutes a secondary analysis of pre-existing commercial platform data and does not involve human subjects research as defined by the Common Rule. Given that (1) data were analyzed at the physician level with no individual patient identifiers, (2) the study was retrospective with no intervention or interaction with participants, and (3) data access was authorized by the platform under commercial research collaboration terms, formal IRB review was not required. All analyses adhered to the Personal Information Protection Law of the People's Republic of China and Haodf.com's data privacy policies. Data Availability: The data used in this study are proprietary and were obtained from Haodf.com under a data use agreement. Due to commercial confidentiality and user privacy agreements, the data are not publicly available. Requests for data access should be directed to Haodf.com or the corresponding author. References Yach D, Hawkes C, Gould CL, Hofman KJ. The global burden of chronic diseases: overcoming impediments to prevention and control [J]. Journal of the American Medical Association, 2004, 291(21): 2616-2622. Yu M, He S, Wu D, Zhu H, Webster C. Examining the Multi-Scalar Unevenness of High-Quality Healthcare Resources Distribution in China [J]. International Journal of Environmental Research and Public Health, 2019, 16(16). Sun S, Li T, Zheng A, Zhang Z, Wang Q, Chen C, Zeng Z. Doctor-patient-family collaboration in community-based chronic disease management to enhance multidimensional value [J]. Patient Education and Counseling, 2025, 132. Shapoo N, Shapoo N, Rehman A, Boma N. Smart Healthcare: The Role of Digital Health in Modern Medicine [J]. Health Care Science, 2025, 4(3): 179-187. Buchanan J, Roope LSJ, Morrell L, Pouwels KB, Robotham JV, Abel L, Crook DW, Peto T, Butler CC, Walker AS, Wordsworth S. Preferences for medical consultations from online providers: evidence from a discrete choice experiment in the United Kingdom [J]. Applied Health Economics and Health Policy, 2021, 19(4): 521-535. Wu J, Huang X, Sun P, Zhang X. What affects patients' choice of consultant: an empirical study of online doctor consultation service [J]. Electronic Commerce Research, 2022. Shen T, Li Y, Chen X. A Systematic Review of Online Medical Consultation Research [J]. Healthcare, 2024, 12(17). Thaler RH. Mental accounting and consumer choice [J]. Marketing Science, 2008, 27(1): 15-25. Gong W, Liu L, Li X, Caine ED, Shi J, Zeng Z, Cheng KK. Quality of asynchronous webchats vs in-person consultations for postpartum depression in China: a cross-sectional, mixed methods study using standardized patients [J]. Lancet Regional Health Western Pacific, 2024, 45. Boulding W, Kirmani A. A consumer-side experimental examination of signaling theory: do consumers perceive warranties as signals of quality? [J]. Journal of Consumer Research, 1993, 20(1): 111-123. Aronson E. The theory of cognitive dissonance: A current perspective [M]. Advances in Experimental Social Psychology. Elsevier. 1969: 1-34. Erdem T, Swait J. Brand equity as a signaling [J]. Journal of Consumer Psychology, 2001, 7(2): 131-157. Erdem T, Swait J. Brand credibility, brand consideration, and choice [J]. Journal of Consumer Research, 2004, 31(1): 191-198. Bagwell K, Riordan MH. High and declining prices signal product quality [J]. The American Economic Review, 1991, 81(1): 224-239. Kirmani A, Rao AR. No pain, no gain: A critical review of the literature on signaling unobservable product quality [J]. Journal of Marketing, 2000, 64(2): 66-79. Zeithaml VA. Consumer perceptions of price, quality, and value: a means-end model and synthesis of evidence [J]. The Journal of Marketing, 1988, 52(3): 2-22. Komiak SX, Benbasat I. Understanding customer trust in agent-mediated electronic commerce, web-mediated electronic commerce, and traditional commerce [J]. Information Technology and Management, 2004, 5(1): 181-207. Vargo SL, Lusch RF. Evolving to a new dominant logic for marketing [M]. The Service Dominant Logic of Marketing. Routledge. 2014: 3-28. Goldfarb A, Tucker C. Digital economics [J]. Journal of Economic Literature, 2019, 57(1): 3-43. Darby MR, Karni E. Free competition and the optimal amount of fraud [J]. The Journal of Law and Economics, 1973, 16(1): 67-88. Stewart KJ. Trust transfer on the world wide web [J]. Organization Science, 2003, 14(1): 5-17. Yaraghi N, Wang W, Gao G, Agarwal R. How online quality ratings influence patients’ choice of medical providers: controlled experimental survey study [J]. Journal of Medical Internet Research, 2018, 20(3): e99. Yang H, Du HS, Shang W. Understanding the influence of professional status and service feedback on patients' doctor choice in online healthcare markets [J]. Internet Research, 2021, 31(4): 1236-1261. Rietveld J, Schilling MA, Bellavitis C. Platform strategy: Managing ecosystem value through selective promotion of complements [J]. Organization Science, 2019, 30(6): 1232-1251. Lambert D. Zero-inflated Poisson regression, with an application to defects in manufacturing [J]. Technometrics, 1992, 34(1): 1-14. Weesie H. On seemingly unrelated estimation and the cluster-adjusted sandwich estimator [J]. Stata Technical Bulletin, 1999, 52: 34-47. Aiken LS. Multiple regression: Testing and interpreting interactions [M]. sage Newbury Park, CA, 1991. Gutman J. A Means-End Chain Model Based on Consumer Categorization Processes [J]. Journal of Marketing, 1982, 46(2): 60-72. Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9463453","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":625769524,"identity":"296f9d6c-aeaa-4ad4-9014-6c5dfa2db85b","order_by":0,"name":"Yumei Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAr0lEQVRIiWNgGAWjYBACPhDx8Z8NDz97A5Fa2ICYcQZbmoxkzwEStDDzsB22MbjhQKwWiewEBh6e8zwMNxgYP3zMIUpL7gYGCYnbPIyzG5glZ24jVouBwW0eZpkDbMy8RGtJSDjHwyaRQIqWAwcO8PAQr4Xn7QbGxoZkHgmeg83E+YWfPXcD898GO3v7480HP3wkRguDQAL7DwiLsYEY9SBrDhCpcBSMglEwCkYuAADaSS+IntUa/gAAAABJRU5ErkJggg==","orcid":"","institution":"School of Economics and Management, Inner Mongolia University","correspondingAuthor":true,"prefix":"","firstName":"Yumei","middleName":"","lastName":"Li","suffix":""}],"badges":[],"createdAt":"2026-04-19 16:03:29","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9463453/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9463453/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107752399,"identity":"723f3441-c9d2-4505-b970-641ad53e75db","added_by":"auto","created_at":"2026-04-24 17:39:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":540148,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9463453/v1/053a0549-0813-4964-ac60-2409c6c0d619.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eSustaining Digital Therapeutic Relationships: The Evolution of Service Quality Cue Utilization from First to Repeat Consultations\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eChronic diseases\u0026mdash;cardiovascular conditions, respiratory disorders, diabetes, and related non-communicable illnesses\u0026mdash;represent a critical global public health challenge. In China, rising prevalence rates, accelerated aging demographics, and lifestyle shifts have exacerbated their socioeconomic burden\u003csup\u003e[1]\u003c/sup\u003e. Asthma, a chronic respiratory disease requiring long-term management, epitomizes this crisis: its complex etiology demands continuous medical supervision, yet healthcare resources remain scarce and geographically unevenly distributed\u003csup\u003e[2]\u003c/sup\u003e. Among chronic conditions, asthma presents an ideal empirical context for examining sustained digital therapeutic relationships: its relapsing-remitting nature necessitates iterative physician input; its symptom-based diagnostic criteria facilitate high-quality asynchronous online consultations; and its treatment protocols require long-term medication adherence monitoring, making continuity of care both clinically critical and technologically feasible through digital platforms. Traditional healthcare systems struggle to deliver accessible, continuous care, particularly for chronic conditions where sustained doctor-patient relationships correlate strongly with clinical outcomes and operational efficiency\u003csup\u003e[3]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eDigital healthcare platforms have emerged as transformative solutions, transcending geographical constraints to connect patients with physicians nationwide\u003csup\u003e[4]\u003c/sup\u003e. Platforms like Haodf.com enable asynchronous consultations, expanding patient choice while generating rich behavioral data. However, the very abundance of physician information (reputation scores, service pricing, response metrics) creates decision complexity for patients. Crucially, while initial consultation choices have been studied\u003csup\u003e[5\u0026ndash;7]\u003c/sup\u003e, how patients sustain relationships through repeat consultations, and how their quality signal preferences evolve, remains poorly understood despite its centrality to chronic disease management.\u003c/p\u003e \u003cp\u003eExisting research exhibits three critical limitations. First, studies predominantly examine static, cross-sectional consultation choice\u003csup\u003e[3]\u003c/sup\u003e, neglecting how patients\u0026rsquo; information processing evolves across the relationship lifecycle. While psychological accounting theory suggests repeat consumers shift from risk-minimization to value-maximization\u003csup\u003e[8]\u003c/sup\u003e, emerging evidence suggests quality signals may undergo verification and reinforcement rather than simple decay. Patients who confirm platform reputation through direct experience may actually increase reliance on such signals, creating a \"trust consolidation\" effect that remains untested in digital healthcare contexts. Second, patient heterogeneity is oversimplified. Patients originate via distinct channels: online-acquired patients (initiating contact through direct platform search without prior interaction) rely on platform-mediated signals, while offline-acquired patients (transitioning from face-to-face encounters via the platform\u0026rsquo;s post-visit reporting service, importing pre-existing therapeutic trust) possess direct quality assessments\u003csup\u003e[9]\u003c/sup\u003e. However, conventional wisdom assumes offline patients rely less on platform signals, a presumption contradicted by preliminary evidence suggesting offline-acquired patients may transfer strong offline trust into sustained online engagement with high-reputation physicians. Third, methodological constraints plague existing studies. Most rely on single-period observations, rendering them unable to disentangle whether quality signals predict subsequent behavior or merely reflect concurrent demand. We address this by employing a time-lagged research design: physician service attributes measured in May 2017 predict first-time consultations in June 2017, which in turn predict repeat consultations from July to December 2017. This temporal sequencing mitigates reverse causality concerns while capturing relationship evolution.\u003c/p\u003e \u003cp\u003eFrom a signal-theoretic perspective, physician reputation may function not merely as a volume determinant but as a threshold gatekeeper\u0026mdash;creating structural barriers that exclude low-reputation physicians from patient consideration sets entirely. This suggests a dual-stage decision architecture wherein reputation governs the extensive margin (whether a physician is considered at all), while process quality shapes the intensive margin (consultation volume conditional on consideration). To address these theoretical and methodological limitations, this study employs a temporal sequencing design to investigate three specific research questions regarding (1) cue evolution across relationship stages, (2) channel-based heterogeneity in signal reliance, and (3) the dual mechanisms governing consideration-set formation versus volume generation. Specifically, we examine how physician service quality cues\u0026mdash;reputation, price, information richness, and response speed\u0026mdash;impact initial and repeat consultation volumes across patient acquisition channels.\u003c/p\u003e \u003cp\u003eOur empirical strategy employs zero-inflated negative binomial regression (ZINB) to accommodate the over-dispersed and zero-heavy distribution of consultation counts across 1,812 asthma specialists, distinguishing between the structural probability of zero consultations and volume conditional on selection. We control for institutional and geographic heterogeneity (physician title, hospital tier, city hierarchy) that may confound signal effects, as physicians in tertiary hospitals or developed cities may possess both higher baseline reputation and differential patient access opportunities.\u003c/p\u003e \u003cp\u003eThis study contributes to platform-mediated service literature by: (1) challenging the signal decay paradigm through temporal analysis of cue evolution across first-time and repeat consultations; (2) delineating how patient acquisition channels moderate signal reliance in sustained digital therapeutic relationships; and (3) distinguishing between consideration-set formation (structural zeros governed by reputation thresholds) and volume drivers (conditional consultation behavior shaped by process quality) in digital healthcare choice. We guide platform designers and physicians in segment-specific engagement strategies\u0026mdash;optimizing reputation and pricing for patient acquisition, while emphasizing response efficiency for chronic disease management retention.\u003c/p\u003e"},{"header":"Research Hypotheses","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eTheoretical Foundation: Signal Evolution in Platform-Mediated Healthcare\u003c/h2\u003e \u003cp\u003eThe prevailing signal decay paradigm in consumer behavior literature posits an inverse relationship between experience accumulation and extrinsic cue reliance\u003csup\u003e[10]\u003c/sup\u003e. This theoretical tradition suggests that consumers progressively substitute direct experience for platform-mediated signals as cognitive investments yield quality information. However, this framework encounters boundaries in high-stakes, credence-dominant service contexts characterized by persistent outcome uncertainty.\u003c/p\u003e \u003cp\u003eSignal Verification and Trust Consolidation. Contrary to the decay thesis, cognitive consistency theory suggests that verified expectations strengthen belief structures\u003csup\u003e[11]\u003c/sup\u003e. When initial service encounters confirm platform reputation signals, patients experience cognitive consonance that validates the platform's signaling infrastructure\u003csup\u003e[12]\u003c/sup\u003e. This generates trust consolidation\u0026mdash;a self-reinforcing cycle wherein validated signals increase in perceived reliability over time. In healthcare, where diagnostic uncertainty persists despite accumulated interactions, reputation signals serve as risk anchors that maintain or increase salience as relationships deepen. Unlike consumer goods where direct experience can fully substitute for extrinsic cues, healthcare platforms create conditions under which verified reputation signals consolidate rather than decay\u003csup\u003e[13]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eFunctional Transmutation of Economic Signals. Information economics distinguishes between the signaling function and allocative function of price\u003csup\u003e[14]\u003c/sup\u003e. Under high information asymmetry, price serves as a diagnostic attribute enabling quality inference \u003csup\u003e[15]\u003c/sup\u003e; as consumers acquire direct experience, the information content of price marginalizes, and its role shifts toward pure cost assessment \u003csup\u003e[16]\u003c/sup\u003e. This transmutation implies that price loses predictive power for consumer choice as experience accumulates, regardless of whether price initially functioned as a quality signal or a budget constraint.\u003c/p\u003e \u003cp\u003eChannel-Contingent Information Processing. Trust transfer theory posits that offline-acquired patients import pre-existing relational trust from face-to-face encounters, creating stable knowledge anchors that reduce sensitivity to experiential variation\u003csup\u003e[17]\u003c/sup\u003e. Conversely, online-acquired patients face pure platform uncertainty, necessitating active learning. Their cue reliance follows constructive trajectories: initial diversification across signals (risk-spreading) followed by concentration on verified signals (risk-reduction) as experience accumulates.\u003c/p\u003e \u003cp\u003eProcess Quality and Value Co-Creation. Service-dominant logic conceptualizes patients as value co-creators\u003csup\u003e[18]\u003c/sup\u003e. As relationships mature from diagnostic establishment to chronic management, the locus of value shifts from information acquisition (content) to relational continuity (process efficiency). This transition implies escalating salience of response speed as a coordination mechanism for sustained care.\u003c/p\u003e \u003cp\u003eGeographic Proximity and Digital Substitution. Information economics suggests that digital platforms function as geographic proximity substitutes by aggregating quality signals that supersede spatial cues\u003csup\u003e[19]\u003c/sup\u003e. However, for patients maintaining offline-originated relationships, online consultations represent geographic continuity-seeking rather than spatial arbitrage.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eHypotheses Development\u003c/h3\u003e\n\u003cp\u003eThe theoretical tension between signal decay and signal verification in healthcare platforms centers on a single diagnostic question: do reputation signals function as temporary heuristics that patients discard upon acquiring direct experience, or as verified risk-reduction mechanisms that consolidate with repeated encounters? The persistence of diagnostic uncertainty in healthcare\u0026mdash;where patients cannot fully assess service quality even after multiple consultations\u0026mdash;suggests that signal verification may dominate over signal decay\u003csup\u003e[20]\u003c/sup\u003e. If verification prevails, reputation effects will demonstrate cross-stage stability or reinforcement trajectories as patients consolidate trust in validated signals..\u003c/p\u003e \u003cp\u003eH1: Physician reputation exhibits non-negative (stabilizing or reinforcing) effects on consultation volume across first-time to repeat consultation stages.\u003c/p\u003e \u003cp\u003eTrust transfer theory suggests that offline-acquired patients possess imported trust that provides cognitive efficiency\u003csup\u003e[21]\u003c/sup\u003e, creating stable high baselines of signal reliance that resist experiential modification. Online-acquired patients, lacking such anchors, exhibit adaptive reliance patterns wherein sensitivity to platform signals intensifies post-verification. Specifically, online-acquired patients\u0026mdash;lacking offline trust anchors\u0026mdash;will exhibit lower reputation sensitivity during initial consultations, but this sensitivity will escalate post-verification as platform signals gain institutional credibility through direct experience. In contrast, offline-acquired patients, having imported face-to-face trust, will maintain stable high reputation reliance across both stages. This creates a channel-contingent moderation wherein the trajectory of reputation reliance diverges by acquisition origin.\u003c/p\u003e \u003cp\u003eH2: Patient acquisition channel moderates the evolution of reputation reliance, such that online-acquired patients exhibit increasing reputation sensitivity across stages, while offline-acquired patients exhibit stable high sensitivity.\u003c/p\u003e \u003cp\u003ePrice functions as a market signal by conveying information under uncertainty\u003csup\u003e[15]\u003c/sup\u003e. However, the information value of any signal depreciates as receivers acquire private information through direct experience\u003csup\u003e[15]\u003c/sup\u003e. In initial consultations, patients may rely on price as a quality inference heuristic or as a cost barrier; in either case, the predictive validity of price for subsequent behavior diminishes as experience accumulation provides superior quality assessments. This information decay hypothesis posits that price loses its decisional weight\u0026mdash;not necessarily reversing direction, but attenuating in magnitude\u0026mdash;as patients transition from signal-dependent to experience-based evaluation.\u003c/p\u003e \u003cp\u003eH3: The functional role of price transmutes from a quality-inference heuristic to a cost barrier as patient-physician relationships mature, with this transmutation manifesting as attenuation for online-acquired patients and directional reversal for offline-acquired patients.\u003c/p\u003e \u003cp\u003eAs relationships mature, service-dominant logic predicts a shift from transactional value (information exchange) to relational value (process efficiency). In initial consultations, high information asymmetry surrounding diagnosis creates substantial demand for detailed information disclosure; patients rely on extensive physician responses to assess service quality and reduce uncertainty. As clinical baselines establish and diagnostic uncertainty reduces, repeat consultations shift focus from information acquisition to therapeutic coordination. The marginal value of information volume thus declines once clinical baselines are established, because patients transition from evaluating diagnostic competence to managing ongoing care. Conversely, response speed gains importance as a coordination mechanism for chronic care management, reflecting an asymmetric escalation in process quality attributes.\u003c/p\u003e \u003cp\u003eH4a: The influence of response speed on consultation volume will escalate across stages for offline-acquired patients, while attenuating for online-acquired patients.\u003c/p\u003e \u003cp\u003eH4b: The influence of information volume on consultation volume attenuates across consultation stages.\u003c/p\u003e \u003cp\u003eIn platform-mediated professional services, physician reputation operates as a digital gatekeeper that determines consideration-set inclusion rather than merely shifting demand continuously. Drawing on endogenous consideration-set theory\u003csup\u003e[22]\u003c/sup\u003e, we argue that reputation signals serve as heuristic screening mechanisms: patients facing information overload use aggregate ratings to exclude low-reputation physicians entirely, creating structural barriers at the extensive margin (whether a physician is considered at all). Recent evidence from online labor and healthcare markets (2019\u0026ndash;2023) confirms that platform ratings generate \"winner-take-all\" visibility dynamics, where sub-threshold providers face near-certain exclusion from patient choice sets\u003csup\u003e[23]\u003c/sup\u003e. Once a physician enters the consideration set, the decision calculus shifts from risk reduction to relational coordination. Here, process attributes\u0026mdash;response speed and information richness\u0026mdash;function as intensity drivers that shape consultation volume conditional on selection (the intensive margin). This hierarchical architecture, wherein aggregate signals govern entry and process attributes govern depth, aligns with recent platform economy research on two-stage consumer evaluation\u003csup\u003e[24]\u003c/sup\u003e. We therefore hypothesize:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eH5\u003c/strong\u003e \u003cp\u003ePhysician reputation primarily governs consideration-set formation (extensive margin), whereas process quality attributes drive consultation volume conditional on selection (intensive margin).\u003c/p\u003e \u003c/p\u003e"},{"header":"Data and Methodology","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eTemporal Research Design and Data Structure\u003c/h2\u003e \u003cp\u003eTo establish predictive temporal precedence while mitigating reverse causality concerns, we employ a time-lagged research design comprising three observation stages:\u003c/p\u003e \u003cp\u003eStage 1 (T0: Baseline Period, 31 May 2017). We extracted comprehensive physician profiles preceding any observed consultation behavior. These time-invariant or pre-determined attributes serve as exogenous predictors: physician demographic characteristics (professional title, hospital tier, city level), platform tenure (registration duration), historical reputation accumulation (site-wide recommendation score aggregated through 31 May 2017), and service pricing structure (minimum posted fees). Crucially, these T0 measurements temporally precede the dependent variables, ensuring that physician attributes predict rather than reflect concurrent demand fluctuations.\u003c/p\u003e \u003cp\u003eStage 2 (T1: First-Consultation Window, 1\u0026ndash;30 June 2017). We captured initial patient acquisition events, distinguishing between two distinct patient acquisition channels: (i) patients initiating contact via \"paid online consultation services\" without prior offline interaction (online-acquired), and (ii) patients transitioning from offline visits through the platform's \"post-visit reporting service\" (offline-acquired). These T1 outcomes represent relationship initiation, establishing the patient base available for subsequent tracking.\u003c/p\u003e \u003cp\u003eStage 3 (T2: Repeat-Consultation Window, 1 July\u0026ndash;31 December, 2017). We tracked sustained engagement behaviors among patients acquired during T1, measuring cumulative repeat consultations within a six-month observation period. This duration accommodates asthma's clinical cyclicity (acute exacerbation, chronic persistence, and remission phases) while capturing meaningful relationship continuity.\u003c/p\u003e \u003cp\u003eThis T0\u0026rarr;T1\u0026rarr;T2 sequencing enables us to model how pre-existing physician characteristics (measured before any observed behavior) predict both initial patient attraction and subsequent relationship sustainability, approximating predictive causal precedence without experimental manipulation.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSample and Variables\u003c/h3\u003e\n\u003cp\u003eThe analytic sample comprises 1,812 asthma specialists practicing across all 31 provincial-level jurisdictions of mainland China. The sample excludes physicians with missing profile data or zero platform tenure, and those without any consultation activity across the entire observation period. Our inclusion criteria require physicians to have complete T0 profile information and at least one measurable quality signal (Rating\u0026thinsp;\u0026ge;\u0026thinsp;3). Asthma constitutes an ideal empirical context: its chronic, relapsing-remitting nature necessitates iterative physician input and long-term medication management, making sustained doctor-patient relationships clinically essential rather than optional. We acknowledge potential selection constraints: Haodf.com physicians may not represent the full population of asthma specialists, and the 2017 data predates post-pandemic telemedicine expansion. However, the platform\u0026rsquo;s market position during the observation period and the national scope of our sample mitigate concerns about external validity for the studied context.\u003c/p\u003e \u003cp\u003eThe study examines four distinct consultation outcomes as dependent variables. Following the temporal structure, we operationalize four stage-specific dependent variables: OnFirstSale (T1), OffFirstSale (T1), OnRepeatSale (T2), and OffRepeatSale (T2), as defined in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Repeat consultation is thus operationalized as any additional paid online interaction within six months of the index (first) consultation, capturing relationship sustainability rather than discrete transactional volume.\u003c/p\u003e \u003cp\u003eIndependent Variables. All physician quality signals\u0026mdash;Rating, Price, ReplySpeed, and InfoAmount\u0026mdash;are measured at T0 baseline, ensuring temporal precedence in predictive relationships. To enable direct comparison of effect sizes across metrics with disparate scales, all independent variables are standardized using z-score transformation (mean\u0026thinsp;=\u0026thinsp;0, SD\u0026thinsp;=\u0026thinsp;1) prior to model estimation (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e for operational definitions and descriptive statistics).\u003c/p\u003e \u003cp\u003eControl Variables (T0-measured). Physician heterogeneity factors including Title, Hospital_3L, City, and Register are measured at baseline to control for institutional status, regional market conditions, and platform tenure effects (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eVariables Definition\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVariables Description\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOnFirstSale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCount of first-time online consultations by patients acquired through digital channels (no prior offline contact).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOnRepeatSale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCumulative repeat consultations within six months(T2) by online-acquired patients initially contacted during T1.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOffFirstSale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCount of first-time online consultations by patients transitioning from offline visits via post-visit reporting (established offline relationship migrating online).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOffRepeatSale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCumulative repeat consultations within six months by offline-acquired patients who first connected online during T1.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRating\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePhysician's platform-generated recommendation heat score (continuous, 1\u0026ndash;5 scale), representing cumulative historical reputation preceding the observation window.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrice\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe minimum service price across consultation modalities (text, voice).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReplySpeed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAverage response latency (days) between patient inquiry and physician reply, calculated from the 100 most recent consultations preceding the baseline observation (T0).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInfoAmount\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean word count per physician response, operationalizing information richness (calculated from 100 most recent interactions).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTitle\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eProfessional hierarchy (Chief Physician; Associate Chief Physician; Attending/Resident ).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHospital_3L\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBinary indicator for Tier-3A hospital affiliation (highest accreditation tier).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRegister\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePlatform tenure (years since registration as of 31 May 2017).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUrban hierarchy ranking (0\u0026thinsp;=\u0026thinsp;most developed, 5\u0026thinsp;=\u0026thinsp;least developed).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eAnalytical Strategy: Zero-Inflated Negative Binomial Regression (ZINB)\u003c/h2\u003e \u003cp\u003e \u003cb\u003eModel Selection.\u003c/b\u003e The dependent variables exhibit two statistical properties necessitating specialized handling. First, zero-inflation: a substantial proportion of physicians record zero consultations in specific categories, reflecting two distinct generative processes\u0026mdash;(i) structural zeros, where physicians are excluded entirely from patient consideration sets due to low visibility or reputation (addressing RQ3 regarding consideration set thresholds), and (ii) random zeros, where patients evaluate but decline specific physicians\u003csup\u003e[25]\u003c/sup\u003e. Second, over-dispersion: variance substantially exceeds mean for all outcome variables, violating Poisson distribution assumptions. We employ Zero-Inflated Negative Binomial (ZINB) regression, which simultaneously models (a) the probability of structural zero occurrence via a logit component, and (b) consultation volume conditional on non-zero occurrence via a negative binomial count component.\u003c/p\u003e \u003cp\u003e \u003cb\u003eEstimation Strategy.\u003c/b\u003e To examine how cue effectiveness evolves across patient experience stages and acquisition channels, we implement a comprehensive analytical framework integrating stratified estimation, coefficient comparison tests, and interaction verification. Separate zero-inflated negative binomial (ZINB) models are estimated for each of the four outcome categories (OnFirstSale, OffFirstSale, OnRepeatSale, OffRepeatSale), enabling direct coefficient comparison across decision contexts. We then employ Seemingly Unrelated Estimation (SUEST) post-estimation procedures to statistically test whether reputation coefficients differ significantly between first-time versus repeat consultations and between online versus offline channels\u003csup\u003e[26]\u003c/sup\u003e. Pooled models with interaction terms (e.g., Rating \u0026times; FirstPurchase, Rating \u0026times; OnlineChannel) following reshape to long-format data, controlling for unobserved physician heterogeneity through clustering\u003csup\u003e[27]\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eDescriptive Statistics\u003c/h3\u003e\n\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents descriptive statistics for the full sample of 1,812 asthma specialists. The four consultation outcomes exhibit substantial dispersion, with standard deviations markedly exceeding means (e.g., OnFirstSale: mean\u0026thinsp;=\u0026thinsp;12.16, SD\u0026thinsp;=\u0026thinsp;42.73; OnRepeatSale: mean\u0026thinsp;=\u0026thinsp;0.37, SD\u0026thinsp;=\u0026thinsp;1.31), indicating right-skewed distributions with heavy tails. Notably, zero-inflation is pronounced in repeat consultation measures: 82.6% of physicians record zero online repeat sales (OnRepeatSale) and 77.8% record zero offline repeat sales (OffRepeatSale), while even first-time consultation variables exhibit substantial zero mass (OnFirstSale: 41.7% zeros; OffFirstSale: 43.7% zeros).\u003c/p\u003e \u003cp\u003eAmong physician quality signals, Rating concentrates at the upper bound (mean\u0026thinsp;=\u0026thinsp;4.00 on a 3\u0026ndash;5 scale, SD\u0026thinsp;=\u0026thinsp;0.35), suggesting ceiling effects in platform reputation systems. Price exhibits wide heterogeneity (mean=\u0026yen;27.13, SD\u0026thinsp;=\u0026thinsp;41.60, range=\u0026yen;5\u0026ndash;\u0026yen;900), reflecting diverse positioning strategies from accessible to premium services. ReplySpeed averages 0.34 days (SD\u0026thinsp;=\u0026thinsp;0.37), indicating generally rapid physician responsiveness, while InfoAmount exhibits substantial variation (mean\u0026thinsp;=\u0026thinsp;36.39 words, SD\u0026thinsp;=\u0026thinsp;35.28). Control variables indicate that 35.4% hold Chief Physician titles, 34.9% hold Associate Chief titles, and 91.4% practice in Tier-3A hospitals, with platform tenure averaging 3.35 years (SD\u0026thinsp;=\u0026thinsp;2.79).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDescriptive Statistics of Variables\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMin\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMax\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eOnFirstSale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e12.156\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e42.729\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e593\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eOffFirstSale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16.793\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e36.282\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e370\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eOnRepeatSale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.311\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eOffRepeatSale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.314\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e41\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eRating\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.352\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eprice\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e27.134\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e41.596\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e900\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003elnInfoAmount\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36.391\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35.277\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.216\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e447.726\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eReplySpeed\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.339\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.367\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eChief\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.354\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.478\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eAssociateChief\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.349\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.477\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eHospital3Level\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.914\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eCity\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.666\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eNote: While Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e reports raw values for descriptive transparency, all continuous independent variables (Rating, Price, ReplySpeed, InfoAmount) are standardized (z-scored) in subsequent regression analyses to facilitate coefficient comparability and mitigate multicollinearity in interaction terms. The dependent variables remain in original count scale to preserve the zero-inflated, over-dispersed properties that motivate ZINB estimation.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eThe Main Results\u003c/h2\u003e \u003cp\u003eWe report ZINB estimates, which simultaneously model (a) the probability of structural zero occurrence (inflation component) and (b) consultation volume conditional on non-zero occurrence (count component). Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents ZINB estimates for the four stage-specific outcomes. Model 1 estimates the determinants of OnFirstSale (initial online consultations by online-acquired patients); Model 2 examines OffFirstSale (initial online consultations by patients transitioning from offline visits); Model 3 assesses OnRepeatSale (sustained online consultations by online-acquired patients within six months); and Model 4 corresponds to OffRepeatSale (sustained online consultations by offline-acquired patients).\u003c/p\u003e \u003cp\u003eConsistent with the signal verification hypothesis, physician reputation (Rating) exhibits strong and persistent predictive power across all stages (β\u0026thinsp;=\u0026thinsp;0.354\u0026ndash;0.587, all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Notably, for online-acquired patients, the reputation coefficient for repeat consultations (β\u0026thinsp;=\u0026thinsp;0.430, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) is comparable to or slightly exceeds that for first-time consultations (β\u0026thinsp;=\u0026thinsp;0.354, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), suggesting that reputation signals retain or potentially increase salience as patient-physician relationships mature\u0026mdash;a pattern consistent with trust verification rather than signal decay. For offline-acquired patients, reputation remains robustly significant across both stages (β\u0026thinsp;=\u0026thinsp;0.587, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01 and 0.525, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), indicating that established offline trust provides a durable foundation for online engagement.\u003c/p\u003e \u003cp\u003ePrice exhibits consistent negative effects on consultation volume, indicating straightforward price sensitivity or cost barrier effects in online healthcare. For online-acquired patients, higher prices significantly deter first-time consultations (β=-0.108, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Model 1), reflecting risk-aversion when quality is uncertain; this sensitivity dissipates for repeat consultations (β\u0026thinsp;=\u0026thinsp;0.081, ns, Model 3) as patients establish quality assessments through direct experience. Strikingly, for offline-acquired patients, price turns strongly negative for repeat consultations (β=-0.229, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Model 4), suggesting that once offline trust is established, excessive pricing becomes a retention barrier\u0026mdash;patients expect continued value from known physicians and are discouraged by high fees despite existing relationships.\u003c/p\u003e \u003cp\u003eProcess quality signals exhibit differential patterns by channel and stage. Response speed (ReplySpeed) exerts consistent negative effects across all specifications (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), indicating that faster response uniformly predicts higher consultation volume. Notably, the absolute magnitude of this effect varies: for online-acquired patients, the coefficient is strongest at the first consultation (β=-0.498, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), suggesting immediate efficiency expectations among platform-native users; the effect persists for repeat consultations (β=-0.310, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), though with somewhat reduced sensitivity. For offline-acquired patients, the speed effect intensifies from the initial (β=-0.228, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) to the repeat consultation stage (β=-0.449, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), suggesting that response efficiency becomes increasingly salient as these patients establish sustained online engagement patterns\u0026mdash;potentially reflecting heightened expectations for digital convenience once offline familiarity is established.\u003c/p\u003e \u003cp\u003eInformation provision volume (InfoAmount) significantly predicts first-time consultations for both channels (β\u0026thinsp;=\u0026thinsp;0.190, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01 for online, 0.112, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01 for offline), suggesting that information richness facilitates patient decision-making under uncertainty. However, this effect diminishes for online repeat consultations (β\u0026thinsp;=\u0026thinsp;0.138, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) and becomes non-significant for offline repeat consultations (β\u0026thinsp;=\u0026thinsp;0.092, ns), consistent with the view that established relationships rely more on efficiency (speed) than on extensive information provision.\u003c/p\u003e \u003cp\u003eOffline-acquired patients exhibit stronger reliance on reputation signals during initial online contact compared to online-acquired patients (OffFirstSale Rating: β\u0026thinsp;=\u0026thinsp;0.587, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01 vs. OnFirstSale Rating: β\u0026thinsp;=\u0026thinsp;0.354, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), supporting the offline-to-online trust transfer hypothesis. These patients bring established face-to-face trust into the digital realm, seeking validation through high-reputation physicians. Conversely, online-acquired patients\u0026mdash;lacking prior direct experience\u0026mdash;distribute their attention across multiple cues, resulting in relatively lower marginal effects for individual signals.\u003c/p\u003e \u003cp\u003eGeographic preferences diverge sharply by channel. City tier (City) exhibits no significant influence on either first-time or repeat consultations for online-acquired patients (β\u0026thinsp;=\u0026thinsp;0.011 and \u0026minus;\u0026thinsp;0.032, both ns), suggesting that platform-mediated quality signals effectively supersede geographic proximity for online-native patients. In stark contrast, offline-acquired patients consistently prefer physicians in lower-tier cities (higher City values) for both initial (β\u0026thinsp;=\u0026thinsp;0.197, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Model 2) and repeat consultations (β\u0026thinsp;=\u0026thinsp;0.182, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Model 4), likely reflecting geographic proximity to their original offline treatment locations.\u003c/p\u003e \u003cp\u003eThe inflation component reveals reputation as a structural barrier to patient consideration. Across all four models, Rating carries a negative and highly significant coefficient in the zero-inflation equation (ranging from \u0026minus;\u0026thinsp;3.698, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01 to -5.973, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), indicating that lower-reputation physicians face substantially elevated probabilities of being excluded entirely from patient choice sets (structural zeros) rather than merely being evaluated and rejected. This confirms RQ3: reputation operates not only as a volume determinant but as a threshold mechanism that determines whether physicians enter patient consideration sets at all.\u003c/p\u003e \u003cp\u003eProfessional titles (Chief, AssociateChief) significantly reduce the probability of structural zeros and increase expected volume conditional on non-zero outcomes. The effects of senior titles are generally stronger for repeat consultations (Chief β\u0026thinsp;=\u0026thinsp;0.655, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Model 3; Chief β\u0026thinsp;=\u0026thinsp;0.817, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, Model 4) than for first-time consultations (Chief β\u0026thinsp;=\u0026thinsp;0.215, p\u0026thinsp;\u0026lt;\u0026thinsp;0.1, Model 1), suggesting that institutional credentials become more salient as patients deepen relationships and seek assurance for sustained engagements. SUEST coefficient comparison tests confirm that the difference in Chief Physician effects between repeat and first-time consultations is statistically significant (χ\u0026sup2; = 7.83, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), supporting the interpretation that institutional credentials gain salience as relationships mature.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMaximum Likelihood Parameter Estimates\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eModel 1\u003c/p\u003e \u003cp\u003e(OnFirstSale)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eModel 2\u003c/p\u003e \u003cp\u003e(OffFirstSale)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eModel 3\u003c/p\u003e \u003cp\u003e(OnRepeatSale)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eModel 4\u003c/p\u003e \u003cp\u003e(OffRepeatSale)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eRating_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.354\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.053)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.587\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.052)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.430\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.093)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.525\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.092)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003ePrice_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.108\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.045)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.052\u003c/p\u003e \u003cp\u003e(0.055)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.081\u003c/p\u003e \u003cp\u003e(0.065)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.229\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.088)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eInfoAmount_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.190\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.050)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.112\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.038)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.138\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.068)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.092\u003c/p\u003e \u003cp\u003e(0.056)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eReplySpeed_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.498\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.044)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.228\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.037)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.310\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.080)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.449\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.079)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eChief\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.215\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.130)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.522\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.111)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.655\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.189)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.817\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.178)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eAssociate\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.263\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.125)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.443\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.110)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.331\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.189)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.576\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.180)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eHospital3Level\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.443\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.205)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.128\u003c/p\u003e \u003cp\u003e(0.186)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.228\u003c/p\u003e \u003cp\u003e(0.351)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.376\u003c/p\u003e \u003cp\u003e(0.322)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eCity\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003cp\u003e(0.041)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.197\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.038)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.032\u003c/p\u003e \u003cp\u003e(0.059)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.182\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.059)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e_cons\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.270\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(1.057)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.896\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.881)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.139\u003c/p\u003e \u003cp\u003e(1.517)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.819\u003c/p\u003e \u003cp\u003e(1.370)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e/lnalpha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ealpha\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.867\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.685\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.176\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.455\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eZero inflation coef.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eRating\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-5.531\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.505)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-3.698\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.292)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-4.138\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.917)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-5.973\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(1.099)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003e_cons\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20.211\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(1.875)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14.236\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(1.125)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e16.131\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(3.429)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e23.449\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(4.129)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eModel summary\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eobs\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1812\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eNonzero obs\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1057\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e946\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e316\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e396\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eZero obs\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e755\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e866\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1496\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1416\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eLR chi2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e163.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e231.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e80.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e108.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eLog likelihood\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-4882.327\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-5055.638\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-1150.085\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-1526.670\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eNotes: * p\u0026thinsp;\u0026lt;\u0026thinsp;0.1, ** p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, *** p\u0026thinsp;\u0026lt;\u0026thinsp;0.01; Standard errors are in parentheses.\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eModerating Effects of ZINB Regression Results by Subgroups\u003c/h2\u003e \u003cp\u003eWe now examine how the effectiveness of quality signals varies across purchase contexts and channels. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents the results of Zero-Inflated Negative Binomial (ZINB) regression models examining the moderating effects of purchase type and channel choice on service sales. The analysis employs a subgroup design: Models 1 and 2 test the moderating effect of first-time purchase (vs. repeat purchase) within online and offline channels respectively; Models 3 and 4 test the moderating effect of online channel (vs. offline) within first-time and repeat purchase samples, respectively. Each model includes main effects of physician rating (Rating_std), service price (price_std), information amount (lnInfoAmount_std), response speed (ReplySpeed_std), and physician/hospital characteristics. The interaction terms capture how the effectiveness of physician reputation and service attributes varies across purchase contexts. The inflation part models the excess zero counts using physician rating as the inflation predictor.\u003c/p\u003e \u003cp\u003eConsistent with the signal verification framework, physician reputation (Rating_std) exhibits strong and statistically significant positive effects on consultation volume across all four subgroups (β\u0026thinsp;=\u0026thinsp;0.467\u0026ndash;0.597, all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). This persistence indicates that platform-generated reputation signals retain predictive power regardless of whether patients are acquired through digital channels or offline referrals, and whether they are engaging in first-time or repeat consultations. Notably, the point estimates suggest relatively higher reputation sensitivity in offline-acquired repeat consultations (β\u0026thinsp;=\u0026thinsp;0.597, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) and first-time consultations overall (β\u0026thinsp;=\u0026thinsp;0.563, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), compared to online-acquired repeat consultations (β\u0026thinsp;=\u0026thinsp;0.467, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Process quality signals demonstrate differential baselines: response speed (ReplySpeed_std) consistently predicts higher volume across all contexts (β=-0.233 to -0.453, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), while information amount (InfoAmount_std) shows modest positive effects (β\u0026thinsp;=\u0026thinsp;0.086\u0026ndash;0.121, p\u0026thinsp;\u0026lt;\u0026thinsp;0.1 or p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) except in the offline channel where significance is marginal.\u003c/p\u003e \u003cp\u003eThe interaction terms reveal significant heterogeneity in signal effectiveness between first-time and repeat consultations, particularly within the online channel. In Model 1 (Online Channel), the negative and significant interaction between First Purchase and Rating_std (β=-0.290, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) indicates that the reputation effect is substantially attenuated for first-time consultations (implied effect\u0026thinsp;\u0026asymp;\u0026thinsp;0.307, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) relative to repeat consultations (baseline β\u0026thinsp;=\u0026thinsp;0.597, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). This pattern is consistent with trust verification mechanisms wherein validated signals gain salience post-experience. Conversely, in Model 2 (Offline Channel), the interaction term is statistically indistinguishable from zero (β=-0.076, n.s.), indicating that offline-acquired patients maintain stable reputation responsiveness across both purchase stages.\u003c/p\u003e \u003cp\u003ePrice signaling exhibits contrasting stage dynamics by channel. In the online channel, the positive price effect observed in repeat consultations (baseline β\u0026thinsp;=\u0026thinsp;0.205, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) dissipates for first-time purchases, as evidenced by the negative interaction term (β=-0.212, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), resulting in a near-zero net effect for initial consultations. This suggests that online-acquired patients may interpret price as a quality signal only after establishing initial contact. In contrast, offline-acquired patients show no significant price sensitivity in repeat consultations (baseline β\u0026thinsp;=\u0026thinsp;0.026, n.s.), but display a positive price effect for first-time consultations (implied effect\u0026thinsp;=\u0026thinsp;0.026\u0026thinsp;+\u0026thinsp;0.181\u0026thinsp;=\u0026thinsp;0.207, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), consistent with price serving as a screening device when offline trust has not yet been established.\u003c/p\u003e \u003cp\u003eResponse speed priorities also diverge by stage and channel. In the online channel, the negative interaction between First Purchase and ReplySpeed_std (β=-0.195, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) indicates that first-time patients exhibit heightened sensitivity to response latency (implied effect \u0026asymp; -0.501) compared to repeat patients (β=-0.306, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). However, in the offline channel, the pattern reverses: the positive interaction term (β\u0026thinsp;=\u0026thinsp;0.20, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) suggests that repeat consultations demonstrate stronger speed sensitivity (baseline β=-0.442, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) than first-time consultations (implied effect \u0026asymp; -0.235).\u003c/p\u003e \u003cp\u003e \u003cem\u003eChannel Differences within Purchase Stages (Models 3\u0026ndash;4).\u003c/em\u003e When examining channel effects within purchase stages, Model 3 (First Purchase) reveals that the online channel significantly dampens reputation effectiveness relative to offline channels (interaction β=-0.245, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), reducing the implied reputation effect for online first-time consultations to approximately 0.318 (compared to 0.563 for offline). This attenuation may reflect greater information diversification among online-acquired patients. For repeat consultations (Model 4), the channel interaction for reputation is negative but statistically insignificant (β=-0.090, n.s.), suggesting comparable reputation sensitivity across channels at the repeat stage.\u003c/p\u003e \u003cp\u003ePrice sensitivity in first-time consultations shows significant channel variation: the negative interaction between Online Channel and ln_price (β=-0.168, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) superimposed on the positive offline baseline (β\u0026thinsp;=\u0026thinsp;0.191, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) results in a null net effect for online first-time purchases, while preserving the positive offline effect. Interestingly, for repeat consultations, the channel interaction for price is positive and significant (β\u0026thinsp;=\u0026thinsp;0.259), transforming the offline null effect (β=-0.063, n.s.) into a positive online effect (implied\u0026thinsp;\u0026asymp;\u0026thinsp;0.196), suggesting that price-quality inferences become salient for online-acquired patients only after initial experience accumulation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eZero-Inflation Component: Extensive Margin Evidence\u003c/h2\u003e \u003cp\u003eThe inflation part of the ZINB models provides complementary evidence regarding the extensive margin of patient choice. Physician rating exhibits large negative coefficients across all specifications (β=-4.084 to -5.518, all p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), indicating that higher reputation significantly reduces the probability of zero consultations (i.e., the likelihood that a physician records no sales in a given period). This finding reinforces the main effects by demonstrating that reputation signals not only increase the volume of consultations conditional on purchase, but also expand the probability of any purchase occurring.\u003c/p\u003e \u003cp\u003eThe likelihood ratio tests (LR chi2) indicate that all four models provide statistically significant improvement over null models, with substantial log-likelihood values ranging from \u0026minus;\u0026thinsp;2686.475 to -9971.521. The dispersion parameters (alpha) range from 1.924 to 2.865, confirming overdispersion in the count data and validating the selection of ZINB over standard Poisson specifications.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModerating Effects of ZINB Regression Results by Subgroups\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOnline Channel\u003c/p\u003e \u003cp\u003e(Online\u0026thinsp;=\u0026thinsp;1)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOffline Channel\u003c/p\u003e \u003cp\u003e(Online\u0026thinsp;=\u0026thinsp;0)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFirst Purchase\u003c/p\u003e \u003cp\u003e(First\u0026thinsp;=\u0026thinsp;1)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRepeat Purchase\u003c/p\u003e \u003cp\u003e(First\u0026thinsp;=\u0026thinsp;0)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eMain Effects (Baseline Group)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eRating_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.597\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.075)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.597\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.073)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.563\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.059)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.467\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.087)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eprice_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.205\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.073)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.026\u003c/p\u003e \u003cp\u003e(0.073)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.191\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.060)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.063\u003c/p\u003e \u003cp\u003e(0.082)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eInfoAmount_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.121\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.066)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.086\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.049)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.103\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.043)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.094\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.055)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eReplySpeed_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.306\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.080)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.442\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.073)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.233\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.042)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.453\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.078)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eChief\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.321\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.108)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.566\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.096)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.332\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.089)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.699\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.133)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eAssociate\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.245\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.104)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.447\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.094)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.305\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.087)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.422\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.133)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eHospital3Level\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.293\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.167)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.221\u003c/p\u003e \u003cp\u003e(0.158)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.164\u003c/p\u003e \u003cp\u003e(0.144)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.301\u003c/p\u003e \u003cp\u003e(0.238)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eCity\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.021\u003c/p\u003e \u003cp\u003e(0.034)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.237\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.033)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.140\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.030)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.115\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.044)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e_cons\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.522\u003c/p\u003e \u003cp\u003e(0.865)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-1.337\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.749)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.448\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.707)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.322\u003c/p\u003e \u003cp\u003e(1.022)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eChannel/Purchase Type Main Effects\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eFirst/Online\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.750\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.091)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.190\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.085)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.145\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.071)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.635\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.113)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eInteraction Effects (Moderating Effects)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eFirst/Online \u0026times; Rating_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.290\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.089)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.076\u003c/p\u003e \u003cp\u003e(0.088)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.245\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.075)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.090\u003c/p\u003e \u003cp\u003e(0.106)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eFirst/Online \u0026times; price_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.212\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.091)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.181\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.088)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.168\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.080)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.259\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.105)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eFirst/Online \u0026times; InfoAmount_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.054\u003c/p\u003e \u003cp\u003e(0.082)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.007\u003c/p\u003e \u003cp\u003e(0.063)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.057\u003c/p\u003e \u003cp\u003e(0.064)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.023\u003c/p\u003e \u003cp\u003e(0.088)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eFirst/Online \u0026times; ReplySpeed_std\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.195\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.091)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.207\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.083)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.255\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.059)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.159\u003c/p\u003e \u003cp\u003e(0.111)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eInflation Part (Zero-Inflation Model)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eRating\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-5.518\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.481)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-4.084\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.311)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-4.733\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.302)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-5.019\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.667)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e_cons\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20.209\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(1.783)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15.744\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(1.186)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17.638\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(1.133)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e19.738\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(2.501)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eDispersion Parameters\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e/lnalpha\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.053\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.048)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.655\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.065)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.909\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.044)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.848\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e \u003cp\u003e(0.110)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ealpha\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.924\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.481\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.335\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eModel Summary Statistics\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eObservations\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3624\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3624\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3624\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3624\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eNonzero obs\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1373\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1342\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2003\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e712\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eZero obs\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2251\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2282\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1621\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2912\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eLR chi2\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1369.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1307.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e403.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e217.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eLog likelihood\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-6042.059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-6596.683\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-9971.521\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-2686.475\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eNotes: * p\u0026thinsp;\u0026lt;\u0026thinsp;0.1, ** p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, *** p\u0026thinsp;\u0026lt;\u0026thinsp;0.01; Standard errors are in parentheses.\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eModels 1\u0026ndash;2: Test the moderating effect of First Purchase (First) in online/offline channels;\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eModels 3\u0026ndash;4: Test the moderating effect of Online Channel (Online) in first/repeat purchases.\u003c/em\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eTheoretical Implications: Dual-Signal Architecture and Channel Heterogeneity\u003c/h2\u003e \u003cp\u003eOur findings illuminate a dual-stage decision architecture in platform-mediated healthcare that extends conventional signaling theory. Consistent with H1, physician reputation and process quality operate as functionally distinct rather than competing mechanisms: reputation serves as a threshold gatekeeper determining consideration-set inclusion (evidenced by significant negative coefficients in the zero-inflation component), while response speed and information depth function as intensity drivers shaping volume conditional on physician entry. This complementarity suggests that in high-stakes credence services, patients rely on hierarchical evaluation schemas wherein platform-aggregated historical signals reduce search risk prior to engagement, and process attributes fine-tune ongoing relationship investment.\u003c/p\u003e \u003cp\u003eOur findings extend recent evidence on digital healthcare quality assessment. Gong et al.\u003csup\u003e[9]\u003c/sup\u003e demonstrate comparable diagnostic quality between asynchronous webchats and in-person consultations for postpartum depression, suggesting that process quality attributes can substitute for physical presence in certain care contexts. Our results complement this by showing that response efficiency, rather than information depth, becomes the dominant process driver in sustained relationships\u0026mdash;potentially because established patients have already resolved initial diagnostic uncertainty. This alignment between cross-sectional quality comparisons and our longitudinal signal evolution patterns strengthens the case for platform-mediated chronic disease management.\u003c/p\u003e \u003cp\u003eH2 reveals that this dual-signal architecture is contingent upon patient knowledge endowments at platform entry. Offline-acquired patients\u0026mdash;those transitioning from face-to-face encounters\u0026mdash;exhibit stable high sensitivity to reputation signals and geographic continuity-seeking preferences. This pattern is consistent with trust transfer theory\u003csup\u003e[21]\u003c/sup\u003e: imported offline trust may serve as cognitive anchors that stabilize platform signal reliance, rendering reputation a confirmatory device rather than an exploratory heuristic. Conversely, online-acquired patients demonstrate spatial substitution effects, exhibiting reduced geographic constraints as platform-mediated quality signals effectively decouple their choices from physical proximity. They display constructive signal reliance, characterized by lower initial reputation sensitivity and heightened responsiveness to process quality, reflecting active knowledge construction in the absence of offline anchors.\u003c/p\u003e \u003cp\u003eThe geographic divergence we observe further nuances the digital disruption narrative. Online-acquired patients exhibit spatial substitution\u0026mdash;null or negative associations with city tier\u0026mdash;as platform signals effectively decouple quality assessment from physical proximity. In contrast, offline-acquired patients demonstrate geographic continuity-seeking, preferring physicians in lower-tier cities proximal to original treatment locations. This suggests that for relationship-maintenance scenarios, online platforms serve complementary infrastructure extending offline therapeutic relationships, rather than pure market mechanisms optimizing for quality independent of geography.\u003c/p\u003e \u003cp\u003eWhile our confirmatory hypotheses address static signal-channel relationships, exploratory analyses of stage-wise variation reveal temporal dynamics that extend our theoretical framework. For online-acquired patients, reputation effects maintain or intensify across consultation stages, contradicting conventional signal decay predictions. This is consistent with a verification mechanism: initial consultations function as calibration opportunities where patients assess platform signal accuracy. When observed quality aligns with reputation indicators, cognitive consonance \u003csup\u003e[11]\u003c/sup\u003e reinforces trust in the platform's signaling infrastructure, creating self-reinforcing reliance cycles. This challenges the universality of signal decay and suggests that in contexts of persistent uncertainty, validated platform signals achieve institutional trust status that deepens with experience.\u003c/p\u003e \u003cp\u003ePrice signals exhibit functional transmutation across relationship stages. For online-acquired patients, price operates as a quality inference heuristic (non-negative initial associations) that attenuates or reverses as experience accumulates, transitioning to pure cost consideration. This aligns with means-end chain theory\u003csup\u003e[28]\u003c/sup\u003e, price shifts from diagnostic attribute to sacrifice attribute as private quality knowledge develops. Offline-acquired patients, possessing pre-validated quality assessments, demonstrate immediate cost sensitivity across stages, suggesting that trust transfer accelerates price function maturation.\u003c/p\u003e \u003cp\u003eProcess quality attributes demonstrate asymmetric escalation. Response speed gains increasing salience across stages\u0026mdash;particularly for offline-acquired patients\u0026mdash;supporting service-dominant logic\u003csup\u003e[18]\u003c/sup\u003e: as relationships mature from diagnostic to maintenance phases, coordination efficiency supersedes information acquisition as the primary value driver. Conversely, information volume exhibits satiation effects, particularly for online-acquired patients, suggesting that detailed explanations facilitate initial decision-making but yield diminishing returns once clinical baselines are established.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003ePractical Implications\u003c/h2\u003e \u003cp\u003ePlatform Design. Interfaces should implement adaptive signal architectures based on patient origin. For online acquisition, prioritize process quality transparency (response time guarantees, detailed service descriptions) to compensate for absent offline trust. For offline acquisition, emphasize geographic proximity indicators linking online physicians to original treatment locations, and prioritize reputation reinforcement over initial trust establishment. Implementing adaptive interfaces requires careful navigation of privacy constraints\u0026mdash;channel identification may inadvertently reveal patient medical history\u0026mdash;and physician workflow considerations, as response time guarantees necessitate compensation mechanisms or workload management tools. The cost-effectiveness of such segmentation should be evaluated against uniform interface designs.\u003c/p\u003e \u003cp\u003ePhysician Strategy. Practitioners must optimize both threshold and intensity mechanisms. Investment in platform reputation (review solicitation, recommendation score maintenance) functions as market entry tickets, while process efficiency (rapid response protocols) drives relationship sustainability. Physicians serving offline-acquired patients should prioritize responsiveness for retention, as this segment exhibits steeply escalating speed sensitivity. Those attracting online-acquired patients should initially emphasize information richness to facilitate risk reduction under uncertainty.\u003c/p\u003e \u003cp\u003eHealthcare Policy. Digital platforms can function as spatial equalization mechanisms for online-acquired patients, enabling access to high-reputation specialists regardless of geographic hierarchy. However, the continuity-seeking behavior of offline-acquired patients indicates that local healthcare infrastructure remains critical for relationship initiation. Policymakers should view online platforms as complementary extensions of traditional care networks rather than substitute solutions, investing in hybrid care pathways that leverage digital efficiency while maintaining offline diagnostic capacity.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eLimitations and Future Research\u003c/h2\u003e \u003cp\u003eSeveral limitations warrant caution in interpreting our findings. First, data from 2017 may not fully capture post-pandemic telemedicine adoption patterns, where patient digital literacy and regulatory environments have shifted substantially. Second, as Haodf.com operates recommendation algorithms that may prioritize certain physicians, platform governance effects may confound pure signal interpretations. Third, our physician-level aggregation precludes observation of patient-level heterogeneity in disease severity and health literacy.\u003c/p\u003e \u003cp\u003eOur physician-level aggregated data precludes observation of patient-level heterogeneity (disease severity, health literacy) and cross-physician learning effects. While temporal sequencing establishes precedence, we cannot fully exclude selection effects wherein unobserved physician characteristics simultaneously influence signal profiles and patient attraction. Future research employing patient-level panel data could disentangle self-selection from true learning effects.\u003c/p\u003e \u003cp\u003eOur findings suggest signal verification as a potential boundary condition to the signal decay paradigm, particularly in high-stakes credence services where diagnostic uncertainty persists despite accumulated interactions. We caution that this pattern\u0026rsquo;s generalizability may be limited to similar professional service contexts. The signal verification mechanism we identify may operate more strongly in contexts of persistent outcome uncertainty (chronic diseases, complex diagnostics). Future research should test boundary conditions across acute versus chronic conditions, insurance-based versus out-of-pocket payment models, and cultural contexts varying in platform trust.\u003c/p\u003e \u003cp\u003eFinally, our binary channel classification (online/offline) could be extended to hybrid trajectories\u0026mdash;patients oscillating between digital and physical channels. Understanding temporal dynamics of trust transfer across these hybrid paths represents a critical avenue for research on healthcare system integration.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study advances platform-mediated service theory by demonstrating that in professional service contexts, institutional trust signals and experiential learning operate as complementary governance mechanisms. Contrary to substitution hypotheses, our findings reveal that platform-verified quality cues gain predictive salience as patient-physician relationships mature, suggesting that patients use direct experience to validate rather than replace institutional signals. The identified acquisition channel boundary condition further indicates that online channels facilitate trust construction through process transparency, while offline channels require sustained relational continuity to maintain transferred trust\u0026mdash;offering a parsimonious framework applicable to legal, financial, and educational platforms.\u003c/p\u003e \u003cp\u003eThese insights carry direct implications for dual-signal platform architecture: reputation investments should target consideration-set formation (search stage), while process quality investments should focus on relationship intensity (retention stage). Resource allocation must further differentiate by patient origin\u0026mdash;prioritizing transparency cues for online-acquired users and geographic continuity for offline-acquired cohorts. For policymakers, the evidence suggests that digital health platforms reduce spatial barriers to specialist access but cannot substitute local infrastructure for relationship initiation; regulatory frameworks should thus incentivize platform-care network integration rather than standalone digital provision.\u003c/p\u003e \u003cp\u003eFuture research should examine hybrid care models combining synchronous telemedicine access with asynchronous follow-up, while addressing self-selection limitations through experimental designs that causally identify channel effects.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u0026nbsp;This study was reviewed and granted an exemption by the Academic Committee of Inner Mongolia University. The research employs retrospective, de-identified administrative data obtained from Haodf.com under a data use agreement. All patient-identifying information was irreversibly removed by the platform prior to data provision; researchers accessed only physician-level aggregates and anonymized interaction metadata (response timestamps and word counts) with no ability to re-identify individual patients. The study involves no human subjects interaction, no intervention, and no collection of identifiable private information. All analyses adhered to the Personal Information Protection Law of the People\u0026apos;s Republic of China and Haodf.com\u0026apos;s data privacy policies.\u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting interests:\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eNatural Science Foundation of Inner Mongolia (2022QN07006).\u003c/p\u003e\u003ch2\u003eAcknowledgments:\u003c/h2\u003e \u003cp\u003eThe authors gratefully acknowledge Haodf.com for providing the data used in this study. This research was supported by the Inner Mongolia Natural Science Foundation (2022QN07006).\u003c/p\u003e \u003cp\u003e \u003cb\u003eEthical Considerations\u003c/b\u003e: This study employed retrospective, aggregated administrative data obtained from Haodf.com via a data use agreement. The research constitutes a secondary analysis of pre-existing commercial platform data and does not involve human subjects research as defined by the Common Rule. Given that (1) data were analyzed at the physician level with no individual patient identifiers, (2) the study was retrospective with no intervention or interaction with participants, and (3) data access was authorized by the platform under commercial research collaboration terms, formal IRB review was not required. All analyses adhered to the Personal Information Protection Law of the People's Republic of China and Haodf.com's data privacy policies.\u003c/p\u003e\u003ch2\u003eData Availability:\u003c/h2\u003e \u003cp\u003eThe data used in this study are proprietary and were obtained from Haodf.com under a data use agreement. Due to commercial confidentiality and user privacy agreements, the data are not publicly available. Requests for data access should be directed to Haodf.com or the corresponding author.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eYach D, Hawkes C, Gould CL, Hofman KJ. The global burden of chronic diseases: overcoming impediments to prevention and control [J]. Journal of the American Medical Association, 2004, 291(21): 2616-2622.\u003c/li\u003e\n\u003cli\u003eYu M, He S, Wu D, Zhu H, Webster C. Examining the Multi-Scalar Unevenness of High-Quality Healthcare Resources Distribution in China [J]. International Journal of Environmental Research and Public Health, 2019, 16(16).\u003c/li\u003e\n\u003cli\u003eSun S, Li T, Zheng A, Zhang Z, Wang Q, Chen C, Zeng Z. Doctor-patient-family collaboration in community-based chronic disease management to enhance multidimensional value [J]. Patient Education and Counseling, 2025, 132.\u003c/li\u003e\n\u003cli\u003eShapoo N, Shapoo N, Rehman A, Boma N. Smart Healthcare: The Role of Digital Health in Modern Medicine [J]. Health Care Science, 2025, 4(3): 179-187.\u003c/li\u003e\n\u003cli\u003eBuchanan J, Roope LSJ, Morrell L, Pouwels KB, Robotham JV, Abel L, Crook DW, Peto T, Butler CC, Walker AS, Wordsworth S. Preferences for medical consultations from online providers: evidence from a discrete choice experiment in the United Kingdom [J]. Applied Health Economics and Health Policy, 2021, 19(4): 521-535.\u003c/li\u003e\n\u003cli\u003eWu J, Huang X, Sun P, Zhang X. What affects patients\u0026apos; choice of consultant: an empirical study of online doctor consultation service [J]. Electronic Commerce Research, 2022.\u003c/li\u003e\n\u003cli\u003eShen T, Li Y, Chen X. A Systematic Review of Online Medical Consultation Research [J]. Healthcare, 2024, 12(17).\u003c/li\u003e\n\u003cli\u003eThaler RH. Mental accounting and consumer choice [J]. Marketing Science, 2008, 27(1): 15-25.\u003c/li\u003e\n\u003cli\u003eGong W, Liu L, Li X, Caine ED, Shi J, Zeng Z, Cheng KK. Quality of asynchronous webchats vs in-person consultations for postpartum depression in China: a cross-sectional, mixed methods study using standardized patients [J]. Lancet Regional Health Western Pacific, 2024, 45.\u003c/li\u003e\n\u003cli\u003eBoulding W, Kirmani A. A consumer-side experimental examination of signaling theory: do consumers perceive warranties as signals of quality? [J]. Journal of Consumer Research, 1993, 20(1): 111-123.\u003c/li\u003e\n\u003cli\u003eAronson E. The theory of cognitive dissonance: A current perspective [M]. Advances in Experimental Social Psychology. Elsevier. 1969: 1-34.\u003c/li\u003e\n\u003cli\u003eErdem T, Swait J. Brand equity as a signaling [J]. Journal of Consumer Psychology, 2001, 7(2): 131-157.\u003c/li\u003e\n\u003cli\u003eErdem T, Swait J. Brand credibility, brand consideration, and choice [J]. Journal of Consumer Research, 2004, 31(1): 191-198.\u003c/li\u003e\n\u003cli\u003eBagwell K, Riordan MH. High and declining prices signal product quality [J]. The American Economic Review, 1991, 81(1): 224-239.\u003c/li\u003e\n\u003cli\u003eKirmani A, Rao AR. No pain, no gain: A critical review of the literature on signaling unobservable product quality [J]. Journal of Marketing, 2000, 64(2): 66-79.\u003c/li\u003e\n\u003cli\u003eZeithaml VA. Consumer perceptions of price, quality, and value: a means-end model and synthesis of evidence [J]. The Journal of Marketing, 1988, 52(3): 2-22.\u003c/li\u003e\n\u003cli\u003eKomiak SX, Benbasat I. Understanding customer trust in agent-mediated electronic commerce, web-mediated electronic commerce, and traditional commerce [J]. Information Technology and Management, 2004, 5(1): 181-207.\u003c/li\u003e\n\u003cli\u003eVargo SL, Lusch RF. Evolving to a new dominant logic for marketing [M]. The Service Dominant Logic of Marketing. Routledge. 2014: 3-28.\u003c/li\u003e\n\u003cli\u003eGoldfarb A, Tucker C. Digital economics [J]. Journal of Economic Literature, 2019, 57(1): 3-43.\u003c/li\u003e\n\u003cli\u003eDarby MR, Karni E. Free competition and the optimal amount of fraud [J]. The Journal of Law and Economics, 1973, 16(1): 67-88.\u003c/li\u003e\n\u003cli\u003eStewart KJ. Trust transfer on the world wide web [J]. Organization Science, 2003, 14(1): 5-17.\u003c/li\u003e\n\u003cli\u003eYaraghi N, Wang W, Gao G, Agarwal R. How online quality ratings influence patients\u0026rsquo; choice of medical providers: controlled experimental survey study [J]. Journal of Medical Internet Research, 2018, 20(3): e99.\u003c/li\u003e\n\u003cli\u003eYang H, Du HS, Shang W. Understanding the influence of professional status and service feedback on patients\u0026apos; doctor choice in online healthcare markets [J]. Internet Research, 2021, 31(4): 1236-1261.\u003c/li\u003e\n\u003cli\u003eRietveld J, Schilling MA, Bellavitis C. Platform strategy: Managing ecosystem value through selective promotion of complements [J]. Organization Science, 2019, 30(6): 1232-1251.\u003c/li\u003e\n\u003cli\u003eLambert D. Zero-inflated Poisson regression, with an application to defects in manufacturing [J]. Technometrics, 1992, 34(1): 1-14.\u003c/li\u003e\n\u003cli\u003eWeesie H. On seemingly unrelated estimation and the cluster-adjusted sandwich estimator [J]. Stata Technical Bulletin, 1999, 52: 34-47.\u003c/li\u003e\n\u003cli\u003eAiken LS. Multiple regression: Testing and interpreting interactions [M]. sage Newbury Park, CA, 1991.\u003c/li\u003e\n\u003cli\u003eGutman J. A Means-End Chain Model Based on Consumer Categorization Processes [J]. Journal of Marketing, 1982, 46(2): 60-72.\u003c/li\u003e\n \u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[{"identity":"277f22ec-98dc-4768-8b62-15fffaee590a","identifier":"10.13039/501100004763","name":"Natural Science Foundation of Inner Mongolia","awardNumber":"2022QN07006","order_by":0}],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":" School of Economics and Management, Inner Mongolia University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"online healthcare, signal verification, trust transfer, quality cues, digital therapeutic relationships","lastPublishedDoi":"10.21203/rs.3.rs-9463453/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9463453/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003ePurpose\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eExisting literature largely presupposes that as patients accumulate clinical encounters, their reliance on extrinsic quality cues—platform reputation chief among them—follows a trajectory of signal decay. Yet this paradigm prematurely generalizes from consumer goods contexts and overlooks boundary conditions in high-stakes credence services where diagnostic uncertainty persists despite repeated interaction. This study examines how physician quality cues differentially shape patient decisions at first versus repeat consultations, and identifies acquisition channel (online versus offline) as a critical moderator of signal evolution.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethodology\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDrawing on a time-lagged research design (T0→T1→T2), we analyze 1,812 asthma specialists from a leading Chinese online healthcare platform. Physician attributes measured at baseline (T0) predict first-time consultations (T1) and subsequent repeat consultations (T2). Zero-inflated negative binomial regression models distinguish between consideration-set formation (structural zeros) and volume conditional on selection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFindings\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOur findings contradict the prevailing signal decay thesis. Physician reputation maintains or intensifies its predictive power across consultation stages—a pattern consistent with signal verification rather than attenuation, and particularly pronounced among online-acquired patients. We identify a dual-mechanism architecture in which reputation operates as a threshold gatekeeper governing consideration-set formation, while process attributes such as response speed drive volume conditional on selection. Acquisition channel significantly moderates these dynamics. Offline-acquired patients exhibit stable reputation reliance and geographic continuity-seeking, preferring physicians in lower-tier cities. Conversely, online-acquired patients display spatial substitution and escalating reputation sensitivity post-verification, while the functional role of price transmutes from a quality-inference heuristic to a pure cost barrier as relationships mature.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImplications\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTheoretically, this study establishes signal verification as a viable alternative mechanism to signal decay in platform-mediated professional services, demonstrating that institutional cues and accumulated direct experience operate as complementary governance mechanisms rather than substitutes. These findings suggest that platform interfaces should be adapted by patient origin: online-acquired patients benefit from prominent process-transparency and efficiency cues during trust formation, while offline-referred patients require geographic proximity signals and reputation confirmation to maintain therapeutic continuity.\u003c/p\u003e","manuscriptTitle":"Sustaining Digital Therapeutic Relationships: The Evolution of Service Quality Cue Utilization from First to Repeat Consultations","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-24 17:39:26","doi":"10.21203/rs.3.rs-9463453/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"e2197c3e-cfe7-4fe7-9d27-28951ea78fb3","owner":[],"postedDate":"April 24th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":66602240,"name":"Information Retrieval and Management"}],"tags":[],"updatedAt":"2026-04-24T17:39:26+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-24 17:39:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9463453","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9463453","identity":"rs-9463453","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}