A Novel Model for Predicting Rehospitalization Risk Incorporating Physical Function, Cognitive Status, and...

Area of Science:

• Health Informatics
• Clinical Prediction Models
• Natural Language Processing in Healthcare

Background:

• Hospital readmissions are a significant concern in the US healthcare system.
• Existing readmission risk models often overlook crucial patient factors like physical function, cognitive status, and psychosocial support.
• Structured electronic health record data inadequately capture these vital aspects.

Purpose of the Study:

• To develop a predictive model for 30-day all-cause hospital readmissions.
• To incorporate patient physical, cognitive, and psychosocial information into readmission risk assessment.
• To identify hospitalized patients at high risk for readmission.

Main Methods:

• Clinician focus groups identified key language related to physical, cognitive, and psychosocial factors.
• Natural language processing (NLP) was used to extract this language from electronic health records of 30,000 inpatients across three hospitals.
• A readmission prediction model was trained on 75% of the data and validated on the remaining 25% and hospital-specific subsets.

Main Results:

• The NLP-derived language was aggregated into 35 variables; the final model included 16 variables.
• The model demonstrated good performance with a validated C-statistic of 0.74 and was well-calibrated.
• Hospital-specific validation yielded C-statistics ranging from 0.70 to 0.75.

Conclusions:

• A 30-day readmission risk model utilizing NLP to identify physical, cognitive, and psychosocial issues performs comparably to leading models.
• The model's strength lies in its use of clinically relevant factors, automation, and scalability.
• Future research can explore interventions targeting identified risks to reduce hospital readmissions.