Unraveling the Associations Between Voice Pitch and Major Depressive Disorder: A Multisite Genetic Study

Unraveling the Associations Between Voice Pitch and Major Depressive Disorder: A Multisite Genetic Study | 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 Article Unraveling the Associations Between Voice Pitch and Major Depressive Disorder: A Multisite Genetic Study Jonathan Flint, Yazheng Di, Elior Rahmani, Joel Mefford, Jinhan Wang, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4135145/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 Major depressive disorder (MDD) often goes undiagnosed due to the absence of clear biomarkers. We sought to identify robust voice biomarkers for MDD and separate trait biomarkers indicative of MDD predisposition from state biomarkers reflecting current depressive symptoms. We investigated the association between voice pitch and MDD in a multisite recurrent MDD case-control cohort and validated our findings in a replication cohort. We then determined the heritability of identified features, their genetic correlation with MDD, and their association with depressive state. We found robust associations between MDD and pitch features, which depicted a slower pitch change and a lower pitch. These features achieved an AUC-ROC of 0.80 in MDD classification. Features measuring the variability and extremity of pitch change speed were heritable and had genetic correlations with MDD. State-related features were also detected. Our results return vocal features to a more central position in clinical and research work on MDD. Biological sciences/Psychology Health sciences/Biomarkers/Diagnostic markers Biological sciences/Genetics/Behavioural genetics Figures Figure 1 Introduction Changes in human pitch and tone of speech have been noted as an important sign in depression for over a century 1,2 . Although not contained in symptomatic criteria for major depressive disorder (MDD) in DSM-III 3 , DSM-IIIR 4 , DSM-IV 5 , or DSM-5 6 , they are found in 26 out of 28 detailed clinical descriptions of melancholia published from 1880-1900 1 and in 19 out of 21 of such descriptions of depression published in the 20th century 2 . Given the current challenges in diagnosing MDD 2,7,8 , where a large proportion of cases (ranging from 50% to 90%) remain untreated 9–11 , the transformation of voice phenomena into diagnostic biomarkers could aid in both clinical and research arenas. Clinical observations describe the speech patterns of depressed patients as slow, weak, low-pitched, and monotonous 1,2,12,13 . These phenomena are typically quantified by increased pause time, lower volume, lower pitch, and reduced pitch variability 14–16 . Many studies have sought to develop features from pitch as a biomarker for depression 17–26 , but none have to date achieved sufficient accuracy and precision for clinical utility. The large number of both vocal features and confounds 24,25 imposes a multi-testing burden that requires large sample sizes which few studies have obtained 16 . The critical distinction between state and trait effects on voice has also never been addressed 27–29 . Furthermore, MDD is likely heterogeneous 30,31 : studies not accounting for this may be underpowered and/or identify effects that do not generalize well 16 . In this paper, we report voice analysis of a large case-control study of MDD and replicate findings in an independent sample, both from China. By including genetic data, we determine differences in voice that reflect either state or trait. Our results return vocal features to a more central position in clinical and research work on MDD. Results We analyzed voice records from interviews conducted as part of the CONVERGE 32 (China, Oxford, and VCU Experimental Research on Genetic Epidemiology) study, a multi-site MDD case-control genetic investigation. CONVERGE recruited 11,670 Han Chinese women through a collaboration involving 58 hospitals in China. The initial and final sample sizes used in each following analysis are presented in Figure 1 . By recruiting only women, we increase homogeneity in both genetic and vocal signals 33,34 . We identified 364,929 voice segments from 7,654 subjects. Summary information is shown in Table S1. 60% of the subjects spoke in standard Mandarin, whereas the rest spoke either local languages or Mandarin with local accents. The mean duration of concatenated voice segments per subject was 297.75 seconds (SD=208.86) for cases and 97.44 seconds (SD=122.96) for controls. We additionally recruited samples for replication purpose with a final sample size of 1,189. Their summary information is in Table S2 . Association Between Pitch Features and Lifetime MDD We extracted 30 voice features ( Table S3 ), providing a comprehensive characterization of the speaker’s prosodic patterns, including pitch and intonation 35,36 . Specifically, we calculated statistics and functions based on the time series of Fundamental Frequency (F0) and its differential values, namely pitch change speed (ΔF0). These statistics and functions include mean values, quartiles, range, and regression coefficients, capturing the pitch trends and pitch dynamics in speech. In an initial analysis, we looked for associations between each pitch features and MDD for the entire 7,654 subjects, adjusting for five demographic variables and ignoring hospital stratification. This naive approach generated many highly significant results ( Table S4 ), which, as we could see from a quantile-quantile plot ( Figure S1 ), were most likely due to inflated P-values. We attributed this inflation to the correlation between MDD and hospitals. Indeed, the case/control ratios were uneven across different hospitals (as shown in Table S 5 and Table S 6 ), a factor that could inflate the results. For example, if cases were mainly recruited from Shanghai (Southern China) and controls were mainly from Beijing (Northern China), the voice features indicating a Shanghai accent would be associated with MDD in the naïve analysis. To account for variations in location and/or hospital, we implemented a two-stage meta-analysis, in which associations were first calculated at the hospital level and subsequently pooled using a random-effects model. From the 30 pitch features, we identified 20 features significantly associated with MDD at FDR<0.05 ( Table 1 ). Using this approach, we found that our results were well-calibrated (QQ Plot in Figure S2 ). As a test of sensitivity of our result to genetic differences between cases and controls, we compared analyses with and without the inclusion of 20 genetic principal components (PCs). Results (shown in Table S7 ) are almost identical, showing no overall decrease in significance with the inclusion of the genetic covariates. We then tried to replicate the identified associations in the replication cohort, using the same two-stage meta-analysis methodology. Out of the 20 features, 16 associations were significantly replicated at FDR<0.05 ( Table 2 ). Notably, 11 of these features were derivatives of the ΔF0 series, or in other words, they were various characterizations of the intonations (speed of pitch change). The distributions of these 11 ΔF0 features are shown in Figure S3 and the other five F0 features in Figure S4 . Genetic Correlations Between Pitch Features and MDD After establishing the association between the voice features and MDD, our next goal was to differentiate trait-related from state-related acoustic measures. We started by estimating the SNP-based heritability for each of the 16 pitch features and we found that four were heritable at FDR<0.05 (see Table 2 for heritable features and Table S8 for results of all 16 features). We then estimated the genetic correlation with MDD for the four features and found that three ΔF0 features had significant genetic correlations ( Table 2 ). They were: 1) the interquartile range (IQR1-3), quantifying the variation of speed in pitch change; 2) the kurtosis, signaling the extremity of speed in pitch change; and 3) the maximum, representing the speed of the fastest pitch change. We performed heritability estimates adjusting for more genetic PCs as a sensitivity test. The heritability of the voice features remained significant even after adjusting for as many as 60 genetic PCs ( Table S9 ). We also performed GWAS for these heritable voice features, but we did not find any significant hits, presumably owing to the limited sample size ( Figures S5-S12 and Table S10). Association Between Pitch Features and Current MDD State We obtained self-assessed current mood for subjects in the replication cohort. Subjects filled in the depressive symptom checklist (SCL) 37 , providing a more immediate assessment of their depressive state. The distributions of SCL scores for cases and controls are presented in Figure S13 . There is an overlap in scores for cases and controls, indicating that many cases were likely not in an episode of MDD at the time of the interview (due to the design which focused on lifetime rather than current MDD episodes). Of the 16 pitch features, 11 showed a significant association with current depressive symptoms after FDR correction ( Table 3 ), including heritable and non-heritable features. We noticed one non-heritable feature significantly associated with current depressive symptoms, the position of the maximum ΔF0 (maxΔF0_Pos), which reflected the time point when the pitch changes in its fastest speech. We asked whether maxΔF0_Pos was specifically predictive of current depressive symptoms, using a case-only design to answer this question. We found that within cases the association with SCL scores was -0.15 (SE = 0.07, P = 0.028, N=326), using the two-stage meta-analysis method. Classification Performance We assessed the classification performance for MDD based on a logistic regression model trained on the identified voice features and covariates, compared with a null model using only the covariates. The null model had a sensitivity of 0.66, and a specificity of 0.59, an AUC-ROC (area under the receiver operating characteristic curve) of 0.70, and an AUC-PR (area under the precision recall curve) of 0.64 in detecting MDD. The full model achieved a sensitivity of 0.74, a specificity of 0.73, an AUC-ROC of 0.80, an AUC-PR of 0.75, and a net reclassification improvement of 0.11. We also tested the classification performances on voice recordings of different lengths, to determine the minimum length of voice necessary for reliable MDD detection. The performance on limited segment durations is shown in Figure S14 . The metrics improve noticeably until about 150 seconds. Around 200 seconds, AUC-ROC and AUC-PR show a slight increase, but sensitivity and specificity exhibit a plateau over a span of 50 seconds. Associations Between Pitch Features and MDD Symptoms, Risk Factors, and Comorbidities. Considering the heterogeneous nature of MDD, we tested for associations between the 30 pitch features and 33 clinical features obtained from cases, including MDD symptoms, environmental risk factors, comorbidities, and suicidality. The results of this within-case two-stage meta-analysis are shown in Table S11 . Six associations were significant after Bonferroni correction (P<0.05/990=), out of which two were significantly replicated (P<0.05/6=). Both associations were between the total number of stressful life events and ΔF0 features, including the IQR1-3 of ΔF0 (16 hospitals in CONVERGE, total N=2,064, ; 4 hospitals in the replication, total N=295, ) and maximum of ΔF0 (16 hospitals in CONVERGE, total N=2,064, ; 4 hospitals in the replication, total N=295, ). Context-Constrained Analysis For all the above analyses, the voice features were extracted from the concatenated segments of spontaneous speech. The speech content and the conversational context were not controlled, as our goal was to identify voice patterns that are persistent and detectable across various speech contexts and build generalizable biomarkers useful in real-world settings. Recognizing that voice acoustic features and their associations with MDD might be sensitive to the context of the interview, we conducted a sensitivity analysis on voice responses to specific questions to check if the effects remain consistent across contexts. We selected two questions from the demographic section of the interview based on their high response rates ( Table S12 ) and neutral nature. These questions—D2.A (“What is your date of birth?”) and D10 (“How much do you weigh while wearing indoor clothing?”)—were chosen because they are unlikely to trigger emotional differences between MDD cases and controls, thus serving as a stable basis for comparison. We identified 533 subjects with voice response to question D2.A and 617 to question D10. The average segment durations were 3.37 seconds (SD=3.05), and 8.47 seconds (SD=2.69), respectively. For each question, we used the corresponding segments to extract the 16 pitch features that were associated with MDD in our main analysis. Using the two-stage meta-analysis method again, we re-estimated their associations and used a one-sided binomial sign test to test consistency in the direction of association effects between the main analysis and analysis here (that is, a one-sided test of whether this fraction is greater than 0.5, see method for our hypothesis). The estimated association effects in context-constrained analysis are reported in Table S13 . We found that for question D10, three out of 16 pitch features maintained significant associations with MDD at FDR<0.05. Remarkably, 15 out of 16 features showed the same direction of association effects, a fraction significantly higher than chance (Binomial P= 0.00026). For D2.A, despite the average duration being a mere 3.37 seconds, three features achieved nominal significance for associations (uncorrected P<0.05), and 12 out of 16 pitch features showed consistent directions of association effects (Binomial P= 0.038). In total, 12 out of 16 voice features showed consistent directions of association effects across all four analyses ( Figure S15 and summarized in Table S1 4 ). We conclude that the findings from the main analyses are not biased by the context of the interview. Discussion We set out to find voice pitch features associated with MDD. By using a large and homogeneous case-control cohort, we provided robust evidence that certain pitch features distinguished MDD cases from matched controls. These associations were further validated by replication. More importantly, three heritable features showed a significant genetic correlation with MDD, and a group of non-heritable features were associated with current depressive symptom severity (as summarized in Table S14 ). Our study advances research efforts to find voice biomarkers of MDD by combining the power of a two-stage meta-analysis with a large, diverse sample collected from numerous hospitals. This rigorous methodology helps to account for potential confounding factors and paves the way for more reliable and generalizable findings. Our findings support and extend previous studies which have indicated potential links between pitch patterns and MDD, but were limited by smaller sample sizes or more heterogeneous cohorts 15,16 . First, our large sample size provided adequate power to test several pitch features from a standardized features set, providing more fine-grained quantitative evidence for the descriptions of the monotonous speech pattern in MDD than in previous studies. Previous studies have found that depressed people speak more slowly with lower pitch and decreased variability 16,25 . Here, our study showed MDD was negatively associated with features measuring how fast pitch changes (the maximum, the 3rd quartile, and the root quadratic mean of ΔF0, Table 1 ), indicating that the reduced rate of change in pitch is a characteristic of voice in MDD patients. We also found that MDD patients spend less time in their upper vocal range (Time with F0>90 th percentile, Table 1 ), affirming the “low-pitched” pattern. Second, our results indicate that MDD's pitch dynamics involve more than reduced variability, showing a broader pitch range and more extreme values (range and kurtosis of F0, Table 1 ). Our analysis also revealed an uneven distribution of the speed with which an MDD patient’s pitch changes, as shown by the negative association between MDD and the flatness of ΔF0 and the positive association with the kurtosis of ΔF0 ( Table 1 ). Overall, these various features enrich our understanding of pitch dynamics, demonstrating a pattern of slower change in pitch, yet with more frequent occurrences of extreme values and pitch change speed. Third, our classification model showed that the accuracy of MDD detection improved with an increase in voice recording duration up to approximately 200 seconds ( Figure S14 ). The AUC-ROC value of 0.80 suggests that these voice features, if combined with additional validated types, could lead to the development of tools providing meaningful diagnostic information about possible cases of MDD. Fourth, our research is the first to distinguish between state- and trait-related voice features associated with MDD. We address the question, are we measuring a depressed state or a trait that is vulnerable to depression? This distinction is crucial, as it suggests that some features might be more reflective of a person’s underlying propensity towards developing MDD (trait biomarkers), while others could be more indicative of a current depressive state (state biomarkers). In our study, the IQR1-3, the kurtosis, and the maximum of ΔF0 showed significant heritability and genetic association with MDD. These findings suggested that people vulnerable to MDD may exhibit different pitch dynamics in speech. Specifically, individuals with an increased genetic risk of MDD may have a smaller value of speed for the fastest pitch change, thus being unable to speak as fast as those without depression. They may show a narrower IQR of pitch change speeds and more frequently occurring extreme changes of pitch (higher kurtosis). Notably, two heritable voice features were also associated with the number of stressful life events. The reason for these associations is unclear, but suggests the possibility that stressful life events reveal a latent predisposition to depression 38,39 , evidenced through a change in vocal features. Fifth, the decision to use spontaneous speech aligns with the goal of capturing naturalistic voice features that are not limited to specific content or situations, thereby increasing the generalizability of the findings. However, this freedom in content introduces additional complexity to our analysis. It leaves a possibility that the associations we detected might be driven by the variations in context, which intrigued different emotional valences and word choices. To exclude this potential bias, we performed a context-constrained analysis. By selecting voice responses to simple demographic questions we avoided context difference. The limited voice duration and small sample size reduce the power to detect a signal and we expected only to observe that the sign of the beta coefficients would be consistent with our prior analyses. However, some effects persisted (e.g., IQR1-3 of ΔF0), even with an average segment length of 3.37 seconds. 12 of the 16 features consistently showed the same direction of effect, demonstrating that the signals we found show a statistical relation with MDD status that goes beyond context and cannot be explained by biases related to the different interview structures between cases and controls. Our results should be assessed with respect to several limitations. First, we only recruited Han Chinese women with recurrent MDD. Our results may not extrapolate to men, those with single episode MDD, or non-Mandarin speakers. Second, we only analyzed pitch features. Investigations into other features in future studies are warranted. Third, replication is needed for trait/state findings. Fourth, there may be changes in speech content and word frequency between MDD cases and controls in general, although our context-constrained analysis demonstrates that the signals we found are persistent across speech content, we cannot separate pitch differences due to word choice from pitch differences due to emotional content without additional experiments directly controlling the linguistic context. In conclusion, we found robust associations between several voice features and MDD. The associated features depicted, in MDD, a slower change in pitch and a particularly uneven distribution of these variations. Features measuring the variability and extremity of pitch change speed were heritable and had genetic correlations with MDD, which highlights their potential use as trait biomarkers for early MDD detection and secondary prevention in Mandarin speakers. There were also non-heritable features associated with the depressive state. Our hope is that these findings will further encourage efforts to assess changes in the voice, long understood by experienced clinicians to be a valuable sign, returning it to a more central position in clinical and research work on MDD. Methods Participants This study is part of the CONVERGE 32 study. Cases of recurrent MDD were recruited from 58 provincial mental health centers and psychiatric departments of general medical hospitals in 45 cities and 23 provinces of China. They were aged between 30-60, with ≥ two episodes of MDD that met the DSM-IV criteria 5 , with the first episode occurring between ages 14-50. Cases were excluded for pre-existing bipolar disorder, nonaffective psychosis, smoking/nicotine dependence (alcohol and substance abuse were virtually absent in this study, so it was not assessed), or mental retardation. Control subjects, screened to exclude a history of MDD, were recruited from patients undergoing minor surgical procedures at general hospitals and individuals attending local community centers. The replication study 40 maintained the same inclusion/exclusion criteria as the CONVERGE study and recruited samples from different hospitals in China, with a final sample size of 1,189 ( Figure 1 and Table S6 ). This study was approved by the Ethical Review Board of Oxford University (Oxford Tropical Research Ethics Committee), Ethics Committee of Bio-x Center, Shanghai Jiao Tong University (M16033), and local hospital review boards. All participants provided written informed consent. Data Collection All subjects went through a semi-structured interview using a computerized assessment system as outlined previously 40 and described in Supplementary . Recordings were not standardized and varied in quality and content. For this study, all recordings were listened to, and segments that contained only the patient’s voice at an adequate quality for the analyses (see Supplementary for details) were identified. All participants provided DNA samples for genetic analysis. Details of DNA sequencing and genotype imputation have been previously reported 32 and described in Supplementary . Covariates The covariates were five demographic variables and two recording quality variables. The demographic variables were age, education level, occupation, marital status, and social class. The recording quality referred to noise level and accent. The noise level and accent label were determined subjectively by the listeners during the process of identifying the patients’ voice segments. The noise was categorized into four levels: 1) No noise; 2) Slight noise but the subject's speech was clear; 3) Noise present but the content of the subject's speech could be clearly heard; 4) High noise levels and unclear speech. During the preprocessing, samples were excluded if they had a very high noise level that made the listeners unable to understand any of the speech content (worse than level 4). Samples included in the analysis (the 7,654 samples in Figure 1 ) all retain the noise level category. The accent was a binary label that indicated whether the subjects’ speech was in standard Mandarin or not. Voice features Our decision to analyze prosodic features of speech, particularly pitch (F0) and change in pitch (ΔF0), prompted the choice of the INTERSPEECH 2016 Computational Paralinguistics Evaluation 35,36 , a well-documented 41 and standardized method that ensures reproducibility. F0 refers to the lowest frequency of a periodic waveform in speech, often perceived as the 'pitch' of the voice 41 . A lower F0 indicates a deeper voice, while reduced variability in F0 reflects a more monotonous tone 15 . The feature set, primarily used for depression detection 24,40,42–45 , captures both temporal and long-term speech information through static utterance level statistics and dynamic ΔF0 coefficients 46,47 . For example, mean, maximum/minimum, quartiles, and kurtosis of the F0 describe the range and distribution of pitch, while the same statistics of the ΔF0 describe the patterns of pitch dynamics (change speed). Calculations were implemented in the openSMILE python package v2.4.2 48 and described in Supplementary . Given that many of the features were highly correlated (for example, the arithmetic and root-quadratic mean of F0, as shown in Figure S 16 ), we removed redundant features (described in Supplementary ), resulting in a set of 15 F0-based features and 15 ΔF0-based features. We provide in Supplemental Table S 3 technical definitions of the 30 features used, along with non-technical explanations of what each feature measures. Two-stage meta-analysis We utilized a two-stage meta-analysis methodology. In the CONVERGE cohort, we selected hospitals with at least 100 individuals and a case/control ratio of at least 1:9 and up to 9:1 for inclusion in the meta-analysis. This selection process resulted in a total of 27 hospitals and a total subject count of N = 5,681 ( Figure 1 and Table S 5 ). At stage 1, for each hospital, a linear regression model was fitted for each F0-related feature as the dependent variable using MDD and covariates ( Table S1 ) as the predictor variables. We applied rank-based inverse normal transformation to the voice features and age. At stage 2, beta coefficients for MDD and standard errors from stage 1 were pooled using random-effects meta-analyses 49 , assuming that the true effect sizes in different sites are not exactly the same but are drawn from a distribution of effect sizes. P-values were FDR-adjusted 50 . To validate our findings, we performed a replication analysis. All four hospitals from the replication cohort with sample sizes ≥ 100 were selected (N=1,084, Table S 6 and Figure 1 ). We performed the same procedure as in the two-stage meta-analysis above. Heritability, Genetic Correlation, and genome wide association study (GWAS) Heritability estimation, genetic correlation, and GWAS were performed on the 7,654 subjects in CONVERGE ( Figure 1 ). The SNP-based heritability used a generalized REML (restricted maximum likelihood) method implemented in LDAK 51 . We applied rank-based inverse normal transformation to the voice features and incorporated the above covariates and 20 genetic PCs. P-values were FDR-adjusted. For heritable voice features, we estimated their genetic correlation with MDD, adjusting for these same covariates and 20 genetic PCs. The genetic correlation was calculated through a bivariate GREML analysis implemented in GCTA 52,53 . P-values were FDR-adjusted based on the number of heritable voice features. We performed GWAS for each one of the heritable voice features adjusting for the covariates and 20 genetic PCs. A genetic relationship matrix (GRM), constructed from the genotype data, was utilized to correct for relatedness among the samples. GWAS was implemented in LDAK 51 . Associations between pitch features and MDD state To identify state biomarkers for MDD, subjects in the replication cohort took part in a 16-item, self-administered questionnaire assessing the severity of depression-related symptoms on a five-point distress scale over the past 30 days (subscales for depression in symptom checklist, SCL) 37 . We also used the same two-stage meta-analysis method to estimate the association between the 16 voice features and SCL scores. All four hospitals from the replication cohort with sample sizes ≥ 100 were selected (N=1,084, Figure 1 ). At stage 1, for each hospital, a linear regression model was fitted for each pitch feature as the dependent variable using SCL scores and the covariates as the predictor variables. At stage 2, beta coefficients for SCL and standard errors from stage 1 were pooled using random-effects meta-analyses 49 . SCL scores were standardized using rank-based inverse normal transformation. P-values were FDR-adjusted. Classification Model To assess the classification performance using the identified voice features in predicting MDD, we constructed a logistic regression model. The training dataset was the CONVERGE data, with the replication data serving as the test dataset. To prevent data leakage from the test set, we incorporated all 20 voice features identified as associated with MDD during the discovery stage, including those not replicated. To provide a baseline for comparison, we also trained a null model using only the covariates to predict MDD. The model’s performance was evaluated in terms of sensitivity, specificity, AUC-ROC, and AUC-PR. To compare the full model (voice + covariates) with the null model (covariates), we also calculated the net reclassification improvement setting the threshold at 0.5. To determine the minimum length of voice recording necessary for reliable MDD detection, we extracted voice features from segments with varying limited durations (ranging from 10 to 300 seconds) and evaluated the model’s performance on these truncated segments. If a segment’s total duration exceeded the set limit, it was truncated; if less, it remained as is. This ensured that the sample size remained consistent in the training and test groups regardless of segment duration. Associations Between Pitch Features and MDD Symptoms, Risk Factors, and Comorbidities. We examined the relationship between the 30 voice F0/ΔF0 features with 33 variables related to MDD, including eight risk factor variables, 11 comorbidity variables, seven symptoms, three variables about suicidality, age of onset, number of MDD episodes, neuroticism, and premenstrual syndrome score. The risk factors we considered included stressful life events, child sexual abuse, and six parenting variables from the parental bonding instrument. These risk factors were previously reported 31,54,55 . The comorbidities included panic, generalized anxiety disorder, dysthymia, melancholia, post-natal depression, and phobia (including general phobia and five specific types of phobia: agoraphobia, social phobia, animal phobia, situational phobia, and blood phobia) 56 . In terms of symptoms, we focused on seven out of the nine criteria for MDD as defined by the DSM 5 . We were unable to test for the symptoms of sad mood and loss of interest due to their high endorsement rate in our recurrent MDD sample. We also evaluated suicidal thoughts, plans, and attempts. It's important to note that these symptoms were assessed based on the participants' recall of their worst episode of MDD, not their current state. We again employed the two-stage meta-analysis procedure, selecting hospitals with a case number ≥ 60. At stage 1, for each hospital, a multivariate linear regression model was fitted for each pitch feature as the dependent variable using one of the above variables and covariates as the independent variables. At stage 2, we used the Q statistics to measure the heterogeneity of the pooled beta coefficients and standard errors. If the heterogeneity test is nominal significant (P<0.05), we then used the random-effects model for meta-analyses, otherwise we used fixed-effects 50 . As the total number of association tests was a lot larger, we use a more conservative multi-testing correction method, the Bonferroni method. Due to the high endorsement rates for certain variables within some hospitals, the hospitals included in the meta-analysis varied depending on the variable being analyzed (For example, if all cases from one hospital did not have suicidal attempts, this hospital would be excluded for the analysis of suicidal attempts at stage 1.). We reported the number of hospitals and sample size for each association along with the meta-results. Context-Constrained Analysis First, we counted the total number of available voice segments for each question in the replication cohort. Then, we selected the two most frequently answered questions from the demographic section of the interview: D2.A (“What is your date of birth?”) and D10 (“How much do you weigh while wearing indoor clothing?”). For each question, we used the corresponding segments to extract the 16 voice features that were associated with MDD in our previous analysis. Finally, we re-assessed the associations between these voice features and MDD through the two-stage meta-analysis method. The limited voice duration and small sample size remarkably reduced the power to detect a significant signal for each one of the voice features. We then applied the one-sided binomial sign test to determine whether the number of voice features demonstrating consistent directions of association effects between the concatenated segments and the context-constrained segments was greater than expected by chance (that is, a one-sided test of whether this fraction is greater than 0.5). The rationale behind this test is as follows: Null Hypothesis : If the association effects observed in our main analysis were predominantly biased by the differences in the interview structure between cases and controls. Then, by selecting voice responses to the same questions, we would expect the effects to be gone and the direction of the association effects in this context-constrained analysis to be random. This randomness would mean that the direction of effects (whether positive or negative) would essentially be a 50-50 chance, showing no consistent pattern with the main analysis. Alternative Hypothesis : Conversely, if the context-constrained analysis reveals a consistent direction in the association effects that significantly exceeds random chance (i.e., more than 50% of the features show the same direction of association at a significance level of P<0.05), it would suggest that our original findings are not merely artifacts of the interview structure. Instead, this outcome would indicate a genuine link between the voice features and MDD that transcends the specifics of how the interview was conducted. Declarations Data Availability The GWAS summary statistics of the voice pitch features associated with MDD are available at https://doi.org/10.6084/m9.figshare.24571321.v1. Due to the sensitive nature of the raw audio files and in adherence to privacy considerations, these files cannot be made publicly available. However, we are committed to facilitating scientific progress and transparency. Thus, secondary data derived from these audio files, specifically voice features, are available upon reasonable request. Researchers interested in accessing these data should contact the corresponding authors, Jonathan Flint at [email protected] . Disclosures : No conflict of interest. Acknowledgments : This work was funded by NIH grant MH-122596 References Kendler, K. S. The genealogy of major depression: symptoms and signs of melancholia from 1880 to 1900. Mol Psychiatry 22 , 1539–1553 (2017). Kendler, K. S. The Phenomenology of Major Depression and the Representativeness and Nature of DSM Criteria. AJP 173 , 771–780 (2016). American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, Third Edition. (American Psychiatric Association, Washington, D.C, 1980). American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, Revised Third Edition. (American Psychiatric Association, Washington, D.C, 1987). American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition. (American Psychiatric Association, Washington, D.C, 1994). American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders (DSM-5®) . (American Psychiatric Association, Washington, D.C, 2013). Lux, V. & Kendler, K. S. Deconstructing major depression: a validation study of the DSM-IV symptomatic criteria. Psychological Medicine 40 , 1679–1690 (2010). Hyman, S. E. Can neuroscience be integrated into the DSM-V? Nat Rev Neurosci 8 , 725–732 (2007). Kessler, R. C. et al. The Epidemiology of Major Depressive Disorder Results From the National Comorbidity Survey Replication (NCS-R). JAMA 289 , 3095–3105 (2003). Lu, J. et al. Prevalence of depressive disorders and treatment in China: a cross-sectional epidemiological study. The Lancet Psychiatry 8 , 981–990 (2021). Thornicroft, G. et al. Undertreatment of people with major depressive disorder in 21 countries. The British Journal of Psychiatry 210 , 119–124 (2017). Guislain, J. Orales sur Les Phrénopathies, ou Traitê Thêorique Et Pratique Des Maladies Mentales: Cours Donné A La Clinique Des Êtablissements D’Aliénés A Gand . vol. 1 (Paris, & Bonn,: Gand, 1852). Kraepelin, E. Manic-Depressive Insanity and Paranoia . (E. & S. Livingstone, Edinburgh, 1921). Sobin, C. Psychomotor Symptoms of Depression. A m J Psychiatry 15 (1997). Cummins, N. et al. A review of depression and suicide risk assessment using speech analysis. Speech Communication 71 , 10–49 (2015). Low, D. M., Bentley, K. H. & Ghosh, S. S. Automated assessment of psychiatric disorders using speech: A systematic review. Laryngoscope Investigative Otolaryngology 5 , 96–116 (2020). Nilsonne, Å. Acoustic analysis of speech variables during depression and after improvement. Acta Psychiatrica Scandinavica 76 , 235–245 (1987). Nilsonne, Å. Speech characteristics as indicators of depressive illness. Acta Psychiatrica Scandinavica 77 , 253–263 (1988). Mundt, J. C., Snyder, P. J., Cannizzaro, M. S., Chappie, K. & Geralts, D. S. Voice acoustic measures of depression severity and treatment response collected via interactive voice response (IVR) technology. Journal of Neurolinguistics 20 , 50–64 (2007). Mundt, J. C., Vogel, A. P., Feltner, D. E. & Lenderking, W. R. Vocal Acoustic Biomarkers of Depression Severity and Treatment Response. Biological Psychiatry 72 , 580–587 (2012). Kuny, St. & Stassen, H. H. Speaking behavior and voice sound characteristics in depressive patients during recovery. Journal of Psychiatric Research 27 , 289–307 (1993). Cannizzaro, M., Harel, B., Reilly, N., Chappell, P. & Snyder, P. J. Voice acoustical measurement of the severity of major depression. Brain and Cognition 56 , 30–35 (2004). Alpert, M., Pouget, E. R. & Silva, R. R. Reflections of depression in acoustic measures of the patient’s speech. Journal of Affective Disorders 66 , 59–69 (2001). Pan, W. et al. Re-examining the robustness of voice features in predicting depression: Compared with baseline of confounders. PLoS ONE 14 , e0218172 (2019). Wang, J. et al. Acoustic differences between healthy and depressed people: a cross-situation study. BMC Psychiatry 19 , 300 (2019). Schultebraucks, K., Yadav, V., Shalev, A. Y., Bonanno, G. A. & Galatzer-Levy, I. R. Deep learning-based classification of posttraumatic stress disorder and depression following trauma utilizing visual and auditory markers of arousal and mood. Psychol. Med. 52 , 957–967 (2022). Di, Y., Wang, J., Liu, X. & Zhu, T. Combining Polygenic Risk Score and Voice Features to Detect Major Depressive Disorders. Frontiers in Genetics 12 , 2451 (2021). Flint, J. The genetic basis of major depressive disorder. Mol Psychiatry (2023) doi:10.1038/s41380-023-01957-9. Hasler, G., Drevets, W. C., Manji, H. K. & Charney, D. S. Discovering Endophenotypes for Major Depression. Neuropsychopharmacol 29 , 1765–1781 (2004). Kendler, K. S., Aggen, S. H. & Neale, M. C. Evidence for multiple genetic factors underlying DSM-IV criteria for major depression. JAMA psychiatry 70 , 599–607 (2013). Peterson, R. E. et al. Molecular Genetic Analysis Subdivided by Adversity Exposure Suggests Etiologic Heterogeneity in Major Depression. AJP 175 , 545–554 (2018). CONVERGE consortium. Sparse whole-genome sequencing identifies two loci for major depressive disorder. Nature 523 , 588–591 (2015). Andrianopoulos, M. V., Darrow, K. N. & Chen, J. Multimodal Standardization of Voice Among Four Multicultural Populations: Fundamental Frequency and Spectral Characteristics. Journal of Voice 15 , 194–219 (2001). Kendler, K. S., Gardner, C., Neale, M. & Prescott, C. Genetic risk factors for major depression in men and women: similar or different heritabilities and same or partly distinct genes? Psychological medicine 31 , 605 (2001). Schuller, B. et al. The INTERSPEECH 2016 Computational Paralinguistics Challenge: Deception, Sincerity & Native Language. in Interspeech 2016 2001–2005 (ISCA, 2016). doi:10.21437/Interspeech.2016-129. Weninger, F., Eyben, F., Schuller, B. W., Mortillaro, M. & Scherer, K. R. On the Acoustics of Emotion in Audio: What Speech, Music, and Sound have in Common. Front. Psychol. 4 , (2013). Derogatis, L. R. SCL-90: an outpatient psychiatric rating scale-preliminary report. Psychopharmacol Bull 9 , 13–28 (1973). Kendler, K. S. & Karkowski-Shuman, L. Stressful life events and genetic liability to major depression: genetic control of exposure to the environment? Psychological Medicine 27 , 539–547 (1997). Kendler, K. S. et al. Stressful Life Events, Genetic Liability, and Onset of an Episode of Major Depression in Women. FOC 8 , 459–470 (2010). Di, Y., Wang, J., Li, W. & Zhu, T. Using i-vectors from voice features to identify major depressive disorder. Journal of Affective Disorders 288 , 161–166 (2021). Eyben, F. Real-Time Speech and Music Classification by Large Audio Feature Space Extraction . (Springer, 2015). Afshan, A. et al. Effectiveness of Voice Quality Features in Detecting Depression. Interspeech 2018 (2018) doi:10.21437/Interspeech.2018-1399. Alghowinem, S., Goecke, R., Epps, J., Wagner, M. & Cohn, J. Cross-Cultural Depression Recognition from Vocal Biomarkers. in Interspeech 2016 1943–1947 (ISCA, 2016). doi:10.21437/Interspeech.2016-1339. Quatieri, T. F. & Malyska, N. Vocal-source biomarkers for depression: a link to psychomotor activity. in Interspeech 2012 1059–1062 (ISCA, 2012). doi:10.21437/Interspeech.2012-311. Syed, Z. S., Schroeter, J., Sidorov, K. & Marshall, D. Computational Paralinguistics: Automatic Assessment of Emotions, Mood and Behavioural State from Acoustics of Speech. in Interspeech 2018 511–515 (ISCA, 2018). doi:10.21437/Interspeech.2018-2019. Schuller, B., Batliner, A., Steidl, S. & Seppi, D. Recognising realistic emotions and affect in speech: State of the art and lessons learnt from the first challenge. Speech Communication 53 , 1062–1087 (2011). Schuller, B. et al. Paralinguistics in speech and language—State-of-the-art and the challenge. Computer Speech & Language 27 , 4–39 (2013). Eyben, F., Wöllmer, M. & Schuller, B. Opensmile: the munich versatile and fast open-source audio feature extractor. in Proceedings of the 18th ACM international conference on Multimedia 1459–1462 (ACM, Firenze Italy, 2010). doi:10.1145/1873951.1874246. Viechtbauer, W. Conducting Meta-Analyses in R with the metafor Package. Journal of Statistical Software 36 , 1–48 (2010). Benjamini, Y. & Hochberg, Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society. Series B (Methodological) 57 , 289–300 (1995). Speed, D., Cai, N., Johnson, M. R., Nejentsev, S. & Balding, D. J. Reevaluation of SNP heritability in complex human traits. Nat Genet 49 , 986–992 (2017). Lee, S. H., Yang, J., Goddard, M. E., Visscher, P. M. & Wray, N. R. Estimation of pleiotropy between complex diseases using single-nucleotide polymorphism-derived genomic relationships and restricted maximum likelihood. Bioinformatics 28 , 2540–2542 (2012). Yang, J., Lee, S. H., Goddard, M. E. & Visscher, P. M. GCTA: A Tool for Genome-wide Complex Trait Analysis. The American Journal of Human Genetics 88 , 76–82 (2011). Tao, M. et al. Examining the relationship between lifetime stressful life events and the onset of major depression in Chinese women. Journal of Affective Disorders 135 , 95–99 (2011). Gao, J. et al. Perceived parenting and risk for major depression in Chinese women. Psychol. Med. 42 , 921–930 (2012). Yang, F. et al. Age at onset of major depressive disorder in Han Chinese women: Relationship with clinical features and family history. Journal of Affective Disorders 135 , 89–94 (2011). Tables Table 1 Association of voice pitch features and MDD. The associations between pitch features and MDD were estimated in the discovery (CONVERGE) sample and the replication sample, using two-stage meta-analysis method. P-values are FDR corrected. The uncorrected P-values are in Table S5. LPC means linear prediction coding. Based Statistical Functionals CONVERGE Replication Beta SE P_FDR Beta SE P_FDR ΔF0 Interquartile range (3rd -1st) -1.070 0.071 1.06E-49 -1.085 0.206 1.46E-06 ΔF0 Maximum (99th percentile) -0.971 0.070 1.78E-42 -0.973 0.214 2.68E-05 F0 Time with F0>90th percentile -0.804 0.060 1.49E-40 -0.779 0.193 1.34E-04 ΔF0 Kurtosis 0.868 0.066 1.11E-38 0.828 0.227 5.37E-04 ΔF0 Root quadratic mean -0.690 0.060 3.80E-30 -0.762 0.211 5.44E-04 ΔF0 3rd quartile -0.781 0.069 2.71E-29 -0.883 0.201 4.64E-05 ΔF0 Position of the minimum -0.602 0.058 1.43E-24 -0.574 0.116 4.70E-06 ΔF0 Mean 0.432 0.047 2.00E-19 0.555 0.102 1.17E-06 ΔF0 Slope of linear regression -0.394 0.043 2.00E-19 -0.553 0.187 4.16E-03 F0 Range 0.566 0.062 3.10E-19 0.483 0.159 3.35E-03 ΔF0 Position of the maximum -0.551 0.061 4.92E-19 -0.614 0.192 2.27E-03 ΔF0 1st quadratic regression coefficient 0.384 0.045 2.59E-17 0.473 0.122 2.31E-04 ΔF0 Flatness* -0.427 0.057 1.27E-13 -0.392 0.094 8.13E-05 F0 2nd LPC coefficient 0.326 0.054 2.70E-09 0.408 0.096 7.39E-05 F0 4th LPC coefficient -0.284 0.048 9.79E-09 -0.248 0.154 0.13 F0 Kurtosis 0.225 0.051 2.03E-05 0.295 0.094 2.78E-03 F0 0th LPC coefficient -0.291 0.074 1.59E-04 -0.385 0.165 0.02 ΔF0 Time with which ΔF0 is rising -0.152 0.043 6.86E-04 -0.248 0.168 0.15 F0 Maximum length of voiced segments 0.194 0.071 9.96E-03 0.117 0.106 0.28 F0 Root quadratic mean 0.143 0.055 0.01 0.000 0.126 1.00 F0 Position of the maximum 0.067 0.043 0.17 ΔF0 Offset of linear regression 0.084 0.055 0.18 F0 Interquartile range (3rd -2nd) -0.063 0.043 0.19 F0 Position of the minimum 0.074 0.053 0.21 F0 Proportion of time with which F0 is rising -0.080 0.059 0.22 F0 3rd quadratic regression coefficient 0.063 0.065 0.38 ΔF0 3rd LPC coefficient -0.028 0.043 0.55 F0 Slope of linear regression 0.029 0.044 0.55 ΔF0 0th LPC coefficient 0.002 0.046 0.97 F0 Minimum (1st percentile) 0.002 0.052 0.97 Table 2. Heritable voice pitch features and their genetic correlation with MDD. Base Statistical Functionals SNP heritability Genetic Correlation h 2 95% CI rg 95% CI P P_FDR ΔF0 Interquartile range (3rd -1st) 0.171 (0.071, 0.272) -0.45 (-0.77, -0.13) 0.03 0.04 Maximum (99th percentile) 0.121 (0.022, 0.221) -0.7 (-1.28, -0.11) 4.17E-05 0.0002 Kurtosis 0.125 (0.025, 0.225) 0.55 (0.23, 0.88) 0.01 0.02 F0 Kurtosis 0.134 (0.035, 0.234) -0.3 (-1.14, 0.54) 0.2 0.2 Table 3. Voice pitch features associated with SCL scores. The associations between pitch features and SCL scores were estimated in the replication sample using two-stage meta-analysis method. LPC means linear prediction coding. Based Feature Statistical Functionals Beta SE P P_FDR ΔF0 Position of the maximum -0.222 0.040 3.42E-08 5.47E-07 Maximum (99th percentile) -0.296 0.074 6.67E-05 5.33E-04 Root quadratic mean -0.238 0.067 3.58E-04 0.001 3rd quartile -0.286 0.081 3.93E-04 0.001 Interquartile range (3rd -1st) -0.309 0.097 0.001 0.004 Kurtosis 0.225 0.081 0.005 0.01 Mean 0.133 0.053 0.01 0.02 Flatness -0.116 0.056 0.04 0.05 Quadratic regression coefficients 1 0.104 0.062 0.09 0.12 Slope of linear regression -0.103 0.067 0.12 0.14 Position of the minimum -0.130 0.086 0.13 0.14 F0 Time of F0>90th percentile -0.246 0.066 1.90E-04 0.001 LPC coefficient 2 0.118 0.042 0.005 0.01 Kurtosis 0.097 0.041 0.02 0.03 Range 0.128 0.057 0.02 0.03 LPC coefficient 0 -0.047 0.104 0.65 0.65 Additional Declarations There is NO Competing Interest. Supplementary Files Supplementary.NatMH.031924.docx Supplemental Text and Figures SupplementalTables.NatMH.031924jf.xlsx Supplemental Tables 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-4135145","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":286809196,"identity":"dc27375a-085a-4645-b265-3af74a0cb942","order_by":0,"name":"Jonathan Flint","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7UlEQVRIiWNgGAWjYBAC9vmPGw4wMNgAmQkQET4QwYNHC+MMRqCWhDSIFqBmBjZitABVHyZNS+Phwh/no/nZcw8+/sBwWJ6N/QDjg7dteLTMb2w4PCPhdu7MnnfJBgcYDhu28SQwG87FpwXosMM8QC0bbuSYSQC1JLAxJLBJ8xLWci53/40c8x9gLfwP2H/j0yII0XIgd4NEjhkDWItEAhszPi3SEiAtacm5M868MZY4Y5Bu2CbxsFlyzjncWvgkmA9/5rGxy+1vzzH8UFFhLc/Pn3zww5sy3FrQgAHYfw1Eqx8Fo2AUjIJRgB0AACO8UuRpqJSOAAAAAElFTkSuQmCC","orcid":"","institution":"UCLA","correspondingAuthor":true,"prefix":"","firstName":"Jonathan","middleName":"","lastName":"Flint","suffix":""},{"id":286809197,"identity":"a3b90e42-e42d-4e24-b8cd-501222689386","order_by":1,"name":"Yazheng Di","email":"","orcid":"https://orcid.org/0000-0003-2483-1696","institution":"Insitute of Psychology Chinese Academy of Science","correspondingAuthor":false,"prefix":"","firstName":"Yazheng","middleName":"","lastName":"Di","suffix":""},{"id":286809198,"identity":"164a176d-b7b8-4d40-96a6-3444f97bd5d6","order_by":2,"name":"Elior Rahmani","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Elior","middleName":"","lastName":"Rahmani","suffix":""},{"id":286809199,"identity":"68c37717-2f08-4011-8b6b-77ad6966ed5b","order_by":3,"name":"Joel Mefford","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Joel","middleName":"","lastName":"Mefford","suffix":""},{"id":286809200,"identity":"529e841a-7eae-410b-a2e9-ae9dc3aaeac0","order_by":4,"name":"Jinhan Wang","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Jinhan","middleName":"","lastName":"Wang","suffix":""},{"id":286809201,"identity":"271974f1-8782-4e9d-ac94-820b0673108f","order_by":5,"name":"Vijay Ravi","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Vijay","middleName":"","lastName":"Ravi","suffix":""},{"id":286809202,"identity":"015bde9c-8e28-4b59-b990-0c222ed55e53","order_by":6,"name":"Aditya Gorla","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Aditya","middleName":"","lastName":"Gorla","suffix":""},{"id":286809203,"identity":"100a3cd5-36cf-4c94-9ba2-88c4b750395e","order_by":7,"name":"Abeer Alwan","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Abeer","middleName":"","lastName":"Alwan","suffix":""},{"id":286809204,"identity":"3bb940c5-4af9-490f-8694-da0ddb8db0b0","order_by":8,"name":"Kenneth Kendler","email":"","orcid":"","institution":"Virginia Commonwealth University","correspondingAuthor":false,"prefix":"","firstName":"Kenneth","middleName":"","lastName":"Kendler","suffix":""},{"id":286809205,"identity":"901bf6fe-2fe5-4be3-ad01-5c9b48897785","order_by":9,"name":"Tingshao Zhu","email":"","orcid":"","institution":"Insitute of Psychology Chinese Academy of Science","correspondingAuthor":false,"prefix":"","firstName":"Tingshao","middleName":"","lastName":"Zhu","suffix":""}],"badges":[],"createdAt":"2024-03-20 07:56:57","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4135145/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4135145/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":54044144,"identity":"9f8f391a-ea9d-4f7e-887f-6e276b3e274e","added_by":"auto","created_at":"2024-04-03 18:38:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":197492,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverview of the study aims, analyses, and sample sizes.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study is based on the CONVERGE cohort as the discovery samples and the Replication as the validation samples. We first discovered the association between voice and lifetime-MDD in CONVERGE (Aim 1a) and replicated it (Aim 1b). We selected a subset of the samples because we used a two-stage meta-analysis method. Then we separated trait biomarkers (Aim 2a) from state biomarkers (Aim 2b) based on the established associations in Aim 1. Additionally, we looked into the association between voice and MDD symptoms, risk factors, and comorbidities within cases (Additional Analysis). Finally, we performed post-hoc analysis following Aim 1 by re-estimating the associations on the voice speech with constrained context.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4135145/v1/31873b91ab00c499656e2b6d.png"},{"id":56210432,"identity":"d42354dc-9839-4f10-86e1-a8e461602035","added_by":"auto","created_at":"2024-05-10 01:11:10","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1275472,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4135145/v1/2c53b2d9-1909-41da-a4cd-c38c5c4ebe3f.pdf"},{"id":54044146,"identity":"2d30a113-b502-4601-a5b0-5344b3d6c302","added_by":"auto","created_at":"2024-04-03 18:38:15","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1276312,"visible":true,"origin":"","legend":"Supplemental Text and Figures","description":"","filename":"Supplementary.NatMH.031924.docx","url":"https://assets-eu.researchsquare.com/files/rs-4135145/v1/8060bf9a474a22d54a66e616.docx"},{"id":54044145,"identity":"2144f483-6b14-47cc-a56a-098e3631b5b2","added_by":"auto","created_at":"2024-04-03 18:38:15","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":224295,"visible":true,"origin":"","legend":"Supplemental Tables","description":"","filename":"SupplementalTables.NatMH.031924jf.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4135145/v1/732ad95ac396409b52340a5d.xlsx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Unraveling the Associations Between Voice Pitch and Major Depressive Disorder: A Multisite Genetic Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eChanges in human pitch and tone of speech have been noted as an important sign in depression for over a century\u003csup\u003e1,2\u003c/sup\u003e. Although not contained in symptomatic criteria for major depressive disorder (MDD) in DSM-III\u003csup\u003e3\u003c/sup\u003e, DSM-IIIR\u003csup\u003e4\u003c/sup\u003e, DSM-IV\u003csup\u003e5\u003c/sup\u003e, or DSM-5\u003csup\u003e6\u003c/sup\u003e, they are found in 26 out of 28 detailed clinical descriptions of melancholia published from 1880-1900\u003csup\u003e1\u003c/sup\u003e and in 19 out of 21 of such descriptions of depression published in the 20th century\u003csup\u003e2\u003c/sup\u003e. Given the current challenges in diagnosing MDD\u003csup\u003e2,7,8\u003c/sup\u003e, where a large proportion of cases (ranging from 50% to 90%) remain untreated\u003csup\u003e9\u0026ndash;11\u003c/sup\u003e, the transformation of voice phenomena into diagnostic biomarkers could aid in both clinical and research arenas.\u003c/p\u003e\n\u003cp\u003eClinical observations describe the speech patterns of depressed patients as slow, weak, low-pitched, and monotonous\u003csup\u003e1,2,12,13\u003c/sup\u003e. These phenomena are typically quantified by increased pause time, lower volume, lower pitch, and reduced pitch variability\u003csup\u003e14\u0026ndash;16\u003c/sup\u003e. Many studies have sought to develop features from pitch as a biomarker for depression\u003csup\u003e17\u0026ndash;26\u003c/sup\u003e, but none have to date achieved sufficient accuracy and precision for clinical utility. The large number of both vocal features and confounds\u003csup\u003e24,25\u003c/sup\u003e imposes a multi-testing burden that requires large sample sizes which few studies have obtained\u003csup\u003e16\u003c/sup\u003e. The critical distinction between state and trait effects on voice has also never been addressed\u003csup\u003e27\u0026ndash;29\u003c/sup\u003e. Furthermore, MDD is likely heterogeneous\u003csup\u003e30,31\u003c/sup\u003e: studies not accounting for this may be underpowered and/or identify effects that do not generalize well\u003csup\u003e16\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eIn this paper, we report voice analysis of a large case-control study of MDD and replicate findings in an independent sample, both from China. By including genetic data, we determine differences in voice that reflect either state or trait. Our results return vocal features to a more central position in clinical and research work on MDD.\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eWe analyzed voice records from interviews conducted as part of the CONVERGE\u003csup\u003e32\u003c/sup\u003e (China, Oxford, and VCU Experimental Research on Genetic Epidemiology) study, a multi-site MDD case-control genetic investigation. CONVERGE recruited 11,670 Han Chinese women through a collaboration involving 58 hospitals in China. The initial and final sample sizes used in each following analysis are presented in \u003cstrong\u003eFigure 1\u003c/strong\u003e. By recruiting only women, we increase homogeneity in both genetic and vocal signals\u003csup\u003e33,34\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eWe identified 364,929 voice segments from 7,654 subjects. Summary information is shown in \u003cstrong\u003eTable S1.\u003c/strong\u003e 60% of the subjects spoke in standard Mandarin, whereas the rest spoke either local languages or Mandarin with local accents. The mean duration of concatenated voice segments per subject was 297.75 seconds (SD=208.86) for cases and 97.44 seconds (SD=122.96) for controls. We additionally recruited samples for replication purpose with a final sample size of 1,189. Their summary information is in \u003cstrong\u003eTable S2\u003c/strong\u003e.\u003c/p\u003e\n\u003ch2\u003eAssociation Between Pitch Features and Lifetime MDD\u003c/h2\u003e\n\u003cp\u003eWe extracted 30 voice features\u0026nbsp;(\u003cstrong\u003eTable S3\u003c/strong\u003e), providing a comprehensive characterization of the speaker\u0026rsquo;s prosodic patterns, including pitch and intonation\u003csup\u003e35,36\u003c/sup\u003e. Specifically, we calculated statistics and functions based on the time series of Fundamental Frequency (F0) and its differential values, namely pitch change speed (\u0026Delta;F0). These statistics and functions include mean values, quartiles, range, and regression coefficients, capturing the pitch trends and pitch dynamics in speech.\u003c/p\u003e\n\u003cp\u003eIn an initial analysis, we looked for associations between each pitch features and MDD for the entire 7,654 subjects, adjusting for five demographic variables and ignoring hospital stratification. This naive approach generated many highly significant results (\u003cstrong\u003eTable S4\u003c/strong\u003e), which, as we could see from a quantile-quantile plot (\u003cstrong\u003eFigure S1\u003c/strong\u003e), were most likely due to inflated P-values. We attributed this inflation to the correlation between MDD and hospitals. Indeed, the case/control ratios were uneven across different hospitals (as shown in \u003cstrong\u003eTable S\u003c/strong\u003e\u003cstrong\u003e5\u0026nbsp;\u003c/strong\u003eand \u003cstrong\u003eTable S\u003c/strong\u003e\u003cstrong\u003e6\u003c/strong\u003e), a factor that could inflate the results. For example, if cases were mainly recruited from Shanghai (Southern China) and controls were mainly from Beijing (Northern China), the voice features indicating a Shanghai accent would be associated with MDD in the na\u0026iuml;ve analysis.\u003c/p\u003e\n\u003cp\u003eTo account for variations in location and/or hospital, we implemented a two-stage meta-analysis, in which associations were first calculated at the hospital level and subsequently pooled using a random-effects model. From the 30 pitch features, we identified 20 features significantly associated with MDD at FDR\u0026lt;0.05 (\u003cstrong\u003eTable 1\u003c/strong\u003e). Using this approach, we found that our results were well-calibrated (QQ Plot in \u003cstrong\u003eFigure S2\u003c/strong\u003e). As a test of sensitivity of our result to genetic differences between cases and controls, we compared analyses with and without the inclusion of 20 genetic principal components (PCs). Results (shown in \u003cstrong\u003eTable S7\u003c/strong\u003e) are almost identical, showing no overall decrease in significance with the inclusion of the genetic covariates.\u003c/p\u003e\n\u003cp\u003eWe then tried to replicate the identified associations in the replication cohort, using the same two-stage meta-analysis methodology. Out of the 20 features, 16 associations were significantly\u0026nbsp;replicated at FDR\u0026lt;0.05\u0026nbsp;(\u003cstrong\u003eTable 2\u003c/strong\u003e). Notably, 11 of these features were derivatives of the \u0026Delta;F0 series, or in other words, they were various characterizations of the intonations (speed of pitch change). The distributions of these 11 \u0026Delta;F0 features are shown in \u003cstrong\u003eFigure S3\u003c/strong\u003e and the other five F0 features in \u003cstrong\u003eFigure S4\u003c/strong\u003e.\u003c/p\u003e\n\u003ch2\u003eGenetic Correlations Between Pitch Features and MDD\u003c/h2\u003e\n\u003cp\u003eAfter establishing the association between the voice features and MDD, our next goal was to differentiate trait-related from state-related acoustic measures. We started by estimating the SNP-based heritability for each of the 16 pitch features and we found that four were heritable at FDR\u0026lt;0.05 (see \u003cstrong\u003eTable 2\u003c/strong\u003e for heritable features and \u003cstrong\u003eTable S8\u003c/strong\u003e for results of all 16 features). We then estimated the genetic correlation with MDD for the four features and found that three \u0026Delta;F0 features had significant genetic correlations (\u003cstrong\u003eTable 2\u003c/strong\u003e). They were: 1) the interquartile range (IQR1-3), quantifying the variation of speed in pitch change; 2) the kurtosis, signaling the extremity of speed in pitch change; and 3) the maximum, representing the speed of the fastest pitch change. We performed heritability estimates adjusting for more genetic PCs as a sensitivity test. The heritability of the voice features remained significant even after adjusting for as many as 60 genetic PCs (\u003cstrong\u003eTable S9\u003c/strong\u003e). We also performed GWAS for these heritable voice features, but we did not find any significant hits, presumably owing to the limited sample size (\u003cstrong\u003eFigures S5-S12\u003c/strong\u003e and \u003cstrong\u003eTable S10).\u003c/strong\u003e\u003c/p\u003e\n\u003ch2\u003eAssociation Between Pitch Features and Current MDD State\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eWe obtained self-assessed current mood for subjects in the replication cohort. Subjects filled in the depressive symptom checklist (SCL)\u003csup\u003e37\u003c/sup\u003e, providing a more immediate assessment of their depressive state. The distributions of SCL scores for cases and controls are presented in \u003cstrong\u003eFigure S13\u003c/strong\u003e. There is an overlap in scores for cases and controls, indicating that many cases were likely not in an episode of MDD at the time of the interview (due to the design which focused on lifetime rather than current MDD episodes).\u003c/p\u003e\n\u003cp\u003eOf the 16 pitch features, 11 showed a significant association with current depressive symptoms after FDR correction (\u003cstrong\u003eTable 3\u003c/strong\u003e), including heritable and non-heritable features. We noticed one non-heritable feature significantly associated with current depressive symptoms, the position of the maximum \u0026Delta;F0 (max\u0026Delta;F0_Pos), which reflected the time point when the pitch changes in its fastest speech. We asked whether max\u0026Delta;F0_Pos was specifically predictive of current depressive symptoms, using a case-only design to answer this question. We found that within cases the association with SCL scores was -0.15 (SE = 0.07, P = 0.028, N=326), using the two-stage meta-analysis method.\u003c/p\u003e\n\u003ch2\u003eClassification Performance\u003c/h2\u003e\n\u003cp\u003eWe assessed the classification performance for MDD based on a logistic regression model trained on the identified voice features and covariates, compared with a null model using only the covariates. The null model had a sensitivity of 0.66, and a specificity of 0.59, an AUC-ROC (area under the receiver operating characteristic curve) of 0.70, and an AUC-PR (area under the precision recall curve) of 0.64 in detecting MDD. The full model achieved a sensitivity of 0.74, a specificity of 0.73, an AUC-ROC of 0.80, an AUC-PR of 0.75, and a net reclassification improvement of 0.11. We also tested the classification performances on voice recordings of different lengths, to determine the minimum length of voice necessary for reliable MDD detection. The performance on limited segment durations is shown in \u003cstrong\u003eFigure S14\u003c/strong\u003e. The metrics improve noticeably until about 150 seconds. Around 200 seconds, AUC-ROC and AUC-PR show a slight increase, but sensitivity and specificity exhibit a plateau over a span of 50 seconds.\u003c/p\u003e\n\u003ch2\u003eAssociations Between Pitch Features and MDD Symptoms, Risk Factors, and Comorbidities.\u003c/h2\u003e\n\u003cp\u003eConsidering the heterogeneous nature of MDD, we tested for associations between the 30 pitch features and 33 clinical features obtained from cases, including MDD symptoms, environmental risk factors, comorbidities, and suicidality. The results of this within-case two-stage meta-analysis are shown in \u003cstrong\u003eTable S11\u003c/strong\u003e. Six associations were significant after Bonferroni correction (P\u0026lt;0.05/990=), out of which two were significantly replicated (P\u0026lt;0.05/6=). Both associations were between the total number of stressful life events and \u0026Delta;F0 features, including the IQR1-3 of \u0026Delta;F0 (16 hospitals in CONVERGE, total N=2,064, ; 4 hospitals in the replication, total N=295, ) and maximum of \u0026Delta;F0 (16 hospitals in CONVERGE, total N=2,064, ; 4 hospitals in the replication, total N=295, ).\u003c/p\u003e\n\u003ch2\u003eContext-Constrained Analysis\u003c/h2\u003e\n\u003cp\u003eFor all the above analyses, the voice features were extracted from the concatenated segments of spontaneous speech. The speech content and the conversational context were not controlled, as our goal was to identify voice patterns that are persistent and detectable across various speech contexts and build generalizable biomarkers useful in real-world settings. Recognizing that voice acoustic features and their associations with MDD might be sensitive to the context of the interview, we conducted a sensitivity analysis on voice responses to specific questions to check if the effects remain consistent across contexts.\u003c/p\u003e\n\u003cp\u003eWe selected two questions from the demographic section of the interview based on their high response rates (\u003cstrong\u003eTable S12\u003c/strong\u003e) and neutral nature. These questions\u0026mdash;D2.A (\u0026ldquo;What is your date of birth?\u0026rdquo;) and D10 (\u0026ldquo;How much do you weigh while wearing indoor clothing?\u0026rdquo;)\u0026mdash;were chosen because they are unlikely to trigger emotional differences between MDD cases and controls, thus serving as a stable basis for comparison. We identified 533 subjects with voice response to question D2.A and 617 to question D10. The average segment durations were 3.37 seconds (SD=3.05), and 8.47 seconds (SD=2.69), respectively. For each question, we used the corresponding segments to extract the 16 pitch features that were associated with MDD in our main analysis. Using the two-stage meta-analysis method again, we re-estimated their associations and used a one-sided binomial sign test to test consistency in the direction of association effects between the main analysis and analysis here (that is, a one-sided test of whether this fraction is greater than 0.5, see method for our hypothesis).\u003c/p\u003e\n\u003cp\u003eThe estimated association effects in context-constrained analysis are reported in \u003cstrong\u003eTable S13\u003c/strong\u003e. We found that for question D10, three out of 16 pitch features maintained significant associations with MDD at FDR\u0026lt;0.05. Remarkably, 15 out of 16 features showed the same direction of association effects, a fraction significantly higher than chance (Binomial P= 0.00026). For D2.A, despite the average duration being a mere 3.37 seconds, three features achieved nominal significance for associations (uncorrected\u0026nbsp;P\u0026lt;0.05), and 12 out of 16 pitch features showed consistent directions of association effects (Binomial P= 0.038). In total, 12 out of 16 voice features showed consistent directions of association effects across all four analyses (\u003cstrong\u003eFigure S15\u0026nbsp;\u003c/strong\u003eand summarized in \u003cstrong\u003eTable S1\u003c/strong\u003e\u003cstrong\u003e4\u003c/strong\u003e). We conclude that the findings from the main analyses are not biased by the context of the interview.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe set out to find voice pitch features associated with MDD. By using a large and homogeneous case-control cohort, we provided robust evidence that certain pitch features distinguished MDD cases from matched controls. These associations were further validated by replication. More importantly, three heritable features showed a significant genetic correlation with MDD, and a group of non-heritable features were associated with current depressive symptom severity (as summarized in \u003cstrong\u003eTable S14\u003c/strong\u003e). Our study advances research efforts to find voice biomarkers of MDD by combining the power of a two-stage meta-analysis with a large, diverse sample collected from numerous hospitals. This rigorous methodology helps to account for potential confounding factors and paves the way for more reliable and generalizable findings.\u003c/p\u003e\n\u003cp\u003eOur findings support and extend previous studies which have indicated potential links between pitch patterns and MDD, but were limited by smaller sample sizes or more heterogeneous cohorts \u003csup\u003e15,16\u003c/sup\u003e. First, our large sample size provided adequate power to test several pitch features from a standardized features set, providing more fine-grained quantitative evidence for the descriptions of the monotonous speech pattern in MDD than in previous studies. Previous studies have found that depressed people speak more slowly with lower pitch and decreased variability\u003csup\u003e16,25\u003c/sup\u003e. Here, our study showed MDD was negatively associated with features measuring how fast pitch changes (the maximum, the 3rd quartile, and the root quadratic mean of \u0026Delta;F0, \u003cstrong\u003eTable 1\u003c/strong\u003e), indicating that the reduced rate of change in pitch is a characteristic of voice in MDD patients. We also found that MDD patients spend less time in their upper vocal range (Time with F0\u0026gt;90\u003csup\u003eth\u003c/sup\u003e percentile, \u003cstrong\u003eTable 1\u003c/strong\u003e), affirming the \u0026ldquo;low-pitched\u0026rdquo; pattern.\u003c/p\u003e\n\u003cp\u003eSecond, our results indicate that MDD\u0026apos;s pitch dynamics involve more than reduced variability, showing a broader pitch range and more extreme values (range and kurtosis of F0, \u003cstrong\u003eTable 1\u003c/strong\u003e). Our analysis also revealed an uneven distribution of the speed with which an MDD patient\u0026rsquo;s pitch changes, as shown by the negative association between MDD and the flatness of \u0026Delta;F0 and the positive association with the kurtosis of \u0026Delta;F0 (\u003cstrong\u003eTable 1\u003c/strong\u003e). Overall, these various features enrich our understanding of pitch dynamics, demonstrating a pattern of slower change in pitch, yet with more frequent occurrences of extreme values and pitch change speed.\u003c/p\u003e\n\u003cp\u003eThird, our classification model showed that the accuracy of MDD detection improved with an increase in voice recording duration up to approximately 200 seconds (\u003cstrong\u003eFigure S14\u003c/strong\u003e). The AUC-ROC value of 0.80 suggests that these voice features, if combined with additional validated types, could lead to the development of tools providing meaningful diagnostic information about possible cases of MDD.\u003c/p\u003e\n\u003cp\u003eFourth, our research is the first to distinguish between state- and trait-related voice features associated with MDD. We address the question, are we measuring a depressed state or a trait that is vulnerable to depression? This distinction is crucial, as it suggests that some features might be more reflective of a person\u0026rsquo;s underlying propensity towards developing MDD (trait biomarkers), while others could be more indicative of a current depressive state (state biomarkers).\u003c/p\u003e\n\u003cp\u003eIn our study, the IQR1-3, the kurtosis, and the maximum of \u0026Delta;F0 showed significant heritability and genetic association with MDD. These findings suggested that people vulnerable to MDD may exhibit different pitch dynamics in speech. Specifically, individuals with an increased genetic risk of MDD may have a smaller value of speed for the fastest pitch change, thus being unable to speak as fast as those without depression. They may show a narrower IQR of pitch change speeds and more frequently occurring extreme changes of pitch (higher kurtosis). Notably, two heritable\u0026nbsp;voice features were also associated with the number of stressful life events. The reason for these associations is unclear, but suggests the possibility that stressful life events reveal a latent predisposition to depression\u003csup\u003e38,39\u003c/sup\u003e, evidenced through a change in vocal features.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFifth, the decision to use spontaneous speech aligns with the goal of capturing naturalistic voice features that are not limited to specific content or situations, thereby increasing the generalizability of the findings. However, this freedom in content introduces additional complexity to our analysis. It leaves a possibility that the associations we detected might be driven by the variations in context, which intrigued different emotional valences and word choices. To exclude this potential bias, we performed a context-constrained analysis. By selecting voice responses to simple demographic questions we avoided context difference. The limited voice duration and small sample size reduce the power to detect a signal and we expected only to observe that the sign of the beta coefficients would be consistent with our prior analyses. However, some effects persisted (e.g., IQR1-3 of \u0026Delta;F0), even with an average segment length of 3.37 seconds. 12 of the 16 features consistently showed the same direction of effect, demonstrating that the signals we found show a statistical relation with MDD status that goes beyond context and cannot be explained by biases related to the different interview structures between cases and controls.\u003c/p\u003e\n\u003cp\u003eOur results should be assessed with respect to several limitations. First, we only recruited Han Chinese women with recurrent MDD. Our results may not extrapolate to men, those with single episode MDD, or non-Mandarin speakers. Second, we only analyzed pitch features. Investigations into other features in future studies are warranted. Third, replication is needed for trait/state findings. Fourth, there may be changes in speech content and word frequency between MDD cases and controls in general, although our context-constrained analysis demonstrates that the signals we found are persistent across speech content, we cannot separate pitch differences due to word choice from pitch differences due to emotional content without additional experiments directly controlling the linguistic context.\u003c/p\u003e\n\u003cp\u003eIn conclusion, we found robust associations between several voice features and MDD. The associated features depicted, in MDD, a slower change in pitch and a particularly uneven distribution of these variations. Features measuring the variability and extremity of pitch change speed were heritable and had genetic correlations with MDD, which highlights their potential use as trait biomarkers for early MDD detection and secondary prevention in Mandarin speakers. There were also non-heritable features associated with the depressive state. Our hope is that these findings will further encourage efforts to assess changes in the voice, long understood by experienced clinicians to be a valuable sign, returning it to a more central position in clinical and research work on MDD.\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003eParticipants\u003c/h2\u003e\n\u003cp\u003eThis study is part of the CONVERGE\u003csup\u003e32\u003c/sup\u003e study. Cases of recurrent MDD were recruited from 58 provincial mental health centers and psychiatric departments of general medical hospitals in 45 cities and 23 provinces of China. They were aged between 30-60, with \u0026ge; two episodes of MDD that met the DSM-IV criteria\u003csup\u003e5\u003c/sup\u003e, with the first episode occurring between ages 14-50. Cases were\u0026nbsp;excluded for pre-existing bipolar disorder, nonaffective psychosis, smoking/nicotine dependence (alcohol and substance abuse were virtually absent in this study, so it was not assessed), or mental retardation. Control subjects, screened to exclude a history of MDD, were recruited from patients undergoing minor surgical procedures at general hospitals and individuals attending local community centers.\u003c/p\u003e\n\u003cp\u003eThe replication study\u003csup\u003e40\u003c/sup\u003e maintained the same inclusion/exclusion criteria as the CONVERGE study and recruited samples from different hospitals in China, with a final sample size of 1,189 (\u003cstrong\u003eFigure 1\u0026nbsp;\u003c/strong\u003eand \u003cstrong\u003eTable S6\u003c/strong\u003e). This study was approved by the Ethical Review Board of Oxford University (Oxford Tropical Research Ethics Committee), Ethics Committee of Bio-x Center, Shanghai Jiao Tong University (M16033), and local hospital review boards. All participants provided written informed consent.\u003c/p\u003e\n\u003ch2\u003eData Collection\u003c/h2\u003e\n\u003cp\u003eAll subjects went through a semi-structured interview using a computerized assessment system as outlined previously\u003csup\u003e40\u003c/sup\u003e and described in \u003cstrong\u003eSupplementary\u003c/strong\u003e. Recordings were not standardized and varied in quality and content. For this study, all recordings were listened to, and segments that contained only the patient\u0026rsquo;s voice at an adequate quality for the analyses (see \u003cstrong\u003eSupplementary\u003c/strong\u003e for details) were identified. All participants provided DNA samples for genetic analysis. Details of DNA sequencing and genotype imputation have been previously reported\u003csup\u003e32\u003c/sup\u003e and described in \u003cstrong\u003eSupplementary\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCovariates\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe covariates were five demographic variables and two recording quality variables. The demographic variables were age, education level, occupation, marital status, and social class. The recording quality referred to noise level and accent. The noise level and accent label were determined subjectively by the listeners during the process of identifying the patients\u0026rsquo; voice segments. The noise was categorized into four levels: 1) No noise; 2) Slight noise but the subject\u0026apos;s speech was clear; 3) Noise present but the content of the subject\u0026apos;s speech could be clearly heard; 4) High noise levels and unclear speech. During the preprocessing, samples were excluded if they had a very high noise level that made the listeners unable to understand any of the speech content (worse than level 4). Samples included in the analysis (the 7,654 samples in \u003cstrong\u003eFigure 1\u003c/strong\u003e) all retain the noise level category. The accent was a binary label that indicated whether the subjects\u0026rsquo; speech was in standard Mandarin or not.\u003c/p\u003e\n\u003ch2\u003eVoice features\u003c/h2\u003e\n\u003cp\u003eOur decision to analyze prosodic features of speech, particularly pitch (F0) and change in pitch (\u0026Delta;F0), prompted the choice of the INTERSPEECH 2016 Computational Paralinguistics Evaluation\u003csup\u003e35,36\u003c/sup\u003e, a well-documented\u003csup\u003e41\u003c/sup\u003e and standardized method that ensures reproducibility. F0 refers to the lowest frequency of a periodic waveform in speech, often perceived as the \u0026apos;pitch\u0026apos; of the voice\u003csup\u003e41\u003c/sup\u003e. A lower F0 indicates a deeper voice, while reduced variability in F0 reflects a more monotonous tone\u003csup\u003e15\u003c/sup\u003e. The feature set, primarily used for depression detection\u003csup\u003e24,40,42\u0026ndash;45\u003c/sup\u003e, captures both temporal and long-term speech information through static utterance level statistics and dynamic \u0026Delta;F0 coefficients\u003csup\u003e46,47\u003c/sup\u003e. For example, mean, maximum/minimum, quartiles, and kurtosis of the F0 describe the range and distribution of pitch, while the same statistics of the \u0026Delta;F0 describe the patterns of pitch dynamics (change speed).\u003c/p\u003e\n\u003cp\u003eCalculations were implemented in the openSMILE python package v2.4.2\u003csup\u003e48\u003c/sup\u003e and described in \u003cstrong\u003eSupplementary\u003c/strong\u003e. Given that many of the features were highly correlated (for example, the arithmetic and root-quadratic mean of F0, as shown in \u003cstrong\u003eFigure S\u003c/strong\u003e\u003cstrong\u003e16\u003c/strong\u003e), we removed redundant features (described in \u003cstrong\u003eSupplementary\u003c/strong\u003e), resulting in a set of 15 F0-based features and 15 \u0026Delta;F0-based features. We provide in Supplemental \u003cstrong\u003eTable S\u003c/strong\u003e\u003cstrong\u003e3\u003c/strong\u003e technical definitions of the 30 features used, along with non-technical explanations of what each feature measures.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTwo-stage meta-analysis\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWe utilized a two-stage meta-analysis methodology. In the CONVERGE cohort, we selected hospitals with at least 100 individuals and a case/control ratio of at least 1:9 and up to 9:1 for inclusion in the meta-analysis. This selection process resulted in a total of 27 hospitals and a total subject count of N = 5,681 (\u003cstrong\u003eFigure 1\u003c/strong\u003e and \u003cstrong\u003eTable S\u003c/strong\u003e\u003cstrong\u003e5\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eAt stage 1, for each hospital, a linear regression model was fitted for each F0-related feature as the dependent variable using MDD and covariates (\u003cstrong\u003eTable S1\u003c/strong\u003e) as the predictor variables. We applied rank-based inverse normal transformation to the voice features and age. At stage 2, beta coefficients for MDD and standard errors from stage 1 were pooled using random-effects meta-analyses\u003csup\u003e49\u003c/sup\u003e, assuming that the true effect sizes in different sites are not exactly the same but are drawn from a distribution of effect sizes. P-values were FDR-adjusted\u003csup\u003e50\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eTo validate our findings, we performed a replication analysis. All four hospitals from the replication cohort with sample sizes \u0026ge; 100 were selected (N=1,084, \u003cstrong\u003eTable S\u003c/strong\u003e\u003cstrong\u003e6\u003c/strong\u003e and \u003cstrong\u003eFigure 1\u003c/strong\u003e). We performed the same procedure as in the two-stage meta-analysis above.\u003c/p\u003e\n\u003ch2\u003eHeritability, Genetic Correlation, and genome wide association study (GWAS)\u003c/h2\u003e\n\u003cp\u003eHeritability estimation, genetic correlation, and GWAS were performed on the 7,654 subjects in CONVERGE (\u003cstrong\u003eFigure 1\u003c/strong\u003e). The SNP-based heritability used a generalized REML (restricted maximum likelihood) method implemented in LDAK\u003csup\u003e51\u003c/sup\u003e. We applied rank-based inverse normal transformation to the voice features and incorporated the above covariates and 20 genetic PCs. P-values were FDR-adjusted. \u0026nbsp;For heritable voice features, we estimated their genetic correlation with MDD, adjusting for these same covariates and 20 genetic PCs. The genetic correlation was calculated through a bivariate GREML analysis implemented in GCTA\u003csup\u003e52,53\u003c/sup\u003e. P-values were FDR-adjusted based on the number of heritable voice features. We performed GWAS for each one of the heritable voice features adjusting for the covariates and 20 genetic PCs. A genetic relationship matrix (GRM), constructed from the genotype\u0026nbsp;data, was utilized to correct for relatedness among the samples. GWAS was implemented in LDAK\u003csup\u003e51\u003c/sup\u003e.\u003c/p\u003e\n\u003ch2\u003eAssociations between pitch features and MDD state\u003c/h2\u003e\n\u003cp\u003eTo identify state biomarkers for MDD, subjects in the replication cohort took part in a 16-item, self-administered questionnaire assessing the severity of depression-related symptoms on a five-point distress scale over the past 30 days (subscales for depression in symptom checklist, SCL)\u003csup\u003e37\u003c/sup\u003e. We also used the same two-stage meta-analysis method to estimate the association between the 16 voice features and SCL scores. All four hospitals from the replication cohort with sample sizes \u0026ge; 100 were selected (N=1,084, \u003cstrong\u003eFigure 1\u003c/strong\u003e). At stage 1, for each hospital, a linear regression model was fitted for each pitch\u0026nbsp;feature as the dependent variable using SCL scores\u0026nbsp;and the\u0026nbsp;covariates as the predictor variables.\u0026nbsp;At stage 2, beta coefficients for SCL\u0026nbsp;and standard errors from stage 1 were pooled using random-effects meta-analyses\u003csup\u003e49\u003c/sup\u003e. SCL scores were standardized using rank-based inverse normal transformation. P-values were FDR-adjusted.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eClassification Model\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTo assess the classification performance using the identified voice features in predicting MDD, we constructed a logistic regression model. The training dataset was the CONVERGE data, with the replication data serving as the test dataset. To prevent data leakage from the test set, we incorporated all 20 voice features identified as associated with MDD during the discovery stage, including those not replicated. To provide a baseline for comparison, we also trained a null model using only the covariates to predict MDD. The model\u0026rsquo;s performance was evaluated in terms of sensitivity, specificity, AUC-ROC, and AUC-PR. To compare the full model (voice + covariates) with the null model (covariates), we also calculated the net reclassification improvement setting the threshold at 0.5.\u003c/p\u003e\n\u003cp\u003eTo determine the minimum length of voice recording necessary for reliable MDD detection, we extracted voice features from segments with varying limited durations (ranging from 10 to 300 seconds) and evaluated the model\u0026rsquo;s performance on these truncated segments. If a segment\u0026rsquo;s total duration exceeded the set limit, it was truncated; if less, it remained as is. This ensured that the sample size remained consistent in the training and test groups regardless of segment duration.\u003c/p\u003e\n\u003ch2\u003eAssociations Between Pitch Features and MDD Symptoms, Risk Factors, and Comorbidities.\u003c/h2\u003e\n\u003cp\u003eWe examined the relationship between the 30 voice F0/\u0026Delta;F0 features with 33 variables related to MDD, including eight risk factor variables, 11 comorbidity variables, seven symptoms, three variables about suicidality, age of onset, number of MDD episodes, neuroticism, and premenstrual syndrome score. The risk factors we considered included stressful life events, child sexual abuse, and six parenting variables from the parental bonding instrument. These risk factors were previously reported\u003csup\u003e31,54,55\u003c/sup\u003e. The comorbidities included panic, generalized anxiety disorder, dysthymia, melancholia, post-natal depression, and phobia (including general phobia and five specific types of phobia: agoraphobia, social phobia, animal phobia, situational phobia, and blood phobia)\u003csup\u003e56\u003c/sup\u003e. In terms of symptoms, we focused on seven out of the nine criteria for MDD as defined by the DSM\u003csup\u003e5\u003c/sup\u003e. We were unable to test for the symptoms of sad mood and loss of interest due to their high endorsement rate in our recurrent MDD sample. We also evaluated suicidal thoughts, plans, and attempts. It\u0026apos;s important to note that these symptoms were assessed based on the participants\u0026apos; recall of their worst episode of MDD, not their current state.\u003c/p\u003e\n\u003cp\u003eWe again employed the two-stage meta-analysis procedure, selecting hospitals with a case number \u0026ge; 60. At stage 1, for each hospital, a multivariate linear regression model was fitted for each\u0026nbsp;pitch\u0026nbsp;feature as the dependent variable using one of the above variables and covariates as the independent variables. At stage 2, we used the Q statistics\u0026nbsp;to measure the heterogeneity of the pooled beta coefficients and standard errors. If the heterogeneity test is nominal significant (P\u0026lt;0.05), we then used the random-effects model for meta-analyses, otherwise we used fixed-effects\u003csup\u003e50\u003c/sup\u003e. As the total number of association tests was a lot larger, we use a more conservative multi-testing correction method, the Bonferroni method.\u003c/p\u003e\n\u003cp\u003eDue to the high endorsement rates for certain variables within some hospitals, the hospitals included in the meta-analysis varied depending on the variable being analyzed (For example, if all cases from one hospital did not have suicidal attempts, this hospital would be excluded for the analysis of suicidal attempts at stage 1.). We reported the number of hospitals and sample size for each association along with the meta-results.\u003c/p\u003e\n\u003ch2\u003eContext-Constrained Analysis\u003c/h2\u003e\n\u003cp\u003eFirst, we counted the total number of available voice segments for each question in the replication cohort.\u0026nbsp;Then, we selected the two most frequently answered questions from the demographic section of the interview: D2.A (\u0026ldquo;What is your date of birth?\u0026rdquo;) and D10 (\u0026ldquo;How much do you weigh while wearing indoor clothing?\u0026rdquo;). For each question, we used the corresponding segments to extract the 16 voice features that were associated with MDD in our previous analysis. Finally, we re-assessed the associations between these voice features and MDD through the two-stage meta-analysis method.\u003c/p\u003e\n\u003cp\u003eThe limited voice duration and small sample size remarkably reduced the power to detect a significant signal for each one of the voice features. We then applied the one-sided binomial sign test to determine whether the number of voice features demonstrating consistent directions of association effects between the concatenated segments and the context-constrained segments was greater than expected by chance (that is, a one-sided test of whether this fraction is greater than 0.5). The rationale behind this test is as follows:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNull Hypothesis\u003c/strong\u003e: If the association effects observed in our main analysis were predominantly biased by the differences in the interview structure between cases and controls.\u0026nbsp;Then, by selecting voice responses to the same questions, we would expect the effects to be gone and the direction of the association effects in this context-constrained analysis to be random. This randomness would mean that the direction of effects (whether positive or negative) would essentially be a 50-50 chance, showing no consistent pattern with the main analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAlternative Hypothesis\u003c/strong\u003e: Conversely, if the context-constrained analysis reveals a consistent direction in the association effects that significantly exceeds random chance (i.e., more than 50% of the features show the same direction of association at a significance level of P\u0026lt;0.05), it would suggest that our original findings are not merely artifacts of the interview structure. Instead, this outcome would indicate a genuine link between the voice features and MDD that transcends the specifics of how the interview was conducted.\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe GWAS summary statistics of the voice pitch features associated with MDD are available at https://doi.org/10.6084/m9.figshare.24571321.v1. Due to the sensitive nature of the raw audio files and in adherence to privacy considerations, these files cannot be made publicly available. However, we are committed to facilitating scientific progress and transparency. Thus, secondary data derived from these audio files, specifically voice features, are available upon reasonable request. Researchers interested in accessing these data should contact the corresponding authors, Jonathan Flint at [email protected].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eDisclosures\u003c/strong\u003e:\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eNo conflict of interest. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eAcknowledgments\u0026nbsp;\u003c/strong\u003e:\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis work was funded by NIH grant MH-122596\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKendler, K. S. The genealogy of major depression: symptoms and signs of melancholia from 1880 to 1900. \u003cem\u003eMol Psychiatry\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 1539\u0026ndash;1553 (2017).\u003c/li\u003e\n\u003cli\u003eKendler, K. S. The Phenomenology of Major Depression and the Representativeness and Nature of DSM Criteria. \u003cem\u003eAJP\u003c/em\u003e \u003cstrong\u003e173\u003c/strong\u003e, 771\u0026ndash;780 (2016).\u003c/li\u003e\n\u003cli\u003eAmerican Psychiatric Association. \u003cem\u003eDiagnostic and Statistical Manual of Mental Disorders, Third Edition.\u003c/em\u003e (American Psychiatric Association, Washington, D.C, 1980).\u003c/li\u003e\n\u003cli\u003eAmerican Psychiatric Association. \u003cem\u003eDiagnostic and Statistical Manual of Mental Disorders, Revised Third Edition.\u003c/em\u003e (American Psychiatric Association, Washington, D.C, 1987).\u003c/li\u003e\n\u003cli\u003eAmerican Psychiatric Association. \u003cem\u003eDiagnostic and Statistical Manual of Mental Disorders, Fourth Edition.\u003c/em\u003e (American Psychiatric Association, Washington, D.C, 1994).\u003c/li\u003e\n\u003cli\u003eAmerican Psychiatric Association. \u003cem\u003eDiagnostic and Statistical Manual of Mental Disorders (DSM-5\u0026reg;)\u003c/em\u003e. (American Psychiatric Association, Washington, D.C, 2013).\u003c/li\u003e\n\u003cli\u003eLux, V. \u0026amp; Kendler, K. S. Deconstructing major depression: a validation study of the DSM-IV symptomatic criteria. \u003cem\u003ePsychological Medicine\u003c/em\u003e \u003cstrong\u003e40\u003c/strong\u003e, 1679\u0026ndash;1690 (2010).\u003c/li\u003e\n\u003cli\u003eHyman, S. E. Can neuroscience be integrated into the DSM-V? \u003cem\u003eNat Rev Neurosci\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 725\u0026ndash;732 (2007).\u003c/li\u003e\n\u003cli\u003eKessler, R. C. \u003cem\u003eet al.\u003c/em\u003e The Epidemiology of Major Depressive Disorder Results From the National Comorbidity Survey Replication (NCS-R). \u003cem\u003eJAMA\u003c/em\u003e \u003cstrong\u003e289\u003c/strong\u003e, 3095\u0026ndash;3105 (2003).\u003c/li\u003e\n\u003cli\u003eLu, J. \u003cem\u003eet al.\u003c/em\u003e Prevalence of depressive disorders and treatment in China: a cross-sectional epidemiological study. \u003cem\u003eThe Lancet Psychiatry\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 981\u0026ndash;990 (2021).\u003c/li\u003e\n\u003cli\u003eThornicroft, G. \u003cem\u003eet al.\u003c/em\u003e Undertreatment of people with major depressive disorder in 21 countries. \u003cem\u003eThe British Journal of Psychiatry\u003c/em\u003e \u003cstrong\u003e210\u003c/strong\u003e, 119\u0026ndash;124 (2017).\u003c/li\u003e\n\u003cli\u003eGuislain, J. \u003cem\u003eOrales sur Les Phr\u0026eacute;nopathies, ou Trait\u0026ecirc; Th\u0026ecirc;orique Et Pratique Des Maladies Mentales: Cours Donn\u0026eacute; A La Clinique Des \u0026Ecirc;tablissements D\u0026rsquo;Ali\u0026eacute;n\u0026eacute;s A Gand\u003c/em\u003e. vol. 1 (Paris, \u0026amp; Bonn,: Gand, 1852).\u003c/li\u003e\n\u003cli\u003eKraepelin, E. \u003cem\u003eManic-Depressive Insanity and Paranoia\u003c/em\u003e. (E. \u0026amp; S. Livingstone, Edinburgh, 1921).\u003c/li\u003e\n\u003cli\u003eSobin, C. Psychomotor Symptoms of Depression. \u003cem\u003eA m J Psychiatry\u003c/em\u003e 15 (1997).\u003c/li\u003e\n\u003cli\u003eCummins, N. \u003cem\u003eet al.\u003c/em\u003e A review of depression and suicide risk assessment using speech analysis. \u003cem\u003eSpeech Communication\u003c/em\u003e \u003cstrong\u003e71\u003c/strong\u003e, 10\u0026ndash;49 (2015).\u003c/li\u003e\n\u003cli\u003eLow, D. M., Bentley, K. H. \u0026amp; Ghosh, S. S. Automated assessment of psychiatric disorders using speech: A systematic review. \u003cem\u003eLaryngoscope Investigative Otolaryngology\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 96\u0026ndash;116 (2020).\u003c/li\u003e\n\u003cli\u003eNilsonne, \u0026Aring;. Acoustic analysis of speech variables during depression and after improvement. \u003cem\u003eActa Psychiatrica Scandinavica\u003c/em\u003e \u003cstrong\u003e76\u003c/strong\u003e, 235\u0026ndash;245 (1987).\u003c/li\u003e\n\u003cli\u003eNilsonne, \u0026Aring;. Speech characteristics as indicators of depressive illness. \u003cem\u003eActa Psychiatrica Scandinavica\u003c/em\u003e \u003cstrong\u003e77\u003c/strong\u003e, 253\u0026ndash;263 (1988).\u003c/li\u003e\n\u003cli\u003eMundt, J. C., Snyder, P. J., Cannizzaro, M. S., Chappie, K. \u0026amp; Geralts, D. S. Voice acoustic measures of depression severity and treatment response collected via interactive voice response (IVR) technology. \u003cem\u003eJournal of Neurolinguistics\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 50\u0026ndash;64 (2007).\u003c/li\u003e\n\u003cli\u003eMundt, J. C., Vogel, A. P., Feltner, D. E. \u0026amp; Lenderking, W. R. Vocal Acoustic Biomarkers of Depression Severity and Treatment Response. \u003cem\u003eBiological Psychiatry\u003c/em\u003e \u003cstrong\u003e72\u003c/strong\u003e, 580\u0026ndash;587 (2012).\u003c/li\u003e\n\u003cli\u003eKuny, St. \u0026amp; Stassen, H. H. Speaking behavior and voice sound characteristics in depressive patients during recovery. \u003cem\u003eJournal of Psychiatric Research\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 289\u0026ndash;307 (1993).\u003c/li\u003e\n\u003cli\u003eCannizzaro, M., Harel, B., Reilly, N., Chappell, P. \u0026amp; Snyder, P. J. Voice acoustical measurement of the severity of major depression. \u003cem\u003eBrain and Cognition\u003c/em\u003e \u003cstrong\u003e56\u003c/strong\u003e, 30\u0026ndash;35 (2004).\u003c/li\u003e\n\u003cli\u003eAlpert, M., Pouget, E. R. \u0026amp; Silva, R. R. Reflections of depression in acoustic measures of the patient\u0026rsquo;s speech. \u003cem\u003eJournal of Affective Disorders\u003c/em\u003e \u003cstrong\u003e66\u003c/strong\u003e, 59\u0026ndash;69 (2001).\u003c/li\u003e\n\u003cli\u003ePan, W. \u003cem\u003eet al.\u003c/em\u003e Re-examining the robustness of voice features in predicting depression: Compared with baseline of confounders. \u003cem\u003ePLoS ONE\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, e0218172 (2019).\u003c/li\u003e\n\u003cli\u003eWang, J. \u003cem\u003eet al.\u003c/em\u003e Acoustic differences between healthy and depressed people: a cross-situation study. \u003cem\u003eBMC Psychiatry\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 300 (2019).\u003c/li\u003e\n\u003cli\u003eSchultebraucks, K., Yadav, V., Shalev, A. Y., Bonanno, G. A. \u0026amp; Galatzer-Levy, I. R. Deep learning-based classification of posttraumatic stress disorder and depression following trauma utilizing visual and auditory markers of arousal and mood. \u003cem\u003ePsychol. Med.\u003c/em\u003e \u003cstrong\u003e52\u003c/strong\u003e, 957\u0026ndash;967 (2022).\u003c/li\u003e\n\u003cli\u003eDi, Y., Wang, J., Liu, X. \u0026amp; Zhu, T. Combining Polygenic Risk Score and Voice Features to Detect Major Depressive Disorders. \u003cem\u003eFrontiers in Genetics\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 2451 (2021).\u003c/li\u003e\n\u003cli\u003eFlint, J. The genetic basis of major depressive disorder. \u003cem\u003eMol Psychiatry\u003c/em\u003e (2023) doi:10.1038/s41380-023-01957-9.\u003c/li\u003e\n\u003cli\u003eHasler, G., Drevets, W. C., Manji, H. K. \u0026amp; Charney, D. S. Discovering Endophenotypes for Major Depression. \u003cem\u003eNeuropsychopharmacol\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 1765\u0026ndash;1781 (2004).\u003c/li\u003e\n\u003cli\u003eKendler, K. S., Aggen, S. H. \u0026amp; Neale, M. C. Evidence for multiple genetic factors underlying DSM-IV criteria for major depression. \u003cem\u003eJAMA psychiatry\u003c/em\u003e \u003cstrong\u003e70\u003c/strong\u003e, 599\u0026ndash;607 (2013).\u003c/li\u003e\n\u003cli\u003ePeterson, R. E. \u003cem\u003eet al.\u003c/em\u003e Molecular Genetic Analysis Subdivided by Adversity Exposure Suggests Etiologic Heterogeneity in Major Depression. \u003cem\u003eAJP\u003c/em\u003e \u003cstrong\u003e175\u003c/strong\u003e, 545\u0026ndash;554 (2018).\u003c/li\u003e\n\u003cli\u003eCONVERGE consortium. Sparse whole-genome sequencing identifies two loci for major depressive disorder. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e523\u003c/strong\u003e, 588\u0026ndash;591 (2015).\u003c/li\u003e\n\u003cli\u003eAndrianopoulos, M. V., Darrow, K. N. \u0026amp; Chen, J. Multimodal Standardization of Voice Among Four Multicultural Populations: Fundamental Frequency and Spectral Characteristics. \u003cem\u003eJournal of Voice\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 194\u0026ndash;219 (2001).\u003c/li\u003e\n\u003cli\u003eKendler, K. S., Gardner, C., Neale, M. \u0026amp; Prescott, C. Genetic risk factors for major depression in men and women: similar or different heritabilities and same or partly distinct genes? \u003cem\u003ePsychological medicine\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 605 (2001).\u003c/li\u003e\n\u003cli\u003eSchuller, B. \u003cem\u003eet al.\u003c/em\u003e The INTERSPEECH 2016 Computational Paralinguistics Challenge: Deception, Sincerity \u0026amp; Native Language. in \u003cem\u003eInterspeech 2016\u003c/em\u003e 2001\u0026ndash;2005 (ISCA, 2016). doi:10.21437/Interspeech.2016-129.\u003c/li\u003e\n\u003cli\u003eWeninger, F., Eyben, F., Schuller, B. W., Mortillaro, M. \u0026amp; Scherer, K. R. On the Acoustics of Emotion in Audio: What Speech, Music, and Sound have in Common. \u003cem\u003eFront. Psychol.\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, (2013).\u003c/li\u003e\n\u003cli\u003eDerogatis, L. R. SCL-90: an outpatient psychiatric rating scale-preliminary report. \u003cem\u003ePsychopharmacol Bull\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 13\u0026ndash;28 (1973).\u003c/li\u003e\n\u003cli\u003eKendler, K. S. \u0026amp; Karkowski-Shuman, L. Stressful life events and genetic liability to major depression: genetic control of exposure to the environment? \u003cem\u003ePsychological Medicine\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 539\u0026ndash;547 (1997).\u003c/li\u003e\n\u003cli\u003eKendler, K. S. \u003cem\u003eet al.\u003c/em\u003e Stressful Life Events, Genetic Liability, and Onset of an Episode of Major Depression in Women. \u003cem\u003eFOC\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 459\u0026ndash;470 (2010).\u003c/li\u003e\n\u003cli\u003eDi, Y., Wang, J., Li, W. \u0026amp; Zhu, T. Using i-vectors from voice features to identify major depressive disorder. \u003cem\u003eJournal of Affective Disorders\u003c/em\u003e \u003cstrong\u003e288\u003c/strong\u003e, 161\u0026ndash;166 (2021).\u003c/li\u003e\n\u003cli\u003eEyben, F. \u003cem\u003eReal-Time Speech and Music Classification by Large Audio Feature Space Extraction\u003c/em\u003e. (Springer, 2015).\u003c/li\u003e\n\u003cli\u003eAfshan, A. \u003cem\u003eet al.\u003c/em\u003e Effectiveness of Voice Quality Features in Detecting Depression. \u003cem\u003eInterspeech 2018\u003c/em\u003e (2018) doi:10.21437/Interspeech.2018-1399.\u003c/li\u003e\n\u003cli\u003eAlghowinem, S., Goecke, R., Epps, J., Wagner, M. \u0026amp; Cohn, J. Cross-Cultural Depression Recognition from Vocal Biomarkers. in \u003cem\u003eInterspeech 2016\u003c/em\u003e 1943\u0026ndash;1947 (ISCA, 2016). doi:10.21437/Interspeech.2016-1339.\u003c/li\u003e\n\u003cli\u003eQuatieri, T. F. \u0026amp; Malyska, N. Vocal-source biomarkers for depression: a link to psychomotor activity. in \u003cem\u003eInterspeech 2012\u003c/em\u003e 1059\u0026ndash;1062 (ISCA, 2012). doi:10.21437/Interspeech.2012-311.\u003c/li\u003e\n\u003cli\u003eSyed, Z. S., Schroeter, J., Sidorov, K. \u0026amp; Marshall, D. Computational Paralinguistics: Automatic Assessment of Emotions, Mood and Behavioural State from Acoustics of Speech. in \u003cem\u003eInterspeech 2018\u003c/em\u003e 511\u0026ndash;515 (ISCA, 2018). doi:10.21437/Interspeech.2018-2019.\u003c/li\u003e\n\u003cli\u003eSchuller, B., Batliner, A., Steidl, S. \u0026amp; Seppi, D. Recognising realistic emotions and affect in speech: State of the art and lessons learnt from the first challenge. \u003cem\u003eSpeech Communication\u003c/em\u003e \u003cstrong\u003e53\u003c/strong\u003e, 1062\u0026ndash;1087 (2011).\u003c/li\u003e\n\u003cli\u003eSchuller, B. \u003cem\u003eet al.\u003c/em\u003e Paralinguistics in speech and language\u0026mdash;State-of-the-art and the challenge. \u003cem\u003eComputer Speech \u0026amp; Language\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 4\u0026ndash;39 (2013).\u003c/li\u003e\n\u003cli\u003eEyben, F., W\u0026ouml;llmer, M. \u0026amp; Schuller, B. Opensmile: the munich versatile and fast open-source audio feature extractor. in \u003cem\u003eProceedings of the 18th ACM international conference on Multimedia\u003c/em\u003e 1459\u0026ndash;1462 (ACM, Firenze Italy, 2010). doi:10.1145/1873951.1874246.\u003c/li\u003e\n\u003cli\u003eViechtbauer, W. Conducting Meta-Analyses in R with the metafor Package. \u003cem\u003eJournal of Statistical Software\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 1\u0026ndash;48 (2010).\u003c/li\u003e\n\u003cli\u003eBenjamini, Y. \u0026amp; Hochberg, Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. \u003cem\u003eJournal of the Royal Statistical Society. Series B (Methodological)\u003c/em\u003e \u003cstrong\u003e57\u003c/strong\u003e, 289\u0026ndash;300 (1995).\u003c/li\u003e\n\u003cli\u003eSpeed, D., Cai, N., Johnson, M. R., Nejentsev, S. \u0026amp; Balding, D. J. Reevaluation of SNP heritability in complex human traits. \u003cem\u003eNat Genet\u003c/em\u003e \u003cstrong\u003e49\u003c/strong\u003e, 986\u0026ndash;992 (2017).\u003c/li\u003e\n\u003cli\u003eLee, S. H., Yang, J., Goddard, M. E., Visscher, P. M. \u0026amp; Wray, N. R. Estimation of pleiotropy between complex diseases using single-nucleotide polymorphism-derived genomic relationships and restricted maximum likelihood. \u003cem\u003eBioinformatics\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 2540\u0026ndash;2542 (2012).\u003c/li\u003e\n\u003cli\u003eYang, J., Lee, S. H., Goddard, M. E. \u0026amp; Visscher, P. M. GCTA: A Tool for Genome-wide Complex Trait Analysis. \u003cem\u003eThe American Journal of Human Genetics\u003c/em\u003e \u003cstrong\u003e88\u003c/strong\u003e, 76\u0026ndash;82 (2011).\u003c/li\u003e\n\u003cli\u003eTao, M. \u003cem\u003eet al.\u003c/em\u003e Examining the relationship between lifetime stressful life events and the onset of major depression in Chinese women. \u003cem\u003eJournal of Affective Disorders\u003c/em\u003e \u003cstrong\u003e135\u003c/strong\u003e, 95\u0026ndash;99 (2011).\u003c/li\u003e\n\u003cli\u003eGao, J. \u003cem\u003eet al.\u003c/em\u003e Perceived parenting and risk for major depression in Chinese women. \u003cem\u003ePsychol. Med.\u003c/em\u003e \u003cstrong\u003e42\u003c/strong\u003e, 921\u0026ndash;930 (2012).\u003c/li\u003e\n\u003cli\u003eYang, F. \u003cem\u003eet al.\u003c/em\u003e Age at onset of major depressive disorder in Han Chinese women: Relationship with clinical features and family history. \u003cem\u003eJournal of Affective Disorders\u003c/em\u003e \u003cstrong\u003e135\u003c/strong\u003e, 89\u0026ndash;94 (2011).\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;Association of voice pitch features and MDD.\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe associations between pitch features and MDD were estimated in the discovery (CONVERGE) sample and the replication sample, using \u003cstrong\u003e\u0026nbsp;\u003c/strong\u003etwo-stage meta-analysis method. P-values are FDR corrected. The uncorrected P-values are in Table S5. LPC means linear prediction coding.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003eBased\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003eStatistical Functionals\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003eCONVERGE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003eReplication\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eBeta\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eSE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eP_FDR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eBeta\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eSE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eP_FDR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eInterquartile range (3rd -1st)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-1.070\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.071\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.06E-49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-1.085\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.206\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.46E-06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMaximum (99th percentile)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.971\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.070\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.78E-42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.973\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.214\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.68E-05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTime with F0\u0026gt;90th percentile\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.804\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.060\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.49E-40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.779\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.193\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.34E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eKurtosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.868\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.066\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.11E-38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.828\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.227\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e5.37E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRoot quadratic mean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.690\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.060\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.80E-30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.762\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.211\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e5.44E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3rd quartile\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.781\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.069\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.71E-29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.883\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.201\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.64E-05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePosition of the minimum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.602\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.058\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.43E-24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.574\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.116\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.70E-06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.432\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.047\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.00E-19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.555\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.102\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.17E-06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSlope of linear regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.394\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.043\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.00E-19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.553\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.187\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.16E-03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRange\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.566\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.062\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.10E-19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.483\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.159\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.35E-03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePosition of the maximum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.551\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.061\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.92E-19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.614\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.27E-03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1st quadratic regression coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.384\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.045\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.59E-17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.473\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.122\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.31E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eFlatness*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.427\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.057\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.27E-13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.392\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.094\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e8.13E-05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2nd LPC coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.326\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.054\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.70E-09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.408\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.096\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e7.39E-05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4th LPC coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.284\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.048\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e9.79E-09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.248\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.154\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eKurtosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.225\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.051\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.03E-05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.295\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.094\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.78E-03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0th LPC coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.291\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.074\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.59E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.385\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.165\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTime with which \u0026Delta;F0 is rising\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.152\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.043\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.86E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.248\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.168\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMaximum length of voiced segments\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.194\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.071\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e9.96E-03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.117\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.106\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRoot quadratic mean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.143\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.055\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.126\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePosition of the maximum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.067\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.043\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eOffset of linear regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.084\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.055\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eInterquartile range (3rd -2nd)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.063\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.043\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePosition of the minimum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.074\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.053\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eProportion of time with which F0 is rising\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.080\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.059\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3rd quadratic regression coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.063\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.065\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3rd LPC coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.028\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.043\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSlope of linear regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.029\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.044\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0th LPC coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.046\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMinimum (1st percentile)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.052\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cbr\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. Heritable voice\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003epitch\u0026nbsp;features and their genetic correlation with MDD.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"99%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.185567010309279%\" rowspan=\"2\"\u003e\n \u003cp\u003eBase\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"28.8659793814433%\" rowspan=\"2\"\u003e\n \u003cp\u003eStatistical Functionals\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.68041237113402%\" colspan=\"2\" valign=\"bottom\"\u003e\n \u003cp\u003eSNP heritability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.2680412371134%\" colspan=\"4\" valign=\"bottom\"\u003e\n \u003cp\u003eGenetic Correlation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"11.290322580645162%\" valign=\"bottom\"\u003e\n \u003cp\u003eh\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.193548387096776%\" valign=\"bottom\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.290322580645162%\" valign=\"bottom\"\u003e\n \u003cp\u003erg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.580645161290324%\" valign=\"bottom\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.129032258064516%\" valign=\"bottom\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.516129032258064%\" valign=\"bottom\"\u003e\n \u003cp\u003eP_FDR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.25%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.166666666666668%\" valign=\"bottom\"\u003e\n \u003cp\u003eInterquartile range (3rd -1st)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.171\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.625%\" valign=\"bottom\"\u003e\n \u003cp\u003e(0.071, 0.272)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e(-0.77, -0.13)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.416666666666666%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.375%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.25%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.166666666666668%\" valign=\"bottom\"\u003e\n \u003cp\u003eMaximum (99th percentile)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.121\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.625%\" valign=\"bottom\"\u003e\n \u003cp\u003e(0.022, 0.221)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e(-1.28, -0.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.416666666666666%\" valign=\"bottom\"\u003e\n \u003cp\u003e4.17E-05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.375%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.0002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.25%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.166666666666668%\" valign=\"bottom\"\u003e\n \u003cp\u003eKurtosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.125\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.625%\" valign=\"bottom\"\u003e\n \u003cp\u003e(0.025, 0.225)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e(0.23, 0.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.416666666666666%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.375%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.25%\" valign=\"bottom\"\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.166666666666668%\" valign=\"bottom\"\u003e\n \u003cp\u003eKurtosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.134\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.625%\" valign=\"bottom\"\u003e\n \u003cp\u003e(0.035, 0.234)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.291666666666667%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e(-1.14, 0.54)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.416666666666666%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.375%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3. Voice\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003epitch\u0026nbsp;features associated with SCL scores.\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe associations between pitch features and SCL scores were estimated in the replication sample using \u003cstrong\u003e\u0026nbsp;\u003c/strong\u003etwo-stage meta-analysis method. LPC means linear prediction coding.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"99%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003eBased Feature\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eStatistical Functionals\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003eBeta\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003eSE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003eP_FDR\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026Delta;F0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003ePosition of the maximum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.222\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.040\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e3.42E-08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e5.47E-07\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eMaximum (99th percentile)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.296\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.074\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e6.67E-05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e5.33E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eRoot quadratic mean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.238\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.067\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e3.58E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.001\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003e3rd quartile\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.286\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.081\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e3.93E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.001\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eInterquartile range (3rd -1st)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.309\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.097\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.001\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.004\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eKurtosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.225\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.081\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.005\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.01\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.133\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.053\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.01\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.02\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eFlatness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.116\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.056\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.04\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.05\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eQuadratic regression coefficients 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.104\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.062\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.09\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.12\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eSlope of linear regression\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.103\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.067\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.12\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.14\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003ePosition of the minimum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.130\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.086\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.13\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.14\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003eF0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eTime of F0\u0026gt;90th percentile\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.246\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.066\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e1.90E-04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.001\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eLPC coefficient 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.118\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.005\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.01\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eKurtosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.097\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.041\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.02\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eRange\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.057\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.02\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"14.583333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"36.458333333333336%\" valign=\"bottom\"\u003e\n \u003cp\u003eLPC coefficient 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e-0.047\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.104\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.5%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.65\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.458333333333334%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.65\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"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":"","lastPublishedDoi":"10.21203/rs.3.rs-4135145/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4135145/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Major depressive disorder (MDD) often goes undiagnosed due to the absence of clear biomarkers. We sought to identify robust voice biomarkers for MDD and separate trait biomarkers indicative of MDD predisposition from state biomarkers reflecting current depressive symptoms. We investigated the association between voice pitch and MDD in a multisite recurrent MDD case-control cohort and validated our findings in a replication cohort. We then determined the heritability of identified features, their genetic correlation with MDD, and their association with depressive state. We found robust associations between MDD and pitch features, which depicted a slower pitch change and a lower pitch. These features achieved an AUC-ROC of 0.80 in MDD classification. Features measuring the variability and extremity of pitch change speed were heritable and had genetic correlations with MDD. State-related features were also detected. Our results return vocal features to a more central position in clinical and research work on MDD.","manuscriptTitle":"Unraveling the Associations Between Voice Pitch and Major Depressive Disorder: A Multisite Genetic Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-04-03 18:38:10","doi":"10.21203/rs.3.rs-4135145/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":"11069236-bb04-4473-b681-fa1e96b17174","owner":[],"postedDate":"April 3rd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":30190377,"name":"Biological sciences/Psychology"},{"id":30190378,"name":"Health sciences/Biomarkers/Diagnostic markers"},{"id":30190379,"name":"Biological sciences/Genetics/Behavioural genetics"}],"tags":[],"updatedAt":"2024-05-10T01:02:57+00:00","versionOfRecord":[],"versionCreatedAt":"2024-04-03 18:38:10","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4135145","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4135145","identity":"rs-4135145","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}