Patient-reported outcome and experience measures are increasingly important in health care and health research. The use of these measures is growing in the US and overseas, and performance measures that incorporate patient-reported outcomes are being considered, particularly in cancer. A major challenge for the use of these measures is patient non-response, especially for diseases such as cancer and dementia. A commonly used approach is to ask a proxy such as the patient’s spouse or child to complete the measure on their behalf. Proxy reporting is used in major surveys, including those used in pay-forperformance approaches. No standards exist regarding how to adjust for the use of proxyreported measures in analyses. As patients requiring proxies likely differ in important ways from those who can self-report, adjusting for these differences is important. In this paper, we evaluate the use of propensity score models when adjusting for proxy-reported data, including weighting, matching with replacement, and non-parametric multiple imputation. Additionally, because previous analyses using propensity scores for proxy reports have employed stepwise or p value based algorithms, we evaluated the sensitivity of our results to the inclusion of respondent-sensitive variables such as proxy reports of patient health status, as well as auxiliary covariates. Under all propensity score methods, estimates obtained from propensity scores using respondent-insensitive variables were different from those obtained when respondent-sensitive variables were incorporated in the propensity score. Propensity score methods have limitations in these contexts and their assumptions should be carefully examined.
History
Publication title
Health Services and Outcomes Research Methodology
Volume
20
Pagination
40-59
ISSN
1387-3741
Department/School
Menzies Institute for Medical Research
Publisher
Springer New York LLC
Place of publication
United States
Rights statement
Copyright 2019 Springer Science+Business Media, LLC, part of Springer Nature