Predictors of Sleep Latency From the Multiple Sleep Latency Test: A Random Forest Investigation in a Community Sample

Area of Science:

• Sleep Medicine
• Computational Biology
• Chronobiology

Background:

• The multiple sleep latency test (MSLT) is crucial for diagnosing sleep disorders.
• Understanding factors predicting MSLT mean sleep latency (MSL) is vital for accurate diagnosis.
• Previous research has not fully leveraged machine learning on high-dimensional data to predict MSLT MSL.

Purpose of the Study:

• To identify key predictors of MSLT mean sleep latency (MSL) using machine learning.
• To explore the relationship between various factors and MSLT MSL in a large community sample.
• To investigate the influence of circadian preference on MSLT MSL.

Main Methods:

• Applied random forest (RF) algorithm on a dataset from the Wisconsin Sleep Cohort Study.
• Utilized 50 potential predictors including demographics, health, sleep (diary, polysomnography [PSG]), and circadian characteristics.
• Conducted regression analyses on the top 10 predictors identified by RF.

Main Results:

• RF model showed low explanatory value (R²=12%) for MSLT MSL.
• Key predictors included PSG sleep onset latency, circadian preference, caffeine use, and Epworth Sleepiness Scale scores.
• Morning circadian preference was associated with longer MSLT MSL compared to other preferences.

Conclusions:

• The complexity and variability of the MSLT limit predictive accuracy with current machine learning models.
• Circadian characteristics significantly influence MSLT MSL and must be considered.
• Further research is needed to improve the predictive power of MSLT MSL models.