Modeling heteroscedastic age‐period‐cohort cancer data