Key result
The individual anesthesiologist had a negligible impact on risk-adjusted mortality among 18,662 patients undergoing cardiac surgery.
Why the study?
Does the individual anesthesiologist impact risk-adjusted mortality in patients undergoing cardiac surgery?
Does the individual anesthesiologist impact risk-adjusted mortality in patients undergoing cardiac surgery?
Answers are grounded in this paper and cite their sources.
Methodological choices, such as using random-effects versus fixed-effects models, can significantly alter conclusions regarding the impact of individual anesthesiologists on cardiac surgery outcomes.
To the Editor In the article by Glance et al.1 on anesthesiologists’ impact on coronary surgery outcomes, the authors state: “To date, no large-scale study has analyzed the effects of different anesthesiologists on patient outcomes.” This is factually incorrect. Our study on the anesthesiologists’ impact on cardiac surgical outcomes in 18,662 patients was published electronically in October 2013 and in print in February 2014.2 In contrast to Glance et al.,1 we found the individual anesthesiologist had a negligible impact on risk-adjusted mortality. These conflicting findings may arise both from different populations and different statistical methodology in the 2 studies. Unlike many practices in New York State, Papworth anesthesiologists’ practices are almost entirely cardiothoracic. Average total cardiac surgical caseload was consequently much greater in our study. Although we were unable to demonstrate a relationship between annual caseload and outcomes at Papworth, differences in annual anesthesiologist cardiac caseload may have been a factor. Higher annual case volumes may help ensure competence. Glance et al.1 analyzed 55% of identified patients and <30% of anesthesiologists performing at least 1 cardiac surgical case, whereas we included 99% of patients and 83% of anesthesiologists. Although not reported in the article, we found wide year-to-year variation and a lack of consistency in anesthesiologist annual rankings after correcting for patient risk. The fixed-effects model of Glance et al.1 did not simultaneously control for surgeon and hospital (which includes the anesthesiologists as a group) effects. They assumed that anesthesiologists and surgeons were randomly paired and that the correlation between surgeon and anesthesiologist performance was small. It is unclear how the reported correlation coefficient of 0.14 was obtained. In contrast, we elected to use random-effects regression analysis, explicitly accounting for grouped data structure (patients grouped within anesthesiologists and surgeons). Random-effects models treat the anesthesiologists as a random sample from a population, allowing the generalization of results to the whole population of anesthesiologists, in contrast to fixed-effects models, which restrict results to the sample of anesthesiologists available. By using random-effects models, one could generalize the estimated results to other anesthesiologists of the same type as those in Papworth. Furthermore, the fixed-effects models fitted did not account for surgeon effect and anesthesiologist effect simultaneously. Our data suggest that the surgeon had a far greater impact on perceived anesthesiologist performance than Glance et al.1 assume. Jonathan H. Mackay, FRCAAnaesthetic DepartmentPapworth HospitalCambridge, United Kingdom[email protected] Samer A. M. Nashef, FRCSSurgical DepartmentPapworth HospitalCambridge, United Kingdom Olympia Paprachristofi, MScMRC Biostatistics UnitCambridge, United Kingdom Linda Sharples, PhDInstitute of Clinical Trials ResearchLeeds, United Kingdom
No takes yet. Share an insight, caveat, or question.
Mackay et al. (2016) conducted a letter in Cardiac surgery (n=18,662). Individual anesthesiologist was evaluated on Risk-adjusted mortality. The individual anesthesiologist had a negligible impact on risk-adjusted mortality among 18,662 patients undergoing cardiac surgery.
Synapse has enriched 10 related papers. Consider them for comparative context: