<p>Composite Outcomes of Mortality and Readmission in Patients with Heart Failure: Retrospective Review of Administrative Datasets</p>

Abstract

Background: Controlling the quality of care through readmissions and mortality for patients with heart failure (HF) is a national priority for healthcare regulators in developed countries. In this longitudinal cohort study, using administrative data such as hospital discharge forms (HDFs), emergency departments (EDs) accesses, and vital statistics, we test new covariates for predicting mortality and readmissions of patients hospitalized for HF and discuss the use of combined outcome as an alternative. Methods: Logistic models, with a stepwise selection method, were estimated on 70% of the sample and validated on the remaining 30% to evaluate 30-day mortality, 30-day readmissions, and the combined outcome. We followed an extraction method for any-cause mortality and unplanned readmission within 30 days after incident HF hospitalization. Data on patient admission and previous history were extracted by HDFs and ED dataset. Results: Our principal findings demonstrate that the model's discriminant ability is consistent with literature both for mortality (AUC=0.738, CI (0.729-0.748)) and readmissions (AUC=0.578, CI (0.562-0.594)). Additionally, the discriminant ability of the composite outcome model is satisfactory (AUC=0.675, CI (0.666-0.684)). Conclusion: Hospitalization characteristics and patient history introduced in the logistic models do not improve their discriminant ability. The composite outcome prediction is led more by mortality than readmission, without improvements for the comprehension of the readmission phenomenon.

An evidence-based data science perspective on the prediction of heart failure readmissions

“…19 Among HFR studies, FPR often exceeds 0.25 before the TPR exceeds 0.5. 11,20,21 Simply put; the false alarm rate often reaches 25% before 50% of discharges resulting in readmission are correctly classified. If an FPR greater than 0.25 were unacceptable, then a C-statistic or AUROC based on the full ROC curve would be misleading.…”

“…Among the comparatively few studies that report it, precision typically ranges from 0.09 to 0.44, meaning that 56 – 91% of predictions for readmission are often incorrect. 3,21,24,28,29 If hospitals acted on the results of such predictions, they would invest their limited time and resources into expectations that are, more often than not, false alarms.…”

“…Predicting HF-related mortality, HFR beyond 30 days, or predicting which patients have HF often leads to greater success than predicting 30-day all-cause HFR. 4,5,10,13,14,20,21,31,33,35 As opposed to using a strong diagnostic such as elevated cardiac troponin to predict whether a patient has HF or will expire within one year, predicting 30-day all-cause HFR often means using data with a weak signal to predict whether a patient will readmit for any unplanned cause within a relatively short period of time. The inherent uncertainties of this problem make it a grand predictive challenge and whether it outstrips the classification problems to which machine learning is often successfully applied is unknown.…”

“…Among HFR studies that report positive predictive values (PPV; precision) and negative predictive values (NPV), only 19% – 44% of predictions for readmission were correct, while 80 – 96% of predictions for non-readmission were correct. 21,24,28 However, no HFR study has explicitly asked whether or why predicting negative outcomes (non-readmissions) is easier than predicting positive ones (readmissions) or how the ability to predict non-readmissions can be used.…”

PRADOC: a trial on the efficiency of a transition care management plan for hospitalized patients with heart failure in France

“…Aside of clinical determinants, it was demonstrated that potential determinants of morbidity in heart failure (HF) patients include demographic factors, 1 use of care, 2,3 primary care organization, 4 primary care accessibility, 1 and hospital care organization 5 . Health service organizations may have transition care programmes that facilitate coordination between health care facilities 4,6 .…”