A Sequence Clustering Approach to Mining Sleep Trajectories from Nursing Narratives and Structured Clinical Data

Area of Science:

• Neuroscience
• Health Informatics
• Rehabilitation Medicine

Background:

• Sleep quality is crucial for neurological patient recovery but difficult to monitor long-term during hospitalization.
• Nursing narrative notes are an underutilized data source for objective, longitudinal sleep tracking.

Purpose of the Study:

• To develop and implement a free software pipeline for monitoring sleep trajectories in post-acute neurorehabilitation patients.
• To integrate structured clinical data with unstructured nursing notes without increasing nursing workload.

Main Methods:

• Extracted and categorized 17,039 nighttime nursing annotations into four sleep quality states.
• Developed a random forest classifier (0.93 sensitivity, 0.94 specificity) to label sleep notes.
• Utilized sequence analysis (TraMineR) and hierarchical clustering (AGNES, Ward's method) to identify sleep trajectory clusters.

Main Results:

• Identified four distinct sleep trajectory clusters (silhouette=0.40) in 303 neurorehabilitation inpatients.
• Clusters showed unique profiles correlating with functional independence, socio-economic status, and hospitalization duration.
• One cluster with poor sleep quality was associated with precarious economic conditions; another with good sleep quality had severe functional impairment.

Conclusions:

• The developed pipeline effectively identified clinically meaningful sleep profiles from nursing notes.
• Sleep trajectories in neurorehabilitation are significantly shaped by functional status and social determinants.
• This approach offers objective insights into sleep patterns, supporting tailored patient recovery strategies.