JMIR Publications

ABSTRACT

Background:

Individuals with heart failure (HF) have a high burden of health care utilization, cost, and morbidity in the year following hospitalization for an acute HF exacerbation. Frailty, described as increased vulnerability to adverse events, is common among those with HF and increases with age1. Health systems worldwide are integrating automated tools within electronic health records (EHR) to measure frailty. However, the consideration of longitudinal data to measure frailty to better predict outcomes among those with HF is lacking2-5.

Objective:

We sought to evaluate the predictive value of adding longitudinal data to a standard frailty index and evaluate prediction of 1-year outcomes in patients with heart failure.

Methods:

This was a retrospective cohort study using national Veterans Health Administration (VA) data. Veterans aged 50 and older with an index hospital admission for heart failure between calendar years 2016-2019 were included. Subjects had 2+ primary care visits in past three years before date of admission to indicate regular use of VA care and a documented ejection fraction (EF). We used the validated VA frailty index (FI) which captures 31 deficits in health based on International Classification of Diseases 10th and Current Procedural Terminology codes.6 We fit a linear line to three calculated FIs for each year prior to index date of admission and reported the slope and intercept individually. This method provides a three-year longitudinal estimate of frailty at admission. We used 1-year, all-cause mortality following the index of admission as the outcome. We reported the Area under the curve (AUC) for predicting outcomes using logistic regression. We estimated two AUCs: a) FI at the time of admission (AUCFI) and b) FI at time of admission plus slope and intercept (AUCFT+FI). Changes in the AUCs are reported as percentage of improvement [ΔAUC=100% * (AUCFT+FI-AUCFI)/AUCFI]. We recursively calculated the AUCs and ΔAUC by including patients whose FI at admission were less than 0.1 and at each step, increased the FI level by 0.01 to 0.4.

Results:

In total, 54,774 Veterans were included. Average age was 73±10 years (BMI 30±7 Kg/m2; 98% male; 55% White, Table 1). Figure 1 shows the AUCFI and AUCFT+FI across the distribution of frailty ranges from prefrail (FI≥0.10 to <0.2) to frail. An FI of 0.2 is equivalent to an accumulation of 7 deficits among 31 variables. The ΔAUC is also displayed. Across all AUCs evaluated for Veterans at different thresholds of FI the AUC improved 4.1% by adding FT to the FI. The highest ΔAUC observed at FI of 0.13 to 0.16 (24%) and reduced to 10% for FI of 0.2 and greater.

Conclusions:

In a national cohort of Veterans admitted with HF, the addition of longitudinal frailty trajectory data resulted in a clinically significant (24%) improvement in one-year mortality prediction compared to a standard FI alone among patients in the prefrail range. In contrast, we observed a modest (4%) improvement in one-year mortality prediction in the overall population. Enhancing AUC prediction in prefrail range is clinically important as interventions to mitigate frailty may be most impactful in this population7. In particular, prefrail patients may benefit from interventions, such as cardiac rehabilitation, to improve frailty status and cardiovascular outcomes1. These results may not generalize to non-Veteran populations. The sample is predominately male but does include a diverse population by race, ethnicity and geographic distribution. In summary, methods for calculating frailty provide useful predictions of adverse outcomes among adults with HF. The addition of longitudinal data particularly improves predictions for prefrail patients with HF. These findings aide clinician and health system decision making as this population benefits most from interventions to slow or prevent progression of frailty.