Could health system factors contribute to residual confounding?
We welcome with interest the study by Vinogradova et al (1). The authors demonstrate increased all-cause mortality associated with the prescription of a direct oral anti-coagulant (DOAC) compared to Warfarin, a vitamin K antagonist (VKA) for anticoagulation in primary care. We share the authors’ concern that the effect size may represent residual confounding.
Health system factors may be contributing unmeasured confounding in this association. This may include for example: local resource implications – e.g. one practice with a better set-up for INR monitoring; or certain regions having local prescribing permissions. These factors would direct prescribing choices and may also influence local population mortality.
A sensitivity analysis or E-value estimate of the impact of potential unmeasured confounding would be a recommended means of estimating robustness of the effect size determined without information on these factors (2). The hazard ratio (HR) of all-cause mortality associated with rivaroxaban in patients without atrial fibrillation was reported as HR 1.51 (95% CI 1.36 to 1.66). Using an E-value calculator (3), an adjusted HR of 1.51 could be explained by an unmeasured confounder that was associated with both the choice of treatment and the outcome by a risk ratio of 2.39-fold each, above and beyond the measured confounders. Uptake of new anticoagulants varied substantially across Clinical Commissioning Groups (CCG) in England from >8% to 60% during the period of study (4). An unmeasured confounding effect of this magnitude may therefore be very reasonable, and the estimated effect size reported may purely represent residual confounding.
We face similar challenges with our own attempts at causal inference using routine data from primary care. We suggest, where available, Lower Layer Super Output Areas (LSOA) mapped onto CCGs and individual prescriber codes as two potential proxy-markers to represent local health system factors as confounders.
1. Vinogradova Y, Coupland C, Hill T, et al. Risks and benefits of direct oral anticoagulants versus warfarin in a real world setting: cohort study in primary care. BMJ 2018;362:k2505
2. VanderWeele TJ,.Ding P. Sensitivity Analysis in Observational Research: Introducing the E-Value. Ann.Intern.Med. 2017;167:268-74.
4. Burn J Pirmohamed M. Direct oral anticoagulants versus warfarin: is new always better than the old? BMJ Open Heart Journal http://dx.doi.org/10.1136/openhrt-2017-000712
Competing interests: No competing interests