Sarcopenia: computable phenotypes and clinical outcomes.

点击翻译按钮获取中文摘要

PROJECT SUMMARY Sarcopenia is a generalized muscle condition that develops with aging and complicates many common chronic diseases, resulting in low muscle mass, weakness, and impaired physical function. Sarcopenia contributes to disability, increased hospitalizations, healthcare costs, and risk of death. Despite being under- recognized clinically, sarcopenia is a major public health concern, with the worldwide prevalence projected to increase by up to 72% in the next 30 years. However, limited knowledge of sarcopenia among clinicians, combined with time pressures in clinical encounters delay its detection, and limit opportunity for intervention or recruitment into clinical trials. To overcome this barrier to detecting sarcopenia, we propose to use advanced big data and machine learning methods to identify additional component variables predicting sarcopenia among the rich electronic health record (EHR) data and develop a validated and portable sarcopenia computable phenotype (which uses a computer algorithm to detect patient characteristics or outcomes from the EHR). This innovative proposal takes advantage of key resources at Indiana University and its affiliation with the Regenstrief Institute and the Indiana Network for Patient Care (INPC), a statewide multi-health system clinical data warehouse including >100 healthcare entities and >18 million unique patients with both coded and text-based data, combined with the ability to perform comprehensive musculoskeletal measurements in the Musculoskeletal Function Imaging and Tissue (MSK-FIT) Core funded through a NIAMS Core Center for Clinical Research grant (P30AR072581). Our long-term goal is to accurately identify patients with, or at risk for, sarcopenia and its consequences in order to provide targeted interventions. We hypothesize that by using medical informatics and machine learning innovations, computable phenotypes can identify patients with sarcopenia from the EHR, predict deficits in measured muscle strength and physical function, and prospectively predict risk of hospitalization and death. In Aim 1, we will categorize >2000 adult participants in the MSK-FIT Core with accessible EHR data, as either sarcopenic or nonsarcopenic according to measurements of muscle strength, muscle mass and physical performance. We will then use 75% of the MSK- FIT Core cohort to train machine deep learning algorithms to detect combinations of variables from these subjects’ EHR predicting whether the patient is sarcopenic or not sarcopenic. The performance of the resulting computable phenotype will then be tested in the remaining 25% of the MSK-FIT Core participants. In Aim 2, we will test the performance of the sarcopenia computable phenotype to detect a clinically meaningful phenotype in the entire INPC adult population (>18 million), by evaluating the ability to predict the rate of hospitalizations and death among patients rated as sarcopenic versus matched controls. Such a computable phenotype will then enable large scale targeted recruitment, pragmatic clinical trials, clinical evaluation and intervention.