New models predict risk of readmission after PCI

Researchers have developed two models to predict patients' risk of readmission within 30 days of percutaneous coronary intervention (PCI).

The models were based on data from all 30-day readmissions after PCI at nonfederal Massachusetts hospitals in 2005 to 2008: 3,760 readmissions out of a total of 36,060 PCI patients surviving to initial discharge. Two-thirds of the patients were used to develop the multivariable models and the other third was used to validate them. Results were published by Circulation: Cardiovascular Quality and Outcomes on July 2.

The first model used only variables known before PCI and showed that significant predictors of readmission included older age, female sex, Medicare or state insurance, congestive heart failure and chronic kidney disease. The second model used variables known at discharge and found readmissions to be associated with lack of a beta-blocker prescription at discharge, post-PCI vascular or bleeding complications, and extended length of stay. The C-statistic, a measure used with logistic regression analysis to indicate the ability of a model to predict the studied outcome (with a range of 0.50 to 1.00, with higher values indicating higher predictive ability), was calculated for both models. The pre-PCI model had a C-statistic of 0.68, which was modestly improved to 0.69 with the addition of the post-PCI variables.

The models could help clinicians and hospitals risk-stratify patients by readmission risk and potentially develop interventions to reduce 30-day readmissions in high-risk patients, the study authors said. The two models could potentially lead to different interventions, for example, more intensive case management or changes in the surgical plan for patients identified as high risk by the pre-PCI model or involvement of home care or more rapid follow-up in patients identified by the post-PCI model.

The causes of the observed associations are not entirely certain, the authors acknowledged. For example, beta-blocker use may actually reduce readmissions, but it's also possible that prescriptions are a marker of high-quality, careful discharge or that patients with more stable blood pressure are likely to both receive beta-blockers and to not be readmitted. The study was also limited by its use of only Massachusetts hospitals and an inability to distinguish between preventable and unpreventable readmissions. The models also include a large number of variables (11 pre-PCI and 19 post) and so will likely require information technology tools for implementation.