A higher machine learning-based automated clinical frailty scale score was independently associated with an increased risk of all-cause death and HF hospitalization (HR 1.38; 95% CI 1.10-1.74).
Cohort (n=731)
Yes
Does a machine learning-based automated frailty rating predict adverse outcomes in elderly patients with chronic heart failure?
A machine learning-based automated clinical frailty scale is an independent predictor of all-cause death and heart failure hospitalization in elderly patients with chronic heart failure.
Effect estimate: HR 1.38 (95% CI 1.10-1.74)
p-value: p=<0.001
Abstract Background Frailty is associated with worse clinical outcomes in patients with heart failure (HF). Although the clinical frailty scale (CFS) is a widely used tool for assessing frailty and outcome prediction, its subjectivity and dependence on physician’s expertise limit its clinical utility. Recently, we have developed a machine learning-based automated CFS rating program application that analyses gait motion and outputs continuous values, showing strong agreement with assessments by 10 trained cardiologists. However, it is unclear whether the automatically rated CFS can stratify the risk of adverse events. Purpose We sought to examine the association between the machine learning-based CFS and clinical outcomes in elderly patients with HF. Methods This prospective study enrolled ambulatory chronic HF patients aged 75 or older from 17 sites between January 2020 and September 2024. A total of 731 patients (mean age 82±4 years, left ventricular ejection fraction LVEF 55 interquartile range 42–65%) were assessed using the automated CFS rating system, output as continuous scores ranging from 3.0 to 7.0, and classified into three frailty groups based on the scores; low ( 4.0), intermediate (4.0 to 5.0), and high (≥ 5.0). The primary outcome was a composite of all-cause death and HF hospitalisation. Results Patients were classified into three frailty groups: low (n = 398, 54.5%), intermediate (n = 216, 29.6%), and high (n = 117, 16.0%). Patients with intermediate and high frailty groups showed older age, predominantly female and higher N-terminal pro-brain natriuretic peptide (NT-proBNP) levels, whereas lower haemoglobin and albumin levels compared to those with low frailty group. Detailed gait analysis revealed that higher frailty scores were significantly associated with decreased gait speed, ankle push-off speed, and elbow angle during walking, while cornering time and variability in spinal posture increased (P 0.001 for all comparisons). During a median follow-up of 584 (interquartile range 286–926) days, cumulative incidences of the primary outcome in the low, intermediate, and high groups were 16.6%, 25.9%, and 27.4%, respectively (P 0.001) (Figure 1). A multivariable regression analysis showed that a higher automated CFS was independently associated with a higher risk of the primary outcome (HR 1.38, 95% CI 1.10–1.74) even after adjustments for age, sex, systolic blood pressure, prior HF history, smoking, diabetes, blood urea nitrogen, sodium, haemoglobin, NT-proBNP, LVEF, and New York Heart Association functional class (Figure 2). Furthermore, when incorporating MAGGIC risk score, a higher automated CFS was independently associated with worse clinical outcomes (HR 1.39, 95% CI 1.14-1.69). Conclusions The machine learning-based automatically rated CFS was independently associated with adverse outcomes in elderly patients with HF, suggesting that the automated frailty scoring system is a reliable tool for risk stratification.Survival analysis Multivariable Cox regression
Tamura et al. (2025) conducted a cohort in chronic heart failure (n=731). Machine learning-based automated clinical frailty scale (CFS) rating was evaluated on Composite of all-cause death and HF hospitalisation (HR 1.38, 95% CI 1.10-1.74, p=<0.001). A higher machine learning-based automated clinical frailty scale score was independently associated with an increased risk of all-cause death and HF hospitalization (HR 1.38; 95% CI 1.10-1.74).