Multimodal hypersensitivity derived from quantitative sensory testing predicts pelvic pain outcome: an observational cohort study.

A total of 354 participants were enrolled in CRAMPP between August 2014 and December 2018 (see Figure 1 ). Participants were recruited by public advertising and flyers in Evanston, Illinois and the surrounding communities. The severity of menstrual pain was confirmed with internet-based prospective symptom diaries for 1–2 months prior to enrollment [ 51 ]. Enrolled participants included women with low menstrual pain (<3 on a 0–10 Numerical Rating Scale [NRS]; 0 = no pain; 10 = worst pain imaginable) or moderate to severe menstrual pain (≥5 on a 0–10 NRS) but no other chronic pain. We also included participants diagnosed with bladder pain syndrome (BPS), and participants diagnosed with a non-pelvic chronic pain condition (general pain ≥5 on a 0–10 NRS for more than three consecutive months). BPS participants were required to meet American Urological Association diagnostic criteria, and report bladder pain ≥3 on a 0–10 NRS for more than 3 consecutive months [ 42 ]. Participants were excluded for the presence of active pelvic or abdominal malignancies, absence of regular menses (except the chronic pain without BPS group), active genitourinary infection in the last four weeks, inability to read or comprehend the informed consent in English, refusal to undergo pelvic examination/testing, hypertension, or refusal to withdraw from oral contraceptives for two months prior to the study visit. All participants provided informed consent and followed protocols approved by NorthShore University HealthSystem’s Institutional Review Board (EH13–094). Participants were monetarily compensated for their time. From the 354 participants, n =154 were excluded for the following reasons: 52 participants were missing one or more data points from QST (e.g., declined participation in task(s), equipment malfunction, migraine sensitivity precluded participation in visual stimulation, etc.) and one participant was under the influence of recreational or illicit substances during the testing appointment. Additionally, 25 participants were lost to follow-up, 16 participants withdrew from the study, and 60 participants were disqualified (e.g., over-recruited dysmenorrhea without bladder pain, started oral contraceptives, or were unable to complete protocol). Ultimately, a total of 200 participants had complete QST data from their baseline assessment visit. Participants completed annual questionnaires following their baseline visit for up to five years. Annual questionnaires were a reduced version of what was asked at their screen and baseline assessment visits and were used to assess pelvic pain outcome. Because the collection of year five annual questionnaires was incomplete at the time of analysis, we included here completed annual questionnaires until year four. Demographic variables of interest are presented in Table 1 . From the 200 participants with complete QST data, 22 were randomized into a 12-month clinical trial that evaluated the efficacy of cyclical ( n =4) and continuous ( n =12) oral contraceptive pills (OCPs) for treating menstrual and bladder pain compared to a control group that did not take OCPs ( n =6). QST was administered by research staff unaware of participant group assignment or history. Because of the low adherence to OCP use and small comparative sample sizes, we did not consider participants’ enrollment in the clinical trial as exclusionary from our analyses of subsequent annual questionnaires. In addition, multimodal sensory testing was performed on all participants before beginning the use of OCPs, so analyses on internal relationships of MMH on longitudinal outcome were not expected to be confounded by OCP use. Eligible participants that were enrolled in the study first participated in a screen visit and then a second baseline assessment visit at Evanston Hospital (Evanston, IL). During the baseline visit, performed in the midluteal (pain-free) phase of the menstrual cycle, participants completed a panel of medical history and psychosocial questionnaires. Participants next underwent a multimodal sensory testing panel that included mechanosensation, cold pressor, visceral provocation, conditioned pain modulation (CPM), temporal summation (TS), and auditory/visual stimulation. All sensory testing measures and self-report questionnaires are detailed below. We have developed a non-invasive bladder filling task [ 90 ] to characterize visceral hypersensitivity observed across CPP conditions like bladder pain syndrome and irritable bowel syndrome [ 6 , 30 , 57 ]. This task has been validated in participants with bladder pain syndrome and chronic pelvic pain [ 91 ]. Notably, even in participants without bladder pain conditions, elevated pain during this task was associated with more frequent report of daily bladder symptoms [ 49 ]. After voiding their bladder, participants ingested 20 fluid ounces of water. They rated their bladder pain and urgency on a 0–100 VAS across four time points: baseline (BL), first sensation (FS) of bladder filling, first urge (FU) which is the usual desire to void their bladder, and maximum tolerance (MT) of bladder filling (corresponding to widely used cystometric sensory thresholds) [ 48 ]. After reaching maximum tolerance (or 2 hours) and voiding their bladder, participants rated their perceived bladder pain on 0–100 VAS according to four McGill pain questionnaire descriptors to potentially differentiate A δ from C fiber pain components: sharp, pressing, dull, and prickling [ 10 ]. In sum, 12 measures from the bladder task were included in this analysis (i.e., four provoked bladder pain ratings and four bladder urgency ratings across the time points, and four McGill descriptor ratings at completion). We examined PPTs transvaginally and externally because local alterations in myofascial pelvic sensitivity [ 50 ] and widespread alterations in bodily sensitivity are thought to underlie centralized pain [ 37 ]. We determined participants’ PPTs using a digital algometer (Wagner Instruments, Greenwich, CT) with a 1-cm 2 rubber tip driven at a ramp rate of 4 Newtons (N)/sec at three fibromyalgia tender point sites [ 97 ]—right trapezius, the right medial knee fat pad, and the right greater trochanter (hip)—and the forehead. Vaginal PPTs were measured using a finger mounted 1 cm 2 diameter force-sensing resistor (Trossen Robotics, Downers Grove, IL) at a ramp rate of 0.5 N/sec at four vaginal sites: right (5 o’clock position) and left iliococcygeus (7 o’clock), anteriorly against the bladder (12 o’clock), and posteriorly against the anorectal raphe (6 o’clock). Body and vaginal PPT procedures utilized software that guided stimulus application and resulted in high ( α > .89) inter- and intra-examiner reliability [ 51 ]. After each set of PPTs, participants were asked to rate their pain on a 0–10 NRS at each site. These after-pain ratings were adjusted for baseline pain ratings recorded before PPT procedures. We previously established that PPTs and after-pain represent two different components of sensation contributing to MMH [ 50 ]. PPTs represent the average force from two independent trials separated by a 2 minute break period. These averaged PPTs were multiplied by −1 so that a greater value indicates increased sensitivity. In total, 16 PPT measures were included in the planned PCA: eight PPTs (four body and four vaginal sites) and eight after-pain ratings (four body and four vaginal sites). We included CPM in our QST panel because prior studies have demonstrated that CPM predicts pain outcome [ 98 , 99 ]. CPM efficiency is thought to be a metric of descending inhibition as tested by “pain inhibits pain” paradigms [ 70 ]. We assessed participants’ CPM by repeat PPT testing of the left medial knee fat pad before and after ice water immersion of the contralateral hand. After an initial PPT measurement, participants waited two minutes before submerging their right hand up to their wrist into a circulating water bath maintained at 0–6°C. After 10 seconds of submersion, participants rated their hand/cold pain on a 0–10 NRS. After 20 seconds of submersion, a repeat PPT measure was taken from the left medial knee fat pad, after which participants were allowed to remove their hand from the cold-water bath. CPM was calculated by subtracting the PPT force (in Newtons) taken before the water bath from the PPT taken after the water bath (i.e., CPM =After-Before). CPM values were then multiplied by −1 so that a greater number denoted reduced/inefficient CPM (i.e., increased impairment) before including results in the PCA. Additionally, cold pain ratings were adjusted for the water temperature by extracting the residuals from a linear model predicting cold pain as a function of water temperature. In total, two measures from CPM testing were included into the PCA: one CPM score and one cold pain rating adjusted for water temperature. Increased response to repeated application of noxious stimuli (i.e., wind-up pain) is thought to reflect a unique component of spinally mediated sensitization that may be associated with increased risk of chronic pain [ 19 , 71 , 88 ]. Therefore, we measured TS using a commonly used strategy: 10 pressure pulses delivered to the right medial knee fat pad using the same body PPT algometer as described above [ 19 ]. Each pulse was delivered at a ramp rate of 4 N/sec with 1 second breaks between pulses using a software-based metronome to guide application. Each pulse was applied until the initial threshold for a pain rating of one was reached. Participants rated their baseline pain at the application site on a 0–10 NRS following each pulse. The TS task ended when participants reached a pain rating ≥ six or after the tenth trial. Each participant’s baseline pain was subtracted from their pain ratings collected after each pulse. A total of three TS measures were entered into the PCA: the average pain experienced during TS, the rate of change in pain ratings as a function of trial (i.e., slope), and the maximum TS trial experienced. The maximum TS trial was multiplied by −1 so that greater values indicated increased sensitivity (i.e., fewer trials allowed due to reaching a pain rating of six). Investigations of MMH are well served to include additional sensory modalities, such as vision, given that light hypersensitivity is commonly reported in conditions with generalized sensory hypersensitivity, like fibromyalgia and migraine [ 32 , 45 , 65 ]. We assessed participants’ visual unpleasantness sensitivity by presenting a periodic pattern-reversal blue/yellow checkerboard stimulus alternating at 25 Hz for 20 seconds across five blocks. Each block contained a single maximal brightness intensity (1, 30, 60, 90, or 120 lux), and block order was randomized across participants. After each block, participants rated stimulus unpleasantness using the Gracely Box Scale which lists the numbers 0 to 20 next to a set of verbal anchors [ 36 ]. A total of two visual sensitivity measures were entered into the PCA: the average visual unpleasantness rating across the blocks, and the rate of change in visual unpleasantness as a function of brightness intensity [ 59 ]. Auditory stimulation assessed an additional sensory modality with reported hypersensitivities in functional pain syndromes [ 53 , 64 , 95 ]. Prior to measuring the participants’ auditory unpleasantness sensitivity, a program first verified that participants maintained less than 20 dB hearing loss (250–8000 Hz) [ 75 ] to equate hearing ability. Next, we presented a series of auditory steady state 80Hz [ 26 ] volume modulated tones (1200 and 1350Hz) in random order of intensity (15, 30 45, or 60 dB) delivered via ground-isolated optimally flat frequency response pneumatic insert earphones (Etymotic, Elk Grove Village, IL). After each 20 second stimulus, the participant rated her perceived unpleasantness on the Gracely Box Scale as described above. A total of two auditory sensitivity measures were entered into the PCA: the average auditory unpleasantness rating across the blocks, and the rate of change in auditory unpleasantness as a function of loudness intensity (i.e., slope). We recorded participants’ scalp electroencephalography (EEG) during visual and auditory sensitivity tasks. Given the behavioral focus of this investigation, these EEG data were out of scope and not included in the present investigation. EEG data from the visual task are published elsewhere [ 59 ]. In the health history profile, we administered several validated self-report questionnaires that assessed various aspects of pelvic pain and associated symptoms. Bladder symptom severity was assessed via the Interstitial Cystitis Symptom Index (ICSI) and Problem Index (ICPI) [ 72 ]. The Genitourinary Pain Index (GUPI) provided a complementary assessment of these urogenital symptoms [ 20 ]. The Complex Medical Symptoms Inventory (CMSI) was used to assess functional symptom burdens that occurred for at least three months in the past year and at any time in the participants’ lifetime [ 96 ]. Somatic symptoms were assessed using the Brief Symptom Inventory (BSI) [ 25 ]. We assessed several health domains from the NIH Patient Reported Outcomes Measurement System (PROMIS) [ 16 ], including anxiety (short form 8a), depression (8b), pain interference (6), pain behavior (7), global physical health (4), and global mental health (4). Specifically, PROMIS raw short form scores were analyzed. Menstrual pain is associated with non-cyclic pelvic pain [ 93 ] and non-pelvic pain hypersensitivity [ 51 , 54 , 73 , 74 ], suggesting that menstrual pain may be a risk factor for developing chronic pain [ 54 , 73 ]. Therefore, we assessed participants’ menstrual pain on a 0–100 VAS on the worst day of their period over the past three months in the absence of pain relievers (e.g., non-steroidal anti-inflammatory drugs [NSAIDs], acetaminophen, etc.) Participants completed annual questionnaires virtually by email in REDCap for up to four years. As part of this questionnaire, participants rated their average feeling of 1) menstrual and non-menstrual pelvic pain, 2) pain with urination, and 3) pain with bowel movements during the past week using a 0–100 VAS (0=no pain; 100=worst pain imaginable). VAS scales are more sensitive to changes than descriptive word-based scales [ 87 ] and have linear properties amenable to averaging [ 68 ]. Also, these three questions were highly collinear at the baseline assessment: 1 vs. 2, r (198) = .61 95% confidence interval (CI) [.51, .69], p < .001; 1 vs. 3, r (198) = .56 [.45, .65], p < .001; and 2 vs. 3, r (198) = .66 [.57, .73], p < .001. Therefore, we averaged these three questions to create a composite pelvic pain outcome variable. Similar composite pain recall variables formed by averaging have demonstrated high validity and reliability comparable to daily diary pain ratings [ 56 ]. To evaluate the ability of different sensory tests to predict pelvic pain outcome, we combined the 40 measures from the multimodal sensory testing panel into three composite variables using summed Z-scores: traditional QST measures, bladder test measures, and audio/visual stimulation. The QST composite comprised 25 measures including PPTs (i.e., thresholds, after pain, and descriptors), TS, CPM, and cold pain. The bladder test composite comprised 11 measures from the bladder test, including pain, urgency and descriptors. The audio/visual sensitivity composite comprised 4 measures, including mean unpleasantness and slope of the stimulus-response function from the auditory and visual tests. Prior to calculating Z-scores, all measures maintained the same directionality such that greater values denoted increased pain/impairment/sensitivity. Final composites were mean centered for regression analyses. To examine the predictive ability of self-reported somatic symptoms, we utilized the Generalized Sensory Sensitivity Brief Scale (GSS Brief) [ 85 ]. The GSS Brief approximates GSS (developed using confirmatory factor analysis) that captures comorbid sensory hypersensitivity often present in chronic overlapping pain conditions. Participants indicate regions on a body map where they have experienced pain during the last week and endorse whether they had any of the following symptoms for at least three months in the past year: 1) dry mouth, 2) rapid heart rate, 3) problems with balance, 4) sensitivity to certain chemicals, such as perfumes, laundry detergents, gasoline, and others, 5) sensitivity to sound, and 6) frequent sensitivity to bright lights. The GSS Brief is a good approximation of GSS, and the GSS factor structure was recently replicated in the cohort used in this study [ 84 ]. A formal power analysis was used to plan the broader clinical trial ( NCT02214550 ). Because the present investigation was a secondary analysis, all participants’ data were included if complete. To assess how well baseline QST, bladder test, and audio/visual measures independently predicted future pelvic pain outcome, we performed four multiple regressions using self-report data from annual follow-up questionnaires. Pelvic pain outcome served as the dependent variable. The independent variables included the summed Z-scores (QST, bladder test, audio/visual sensitivity) defined above. We also included baseline pelvic pain outcome as a covariate. All independent variables were measures collected at the participants’ baseline visit. Additionally, we reduced the dimensionality of our sensory testing panel (40 measures/columns) using principal components analysis (PCA) [ 2 ]. All measures maintained the same directionality such that greater values denoted increased pain/impairment/sensitivity/etc. Each column was then Z-scored prior to decomposing the matrix via singular value decomposition [ 1 ]. Inferential statistics were performed using data resampling techniques [ 2 , 9 ]. Permutation testing for the PCA was conducted by creating null distributions for each principal component (PC) by shuffling each column’s values without replacement and repeating the PCA for 2,000 iterations. Probability values for each PC were calculated by comparing our fixed-effects eigenvalues to their respective null distributions. Contributions were calculated by dividing each measure’s squared factor score by the component eigenvalue [ 9 ]. Bootstrap samples were formed by selecting participants at random with replacement. Bootstrap distributions were formed by supplementary projecting the bootstrap samples onto the eigenspace generated from the fixed-effects analysis [ 9 ]. This procedure was repeated 2,000 times. Bootstrapping quantified each measure’s loading stability/contribution importance using bootstrap ratios (BSRs). A BSR is the ratio between a measure’s fixed-effect factor score (i.e., loading) and the standard deviation of its bootstrapped distribution. BSRs are interpreted like Student’s t value. Therefore, significantly contributing measures have | BSRs | > 1.96 ( p < .05). To compare PCs with measures not included in their initial formulation, we calculated bootstrapped correlations between the row-wise (i.e., participant) factor scores and validated self-report questionnaires. To assess how well QST-based PCs predicted future pelvic pain outcome, we repeated the multiple regression procedure as described above except that pertinent PCs served as the independent variables instead of the three Z-scored measures. A sensitivity analysis was performed to examine whether adjusting for prevalence rates of dysmenorrhea, bladder pain syndrome, and other types of chronic pain in the general population altered regression results. Keeping our sample size constant ( n =200), participants were sampled with replacement according to the following prevalence rates [ 12 , 24 , 82 ]: 50% pain-free healthy controls ( n =100), 40% moderate-severe dysmenorrhea ( n =80), 5% bladder pain syndrome ( n =10), 5% other chronic pain ( n =10). Regression analyses were recomputed using these bootstrapped data samples. This procedure was repeated for 2,000 iterations. Mean regression estimates and effect sizes were then compared to original unadjusted values. We conducted post hoc model comparison analyses to assess the relative performance of PCs over a priori defined sensory testing composites (summed Z-scores) in predicting future pelvic pain. Given the multicollinearity between PCs and the composites, we compared the Akaike and Bayesian information criterion (AIC and BIC, respectively) values between two models estimated separately at each year of follow-up in accordance with previously published guidelines [ 18 ]. Evidence ratios for the best model versus a comparator model were calculated as exp { - 1 2 Δ } where Δ is the change in AIC between models [see 18 ]. Analyses were performed in R (4.1.0) within RStudio (1.4.1106) using the following packages: fixed- and random-effects PCA were performed using ExPosition [ 9 ], bootstrapped correlations were computed using psych [ 76 ], effect sizes were calculated using effectsize [ 11 ], data processing and figures were generated using tidyverse packages [ 94 ], color palettes were inspired by RColorBrewer [ 69 ] and ghibli [ 52 ]. All code used to process, analyze, and visualize the data in this manuscript is available on GitHub ( https://github.com/mkmiecik14/mmh ), and data are available on Open Science Framework ( https://doi.org/10.17605/OSF.IO/27KY9 ).

Distributions of predictor variables are visualized in Figure 2A (see Table 2 and Table 3 for complete sensory testing data). Median pelvic pain outcome across the four-year follow up was stable and many participants reported moderate pelvic pain (see Figure 2B left panel). Cross-sectional relationships between sensory tests and baseline pelvic pain demonstrated that sensitivities across the QST, bladder test, or audio/visual composites were associated with worse baseline pelvic pain (see Figure 2B right panel). Multiple regressions were used to assess how well QST, bladder test, and audio/visual testing predicted pelvic pain outcome on annual questionnaires administered up to four years following the baseline visit (see Table 4 ). We accounted for baseline pelvic pain by including it as a covariate (see Table S1 for descriptive statistics). Baseline pelvic pain was the strongest predictor of year 1 pelvic pain, but steadily declined over time (see Figure 2C left panel). The bladder test and QST predicted pelvic pain at years 3 and 4, respectively. Audio/visual testing did not predict outcome at any year. After adjusting for population-based prevalence rates of patient groups, baseline pelvic pain was a stronger predictor of year 4 pelvic pain outcome (see Figure S1 ). Also, QST explained little to no variance in pelvic pain outcome at any year. Thus, overall baseline pelvic pain was a better predictor of future pelvic pain than QST, bladder test, and audio/visual composites during the first 3 years, but was not significant at year 4. We determined the number of PCs underlying QST variability across the cohort by examining the scree plot (see Figure S2 and Table S2 ), permutation testing results, and loadings using geometrically plotted factor scores of QST measures [ 2 ]. Accordingly, we identified three components as interpretable. The first PC (PC1) explained 20.6% of the variance ( p = .0005), the second (PC2) 12.4% ( p = .0005), and the third (PC3) 446 9.5% ( p = .0005). Factor score plots for the first three PCs are shown in Figure 3 (see Figure S3 for contributions and Figure S4 for bootstrapped significance of factor loadings). Given that all measures loaded positively on PC1, we interpret PC1 to represent MMH (i.e., increased sensitivity on one measure was associated with an increased sensitivity on another). Forehead, hip, knee, shoulder, and vaginal PPTs positively loaded on PC2, while their respective after-pain ratings were opposed on PC2. This factor is representative of a stimulus-response function of pressure and after-pain resulting from PPT testing, hereafter referred to as PPT S-R (PPT stimulus-response). In other words, participants with lower PPTs (i.e., less force, greater sensitivity) reported less after-pain ratings. PC3 depicted an opposing relationship between bladder task measures and PPTs (thresholds and after-pain ratings). Given that additional measures (e.g., visual mean) had weak loadings on PC2, PPT-SR may represent another complex integratory mechanism. However, the contribution of these additional measures steeply drops off ( Figure S3 ). Given the orthogonality of PCs, PC3 captured bladder pain hypersensitivity that was distinct from PC1 ( MMH ). Hereafter, we distinguish our interpretations of the observed PCs from theoretical constructs by using italics for PC interpretations, e.g., MMH = PC1, and unitalicized text for theoretical constructs (e.g., MMH). Figure 4 depicts the bootstrapped correlations between row-wise factor scores (i.e., participants) of PCs and validated questionnaires of self-reported menstrual pain, genitourinary symptoms, depression, anxiety, and health (see Table S3 for descriptive statistics of self-report measures). MMH and bladder hypersensitivity correlated strongly with every measure included, while PPT S-R only weakly correlated with two standardized clinical questionnaires for bladder pain: the ICSI and GUPI. These widespread correlations observed across PCs 1 and 3, but not PC2, demonstrate that these two dimensions ( MMH and bladder pain hypersensitivity ) explain variability in participants’ current pain- and health-related quality of life. Also, given the orthogonality of PCs, these results suggest the contribution of two mechanisms to explain patients’ current pelvic pain health-related quality of life: 1) MMH and 2) bladder hypersensitivity. Multiple regressions were used to assess how well the three obtained PCs (i.e., MMH , PPT S-R , and bladder hypersensitivity ) predicted pelvic pain outcome on annual questionnaires administered up to four years following the baseline visit (see Table 5 for regression results). Distributions of the PCs are presented in Figure 2A . Similar to the self-report measures, baseline pelvic pain correlated with both MMH and bladder hypersensitivity , but not with PPT-SR (see Figure 2B right panel). PCs were orthogonal to each other ( r =0). Baseline pelvic pain was the strongest predictor of year 1 pelvic pain, but that association steadily decreased over time. In contrast, MMH increased in its predictability of pelvic pain outcome over time and predicted worse pelvic pain outcome continuously up to four years later (see Figure 2C right panel). A 1 SD increase in MMH at baseline predicted a .44 SD increase, or nearly 6 VAS points, in pelvic pain ratings four years later. PPT S-R and bladder pain hypersensitivity did not predict outcome at any year. Adjusting for population-based prevalence rates of included diagnostic groups replicated the observed sample-wise regression results (see Figure S1 ). Thus, the positive association between MMH and pelvic pain was robust and differentiated participant four-year trajectories (see Figure 5 ). Additional analyses examining the effect of annual questionnaire attrition determined that longitudinal data were missing at random and did not selectively depend on recruited participant groups, pelvic pain outcome, nor predictor variables of interest (see the Supplementary Material section on attrition for details). We compared the added predictive value of MMH over sensory testing composites by comparing the AIC and BIC values between two models estimated separately at each year of follow-up: 1) the MMH model regressed pelvic pain outcome on MMH and 2) the composite model regressed pelvic pain outcome on the QST, bladder test, and audio/visual summated Z-scored composite variables. Baseline pelvic pain served as a covariate in both models. Across all four follow-up years, the MMH model was the best model with the lowest AIC and BIC values. The strength of the evidence (i.e., evidence ratio) for the MMH model over the composite model (i.e., n x denotes evidence is n times stronger for the MMH than composite model) using the more conservative AIC was 7.28x at year 1 (ΔAIC = 3.97, ΔBIC = 9.90), 9.50x at year 2 (ΔAIC = 4.50, 508 ΔBIC = 10.2), 1.73x at year 3 (ΔAIC = 1.10, ΔBIC = 6.25), and 5.58x at year 4 (ΔAIC = 3.44, ΔBIC = 8.37) [ 18 ]. These results suggest a preference for the simple and more parsimonious MMH model in predicting pelvic pain outcome over the composite model. To compare MMH to the GSS Brief—a questionnaire-based measurement of increased sensory and diffuse pain sensitivity—we examined their correlations and compared their performance in predicting pelvic pain using multiple regression. Zero-order Pearson correlations demonstrated a strong positive association between baseline pelvic pain and the GSS Brief, r = .56 95% [.45, .64], p < .001, but a weaker positive relationship between GSS Brief and MMH , r = .41 [.28, .52], p < .001. Despite these correlations, all predictors demonstrated low multicollinearity (variance inflation factor <2) across all four years in regression models predicting pelvic pain outcome as a function of MMH , the GSS Brief, and baseline pelvic pain serving a covariate. Baseline pelvic pain decreased in predictive strength over time, while MMH increased in predictive strength. However, the GSS Brief did not predict pelvic pain outcome at any year, suggesting that MMH is a better predictor of outcome than questionnaire-based methods (see Table S4 ).

Inadequate understanding of mechanisms underlying pain sensitivity, and how sensitivity conveys risk for developing chronic pain, remains a barrier to treating and preventing chronic pain conditions. QST is the most widely used method for systematically measuring pain sensitivity [ 21 , 22 , 77 ]; however, traditional assays that measure single nociceptive modalities (e.g., thermal, pressure) have demonstrated inconsistent predictive power for pain outcome [ 15 , 31 , 80 , 89 ]. Likewise, the ability of dynamic QST paradigms assessing pain modulation, such as CPM and TS, to predict pain outcome have also been mixed [ 27 , 66 , 70 , 71 ]. Our results demonstrate that the utility of QST can be improved by measuring sensory hypersensitivity more broadly (i.e., including several disparate sensory modalities to evaluate MMH). This approach is aligned with the repeated observation that many functional pain syndromes additionally report increased sensitivity to environmental stimuli (e.g., lights, sounds, odors). Thus, dysfunction in global sensory processing may be more crucial to chronic pain risk than any nociceptive mechanism underlying hyperalgesia. In a cohort of women with a range of chronic pelvic pain risk, we found that including both nociceptive and non-nociceptive (e.g., visual and auditory) assays identified that PC1, MMH, was a robust common denominator underlying sensory testing variability. Notably, baseline MMH predicted pelvic pain annually for four years even when accounting for baseline pelvic pain. MMH outperformed sensory sensitivity composite models and was a better predictor of outcome than a questionnaire-based method of assessing generalized sensory sensitivity. MMH, rather than solitary dysfunction in specific nociceptive modalities or pain modulation, appears to underlie evolution of pain vulnerability and likely plays a substantial role in the development of chronic pain conditions. QST studies often infer centralized or generalized mechanisms of pain sensitivity to explain observed hypersensitivity in chronic pain conditions [ 30 , 34 , 40 , 41 , 45 , 46 , 51 , 53 , 59 , 65 ]. Relationships between and within QST modalities supports the hypothesis that QST assays measure distinct mechanisms of pain sensitivity [ 39 , 47 , 55 , 61 ]. However, given that most studies relied on singular noxious modalities of QST, or combinatory approaches of select modalities (e.g., heat and pressure), they provide only a limited ability to evaluate generalized mechanisms of hypersensitivity in centralized pain conditions [ 13 ]. The current investigation improves upon these previous attempts by administering a broader panel of nociceptive sensory tests (PPTs, CPM, TS, cold pressor, bladder provocation) and non-nociceptive supraspinal tests (visual and auditory sensitivity). Like previous QST studies [ 39 , 47 ], a PCA of these 40 QST measures resulted in modality-specific components: PPT S-R [ 79 ] and bladder hypersensitivity [ 91 ]. In contrast, the largest source of underlying QST variability was MMH , a component that was not modality specific. Although previous studies have conjectured that centralized hypersensitivity (e.g., generalized sensory sensitivity, somatization, somatic symptoms disorders, sensory modulation disorder, etc.) [ 8 , 14 , 23 , 85 ] underlies chronic pain conditions, our results provide evidence of MMH being a broad construct that correlates with affective symptoms, somatic symptoms, genitourinary pain, menstrual pain, and overall pain and health. Given the consistent and robust correlation between MMH and these other factors crucially affecting chronic pain, we hypothesize neural regions underlying MMH, such as the insula and cingulate cortex [e.g., 4 , 45 , 64 , 83 ], contain critical circuits in functional pain syndromes. Studies that have used various QST measures to predict long-term outcome have not generated uniform conclusions [e.g., 27 , 35 , 38 , 46 , 66 , 67 , 86 , 98 , 99 ]. In contrast, current pain intensity has consistently been associated with future pain experience [ 3 , 33 , 60 , 98 ]. This investigation may explain the equivocal findings for QST on future pain outcomes: risk for worse pain outcome is possibly not due to alterations in normal interpretation of specific nociceptive modalities, but rather MMH. When testing whether current pain (baseline pelvic pain), MMH , or nociceptive modality-specific components (PCs 2 and 3) best predicted future pelvic pain outcome, we found that current pain provided a diminishing degree of predictive power annually, while MMH provided an increasing degree of predictive power over time. Additionally, MMH was the only significant predictor of year four pelvic pain intensity. This pattern of results held consistent even when adjusting for population prevalence rates of dysmenorrhea, bladder pain syndrome, and other types of chronic pain (see Figure S1 ). In contrast, sensory testing composites sporadically predicted pelvic pain outcome: a more sensitive bladder test and QST assessment predicted worse pelvic pain at years three and four, respectively, with baseline pelvic pain decreasing in predictive power annually. A similar pattern of results was observed when adjusting for population prevalence rates; however, baseline pelvic pain became an important predictor of year four pelvic pain (see Figure S1 ). When sensory testing was split into composite measures, current (baseline) pelvic pain was the best predictor of future pelvic pain, but longitudinal estimates became unstable. Therefore, the modality-specific QST differences often observed in cross-sectional studies comparing pain patients to controls [e.g., 40 , 46 ] could reflect dynamic states of hypersensitivity. In contrast, the increased MMH in participants that develop worse pelvic pain suggest that a pervasive sensory processing mechanism is stable across time. The multimodal sensory testing panel used here is one of the largest efforts to characterize MMH in individuals harboring variable degrees of risk for CPP and includes one of the longest follow-up periods of any previous QST pelvic pain study. The current investigation leveraged PCA that facilitated dimensionality reduction (from 40 to 3 variables) in a racially diverse sample enriched with at-risk individuals. However, some limitations may affect the generalizability of our results. The sample included a large number of college students. As a result, the sample was young ( M =25, SD =6 years), and the vast majority were nulliparous (85% without a prior pregnancy). A strength of our study is the long-term follow-up, although there was some attrition. Prior studies of long-term pain had comparable attrition rates (50–80%), and attrition had a marginal impact on outcome [ 28 , 81 ]. Together, our analyses suggest that longitudinal data were missing at random and did not selectively depend on recruited participant groups, pelvic pain outcome, nor predictor variables of interest calculated either via sensory testing composites or PCA. Given that comprehensive multimodal sensory testing is onerous on patients and staff, it remains challenging to validate these findings within large clinical populations. Although the longitudinal stability of MMH is unknown, it is encouraging that MMH derived from a one-day sensory testing panel predicted a single week of average pelvic pain four years later. Future work would be well served to establish the minimum battery of sensory tests needed to evaluate MMH. Thus, we envision that with further refinements of MMH testing, such as automated methods [e.g., 46], it may possible to validate a clinically oriented short protocol. Developing early predictive methods for long-term chronic pain risk are essential for preventing its deleterious effects. Because the goal of CRAMPP was to evaluate the role of menstrual pain and other nociceptive mechanisms in the development of chronic pelvic pain in an at-risk cohort, the follow-up questionnaires were specifically focused towards interrogating pelvic pain. Future studies should investigate the role of MMH on non-pelvic pain risk. This analysis provided crucial evidence supporting MMH, a hypothesized construct underlying “centralized” mechanisms of pain sensitivity, and its predictive ability of worse pelvic pain outcome. Our study demonstrates that multimodal sensory testing improves the prediction of pain status or outcome over parsimonious unimodal approaches or questionnaire-based strategies. MMH exists on a continuum, and individuals that report increased sensory sensitivity or demonstrate hypersensitive QST responses are more vulnerable to worse future pain [ 8 , 38 , 66 , 89 ]. Neuroimaging paradigms have implicated the anterior insula and cingulate cortex as important for multimodal sensory integration and nociceptive appraisal [ 44 , 45 , 64 ]. Therefore, future work to abrogate the course of chronic pain would be well served to understand and target the neural mechanisms that underlie MMH.