The PRED'IC algorithm achieved an AUC of 0.81 for predicting 90-day readmission in decompensated heart failure, demonstrating higher specificity (69.7% vs 35.8%) than cardiologist assessment.
Observational (n=120)
No
Does the PRED'IC predictive algorithm improve the prediction of 90-day heart failure-related readmission compared to cardiologist assessment in older patients hospitalized for congestive heart failure?
An AI-based predictive algorithm demonstrated fair diagnostic accuracy and higher specificity than clinical judgment for predicting 90-day readmission in heart failure patients, highlighting its potential to optimize follow-up strategies.
Absolute Event Rate: 0.81% vs 0.65%
Heart failure is a leading cause of hospital readmission globally. Few studies have compared the performance of artificial intelligence-based prediction models directly with expert clinical judgment. The objective was to compare the performance of the PRED’IC predictive algorithm with cardiologist assessments for predicting 90-day readmissions following hospitalization for acute decompensated chronic heart failure. This retrospective, single-center observational study included 120 patients aged ≥ 60 years hospitalized for congestive heart failure at Montpellier University Hospital between 2017 and 2019. The PRED’IC algorithm estimated individual readmission risks using clinical, biological, and administrative data from electronic medical records. Three cardiologists independently classified patients as low, moderate, or high risk. The reference standard was 90-day heart failure-related readmission. Performance metrics included sensitivity, specificity, positive and negative predictive values (PPV, NPV), and area under the receiver operating characteristic curve (AUC). Exploratory subgroup analyses assessed fairness across age and sex. The cohort comprised 120 patients (mean age 81 ± 9.8 years; 49.2% women); 11 (9.2%) were readmitted within 90 days. PRED’IC identified 42 (35%) patients as high risk versus 80 (67%) by cardiologists. PRED’IC achieved an AUC of 0.81 (95% CI 0.67–0.93), with sensitivity 81.8% (95% CI 59.0–100), specificity 69.7% (95% CI 61.1–78.4), PPV 21.4% (95% CI 9.0–33.8), and NPV 97.4% (95% CI 93.9–100). Cardiologists demonstrated higher sensitivity (90.9%) but lower specificity (35.8%). Subgroup analyses showed consistent algorithmic performance across sex and age categories, with slightly higher specificity among older men. The PRED’IC algorithm achieved fair diagnostic accuracy and higher specificity than cardiologist assessment while maintaining excellent negative predictive value. These findings suggest that artificial intelligence-based prediction tools can complement clinical judgment by improving risk stratification and optimizing follow-up strategies for heart failure patients. Prospective multicenter evaluations are warranted to assess real-world clinical impact.
Mercier et al. (Thu,) conducted a observational in Decompensated chronic heart failure (n=120). PRED'IC predictive algorithm vs. Cardiologist assessment was evaluated on Area under the receiver operating characteristic curve (AUC) for predicting 90-day heart failure-related readmission (95% CI 0.67-0.93). The PRED'IC algorithm achieved an AUC of 0.81 for predicting 90-day readmission in decompensated heart failure, demonstrating higher specificity (69.7% vs 35.8%) than cardiologist assessment.