Author(s): Mona Izadnegahdar, May K Lee, Min Gao, Melissa Pak and Karin H Humphries*
Background: Acute myocardial infarction (AMI) patients are frequently transferred to tertiary care facilities for advanced procedures. Exclusion of transferred patients can result in biased estimates, especially if the focus is on a population-based cohort. The potential magnitude of this bias remains unknown. We assessed the impact of excluding transferred patients on 30-day AMI mortality estimates, over a 10-year period. Methods: All AMI hospitalizations to acute care hospitals in British Columbia, from 2000 to 2009, were captured. Transfers were defined as a discharge from the index AMI hospitalization to another hospital. We compared transfer rates by sex, age and year using logistic regression models. Age- and sex-specific 30-day AMI mortality rates and the sex odds ratios (OR) were estimated, regardless of transfer status, and for the sub-cohort excluding transferred patients. Results: Of 63,310 AMI patients, 40.6% had at least one transfer out of the index AMI hospital. Men and younger patients were more likely to be transferred. Transfer rates increased over time in all age groups regardless of sex. Overall, when transfers were excluded, 30-day AMI mortality rates were overestimated (absolute difference of 5.5% in women and 6.7% in men). Furthermore, the 30-day mortality OR (women vs. men) was underestimated: transfer outs excluded (OR=1.25, 95% CI: 1.19, 1.31) vs. all patients included (OR=1.49, 95% CI: 1.42, 1.56). Conclusions: 30-day mortality may be overestimated in population-based analyses when transfers are excluded, while the sex difference is underestimated. The observed bias is strongly affected by the magnitude and time trends of transfer rates.
