- Email: [email protected]
Intrinsic racial differences in the risk of cesarean delivery are not explained by differences in caregivers or hospital site of delivery
American Journal of Obstetrics and Gynecology (2006) 194, 1323–8
www.ajog.org
Intrinsic racial differences in the risk of cesarean delivery are not explained by differences in caregivers or hospital site of delivery Judith H. Chung, MD,a,b,* Thomas J. Garite, MD,b Adele M. Kirk, MHA,c Amie L. Hollard, MD,b,d Deborah A. Wing, MD,b David C. Lagrew, MDb,e Department of Obstetrics and Gynecology, Women’s Hospital, Long Beach Memorial Medical Center, Long Beach, CAa; Division of Maternal-Fetal Medicine, University of California, Irvine, Orange, CAb; Department of Health Services, University of California, Los Angeles, School of Public Health, Los Angeles, CAc; Department of MaternalFetal Medicine, Saint Elizabeth Regional Medical Center, Lincoln, NE d; Saddleback Memorial Medical Center, Laguna Hills, CAe Received for publication June 22, 2005; revised November 3, 2005; accepted November 28, 2005
KEY WORDS Cesarean delivery Race Ethnicity
Objective: This study was undertaken to assess whether racial differences in the risk of cesarean delivery result from differing practices of their caregivers or the hospitals at which they deliver. Study design: A retrospective cohort study was performed using the Perinatal Database of the Memorial Health Care System. Logistic regression was used to estimate the risk of primary cesarean delivery among patients eligible for labor. The contribution of hospital and physician level cluster correlation was evaluated using fixed and random effects regression models. Results: Compared with white patients, black and Hispanic patients were 75% and 22% more likely to undergo primary cesarean delivery. Further adjustment for hospital and physician level cluster correlation resulted in persistently increased risks of primary cesarean delivery in black (54%) and Hispanic patients (12%). Conclusion: Hospital site of delivery and individual physician practices do not fully explain racial differences in the risk of primary cesarean delivery. Ó 2006 Mosby, Inc. All rights reserved.
The rate of cesarean delivery in the United States has increased 5-fold compared with the mid 1960s, when the rate was a mere 4.5% of all births.1 This has resulted in rising health care costs and morbidity for obstetric patients, with an unproven benefit with respect to birth * Reprint requests: Judith H. Chung, MD, Long Beach Memorial Medical Center Women’s Hospital, 2801 Atlantic Ave, Long Beach, CA 90806. E-mail: [email protected] 0002-9378/$ - see front matter Ó 2006 Mosby, Inc. All rights reserved. doi:10.1016/j.ajog.2005.11.043
outcomes.2,3 As such, there has been an increased emphasis placed on attempts to decrease the cesarean delivery rate.4 In fact, reducing the number of cesarean births is 1 of the Maternal, Infant, and Child Objectives of Healthy People 2010.5 In the 1990s, vaginal birth after previous cesarean delivery was identified as 1 method of reducing the number of cesarean deliveries performed.6,7 However, recent concerns regarding the safety of vaginal birth after cesarean delivery have resulted in a decline in the
1324 number of such patients undergoing a trial of labor.8 Therefore, there has been an increased emphasis on the prevention of primary cesarean delivery as a better means of decreasing the overall cesarean delivery rate in this country. As such, a better understanding of the risk factors for primary cesarean delivery is needed. Previous studies have shown a difference in the rate of cesarean delivery among various race/ethnic groups.1,9-11 However, these studies may be biased, as the potential effects of delivering physician and hospital site of delivery were not addressed. Therefore, this study attempts to evaluate ethnic differences in the risk of cesarean delivery, after adjusting for these potential omitted variables using fixed and random effects regression modeling.
Methods With consent from the Institutional Review Board at Long Beach Memorial Medical Center, the Memorial Care Perinatal Database, from January 1, 1997, to June 30, 2002, was used for this study. As previously described, this is a database that contains antenatal and birth information on all pregnant women delivering within the Memorial Health Care System, a 4-hospital network that serves a diverse patient population in Southern California.12 Two of the 4 hospitals (hospitals B and D) provide both 24-hour perinatal consultative services and regional neonatal intensive care services. One of the 4 hospitals (hospital B) has a resident service, supervised by the division of perinatalogy. All 4 hospitals have 24-hour anesthesia availability. Rates of public insurance coverage vary from 5% in hospital D to 55% in hospital A. Of 54,144 observations, multiple gestations, patients with mixed, other, or unknown ethnicity, and patients with a previous cesarean delivery were excluded from the study, leaving 39,522 observations. Of these patients, only those between 24 and 42 weeks’ gestation and who were eligible for labor were included in the study (37,688 observations). Ethnicity was determined by the patient. This was recorded during the course of her prenatal care and confirmed on admission to the hospital for delivery. Intercooled Stata version 8.0 (Stata Corp, College Station, TX) was used for data management and statistical analysis. Demographic data were presented as a function of ethnic group (black, Asian, Hispanic, and white). The c2 test was used for bivariate analysis of categorical data and analysis of variance (ANOVA) for continuous data. Logit regression was then performed with cesarean delivery as the dependent variable and ethnicity as the primary independent variable. The adjusted logistic regression was designated as the ‘‘Patient Characteristics Model.’’ Covariates in this model were identified a priori and included induction of labor,
Chung et al maternal age, birth weight, excess maternal weight gain, defined as a weight gain of greater than 30 lbs, epidural anesthesia, diabetes mellitus, parity, preeclampsia, placental abruption, gestational age, and the following interactionsdinduction*preeclampsia and diabetes* preeclampsia. The clustering of patients within physician providers and delivering hospitals was addressed by using fixed and random effects regression modeling. Such regression modeling is imperative for the correct calculation of standard errors, which can be underestimated if clustering is unaccounted. In addition, the use of fixed and random effects regression modeling accounts for unmeasured factors at the physician and hospital level. These unmeasured factors can influence the outcome and be correlated with patient-level covariates, leading to biased estimates of the effect of patient-level factors, such as race. Therefore, hospital-level clustering was addressed by using dummy variable logit regression with fixed effects modeling, because the unobservable effects of hospital site of delivery were assumed to be correlated with other covariates in the model. This model was designated as the ‘‘Hospital Fixed Effects Model.’’ In a similar fashion, physician-level clustering was addressed with the use of fixed effects. This model was designated as the ‘‘Physician Fixed Effects Model.’’ Finally, physician-random effects and hospital-fixed effects were simultaneously modeled by using the Stata ‘‘xtlogit’’ command with the random effects option, where hospital-fixed effects were included as dummy variables in the main regression equation. Physician-fixed effects were not included in this model, as physicians delivered primarily at 1 hospital. Thus, any physician-fixed effects should be subsumed within the hospital-fixed effects. Hospital site of delivery information was available for all patients in the study. Delivering physician information was available for 37,644 of the 37,688 observations in the study. The 44 patients who did not have a designated delivery physician were eliminated from the regression model. There were 250 different delivering physicians in the data set. During the study period, physicians delivered between 1 and 2,033 patients, with an average of 151 patients per physician. The majority of physicians delivered patients belonging to each of the 4 ethnic groups in the study, and for the most part physicians performed deliveries at only 1 of the 4 hospitals in the study. Of the 250 physicians, 71 physicians performed either all cesarean deliveries or all vaginal deliveries, accounting for 2,137 observations. Because these physicians had no variability in the outcome measure, they were dropped from the Physician Fixed Effect Model. Predicted probabilities for cesarean delivery in all models were obtained and used to calculate the relative risks of cesarean delivery, with white patients as the reference group. Biased-corrected, bootstrapped 95%
Chung et al Table I
1325
Demographic data by ethnic group*
Maternal age (y) Parity R 1 Gestational age (wks) Hospital site of delivery Hospital A Hospital B Hospital C Hospital D Total Diabetes mellitus Preeclampsia Placental abruption Excess weight gain (O30 lbs) Induction of labor Epidural in labor Overall cesarean delivery rate Cesarean delivery rate in patients eligible for labor Birth weight (g) Macrosomia (O4000 g) Low birth weight (!2500 g) 1-min Apgar !5 5-min Apgar !7
White
Black
30.2 G 5.6 (18,569) 53.5% (9,936/18,569) 39.4 G 1.7 (18,569)
25.8 G 6.1 (3,712) 30.0 G 5.6 (3,348) 26.4 G 6.0 (12,059) !.0001 59.8% (2,219/3,712) 51.4% (1,721/3,348) 63.3% (7,628/12,059) !.001 39.0 G 2.3 (3,712) 39.1 G 1.8 (3,348) 39.2 G 2.0 (12,059) !.0001
Asian
10.3% (1,916) 39.7% (7,374) 5.2% (966) 44.8% (8,313) (18,569) 2.7% (500/18,569) 2.6% (485/18,569) 0.2% (45/18,569) 21.4% (3,964/18,569)
3.7% (135) 91.5% (3,398) 1.1% (40) 3.7% (139) (3,712) 3.3% (121/3,712) 3.5% (128/3,712) 0.6% (22/3,712) 21.4% (795/3,712)
15.3% (512) 59.3% (1,988) 6.5% (217) 18.9% (631) (3,348) 4.9% (165/3,348) 2.3% (77/3,348) 0.4% (13/3,348) 16.5% (551/3,348)
Hispanic
29.5% (3,562) 53.9% (6,505) 6.8% (807) 9.8% (1,185) (12,059) 5.1% (615/12,059) 3.2% (391/12,059) 0.3% (39/12,059) 13.4% (1,615/12,059)
P value
!.001 !.001 !.001 .005 !.001
27.3% (5,038/18,569) 22.9% (850/3,712) 16.6% (555/3,348) 18.9% (2,278/12,059) !.001 76.3% (14,136/18,569) 54.9% (2,037/3,712) 56.9% (1,904/3,348) 41.5% (4,999/12,059) !.001 15.6% (3,053/19,564) 17.2% (670/3,898) 15.1% (528/3,496) 13.2% (1,658/12,594) !.001 11.1% (2,058/18,569)
13.0% (484/3,712)
11.4% (380/3,348)
3490 G 540 (18,569) 14.6% (2,717/18,569) 3.5% (648/18,569)
3233 G 601 (3,712) 3267 G 513 (3,348) 3391 G 581 (12,058) !.0001 7.1% (264/3,712) 6.4% (215/3,348) 11.4% (1,369/12,059) !.001 8.6% (320/3,712) 5.2% (174/3,348) 5.2% (622/12,059) !.001
2.2% (413/18,569) 0.9% (158/18,569)
3.1% (116/3,712) 1.6% (58/3,712)
1.5% (49/3,348) 0.6% (21/3,348)
9.3% (1,123/12,059)
2.2% (269/12,059) 1.0% (125/12,059)
!.001
!.001 !.001
* Analyzed using the c2 test for categorical data and ANOVA for continuous data.
CIs were determined for each predicted probability and relative risk. For each bootstrapped confidence interval, 1000 repetitions were performed, as recommended in the literature.
Results Demographic data by ethnic group are presented in Table I. Overall, there were 37,688 patients that met inclusion criteria. Black and Hispanic women were more likely to be younger and parous compared with white and Asian women. The highest rates of diabetes mellitus were noted in the Asian and Hispanic ethnic groups. In contrast, excess maternal weight gain was seen more commonly among white and black women. White women were more likely to undergo labor induction and receive epidural anesthesia in labor, whereas black women had the highest rate of low birth weight infants and low 1/5-minute Apgar scores. The majority of black women delivered at hospital B. White women were as likely to deliver at hospital B as they were to deliver at hospital D. Hispanic women were most likely to deliver at hospitals A and B. Unadjusted cesarean delivery rates were highest among black women and lowest among Hispanic women.
Parameter estimates with standard errors, as well as odds ratios for cesarean delivery with 95% CIs for each of the covariates in the Patient Characteristics Model are presented in Table II. Patient characteristics associated with an increased odds of cesarean delivery included induction of labor, increasing maternal age, low birth weight (!2500 g), macrosomia (O4000 g), excessive weight gain, diabetes mellitus, preeclampsia, and placental abruption. Patient characteristics associated with a decreased odds of cesarean delivery included epidural anesthesia and a history of a previous vaginal delivery. The relative risks for cesarean delivery by ethnic group are shown in Table III. The unadjusted risks of cesarean delivery were relatively similar among the 4 ethnic groups, although there were some notable differences. Specifically, the unadjusted risk of cesarean delivery in Asian women was not significantly different from that of white women, whereas Hispanic women were 16% less likely and black women were 18% more likely to undergo cesarean delivery, as compared with white women. Among each of the 3 ethnic groups (black, Asian, and Hispanic), the Hospital Fixed Effects Model and the Physician Fixed Effects Models yielded comparable
1326
Chung et al
Table II Coefficient estimates with standard errors, odds ratios for cesarean delivery with 95% CI, and P-values for the patient characteristics model Patient characteristics model* Covariate
Coefficient estimate
Standard error
Odds ratio
95% CI
P value
0.20 0.62 0.90 1.24 0.63 0.066 0.22 1.93
0.08 0.07 0.07 0.08 0.06 0.06 0.05 0.04
1.22 1.85 2.46 3.45 1.89 1.07 1.25 0.15
1.05-1.41 1.61-2.14 2.13-2.85 2.96-4.02 1.67-2.12 0.94-1.21 1.14-1.36 0.13-0.16
.008 !.001 !.001 !.001 !.001 .30 !.001 !.001
0.53 0.10 0.18 0.89 0.87 1.30 0.30 0.45 0.28
0.27 0.24 0.24 0.08 0.11 0.22 0.04 0.04 0.04
1.70 1.11 1.20 2.43 2.40 3.67 1.35 1.58 0.76
1.00-2.88 0.69-1.76 0.76-1.91 2.10-2.82 1.92-3.00 2.37-5.69 1.24-1.47 1.45-1.71 0.70-0.82
.05 .67 .43 !.001 !.001 !.001 !.001 !.001 !.001
0.95 0.44 0.48 0.12 2.60
0.05 0.10 0.16 0.24 0.25
2.59 1.55 0.62 0.89 d
2.36-2.84 1.28-1.87 0.45-0.84 0.56-1.42 d
!.001 !.001 .002 .63 !.001
y
Maternal age 20-25 y 25-30 y 35-40 y O40 y Blackz Asianz Hispanicz Parity R 1 Gestational agex 28-32 wks 32-26 wks O36 wks Diabetes mellitus Preeclampsia Placental Abruption Excess weight gain Induction Epidural anesthesia Birth weight{ !2500 g O4000 g Induction*preeclampsia Diabetes*preeclampsia Constant
* Estimates are derived from a multinomial logit regression model. y Reference group: !20 y old. z Reference group: white. x Reference group !28 wks’ gestation. { Reference group: Birth weight between 2500 and 4000 g.
estimates for the ethnic-dependent risks of cesarean delivery, when compared with the white group. With respect to the Hospital Fixed Effects and Physician Random Effects Models, the largest difference between the ordinary logistic results and the fixed- and randomeffects models was noted in the black women, where the simultaneous adjustment for both hospital-fixed effects and physician-random effects reduced the estimated relative risk of cesarean delivery with respect to white women from 1.75 in the ordinary logistic model (Patient Characteristics Model) to 1.54 (Hospital Fixed Effects and Physician Random Effects Model). However, a significant effect of black ethnicity on the risk of cesarean delivery was still noted in this model. Among Asian and Hispanic women, the simultaneous adjustment for hospital-fixed effects and physician-random effects did not yield estimates that were notably different from the ordinary logistic results, leading to only a 5% reduction in the effect of Asian ethnicity on the risk of cesarean delivery and a 10% reduction in the effect of Hispanic ethnicity on the risk of cesarean delivery compared with white women.
Comment Racial differences in health care use have been reported in many areas of medicine, which persist even after adjusting for patient-level characteristics such as disease severity, insurance status, and ability to pay.13 This has prompted researchers to investigate the role of physician- and hospital-level characteristics in accounting for these observed racial differences.14,15 With respect to the propensity to perform a cesarean delivery, it is well known that patient characteristics have a strong influence. For example, induction of labor increases the risk of cesarean delivery. However, potential bias of the managing physician and the general trend of practice at the hospital site of delivery may also be contributing factors. Therefore, we performed this study to evaluate the risk of cesarean delivery with respect to ethnicity and implemented fixed and random effects regression modeling to address potential physician- and hospitallevel clustering. The current literature, with respect to the risk of cesarean delivery by ethnic group, is varied.1,9-11 Some
Chung et al Table III
1327 Relative risks and 95% CI of cesarean delivery by ethnic group with white as the reference group Relative risks (95% CI*) for cesarean section
Unadjusted model Patient characteristics modely Hospital-fixed effects modelz Physician-fixed effects modelx Hospital-fixed effects C physician-random effects model{
White (N = 18,554)
Black (N = 3,710)
Asian (N = 3,345)
Hispanic (N = 12,055)
1.0 1.0 1.0 1.0 1.0
1.18 1.75 1.65 1.63 1.54
1.02 1.06 1.02 1.02 1.01
0.84 1.22 1.16 1.13 1.12
(1.06-1.29) (1.57-1.92) (1.47-1.83) (1.45-1.84) (1.41-1.69)
(0.92-1.13) (0.95-1.12) (0.91-1.12) (0.90-1.13) (0.90-1.10)
(0.79-0.91) (1.12-1.33) (1.06-1.27) (1.03-1.26) (1.02-1.24)
* Reported 95% CIs are bootstrapped and bias corrected. Bootstrapped confidence intervals were estimated with 1000 repetitions, unless otherwise specified. y Estimates are derived from a multinomial logit regression model adjusted for induction of labor, maternal age, birth weight, excess maternal weight gain (O30 lbs), epidural anesthesia, diabetes mellitus, parity R 1, preeclampsia, abruption, gestational age, induction*preeclampsia, diabetes*preeclampsia. z Estimates are derived from the multinomial logit regression with hospital-level clustering fixed-effect corrections. x Estimates are derived from the multinomial logit regression with physician-level clustering fixed-effect corrections. { Estimates are derived from the multinomial logit regression with hospital-level clustering fixed-effect corrections and physician-level random-effect corrections.
have suggested that there is an increased risk of cesarean delivery among white women compared with women of other ethnic groups, whereas others have suggested the converse. This inconsistency may be related to the fact that these studies have incompletely adjusted for 1 or more of the following: clinical risks of cesarean delivery, socioeconomic factors, and/or the contribution of physician choice and practice patterns of the delivery physician and hospital. Our results support the previously made observation that risk adjustment is essential when evaluating racial variation in cesarean delivery rates.9 Furthermore, our study suggests that racial differences in the risk of cesarean delivery persist even after adjusting for both clinical risk and physician and hospital factors, although the cause of these residual differences remains unknown. Such remaining differences may be due to differential practices of hospitals or physicians toward different ethnic groups, or to other, unobserved patientlevel factors such as clinical differences or patient preferences. Therefore, further study is necessary to elucidate the reasons behind these observed racial differences. One potential limitation of this study is the absence of information regarding measures of socioeconomic status and insurance status at the individual level. Such differences have been associated with adverse health outcomes and have long been blamed for the differences in health care outcomes seen among varying ethnic groups. Although our study lacked socioeconomic data, differences in health care outcomes related to socioeconomic status generally refers to differences in risk of disease, rather than the propensity to perform a medical procedure. Therefore, risk adjustment that uses clinical characteristics associated with cesarean delivery should account for differences related to socioeconomic status.16 Interestingly, previous authors have suggested that patients of high socioeconomic are more likely to desire cesarean delivery.17 This may be due to the
possibility that such populations may be less accepting of the recommendations of their health care providers compared with patients of lower socioeconomic status or of clinic populations.17 Therefore, had we been able to fully adjust for socioeconomic status our analysis, an even greater differential in the risk of cesarean delivery between white and non-white subgroups might have been identified. Although we did not have insurance information at the individual level, we do know that publicly funded patients were more likely to deliver at certain hospitals. Therefore, such differences in the demographic makeup of one hospital versus another should be accounted for by the fixed effect modeling technique used in our study. One drawback of this modeling technique is that the exact impact of insurance status cannot be determined. However, obtaining a parameter estimate and odds ratio for cesarean delivery as a function of insurance status was not the primary goal of our study. The major strength of this study includes the implementation of fixed and random effects regression modeling to account for the clustering of patients within physician providers and delivering hospitals. Unmeasured factors at these higher-order levels can influence the outcome and be correlated with individual-level covariates, leading to biased estimates of the effect of individual-level factors, such as race. The addition of fixed and random effects to the ordinary logistic model (Patient Characteristics Model) resulted in a diminution of the effect of ethnicity on the risk of cesarean delivery, suggesting that although racial differences in cesarean delivery appear to exist, they may have been overestimated by previous studies in which physician- and hospital-level factors were not assessed or accounted for. This observation can be further generalized to any study in which patient-level data are used. Such studies should be closely evaluated for the possibility of higher-level
1328 clustering, and techniques such as fixed or random effects regression modeling or multilevel modeling should be implemented to account for its effect. In conclusion, our study suggests that intrinsic racial differences in the risk of cesarean delivery appear to exist, even after adjusting for physician- and hospital-level factors. Although differing physician and hospital practices do not appear to fully account for the differences seen in the risk of cesarean delivery among various ethnic groups, some effect is noted. In addition to determining the reasons behind the observed racial differences in cesarean delivery rates, future studies might also include further investigation and identification of particular physician and hospital characteristics that may predispose to a higher risk of performing a cesarean delivery.
Chung et al
6.
7. 8.
9.
10. 11.
12.
References 1. Bravemen P, Egerter S, Edmonston F, Verdon M. Racial/ethnic differences in the likelihood of cesarean delivery, California. Am J Obstet Gynecol 1995;85:625-30. 2. Scheller JM, Nelson KB. Does cesarean delivery prevent cerebral palsy or other neurologic problems of childhood? Obstet Gynecol 1994;83:624-30. 3. Shearer EL. Cesarean section: medical benefits and costs. Soc Sci Med 1993;37:1223-37. 4. Healthy People 2010. Maternal, infant, and child health indicator 16-09: reduce cesarean births among low-risk (full-term, singleton, vertex presentation) women. Available at: http:www.healthypeople. gov/document/html/objectives/16-09.htm. Accessed on January 12, 2006. 5. Centers for Disease Control and Prevention Health Resources and Service Administration. Healthy People 2010: objectives for improving health. Maternal, Infant, and Child Indicators [monograph on the Internet]. Rockville, MD: US Department of Health and Human Services. 2000 [cited 2005 May 17]. Available at:
13.
14.
15.
16.
17.
http://www.healthypeople.gov/Document/pdf/Volume2/16MICH. pdf. Accessed on January 12, 2006. Menacker F, Curtin SC. Trends in cesarean birth and vaginal birth after previous cesarean section 1991-99. Natl Vital Stat Rep 2001; 49:1-15. Vaginal birth after previous cesarean delivery. Washington, DC: The College; 1999. ACOG Practice Bulletin No. 5. Lydon-Rochelle M, Holt VL, Easterling TR, Martin DP. Risk of uterine rupture during labor among women with a prior cesarean delivery. N Eng J Med 2001;345:3-8. Aron DC, Howard GS, DiGiuseppe DL, Harper DL, Rosenthal GE. Variations in risk-adjusted cesarean delivery rates according to race and health insurance. Med Care 2000;38:35-44. Gould JB, Davey B, Stafford RS. Socioeconomic differences in rates of cesarean section. N Engl J Med 1999;321:233-9. Butcher AH, Fos PJ, Zuniga M, Pane G. Racial variations in cesarean section rates: an analysis of Medicaid data in Louisiana. J Public Health Manag Pract 1997;3:41-8. Chung JH, Boscardin WJ, Garite TJ, Lagrew DC, Porto M. Ethnic differences in birth weight by gestational age: at least a partial explanation for the Hispanic epidemiologic paradox? Am J Obstet Gynecol 2003;189:1058-62. Ford ES, Cooper RS. Racial/ethnic differences in health care utilization of cardiovascular procedures: a review of the evidence. Health Serv Res 1995;30:237-52. Harris DR, Andrews R, Elixhauser A. Racial and gender differences in use of procedures for black and white hospitalized adults. Ethn Dis 1997;7:91-105. Schulman KA, Berlin JA, Harless W, Kerner JF, Sistrunk S, Gersh BJ, et al. The effect of race and sex on physicians’ recommendations for cardiac catheterization. N Engl J Med 1999;340: 618-26. Jonas O, Roder D, Chan A. The association of maternal and socioeconomic characteristics in metropolitan Adelaide with medical, obstetric and labor complications and pregnancy outcomes. Aust N Z J Obstet Gynaecol 1992;32:1-5. Dunsmoor-Su R, Sammel M, Stevens E, Peipert JL, Macones G. Impact of sociodemographic and hospital factors on attempts at vaginal birth after cesarean delivery. Obstet Gynecol 2003;102: 1358-65.
www.ajog.org
Intrinsic racial differences in the risk of cesarean delivery are not explained by differences in caregivers or hospital site of delivery Judith H. Chung, MD,a,b,* Thomas J. Garite, MD,b Adele M. Kirk, MHA,c Amie L. Hollard, MD,b,d Deborah A. Wing, MD,b David C. Lagrew, MDb,e Department of Obstetrics and Gynecology, Women’s Hospital, Long Beach Memorial Medical Center, Long Beach, CAa; Division of Maternal-Fetal Medicine, University of California, Irvine, Orange, CAb; Department of Health Services, University of California, Los Angeles, School of Public Health, Los Angeles, CAc; Department of MaternalFetal Medicine, Saint Elizabeth Regional Medical Center, Lincoln, NE d; Saddleback Memorial Medical Center, Laguna Hills, CAe Received for publication June 22, 2005; revised November 3, 2005; accepted November 28, 2005
KEY WORDS Cesarean delivery Race Ethnicity
Objective: This study was undertaken to assess whether racial differences in the risk of cesarean delivery result from differing practices of their caregivers or the hospitals at which they deliver. Study design: A retrospective cohort study was performed using the Perinatal Database of the Memorial Health Care System. Logistic regression was used to estimate the risk of primary cesarean delivery among patients eligible for labor. The contribution of hospital and physician level cluster correlation was evaluated using fixed and random effects regression models. Results: Compared with white patients, black and Hispanic patients were 75% and 22% more likely to undergo primary cesarean delivery. Further adjustment for hospital and physician level cluster correlation resulted in persistently increased risks of primary cesarean delivery in black (54%) and Hispanic patients (12%). Conclusion: Hospital site of delivery and individual physician practices do not fully explain racial differences in the risk of primary cesarean delivery. Ó 2006 Mosby, Inc. All rights reserved.
The rate of cesarean delivery in the United States has increased 5-fold compared with the mid 1960s, when the rate was a mere 4.5% of all births.1 This has resulted in rising health care costs and morbidity for obstetric patients, with an unproven benefit with respect to birth * Reprint requests: Judith H. Chung, MD, Long Beach Memorial Medical Center Women’s Hospital, 2801 Atlantic Ave, Long Beach, CA 90806. E-mail: [email protected] 0002-9378/$ - see front matter Ó 2006 Mosby, Inc. All rights reserved. doi:10.1016/j.ajog.2005.11.043
outcomes.2,3 As such, there has been an increased emphasis placed on attempts to decrease the cesarean delivery rate.4 In fact, reducing the number of cesarean births is 1 of the Maternal, Infant, and Child Objectives of Healthy People 2010.5 In the 1990s, vaginal birth after previous cesarean delivery was identified as 1 method of reducing the number of cesarean deliveries performed.6,7 However, recent concerns regarding the safety of vaginal birth after cesarean delivery have resulted in a decline in the
1324 number of such patients undergoing a trial of labor.8 Therefore, there has been an increased emphasis on the prevention of primary cesarean delivery as a better means of decreasing the overall cesarean delivery rate in this country. As such, a better understanding of the risk factors for primary cesarean delivery is needed. Previous studies have shown a difference in the rate of cesarean delivery among various race/ethnic groups.1,9-11 However, these studies may be biased, as the potential effects of delivering physician and hospital site of delivery were not addressed. Therefore, this study attempts to evaluate ethnic differences in the risk of cesarean delivery, after adjusting for these potential omitted variables using fixed and random effects regression modeling.
Methods With consent from the Institutional Review Board at Long Beach Memorial Medical Center, the Memorial Care Perinatal Database, from January 1, 1997, to June 30, 2002, was used for this study. As previously described, this is a database that contains antenatal and birth information on all pregnant women delivering within the Memorial Health Care System, a 4-hospital network that serves a diverse patient population in Southern California.12 Two of the 4 hospitals (hospitals B and D) provide both 24-hour perinatal consultative services and regional neonatal intensive care services. One of the 4 hospitals (hospital B) has a resident service, supervised by the division of perinatalogy. All 4 hospitals have 24-hour anesthesia availability. Rates of public insurance coverage vary from 5% in hospital D to 55% in hospital A. Of 54,144 observations, multiple gestations, patients with mixed, other, or unknown ethnicity, and patients with a previous cesarean delivery were excluded from the study, leaving 39,522 observations. Of these patients, only those between 24 and 42 weeks’ gestation and who were eligible for labor were included in the study (37,688 observations). Ethnicity was determined by the patient. This was recorded during the course of her prenatal care and confirmed on admission to the hospital for delivery. Intercooled Stata version 8.0 (Stata Corp, College Station, TX) was used for data management and statistical analysis. Demographic data were presented as a function of ethnic group (black, Asian, Hispanic, and white). The c2 test was used for bivariate analysis of categorical data and analysis of variance (ANOVA) for continuous data. Logit regression was then performed with cesarean delivery as the dependent variable and ethnicity as the primary independent variable. The adjusted logistic regression was designated as the ‘‘Patient Characteristics Model.’’ Covariates in this model were identified a priori and included induction of labor,
Chung et al maternal age, birth weight, excess maternal weight gain, defined as a weight gain of greater than 30 lbs, epidural anesthesia, diabetes mellitus, parity, preeclampsia, placental abruption, gestational age, and the following interactionsdinduction*preeclampsia and diabetes* preeclampsia. The clustering of patients within physician providers and delivering hospitals was addressed by using fixed and random effects regression modeling. Such regression modeling is imperative for the correct calculation of standard errors, which can be underestimated if clustering is unaccounted. In addition, the use of fixed and random effects regression modeling accounts for unmeasured factors at the physician and hospital level. These unmeasured factors can influence the outcome and be correlated with patient-level covariates, leading to biased estimates of the effect of patient-level factors, such as race. Therefore, hospital-level clustering was addressed by using dummy variable logit regression with fixed effects modeling, because the unobservable effects of hospital site of delivery were assumed to be correlated with other covariates in the model. This model was designated as the ‘‘Hospital Fixed Effects Model.’’ In a similar fashion, physician-level clustering was addressed with the use of fixed effects. This model was designated as the ‘‘Physician Fixed Effects Model.’’ Finally, physician-random effects and hospital-fixed effects were simultaneously modeled by using the Stata ‘‘xtlogit’’ command with the random effects option, where hospital-fixed effects were included as dummy variables in the main regression equation. Physician-fixed effects were not included in this model, as physicians delivered primarily at 1 hospital. Thus, any physician-fixed effects should be subsumed within the hospital-fixed effects. Hospital site of delivery information was available for all patients in the study. Delivering physician information was available for 37,644 of the 37,688 observations in the study. The 44 patients who did not have a designated delivery physician were eliminated from the regression model. There were 250 different delivering physicians in the data set. During the study period, physicians delivered between 1 and 2,033 patients, with an average of 151 patients per physician. The majority of physicians delivered patients belonging to each of the 4 ethnic groups in the study, and for the most part physicians performed deliveries at only 1 of the 4 hospitals in the study. Of the 250 physicians, 71 physicians performed either all cesarean deliveries or all vaginal deliveries, accounting for 2,137 observations. Because these physicians had no variability in the outcome measure, they were dropped from the Physician Fixed Effect Model. Predicted probabilities for cesarean delivery in all models were obtained and used to calculate the relative risks of cesarean delivery, with white patients as the reference group. Biased-corrected, bootstrapped 95%
Chung et al Table I
1325
Demographic data by ethnic group*
Maternal age (y) Parity R 1 Gestational age (wks) Hospital site of delivery Hospital A Hospital B Hospital C Hospital D Total Diabetes mellitus Preeclampsia Placental abruption Excess weight gain (O30 lbs) Induction of labor Epidural in labor Overall cesarean delivery rate Cesarean delivery rate in patients eligible for labor Birth weight (g) Macrosomia (O4000 g) Low birth weight (!2500 g) 1-min Apgar !5 5-min Apgar !7
White
Black
30.2 G 5.6 (18,569) 53.5% (9,936/18,569) 39.4 G 1.7 (18,569)
25.8 G 6.1 (3,712) 30.0 G 5.6 (3,348) 26.4 G 6.0 (12,059) !.0001 59.8% (2,219/3,712) 51.4% (1,721/3,348) 63.3% (7,628/12,059) !.001 39.0 G 2.3 (3,712) 39.1 G 1.8 (3,348) 39.2 G 2.0 (12,059) !.0001
Asian
10.3% (1,916) 39.7% (7,374) 5.2% (966) 44.8% (8,313) (18,569) 2.7% (500/18,569) 2.6% (485/18,569) 0.2% (45/18,569) 21.4% (3,964/18,569)
3.7% (135) 91.5% (3,398) 1.1% (40) 3.7% (139) (3,712) 3.3% (121/3,712) 3.5% (128/3,712) 0.6% (22/3,712) 21.4% (795/3,712)
15.3% (512) 59.3% (1,988) 6.5% (217) 18.9% (631) (3,348) 4.9% (165/3,348) 2.3% (77/3,348) 0.4% (13/3,348) 16.5% (551/3,348)
Hispanic
29.5% (3,562) 53.9% (6,505) 6.8% (807) 9.8% (1,185) (12,059) 5.1% (615/12,059) 3.2% (391/12,059) 0.3% (39/12,059) 13.4% (1,615/12,059)
P value
!.001 !.001 !.001 .005 !.001
27.3% (5,038/18,569) 22.9% (850/3,712) 16.6% (555/3,348) 18.9% (2,278/12,059) !.001 76.3% (14,136/18,569) 54.9% (2,037/3,712) 56.9% (1,904/3,348) 41.5% (4,999/12,059) !.001 15.6% (3,053/19,564) 17.2% (670/3,898) 15.1% (528/3,496) 13.2% (1,658/12,594) !.001 11.1% (2,058/18,569)
13.0% (484/3,712)
11.4% (380/3,348)
3490 G 540 (18,569) 14.6% (2,717/18,569) 3.5% (648/18,569)
3233 G 601 (3,712) 3267 G 513 (3,348) 3391 G 581 (12,058) !.0001 7.1% (264/3,712) 6.4% (215/3,348) 11.4% (1,369/12,059) !.001 8.6% (320/3,712) 5.2% (174/3,348) 5.2% (622/12,059) !.001
2.2% (413/18,569) 0.9% (158/18,569)
3.1% (116/3,712) 1.6% (58/3,712)
1.5% (49/3,348) 0.6% (21/3,348)
9.3% (1,123/12,059)
2.2% (269/12,059) 1.0% (125/12,059)
!.001
!.001 !.001
* Analyzed using the c2 test for categorical data and ANOVA for continuous data.
CIs were determined for each predicted probability and relative risk. For each bootstrapped confidence interval, 1000 repetitions were performed, as recommended in the literature.
Results Demographic data by ethnic group are presented in Table I. Overall, there were 37,688 patients that met inclusion criteria. Black and Hispanic women were more likely to be younger and parous compared with white and Asian women. The highest rates of diabetes mellitus were noted in the Asian and Hispanic ethnic groups. In contrast, excess maternal weight gain was seen more commonly among white and black women. White women were more likely to undergo labor induction and receive epidural anesthesia in labor, whereas black women had the highest rate of low birth weight infants and low 1/5-minute Apgar scores. The majority of black women delivered at hospital B. White women were as likely to deliver at hospital B as they were to deliver at hospital D. Hispanic women were most likely to deliver at hospitals A and B. Unadjusted cesarean delivery rates were highest among black women and lowest among Hispanic women.
Parameter estimates with standard errors, as well as odds ratios for cesarean delivery with 95% CIs for each of the covariates in the Patient Characteristics Model are presented in Table II. Patient characteristics associated with an increased odds of cesarean delivery included induction of labor, increasing maternal age, low birth weight (!2500 g), macrosomia (O4000 g), excessive weight gain, diabetes mellitus, preeclampsia, and placental abruption. Patient characteristics associated with a decreased odds of cesarean delivery included epidural anesthesia and a history of a previous vaginal delivery. The relative risks for cesarean delivery by ethnic group are shown in Table III. The unadjusted risks of cesarean delivery were relatively similar among the 4 ethnic groups, although there were some notable differences. Specifically, the unadjusted risk of cesarean delivery in Asian women was not significantly different from that of white women, whereas Hispanic women were 16% less likely and black women were 18% more likely to undergo cesarean delivery, as compared with white women. Among each of the 3 ethnic groups (black, Asian, and Hispanic), the Hospital Fixed Effects Model and the Physician Fixed Effects Models yielded comparable
1326
Chung et al
Table II Coefficient estimates with standard errors, odds ratios for cesarean delivery with 95% CI, and P-values for the patient characteristics model Patient characteristics model* Covariate
Coefficient estimate
Standard error
Odds ratio
95% CI
P value
0.20 0.62 0.90 1.24 0.63 0.066 0.22 1.93
0.08 0.07 0.07 0.08 0.06 0.06 0.05 0.04
1.22 1.85 2.46 3.45 1.89 1.07 1.25 0.15
1.05-1.41 1.61-2.14 2.13-2.85 2.96-4.02 1.67-2.12 0.94-1.21 1.14-1.36 0.13-0.16
.008 !.001 !.001 !.001 !.001 .30 !.001 !.001
0.53 0.10 0.18 0.89 0.87 1.30 0.30 0.45 0.28
0.27 0.24 0.24 0.08 0.11 0.22 0.04 0.04 0.04
1.70 1.11 1.20 2.43 2.40 3.67 1.35 1.58 0.76
1.00-2.88 0.69-1.76 0.76-1.91 2.10-2.82 1.92-3.00 2.37-5.69 1.24-1.47 1.45-1.71 0.70-0.82
.05 .67 .43 !.001 !.001 !.001 !.001 !.001 !.001
0.95 0.44 0.48 0.12 2.60
0.05 0.10 0.16 0.24 0.25
2.59 1.55 0.62 0.89 d
2.36-2.84 1.28-1.87 0.45-0.84 0.56-1.42 d
!.001 !.001 .002 .63 !.001
y
Maternal age 20-25 y 25-30 y 35-40 y O40 y Blackz Asianz Hispanicz Parity R 1 Gestational agex 28-32 wks 32-26 wks O36 wks Diabetes mellitus Preeclampsia Placental Abruption Excess weight gain Induction Epidural anesthesia Birth weight{ !2500 g O4000 g Induction*preeclampsia Diabetes*preeclampsia Constant
* Estimates are derived from a multinomial logit regression model. y Reference group: !20 y old. z Reference group: white. x Reference group !28 wks’ gestation. { Reference group: Birth weight between 2500 and 4000 g.
estimates for the ethnic-dependent risks of cesarean delivery, when compared with the white group. With respect to the Hospital Fixed Effects and Physician Random Effects Models, the largest difference between the ordinary logistic results and the fixed- and randomeffects models was noted in the black women, where the simultaneous adjustment for both hospital-fixed effects and physician-random effects reduced the estimated relative risk of cesarean delivery with respect to white women from 1.75 in the ordinary logistic model (Patient Characteristics Model) to 1.54 (Hospital Fixed Effects and Physician Random Effects Model). However, a significant effect of black ethnicity on the risk of cesarean delivery was still noted in this model. Among Asian and Hispanic women, the simultaneous adjustment for hospital-fixed effects and physician-random effects did not yield estimates that were notably different from the ordinary logistic results, leading to only a 5% reduction in the effect of Asian ethnicity on the risk of cesarean delivery and a 10% reduction in the effect of Hispanic ethnicity on the risk of cesarean delivery compared with white women.
Comment Racial differences in health care use have been reported in many areas of medicine, which persist even after adjusting for patient-level characteristics such as disease severity, insurance status, and ability to pay.13 This has prompted researchers to investigate the role of physician- and hospital-level characteristics in accounting for these observed racial differences.14,15 With respect to the propensity to perform a cesarean delivery, it is well known that patient characteristics have a strong influence. For example, induction of labor increases the risk of cesarean delivery. However, potential bias of the managing physician and the general trend of practice at the hospital site of delivery may also be contributing factors. Therefore, we performed this study to evaluate the risk of cesarean delivery with respect to ethnicity and implemented fixed and random effects regression modeling to address potential physician- and hospitallevel clustering. The current literature, with respect to the risk of cesarean delivery by ethnic group, is varied.1,9-11 Some
Chung et al Table III
1327 Relative risks and 95% CI of cesarean delivery by ethnic group with white as the reference group Relative risks (95% CI*) for cesarean section
Unadjusted model Patient characteristics modely Hospital-fixed effects modelz Physician-fixed effects modelx Hospital-fixed effects C physician-random effects model{
White (N = 18,554)
Black (N = 3,710)
Asian (N = 3,345)
Hispanic (N = 12,055)
1.0 1.0 1.0 1.0 1.0
1.18 1.75 1.65 1.63 1.54
1.02 1.06 1.02 1.02 1.01
0.84 1.22 1.16 1.13 1.12
(1.06-1.29) (1.57-1.92) (1.47-1.83) (1.45-1.84) (1.41-1.69)
(0.92-1.13) (0.95-1.12) (0.91-1.12) (0.90-1.13) (0.90-1.10)
(0.79-0.91) (1.12-1.33) (1.06-1.27) (1.03-1.26) (1.02-1.24)
* Reported 95% CIs are bootstrapped and bias corrected. Bootstrapped confidence intervals were estimated with 1000 repetitions, unless otherwise specified. y Estimates are derived from a multinomial logit regression model adjusted for induction of labor, maternal age, birth weight, excess maternal weight gain (O30 lbs), epidural anesthesia, diabetes mellitus, parity R 1, preeclampsia, abruption, gestational age, induction*preeclampsia, diabetes*preeclampsia. z Estimates are derived from the multinomial logit regression with hospital-level clustering fixed-effect corrections. x Estimates are derived from the multinomial logit regression with physician-level clustering fixed-effect corrections. { Estimates are derived from the multinomial logit regression with hospital-level clustering fixed-effect corrections and physician-level random-effect corrections.
have suggested that there is an increased risk of cesarean delivery among white women compared with women of other ethnic groups, whereas others have suggested the converse. This inconsistency may be related to the fact that these studies have incompletely adjusted for 1 or more of the following: clinical risks of cesarean delivery, socioeconomic factors, and/or the contribution of physician choice and practice patterns of the delivery physician and hospital. Our results support the previously made observation that risk adjustment is essential when evaluating racial variation in cesarean delivery rates.9 Furthermore, our study suggests that racial differences in the risk of cesarean delivery persist even after adjusting for both clinical risk and physician and hospital factors, although the cause of these residual differences remains unknown. Such remaining differences may be due to differential practices of hospitals or physicians toward different ethnic groups, or to other, unobserved patientlevel factors such as clinical differences or patient preferences. Therefore, further study is necessary to elucidate the reasons behind these observed racial differences. One potential limitation of this study is the absence of information regarding measures of socioeconomic status and insurance status at the individual level. Such differences have been associated with adverse health outcomes and have long been blamed for the differences in health care outcomes seen among varying ethnic groups. Although our study lacked socioeconomic data, differences in health care outcomes related to socioeconomic status generally refers to differences in risk of disease, rather than the propensity to perform a medical procedure. Therefore, risk adjustment that uses clinical characteristics associated with cesarean delivery should account for differences related to socioeconomic status.16 Interestingly, previous authors have suggested that patients of high socioeconomic are more likely to desire cesarean delivery.17 This may be due to the
possibility that such populations may be less accepting of the recommendations of their health care providers compared with patients of lower socioeconomic status or of clinic populations.17 Therefore, had we been able to fully adjust for socioeconomic status our analysis, an even greater differential in the risk of cesarean delivery between white and non-white subgroups might have been identified. Although we did not have insurance information at the individual level, we do know that publicly funded patients were more likely to deliver at certain hospitals. Therefore, such differences in the demographic makeup of one hospital versus another should be accounted for by the fixed effect modeling technique used in our study. One drawback of this modeling technique is that the exact impact of insurance status cannot be determined. However, obtaining a parameter estimate and odds ratio for cesarean delivery as a function of insurance status was not the primary goal of our study. The major strength of this study includes the implementation of fixed and random effects regression modeling to account for the clustering of patients within physician providers and delivering hospitals. Unmeasured factors at these higher-order levels can influence the outcome and be correlated with individual-level covariates, leading to biased estimates of the effect of individual-level factors, such as race. The addition of fixed and random effects to the ordinary logistic model (Patient Characteristics Model) resulted in a diminution of the effect of ethnicity on the risk of cesarean delivery, suggesting that although racial differences in cesarean delivery appear to exist, they may have been overestimated by previous studies in which physician- and hospital-level factors were not assessed or accounted for. This observation can be further generalized to any study in which patient-level data are used. Such studies should be closely evaluated for the possibility of higher-level
1328 clustering, and techniques such as fixed or random effects regression modeling or multilevel modeling should be implemented to account for its effect. In conclusion, our study suggests that intrinsic racial differences in the risk of cesarean delivery appear to exist, even after adjusting for physician- and hospital-level factors. Although differing physician and hospital practices do not appear to fully account for the differences seen in the risk of cesarean delivery among various ethnic groups, some effect is noted. In addition to determining the reasons behind the observed racial differences in cesarean delivery rates, future studies might also include further investigation and identification of particular physician and hospital characteristics that may predispose to a higher risk of performing a cesarean delivery.
Chung et al
6.
7. 8.
9.
10. 11.
12.
References 1. Bravemen P, Egerter S, Edmonston F, Verdon M. Racial/ethnic differences in the likelihood of cesarean delivery, California. Am J Obstet Gynecol 1995;85:625-30. 2. Scheller JM, Nelson KB. Does cesarean delivery prevent cerebral palsy or other neurologic problems of childhood? Obstet Gynecol 1994;83:624-30. 3. Shearer EL. Cesarean section: medical benefits and costs. Soc Sci Med 1993;37:1223-37. 4. Healthy People 2010. Maternal, infant, and child health indicator 16-09: reduce cesarean births among low-risk (full-term, singleton, vertex presentation) women. Available at: http:www.healthypeople. gov/document/html/objectives/16-09.htm. Accessed on January 12, 2006. 5. Centers for Disease Control and Prevention Health Resources and Service Administration. Healthy People 2010: objectives for improving health. Maternal, Infant, and Child Indicators [monograph on the Internet]. Rockville, MD: US Department of Health and Human Services. 2000 [cited 2005 May 17]. Available at:
13.
14.
15.
16.
17.
http://www.healthypeople.gov/Document/pdf/Volume2/16MICH. pdf. Accessed on January 12, 2006. Menacker F, Curtin SC. Trends in cesarean birth and vaginal birth after previous cesarean section 1991-99. Natl Vital Stat Rep 2001; 49:1-15. Vaginal birth after previous cesarean delivery. Washington, DC: The College; 1999. ACOG Practice Bulletin No. 5. Lydon-Rochelle M, Holt VL, Easterling TR, Martin DP. Risk of uterine rupture during labor among women with a prior cesarean delivery. N Eng J Med 2001;345:3-8. Aron DC, Howard GS, DiGiuseppe DL, Harper DL, Rosenthal GE. Variations in risk-adjusted cesarean delivery rates according to race and health insurance. Med Care 2000;38:35-44. Gould JB, Davey B, Stafford RS. Socioeconomic differences in rates of cesarean section. N Engl J Med 1999;321:233-9. Butcher AH, Fos PJ, Zuniga M, Pane G. Racial variations in cesarean section rates: an analysis of Medicaid data in Louisiana. J Public Health Manag Pract 1997;3:41-8. Chung JH, Boscardin WJ, Garite TJ, Lagrew DC, Porto M. Ethnic differences in birth weight by gestational age: at least a partial explanation for the Hispanic epidemiologic paradox? Am J Obstet Gynecol 2003;189:1058-62. Ford ES, Cooper RS. Racial/ethnic differences in health care utilization of cardiovascular procedures: a review of the evidence. Health Serv Res 1995;30:237-52. Harris DR, Andrews R, Elixhauser A. Racial and gender differences in use of procedures for black and white hospitalized adults. Ethn Dis 1997;7:91-105. Schulman KA, Berlin JA, Harless W, Kerner JF, Sistrunk S, Gersh BJ, et al. The effect of race and sex on physicians’ recommendations for cardiac catheterization. N Engl J Med 1999;340: 618-26. Jonas O, Roder D, Chan A. The association of maternal and socioeconomic characteristics in metropolitan Adelaide with medical, obstetric and labor complications and pregnancy outcomes. Aust N Z J Obstet Gynaecol 1992;32:1-5. Dunsmoor-Su R, Sammel M, Stevens E, Peipert JL, Macones G. Impact of sociodemographic and hospital factors on attempts at vaginal birth after cesarean delivery. Obstet Gynecol 2003;102: 1358-65.