Outcomes based on registry data can be skewed if cardiologists can easily identify a high-risk subgroup of patients and then preferentially allocate them to one treatment over another, according to a study designed to explain conflicting results between randomized clinical trials (RCTs) and observational studies on STEMI patients.
RCTs in the treatment of STEMI patients repeatedly have shown that patients treated with primary PCI have better outcomes than those treated with fibrinolysis. But the designs of clinical trials typically exclude higher-risk patients and therefore, may not accurately reflect patient populations seen in practice.
Some observational studies, on the other hand, have shown no comparative benefit for primary PCI. Observational studies such as those based on registry data incorporate a real-world setting but cannot capture confounders, factors that can’t be measured but still may influence results. Among those confounders may be physicians’ preference to treat higher risk patients with one therapy and not the other,
Sayan Sen, MRCP, of St. Mary’s Hospital in London, and colleagues sought to put the debate to rest by attempting to quantify the impact of allocation bias in STEMI registry data. They first designed a model to gauge the potential effect of allocation bias. They then analyzed registry data for 55,022 STEMI patients to determine if high-risk patients were evenly distributed between treatment groups and assessed the effect of imbalances on mortality results.
Sen and colleagues found that STEMI registries were very sensitive to physician preference. Based on their analysis, it took only a small number of high-risk patients allocated to one group or the other to distort findings. For primary PCI, a mere 5.4 percent more prevalence of high-risk patients in primary PCI groups disguised the mortality benefit of primary PCI.
“We have shown that the effect of allocation bias on mortality within registries is far larger than might be suspected,” Sen and colleagues wrote. “Once this allocation bias is accounted for, registries predict that primary PCI has 22 percent lower long-term mortality than fibrinolysis.”
Their results are consistent with meta-analyses of RCTs, they wrote.
“The principal challenge to comparative effectiveness research from an observational data set may be insuperable: the investigators can adjust only for the information that is documented systematically, which is only a subset of the information used by the clinician on the spot,” they concluded. “Observational comparative effectiveness research is therefore led astray, not by poor research but by good medical practice.”
The study was published online Nov. 13 in Circulation: Cardiovascular Quality and Outcomes.