Measurement properties of comorbidity indices in maternal health research: a systematic review.

BACKGROUND

Maternal critical illness occurs in 1.2 to 4.7 of every 1000 live births in the United States and approximately 1 in 100 women who become critically ill will die. Patient characteristics and comorbid conditions are commonly summarized as an index or score for the purpose of predicting the likelihood of dying; however, most such indices have arisen from non-pregnant patient populations. We sought to systematically review comorbidity indices used in health administrative datasets of pregnant women, in order to critically appraise their measurement properties and recommend optimal tools for clinicians and maternal health researchers.

METHODS

We conducted a systematic search of MEDLINE and EMBASE to identify studies published from 1946 and 1947, respectively, to May 2017 that describe predictive validity of comorbidity indices using health administrative datasets in the field of maternal health research. We applied a methodological PubMed search filter to identify all studies of measurement properties for each index.

RESULTS

Our initial search retrieved 8944 citations. The full text of 61 articles were identified and assessed for final eligibility. Finally, two eligible articles, describing three comorbidity indices appropriate for health administrative data remained: The Maternal comorbidity index, the Charlson comorbidity index and the Elixhauser Comorbidity Index. These studies of identified indices had a low risk of bias. The lack of an established consensus-building methodology in generating each index resulted in marginal sensibility for all indices. Only the Maternal Comorbidity Index was derived and validated specifically from a cohort of pregnant and postpartum women, using an administrative dataset, and had an associated c-statistic of 0.675 (95% Confidence Interval 0.647-0.666) in predicting mortality.

CONCLUSIONS

Only the Maternal Comorbidity Index directly evaluated measurement properties relevant to pregnant women in health administrative datasets; however, it has only modest predictive ability for mortality among development and validation studies. Further research to investigate the feasibility of applying this index in clinical research, and its reliability across a variety of health administrative datasets would be incrementally helpful. Evolution of this and other tools for risk prediction and risk adjustment in pregnant and post-partum patients is an important area for ongoing study.