Graduate Assistant University of Florida College of Pharmacy Gainesville, United States
Background: We can leverage the rich information in automated healthcare data to find a valid negative control outcome (NCO), which can be used to assess and remove bias. Systemic quinolones are known to increase tendon rupture risk. However, recent studies did not find or found negligible magnitudes of this adverse effect, suggesting residual confounding due to channeling quinolones away from high-risk patients.
Objectives: To adjust risk estimates of oral quinolone-associated Achilles tendon rupture (ATR), using sports injuries as NCO to capture unmeasured physical activity.
Methods: We conducted a retrospective cohort study using 2005-Oct 2015 MarketScan® Commerical Claims Databases. We formed a cohort of adults aged 19-64 years with acute sinusitis or urinary tract infection and new initiation of quinolones or comparative antibiotics (e.g., amoxicillin, nitrofurantoin). We used a 1-year lookback period to measure baseline characteristics (e.g., diabetes, history of tendon rupture) and followed patients for 35 days or till they met censoring criteria (e.g., hospitalized not related to ATR). Inverse probability of treatment weighted (IPTW) Cox proportional hazard models were used to adjust for measured confounding. The corrected HR was the proportion of the adjusted HR of ATR over that of sports injuries and 95% confidence intervals were obtained by a joint model using stratified Cox models.
Results: We found an unadjusted HR of 1.08 (0.65-1.79) and IPTW adjusted HR of 0.91 (0.47-1.79) for ATR. The sports injury outcome had an unadjusted HR of 0.67 (0.48-0.93) and IPTW adjusted HR 0.60 (0.40-0.99). After adjusting for the departure from 1 of HR of NCO, the corrected HR for ATR was 1.51 (1.06-2.14).
Conclusions: NCO-corrected results moved the ATR HR closer to the pooled risk estimate of previous quinolone ATR studies [pooled odds ratio 2.52 (1.81-3.52)]. Although the NCO enhanced traditional adjustments for confounding, the adjusted risk estimate was still smaller than the pooled OR, suggesting some residual confounding was retained.
