In the July 2000 issue of Diabetes Care, Hummel et al.(1) reported a 2-year follow-up of the German BABYDIAB Study. Indeed, the BABYDIAB Study is an interesting and comparably large prospective study that follows infants of mothers or fathers with type 1 diabetes. So far, 10 children have developed diabetes and an additional 21 have developed islet cell antibodies. In the present article,Hummel et al. (2) attempted to study the possible effects of environmental risk factors, such as breast-feeding and vaccination, as well as measles, mumps, and rubella infections, to determine the likelihood of genetically susceptible children to develop diabetes or β-cell autoimmunity, as mirrored by islet autoantibodies. The authors found no significant effect of breast-feeding prevalence, duration of vaccinations, or reported childhood viral infections,and they concluded that these environmental factors are unlikely to have a major causal influence on initiating islet autoimmunity in genetically susceptible children. Nevertheless, it is important to identify what is meant by the expression “major.” Does it relate to relative or absolute risk, to the attributable proportion, or to the total number of type 1 diabetes cases (only a small fraction of which occur in first-degree family members)? The magnitude of risk considered to be important and the basis for this consideration should be defined when designing any study. Moreover, it should lead to a comprehensive power calculation that would establish the possibility of the study to detect such effects.
For example, the meta-analysis by Gerstein et al.(3), which is cited by Hummel et al. and was based on thousands of cases and controls, indeed showed a rather small overall effect in terms of relative risk (odds ratio [OR] 1.63,95% CI 1.22-2.17). In populations with a low breast-feeding frequency and duration, the attributable proportion of cases and the total number due to this specific factor might still be of significance. In populations with a high compliance to vaccination programs, the number of cases due to this factor could be considered to be of major importance if identified as having a risk exposure with an OR of 1.5-2.0.
When applying a power calculation test for cohort studies with internal comparisons (4) on breast-feeding data presented by Hummel et al., the number of expected events(i.e., clinical diabetes or autoantibody appearance) necessary to detect a relative risk of 1.5 would be estimated as 156 cases among the nonexposed children, given that the proportion of exposed (e.g., no breast-feeding) to nonexposed is 1:4 and the desired power is 80%, (P < 0.05). However, the expected number of events (under the null hypothesis) of independent component analysis positivity in the report by Hummel et al. was∼25. Eight cases of clinical diabetes were among the unexposed(breast-fed) children, which suggests a very low power to detect a risk increase of 1.5. With this sample size, only risk increases of ∼5-10 times would be detectable. With 16% of bacillus Calmette-Guerin-vaccinated individuals, approximately the same high probability of missing a risk increase would be expected. As for the diphtheria and tetanus toxoids and pertussis vaccine -vaccination, to which almost all the children in the cohort were exposed, even extremely large effects would remain undetected. In addition, for measles, mumps, and rubella vaccines and exposure to measles,mumps, or rubella infections the study gives no meaningful assessment of risk,because of the very low power.
For the study of a complex disease like type 1 diabetes, which requires the consideration of large numbers of risk genes and environmental factors, much larger studies are needed for estimating possible effects. In general,case-control studies are preferable in studying low-prevalence diseases because they allow enough power to effectively estimate low relative risks and to identify exposure interactions and confounding effects in multivariate analyses. The problems with potential disease-dependent biases in case-control studies certainly must be recognized, but not over-estimated, because they can be avoided by using a prerecorded hospital or register data. The lower sensitivity/specificity of exposure estimates often experienced in retrospective studies and unrelated to disease, will only lead to more conservative risk estimates, and can be readily compensated for by larger study populations (5). Indeed,prospective cohort studies now taking place in different parts of the world may be important for the assessment of the predictive value of immune markers at different ages before clinical onset of type 1 diabetes and for the presentation of viral antibodies. Even for these purposes, though, larger studies are preferable.
A meaningful assessment of the impact of potential environmental risks on study populations is of interest for understanding the mechanisms of type 1 diabetes and for developing preventive strategies. However, despite the prospective design, Hummel et al.'s study on the children of mothers or fathers with type 1 diabetes is insufficient. Conclusive negative results cannot be attained until studies with substantially larger populations are conducted.
