Background: Despite extensive studies on exposure and disease misclassification, few studies have investigated misclassification of confounders. This study aimed to identify differentially misclassified confounders in a comparative evaluation of hospital care quality and to quantify their impact on hospital-specific risk-adjusted estimates, focusing on the appropriateness of caesarean sections (CS).
Methods: We gathered data from the Hospital Information System in Italy for women admitted in 2005-2010. We estimated adjusted proportions of CS with logistic regression models. Among several confounders, we focused on high fetal head at term (HFH), which is seldom objectively documentable in medical records.
Results: A total of 540 maternity units were compared. The median HFH prevalence was 0.9%, ranging from 0 to 70%. In some units, HFH was coded so frequently that it was unlikely to reflect a natural heterogeneity. This "over-coding" was conditional on the outcome because it occurred more frequently for women that underwent CS. This suggested an opportunistic coding to justify the choice of a CS. HFH misclassification was not randomly distributed over Italy; it had an excess in the Campania region where, in some units, the proportion of HFHs gradually increased from 2005 to 2010 (e.g., from 0 to 26%), but the national average remained constant (2.5%). The inclusion of the misclassified diagnosis in the models favored those hospitals that codified in a less-than-fair manner.
Conclusions: Our findings emphasized the importance of rigorously inspecting for differential misclassification of confounders. Their validity may be subject to substantial heterogeneity over hospitals, over time and geographical areas.