ON COMBINING DOSE-RESPONSE DATA FROM EPIDEMIOLOGIC STUDIES BY METAANALYSIS

Using data from a meta-analysis of the effects of oestrogen replacemen t therapy on the development of breast cancer, we compared alternative methods for combining dose-response slopes from epidemiological studi es. We evaluated issues related both to summarizing data from single s tudies and to combining results from multiple studies. Findings relate d to the analysis of individual dose-response studies include: (I) a m ethod of weighing studies that gives greater influence to dose-respons e slopes that conform to the linear relation of relative risk to durat ion can lead to large differences in calculated weights as a function of non-linearity; (2) a regression model using a variable-intercept re sulted in a mean dose-response slope that increased as much as threefo ld when compared with the values obtained with a zero-intercept model. When combining results from multiple studies, we found: (1) calculati ng standard errors of mean dose-response slopes by methods that allow for both among-study and within-study variability (a random-effects ty pe model) gave values different from a method that assumes homogeneity and equal within-study precision (a fixed-effects model); (2) the ran dom-effects model gives mean and standard error results most similar t o a bootstrap resampling method as increasing heterogeneity is observe d (however, this model could give biased mean estimates compared with the bootstrap method); (3) a components-of-variance model compares fav ourably with the bootstrap and is easier to apply than the random-effe cts model. Based on these findings, we recommend the use of methods wh ich incorporate heterogeneity to guard against underestimating the sta ndard error. However, caution is urged because bias in point estimates can occur if extreme heterogeneity is present. Two other observations affect the interpretation of data combined from multiple studies. Fir st, inclusion into a model of quality scores assigned by blinded revie wers had little effect on the mean dose-response slope and its standar d error. Second, the number of studies required to achieve desired sta tistical power, varies with effect size.