Prediction of readmission or death in heart failure (HF) is of critical importance and is currently based on a limited number of variables selected by experts/literature. Artificial intelligence (AI) allows to include a non-limited and non-selected number of variables.

The objective was to predict the probability of HF readmission and/or death inpatients with a new diagnosis of HF, using medical records and machine learning models, without a priori.

This pilot monocentric study included all patients having a first HF admission between January 1st, 2015 and December 31, 2018, enrolled by data record (ICD-10) from the department of cardiology Paris Saint-Joseph Hospital. One thousand variables extracted from electronic health records/local hospital discharge database (PMSI) were used to create models. Data from the national PMSI were also included using the Hawkes process. Models were constructed on patients admitted between 2015 and 2017 to predict 1 and 3 months probability of HF readmission (n=905 and n=829 respectively) and HF readmission or mortality (n=923 and n=868 respectively). Discrimination was tested using the area under the ROC curves (AUC) inpatients admitted in 2018.

At 1 month, AUC was 0.56 [95%CI 0.47–0.66] (n=294) and 0.55 [95%CI 0.47–0.64] (n=302) for the HF readmission and the composite outcome respectively. At 3 months, AUCs were 0.51 [95%CI 0.41–0.59] (n=224) and 0.52 [95%CI 0.45–0.60] (n=235) respectively.

The predictive value of these models using AI in this population is limited and performed better at 1 month than 3 months. Potential biases include the accuracy of HF diagnosis (based on ICD-10) and the limited size of the derivation cohort compared to the number of variables included in the models. This approach should therefore be assessed in a larger population.