Article
Development and validation of the PROPERmed instrument to identify older patients in general practice at risk of hospital admissions: an individual participant data meta-analysis (IPD-MA)
Search Medline for
Authors
Published: | September 11, 2019 |
---|
Outline
Text
Background: Elderly patients with multimorbidity and polypharmacy are at risk of inappropriate prescriptions and undertreatment, which may lead to increased number of hospital admissions (HAs). For designing preventive interventions and applying them to heterogeneous primary care populations, it would be helpful to identify those patients at highest risk of HAs.
Objective: To develop and validate a prognostic model to predict HAs within six-month follow-up in older patients in general practice with ≥1 chronic condition and ≥1 chronic medication.
Methods: We harmonized individual participant data (IPD) from four cluster-randomized trials conducted in the Netherlands and Germany.The model was developed using logistic regression with a stratified-intercept to account for between-study heterogeneity in baseline risk. Variables were selected in complete cases and then refitted in multiply imputed data to obtain the final model equation. Between-study heterogeneity in predictor effects was explored by meta-analytic techniques and interaction terms accounted for it, if indicated. Predictive performance and generalisability were derived by bootstrap internal validation and internal-external cross-validation (IECV), respectively.
Results: We included 3,832 participants with a mean age of 78 years, 60% females, 95% living at home, 3 chronic conditions on average as well as 7 chronic prescriptions. Selected predictors related to demographics (e.g., age), disease and health status (e.g., heart failure, pain), and medication-related risks (internal performance: calibration slope of 0.85 [0.33;1.36] and c-statistics of 0.64 [0.62;0.66]; generalisability: pooled c-statistics in the IECV loop of 0.61). Development sample sizes, event frequencies in validation data and effect heterogeneity may partly explain obtained performance estimates.
Discussion: This very first IPD-based prediction model for HAs in older patients already performed satisfactorily. Nevertheless, it may be further improved, e.g., by considering preventable HAs instead of all-cause HAs.
Take home message for practical use: IPD-based modelling is a promising approach to address the challenging prediction of future HAs in general practice.