Frequency of adjustment with comorbidity and illness severity scores and indices in cardiac arrest research

Abstract Background Previous research demonstrates that results from observational research correlate well with results from clinical trials, and if the former are well designed these can guide clinical practice. Observational designs are commonly used in cardiac arrest research, but it is known that this research area is beset by confounding, especially due to illness severity and comorbidity. The aim of this study was to count the number of studies that account for comorbidity and illness severity with scores and indices, and to measure the change in results across analyses that adjust for these scores and indices. Methods A systematic search of Medline, Embase and Cochrane Central for cardiac arrest studies that report survival outcomes for the year 2015 and that utilize illness severity and comorbidity scores and indices was conducted. Eligibility assessment and data extraction were completed independently by two reviewers. The proportion of studies that employ these scores and indices and also those studies that adjusts for such scores/indices were noted. The change in magnitude of estimates in analysis that incorporate illness severity and comorbidity measures were compared to analysis that do not. Results Sixty (28%) of 213 cardiac arrest studies that report survival outcomes utilize illness severity or comorbidity indices and scores, of which 39 studies (65%) used risk scores and indices to account for the confounding effect of comorbidity or illness severity. A −14% change (minus indicating change towards the null) in the magnitude of effect sizes was apparent when models included illness severity or comorbidity adjustment (interquartile range −37.7 to 4.4%), compared to when they did not. Conclusions A relatively small proportion of cardiac arrest studies account for illness severity and comorbidity with scores and indices, and such adjustment tend to drive estimates towards the null (no difference in groups being compared). Confounding by illness severity and comorbidity is a significant source of bias in non-randomized cardiac arrest studies and should be a key part of the design of such studies.