| DOI: DOI:10.29090/psa.2024.03.24.AP0713 | Pharm Sci Asia 2024; 51(4), 314-325 |
Development and validation of machine learning-based predictive clinical decision support system for olanzapine in patients with schizophreniaAnh Mai Kieu1,2, Ha Nguyet Dang2, Khanh Thi Cat Tran2, Tuyen Thi Thanh Nguyen3, Chien Huu Nguyen3,
Thanh Chi Nguyen4, Huong Thu Pham4, Jennifer Le5, Hai Thanh Nguyen2,*
1 Department of Pharmacy, Vinh Medical University, Vinh, Vietnam
2
Department of Clinical Pharmacy, Hanoi University of Pharmacy, Hanoi, Vietnam
3
Department of Pharmacy, Vietnamese National Psychiatric Hospital No.1, Hanoi, Vietnam
4
Academy of Military Science and Technology, Hanoi, Vietnam
5
Skaggs School of Pharmacy and Pharmaceutical Sciences, University of California San Diego, San Diego, USA
Olanzapine is an atypical antipsychotic used to treat schizophrenia but can cause metabolic syndrome leading to severe cardiovascular events. This study aimed to develop a predictive decision tree model for clinical responses and adverse events of olanzapine, and integrate this model into the clinical decision support system (CDSS). The study consisted of three phases: (1) prospectively analyzed clinical responses and safety for hospitalized schizophrenic patients receiving olanzapine at Vietnamese National Psychiatric Hospital No.1; (2) determined the statistically significant predictors and developed predictive algorithms in machine learning (Decision Tree) to build the CDSS that incorporated warnings and predictive models for effectiveness and metabolic syndrome; and (3) conducted a longitudinal study on interventions after CDSS integration. Of 232 patients evaluated in phase 1, 76% responded positively to olanzapine, and 31% developed metabolic syndrome. 24 predictive variables were analyzed for effectiveness and 10 others were analyzed for metabolic syndrome. In phase 2, the decision tree model using Bayesian Model Averaging identified important predictive factors for effectiveness, retaining three important nodes: early response, response history, and olanzapine dose, with performance metrics of accuracy 0.89, precision 0.92, recall 0.94 and F1-score 0.93. Besides, another model using univariate regression identified important predictive factors for metabolic syndrome, retaining three important nodes: baseline waist < 89 cm, baseline triglyceride < 3.1 mmol/L, and age < 36 years with performance metrics of accuracy 0.88, precision 0.90, recall 0.69, and F1-score 0.78. Phase 3 evaluated 70 patients using CDSS, with 87% receiving “positively-responded” predictions, and 30% receiving metabolic syndrome predictions in the first week. 22 clinical pharmacist interventions led to doctors changing “clinical decisions”, while 389 interventions resulted in the “monitoring plan” of doctors. Incorporating machine learning models into CDSS is valuable in helping physicians identify and make interventions to ensure effective and safe use of olanzapine in schizophrenic patients
Keyword:
Olanzapine; Schizophrenia; Decision Tree; Machine Learning; Clinical Decision Support System
Download full paper (PDF File size: 1,584.54 KB.)
|
