Correspondence
Lack of Health Insurance and Overall Health
N Engl J Med 2002; 346:626-627February 21, 2002
- Article
To the Editor:
Baker et al. (Oct. 11 issue)1 used disability data from the 1992 and 1996 surveys of the Health and Retirement Study to assess the effect of health insurance status on the risk of the development of a physical disability. Unfortunately, as pointed out in the article, the wording of the questions related to disability on the Health and Retirement Study surveys changed between 1992 and 1996.
The initial survey, in 1992, inquired about the performance of physical tasks. The wording was, “How difficult is it for you to [perform this task]?” and the potential answers were “not at all difficult,” “a little difficult,” “somewhat difficult,” “very difficult,” and “something you can't do at all.” In 1994, subjects were asked about the same tasks, but the question was changed to, “Do you have any difficulty with [this task]?” Subjects who reported difficulty were asked, “Is that a little or a lot of difficulty?” In 1996, the question was changed again and became, “Because of a health problem do you have any difficulty with [this task]?” The responses (“yes,” “no,” “can't do,” or “don't do”) were recorded for the same tasks as in the previous surveys.
To deal with these changes in format, Baker et al. restricted their analyses to participants who reported a new disability in 1996. They made the assumption that participants who answered “not at all difficult” in 1992 and then either “yes” or “can't do” in 1996 had a new disability. We examined the effect the wording change may have had on the outcome by using an experimental module from the 1994 survey in which 771 subjects were asked questions about disability in both the 1992 and the 1994 formats. Since the first part of the 1994 question format is almost identical to the 1996 format, we were able to compare the participants' responses to the old (1992) and new (1996) questions.
We used the algorithm described by Baker et al. to determine the number of participants in whom a new mobility- or agility-related disability would appear to have developed solely on the basis of the change in the question (Table 1Table 1
Participants Who Reported a New Disability in Mobility or Agility in 1994 as a Result of the Change in Question Format.). We found that 14 percent of the subjects would appear to have a new agility-related disability but that only 4 percent would appear to have a new mobility-related disability. Baker et al. reported that 17.1 percent of the continuously insured participants and 28.8 percent of the continuously uninsured participants had a new disability related to mobility (P<0.001) and that 32.2 percent and 38.7 percent, respectively, had a new agility-related disability (P=0.003). The lack of concordance with the results from the experimental module suggests that some of the differences between the insured and uninsured participants may have reflected the change in question format.It is not our intention to suggest that health insurance status does not have an important effect on health status and disability status. However, because of the change in question format in these surveys, it is difficult to determine the magnitude of the effect from the data used in the analysis. Repeating the analysis with data from 1996 through 2000 would allow the authors to estimate the true effect of insurance on disability with the use of identically worded questions.
1 ReferencesDavid J. Harrison, M.A.
Andrea K. Biddle, M.P.H., Ph.D.
University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-7411
davidharrison99@hotmail.com 1
Baker DW ,Sudano JJ ,Albert JM ,Borawski EA ,Dor A . Lack of health insurance and decline in overall health in late middle age. N Engl J Med 2001;345:1106-1112
Full Text | Web of Science | Medline
Author/Editor Response
The authors reply:
To the Editor: Harrison and Biddle present data suggesting that changes in the questions about physical functioning in the Health and Retirement Study could have led us to overestimate the rate at which new physical difficulties developed in the participants. They used an “experimental module” from the main 1994 interview as their gold standard to judge the validity of the measured outcomes. This approach is problematic, because although the questions in the experimental module were identical to the questions administered at base line (1992), the mode of administration was different (a face-to-face interview in 1992 in contrast to a self-administered questionnaire in the experimental module in 1994). It is therefore uncertain that the experimental module provides the most valid estimate of the rate of new physical difficulties. Moreover, we measured outcomes with data from the 1996 questions, not the 1994 questions used in the analysis conducted by Harrison and Biddle. It is unclear whether their results can be generalized to the 1996 data, because the 1996 questions had a different set of response options than either the 1992 or the 1994 questions.
It remains possible that changes in the survey questions may have led us to overestimate the true number of persons in whom new difficulties developed, particularly with respect to agility. However, we disagree that the “differences between the insured and uninsured participants may have reflected the change in question format.” This would be true only if any misclassification of the outcomes was larger among the uninsured participants than among the insured participants (i.e., if there was differential misclassification bias),1 and Harrison and Biddle do not present evidence suggesting such a bias. In general, unbiased error in the measurement of an outcome leads to underestimation of the true difference between groups (i.e., a bias toward the null).1 For example, we reported that the rate of new mobility difficulties was 17.1 percent among the continuously insured participants and 28.8 percent among those who were continuously uninsured. If both of these figures were overestimated by 4 percentage points, then the true rates of new difficulties with mobility would be 13.1 percent and 24.8 percent, respectively. Thus, the correct estimate of the crude relative risk of new mobility difficulties among the continuously uninsured participants as compared with those who were continuously insured would be 1.89, which is substantially higher than our original estimate of 1.68.
Therefore, our findings may have underestimated the increased risk of a decline in physical functioning in uninsured persons. We agree that it is important to examine whether the uninsured participants were at increased risk for the development of new physical difficulty between 1996 and 2000, a period when the questions about physical functioning in the Health and Retirement Study remained the same.
1 ReferencesDavid W. Baker, M.D., M.P.H.
Joseph J. Sudano, Ph.D.
MetroHealth Medical Center, Cleveland, OH 44109-1998
dwb@po.cwru. edu
