Predicting and differentiating accidental and self-harm drug poisonings using health records data Authors: Gregory E. Simon, Robert D. Wellman, Susan M. Shortreed, Eric Johnson, Stacy A. Sterling, Karen J. Coleman, Brian K. Ahmedani, Zimri S. Yaseen, Andrew D. Mosholder, Jason Morgan, Jason Morgan, Karli Montague-Cardoso, Jason Morgan, Karli Montague-Cardoso, Karli Montague-Cardoso Published: 2026 ABSTRACT Preventing drug poisonings or overdoses will involve both generic poisoning prevention strategies and strategies specific to self-harm poisonings. This research used health records to develop statistical models predicting drug poisoning in general and self-harm poisoning in particular after an outpatient mental health specialty visit or primary care visit with mental health diagnosis. Records regarding visits from 2015 to 2019 in four health systems were used to develop models predicting poisoning following a two-step process: prediction of any poisoning or overdose followed by differentiating self-harm from accidental or undetermined intent poisoning or overdose. Separate random forest models for mental health specialty and primary care mental health visits were developed in 70% random samples of visits and validated in the remaining 30%. Among 19,130,028 visits, 114,911 (0.60%) were followed by any poisoning and 52,063 (0.27%) by self-harm poisoning. Models had moderate performance predicting any poisoning (Area under the Receiver Operating Curve or AUC = 0.778 among mental health specialty visits and 0.767 among primary care mental health visits) and good performance for specifically predicting self-harm poisoning (AUCs = 0.836 and 0.858 respectively). Mental health visits with risk above the 95th percentile accounted for 39.7% of self-harm poisoning (sensitivity) and actual risk of 2.57% (positive predictive value). Predictors of any poisoning included prior self-harm, accidental poisoning, and mental health service use. Specific predictors of self-harm poisoning included age and prior mental health diagnoses or treatments. In conclusion, among outpatients with mental health diagnoses, prediction models using records data have moderate accuracy predicting any overdose or poisoning and good accuracy predicting self-harm poisoning. Self-harm and accidental poisoning have both shared and distinct predictors. Introduction Overdoses or poisonings account for over 100,000 deaths annually in the US, an increase of approximately 50% over the last 5 years [1] that accelerated during the COVID-19 pandemic [2]. Nonfatal poisonings account for nearly 1,000,000 emergency department (ED) visits annually [3], and ED visits for poisoning increased during the COVID-19 pandemic despite decreases in overall ED utilization. Approximately 70% of nonfatal poisonings and 90% of fatal poisonings are classified as accidental, with the remainder coded as self-harm or undetermined intent [4,5]. Self-harm poisonings and accidental poisonings affect overlapping populations and involve shared risk factors [6,7]. Interventions to prevent accidental poisoning and interventions to prevent self-harm poisoning may be either overlapping or distinct. Preventive interventions to reduce availability of toxic drugs or medications, such as dispensing smaller quantities, safer packaging, drug “take-back” programs, or promotion of safe drug storage, aim to prevent both accidental and self-harm poisonings [8–10]. In contrast, clinical interventions such as dialectical behavior therapy [11,12] or cognitive behavioral therapy [13] specifically aim to reduce risk of intentional self-harm. Effective targeting of interventions to prevent either accidental or self-harm poisoning will require accurate identification of people at increased risk for specific types of poisoning events. While traditional clinical risk factors have limited accuracy in predicting self-harm [14], prediction models developed from health records data can accurately identify those at risk of fatal or non-fatal self-harm [15,16]. Large health systems have begun to implement those prediction models to prompt outreach or notify treating clinicians [17,18]. In parallel, prediction models using health records data can accurately identify risk of poisoning coded as accidental [19–21]. Previous research has not, however, evaluated overlap of and differences between prediction of poisonings in general and self-harm poisonings in particular. Here we use records data from four large health systems to examine similarities and differences between prediction of medication or drug poisoning in general and self-harm poisoning in particular. We examine two specific questions: 1)How well do statistical models developed using health records data perform in sequentially predicting any medication or drug poisoning followed by distinguishing self-harm poisonings from poisonings with accidental or undetermined intent? How well do statistical models developed using health records data perform in sequentially predicting any medication or drug poisoning followed by distinguishing self-harm poisonings from poisonings with accidental or undetermined intent? 2)How do the predictors of any medication or drug poisoning visit differ from the predictors differentiating self-harm poisonings from those with accidental or undetermined intent? How do the predictors of any medication or drug poisoning visit differ from the predictors differentiating self-harm poisonings from those with accidental or undetermined intent? We focus on specialty mental health specialty visits and primary care visits with mental health diagnosis, because over half of suicide deaths and half of nonfatal suicide attempts are preceded by one of those visits within three months [22,23]. We focus on the 90 days following a health care visit, since that is the period in which potential interventions (such as safer dispensing of medications or psychotherapy addressing risk of self-harm) might take effect. Methods Ethics statement The Kaiser Permanente institutional review board approved use of deidentified records data for this research and waived the requirement for individual informed consent. Records data were accessed between 4/16/2021 and 9/26/2022. Investigators had no access to identifying information. Setting and population The four participating health systems (Henry Ford Health and the Northern California, Southern California, and Washington regions of Kaiser Permanente) provide health insurance coverage and comprehensive primary care and mental health care to a combined population of approximately 11 million members enrolled through employer-sponsored insurance, individual insurance, subsidized insurance exchange programs, Medicare, and Medicaid. Membership of each health system is representative of its service area population in terms of age, sex, race, and ethnicity [24]. Each participating health system maintains a research data warehouse integrating information from electronic health records (for all services provided by the health system) and insurance claims (for all services provided at outside facilities), including medication dispensing as well as procedure codes and diagnoses from outpatient, emergency department, and inpatient encounters [25]. Membership records are linked to state mortality data to ascertain date and cause of death for all members. The sample included specialty mental health visits and primary care visits with mental health or substance use diagnoses beginning 10/1/2015 (when the ICD-10-CM coding system was implemented) and ending 12/31/2019 (avoiding the disruptions in care during the COVID-19 pandemic) by health system members aged 11 years or older, including all eligible visits for any individual. To ensure adequate ascertainment of outcomes, visits by patients disenrolling from the health system within 90 days were excluded. Outcomes Nonfatal medication or drug poisoning events over 90 days following each eligible visit were identified by ICD-10-CM encounter diagnoses in the range T36-T65 (all poisonings with substances “chiefly medicinal in nature”) recorded at any encounter (including outpatient, emergency department, or inpatient encounters) over 90 days following an eligible visit. This included poisonings or overdoses involving prescribed and non-prescribed drugs, including poisonings with opioids, cocaine, methamphetamine, and other stimulants. Poisonings with self-harm, accidental, and undetermined intent were identified by relevant modifiers, and poisonings coded as assault were excluded. Previous research in these health systems supports the accuracy of ICD-10-CM encounter diagnoses for identifying opioid poisonings [26] and self-harm poisonings [27]. Fatal poisonings were identified from state mortality records using cause of death codes for poisoning (T36-T65) and related codes for intent, including X60-X69 (self-harm), X40-X49 (accidental), and Y10-Y19 (undetermined intent). For both nonfatal and fatal poisonings, a single event with diagnoses indicating both self-harm intent and accidental/undetermined intent was considered self-harm. For each sampled visit, classification of outcomes considered the first poisoning event during the outcome period. Predictors Following methods developed in our previous research [28], approximately 2500 potential predictors of poisoning events included sociodemographic and clinical characteristics were extracted from research data warehouses for up to 60 months prior to each eligible visit. Sociodemographic characteristics included age, sex, self-reported race, self-reported Hispanic ethnicity, and neighborhood income and educational attainment from linkage of residential address to census data. Race and ethnicity data were typically reported in response to standard categories during outpatient visit registration and could be missing because a relevant visit did not occur, because questions were not asked or because a patient declined to answer. Mental health diagnoses were represented in 11 categories (e.g., depressive disorder, anxiety disorder, personality disorder), substance use disorder were represented in 5 categories (e.g., alcohol use disorder, opioid use disorder), and two additional categories represented pain and traumatic brain injury diagnoses. Mental health and substance use-related utilization were represented in 5 categories (e.g., hospitalization with mental health diagnosis, emergency department visit with substance use diagnosis). Prior injury and poisoning diagnoses were represented in 12 categories (e.g., benzodiazepine poisoning coded as accidental, self-harm). For each category of diagnosis, utilization, or prior injury/poisoning, 51 separate predictors represented different potential time patterns over the prior 60 months (e.g., number of occurrences in prior 3 months, most recent month with any occurrence). Psychiatric medication dispensings were represented in 10 categories (e.g., antidepressants, second-generation antipsychotics), and opioid-related medications were represented in 16 categories. For each medication category, 29 separate predictors represented different time patterns over the prior 60 months (e.g., days’ supply in prior 12 months, most recent month with no dispensing). Responses to each item of Patient Health Questionnaire (PHQ-9) depression questionnaires [29] completed at the sampled visit were represented by 12 potential predictors. Responses to PHQ-9 questionnaires at previous visits were represented by an additional 95 potential predictors representing various combinations and time patterns of responses. Chronic medical conditions were represented by the Charlson [30,31] comorbidity score and by binary indicators for each of the 17 conditions in Charlson score, calculated for the prior 12 months. Training and validating two-step prediction models for poisoning outcomes Because mental health specialty and primary care visits with mental health diagnoses represent different patient populations and different prevention opportunities, mental health specialty and primary care visit samples were analyzed separately, with each sample divided into a 70% training set and a 30% held-out validation or test set, split at the patient level [32]. Random forest models [33] predicting and differentiating poisoning events were developed using a two-step process. In the first step, a random forest model estimated among all training visits predicted probability of any drug poisoning in the following 90 days (p^p). In the second step, among training visits followed by any poisoning, a random forest model predicted if the poisoning was coded as self-harm (p^sh|p). The predicted probability from the second model, differentiating self-harm intent, was then calculated for all training visits. Predicted probability of self-harm poisoning among all training visits (p^sh) was calculated as the product of both steps (p^sh= p^sh|pp^p). The same predictors were considered at each step, and both models were estimated using random forests with probability trees [33,34]. Tuning parameters (minimum node size and number of trees) were selected for each step using five-fold cross validation in the training dataset with folds divided at the person level and parameters selected using out-of-fold AUC (details in S2 Table and S3 Table). The number of predictors considered at each decision point was set at 50, approximately the square root of the total number of predictors. Evaluating model performance in the validation dataset Final random forests were applied to all the visits in the validation sets to obtain predicted probability of any poisoning (p^p) and predicted probability of a poisoning coded as self-harm (p^sh|p). The performance of the two-step model was evaluated by estimating the performance of p^sh= p^sh|pp^p among all validation visits, indicated by AUC as well as the sensitivity, specificity, and positive predicted value using a cut-point at the 95th percentile in training data. Confidence intervals were calculated using 10,000 bootstrap iterations performed at the visit level. All analyses were conducted with R version 3.5.3 and R studio version 1.1.463 for windows. Random forests were estimated using the R package ranger version 0.11.2 [35]. Variable importance for predicting and differentiating poisoning events Variable importance based on the gini index was calculated on the training data for all predictors used in estimating the final random forest models [33]. To allow comparison between step 1 and step 2 models, variable importance measures were normalized to a maximum of 100. Results Study sample Table 1 displays the characteristics of 13,573,632 eligible mental health specialty visits (including 91,411 followed by any poisoning and 45,340 followed by self-harm poisoning) and characteristics of 5,556,396 eligible primary care mental health visits (including 23,500 followed by any poisoning and 6,723 followed by self-harm poisoning). Characteristics of visits assigned to the training and validations datasets are shown in S1 Table. 10.1371/journal.pmen.0000630.t001Table 1Characteristics of full visit samples, visits followed by any poisoning, and visits followed by self-harm poisoning.MENTAL HEALTH SPECIALTY VISITSPRIMARY CARE VISITS, MENTAL HEALTH DX.All VisitsAny PoisoningSelf-harm PoisoningAll VisitsAny PoisoningSelf-harm PoisoningN%N%N%N%N%N%Visits, n (row %)13,573,63291,41145,3405,556,39623,5006,723 Female 8,574,38963.260,37966.133,99775.03,447,35062.014,13862.04,47362.0 Age group (years) 11-171,789,58713.218,98920.813,82730.5412,1467.41,8607.91,05015.618-292,617,89719.323,20925.411,71725.8789,29514.25,05221.51,76626.330-443,452,89225.420,26022.29,04620.01,114,30720.14,90320.91,54523.045-644,140,95130.522,54024.79,03519.91,626,29229.36,82729.11,81327.065 or older1,572,30511.66,4137.01,7153.81,614,35629.14,85820.75498.2 Race and ethnicity* Asian1,091,6838.06,6747.33,9568.7410,4267.41,8607.95037.5American Indian/Native Alaskan123,0350.91,1231.25951.391,0151.62521.1861.3Black/African American1,291,9819.58,0188.83,8788.6499,3079.01,9038.15257.8Native Hawaiian/Pacific Islander92,1000.75980.73190.726,2920.51270.5530.8White, non-Hispanic7,409,66654.652,54557.524,70554.52,914,70252.513,80858.83,74955.8Hispanic ethnicity3,618,81726.724,05426.312,71928.11,502,80727.05,22922.31,72525.7Not recorded329,0962.41,5831.78681.9111,8472.03211.4821.2 Insurance Type Commercial group10,773,46379.472,18579.036,66780.93,850,34569.315,91667.74,97974.1High-deductible Health Plan1,005,7777.45,8976.52,9776.6416,7567.51,5106.44536.7Individual coverage2,851,48221.020,57622.59,37020.71,643,13529.67,03829.91,71425.5Medicaid1,604,18011.813,10214.36,73114.8617,19411.13,47714.81,12516.7Medicare2,262,05016.714,85716.36,02813.31,815,48232.77,16930.51,27218.9 PHQ-9 item 9 recorded at index visit Not Recorded12,673,50193.485,39293.442,89094.65,323,43095.821,85793.06,16591.7Response: 0690,9405.13,4023.79372.1185,8463.31,1114.72924.3Response: 1137,8871.01,3511.56701.531,4750.63001.31281.9Response: 242,3800.35700.63880.99,7000.21230.5681.0Response: 328,9240.26960.84551.05,9450.11090.5701.0 Sum of PHQ-9 Items 1–8 recorded at index visit Not Recorded12,712,15993.785,62493.742,95294.75,344,57696.222,01393.76,19692.2Response: 0–4203,7151.57890.92080.540,9910.71540.7480.7Response: 5–10301,4742.21,6061.85321.270,0111.33811.61161.7Response: 11–15183,3801.41,2781.45761.349,4270.93781.61322.0Response: 16–20117,6580.91,2311.35931.334,7350.63071.31241.8Response: 21 or higher55,2460.48831.04791.116,6560.32671.11071.6 Diagnoses Recorded in Prior 5 Years Anxiety10,680,68478.778,76486.239,78087.73,352,91560.318,00376.65,47581.4Bipolar disorder1,728,00812.725,18027.513,90330.7318,0645.74,58119.51,75526.1Depression9,685,68471.476,27983.440,73589.82,775,22849.916,35369.65,44280.9Personality disorder1,825,98813.522,73824.913,03928.8376,6756.83,86616.51,58223.5Schizophrenia spectrum disorder610,0904.58,8669.74,62610.2112,1802.01,6206.96079.0Traumatic brain injury510,5603.86,3897.02,8766.3224,5184.01,9258.25317.9Opioid use disorder593,0274.411,80012.93,4847.7218,3743.94,50319.21,03615.4 Mental Health Service Use in Prior 5 Years Mental health inpatient stay3,170,38923.449,32754.027,13859.91,118,13920.111,30048.13,62553.9Mental health emerg dept visit5,217,47738.461,58967.431,74170.01,947,79835.115,22964.84,66069.3Mental health outpatient visit12,542,90392.487,92796.243,85196.72,734,23749.217,26973.55,64283.9 Medications Dispensed in Prior 5 Years Antidepressant9,041,56466.674,50381.538,21184.33,167,54857.017,21273.25,24978.1Benzodiazepine5,462,90840.248,05152.621,83748.22,104,33737.913,24856.43,71655.31st Generation antipsychotic528,3353.97,6788.43,9448.7182,0663.31,9758.45718.5Lithium559,3114.110,22311.25,44812.078,9341.41,4816.35858.72nd Generation antipsychotic2,957,90621.840,11743.921,76048.0629,71111.37,61432.42,86142.6* Individuals who reported identifying with more than one listed race and ethnicity contribute to all selected racial and ethnic subgroups. Characteristics of full visit samples, visits followed by any poisoning, and visits followed by self-harm poisoning. * Individuals who reported identifying with more than one listed race and ethnicity contribute to all selected racial and ethnic subgroups. Validating two-step prediction models for poisoning outcomes Table 2 summarizes classification performance of models for prediction of any drug poisoning (Step 1) or specific prediction of self-harm poisoning (Step 1 and Step 2 combined). For prediction of any poisoning, overall performance was moderate, as indicated by an AUCs of 0.778 (95% CI 0.775 – 0.781) in the mental health specialty sample and 0.767 (95% CI 0.761 – 0.773) in the primary care mental health sample. Selecting the highest-ranked 5 percent of visits would identify 31.1% (95% CI 30.5 - 31.7) of visits in the mental health specialty sample and 33.3% (95% CI 32.2 - 34.4) of visits in the primary care sample followed by any poisoning event, representing six-fold concentrations of risk. Area under the Precision-Recall Curve (AUPRC) value were approximately eight times the event rate (the expected AUPRC value if classification was no better than chance). For specific prediction of self-harm poisoning, overall performance was good, as indicated by AUCs of 0.836 (95% CI 0.833 – 0.840) in the mental health specialty sample and 0.858 (95% CI 0.850 – 0.867) in the primary care mental health sample. Selecting the highest-ranked 5% of visits would identify 39.7% (95% CI 19.7 – 40.6) of visits in the mental health specialty sample and 47.5% (95% CI 45.3 – 49.8) of visits in the primary care mental health sample followed by a self-harm poisoning – representing eight-fold and ten-fold concentrations of risk. AUPRC values were ten to twenty times the event rate (the expected AUPRC value if classification were no better than chance). 10.1371/journal.pmen.0000630.t002Table 2Classification performance of two-step models for prediction of any poisoning event and self-harm poisoning event. Performance was assessed in validation dataset; 95% confidence intervals were constructed using 10,000 bootstrap samples.Event RateAUC(95% CI)Brier score(95% CI)F-score of 95th %ile(95% CI)Sensitivity of 95th %ile(95% CI)PPV of 95th %ile(95% CI)Specificity of 95th %ile(95% CI)AUPRC(95% CI)STEP 1 MODEL PREDICTING ANY POISONINGMental Health Specialty Visits0.00670.778(0.775, 0.781)0.00623(0.00615, 0.0063)0.073(0.0712, 0.0742)0.311(0.305, 0.317)0.0412(0.0403, 0.042)0.9517(0.9516, 0.9518)0.044(0.043, 0.045)Primary care Visits0.00420.767(0.761, 0.773)0.0040(0.0039, 0.0040)0.052(0.05, 0.055)0.333(0.322, 0.344)0.028(0.027, 0.03)0.9511(0.9510, 0.9512)0.034(0.032, 0.035)COMBINED STEP 1 AND STEP 2 MODEL PREDICTING SELF-HARM POISONINGMental Health Specialty Visits0.00330.836(0.833, 0.840)0.003(0.0031, 0.0032)0.048(0.047, 0.050)0.397(0.389, 0.406)0.0257(0.0250, 0.0264)0.9511(0.9510, 0.9514)0.029(0.027, 0030)Primary care Visits0.00120.858(0.85, 0.867)0.0011(0.0011, 0.0012)0.022(0.021, 0.023)0.475(0.453, 0.498)0.011(0.011, 0.012)0.9505(0.9503, 0.9507)0.024(0.022, 0.026) Classification performance of two-step models for prediction of any poisoning event and self-harm poisoning event. Performance was assessed in validation dataset; 95% confidence intervals were constructed using 10,000 bootstrap samples. Table 3 displays calibration performance for prediction of poisoning events in the validation sample. For prediction of any poisoning (lower left), predicted risk closely approximated observed risk, and observed risk among visits above the 99.5th percentile was approximately 50 times higher than among visits below the 50th percentile. For prediction of self-harm poisoning, (upper right), predicted risk closely approximated observed risk across all strata, and observed risk among visits above the 99.5th percentile was approximately 100 times higher than among visits below the 50th percentile. Results were similar when the primary care visit model was applied to prediction of either any poisoning or self-harm poisoning in the primary care mental health visit validation sample (S4 Table). 10.1371/journal.pmen.0000630.t003Table 3Calibration tables for performance of two-step model predicting any poisoning event and predicting self-harm poisoning in validation sample of mental health specialty visits.Predicting any poisoningPredicting self-harm poisoningPercentileAverage predicted probabilityProportion of visits followed by an eventAverage predicted probabilityProportion of visits followed by an event99.5<0.0900.1140.0600.06699.0-99.50.0550.0490.0370.03195.0-99.00.0330.0300.0200.02090.0-95.00.0180.0170.0090.00975.0-90.00.0100.0100.0040.00550.0-75.00.0050.0050.0020.002<50.00.0020.0020.0010.001 Calibration tables for performance of two-step model predicting any poisoning event and predicting self-harm poisoning in validation sample of mental health specialty visits. Variable importance for predicting and differentiating poisoning events The left portion of Fig 1 describes the most influential predictors in the Step 1 model predicting any poisoning or overdose following a mental health specialty visit. Among the 100 most influential predictors, 59 were indicators of prior self-harm injury or poisoning, 21 were indicators of prior poisonings or overdoses coded as accidental or undetermined intent, and 14 were indicators of prior emergency department or inpatient care with mental health or substance use diagnoses. S1 Fig lists those 100 individual predictors in order of influence. The right portion of Fig 2 describes the most influential predictors in the Step 1 model predicting any poisoning or overdose following a primary care visit with mental health or substance use diagnosis. The most influential predictors were generally similar to those for mental health specialty visits, with a greater representation of indicators for emergency department or inpatient care with mental health or substance use diagnosis. S2 Fig lists those 100 individual predictors in order of influence. Classification of 100 most influential predictors for Step 1 model predicting any poisoning following mental health specialty visit (left) or primary care visit with mental health or substance used diagnosis (right). Classification of 100 most influential predictors for Step 2 model differentiating self-harm poisonings from other poisonings among mental health specialty visits (left) and general mental health visits with mental health or substance use diagnoses (right). The left portion of Fig 2 describes the most influential predictors in the Step 2 model differentiating self-harm overdose or poisoning from other overdose or poisoning events. The 100 most influential predictors included prior self-harm injury or poisoning, prior mental health diagnoses prior emergency department or inpatient care with mental health or substance use diagnosis, prior substance use diagnose, and prior pain diagnoses. S3 Fig lists those 100 predictors in order of influence. The right portion of Fig 2 describes parallel results for differentiating self-harm overdose or poisoning from other overdose or poisoning following a primary care visit with mental health or substance use diagnosis. Results were generally similar to those for mental health specialty visits. S4 Fig lists those 100 predictors in order of influence. Discussion Prediction models developed from health records data had moderate performance in identifying patients who subsequently experienced any medication overdose or poisoning. Selecting visits in the highest 5% of the risk score distribution would identify approximately one third of visits followed by a poisoning event within 90 days – a six-fold concentration of risk. Among visits in that top 5%, the proportion actually experiencing a poisoning event in the following 90 days was approximately 1 out of 25 for mental health specialty visits and 1 out of 35 for primary care visits with mental health diagnoses. Models using records data had greater overall classification accuracy in predicting self-harm overdose of poisoning. Selecting visits in the highest 5% of the risk score distribution would identify over 40% of visits followed by a self-harm poisoning within 90 days – an eight- to nine-fold concentration of risk. Among visits in that top 5%, the proportion actually experiencing a poisoning even in the following 90 days was approximately 1 out of 40 for mental health specialty visits and 1 out of 100 for primary care visits with mental health diagnoses. Those lower event rates or positive predictive values for self-harm poisoning reflect the lower overall prevalence of self-harm poisonings – accounting for approximately one half of all poisonings following mental health specialty visits and fewer than one-third of all poisonings following primary care visits with mental health diagnoses. Specific predictors differentiating self-harm poisoning were moderately different from overall predictors of any overdose or poisoning. Prior self-harm events and prior use of inpatient or emergency department care for mental health or substance use diagnoses were influential as general predictors of any poisoning and as specific predictors of self-harm poisoning. Prior mental health diagnoses, prior substance use diagnoses, and prior pain diagnoses were identified as more specific predictors of self-harm poisoning or overdose. It is notable that prior self-harm injuries or poisonings were influential predictors of any poisoning event, including those coded as accidental. Similar to previous research regarding prediction of any self-harm or suicide attempt [36], these analyses do not identify new or unexpected predictors of overdose in general or self-harm overdose. The advantage of machine learning methods lies not in discovery of new risk factors but in optimally combining and weighting known risk factors to yield more accurate composite risk scores. Neither the individual risk factors identified in Table 1 nor a simple checklist could identify risk as accurately as a high-dimensional risk score [37,38]. The predictive relationships identified by these statistical models cannot necessarily be considered causal factors or targets for intervention. Predictors should instead be considered correlates of or markers for risk. For example, we should not conclude that use of mental health medications causes increased risk of poisoning or self-harm or that use of those medications should be avoided. Instead, we would conclude that mental health medications are more often prescribed to patients already at increased risk. Similarly, we should not assume that inpatient or emergency department mental health care cause increased risk or that those services should be avoided. Table 2 illustrates a practical trade-off between accuracy of prediction and potential public health impact. Compared to the step 1 model predicting any poisoning, the combined model predicting self-harm poisoning had superior classification accuracy as measured by AUC and sensitivity. But that more specific model had lower PPV, reflecting the lower prevalence of self-harm poisoning. Any intervention focused on patients above the 95th percentile of risk in the two-step model would reach only half as many patients as an intervention guided by the broader one-step model. Acceptable accuracy for any prediction models depends on the action or intervention that the model might inform. A positive predictive value in the range of 4%, as seen in this sample for prediction of any poisoning after a mental health specialty visit, might be appropriate for a low-intensity and low-risk intervention, such as safer packaging of medications often involved in overdose. An expected overdose or poisoning risk of 4% would likely be high enough to warrant additional clinical attention and risk assessment. Given the relatively low positive predictive value, any implementation must consider the added burden on clinical staff. But, absent a more detailed clinical assessment, that predicted rate of overdose or poisoning alone would not justify an intervention with significant risk or burden, such as involuntary hospitalization. Any clinical implementation of risk prediction scores must consider that false-positive identification of self-harm risk can have negative consequences such as unnecessary stigmatization or diverting attention from other clinical needs. Limitations Interpretation of these findings should consider several limitations. Ascertainment of poisoning events was limited to poisonings that resulted in death or presented to health care settings and were formally diagnosed. Some poisonings of undetermined intent may have represented self-harm. These findings might not generalize to people without recorded mental health diagnoses or to other health systems with less complete records systems or patterns of diagnosis or service use. The discrete or coded data used in these analyses (diagnosis codes, procedure codes, medication dispensings) would not capture use of non-prescribed substances, negative life events, or social determinants of health that may significantly influence likelihood of poisoning in general or self-harm poisoning in particular [39]. Excluding visits after 12/31/2019 avoids disruptions in care during the early COVID-19 pandemic but also avoids changes in overdose patterns with increasing use of fentanyl. Implementation of this or a similar model in current practice would require testing accuracy using more recent data, possibly leading to re-calibration to account for changes in overdose rates or new model fitting to account for changes in predictive relationships. Conclusions For health systems hoping to use health records data to target poisoning prevention interventions, statistical models developed using health records data have moderate accuracy in identifying patients at increased risk for any poisoning. Models developed to specifically predict self-harm poisoning have good performance, with the most influential predictors of self-harm poisoning including both prior poisoning events and indicators of prior mental health diagnosis or treatment. Supporting information Training and validation samples for mental health specialty visits and general medical visits with mental health diagnoses. (DOCX) Cross validation results for selecting optimal tuning parameter for random forest model predicting any poisoning among all training visits. Optimal out-of-sample AUCs are shown in bold. (DOCX) Cross validation results for selecting optimal tuning parameters for random forest model predicting a self-harm diagnosed poisoning among visits followed a poisoning. Optimal out-of-sample AUCs are shown in bold parameters in bold. (DOCX) Calibration tables for performance of two-step model for prediction of self-harm poisoning events in validation sample of primary care visits with mental health diagnoses. (DOCX) Normalized variable importance for 100 most influential predictors in Step 1 model predicting any poisoning event in full sample of mental health specialty visits. (TIF) Normalized variable importance for 100 most influential predictors in Step 1 model predicting any poisoning event in full sample of general medical visits with mental health diagnoses. (TIF) Normalized variable importance for 100 most influential predictors in Step 2 model predicting self-harm poisoning, limited to mental health specialty visits followed by a poisoning event. (TIF) Normalized variable importance for 100 most influential predictors in Step 2 model predicting self-harm poisoning, limited to general medical visits followed by a poisoning event. (TIF) 18 Jan 2026 PMEN-D-25-00516 Predicting and differentiating accidental and self-harm drug poisonings using health records data PLOS Mental Health Dear Dr. Simon, Thank you for submitting your manuscript to PLOS Mental Health. After careful consideration, we feel that it has merit but does not fully meet PLOS Mental Health’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Your manuscript has been assessed by one reviewers. While they indicate some interest in your work, they have raised several concerns that require your attention. Please consider these points carefully when preparing your revised manuscript and point-by-point response document. Please note that we have only been able to secure a single reviewer to assess your manuscript. We are issuing a decision on your manuscript at this point to prevent further delays in the evaluation of your manuscript. Please be aware that the editor who handles your revised manuscript might find it necessary to invite additional reviewers to assess this work once the revised manuscript is submitted. However, we will aim to proceed on the basis of this single review if possible. Please submit your revised manuscript by Feb 28 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at mentalhealth@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pmen/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. A letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. We look forward to receiving your revised manuscript. Kind regards, Jason Morgan Staff Editor PLOS Mental Health Journal Requirements: -->1. Please amend your detailed Financial Disclosure statement. This is published with the article. It must therefore be completed in full sentences and contain the exact wording you wish to be published.-->--> -->-->i. Please clarify all sources of financial support for your study. List the grants, grant numbers, and organizations that funded your study, including funding received from your institution. Please note that suppliers of material support, including research materials, should be recognized in the Acknowledgements section rather than in the Financial Disclosure. -->-->ii. State the initials, alongside each funding source, of each author to receive each grant. For example: "This work was supported by the National Institutes of Health (####### to AM; ###### to CJ) and the National Science Foundation (###### to AM)."-->-->iii. State what role the funders took in the study. If the funders had no role in your study, please state: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.”-->-->iv. If any authors received a salary from any of your funders, please state which authors and which funders.-->--> -->--> -->-->2. Please ensure that your Ethics Statement is available in its entirety at the beginning of your Methods section, under a subheading 'Ethics Statement'. It must include:-->--> -->-->i) The name(s) of the Institutional Review Board(s) or Ethics Committee(s)-->--> -->-->ii) The approval number(s), or a statement that approval was granted by the named board(s) -->--> -->-->iii) (for human participants or donors) - A statement that formal consent was obtained (must state whether verbal/written) OR the reason consent was not obtained (e.g., anonymity).-->--> -->--> -->-->3. We have noticed that you have uploaded Supporting Information files, but you have not included a list of legends. Please add a full list of legends for your Supporting Information files after the references list. -->--> -->--> -->-->4. We note that you have indicated that there are restrictions to data sharing for this study. For studies involving human research participant data or other sensitive data, we encourage authors to share de-identified or anonymized data. However, when data cannot be publicly shared for ethical reasons, we allow authors to make their data sets available upon request. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. -->--> -->--> Before we proceed with your manuscript, please address the following prompts:-->--> -->--> a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., a Research Ethics Committee or Institutional Review Board, etc.). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.-->--> -->--> b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of recommended repositories, please see https://journals.plos.org/plosone/s/recommended-repositories. You also have the option of uploading the data as Supporting Information files, but we would recommend depositing data directly to a data repository if possible.-->--> -->--> Please update your Data Availability statement in the submission form accordingly.--> If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise. Additional Editor Comments (if provided): [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions -->Comments to the Author 1. Does this manuscript meet PLOS Mental Health’s publication criteria? Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe methodologically and ethically rigorous research with conclusions that are appropriately drawn based on the data presented.--> Reviewer #1: Yes ********** -->2. Has the statistical analysis been performed appropriately and rigorously?--> Reviewer #1: Yes ********** -->3. Have the authors made all data underlying the findings in their manuscript fully available (please refer to the Data Availability Statement at the start of the manuscript PDF file)? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception. The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.--> Reviewer #1: No ********** -->4. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS Mental Health does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.--> Reviewer #1: Yes ********** -->5. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)--> Reviewer #1: Thak you for the opportunity to review this paper demonstrating the modelling of poisoning, including accidental vs intentional poisoning. The paper is generally sound, although I would like the authors to consider some issues. First, I spotted two minor typos: P3L38 add “to” P14L281 change “even tin” to “event in” There are some major issues I would like the authors to consider. First, there is lack of clarity about the sample. P7L115–121 notes: “The sample included specialty mental health visits and primary care visits with mental health or substance use diagnoses… including all eligible visits for any individual.” So first, is an “eligible visit” a mental health visit (either specialty or primary care with mental health/substance use)? Does this mean that all other clients were excluded? From Table 1 (which is too late to be certain about this point), it seems so. If so, this should probably be reflected in the title and abstract, as the underlying populations are quite different. The abstract only mentions “after an outpatient mental health or primary care visit”, which is ambiguous. Obviously, a mental health/substance use visit is, in and of itself, a predictor of subsequent drug poisoning. So: there are two different research questions for the two different populations: “What are predictors (one of which is possibly mental health/substance use visit) of drug poisoning in a general population?” vs “What are predictors of drug poisoning in a population of patients who have had a mental health/substance use visit?” I gather that it is the latter which the authors are investigating, but this should be made clear right from the title onward. Generally, I think the authors should also more clearly define the term “eligible visit” in the context of their study. They give the ICD codes for the outcomes, but very little detail for defining the eligible visits. Depending on what exactly is their underlying population, the authors could also, if their underlying population is only those patients with specialty mental health or primary care with mental health/substance use), rephrase sentences like (P9L173): “Because mental health specialty and primary care visits” to “Because specialty- and primary care mental health visits” in order to avoid confusion. Second, please redo both figures. Pie charts are almost never a good graphical choice (https://theconversation.com/heres-why-you-should-almost-never-use-a-pie-chart-for-your-data-214576), and I’m afraid this paper is one more example of where they are not. In fact, if the authors chose a simple table, Figures 1 and 2 could be collapsed into a single, more readable, entity. And if they insist on using figures, then four stacked bar charts in a 2x2 small-multiple chart could also collapse them into a single entity. Regardless, the legend shows that there are nine categories, and yet the charts only show six or seven. Even in a combined bar chart as suggested, the authors should ensure that all nine category value labels are displayed (in a table, this is not a problem). Third, I am glad that the authors note that “The predictive relationships identified by these statistical models cannot necessarily be considered causal factors or targets for intervention.” (P15L303–4). I think the discussion of this on P16 is quite good. However, I would like the authors to more explicitly consider the issue of Specificity. What are the implications, in this context, of not correctly identifying the patients who do not require an intervention. Commendably, the authors recommend at most “low-intensity and low-risk intervention” (P16L320). On P10L197 the authors mention specificity, but it is never reported nor discussed. The issue here might be illustrated with the similar situation in breast cancer screening. Better screening leads to more cases being identified, but also leads to greater numbers of people receiving treatment unnecessarily. In breast cancer screening, emphasis is put on the NNT and NNH. The point here is that interventions, even low intensity and low risk, can have unintended consequences. The authors rightly point out that their model is not enough to justify interventions alone. But one has to ask, then, what the point of the model is? This paper, in my opinion, needs comprehensive reporting showing not only how many patients are correctly identified as being at risk of future poisoning, but also how many patients at risk were missed, how many patients deemed not at risk did actually experience a subsequent poisoning event, and how many patients deemed at risk did not poisoning (i.e., reporting that covers all four quadrants of the confusion matrix). At present, the paper only considers sensitivity and PPV. This leads to the last suggestion, which relates to the scope of the paper. First, no justification is given for the 90-day follow-up period. There is some literature to indicate that indeed, most repeat poisonings do occur quite soon after the index, but the authors here are not investigating repeat poisoning. I don't think 90 days is necessarily wrong, just that it should be justified. And it should also be recommended, I think, that more work be done on slightly longer time periods. Also, in terms of scope, it was interesting to note that the authors recorded mortality (P8L133-4), but did not examine the performance of their model considering that outcome. It is not a faultless distinction, but when examining poisoning populations, there is a clear distinction and clear population demographic differences between those with suicidal intent and those who are self-harming. ********** -->6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. Do you want your identity to be public for this peer review? If you choose “no”, your identity will remain anonymous but your review may still be made public. For information about this choice, including consent withdrawal, please see our Privacy Policy.--> Reviewer #1: Yes: Jacques Eugene Raubenheimer ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] -->Figure Resubmissions: -->-->While revising your submission, we strongly recommend that you use PLOS’s NAAS tool (https://ngplosjournals.pagemajik.ai/artanalysis) to test your figure files. NAAS can convert your figure files to the TIFF file type and meet basic requirements (such as print size, resolution), or provide you with a report on issues that do not meet our requirements and that NAAS cannot fix.-->--> After uploading your figures to PLOS’s NAAS tool - https://ngplosjournals.pagemajik.ai/artanalysis, NAAS will process the files provided and display the results in the "Uploaded Files" section of the page as the processing is complete. If the uploaded figures meet our requirements (or NAAS is able to fix the files to meet our requirements), the figure will be marked as "fixed" above. If NAAS is unable to fix the files, a red "failed" label will appear above. When NAAS has confirmed that the figure files meet our requirements, please download the file via the download option, and include these NAAS processed figure files when submitting your revised manuscript.--> 21 Jan 2026 Submitted filename: response to PLOS MH reviews.docx 5 Apr 2026 PMEN-D-25-00516R1 Predicting and differentiating accidental and self-harm drug poisonings using health records data PLOS Mental Health Dear Dr. Simon, Thank you for submitting your revised manuscript to PLOS Mental Health. As explained in a prior email, I have reached out to a second reviewer, who has now provided their comments, which you can access in the attached document. In light of these comments, I would like to request a final round of minor revisions. Thank you for your patience and understanding. Please submit your revised manuscript by May 03 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at mentalhealth@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pmen/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. A letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only the individual author can complete the verification step; PLOS staff cannot verify ORCID iDs on behalf of authors. We look forward to receiving your revised manuscript. Kind regards, Karli Montague-Cardoso Staff Editor PLOS Mental Health Journal Requirements: If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Additional Editor Comments (if provided): [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions -->Comments to the Author 1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.--> Reviewer #1: All comments have been addressed Reviewer #2: (No Response) ********** -->2. Does this manuscript meet PLOS Mental Health’s publication criteria? Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe methodologically and ethically rigorous research with conclusions that are appropriately drawn based on the data presented.--> Reviewer #1: Yes Reviewer #2: Yes ********** -->3. Has the statistical analysis been performed appropriately and rigorously?--> Reviewer #1: Yes Reviewer #2: Yes ********** -->4. Have the authors made all data underlying the findings in their manuscript fully available (please refer to the Data Availability Statement at the start of the manuscript PDF file)? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception. The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.--> Reviewer #1: Yes Reviewer #2: Yes ********** -->5. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS Mental Health does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.--> Reviewer #1: Yes Reviewer #2: Yes ********** -->6. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)--> Reviewer #1: (No Response) Reviewer #2: Overall, this is a well-designed, highly rigorous retrospective cohort study. The use of a sequential, two-step modeling pipeline to parse apart spectral overlap in poisoning intent is a strong methodological contribution to the health informatics literature. Please see the attachment for my comments that may help the authors improve the manuscript. ********** -->7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. Do you want your identity to be public for this peer review? If you choose “no”, your identity will remain anonymous but your review may still be made public. For information about this choice, including consent withdrawal, please see our Privacy Policy.--> Reviewer #1: Yes: Jacques Eugene Raubenheimer Reviewer #2: No ********** [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] -->Figure Resubmissions: -->-->While revising your submission, we strongly recommend that you use PLOS’s NAAS tool (https://ngplosjournals.pagemajik.ai/artanalysis) to test your figure files. NAAS can convert your figure files to the TIFF file type and meet basic requirements (such as print size, resolution), or provide you with a report on issues that do not meet our requirements and that NAAS cannot fix.-->--> After uploading your figures to PLOS’s NAAS tool - https://ngplosjournals.pagemajik.ai/artanalysis, NAAS will process the files provided and display the results in the "Uploaded Files" section of the page as the processing is complete. If the uploaded figures meet our requirements (or NAAS is able to fix the files to meet our requirements), the figure will be marked as "fixed" above. If NAAS is unable to fix the files, a red "failed" label will appear above. When NAAS has confirmed that the figure files meet our requirements, please download the file via the download option, and include these NAAS processed figure files when submitting your revised manuscript.--> 7 May 2026 Submitted filename: response to 2nd PLOS review.docx 12 May 2026 Predicting and differentiating accidental and self-harm drug poisonings using health records data PMEN-D-25-00516R2 Dear Dr. Simon, We are pleased to inform you that your manuscript 'Predicting and differentiating accidental and self-harm drug poisonings using health records data' has been provisionally accepted for publication in PLOS Mental Health. Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. A member of our team will be in touch with a set of requests. Please note that your manuscript will not be scheduled for publication until you have made the required changes, so a swift response is appreciated. IMPORTANT: The editorial review process is now complete. PLOS will only permit corrections to spelling, formatting or significant scientific errors from this point onwards. Requests for major changes, or any which affect the scientific understanding of your work, will cause delays to the publication date of your manuscript. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they'll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact mentalhealth@plos.org. Thank you again for supporting Open Access publishing; we are looking forward to publishing your work in PLOS Mental Health. Best regards, Karli Montague-Cardoso Staff Editor PLOS Mental Health *********************************************************** Reviewer Comments (if any, and for reference):