Levent Kirisci
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Levent Kirisci has a total of 92 co-authors. Here are the top co-authors by number of shared publications:
Maureen Reynolds
Department of Pharmaceutical Sciences, University of Pittsburgh
15 shared publicationsMichael Vanyukov
Department of Pharmaceutical Sciences, University of Pittsburgh
14 shared publicationsRalph Tarter
Department of Pharmaceutical Sciences, University of Pittsburgh
11 shared publicationsRalph E. Tarter
Department of Pharmaceutical Sciences, Center for Education and Drug Abuse Research, University of Pittsburgh, Pittsburgh, PA, USA
6 shared publicationsAda Mezzich
Center for Education and Drug Abuse Research, University of Pittsburgh, School of Pharmacy, Pittsburgh, Pennsylvania
5 shared publications- Lirong Wang
Department of Pharmaceutical Sciences, University of Pittsburgh School of Pharmacy
5 shared publications
Latest Publications
- Open Access
Exploring the relationship between genetic and environmental influences on the initiation and progression of cannabis use from adolescence to early adulthood
Aims: To estimate the genetic and environmental contributions to variance in liabilities to the initiation and progression stages of cannabis use. Design: A two‐stage, causal‐common‐contingent model was applied to data from a population‐based twin study to assess the relationship between initiation and progression of cannabis use and the contributions of genetic, shared environmental and unique environmental influences on both stages. Setting: Virginia, United States. Participants: 1366 self‐identified White twin pairs from the Virginia Twin Study of Adolescent Behavioral Development (VTSABD), enrolled between 1990 and 2006. Data were drawn from the first four waves for the 10–18 age group ( n = 1331 pairs). Two subsequent adult follow‐up waves provided data for the 18–31 age group ( n = 1183 pairs) and the 22–32 age group ( n = 666 pairs). Measurements: Cannabis initiation was defined as any lifetime use. Progression was defined as current or lifetime use occurring more than 10 times per month. Results: A statistically significant association was detected between liabilities to cannabis initiation and progression for two early adult groups [ages 18–31: b = 0.88, 95% confidence interval (CI) = 0.75–0.95, P ‐value < 0.001; and ages 22–32: b = 0.92, CI = 0.72–0.99, P ‐value < 0.001], but not for adolescents (ages 10–18, P ‐value = 1). Relative magnitudes of genetic and environmental contributions differed between adolescence and early adulthood. Variance in liability to cannabis use initiation was explained by genetic (56%), shared environmental (38%) and unique environmental influences (6%) in adolescence, and by genetic (79–84%) and unique environmental influences (16–22%) in early adulthood. Variance in liability to cannabis use progression was accounted for by genetic (37%) and unique environmental influences (63%) in adolescence. In early adulthood, 77–84% of the variance in liability to progression was shared with initiation, with the remaining variance attributable to environmental influences specific to progression. Summary: Among United States adolescents who initiate cannabis use, liability to progression does not appear to be explained by the same genetic or environmental influences underlying liability to initiation. In early adulthood, the two stages appear to share some genetic and unique environmental influences.
Open AccessExploring the relationship between genetic and environmental influences on the initiation and progression of cannabis use from adolescence to early adulthood
Aims: To estimate the genetic and environmental contributions to variance in liabilities to the initiation and progression stages of cannabis use. Design: A two‐stage, causal‐common‐contingent model was applied to data from a population‐based twin study to assess the relationship between initiation and progression of cannabis use and the contributions of genetic, shared environmental and unique environmental influences on both stages. Setting: Virginia, United States. Participants: 1366 self‐identified White twin pairs from the Virginia Twin Study of Adolescent Behavioral Development (VTSABD), enrolled between 1990 and 2006. Data were drawn from the first four waves for the 10–18 age group ( n = 1331 pairs). Two subsequent adult follow‐up waves provided data for the 18–31 age group ( n = 1183 pairs) and the 22–32 age group ( n = 666 pairs). Measurements: Cannabis initiation was defined as any lifetime use. Progression was defined as current or lifetime use occurring more than 10 times per month. Results: A statistically significant association was detected between liabilities to cannabis initiation and progression for two early adult groups [ages 18–31: b = 0.88, 95% confidence interval (CI) = 0.75–0.95, P ‐value < 0.001; and ages 22–32: b = 0.92, CI = 0.72–0.99, P ‐value < 0.001], but not for adolescents (ages 10–18, P ‐value = 1). Relative magnitudes of genetic and environmental contributions differed between adolescence and early adulthood. Variance in liability to cannabis use initiation was explained by genetic (56%), shared environmental (38%) and unique environmental influences (6%) in adolescence, and by genetic (79–84%) and unique environmental influences (16–22%) in early adulthood. Variance in liability to cannabis use progression was accounted for by genetic (37%) and unique environmental influences (63%) in adolescence. In early adulthood, 77–84% of the variance in liability to progression was shared with initiation, with the remaining variance attributable to environmental influences specific to progression. Summary: Among United States adolescents who initiate cannabis use, liability to progression does not appear to be explained by the same genetic or environmental influences underlying liability to initiation. In early adulthood, the two stages appear to share some genetic and unique environmental influences.
- 1 August 2025
- Published by American Psychological Association (APA) in Experimental and Clinical Psychopharmacology
- Vol. 33 (4) , 407-415
Interplay of cultural beliefs, psychological disposition, and social milieu during childhood on development of substance use disorder by early adulthood.
This study prospectively examined the interplay of culturally ingrained beliefs regarding the benefits of substance use, social milieu featured by disengaged parents and deviant friends, and behavior undercontrol on the development of substance use disorder (SUD) by early adulthood. The sample at baseline consisted of 497 youths between 10 and 12 years of age. Subsequent evaluations were conducted when the participants attained 12-14, 16, and 22 years of age. The research protocol was composed of self-report, informant report, and performance measures. Positive beliefs about substance use effects were evaluated using the Adolescent Alcohol and Drug Expectancy Questionnaire. The social milieu was characterized by the environtype index encompassing parental engagement and normative behavior of friends. Behavior control was quantified using the neurobehavior disinhibition scale. Substance use frequency and intensity of reward experience during consumption were recorded using the Drug Use Screening Inventory and the Substance Use Questionnaire. The outcome variable, lifetime SUD, was diagnosed using the Structured Clinical Interview for Diagnosis. Structural equation modeling evaluated the association among these variables. Beliefs regarding the benefits of consumption mediated the associations of childhood social milieu with substance use frequency and reward intensity during the transition into adolescence. Reward intensity during this developmental period predicted consumption frequency and severity of behavior undercontrol in midadolescence, which conjointly predicted SUD by early adulthood. This study demonstrated that SUD manifested by early adulthood culminates from the coaction of acquired cultural beliefs, social context facilitating access to addictive chemicals, and suboptimal psychological self-regulation. (PsycInfo Database Record (c) 2025 APA, all rights reserved).
- 1 August 2025
- Published by American Psychological Association (APA) in Experimental and Clinical Psychopharmacology
- Vol. 33 (4) , 407-415
Interplay of cultural beliefs, psychological disposition, and social milieu during childhood on development of substance use disorder by early adulthood.
This study prospectively examined the interplay of culturally ingrained beliefs regarding the benefits of substance use, social milieu featured by disengaged parents and deviant friends, and behavior undercontrol on the development of substance use disorder (SUD) by early adulthood. The sample at baseline consisted of 497 youths between 10 and 12 years of age. Subsequent evaluations were conducted when the participants attained 12-14, 16, and 22 years of age. The research protocol was composed of self-report, informant report, and performance measures. Positive beliefs about substance use effects were evaluated using the Adolescent Alcohol and Drug Expectancy Questionnaire. The social milieu was characterized by the environtype index encompassing parental engagement and normative behavior of friends. Behavior control was quantified using the neurobehavior disinhibition scale. Substance use frequency and intensity of reward experience during consumption were recorded using the Drug Use Screening Inventory and the Substance Use Questionnaire. The outcome variable, lifetime SUD, was diagnosed using the Structured Clinical Interview for Diagnosis. Structural equation modeling evaluated the association among these variables. Beliefs regarding the benefits of consumption mediated the associations of childhood social milieu with substance use frequency and reward intensity during the transition into adolescence. Reward intensity during this developmental period predicted consumption frequency and severity of behavior undercontrol in midadolescence, which conjointly predicted SUD by early adulthood. This study demonstrated that SUD manifested by early adulthood culminates from the coaction of acquired cultural beliefs, social context facilitating access to addictive chemicals, and suboptimal psychological self-regulation. (PsycInfo Database Record (c) 2025 APA, all rights reserved).
- Open Access
- 5 December 2024
- Published by MDPI AG in Biomedicines
- Vol. 12 (12) , 2772
Integrating Drug Target Information in Deep Learning Models to Predict the Risk of Adverse Events in Patients with Comorbid Post-Traumatic Stress Disorder and Alcohol Use Disorder
Background/Objectives: Comorbid post-traumatic stress disorder (PTSD) and alcohol use disorder (AUD) patients are at a significantly higher risk of adverse outcomes, including opioid use disorder, depression, suicidal behaviors, and death, yet limited treatment options exist for this population. This study aimed to build on previous research by incorporating drug target information into a novel deep learning model, T-DeepBiomarker, to predict adverse outcomes and identify potential therapeutic medications. Methods: We utilized electronic medical record (EMR) data from the University of Pittsburgh Medical Center (UPMC), analyzing 5565 PTSD + AUD patients. T-DeepBiomarker was developed by integrating multimodal data, including lab results, drug target information, comorbidities, neighborhood-level social determinants of health (SDoH), and individual-level SDoH (e.g., psychotherapy and veteran status). The model was trained to predict adverse events, including opioid use disorder, suicidal behaviors, depression, and death, within three months following any clinical encounter. Candidate medications targeting significant proteins were identified through literature reviews. Results: T-DeepBiomarker achieved high predictive performance with an AUROC of 0.94 for adverse outcomes in PTSD + AUD patients. Several medications, including OnabotulinumtoxinA, Dronabinol, Acamprosate, Celecoxib, Exenatide, Melatonin, and Semaglutide, were identified as potentially reducing the risk of adverse events by targeting significant proteins. Conclusions: T-DeepBiomarker demonstrates high accuracy in predicting adverse outcomes in PTSD + AUD patients and highlights candidate drugs with potential therapeutic effects. These findings advance pharmacotherapy for this high-risk population and identify medications that warrant further investigation.
- 5 December 2024
- Published by MDPI AG in Biomedicines
- Vol. 12 (12) , 2772
Open AccessIntegrating Drug Target Information in Deep Learning Models to Predict the Risk of Adverse Events in Patients with Comorbid Post-Traumatic Stress Disorder and Alcohol Use Disorder
Background/Objectives: Comorbid post-traumatic stress disorder (PTSD) and alcohol use disorder (AUD) patients are at a significantly higher risk of adverse outcomes, including opioid use disorder, depression, suicidal behaviors, and death, yet limited treatment options exist for this population. This study aimed to build on previous research by incorporating drug target information into a novel deep learning model, T-DeepBiomarker, to predict adverse outcomes and identify potential therapeutic medications. Methods: We utilized electronic medical record (EMR) data from the University of Pittsburgh Medical Center (UPMC), analyzing 5565 PTSD + AUD patients. T-DeepBiomarker was developed by integrating multimodal data, including lab results, drug target information, comorbidities, neighborhood-level social determinants of health (SDoH), and individual-level SDoH (e.g., psychotherapy and veteran status). The model was trained to predict adverse events, including opioid use disorder, suicidal behaviors, depression, and death, within three months following any clinical encounter. Candidate medications targeting significant proteins were identified through literature reviews. Results: T-DeepBiomarker achieved high predictive performance with an AUROC of 0.94 for adverse outcomes in PTSD + AUD patients. Several medications, including OnabotulinumtoxinA, Dronabinol, Acamprosate, Celecoxib, Exenatide, Melatonin, and Semaglutide, were identified as potentially reducing the risk of adverse events by targeting significant proteins. Conclusions: T-DeepBiomarker demonstrates high accuracy in predicting adverse outcomes in PTSD + AUD patients and highlights candidate drugs with potential therapeutic effects. These findings advance pharmacotherapy for this high-risk population and identify medications that warrant further investigation.
- Open Access
- 14 January 2024
- Published by MDPI AG in Journal of Personalized Medicine
- Vol. 14 (1) , 94
DeepBiomarker2: Prediction of Alcohol and Substance Use Disorder Risk in Post-Traumatic Stress Disorder Patients Using Electronic Medical Records and Multiple Social Determinants of Health
Prediction of high-risk events amongst patients with mental disorders is critical for personalized interventions. We developed DeepBiomarker2 by leveraging deep learning and natural language processing to analyze lab tests, medication use, diagnosis, social determinants of health (SDoH) parameters, and psychotherapy for outcome prediction. To increase the model’s interpretability, we further refined our contribution analysis to identify key features by scaling with a factor from a reference feature. We applied DeepBiomarker2 to analyze the EMR data of 38,807 patients from the University of Pittsburgh Medical Center diagnosed with post-traumatic stress disorder (PTSD) to determine their risk of developing alcohol and substance use disorder (ASUD). DeepBiomarker2 predicted whether a PTSD patient would have a diagnosis of ASUD within the following 3 months with an average c-statistic (receiver operating characteristic AUC) of 0.93 and average F1 score, precision, and recall of 0.880, 0.895, and 0.866 in the test sets, respectively. Our study found that the medications clindamycin, enalapril, penicillin, valacyclovir, Xarelto/rivaroxaban, moxifloxacin, and atropine and the SDoH parameters access to psychotherapy, living in zip codes with a high normalized vegetative index, Gini index, and low-income segregation may have potential to reduce the risk of ASUDs in PTSD. In conclusion, the integration of SDoH information, coupled with the refined feature contribution analysis, empowers DeepBiomarker2 to accurately predict ASUD risk. Moreover, the model can further identify potential indicators of increased risk along with medications with beneficial effects.
- 14 January 2024
- Published by MDPI AG in Journal of Personalized Medicine
- Vol. 14 (1) , 94
Open AccessDeepBiomarker2: Prediction of Alcohol and Substance Use Disorder Risk in Post-Traumatic Stress Disorder Patients Using Electronic Medical Records and Multiple Social Determinants of Health
Prediction of high-risk events amongst patients with mental disorders is critical for personalized interventions. We developed DeepBiomarker2 by leveraging deep learning and natural language processing to analyze lab tests, medication use, diagnosis, social determinants of health (SDoH) parameters, and psychotherapy for outcome prediction. To increase the model’s interpretability, we further refined our contribution analysis to identify key features by scaling with a factor from a reference feature. We applied DeepBiomarker2 to analyze the EMR data of 38,807 patients from the University of Pittsburgh Medical Center diagnosed with post-traumatic stress disorder (PTSD) to determine their risk of developing alcohol and substance use disorder (ASUD). DeepBiomarker2 predicted whether a PTSD patient would have a diagnosis of ASUD within the following 3 months with an average c-statistic (receiver operating characteristic AUC) of 0.93 and average F1 score, precision, and recall of 0.880, 0.895, and 0.866 in the test sets, respectively. Our study found that the medications clindamycin, enalapril, penicillin, valacyclovir, Xarelto/rivaroxaban, moxifloxacin, and atropine and the SDoH parameters access to psychotherapy, living in zip codes with a high normalized vegetative index, Gini index, and low-income segregation may have potential to reduce the risk of ASUDs in PTSD. In conclusion, the integration of SDoH information, coupled with the refined feature contribution analysis, empowers DeepBiomarker2 to accurately predict ASUD risk. Moreover, the model can further identify potential indicators of increased risk along with medications with beneficial effects.
- Open Access
- 19 August 2022
- Published by Elsevier in Journal of the American Pharmacists Association
- Vol. 63 (1) , 66-73.e1
Impact of pharmacist motivational interviewing on hepatitis B vaccination in adults with diabetes
- 19 August 2022
- Published by Elsevier in Journal of the American Pharmacists Association
- Vol. 63 (1) , 66-73.e1
Open AccessImpact of pharmacist motivational interviewing on hepatitis B vaccination in adults with diabetes
- 1 September 2021
- Published by American Psychological Association (APA) in Psychological Assessment
- Vol. 33 (9) , 890-903
Elaborating on the longitudinal measurement invariance and construct validity of the triarchic psychopathy scales from the Multidimensional Personality Questionnaire.
Because the construct of psychopathy is of chief interest across different disciplines, spanning developmental, clinical, and forensic psychology, its assessment bears far-reaching implications. One prominent contemporary conceptualization of psychopathy, the Triarchic Model, posits that a psychopathic personality encompasses three phenotypic constructs: boldness, meanness, and disinhibition. Recently, triarchic scales have been derived based on items from the Multidimensional Personality Questionnaire (MPQ), and the psychometric characteristics of this approach (MPQ-triarchic [MPQ-Tri]) are promising. The present study examined the longitudinal measurement invariance and the construct validity of the MPQ-Tri scales in a large and diverse high-risk sample (N = 716) across four time points from age 16-25. First, we report and discuss implications of confirmatory and exploratory factor analyses of the MPQ-Tri scales. Next, we report evidence for longitudinal configural and partial scalar invariance. In addition, in line with previous studies, MPQ-Boldness showed relatively higher levels of rank-order and mean-level stability compared to MPQ-Meanness and Disinhibition. Finally, in terms of construct validity, the MPQ-Tri scales showed a pattern of association with external correlates across internalizing and externalizing domains that were largely in line with theoretical expectations. One partial exception concerned the limited discriminant validity of the MPQ-Meanness and Disinhibition scales. On balance, the present findings suggest that the MPQ-Tri scales fulfill their intended purpose, with some noted limitation, and provide grounds for the use of the MPQ-Tri scales in developmentally-informed studies on the etiology and consequences of psychopathy.
- 1 September 2021
- Published by American Psychological Association (APA) in Psychological Assessment
- Vol. 33 (9) , 890-903
Elaborating on the longitudinal measurement invariance and construct validity of the triarchic psychopathy scales from the Multidimensional Personality Questionnaire.
Because the construct of psychopathy is of chief interest across different disciplines, spanning developmental, clinical, and forensic psychology, its assessment bears far-reaching implications. One prominent contemporary conceptualization of psychopathy, the Triarchic Model, posits that a psychopathic personality encompasses three phenotypic constructs: boldness, meanness, and disinhibition. Recently, triarchic scales have been derived based on items from the Multidimensional Personality Questionnaire (MPQ), and the psychometric characteristics of this approach (MPQ-triarchic [MPQ-Tri]) are promising. The present study examined the longitudinal measurement invariance and the construct validity of the MPQ-Tri scales in a large and diverse high-risk sample (N = 716) across four time points from age 16-25. First, we report and discuss implications of confirmatory and exploratory factor analyses of the MPQ-Tri scales. Next, we report evidence for longitudinal configural and partial scalar invariance. In addition, in line with previous studies, MPQ-Boldness showed relatively higher levels of rank-order and mean-level stability compared to MPQ-Meanness and Disinhibition. Finally, in terms of construct validity, the MPQ-Tri scales showed a pattern of association with external correlates across internalizing and externalizing domains that were largely in line with theoretical expectations. One partial exception concerned the limited discriminant validity of the MPQ-Meanness and Disinhibition scales. On balance, the present findings suggest that the MPQ-Tri scales fulfill their intended purpose, with some noted limitation, and provide grounds for the use of the MPQ-Tri scales in developmentally-informed studies on the etiology and consequences of psychopathy.
- Open Access
- 4 March 2021
- Published by MDPI AG in Journal of Personalized Medicine
- Vol. 11 (3) , 178
An Emulation of Randomized Trials of Administrating Antipsychotics in PTSD Patients for Outcomes of Suicide-Related Events
Post-traumatic stress disorder (PTSD) is a prevalent mental disorder marked by psychological and behavioral changes. Currently, there is no consensus of preferred antipsychotics to be used for the treatment of PTSD. We aim to discover whether certain antipsychotics have decreased suicide risk in the PTSD population, as these patients may be at higher risk. A total of 38,807 patients were identified with a diagnosis of PTSD through the ICD9 or ICD10 codes from January 2004 to October 2019. An emulation of randomized clinical trials was conducted to compare the outcomes of suicide-related events (SREs) among PTSD patients who ever used one of eight individual antipsychotics after the diagnosis of PTSD. Exclusion criteria included patients with a history of SREs and a previous history of antipsychotic use within one year before enrollment. Eligible individuals were assigned to a treatment group according to the antipsychotic initiated and followed until stopping current treatment, switching to another same class of drugs, death, or loss to follow up. The primary outcome was to identify the frequency of SREs associated with each antipsychotic. SREs were defined as ideation, attempts, and death by suicide. Pooled logistic regression methods with the Firth option were conducted to compare two drugs for their outcomes using SAS version 9.4 (SAS Institute, Cary, NC, USA). The results were adjusted for baseline characteristics and post-baseline, time-varying confounders. A total of 5294 patients were eligible for enrollment with an average follow up of 7.86 months. A total of 157 SREs were recorded throughout this study. Lurasidone showed a statistically significant decrease in SREs when compared head to head to almost all the other antipsychotics: aripiprazole, haloperidol, olanzapine, quetiapine, risperidone, and ziprasidone (p < 0.0001 and false discovery rate-adjusted p value < 0.0004). In addition, olanzapine was associated with higher SREs than quetiapine and risperidone, and ziprasidone was associated with higher SREs than risperidone. The results of this study suggest that certain antipsychotics may put individuals within the PTSD population at an increased risk of SREs, and that careful consideration may need to be taken when prescribed.
- 4 March 2021
- Published by MDPI AG in Journal of Personalized Medicine
- Vol. 11 (3) , 178
Open AccessAn Emulation of Randomized Trials of Administrating Antipsychotics in PTSD Patients for Outcomes of Suicide-Related Events
Post-traumatic stress disorder (PTSD) is a prevalent mental disorder marked by psychological and behavioral changes. Currently, there is no consensus of preferred antipsychotics to be used for the treatment of PTSD. We aim to discover whether certain antipsychotics have decreased suicide risk in the PTSD population, as these patients may be at higher risk. A total of 38,807 patients were identified with a diagnosis of PTSD through the ICD9 or ICD10 codes from January 2004 to October 2019. An emulation of randomized clinical trials was conducted to compare the outcomes of suicide-related events (SREs) among PTSD patients who ever used one of eight individual antipsychotics after the diagnosis of PTSD. Exclusion criteria included patients with a history of SREs and a previous history of antipsychotic use within one year before enrollment. Eligible individuals were assigned to a treatment group according to the antipsychotic initiated and followed until stopping current treatment, switching to another same class of drugs, death, or loss to follow up. The primary outcome was to identify the frequency of SREs associated with each antipsychotic. SREs were defined as ideation, attempts, and death by suicide. Pooled logistic regression methods with the Firth option were conducted to compare two drugs for their outcomes using SAS version 9.4 (SAS Institute, Cary, NC, USA). The results were adjusted for baseline characteristics and post-baseline, time-varying confounders. A total of 5294 patients were eligible for enrollment with an average follow up of 7.86 months. A total of 157 SREs were recorded throughout this study. Lurasidone showed a statistically significant decrease in SREs when compared head to head to almost all the other antipsychotics: aripiprazole, haloperidol, olanzapine, quetiapine, risperidone, and ziprasidone (p < 0.0001 and false discovery rate-adjusted p value < 0.0004). In addition, olanzapine was associated with higher SREs than quetiapine and risperidone, and ziprasidone was associated with higher SREs than risperidone. The results of this study suggest that certain antipsychotics may put individuals within the PTSD population at an increased risk of SREs, and that careful consideration may need to be taken when prescribed.
- Open Access
- 27 October 2020
- Published by MDPI AG in Brain Sciences
- Vol. 10 (11) , 784
Prediction of Suicide-Related Events by Analyzing Electronic Medical Records from PTSD Patients with Bipolar Disorder
Around 800,000 people worldwide die from suicide every year and it’s the 10th leading cause of death in the US. It is of great value to build a mathematic model that can accurately predict suicide especially in high-risk populations. Several different ML-based models were trained and evaluated using features obtained from electronic medical records (EMRs). The contribution of each feature was calculated to determine how it impacted the model predictions. The best-performing model was selected for analysis and decomposition. Random forest showed the best performance with true positive rates (TPR) and positive predictive values (PPV) of greater than 80%. The use of Sertraline, Fentanyl, Aripiprazole, Lamotrigine, and Tramadol were strong indicators for no SREs within one year. The use of Haloperidol, Trazodone and Citalopram, a diagnosis of autistic disorder, schizophrenic disorder, or substance use disorder at the time of a diagnosis of both PTSD and bipolar disorder, predicted the onset of SREs within one year. Additional features with potential protective or hazardous effects for SREs were identified by the model. We constructed an ML-based model that was successful in identifying patients in a subpopulation at high-risk for SREs within a year of diagnosis of both PTSD and bipolar disorder. The model also provides feature decompositions to guide mechanism studies. The validation of this model with additional EMR datasets will be of great value in resource allocation and clinical decision making.
- 27 October 2020
- Published by MDPI AG in Brain Sciences
- Vol. 10 (11) , 784
Open AccessPrediction of Suicide-Related Events by Analyzing Electronic Medical Records from PTSD Patients with Bipolar Disorder
Around 800,000 people worldwide die from suicide every year and it’s the 10th leading cause of death in the US. It is of great value to build a mathematic model that can accurately predict suicide especially in high-risk populations. Several different ML-based models were trained and evaluated using features obtained from electronic medical records (EMRs). The contribution of each feature was calculated to determine how it impacted the model predictions. The best-performing model was selected for analysis and decomposition. Random forest showed the best performance with true positive rates (TPR) and positive predictive values (PPV) of greater than 80%. The use of Sertraline, Fentanyl, Aripiprazole, Lamotrigine, and Tramadol were strong indicators for no SREs within one year. The use of Haloperidol, Trazodone and Citalopram, a diagnosis of autistic disorder, schizophrenic disorder, or substance use disorder at the time of a diagnosis of both PTSD and bipolar disorder, predicted the onset of SREs within one year. Additional features with potential protective or hazardous effects for SREs were identified by the model. We constructed an ML-based model that was successful in identifying patients in a subpopulation at high-risk for SREs within a year of diagnosis of both PTSD and bipolar disorder. The model also provides feature decompositions to guide mechanism studies. The validation of this model with additional EMR datasets will be of great value in resource allocation and clinical decision making.