Data source

We used Vital Statistics birth records linked with New York state discharge abstract data, the Statewide Planning and Research Cooperative System, for all delivery hospitalizations in New York City from 2011 through 2013. Data linkage was conducted by the New York State Department of Health, and 98.8% of maternal discharge abstracts were linked with infant live birth certificates.

Institutional review board approvals were obtained from the New York City Department of Health and Mental Hygiene, the New York State Department of Health, and the Icahn School of Medicine at Mount Sinai. Delivery hospitalizations were identified based on International Classification of Diseases , ninth revision, Clinical Modification diagnosis and procedure codes and Diagnosis-Related Group delivery codes. From linked records, 4 hospitals with annual delivery volumes less than 5 births and 1360 deliveries with missing hospital identifiers were excluded. The final sample included 353,773 deliveries at 40 hospitals.

Severe maternal morbidity

We used a published algorithm to identify severe maternal morbidity, using diagnoses for life-threatening conditions (eg, renal failure, eclampsia) and procedure codes for life-saving procedures (eg, hysterectomy, ventilation, blood transfusion) defined by investigators from the Centers for Disease Control and Prevention.

As specified by the algorithm, we excluded hospitalizations with a length of stay less than the 90th percentile as calculated separately for vaginal, primary, and repeat cesarean deliveries. All severe maternal morbidity hospitalizations associated with in-hospital mortality and transfer as well as severe complications identified by procedure codes were included, regardless of the length of stay, as recommended. Transfers were defined as discharge disposition after delivery or source of admission for delivery as specified.

Covariates

To risk-adjust hospital-level rates of maternal morbidity, we used variables from the vital statistics records, including mothers’ sociodemographic characteristics (maternal age, self-identified race and ethnicity, parity, education), prenatal care visits, and clinical and obstetric factors (multiple pregnancy, history of previous cesarean delivery, body mass index).

New York City Vital Statistics collect self-identified race and ethnicity data. We ascertained patient insurance status from the Statewide Planning and Research Cooperative System. We also included diagnoses for patient risk factors that could lead to maternal morbidity but were likely present on admission to the hospital (eg, diabetes, hypertension, obesity, premature rupture of membranes, disorders of placentation). These conditions have been used to risk-adjust for severe maternal morbidity, cesarean deliveries, and other maternal outcomes.

We obtained teaching status from the American Hospital Association, ownership and nursery level from the New York State Department of Health, and volume of deliveries in each hospital from the Statewide Planning and Research Cooperative System to assess how other hospital characteristics are correlated with severe maternal morbidity.

Analysis

We compared the sociodemographic characteristics and clinical conditions of black and white women using χ ² tests. We used a mixed-effects logistic regression with a random hospital-specific intercept to generate risk-standardized severe maternal morbidity rates for each hospital. The models included the covariates described in previous text. Hospital risk-standardized rates were computed from these models using methods recommended by the Centers for Medicare and Medicaid Services Hospital Compare. These rates were the ratio of predicted to expected severe maternal morbidity rates, multiplied by the New York City average severe maternal morbidity rate. For each hospital, the numerator of the ratio is the number of severe maternal morbidity cases predicted on the bases of the hospital’s performance with its case-mix, and the denominator is the number of severe maternal morbidity cases expected on the bases of the New York City performance with that hospital’s case mix. We ranked hospitals from lowest to highest risk-standardized severe maternal morbidity rates. These analyses did not include hospital-level variables.

We conducted a sensitivity analysis using observed to expected rates for hospital ranking and found that rankings differed very little between the Centers for Medicare and Medicaid Services model and the standard observed to expected ratio. In addition, because blood transfusions are an important component of severe maternal morbidity, we examined the correlation between hospital rankings based on severe maternal morbidity with and without blood transfusion.

To assess racial disparities in the use of hospitals with the lowest morbidity rates, we calculated the cumulative distributions of births among hospitals ranked from the lowest to the highest standardized morbidity rate for black and white mothers. We used the Kolmogorov-Smirnov test to assess whether the distributions of deliveries among hospitals differed for white and black women.

To address the effects on black severe maternal morbidity rates of these differences in delivery location, we conducted a thought experiment and asked what would happen if black mothers went to the same hospitals as white mothers? We used the same risk-standardized morbidity model and kept all individual patient characteristics the same. We calculated the predicted probability of morbidity for each black mother at each hospital.

For each black mother, we took the weighted average of these probabilities, in which weights were the percentage of white mothers who went to each hospital. The difference between the predicted probability at the hospital a black mother went to and the weighted average probability if the black mother delivered at the white mother’s hospital is the decrease or increase in the probability of a morbid event. The sum of the difference in probabilities across all black women is the morbid events avoided if black mothers went to the same hospitals as white mothers or the morbid events because of between-hospital disparities.

A recent simulation study tested this approach against the more common approach of identifying minority-serving facilities based on the percentage of black patients at a hospital and found that it more accurately measured the magnitude of between-hospital disparities, although both were successful at identifying the existence of disparities.

To investigate the association between hospital characteristics and severe maternal morbidity rates, we estimated the mixed-effects logistic regression that included maternal sociodemographic and clinical factors as well as the hospital characteristics described in the previous text.

All statistical analysis was performed using the SAS system software version 9.3 (SAS Institute Inc, Cary, NC).