Longitudinal Risk Adjustment for Maternal End-Organ Injury and Death

To determine whether adjusting for healthcare utilization and comorbidity diagnosed in the year before delivery improves the prediction of adverse maternal outcomes.

The Truven Health MarketScan database was used to determine whether healthcare utilization and comorbidity diagnosed in the year before pregnancy improved prediction of acute organ injury or death during the delivery hospitalization through 30 days postpartum in this retrospective cohort study. In an initial model, we analyzed the risk for adverse outcomes controlling for underlying comorbidity, obesity, and demographic risk factors present during pregnancy. Subsequent models included diagnoses from the year before pregnancy as well as whether patients had emergency department encounters, inpatient hospitalizations, or received medications from a pharmacy. We compared risk estimates and whether prediction of acute organ injury or death improved with data from the year before pregnancy. Unadjusted and adjusted log-linear regression models were performed to demonstrate the association between exposures and outcomes with unadjusted (RR) and adjusted risk ratios (aRR) with 95% CIs as measures of effects. Logistic regression was performed to calculate the c-statistic of the adjusted models. Separate analyses were performed for patients with Medicaid and commercial insurance. An analysis of Medicaid patients by maternal race and ethnicity was performed to determine if diagnoses and utilization before pregnancy accounted for maternal disparities.

A total of 740,002 patients were analyzed in this study. In unadjusted analyses of patients with commercial insurance, ≥2 compared to 0 emergency department encounters (RR = 1.82, 95% CI = 1.61, 2.07), ≥2 compared to 0 inpatient hospitalizations (RR = 4.43, 95% CI = 3.20, 6.13), and receipt of medications from ≥5 prescription groups compared to no prescriptions (RR = 1.97, 95% CI = 1.74, 2.24) were all associated with increased risk for acute organ injury or death. Higher underlying comorbidity and obesity were also associated with increased risk. These risks were attenuated in adjusted analyses but retained significance. Risk estimates were similar for patients with Medicaid insurance with the exception of receipt of medications from ≥5 prescription groups which was non-significant in adjusted analyses (aRR = 1.12, 95% CI = 0.90, 1.40). C-statistics from logistic regression models were similar for models with and without pre-pregnancy data. When race was added to the adjusted models, risk among black women in the adjusted models did not differ significantly from the unadjusted estimate.

ED encounters and inpatient admissions the year before pregnancy were associated with increased risk of adverse maternal outcomes. However, adding these risk factors to adjusted models did not meaningfully improve the amount of variance accounted for. Further research is indicated to determine to what degree longitudinal care quality is associated with maternal risk.