Longitudinal trajectories of hospitalizations in US older adults with heart failure.

Abstract

Background: Little is known about long-term risks of hospitalizations among heart failure (HF) older adults at a national level. We identified long-term patterns of hospitalizations following the diagnosis of HF and assessed whether high-risk patients can be identified prior to discharge.

Methods: This is a retrospective cohort study from 2010 to 2020. Fee-for-service Medicare beneficiaries with newly diagnosed HF during an inpatient stay and followed for up to 5 years. Group-based trajectory models (GBTM) identified 4 trajectories of all-cause hospitalizations during follow-up. Least absolute shrinkage and selection operator regression was used to identify patients' baseline characteristics to predict their hospitalization trajectories. Model discrimination was assessed using C-statistics, and calibration was evaluated using expected-to-observed ratios.

Results: Of the 84 597 Medicare beneficiaries with newly diagnosed HF (mean age: 77.4 [± 7.1] years and 58.3% male), we identified 4 distinct trajectories of hospitalizations: Group 1 (n = 19 340; 22.9%) had consistently "low risks" of hospitalization, Group 2 (n = 53 922; 57.9%) had elevated risks shortly after discharge ("high-to-low risk"), Group 3 (n = 5035; 9.8%) had elevated risks at later stages of illness ("low-to-high risk"), and Group 4 (n = 6300; 9.4%) had consistently "high risks" of hospitalization. Models were well-calibrated predicting hospitalization trajectories (compared to low-risk patients [Group 1]), and C-statistics were 0.65 (Group 2), 0.68 (Group 3), and 0.79 (Group 4). Discrimination was consistently better among patients ≤70 years, and calibration remained excellent across all subgroups with expected-to-observed ratios near 1.0.

Conclusions: High-risk HF patients could be identified at the time-of-diagnosis using routinely available clinical characteristics, enabling targeted interventions during critical periods to improve long-term outcomes.