Response bias: effect on outcomes evaluation by mail surveys after total knee arthroplasty.

BACKGROUND: Mail survey questionnaires are increasingly being used for follow-up evaluations to gauge satisfaction and performance after total joint arthroplasty. Responses to questionnaires are subject to a variety of possible biases. We evaluated response behavior in a mail survey of patients who had had a total knee arthroplasty.
METHODS: A ten-question survey that evaluated satisfaction, general health, and Knee Society knee function and clinical scores was mailed to 472 patients who had undergone consecutive primary total knee arthroplasties from 1996 to 1998. The 83% who responded were stratified as early, late, and repeat-mailing responders. The 17% who failed to respond after two mailings were considered nonresponders. All of the nonresponders were eventually contacted. The groups were compared with regard to their scores at the preoperative office visit, at the most recent office visit, and on the mail survey.
RESULTS: In the mail survey, the patients who responded earliest gave the highest satisfaction ratings and the nonresponders gave the poorest ratings (p < 0.001). Similarly, the mean Knee Society knee score (and standard deviation) was significantly higher for the early responders (82.7 +/- 19.0) than for the nonresponders (66.9 +/- 16.0), as was the mean function score (68.8 +/- 24.1 compared with 48.4 +/- 12.5) and the mean pain score (39.8 +/- 13.9 compared with 27.0 +/- 9.7) (all p < 0.0001). The change between the preoperative and mail survey Knee Society knee scores was significantly higher for the early responders (46.12 +/- 25.71) than for the nonresponders (28.45 +/- 23.62), as was the change in the mean function scores (18.87 +/- 22.52 compared with 5.34 +/- 20.05) and the change in the mean pain scores (23.57 +/- 17.76 compared with 10.67 +/- 12.93) (all p < 0.0001).
CONCLUSIONS: Patients who do not respond to mail surveys used for follow-up are unique in that they report significantly poorer outcomes than do responders. This potential response bias should be considered in all follow-up analyses. Because it may be difficult to attain 100% response rates in very large series of patients, division of the study cohort into more manageable segments is advised to achieve a more complete response rate. The assessment of patients who are lost to follow-up is an important and necessary component in the accurate analysis of outcomes after arthroplasty.