Intensive Care Utilization during Hospital Admission for Delivery : Prevalence, Risk Factors, and Outcomes in a Statewide Population

Under Maryland law, information regarding care given during all hospital admissions at the 52 nonfederal, short-stay hospitals in Maryland must be reported to the state Health Services Cost Review Commission (HSCRC), Baltimore, Maryland, which maintains an anonymous database: the Uniform Health Discharge Data Set (UHDDS). Although there are federal hospitals in Maryland in which discharge data are not included in the UHDDS, the majority of deliveries occur in these nonfederal, short-stay hospitals and, therefore, are subject to the mandatory reporting requirements. Thirty-seven of those 52 hospitals provide obstetric services for childbirth in the facility. Home and birthing center births were not recorded in the UHDDS.

The UHDDS data were collected beginning in January 1984. Data for the current study were extracted from the UHDDS for a 14-yr period from January 1984 through December 1997. The sampling frame included all patients with Diagnosis Related Grouping (DRG) codes 370–375 (370–371 for cesarean section and 372–375 for vaginal delivery of a neonate) discharged from a Maryland hospital. All patients admitted to the ICU sometime during hospital admission for childbirth were selected from the HSCRC database and defined as “cases.” In addition, a “control,” defined as a woman who was delivered of a neonate in the same year, but was not admitted to the ICU during hospital admission for delivery, was randomly selected for each case. Controls were randomly chosen for each of the 14 yr using Microsoft Access (Redmond, WA). A random-number generator was used to determine a number (n) between 1 and 10. Every nth patient was selected from the patients not admitted to the ICU with a DRG code for vaginal delivery or cesarean section.

A hospital was given a “major teaching status” if the hospital included both anesthesiology and obstetrics/gynecology accredited graduate medical training programs; a “minor teaching status” if the hospital included one of these training programs; and a “community status” if the hospital did not have any of these training programs. Quota sampling procedures were used such that the proportion of the major teaching hospitals in the control group mirrored the proportion in the population (approximately 11%).

Demographic variables included age, race, payment source, and marital status. Hospital variables included source of admission (home or another hospital) and hospital type (major teaching, minor teaching, or community). Of 37 hospitals that perform obstetric deliveries, 3 met the criteria for major teaching hospital; 5 met the criteria for minor teaching hospital; and 29 met the criteria for community hospital. This classification system was created to allow comparison of ICU use among the three defined hospital types. Hospital delivery volume was computed by averaging the number of deliveries that occurred from 1984 to 1997. To aid in interpretation of the results, hospital delivery volume was divided into three categories: low (< 500 deliveries), medium (500–1,499 deliveries), and high (> 1,500 deliveries). 10 

International Classification of Diseases, 9th Revision, Clinical Modification  (ICD-9-CM) codes for the principal discharge diagnosis, the principal procedure, up to 14 secondary discharge and procedure diagnoses from 1994–1997, and up to five secondary discharge and procedure diagnoses from 1984–1993 were reviewed. The Romano-Charlson comorbidity index was used with a modification of diagnosis codes to reflect the obstetric population. 11–16 

The outcome variables studied included ICU admission and ICU mortality rate. A case–control framework was used to estimate the strength of the association between suspected risk factors and these outcomes.

The first set of analyses compared the 1,023 patients admitted to the ICU after childbirth, with the 1,023 patients matched for calendar year who were not admitted to the ICU after childbirth. These groups were compared for demographics, hospital characteristics, preexisting medical diagnoses, and pregnancy-related medical diagnoses. Significant differences of demographics and hospital characteristics were assessed using univariable logistic regression modeling. Unadjusted odds ratios were used as an index of the strength of the association between each risk factor and ICU use. A final multivariable regression model was constructed to evaluate the association between each risk factor and ICU use, holding constant all other variables in the model. Because of the very low prevalence of medical diagnoses among the control group, significant differences were assessed using the Fisher exact test from two-by-two tables.

In a similar fashion, the second set of analyses aimed to evaluate the risk factors associated with ICU mortality. Cases were defined as the 34 patients who died after ICU admission (during same admission for childbirth); controls were defined as the remaining 989 patients admitted to the ICU after childbirth who were ultimately discharged. Logistic regression modeling, as described previously, was used to assess significant differences between cases and controls for demographics and hospital characteristics. One important difference in the modeling procedure was the evaluation of disease severity between the two groups and the inclusion of a composite variable in the final model to evaluate the influence of the number of medical diagnoses on the risk of ICU death. Because the small cell sizes again precluded the use of logistic regression modeling, significant differences between the two groups regarding medical diagnoses were assessed using the Fisher exact test from two-by-two tables.

During the period from January 1984 through December 1997, there were 822,591 hospital admissions for childbirth in Maryland for the institutions studied. There were 1,023 ICU admissions and 34 ICU deaths during these hospital admissions for childbirth. During the 14-yr study period the overall ICU use rate was 0.14% for major teaching hospitals, 0.13% for minor teaching hospitals, and 0.11% for community hospitals. Although the median ICU length (duration) of stay (LOS) was 2 days (range, 1–38 days), 45% of ICU admissions had an ICU LOS of 1 day.

The results of the multivariable logistic regression model for prediction of ICU admission are presented in table 1. Patients admitted to the ICU were significantly more likely to be older than 35 yr (23.1 vs.  17.1%; odds ratio [OR]= 1.4; confidence interval [CI]= 1.05–1.81;P = 0.02); to be black (44.6 vs.  31.8%; OR = 1.8; CI = 1.38–2.30;P < 0.001), or to be from another racial category other than white or black (5.1 vs.  1.2%; OR = 5.9; CI = 2.60–12.77;P < 0.001); to be cared for in a minor teaching hospital (23.2 vs.  14.5%; OR = 2.0; CI = 1.48–2.60;P < 0.001); or to be transferred from another hospital (3.2 vs.  1.6%; OR = 2.51; CI = 1.23–5.14;P = 0.01). Notably, the adjusted odds ratios did not vary substantially from the unadjusted odds ratios obtained from the univariable models, suggesting that age, race, hospital type, and source of admission independently and in combination were associated with ICU admission. Marital status was not observed to be significantly associated with ICU admission.

More than 70% of ICU admissions during hospital stay for childbirth were associated with cesarean section, compared with 22.9% of controls (OR = 9.0; CI = 7.24–11.16;P < 0.001). Table 2illustrates the diagnoses associated with ICU admission, ordered by frequency of occurrence during hospital admission for childbirth. The six most common diagnoses associated with ICU admission were preeclampsia or eclampsia (36.6%), postpartum hemorrhage (28.3%), placental abruption or previa (15.8%), pulmonary complications (13.5%), infection (5.2%), and anesthesia-related complications (4.4%) (P < 0.001). The prevalence of preexisting and pregnancy-related medical diagnoses among the ICU admissions and controls is shown in table 2. Compared to controls, cases had a significantly higher proportion of asthma, cardiac disease, drug dependence, diabetes mellitus, and all pregnancy-related medical diagnoses (P < 0.05).

Anesthesia-related complications during labor and delivery are classified as pulmonary, cardiac, central nervous system, or other complications of anesthesia or sedation in labor and delivery (ICD-9-CM codes 668.0–668.2, 668.8, and 668.9). First, of the 45 cases (4.4%) with anesthesia-related complications during labor and delivery, 84% had one anesthesia-related complication diagnosis, 9% had two anesthesia-related diagnoses, and 7% had three anesthesia-related diagnoses. Second, of these 45 cases with anesthesia-related complications, 67% were classified as pulmonary, 29% were classified as cardiac, 13% were classified as central nervous system, and 13% were classified as other complications of anesthesia or sedation in labor and delivery. Third, of these 45 cases, 64% were associated with cesarean section. Fourth, of these 45 cases, 9% were associated with ICU mortality. Cardiac complications were defined as postpartum cardiomyopathy (ICD-9-CM code 674.8); cardiac disease was defined as valvular–endocardial disorders (ICD-9-CM code 424); shock during or after labor and delivery was defined as obstetric shock (ICD-9-CM code 669.1); pulmonary complications were defined as aspiration pneumonia, pulmonary edema, acute respiratory failure, and acute respiratory distress syndrome (ICD-9-CM codes 507.0, 518.4, 518.5, and 518.8); and cerebrovascular event was defined as cerebrovascular disorder in the puerperium, subarachnoid or intracerebral hemorrhage, and seizure or cerebrovascular accident (ICD-9-CM codes 674.0, 430, 431, 432, and 436). Nineteen of 1,023 ICU admissions (1.9%) had an ICD-9-CM diagnosis (V237) of “insufficient prenatal care.”

The annual ICU mortality rate (defined as number of ICU deaths per 1,000 ICU admissions) over the 14-yr period studied ranged from 1.0 to 8.3%, with an overall mortality rate of 3.3%. ICU deaths were associated with a significantly longer duration ICU LOS (mean 7.7 days with an SD of 9.9 days for ICU deaths vs.  2.7 days with an SD of 3.3 days for ICU survivors;P < 0.001).

Table 3shows the results of the final multivariable logistic regression model for prediction of ICU mortality. The groups did not significantly differ with regard to age, marital status, payment source, hospital type, or source of admission. However, black race was associated with approximately a twofold increased risk of ICU death (65% of ICU deaths were of black people; OR = 2.90; CI = 1.18–7.20;P = 0.02), and being a race other than white or black was associated with a threefold increased risk of ICU death (OR = 3.9; CI = 0.96–15.8;P = 0.06). In addition, 91.2% of ICU deaths occurred in high-volume hospitals versus  8.8% in medium-volume hospitals and 0% in low-volume hospitals (P < 0.05; data not shown in a table).

Although the groups differed with respect to hospital type, with 8.1% of ICU admissions in major teaching hospitals resulting in ICU deaths versus  2.5% of ICU admissions in minor teaching hospitals and 2.8% of ICU admissions in community hospitals, this association became nonsignificant (P = 0.08) with the introduction of the disease severity index to the logistic regression model.

Table 4lists the diagnoses associated with ICU deaths by frequency of occurrence during hospital admission for childbirth. The proportion of cesarean sections did not differ between the two groups studied (67.6 vs.  72.7%). The six most common diagnoses associated with ICU death were postpartum hemorrhage (44.1%), preeclampsia or eclampsia (38.2%), pulmonary complications (29.4%), shock (26.5%), cerebrovascular event (20.6%), and anesthesia-related complications (11.8%).

This study evaluated a large statewide population database to determine ICU use and intensive care outcomes during hospital admission for childbirth. In London, the Confidential Enquiries into Maternal Deaths (CEMD) attempts to define morbid and mortal outcomes but does not address ICU use. 17–19In the United States, others have described anesthesia-related mortality, but the methods used did not allow for evaluation of ICU use. 2Kaunitz et al.  20stratified safety of obstetric care by the number of annual deliveries in a hospital, but nonfatal morbidities could not be assessed by the methods used.

The frequency of use of ICU resources by obstetric patients is of interest to healthcare providers who may be involved in both the critical care of obstetric patients and the planning of intensive care facilities for obstetric patients. Such use statistics may allow for the most efficient allocation of patient care resources. Estimates of perinatal mortality rates and the need for perinatal ICU services previously published vary widely among institutions. 1–9,17–28The current study shows that the incidence of ICU admission after childbirth is low and that ICU mortality in these patients is not as high as generally reported. 7–9,21–24These findings may reflect lower ICU use comorbidities found in a statewide population versus  patients previously reported by others receiving care predominantly in a tertiary care facility. This uniformity of ICU use across different hospital types is interesting because of the probability that obstetric patients in teaching hospitals have a higher acuity index for illness, as shown in the current study. Although the ICU use rates were similar among the hospital types, the criteria for ICU admission is unclear without individual patient chart review. Community hospitals possibly may have a lower threshold for ICU admission.

Previous investigations of obstetric patient ICU morbidity and mortality outcomes have evaluated data from a single center. 7–9,21–25The current study, however, included data from 37 hospitals, used a matched control group to compare the frequency of morbid complications, and used a statewide population. Previous studies, therefore, may be misleading in developing statewide or national healthcare policy regarding perinatal ICU location, design, and use.

The volume of deliveries performed by a particular hospital may also influence the provision of care for the critically ill obstetric patient. Several investigators have reported that obstetric patients who experience complications are selectively transferred to larger hospitals. 20,29Overall, the current study showed a trend for higher volume hospitals to receive more transfers. Because the actual hospital source of admission for the transfer could not be determined using the UHDDS, we could not determine whether these transfers were from hospitals with low volumes of deliveries. Some have postulated that hospitals with less active obstetric services may not be well-staffed and equipped to respond effectively to life-threatening obstetric situations. Implications regarding individual hospital volume of deliveries as the volume relates to the care of the critically ill obstetric patient is yet to be determined and requires further research.

Perinatal ICU services varies widely from the dedicated obstetric ICU 7to the transfer of a critically ill obstetric patient to the ICU 9,22–25,27,30or to another hospital ICU. 8Although the current study indicates that an ICU located within and solely for the use of the obstetric suite would likely be relatively underused (even in major or minor teaching hospitals), Mabie et al.  12suggest there may be benefits to that approach. In their study, the convenient location of the obstetric ICU may have prompted more frequent ICU admissions, especially for closer observation, giving rise to a high ICU use rate of 0.90% for their patients versus  the ICU use rate of 0.12% observed in the current study.

The optimal way to treat critically ill obstetric patients may depend on many factors, some unique to a given healthcare institution. There are various policies in a particular institution for intensive treatment of critically ill obstetric patients, often dictated by the availability of facilities. 22Because the volume of deliveries and the need for perinatal ICU services varies widely, consideration for planning of obstetric services for the future may be warranted. Future implications may include “regionalized” perinatal ICU services in settings in which ICU use is low.

The median ICU LOS in the current study was 2 days, and 45% of ICU admissions had an ICU LOS of 1 day. This short ICU stay has also been shown in other studies. 27Some obstetric patients may have morbidity associated with pregnancy-related complications, whereas other obstetric patients may have morbidity associated with exacerbation of underlying medical conditions during pregnancy. However, the short-duration ICU stay shown in obstetric patients during hospital admission for childbirth highlights the finding that the majority of obstetric patients, even those requiring ICU care, recover rapidly after childbirth, and that long-term ICU use is rare.

We note limitations to the current study. First, the UHDDS did not include the type of anesthesia (regional vs.  general vs.  intravenous sedation) for labor and delivery; thus, making it difficult to determine the incidence of anesthesia-related complications based on anesthesia technique. Second, the coding of comorbid diseases and complications in the HSCRC database may not be as accurate as the coding of the principal procedure. However, analysis of the quality of the Maryland HSCRC database indicates a low error rate for the coding of our outcome variables. 14Third, before 1993, the data in the records of the UHDDS included only up to five discharge diagnosis codes and procedure codes for each patient. Thus, morbidity diagnoses may be underreported for these early years. Although this may have limited our assessment of trends in morbidity, it did not bias the current results because all hospitals were included in the analysis for every year. Fourth, some patients may have been missed from analysis if these particular patients were transferred from other facilities at which the actual delivery occurred (e.g. , death after a postpartum transfer might be missed). Fifth, postpartum patients readmitted to the hospital after discharge from childbirth admission with complications may be missed from analysis solely based on DRG coding 370–375. Thus, the UHDDS may underestimate ICU admission and mortality rate. Sixth, mortality data only included deaths that occurred during that hospital admission and would underestimate pregnancy-related mortality, conventionally defined as death occurring up to 1 yr after childbirth. 3,4Seventh, the UHDDS precluded (1) assessment of the urgency of the cesarean section because all admissions were coded as “admission for delivery” (e.g. , without “elective” or “emergent” coding) and (2) determination of whether ICU admission occurred before or after childbirth.

In conclusion, we demonstrate a low incidence of ICU admission and mortality in obstetric patients during hospital admission for childbirth. A dedicated obstetric ICU may thus be unlikely to be cost-effective in most hospital settings. Our findings may have implications (1) in planning perinatal hospital ICU facilities, (2) in shaping healthcare delivery policies, and (3) for reassessing the influence of reallocation and consolidation of manpower and healthcare services.

The authors thank Dr. Steve Frank, Department of Anesthesiology and Critical Care Medicine, Johns Hopkins University School of Medicine, for his invaluable review of the manuscript.