We used data from the Nationwide Inpatient Sample (NIS) from 1994–2007. We included several years of data in order to have a sufficient number of discharge records among HIV-infected women to create reliable estimates for selected surgical procedures. The Healthcare Cost and Utilization Project (HCUP) includes databases and software tools developed through a partnership among private industry, states, and the federal government. The NIS, the largest all-payer database of inpatient stays in the United States, is a key component of HCUP. The NIS incorporates data from approximately 8 million hospital stays per year, and it approximates a 20% stratified sample of community hospitals in the US (AHRQ, 2010). Sampling is stratified on location (rural or urban), hospital size, region of the country, teaching status, and type of ownership (public or private). As of 2007, 40 states contributed data to the NIS, and hospitals in the sampling frame comprised approximately 90% of US hospital discharges [14
We analyzed discharge records from women aged 15 and older, excluding hospitalizations that included delivery (International Classification of Diseases, 9th revision, Clinical Modification (ICD-9-CM) codes 650; V27). We further limited the study population to include only hospitalizations during which one of the following procedures had been performed: oophorectomy (procedure codes 65.3, 65.31, 65.39, 65.4, 65.41, 65.49, 65.5, 65.51, 65.52, 65.53, 65.54, 65.6, 65.61, 65.62, 65.63, and 65.64), salpingectomy for ectopic pregnancy (procedure code 66.62), bilateral tubal sterilization (procedure codes 66.2, 66.21, 66.22, 66.29, 66.3, 66.31, 66.32, and 66.39), dilation and curettage (procedure codes 69.0, 69.01, 69.02, and 69.09) or hysterectomy. Hysterectomy was defined as abdominal hysterectomy (68.3, 68.39, 68.4, and 68.49), vaginal hysterectomy (68.5; 68.59), or total laparoscopic hysterectomy/laparoscopic-assisted hysterectomy (68.31, 68.41, and 68.51). Any type of hysterectomy with the code 54.21 was also coded as laparoscopic-assisted. We focused on these gynecologic surgical procedures because they were the most common (at least 150 surgeries performed) among hospitalizations of HIV-infected women in our dataset. With the exception of hysterectomy with concomitant oophorectomy, we excluded hospitalizations during which multiple gynecologic surgeries were performed. The NIS does not include patient identifiers, and the unit of analysis is the hospital discharge record. Although some patients may have been admitted multiple times during the study period for procedures we examined, we expect this to be rare.
Our primary outcome, experiencing at least one complication of surgical procedures, was defined as experiencing extended length of stay; transfusion; anemia due to acute blood loss; accidental puncture or laceration during a procedure; hemorrhage, hematoma, or seroma complicating a procedure; urinary tract infection; fever; other postoperative infection; urinary tract complications including urinary retention and ureteral obstruction; paralytic ileus; any of several less common complications (e.g., thromboembolism and postoperative shock). Extended length of stay was defined as being at or above the 90th percentile for that specific surgical procedure. This was equivalent to ≥5 days for hysterectomy with oophorectomy, ≥4 for hysterectomy alone, ≥9 for oophorectomy alone, ≥4 for salpingectomy for ectopic pregnancy, ≥5 for bilateral tubal sterilization, and ≥6 for dilation and curettage. Other complications were defined based on relevant ICD-9 codes.
Our primary independent variable was HIV status (ICD-9-CM codes 042, 043, 044, 079.53, 279.10, 279.19 795.71, 795.8, and V08). We defined comorbidity as presence of ≥1 of the following conditions/behaviors that could put women at increased risk for complications of the gynecologic surgeries we examined: obesity, diabetes, cardiac condition or hypertension, anemia, gastrointestinal ulcers, smoking, and alcohol or substance abuse. Based on review of the literature, we selected relevant ICD-9 codes for these conditions/behaviors, and we defined them accordingly.
We compared discharge records of HIV-infected and -uninfected women undergoing the gynecologic surgeries we examined on various descriptive characteristics of patients and hospitals, including age, primary payer, hospital teaching status/location, hospital region, and presence of any comorbidity. Race was not examined because some states do not report race/ethnicity data, and, among states that do report this, there are often inconsistencies and missing values in the data. Comparisons were evaluated with chi-squared tests (alpha = 0.05).
For each surgery, we used multivariable logistic regression to estimate the association between HIV infection status and experiencing ≥1 complication of surgery, adjusting for patient age, primary payer, year of hospitalization, and presence of any comorbidity. Because of the possibility that associations between HIV infection status and the occurrence of complications might differ depending on whether comorbidity was present, we tested for interaction between HIV infection status and presence of any comorbidity. Associations for which statistically significant interaction was detected (alpha = 0.05) are presented separately for women with and without comorbidity. In addition, we conducted multivariable logistic regression to estimate the association between HIV infection and the 4 most common complications in our sample. These included extended length of stay, transfusion, anemia due to blood loss, and all infectious complications combined (i.e., experience of urinary tract infection; fever; other postoperative infection; or contaminated or infected blood, other fluid, drug, or biological substance). Again, we adjusted models for patient age, primary payer, year of hospitalization, and presence of comorbidity, and we tested for interaction between HIV infection and any comorbidity. Finally, for each surgery we examined, we tested for major shifts over time by using multivariable logistic regression (with adjustment for the same variables), to compare, for hospitalizations among HIV infected women, the odds of extended length of stay, infectious complications, and all other complications combined during the time periods preceding (1994–2000) and during (2001–2007) widespread implementation of highly active antiretroviral treatment (HAART) in the US.
We used SAS-callable SUDAAN 9.0 software (RTI International, Research Triangle, Durham, NC, USA) to account for the multistage probability sampling design. All results are based on weighted estimates of hospitalizations in the US during the period of study. In 1998, the NIS sample design changed to better reflect the population of hospitals in the sample. Specifically, short-term rehabilitation hospitals were excluded, stratification variables were redefined, the discharge definition was changed, and previous-year NIS hospitals were no longer given sampling precedence. To account for the change in sample design, we applied an alternate set of NIS discharge and hospital weights (based on the 1998 design) to 1994–1997 data [15
]. All programming was independently duplicated by a second data analyst. Because the study utilized deidentified data from a publicly available data set, the Centers for Disease Control and Prevention determined that human-subject research review was not required.