Insomnia-LCA classifier: an open web application for insomnia subtype classification using latent class analysis

Area of Science:

• Psychiatry and Behavioral Science
• Computational Psychology

Background:

• Insomnia heterogeneity complicates diagnosis and treatment.
• Latent class analysis (LCA) has identified distinct insomnia subtypes.
• Previous LCA findings were limited to original datasets, hindering broader application.

Purpose of the Study:

• To develop a practical, open-source tool for classifying insomnia subtypes.
• To enable the deployment and testing of LCA-derived phenotypes using the Insomnia Severity Index (ISI).

Main Methods:

• Developed the insomnia-LCA classifier, a web application.
• Utilized class priors and conditional response probabilities from a prior LCA study.
• Applied the classifier to assign new ISI profiles to four identified subtypes: no insomnia (NI), subthreshold insomnia (SI), high insomnia risk (HI), and predominant daytime symptoms (DS).
• Enabled individual and batch processing of ISI responses.

Main Results:

• The classifier demonstrated high accuracy (accuracy = 0.999) and reliability (Cohen's kappa = 0.999) when reclassifying the original dataset.
• Synthetic profiles generated by the tool behaved as expected.
• Outputs include class probabilities, modal assignment, ISI scores, and profile comparisons.

Conclusions:

• The insomnia-LCA classifier is a practical and reproducible tool.
• Facilitates the application and validation of insomnia subtypes in clinical research.
• Aids in understanding and utilizing data-driven insomnia phenotypes.