Lead/Presenter: Ralph Ward, COIN - Charleston
All Authors: Ward RC (Health Equity and Rural Outreach Innovation Center, Charleston), Weeda,E (Medical University of South Carolina), Taber, DJ (Health Equity and Rural Outreach Innovation Center, Charleston) Axon, RN (Health Equity and Rural Outreach Innovation Center, Charleston) Gebregziabher, M (Health Equity and Rural Outreach Innovation Center, Charleston)
Objectives:
To develop and validate a risk score to screen for frailty using administrative data from Veterans Affairs (VA) patients.
Methods:
In a development cohort of 127,350 randomly sampled patients admitted to VA hospitals between 10/1/2015 and 9/30/2016, we used hierarchical clustering analysis to identify groups of patients with diagnostic codes associated with frailty and high resource use. We developed a frailty risk score based only on diagnostic codes that were at least twice as prevalent in the frailty cluster versus the rest of the cohort. The score was validated in two ways. First, we determined how well the score predicted adverse outcomes in a separate validation cohort of 75,453 VA inpatients with type 2 diabetes and followed between 2000 and 2010. Second, for randomly selected patients (n = 9,565), we determined frailty status based on clinical observations recorded as part of VA clinical reminder decision support modules and assessed how well the frailty score predicted frailty status in this group using a univariate logistic regression model.
Results:
In the development cohort, those in the frailty cluster were older (mean age of 71.9 versus 67.3 years overall) and had a substantially higher rate of mortality (22.7% versus 11.8%). Upon validation, patients in the highest frailty risk score group had a higher adjusted odds of hospital stays lasting > 10 days (odds ratio [OR] = 2.39; 95% confidence interval [CI] = 2.15-2.66), 30-day all-cause mortality (OR = 2.14; 95%CI = 1.84-2.49) and 30-day readmission rate (OR = 2.60; 95%CI = 2.34-2.89), as compared to those in the lowest frailty risk group. Validation with frailty status based on clinical observation showed the score had high predictability: OR 1.20; 95%CI = 1.19-1.21, c-statistic 0.82 for each unit-increase in the frailty score.
Implications:
We produced a highly predictive model to identify Veterans who are at increased risk of poor outcomes due to frailty. The score's performance was very strong in predicting clinically observed frailty status. High risk scores were associated with substantially higher odds of mortality, readmission and hospital resource utilization.
Impacts:
Frailty score applications include use by clinicians for triaging patients, by hospitals for resource allocation and by health systems for better risk adjustment when comparing quality outcome measures.