Validating the Insomnia Severity Index and Its 3-Item Shortened Version Among Cancer Patients: A Machine Learning Approach

Background and Objective Insomnia is a prevalent yet often underrecognized symptom among cancer patients, who already face significant psychological and physical burdens. While the 7-item Insomnia Severity Index (ISI) is widely used to assess insomnia, its length can impose additional strain on patients undergoing treatment. A shortened, machine learning–derived version of the ISI, known as ISI-3m, has shown strong predictive performance in general populations but has not been validated among cancer patients.Methods We retrospectively analyzed data from 865 cancer patients who completed the ISI at their initial visit to the Stress Clinic at Asan Medical Center between 2017 and 2024. We first evaluated the full ISI’s psychometric properties using exploratory and confirmatory factor analysis (EFA and CFA) and internal consistency (McDonald’s omega). We then assessed the predictive performance of ISI-3m, using XGBoost models across binary and multiclass insomnia classification tasks, comparing it with four existing short forms.Results The ISI showed a unidimensional structure with excellent model fit (comparative fit index=0.964, standardized root mean square residual=0.028) and high internal consistency (ω=0.914). ISI-3m outperformed other short forms, achieving the highest accuracy for mild (0.935), moderate (0.857), and multi-class (0.827) insomnia classification. These findings affirm the ISI-3m’s robust predictive ability in cancer population.Conclusions The ISI-3m is a brief, valid, and effective screening tool for assessing insomnia severity among cancer patients. Its strong performance and reduced burden make it especially suitable for oncology care, supporting timely and efficient sleep assessments in high-demand clinical settings.