Hospitalizations of pregnant HIV-infected women in the USA prior to and during the era of HAART, 1994–2003

Abstract

Background: 

The literature on whether HIV infection and its complex antiretroviral treatments confer a higher risk for adverse pregnancy outcomes is controversial.

Objective: 

We compared rates of hospitalization for select morbidities among HIV-infected and uninfected pregnant women in the USA.

Design and Methods: 

Using data from the 1994–2003 Nationwide Inpatient Sample, we used descriptive statistics and multivariate logistic regression to examine socio-demographic characteristics, morbidity outcomes and time trends.

Results: 

There were approximately 6000 hospitalizations per year of HIV-infected pregnant women in the USA. HIV-infected women were more likely to be hospitalized in urban hospitals, in the South, have Medicaid as the expected payer, have longer hospitalizations and incur higher charges than uninfected women. Hospitalizations for major puerperal sepsis, genitourinary infections, influenza, bacterial infections, preterm labor/delivery, and liver disorders were more frequent among pregnant HIV-infected women than their uninfected counterparts. However, rates of pre-eclampsia and antepartum hemorrhage were not significantly different. While rates of inpatient mortality and various infectious conditions decreased between 1994 and 2003, the rate of gestational diabetes increased among HIV-infected pregnant women.

Conclusions: 

HIV-infected pregnant women in the USA continue to be at higher risk for morbidity and adverse obstetric outcomes. With the introduction of antiretroviral therapy, rates of most of the conditions examined have either decreased or remained stable, hence current antiretroviral regimens do not seem to be associated with major adverse pregnancy outcomes on a population basis. The increase in gestational diabetes among HIV-infected women may be associated, in part, with antiretroviral therapy and merits further attention.

Introduction

Widespread use of HAART among HIV-infected women in the USA has led to dramatic decreases in the rate of perinatal HIV transmission ^[1]. Guidelines issued by the US Public Health Service Task Force emphasize the importance of HAART during pregnancy for maternal health and for preventing the transmission of HIV to the infant ^[2]. Many pregnant women who know of their status are already in treatment prior to their pregnancy. Current estimates are that 6000–7000 women in the USA with HIV deliver each year ^[1].

The benefits of HAART notwithstanding, concerns have been raised about the safety of new antiretroviral agents, including protease inhibitors (PI), particularly during pregnancy. Antiretroviral medications can cause anemia, lactic acidosis, liver dysfunction, disorders of glycose metabolism or diabetes, and hyperlipidemia ^[3–7]. Reports, mainly from Europe, have suggested that combination antiretroviral therapy might be associated with increased rates of prematurity ^[4–8], particularly when PI are used during early pregnancy or even before pregnancy ^[9]. Other studies, however, including studies of large US cohorts have not confirmed this association ^[10–13]. Whereas studies from the pre-HAART era reported a higher incidence of low-birth-weight infants and premature delivery among HIV-infected women than among uninfected women ^[14], results from a more recent US study suggested that HAART is associated with improved obstetric outcomes and few maternal complications ^[13]. In addition, pregnancy outcomes among women with HIV were reported to be similar to those of HIV-negative women in a recent small prospective study of women at six US centers ^[15].

Many of the studies of pregnancy outcomes among women with HIV, however, have had limitations such as relatively small numbers of subjects, a retrospective design, and a lack of adjustment for other factors that could influence rates of adverse pregnancy outcomes. Given the conflicting evidence concerning the impact of HIV infection on adverse pregnancy outcomes in the USA, a study of this impact based on reliable nationwide data is clearly needed. The Healthcare Cost and Utilization Project (HCUP) provides a unique and extensive dataset containing a wide range of information on hospitalizations during pregnancy among US women with and without HIV infection from 1994 (before HAART was introduced) through 2003. Using the HCUP database, we first compared the demographic characteristics, length of hospital stay, and charges associated with hospitalization among HIV-infected women with those among uninfected women. We then compared rates of select adverse pregnancy outcomes that could be related to HIV infection or its treatments among the two groups of women after adjusting for maternal demographic characteristics, hospital characteristics, and the presence of other co-morbidities. We also examined trends in the rates of these outcomes from 1994 through 2003, a period that encompasses the advent of HAART around 1997.

Methods

We obtained hospital discharge data from the HCUP Nationwide Inpatient Database (NIS), one of a family of research databases and software tools sponsored by the Agency for Healthcare Research and Quality (AHRQ) in partnership with state-level data-collection organizations to provide national estimates of inpatient care delivered in the USA ^[16].

The NIS is a stratified probability sample of approximately 20% of all US community hospitals as defined by the American Hospital Association (AHA). The AHA defines community hospitals as nonfederal, short-term (average length of stay < 30 days) general and specialty hospitals whose facilities are open to the public. Hospitals are selected for the NIS on the basis of five characteristics: geographic region, ownership, rural/urban location, teaching status, and bed size. The NIS includes all inpatient stays from an average of roughly 1000 hospitals; when weighted, the NIS data on seven million discharges provide national estimates of inpatient care ^[17]. It is the largest collection of inpatient care data in the USA and includes patient demographic data and diagnostic/procedural data, as well as facility information.

We analyzed NIS data from 1994 to 2003, a period during which several revisions were made to the NIS sample design ^[18]. One revision was that as more states made their data available to HCUP, the sampling frame increased from 17 states in 1994 to 37 states in 2003. Another revision occurred in 1998 when the NIS sampling methods were changed to better reflect the cross-sectional population of the hospitals represented by the sample: the hospital stratification variables were redefined, rehabilitation hospitals were excluded from the NIS hospital universe, sampling preferences were no longer given to prior year NIS hospitals and AHA hospital unit discharges were used instead of total facility discharges which included nursing home unit discharges. In addition, there were changes over time to definitions and availability of NIS database variables. To account for these changes and to facilitate a trend analysis based on multiple years of NIS data, we used an alternate set of NIS discharge and hospital weights for the 1994–1997 HCUP NIS data sets that were calculated in the same way as those for later NIS data sets ^[18].

Our analysis was restricted to pregnant women 15–44 years of age. We attempted to identify all pregnancy hospitalizations (excluding those for ectopic/molar pregnancy or spontaneous abortion) by searching for those with a primary or secondary International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) discharge code of 640–677, V22–V24, V27–V28, or 792.3. Pregnancy hospitalizations that resulted in a delivery were identified by ICD-9-CM code with a fifth digit of 1 or 2 and V27.

We also used ICD-9-CM codes to identify hospitalizations in which the following morbidities or adverse pregnancy outcomes were involved: (i) preterm labor or rupture of membranes (ICD-9-CM codes 644.0, 644.2, 658.1); (ii) preterm delivery (644.21, 658.11); (iii) gestational diabetes ( 648.8); (iv) liver disorders of pregnancy including hepatic steatosis and lactic acidosis (276.2, 571.8, 646.7); (v) pre-eclampsia/hypertensive disorders of pregnancy (642); (vi) antepartum hemorrhage (641); (vii) major puerperal sepsis (670); (viii) urinary tract infections (646.6, 590, 595, 597, 599.0, 614.0–614.5, 614.7–614.9, 615, 616); (ix) sexually transmitted infections (647.0–647.2, 090.0–099.9, 054.5, 054.7, 079.88, 078.98); (x) bacterial infections and sepsis during pregnancy (003.1, 020.0, 020.2, 020.8, 020.9, 021.8, 021.9, 022.3, 022.8, 022.9, 023.0–023.9, 030.0–030.9, 031.2–031.9, 032.89, 032.9–033.1, 034.1, 036.3, 036.81, 036.89, 036.9, 037, 038.0–038.9, 039.2–039.9, 054.5, 790.7, V09); (xi) influenza (480–487); (xii) anemia (648.2, 280–285); (xiii) drug dependence (303.0, 303.9); and (xiv) hepatitis B and C (070.2, 070.3, 070.41, 070.42, 070.44, 070.51, 070.52, 070.54, 070.70, 070.71). To identify women infected with HIV, we used ICD-9-CM codes: 042, 0420–0422, 0429, 0430–0433, 0439, 0440, 0449, 07953, 27910, 27919, 79571, 7958 and V08.

Using Student's t test and the χ² test with a significance level of 0.05, we compared the following data for HIV-infected women with that for uninfected women: age distribution, expected primary payer (a proxy for socioeconomic status), urban/rural location of the hospital, geographic region of the hospital, total and mean length of hospital stay, total and mean hospital charges associated with the hospitalization, and inpatient mortality rate. For each year, we also compared the rate of each adverse outcome per 100 deliveries among HIV-infected women with that among uninfected women using a pair-wise t test and multivariate logistic regression and adjusting for clinically relevant co-morbidities and demographic variables such as maternal age group, expected payer, drug abuse, infections, and anemia and for hospital characteristics such as location. We did not adjust for race, as race information in the database is incomplete. To assess trends in adverse pregnancy outcomes, we compared data from 2003, 2000, and 1997 with that from1994 (the reference year), adjusting for maternal age, expected payer, and delivery status. Hospitalizations in 1994 occurred during what we considered to be the 'pre-HAART era’ when few pregnant women, if any, would have been on monotherapy, given that the US Public Health guidelines for the use of zidovudine during pregnancy were first published in the end of that year; by 2003, HAART was widely available and the preferred prophylaxis for HIV-infected pregnant women, or as treatment for their own health. Our trend analysis thus tracked the outcomes of US pregnant women from the pre-HAART era through the gradual introduction and current widespread use of HAART in the USA. All statistical analyses were conducted with SUDAAN version 9 (Research Triangle Institute, North Carolina, USA).

Results

Characteristics of hospitalizations of pregnant women by HIV status

We estimated that there were slightly over 6000 hospital discharges during pregnancy per year in the USA in women who had an HIV diagnosis from 1994 to 2003, representing approximately 0.14% of all hospitalizations of pregnant women in the USA. We also found that the number of hospitalizations per delivery was slightly higher among HIV-infected women than among uninfected women during each year examined (Table 1). The HIV-infected pregnant women spent a total of approximately 20 000–25 000 days in the hospital per year, and their annual hospital charges ranged from $43.5 million in 1994 to $86 million in 2003 (Table 1). Pregnancy-related hospitalizations of HIV-infected women were, on average, both longer (3.7 days versus 2.64 days in 2003, P < 0.001) and more costly ($13 873 versus $8415 in 2003, P < 0.001) than those of uninfected women (Table 1).

Table 1: Demographic and hospital characteristics and select outcomes among pregnant women aged 15–44 years admitted to US hospitals in 2003 and in 1994, by HIV status [n (%)].

Although the age distribution of discharged pregnant women with HIV infection mirrored that of their uninfected counterparts, the distribution of their expected primary payer, geographic location, and type of hospital differed (Table 1). Medicaid was the expected payer for over two-thirds of HIV-infected pregnant women, and private insurance was the expected payer for fewer than one-fifth, whereas Medicaid was the expected payer for only about a third of HIV-uninfected women, and private insurance was the expected payer for over a half. We also found that pregnant women with HIV were less likely than those without HIV to be hospitalized in rural hospitals (P < 0.001), more likely to be in the South (P < 0.001), and, in 1994, more likely to die during the hospitalization (P = 0.02). Table 1 depicts findings from 2003 and 1994, the last and first year of this analysis, respectively.

Adverse outcomes among HIV-infected pregnant women

The adjusted rates of hospitalizations that included a diagnosis of preterm delivery, major puerperal sepsis, sexually transmitted infections, influenza, and bacterial infections were each significantly higher among HIV-infected women than among uninfected women during each year from 1994 through 2003 (Table 2). In contrast, the average rate of hypertensive disorders of pregnancy/pre-eclampsia and antepartum hemorrhage did not differ significantly between the two groups and the rate of gestational diabetes was significantly lower among HIV-infected women in 1994 but did not differ significantly between the two groups in 2003. In contrast, the rate of liver disorders of pregnancy was significantly higher among HIV-infected women, compared with uninfected ones in 1994 but not in 2003.

Table 2: Rates of select conditions among pregnant women aged 15–44 years hospitalized in the USA in 2003 and in 1994, by HIV status.

Trends over time

The annual number of hospitalizations per delivery decreased between 1994 and 2003 among both groups of women, though the trend was statistically significant only among uninfected women (Table 3). Patient distribution by age, expected payer, and hospital location did not change significantly over the 10-year period, while hospital charges associated with the hospitalization increased sharply for both infected and uninfected women. (Table 3).

Table 3: Multivariate logistic regression analysis for hospitalizations with each outcome among HIV infected and HIV-uninfected pregnant women aged 15–44 years, 1994, 1997, 2000 and 2003 (adjusted for age group, primary payer and delivery status).

Though the mortality rate among hospitalized pregnant women in the USA is very low both among those with HIV infection and those without, it was still substantially higher among those with HIV infection in 2003 (0.13% versus 0.01%) even though the mortality rate among infected women had dropped from 0.47% in 1994. (Table 3). We also found a statistically significant decrease in the rate of sexually transmitted infections, as well as decreasing trends in rates of bacterial infections and sepsis, urinary tract infections, and major puerperal sepsis in HIV+ women (Table 3 and Fig. 1). Although there were similar decreases for these outcomes among uninfected women (Table 3 and Fig. 1), the drop in hospitalizations for sexually-transmitted diseases was more marked among HIV-infected women.

Fig. 1: Annual rates (based on data from the Healthcare Cost and Utilization Project) of select adverse pregnancy conditions per 100 deliveries among hospitalized HIV-infected and uninfected women aged 15–44 years in the USA 1994–2003. (a) Gestational diabetes. (b) Liver disorders. (c) Preterm labor. (d) Preterm delivery. (e) Sexually transmitted infections. (f) Major puerperal sepsis. ^*Based on data from the Healthcare Cost and Utilization Project.

In contrast to the drop in rates for the conditions discussed above, the rate of hospitalizations for gestational diabetes increased between 1994 and 2000 among both HIV-infected and uninfected women, though much more severely among infected women (Table 3). Similarly, there was a trend for an increase in liver disorders of pregnancy since 2000, both in infected and uninfected women (Table 3). We found no significant change in the rates of preterm labor/delivery or antepartum hemorrhage (Table 3 and Fig. 1) or in the rate of influenza hospitalizations among either group of women; pre-eclampsia did not change among HIV-infected women but increased slightly among uninfected ones (Table 3).

Discussion

HIV infection is thought to directly affect pregnancy outcomes only modestly, if at all ^{[10,14,19,20]}. However, the effects of advanced disease and immunosuppression including anemia, malnutrition and infections, as well as other co-existing factors frequently encountered among HIV-infected women such as smoking, alcohol and illicit drug use and other sexually-transmitted infections place them at increased risk for an adverse outcome ^[21–23]. Thus the etiology of adverse obstetric outcomes in HIV-infected women may be difficult to determine. Additional considerations are the effects of antiretroviral agents used in pregnancy. Early reports of antiretroviral treatment of pregnant women observed improved pregnancy outcomes ^[3,6,9,12]. However, with the introduction of multiple antiretroviral classes of drugs with distinct side-effect profiles, reports of adverse effects in pregnant women and their offspring emerged. More recent findings from a large prospectively followed US cohort seem encouraging; antiretroviral use in pregnancy was independently associated with few maternal adverse outcomes (such as anemia, thrombocytopenia, gastrointestinal side-effects and gestational diabetes) and a decreased risk for some obstetric outcomes such as stillbirth, low birth weight, premature delivery and rupture of the membranes ^[13].

Our study has a different design than those cited above: We followed hospitalization patterns among HIV-infected and HIV-uninfected women in the USA from 1994 through 2003 using the HCUP database. Even though we did not have treatment information on the HIV-infected women, our data span the era before the introduction of HAART in 1997 through the years of HAART's widespread implementation during pregnancy and allowed us to identify ecologic associations between HAART use and pregnancy outcomes.

We found that hospitalizations of HIV-infected pregnant women tend to cluster in distinct geographic locations and hospital types, last longer and cost significantly more than hospitalizations of uninfected pregnant women. Medicaid is the expected payer for the great majority of them. These characteristics did not change between 1994 and 2003. However, we were encouraged to find that the inpatient mortality rate among HIV-infected pregnant women declined during this period even as hospital charges approximately doubled for both infected and uninfected women, reflecting both general and HIV-specific hospital practice and reimbursement changes and trends in costs of medical care. Of interest, after adjusting for inflation over the 10-year period 1994–2003, the total charges of $43.5 million associated with hospitalizations of HIV-infected pregnant women in 1994 would translate to $52.9 million in 2003. Thus, the inflation-adjusted increase between 1994 ($43.5 million) and 2003 ($86.1million) was actually less than the apparent doubling.

We did find the rate of several morbidities and adverse pregnancy outcomes, including preterm labor and delivery and a variety of infectious conditions during pregnancy and puerperium to be significantly higher among HIV-infected women than among uninfected women throughout the study period, even after adjusting for sociodemographic factors and co-morbidities. We are unable to determine whether these higher rates were a direct result of the HIV infection or of differences between the two populations unaccounted for by the information available in the NIS database that we were able to control for. We found no significant difference in the rates of hypertensive disorders of pregnancy and antepartum hemorrhage in the two groups of women.

Perhaps more interesting to examine is the trend in these outcomes over the 10-year period. These trends will reflect the changing practices of care overall, as well as the introduction of HAART regimens in particular. The ratio of number of pregnancy hospitalizations per delivery decreased slightly for both infected and uninfected women over the 10-year period. Further, distinct patterns of change have emerged. The rates of bacterial infections, urinary tract infections, sexually transmitted infections, and major puerperal sepsis all had a declining trend among HIV-infected women. Even though this could be ascribed, partially, at least, to HAART, it is of interest that these outcomes had a similar decreasing trend among HIV-uninfected pregnant women as well. This indicates that other changes in care, increased use of outpatient care and/or changes in coding practices have contributed to these trends.

Glucose intolerance and diabetes is a recognized complication associated with the use of PI ^[13]. We found that the rate of gestational diabetes was lower among HIV-infected women than among uninfected women in 1994 but that the rates had equalized by 2003 as the rate among infected women increased more rapidly. Although obesity, a known risk factor for gestational diabetes, is becoming ever more prevalent in the USA ^[24] and is likely to be more prevalent among uninfected women than among those with HIV, we did not have information on maternal body mass index, so we could not account for this factor in our analysis. The increase in the rate of gestational diabetes among HIV-infected women may reflect general trends ^[25,26] or possibly increasing recognition among HIV-infected women given the association with PI; however the widening use of HAART, and particularly the increased use of PI should be considered as a cause. Other studies of the association between PI use and gestational diabetes ^[13,27] have produced contradictory results. Interestingly, we found that the rate of hospitalizations for liver disorders of pregnancy, including hepatic steatosis and lactic acidosis, increased among both infected and uninfected women since 2000. Of interest is the very recent report by Gebo et al., describing an increase in hospitalizations for liver-related complications among HIV-infected adults in the USA following the introduction of HAART ^[28]. The increase observed among uninfected women could suggest that other factors, including changes in medical or coding practices, may in part account for this finding.

In contrast to the Swiss and European Collaborative Studies ^[6,9], but consistent with the US studies ^[10,11] and the French study by Mandelbrot ^[3], we did not observe an increase in the rate of preterm labor and preterm delivery among HIV-infected women from 1994 (pre-HAART era) to 2003 (widespread use of HAART). This result argues against HAART having a major effect on preterm delivery rates. Of interest, we found that HIV-infected women had a consistently higher rate of preterm labor/delivery than HIV-uninfected women, even after adjusting for several covariates. Since, however, we did not have direct individual treatment information or information on stage of disease, we cannot rule out effects of HAART on preterm delivery among particular groups of HIV-infected women or according to length or type of antiretroviral therapy used before or during pregnancy and our result should be interpreted with caution.

Our study has some limitations: Since the unit of measurement was the hospital admission, rather than the individual patient, some patients may be represented more than once in a given year. Further, hospitalizations usually capture the more serious morbidities, and may not give a full account of all existing conditions. To circumvent this problem we included all, and not just the principal, diagnostic codes recorded for each hospitalization. However recording all co-existing conditions may not be thorough. In addition, diagnosis coding may be influenced by other factors, such as reimbursement policies, temporal changes, or differential practices in HIV-infected versus uninfected women. Particularly when examining trends, one should also keep in mind that hospitalization practices for some conditions may have changed over time and that might have affected chronic and acute conditions differently. In addition, information such as race, direct measures of socioeconomic status, smoking, previous obstetric history, other prescription drug use, and obesity, that might affect certain pregnancy outcomes is not available, thus rendering the comparison populations potentially different in aspects that we cannot account for. Another limitation was that other temporal changes in clinical practices or in the outcomes studied, and not solely the HAART, may account for some of our findings. However, the trends in most outcomes among HIV-negative women remained remarkably constant arguing against major changes in practice. Our study is not large enough to exclude small but potentially increased rates of infrequent toxicities, such as lactic acidosis or liver toxicity. Further, the codes we used for capturing preterm delivery many not include some medically indicated ones, which could represent up to 20% of preterm births (authors' unpublished observations).

The strengths of our study include its large sample size, the extent to which the sample was representative of the US population and its presentation of data from before the introduction of HAART through HAART's widespread adoption in US hospitals. This allowed us to compare rates of various pregnancy outcomes among HIV-infected women with those among uninfected women throughout this pivotal period.

This study concurs with previously published US findings that antiretroviral regimens currently being used to treat HIV-infected women during pregnancy are not associated with major adverse pregnancy outcomes on a population basis and seem to confer health benefits. However, HIV-infected women in the USA continue to be at higher risk than uninfected women for adverse maternal and obstetric outcomes. To assess that risk for specific subpopulations, researchers will need to analyze additional prospectively collected data representing different ethnic and geographic populations of pregnant HIV-infected women treated with different types of antiretroviral regimens. As these regimens become increasingly complex and newer agents are introduced, such analyses will be important for ensuring that antiretroviral regimens are providing optimal health benefits for both mothers and their infants.

Acknowledgments

The views expressed in this article are those of the authors and do not necessarily reflect those of the Agency for Healthcare Research and Quality, the Centers for Disease Control and Prevention, or the US Department of Health and Human Services. The authors had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.