Copyright © 2005 by the Johns Hopkins Bloomberg School of Public Health
LETTERS TO THE EDITOR |
RE: "STUDIES WITH LOW RESPONSE PROPORTIONS MAY BE LESS BIASED THAN STUDIES WITH HIGH RESPONSE PROPORTIONS"
1 Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, WA.
2 Department of Epidemiology, School of Public Health and Community Medicine, University of Washington, Seattle, WA.
3 Center for Health Studies, Group Health Cooperative of Puget Sound, Seattle, WA.
4 Department of Pharmacy, School of Pharmacy, University of Washington, Seattle, WA.
We agree with Stang and Jöckel’s (1) conclusion that if exposure misclassification is substantial and increases with the difficulty of respondent recruitment, studies with low response proportions may be less biased than studies with high response proportions. However, Stang and Jöckel acknowledged that there had been no studies of the magnitude of such potential misclassification in relation to the difficulty of respondent recruitment. They encouraged investigators to monitor exposure misclassification by response "wave."
Within a large case-control study of breast cancer, we used data from a validation substudy of prescription drug use (2) to assess whether information provided by respondents from whom it was more difficult to obtain an interview (defined as taking a longer time to recruit them) is indeed less accurate than information provided by more easily interviewed respondents. We conducted the substudy among participants who either were enrollees of the Group Health Cooperative health maintenance organization or reported filling their prescription drugs exclusively at two retail pharmacy chains. Three categories of medication were assessed in the substudy: statins, antidepressants, and antihypertensives. Of the three types of medication, antihypertensive agents were the most commonly used and are our focus here.
We computed sensitivity and specificity for use of any antihypertensive medications during the 2 years prior to the reference date (the date of diagnosis for cases and a similar assigned date for controls) as reported by the respondent during the in-person interview, using pharmacy records as the "gold standard." Women were considered "true users" if they had filled at least two prescriptions for antihypertensive drugs during the 2-year time period. We examined sensitivity and specificity by case-control status and time between first contact and interview date (<2 months, 
2 months) (table ). We used the date on which a letter describing the study was mailed as the date of first contact. The sensitivity and specificity of self-reported data on antihypertensive drug use were slightly lower for cases interviewed 2 or more months after first contact than for those interviewed promptly. However, these measures were similar for controls interviewed within 2 months and those interviewed later.
|
Using the pharmacy data as our measure of use of antihypertensive medications, we assessed the impact of restricting our analysis to study participants interviewed promptly (table ). With this approach, respondents who were interviewed within 2 months of letter contact had a 1.25-fold increased risk of breast cancer, whereas the "true" underlying odds ratio based on pharmacy data for all women in the substudy was 0.94. There was very little exposure misclassification among controls, but late-responding controls were much more likely to have used antihypertensive medication than were early responders (table ). Our finding that the "true" overall prevalence of exposure in controls was less biased when late responders were included is consistent with figure 2 in Stang and Jöckel’s paper (1). The lowest line in this figure shows that observed exposure prevalence is less biased when late responders are included, if exposure prevalence increases with the difficulty of recruitment and there is no exposure misclassification.
|
Including late responders decreased bias in our data. Our results are limited by the small number of respondents who were interviewed 2 or more months after first contact and may be the result of random variation. Nevertheless, our findings suggest that respondents who are difficult to interview do not provide less accurate information than respondents who are easily interviewed and argue in favor of continued vigilance in maintaining high response proportions in epidemiologic case-control studies. We encourage additional study of exposure misclassification by difficulty of respondent recruitment in mail, telephone, and additional in-person interview studies.
 |
REFERENCES |
|---|
 TOP REFERENCES |
|---|
- Stang A, Jöckel KH. Studies with low response proportions may be less biased than studies with high response proportions. Am J Epidemiol 2004;159:204–10.
[Abstract/Free Full Text] - Boudreau DM, Daling JR, Malone KE, et al. A validation study of patient interview data and pharmacy records for antihypertensive, statin, and antidepressant medication use among older women. Am J Epidemiol 2004;159:308–17.
[Abstract/Free Full Text]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  K. Olson Survey Participation, Nonresponse Bias, Measurement Error Bias, and Total Bias Public Opin Q, January 1, 2006; 70(5): 737 - 758. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Stang and K.-H. Jockel THE AUTHORS REPLY Am. J. Epidemiol., February 15, 2005; 161(4): 403 - 403. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||