This cohort study used data from 4,063,106 recorded live births that occurred in California from January 1, 2007 through December 31, 2014. California adopted the revised U.S. birth certificate in 2007, which added the collection of maternal weight and height information. The International Classification of Disease, 10th Edition, Clinical Modification was implemented in 2015. We restricted to delivery years in which the International Classification of Disease, 9th Edition, Clinical Modification (ICD-9-CM) was used for consistency in hospital coding practices. Birth certificate data were obtained from the California Department of Public Health (2007–2011) and the California Maternal Quality Care Collaborative (2012–2014). For 99% of deliveries, birth certificate data were previously linked to maternal delivery hospitalization discharge data collected by the Office of Statewide Health Planning and Development. If twins or other multiples were delivered, the maternal discharge data were duplicated in separate delivery records for each infant. We therefore selected the first delivery record if multiples were delivered. The final sample of complete cases included 3,556,206 deliveries with linked records and plausible gestational duration (> 20 weeks). We assessed differences between included and excluded subjects.
The risk factors of interest were advanced maternal age, pre-pregnancy obesity, pre-pregnancy comorbidity, and cesarean delivery. Maternal age at delivery was collected on the birth certificate, and advanced maternal age was defined as ≥35 years. Pre-pregnancy body mass index (BMI) was calculated from self-reported pre-pregnancy weight and height on the birth certificate and maternal obesity was defined as BMI ≥30 kg/m
2. Pre-pregnancy comorbidity was a binary variable defined as indication of any of the 12 pre-pregnancy medical conditions included in the obstetric comorbidity index created by Bateman et al. [
26], reported either in the hospitalization record or on the birth certificate. These conditions included pulmonary hypertension, sickle cell disease, chronic renal disease, preexisting hypertension, chronic ischemic heart disease, congenital heart disease, systemic lupus erythematosus, human immunodeficiency virus, cardiac valvular disease, chronic congestive heart failure, asthma, and preexisting diabetes mellitus. Cesarean delivery was reported either in the hospitalization record (ICD-9-CM procedure code 74) or on the birth certificate. In descriptive analyses, we additionally categorized maternal age as < 20, 20–24, 25–29, 30–34, 35–39, or ≥ 40 years; pre-pregnancy BMI (kg/m
2) as < 18.5 (underweight), 18.5–24.9 (normal weight), 25–29.9 (overweight), 30–34.5 (obesity class 1), 35–39.9 (obesity class 2), and ≥ 40 (obesity class 3); pre-pregnancy comorbidity as hypertension, diabetes, asthma, or other; and delivery method as vaginal without induction, vaginal with induction, primary cesarean without induction, primary cesarean with induction, and repeat cesarean. We used ICD-9-CM procedure codes 73.1, 73.4, 73.01 in delivery hospitalization records to identify labor induction, as previously validated [
27]. Prior cesarean delivery was identified using birth certificate data and ICD-9-CM diagnosis code 654.2x in delivery hospitalization records. Confounding factors included in all multivariable regression models were selected a priori based on prior knowledge, directed acyclic graphs, and variables available in the dataset [
2,
5,
6,
14,
17,
22]. These factors were reported on the birth certificate and included maternal race/ethnicity (U.S.-born Hispanic/Latina, foreign-born Hispanic/Latina, non-Hispanic White, Asian/Pacific Islander, non-Hispanic Black, Other), educational attainment (less than high school degree, high school degree or equivalent, some college, college degree), expected method of payment for delivery (private insurance or other), obstetric history (nulliparous, multiparous without previous cesarean delivery, multiparous with previous cesarean delivery), and twin/multiple birth. We additionally adjusted cesarean delivery models for placental conditions (abruption or previa; ICD-9-CM diagnosis code 641.x), preeclampsia (ICD-9-CM diagnosis codes 642.4, 642.5, 642.7), and gestational age because of their associations with cesarean delivery and severe maternal morbidity.
Statistical analysis
We first assessed the distribution of maternal and delivery characteristics in women with and without severe maternal morbidity. We then tested the association of each risk factor of interest (advanced maternal age, pre-pregnancy obesity, pre-pregnancy comorbidity, and cesarean delivery) with severe maternal morbidity using multivariable logistic regression models, adjusted for race/ethnicity, education, payment method, obstetric history, and twin/multiple birth. Because of differences in temporality, which distinguish confounders and mediators, we additionally adjusted for advanced maternal age with obesity as the predictor; advanced maternal age and obesity with comorbidity as the predictor; and advanced maternal age, obesity, comorbidity, placental condition, preeclampsia, and gestational age with cesarean delivery as the predictor.
We then estimated population attributable risk percentages to understand the population-level implications of the associations between the risk factors of interest and severe maternal morbidity. Population attributable risk percentages account for both the strength of an association and the prevalence of the risk factor in the population. To calculate this measure, we used the multivariable logistic regression models to predict the prevalence of severe maternal morbidity if the risk factor of interest were eliminated. For example, severe maternal morbidity prevalence was predicted from the models after re-coding all cesarean deliveries to vaginal deliveries in the dataset (for heuristic purposes). We then calculated population attributable risk percentages as:
$$ 100\times \left(\frac{Observed\ prevalence- Predicted\ prevalence\ if\ risk\ factor\ eliminated}{Observed\ prevalence}\right) $$
We bootstrapped these simple substitution models 1000 times to obtain 95% confidence intervals.
We followed a similar approach to assess the effect of the risk factors of interest on the temporal trend of severe maternal morbidity. First, we calculated and plotted the annual prevalence of severe maternal morbidity and each risk factor. We then examined the trend in severe maternal morbidity using multivariable logistic regression models with delivery year as the independent variable. We sequentially adjusted the model for confounders and each risk factor of interest (advanced age, obesity, comorbidity, cesarean delivery) to determine their contributions to the trend. Finally, we calculated the predicted prevalence of severe maternal morbidity over time by stratifying the data by delivery year and modeling the association between each risk factor and severe maternal morbidity in each delivery year using multivariable regression models. This analytical step facilitates understanding the impact of the risk factors of interest on the population burden of severe maternal morbidity over time.
The CDC has reported that blood transfusions have driven the national increasing trend of severe maternal morbidity [
3]. We therefore repeated all analyses for the outcome of severe maternal morbidity excluding cases for which blood transfusion was the only qualifying indicator (referred to below as ‘transfusion-only’ cases). All analyses were performed in R 3.4.2 and SAS 9.4.