OBJECTIVES: Fall injuries are most prevalent among the elderly population leading to medical complications and financial burden. Therefore, predicting the risk of fall can help avert fall incidences and effectively capture the associated risk factors.

METHODS:

A total of 100,061 subjects aged 60 years and older were considered for the index period 2019-2021, with 35,851 incident subjects identified in the event cohort. The intrinsic factors for predicting fall risk included comorbidities and demographic characteristics, which were accrued from both claims and electronic health records (EHR) data of Optum® de-identified Market Clarity database. Patients signs and symptoms that can accentuate a fall risk were identified from EHR data. Supervised ML techniques were used to develop the algorithm for fall prediction, which included logistic regression, XGBoost, and random forest techniques. Using the integrated database, 39 predictors were used for testing and training purposes, considering 80:20 ratio for training the model. A recursive feature elimination technique was used to identify the most relevant independent variables impacting the fall forecast.

RESULTS:

In the logistic regression, it was observed that older population (OR: 1.76), having cognitive impairment (OR: 1.54), past brain injuries (OR: 2.24), lower extremity orthopedic conditions (OR: 1.36), parkinsonism (OR: 2.13), peripheral neuropathies (OR: 1.12), hypertension (OR: 1.16), weakness (OR: 1.91) and dizziness (OR: 1.14) have a significant impact on the propensity to fall. From patient’s signs and symptoms data, past fall history (1.39) and the presence of dementia (1.21) were found to be important predictors of fall risk. This model depicted an accuracy level of 68.65%, with a sensitivity score of 0.77 and specificity score of 0.54, which was substantiated by XGBoost and Random Forest models.

CONCLUSIONS:

Linking EHR and claims data has provided a broader spectrum of information for a more in-depth analyses on the prediction of future falls.