Novel predictors of infection-related rehospitalization in older patients with heart failure in Japan

Abstract

Aim: Rehospitalization of patients with heart failure (HF) incurs high health care costs and increased mortality. Infection-related rehospitalizations in patients with HF occur frequently, and the risk increases with age. This study aimed to identify the factors associated with infection-related rehospitalizations in older patients with HF.

Methods: Demographic, clinical, and pharmacological data from 1061 patients with acute HF who were enrolled in the Kochi Registry of Subjects With Acute Decompensated Heart Failure (Kochi YOSACOI study) were analyzed. Additionally, a machine learning approach was applied in addition to the traditional statistical analysis model. Of the patients hospitalized for HF, 729 were ultimately analyzed.

Results: During the 2-year postdischarge follow-up period, 121 (17%) patients were readmitted for infections. Logistic regression analysis identified a Japanese Cardiovascular Health Study (J-CHS) score of ≥3 (odds ratio, 1.83 [95% confidence interval, 1.18-2.83]; P = 0.007) at discharge as a key factor for infection-related rehospitalizations. Machine learning models confirmed that a higher J-CHS score and lower estimated glomerular filtration rate (eGFR) increased the risk of infection-related rehospitalizations. Decision tree analysis classified the risk into high (J-CHS score ≥3), medium (J-CHS score <3; eGFR ≤35.0) and low (J-CHS score <3; eGFR >35.0) groups.

Conclusions: Infection-related rehospitalizations occur in older patients with HF and are associated with frailty and eGFR. These findings provide valuable insights for health care providers to better manage the risk of infection-related rehospitalizations in older patients with HF, potentially improving patient outcomes. Geriatr Gerontol Int 2025; 25: 543-552.

Keywords: J‐CHS score; decision tree; frail; heart failure; infection‐related rehospitalization.

Figure 1

Shapley additive explanation (SHAP) values for each variable were calculated from the individual values of each patient within the prediction model. Each data point is represented by a dot with a color based on the variable value. The red dots indicate high values for a particular patient's variable, whereas the blue dots indicate low values. For binary categorical variables, a blue dot indicates the absence of the category, and a red dot indicates its presence. By visualizing these patient‐specific SHAP values, the relationship between the value of each variable (indicated by the dot color) and its impact on the model output becomes clear. BMI, body mass index; BNP, brain natriuretic peptide; COPD, chronic obstructive pulmonary disease; eGFR, estimated glomerular filtration rate; GNRI, Geriatric Nutritional Risk Index; J‐CHS, Japanese Cardiovascular Health Study; MRA, mineralocorticoid receptor antagonist; NYHA, New York Heart Association; SBP, systolic blood pressure.

Figure 2

Shapley additive explanation (SHAP) main effect value for the (a) Japanese Cardiovascular Health Study (J‐CHS) score and (b) estimated glomerular filtration rate (eGFR). SHAP values for each variable were calculated from the individual values of each patient within the prediction model. Each data point is represented by a dot with a color based on the variable value. By visualizing these patient‐specific SHAP values, the relationship between each variable value (indicated by the dot color) and its impact on the model output becomes clear. (c) The interaction effects of J‐CHS score influence the predicted probability of infection‐related hospitalization based on age. The SHAP interaction values quantify these effects and show the interaction between age and the J‐CHS score. When the SHAP interaction values (y axis) are plotted as a function of patient age (x axis) and the value of the interacting variable (indicated by the color of the dot—red for a J‐CHS score of ≥3 and blue for a J‐CHS score of <2), trends in variables and their values that have a greater interaction effect emerge. In the case of age, an interaction effect with a J‐CHS score of ≥3 becomes apparent at approximately 80 years of age.

Figure 3

Decision tree analysis used to stratify the probability of infection‐related hospitalization. Patients were classified into the high‐risk (Japanese Cardiovascular Health Study [J‐CHS] score ≥3; n = 394), medium‐risk (J‐CHS score <3; eGFR ≤35.0; n = 83) and low‐risk groups (J‐CHS score <3; eGFR >35.0; n = 252). CDs, cardiovascular deaths; eGFR, estimated glomerular filtration rate; NCDs, non‐cardiovascular deaths.