Predicting non-response in patient-reported outcome measures: results from the Swiss quality assurance programme in cardiac inpatient rehabilitation

Abstract

Abstract Background Quality assurance programmes measure and compare certain health outcomes to ensure high-quality care in the health-care sector. The outcome of health-related quality of life is typically measured by patient-reported outcome measures (PROMs). However, certain patient groups are less likely to respond to PROMs than others. This non-response bias can potentially distort results in quality assurance programmes. Objective Our study aims to identify relevant predictors of non-response during assessment using the PROM MacNew Heart Disease questionnaire in cardiac rehabilitation. Methods This is a cross-sectional study based on data from the Swiss external quality assurance programme. All patients aged 18 years or older who underwent inpatient cardiac rehabilitation in 16 Swiss rehabilitation clinics between 2016 and 2019 were included. Patients’ socio-demographic and basic medical data were analysed descriptively by comparing two groups: non-responders and responders. We used a random intercept logistic regression model to estimate the associations of patient characteristics and clinic differences with non-response. Results Of 24 572 patients, there were 33.3% non-responders and 66.7% responders. The mean age was 70 years, and 31.0% were women. The regression model showed that being female was associated with non-response [odds ratio (OR) 1.22; 95% confidence interval (CI) 1.14–1.30], as well as having no supplementary health insurance (OR 1.49; 95% CI 1.39–1.59). Each additional year of age increased the chance of non-response by an OR of 1.02 (95% CI 1.02–1.02). Not being a first language speaker of German, French or Italian increased the chance of non-response by an OR of 6.94 (95% CI 6.03–7.99). Patients admitted directly from acute care had a higher chance of non-response (OR 1.23; 95% CI 1.10–1.38), as well as patients being discharged back into acute care after rehabilitation (OR 3.89; 95% CI 3.00–5.04). Each point on the cumulative illness rating scale total score increased the chance of non-response by an OR of 1.05 (95% CI 1.04–1.05). Certain diagnoses also influenced the chance of non-response. Even after adjustment for known confounders, response rates differed substantially between the 16 clinics. Conclusion We have found significant non-response bias among certain patient groups, as well as across different treatment facilities. Measures to improve response rates among patients with known barriers to participation, as well as among different treatment facilities, need to be considered, particularly when PROMs are being used for comparison of providers in quality assurance programmes or outcome evaluation.