gms | German Medical Science

Introduction: In a previous individual data based metaanalysis of 49 gene association studies analysis we checked whether the TP53 polymorphism PRO72ARG is associated with the risk of cervical neoplasms [1]. Various case definitions (invasive, high and low grad lesion) were considered by fitting separate logistic models to the respective case control status outcome while accounting for study, ethnic group and material used for genotyping (separate modelling approach).

In the present analysis, we investigated the benefits and drawbacks of a reverse modelling approach: The genotype distribution was modelled in dependence on the case control status and the covariates thus allowing for the inclusion of several case entities in one model.

Patients and methods: For the reverse approach all 49 studies could be analysed, but only 33 studies and 13419 individuals could be used in separate modelling. Separate modelling was based on binary, reverse modelling on generalized logistic regression. Solely controls confirmed by negative cytology were used in analyses.

Because less than half of the studies comprised all case entities, the reverse approach implicitly mixes direct and indirect comparisons. The role of indirect comparisons was investigated by definition of study level variables that reflect the types of cases contained in a study.

Results: In both approaches no significant association was found in studies using white blood cells for all entities, but a significant association between invasive cancer and genotype for studies using tumor specimen, reflecting loss of heterozygosity. Effect estimates were very similar in both approaches.

Standard errors for log odds ratios for all comparisons between case and control groups and for all genotype contrasts were consistently reduced by reverse modelling. But the gain in precision was small.

Conclusion: Reverse modelling allows for a comprehensive analysis of different case entities within a single model and yields more precise effect estimates.