Data source

Data from the 2010–2014 Nationwide Readmissions Database (NRD) were utilized for this study. The NRD is one of the largest all-payer databases assembled annually by the Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project. Hospital admissions for individual patients are tracked on a state level within a given year. The data contain weights that allow for the calculation of national readmission estimates for both insured and uninsured patients. As of 2014, 22 states accounting for 51% of US residents and 49% of US hospitalizations were included in the NRD. This study was granted institutional review board exemption by both the Columbia University and University of Southern California Institutional Review Boards, as the data are de-identified and publicly available.

Study population

Readmissions for obstetric and postpartum indications within 60 days of delivery hospitalization discharge for women aged 15–54 years were identified. Delivery hospitalizations were identified utilizing International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes of 650 and V27.x; these codes capture >95% of all deliveries. All delivery hospitalizations, including preterm births and stillbirths, were included. We identified readmissions for obstetric and postpartum indications based on primary ICD-9-CM diagnosis and procedure codes. Admission indications included (1) sepsis, (2) heart failure, (3) wound complications, (4) hemorrhage, (5) breast complications, (6) urinary tract infection, (7) uterine infections, (8) hypertensive disorders, (9) thrombotic disorders, (10) upper respiratory infection, (11) dilation and curettage, (12) hysterectomy, (13) transfusion, and (14) laparotomy. Prior studies evaluating postpartum readmissions have included psychiatric, gallbladder disease, and appendicitis diagnoses ; because these indications for admission may not have been related to postpartum care or may have been more likely to be planned to occur at a different hospital than the delivery hospital, they were not included in this analysis.

Within the NRD, patients have a unique identifier that allows tracking across multiple hospitalizations within a given state within a calendar year. Only admissions from January through October were included, as the NRD is assembled annually and cannot be linked across years; delivery hospitalizations during November and December were not included because readmissions for the subsequent 60 days cannot be fully ascertained. Readmissions were categorized either as occurring at the same hospital as the delivery hospitalization discharge or as fragmented and occurring at a different hospital. For patients with >1 readmission, only the first readmission within 60 days of discharge was included in the analysis.

Outcomes

The primary outcome for this analysis was readmission fragmentation. Both unadjusted and adjusted analyses with demographic, hospital, and obstetric and medical factors with fragmentation as the outcome were performed. In addition to the primary outcome, we performed several secondary outcome analyses. First, we evaluated whether readmissions for specific indications were more or less likely to be fragmented. Individual indications for postpartum readmissions were determined using a previously described algorithm based on ICD-9-CM procedure and diagnosis codes, excluding diagnoses for appendicitis, gallbladder disease, and psychiatric indications. For each indication for readmission, this analysis determined the proportion of readmissions that were fragmented.

Second, this analysis assessed risk for SMM during readmissions and whether fragmentation was associated with increased risk for this outcome. SMM was based on criteria from the Centers for Disease Control and Prevention (CDC). The CDC composite includes ICD-9-CM codes for 18 conditions, including eclampsia, stroke, renal failure, hysterectomy, thromboembolism, heart failure, and blood transfusion. Readmissions with and without fragmentation were compared with the CDC’s SMM criteria as the outcome. For the SMM analysis, patients with SMM during their delivery admission were excluded to avoid misclassification.

Third, we evaluated risk for high readmission costs and prolonged length of stay and determined whether fragmentation was associated with increased risk for these outcomes. High readmission cost was defined as cost >90th percentile (≥$11,594.72). Prolonged readmission length of stay was defined as length of readmission stay >90th percentile (≥5 days).

Fourth, we performed a set of ancillary analyses. As the first ancillary analysis we evaluated maternal mortality risk comparing fragmented vs nonfragmented readmissions. Second, because fragmentation may to some degree be a proxy for regionalization of care, we analyzed the proportion of readmitted women who were transferred during their delivery hospitalization to determine to what degree outgoing transfers on a hospital level may be associated with readmission fragmentation. Third, because postpartum hemorrhage and preeclampsia may be highly predictive of SMM we performed a sensitivity analysis excluding these factors as SMM predictors.

Patient, hospital, and clinical factors

Demographic, hospital, and clinical factors from the delivery hospitalization were included in the analysis. Patient factors including payer status (Medicare, Medicaid, private insurance, self-pay, no charge, and other), median income quartile by ZIP code, and patient age (categorized as 15–19, 20–24, 25–29, 30–34, 35–39, and 40–54 years old). Hospital factors included hospital bed size (small, medium, large) and teaching status (metropolitan teaching, non-teaching, and nonmetropolitan). Clinical factors from the delivery hospitalization were identified using ICD-9-CM codes and included cesarean delivery, pregestational diabetes, gestational diabetes, postpartum hemorrhage, systemic lupus erythematosus, multiple gestation, asthma, chronic kidney disease, hypertensive diseases of pregnancy, and chronic hypertension.

Statistical analysis

Demographic analyses were conducted to describe the differences in the baseline characteristics stratified by fragmentation (same-hospital readmission vs fragmented readmission) status. For our primary analysis, we determined the association between the aforementioned patient, hospital, and clinical factors and risk for fragmentation. Univariable log-linear regression models with a Poisson distribution were fit with unadjusted risk ratios (RR) and 95% confidence intervals (CI) as the measures of association. Adjusted multivariable log-linear regression models including all the demographic, hospital, and obstetric and medical factors were performed with fragmented readmission as the primary outcome and adjusted risk ratios (aRR) with 95% CIs as the measure of association. Because transfer to other facilities could affect readmission outcomes, a sensitivity analysis was performed restricted to only patients with routine discharges; routine discharge excludes transfers to other hospitals and skilled nursing facilities. Adjusted log-linear regression models including the same demographic, hospital, and clinical factors as the primary analysis along with fragmented readmission added as a risk factor were additionally performed for the 3 following secondary outcomes: (1) SMM, (2) high readmission costs >90th percentile, and (3) prolonged lengths of stay >90th percentile. Costs were calculated utilizing NRD provided cost-to-charge ratios and adjusted for the cost of inflation to represent 2016 dollars using the Consumer Price Index. Because specific indication for readmission could be associated with cost, length of stay, and risk for severe morbidity, sensitivity analyses were performed repeating the adjusted models with indication for admission as a risk factor. All analyses were performed with SAS 9.4 (SAS Institute, Cary, North Carolina).