Investigating the relationship between Severity of Illness and the modified Rankin Scale in ischemic stroke patients with response mapping

Authors: Dewilde S, Annemans L, Thijs V
Published in: Value in Health. 2015 Nov;18(7):A405

Abstract

Objectives: To investigate the relationship between the APR-DRG severity of illness index (SOI) used to determine the reimbursement/payment level for stroke hospitalisation, and the modified Rankin Scale (mRS), the most frequently used outcome measure in stroke trials.

Methods: Data from all ischemic stroke hospitalizations from a teaching hospital in Belgium were collected between 2006 and 2009. Data collection included the SOI, patient characteristics (age, gender), risk factors (hypertension, smoking, hypercholersterolemia, diabetes, CAD, PAD, previous stroke), clinical parameters (aorta-atherosclerosis, cancer, TOAST (large-artery atherosclerosis, cardioembolism, small-vessel occlusion, other), microbleeds, atrial fibrillation, akinesiahypokinesa, endocarditis, MI), functional scales (NiH, mRS), repeat events. An ordered multinomial regression estimated the relationship between the SOI and these covariates. Using the regression parameters and the mean value of the other covariates, predicted values were generated for each combination of the mRS and the SOI. Monte Carlo simulations generated a set of predicted SOI values per patient (response mapping). Data from 2010 and 2011 were used for validation of the regression model.

Results: 559 hospitalizations were used for the regression analysis. Factors that were discriminating in predicting the correct SOI category were the mRS (p<0.001), age (p=0.0017), NIH at arrival in hospital (p<0.001), TOAST (p=0.0129), atrial fibrillation (0.0217) and repeat in-hospital event (p=0.0031). Generating Monte Carlo predicted values demonstrated good concordance across SOI levels at the population level (2.3% vs 2.0% categorized in SOI1, 49.8% vs 50.3% in SOI2, 32.9% vs 31.4% in SOI3, 15.4% vs 16.0% in SOI4, for the true and the simulated proportions respectively), and the root mean-squared error was 0.33. Validation of the data with 588 hospitalizations from 2010 and 2011 confirmed the good fit of the model.

Conclusions: Factors affecting the reimbursement/payment level of a stroke admission are age, location of the ischemia, atrial fibrillation, scores on stroke functional scales and new in-hospital events.