We used data from the Nationwide Inpatient Sample (NIS) of the Healthcare Cost and Utilization Project (HCUP), a federal-state-industry partnership sponsored by the Agency for Healthcare Research and Quality (AHRQ).(7
) The NIS is the largest all-payer inpatient care database in the United States and uses discharges from the sampled hospitals to produce nationwide estimates. The number of states participating in the NIS ranged from 22 in 1998 to 37 in 2005; the sampling frame for the 2005 NIS is a sample of hospitals that comprises approximately 90 per cent of all hospital discharges in the United States. Details of the sampling strategy are described elsewhere.(7
) The weighted nationwide annual estimates of the total number of discharges from all U.S. hospitals ranged from 34,874,001 in 1998 to 39,163,834 in 2005. Because the NIS excludes data elements that could directly or indirectly identify individuals, this research was determined to be exempt by the institutional review board of the Centers for Disease Control and Prevention.
Our analysis included all 1998–2005 delivery hospitalizations. To identify delivery hospitalizations, we used a combination of delivery-related diagnosis and procedure ICD-9-CM codes as described in detail elsewhere.(8
) Hospitalization records were excluded if their discharge summary contained ICD-9-CM codes for hydatidiform mole (630), other abnormal product of conception (631), ectopic pregnancy (633), or abortion (632, 634, 635, 636, 637, 638, 639 and 69.01, 69.51, 74.91, 75.0).
We assessed severe/“near-miss” maternal morbidity using the WHO method of disease-based and management-based groups.(9
) The disease-based group consisted of severe anesthesia complications, renal failure, heart failure, puerperal cerebrovascular disorders, obstetric pulmonary embolism, pulmonary edema, adult respiratory syndrome, deep venous thrombosis, disseminated intravascular coagulation, sepsis, and shock. The management-based group included hysterectomy, blood transfusions, and ventilation.
First, we identified hospitalizations with severe obstetric complications using condition-specific ICD-9 codes. Second, we used mortality, transfer from or to another health care facilities, and length of stay as criteria to distinguish between hospitalizations with severe complications and hospitalizations with preexisting conditions. Since hospitalizations with mortality or transfer may have a short length of stay, we considered them as hospitalizations with severe complications regardless of the length of stay. Among the remaining hospitalizations we reclassified hospitalizations with short length of stay (defined as hospitalizations with length of stay <90th percentiles calculated separately for vaginal, primary and repeat cesarean deliveries), as hospitalizations without severe complications unless ventilation or hysterectomy were also included. The percent of excluded hospitalizations ranged from 10.7% to 46.4% for renal failure and complications of anesthesia, respectively when this criterion was applied. The ICD-9-CM codes for pregnancy and labor/delivery complications included in the study are listed in .
International Classification of Diseases, Ninth Revision complication codes used to describe severe maternal morbidity in the United States, 1998–2005, Nationwide Inpatient Sample.
Since we planned to conduct trend analyses, all changes in ICD-9-CM codes documented by the National Center for Health Statistics that took place during the period of study were assessed and included in our coding algorithm.(10
) The only substantive changes documented were for codes for sepsis in 2002 and codes for cardiomyopathy that have been expanded into a specific code; thus we could not investigate trends in rates of cardiomyopathy. In-hospital mortality was identified using the variable “died during hospitalization” in the HCUP data system.
The unit of analysis was a delivery hospitalization, not an individual. We performed two type of testing: a linear trend in prevalence of severe labor/delivery complications as well as in prevalence of other potential determinants of the complications across the four intervals (1998–99, 2000–01, 2002–03, and 2004–05) and a difference in the prevalence of severe labor/delivery complications between each two-year intervals adjusted for other factors. Although we reported linear trends for all four intervals, we showed prevalence and difference in the prevalence only for 1998–99 and 2004–05. We used orthogonal polynomial coefficients that are calculated recursively according to the method of Fisher and Yates to linear trend testing.(11
) The significance level used to test linear trends was set at 99% (i.e. p-value = 0.01 threshold). We applied logistic regression to examine the changes in rates of severe obstetric complications between 1998–99 and 2004–05.
We reported prevalence and examined the trend in prevalence of age groups, insurance status, mode of delivery, and prevalence of selected pregnancy complications from 1998–99 to 2004–2005. We also reported overall rates of each complication as well as rates of each complication by age group or mode of delivery per 1,000 deliveries and assessed their trend by four two-year intervals.
Odds ratios (OR) and 95% confidence intervals (CIs) were calculated to estimate the risk of each severe obstetric complication in three time intervals (2000–01, 2002–03, 2004–05) relative to 1998–99. Only results for 2004–05 are presented. Sequential models were constructed with maternal age only, mode of delivery only, both maternal age and mode of delivery, and a full model also including other potential determinants of severe obstetric complications in order to assess the effect of each factor or group of factors on the relationship between time period (2004–05 vs. 1998–99) and the occurrence of each severe obstetric complication. In addition to maternal age and mode of delivery, we also investigated selected pregnancy conditions and hospital characteristics (region [Northwest, Midwest, South, West], location [rural/urban], teaching status [yes/no], and bed size [small, medium, large]) as potential predictors or covariates of obstetric complications. Since some states do not report race/ethnicity information to HCUP, we did not include this variable in the analysis.(7
We used SAS software (version 9.1, SAS Institute Inc. Cary, NC) to manage data and used SAS-callable SUDAAN software (version 9.0, RTI International, Research Triangle, NC) to account for the multistage probability sampling design. Thus, all results are based on the weighted estimates of delivery hospitalizations in the U.S during the period of study. All programming was independently duplicated by a second analyst.