Determining the Importance of Clinical Modalities for NeuroDegenerative Disorders and Risk of Patient Injury Using...

Area of Science:

• Gerontology
• Medical Informatics
• Machine Learning

Background:

• Falls significantly reduce life expectancy in the elderly, particularly those with neurodegenerative disorders (NDD).
• Predictive models for fall-related injuries are crucial for improving patient outcomes and longevity.

Purpose of the Study:

• To explore Machine Learning (ML) on Electronic Health Records (EHR) for predicting time-to-event survival analysis of injuries.
• To investigate the influence of sensitive attributes like race, ethnicity, and sex on these predictive models.
• To assess the impact of combining multiple data modalities (medications, laboratory tests) on prediction accuracy and fairness.

Main Methods:

• Utilized multiple survival analysis methods on a cohort of 29,045 patients aged 65+ from PennMedicine.
• Included patients with NDD, Mild Cognitive Impairment (MCI), or other diseases.
• Compared prediction performance using the C-index and analyzed the role of medication and laboratory data, along with sensitive attributes.

Main Results:

• Medication features influenced Hazard Ratios (HR) variably based on NDD type.
• Black race was associated with increased fall/injury risk in models using only medication and sensitive attributes.
• Combining medication and laboratory data eliminated the association between Black race and increased fall/injury risk.
• Combined modality models demonstrated improved survival analysis performance across various methods.

Conclusions:

• Combining medication and laboratory data in survival models for NDD and MCI patients yields robust, unbiased predictions of fall/injury risk.
• Multimodal data integration enhances the accuracy and fairness of predictive models for elderly fall risk assessment.
• This approach mitigates biases related to race and ethnicity in injury prediction.