In a recent article and related press release, Yang and Schuler1 claim that marital distress among breast cancer patients is associated with poorer health and a steeper decline in physical activity. Should patients with breast cancer be eager to do everything they can to promote their health by working on their marriages? Maybe, but not on the basis of Yang and Schuler's claims. The authors acknowledge that their article was based on “previous research,” but they should have disclosed that it was based on extensive analyses of the same dataset from which their sample was drawn. Those previous analyses failed to demonstrate any effect whatsoever of marital distress on disease progression.2 Apparently, there was no press release for these reassuring findings. Other reports on the same data claimed robust effects of a relaxation and health education intervention when at least 7 measures of mood and well being, numerous measures of health behaviors, and 15 measures of immune function were examined.3
Putting aside the issue of whether effects persist when appropriate controls for Type I error are introduced, it is left unclear how these analyses informed Yang and Schuler's post hoc selection of a convenience sample of women for their study or choice of independent and control variables. The measurement of physical health in their study is inadequate. The Karnofsky performance status is used, which confounds health and role functioning and is not validated as an outcome measure in such a relatively healthy population. Similarly, the second measure of health is not validated, seems most sensitive to side effects of treatment, and, among the women that they studied, yielded modest associations with marital distress unlikely to have any clinical significance.
Problems in a post hoc selection of patients and measures are compounded by multivariate analyses inappropriate to the sample size, over fitted regression equations, and graphic displays of results with truncated axes that easily are misinterpreted as depicting dramatic differences where there are only marginally different and clinically insignificant effects. Given previous claims of a robust treatment effect, the question is begged whether marital distress moderates any effects of intervention. Unfortunately, examination of such an interaction is precluded by the small size of the convenience sample. Authors should exercise restraint in interpreting post hoc analyses of clinical trials, particularly when vulnerable patients may be listening.