Nonresponse bias: does it affect measurement of clinician behavior?

Abstract

Background: Previous studies of nonresponders have not assessed the effects of nonresponse on the accuracy of clinician behavior measurements. Knowledge of these effects is critical to both research and quality improvement.

Objective: To evaluate the hypothesis that nonresponders to a survey would not adversely affect the ability to measure rates of preventive services.

Research design: Four primary-care medical practices participating in a randomized clinical trial provided an unusual opportunity to compare the medical record-documented care of both responders and nonresponders to a survey of their patients.

Subjects: Three hundred forty-five nonresponders and 321 responders to a questionnaire requesting participation in the study.

Measures: Differences in patient characteristics and diseases and documentation of screening and management of tobacco use, hypertension, and hypercholesterolemia.

Results: Although the survey process resulted in a response rate of only 52.5% and some statistically significant differences in responder and nonresponder characteristics, there were no differences in management behavior regarding cardiovascular risk factors. Responders were more likely to have adjusted documentation of tobacco use (OR = 1.4), blood pressure measurement (OR = 9.8), and cholesterol testing (OR = 2.0), but not family history of cardiovascular disease. The most striking difference in subject characteristics was that 22.0% of nonresponders and only 12.1% of responders were tobacco users (P = 0.002).

Conclusions: This study confirms that survey nonresponders may have some different characteristics and risk factor screening rates than responders. However, if confirmed by others, nonresponders who have risk factors identified may not be managed differently than responders.