Evaluating machine learning algorithms for prediction of treatment response for sleep disturbances in patients with schizophrenia - A post-hoc analysis from a randomized controlled trial

Aug 5, 2026

Upload by Eden

Post-hoc machine learning analyses were conducted to predict treatment response to sleep disturbances in patients with schizophrenia, using data from a randomized controlled trial evaluating add-on ramelteon. Models were trained with random forest, k-nearest neighbors, extreme gradient boosting, R part classification and regression trees, and logistic regression. Visual comparisons used box plots and dot plots. Logistic regression achieved the best fit with specificity 0.93, sensitivity 0.45, ROC 0.78, and 90% prediction accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Summary

Background

Subjects and Methods

Results

Conclusions

INTRODUCTION

Psychiatria Danubina, 2025 Vol􀀑 37, No 􀀑 1, pp 46-54 [https://doi](https://doi) 􀀑 org/10 􀀑 24869/psyd􀀑 2025 􀀑 46 Original research © Medicinska naklada – Zagreb, Croatia Evaluating machine learning algorithms for prediction of treatment response for sleep disturbances in patients with schizophrenia: A post-hoc analysis from a randomized controlled trial Archana Mishra, Rituparna Maiti, Monalisa Jena & Anand Srinivasan All India Institute of Medical Sciences (AIIMS), Bhubaneswar, Odisha, India received: 28 􀀑 4 􀀑 2023; revised: 8 􀀑 5 􀀑 2024; accepted: 13 􀀑 11 􀀑 2024 Summary Background: A post-hoc analysis was planned to create and compare machine learning algorithms to predict treatment responses to sleep disturbances inpatients with schizophrenia 􀀑 Subjects and Methods: This post-hoc analysis was done on a randomized controlled trial (NCT03075657), studying the effect of add-on ramelteon on sleep and circadian rhythm disturbances in 120 patients with schizophrenia 􀀑 We created models using random forest, k-nearest neighbors, extreme gradient boosting machine, R part Classification and regression trees and logistic regression algorithms 􀀑 R language with mlbench, caret, MASS, rPARTpackages were used􀀑 Box plot and dot plot were plotted to visualize comparisons among the models 􀀑 Results: The logistic regression algorithm was found to be the best-fit model with a specificity of 0 􀀑 93 and sensitivity of 0 􀀑 45, and ROC 0 􀀑 78 􀀑 Predominant symptom domain (positive or negative), urinary melatonin and global PSQI score at baseline were the most important variables when plotted in terms of mean decrease accuracy􀀑 These variables contributed significantly to the final model in the logistic regression algorithm, and the accuracy of this algorithm was found to be 90% for prediction 􀀑 Conclusions: Machine learning models are an emerging trend in clinical research and should be translated into clinical practice 􀀑 The logistic regression model predicted responders with 90% accuracy􀀑 Keywords: Schizophrenia; sleep disorder; machine learning * * * * * INTRODUCTION Sleep and circadian rhythm disturbances are frequent in patients with schizophrenia in addition to the cardinal features, which are used to diagnose the disease and ascertain its severity (Kaskie et al., 2017) . Antipsychotic therapy helps alleviate the positive and negative symptoms, but they are ineffective in improving sleep disturbances. Patients report adverse consequences of these sleep disorders in the form of worsening symptoms that warrant treatment (Chemerinski et al., 2002) . Conventional sedatives like benzodiazepines may impact phase shifts in circadian rhythm, thus setting up a vicious cycle in sleep disorders and psychosis symptoms (Hajak et al., 1996) . Melatonin is a robust biological marker of circadian rhythmicity (Claustrat & Leston, 2015) . Though there are pieces of evidence supporting the detection of schizophrenia with the help of various characteristics including structural MRI and multiple kernel learning identifiers; algorithms for prediction of response to interventions are lacking (Oh et al,. 2020; Chilla et al., 2022). In our previous study, augmentation therapy with melatonin receptor agonist ramelteon improved sleep and circadian rhythm in schizophrenia (Mishra et al., 2020) . The data for this study was taken from the above-mentioned randomized controlled trial studying the effect of add-on ramelteon on sleep and circadian rhythm in patients with schizophrenia. The primary objective of the clinical trial was to evaluate the change in serum melatonin (a robust biomarker of circadian rhythm), and the sample size was calculated accordingly. A regression model was built for the same in the original study. The present study aimed to generate multivariate prediction models for treatment response in terms of improvement in sleep using the Pittsburgh Sleep Quality Index (PSQI) scores in patients with schizophrenia. As a consi

FAQs

What data and trial design were used for the post-hoc analysis?

The analysis used data from a randomized controlled trial (NCT03075657) assessing add-on ramelteon in 120 patients with schizophrenia.

Which machine learning algorithms were compared to predict treatment response?

Random forest, k-nearest neighbors, extreme gradient boosting, classification and regression trees (rpart), and logistic regression were evaluated.

Which model performed best and how accurate was it?

Logistic regression was the best-fit model, with specificity 0.93, ROC 0.78, and 90% accuracy for predicting responders.

Aug 5, 2026

Upload by Eden