I'm doing a case-control study, and I have made stratified analyses based on two age groups, as well as subtype of the type of cancer I'm doing my research on. My exposure of interest seems to be associated with an increased risk for the disease in only one of the groups. When examining both groups together (using multivariate conditional logistic regression for all analyses), the association is also statistically significant. My analyses both include binary exposure variables and continuous.
I stumbled upon this article (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4026206/), which discusses the error in concluding there must be a difference between the two groups, only because of the confidence intervals and P-values. As I understood, I need to do some sort of test of heterogeneity for my RRs, CIs and P-values, to fully see if the difference between the group is significant. The tests mentioned in the article is the Breslow-Day test, Wald test and regression-based test of interaction. As I understand, the Breslow-Day test is only applicable for contingency tables when I have binary variables of exposure.
So, what test should I use for my different type of exposure variables, when I have varying number of groups in my different stratifications, and how do I interpret the results? I use SPSS for statistical analysis.