Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Patient-Level vs Group-Level Data to Adjust Meta-analysis on Transfusion and Mortality

Patient-Level vs Group-Level Data to Adjust Meta-analysis on Transfusion and Mortality When patient-level data are not available, confounding in meta-analysis can be minimized by using adjusted odds and/or hazard ratios from each source study. In their analysis, Chatterjee et al1(p133) instead chose to use unadjusted odds ratio (ORs) to “avoid bias from different types of adjustments in the various studies.” They adjusted their meta-regression with a limited number of group-level candidate predictors, which—except for baseline creatinine and hemoglobin levels—have not been validated as significant predictors of events in previous multivariate studies on the topic. This choice is not without serious limitations because it may lead to inaccurate inferences, especially when used with observational data.2 While the study by Chatterjee et al1 is methodologically beyond any reproach, one cannot help but wonder how much statistical heterogeneity can be sacrificed to avoid the “bias from different type of adjustments.” The observational studies selected in the meta-analysis were all done with a surprisingly homogenous methodology. All included a multivariate modeling using mortality as the dependent predictors and baseline demographics plus comorbid conditions as the independent predictors. Most accounted for transfusion propensity. By choosing to favor unadjusted ORs, the authors sacrificed patient-level data adjustment performed with a comprehensive set of predictors, which may have led to an overestimation in the strength of association between transfusion and mortality. In most instances, the risk ratios obtained by Chatterjee et al1 were significantly closer to the crude ORs reported by the investigators than the actual adjusted ORs obtained with patient-level data. For instance, in the study by Jani et al,3 the reported crude and adjusted transfusion ORs (95% CIs) were 5.48 (4.23-7.09) and 2.02 (1.47-2.79), respectively, whereas the risk ratio obtained by Chatterjee et al1 was 4.83 (3.81-6.12). Similarly, in a study by Jolicœur et al,4 the reported crude and adjusted transfusion ORs (95% CIs) were 6.30 (4.14-9.59) and 2.16 (1.20-3.88), respectively, whereas the risk ratio used by Chatterjee et al1 was 6.38 (4.88-8.34). Such an overestimation was seen in 6 of the 8 studies observational studies in which such information was available (of note, the transfusion data could not be extracted from the study by Shishehbor et al, cited by the authors). While I enjoyed reading the article by Chatterjee et al,1 and agree with the overall message in favor for a reasonably restrictive blood transfusion policy in critically ill patients, I wonder if the effect size linking transfusion to mortality may not have been overestimated. Back to top Article Information Correspondence: Dr Jolicœur, Montreal Heart Institute, 5000 rue Bélanger E, Montréal, QC H1T 1C8, Canada (marc.jolicoeur@icm-mhi.org). Conflict of Interest Disclosures: None reported. References 1. Chatterjee S, Wetterslev J, Sharma A, Lichstein E, Mukherjee D. Association of blood transfusion with increased mortality in myocardial infarction: a meta-analysis and diversity-adjusted study sequential analysis. JAMA Intern Med. 2013;173(2):132-13923266500PubMedGoogle ScholarCrossref 2. Biondi-Zoccai G, Agostoni P, Abbate A, D’Ascenzo F, Modena MG. Potential pitfalls of meta-analyses of observational studies in cardiovascular research. J Am Coll Cardiol. 2012;59(3):292-29322240139PubMedGoogle ScholarCrossref 3. Jani SM, Smith DE, Share D, et al. Blood transfusion and in-hospital outcomes in anemic patients with myocardial infarction undergoing percutaneous coronary intervention. Clin Cardiol. 2007;30(10):(suppl 2) II49-II5618228652PubMedGoogle ScholarCrossref 4. Jolicœur EM, O’Neill WW, Hellkamp A, et al; APEX-AMI Investigators. Transfusion and mortality in patients with ST-segment elevation myocardial infarction treated with primary percutaneous coronary intervention. Eur Heart J. 2009;30(21):2575-258319596659PubMedGoogle ScholarCrossref http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png JAMA Internal Medicine American Medical Association

Patient-Level vs Group-Level Data to Adjust Meta-analysis on Transfusion and Mortality

JAMA Internal Medicine , Volume 173 (12) – Jun 24, 2013

Loading next page...
 
/lp/american-medical-association/patient-level-vs-group-level-data-to-adjust-meta-analysis-on-2aK73gG7xv
Publisher
American Medical Association
Copyright
Copyright © 2013 American Medical Association. All Rights Reserved.
ISSN
2168-6106
eISSN
2168-6114
DOI
10.1001/jamainternmed.2013.6518
Publisher site
See Article on Publisher Site

Abstract

When patient-level data are not available, confounding in meta-analysis can be minimized by using adjusted odds and/or hazard ratios from each source study. In their analysis, Chatterjee et al1(p133) instead chose to use unadjusted odds ratio (ORs) to “avoid bias from different types of adjustments in the various studies.” They adjusted their meta-regression with a limited number of group-level candidate predictors, which—except for baseline creatinine and hemoglobin levels—have not been validated as significant predictors of events in previous multivariate studies on the topic. This choice is not without serious limitations because it may lead to inaccurate inferences, especially when used with observational data.2 While the study by Chatterjee et al1 is methodologically beyond any reproach, one cannot help but wonder how much statistical heterogeneity can be sacrificed to avoid the “bias from different type of adjustments.” The observational studies selected in the meta-analysis were all done with a surprisingly homogenous methodology. All included a multivariate modeling using mortality as the dependent predictors and baseline demographics plus comorbid conditions as the independent predictors. Most accounted for transfusion propensity. By choosing to favor unadjusted ORs, the authors sacrificed patient-level data adjustment performed with a comprehensive set of predictors, which may have led to an overestimation in the strength of association between transfusion and mortality. In most instances, the risk ratios obtained by Chatterjee et al1 were significantly closer to the crude ORs reported by the investigators than the actual adjusted ORs obtained with patient-level data. For instance, in the study by Jani et al,3 the reported crude and adjusted transfusion ORs (95% CIs) were 5.48 (4.23-7.09) and 2.02 (1.47-2.79), respectively, whereas the risk ratio obtained by Chatterjee et al1 was 4.83 (3.81-6.12). Similarly, in a study by Jolicœur et al,4 the reported crude and adjusted transfusion ORs (95% CIs) were 6.30 (4.14-9.59) and 2.16 (1.20-3.88), respectively, whereas the risk ratio used by Chatterjee et al1 was 6.38 (4.88-8.34). Such an overestimation was seen in 6 of the 8 studies observational studies in which such information was available (of note, the transfusion data could not be extracted from the study by Shishehbor et al, cited by the authors). While I enjoyed reading the article by Chatterjee et al,1 and agree with the overall message in favor for a reasonably restrictive blood transfusion policy in critically ill patients, I wonder if the effect size linking transfusion to mortality may not have been overestimated. Back to top Article Information Correspondence: Dr Jolicœur, Montreal Heart Institute, 5000 rue Bélanger E, Montréal, QC H1T 1C8, Canada (marc.jolicoeur@icm-mhi.org). Conflict of Interest Disclosures: None reported. References 1. Chatterjee S, Wetterslev J, Sharma A, Lichstein E, Mukherjee D. Association of blood transfusion with increased mortality in myocardial infarction: a meta-analysis and diversity-adjusted study sequential analysis. JAMA Intern Med. 2013;173(2):132-13923266500PubMedGoogle ScholarCrossref 2. Biondi-Zoccai G, Agostoni P, Abbate A, D’Ascenzo F, Modena MG. Potential pitfalls of meta-analyses of observational studies in cardiovascular research. J Am Coll Cardiol. 2012;59(3):292-29322240139PubMedGoogle ScholarCrossref 3. Jani SM, Smith DE, Share D, et al. Blood transfusion and in-hospital outcomes in anemic patients with myocardial infarction undergoing percutaneous coronary intervention. Clin Cardiol. 2007;30(10):(suppl 2) II49-II5618228652PubMedGoogle ScholarCrossref 4. Jolicœur EM, O’Neill WW, Hellkamp A, et al; APEX-AMI Investigators. Transfusion and mortality in patients with ST-segment elevation myocardial infarction treated with primary percutaneous coronary intervention. Eur Heart J. 2009;30(21):2575-258319596659PubMedGoogle ScholarCrossref

Journal

JAMA Internal MedicineAmerican Medical Association

Published: Jun 24, 2013

Keywords: transfusion

References