Janakiraman, Vanitha MD, MPH; Lazar, Jane RN, MPH; Joynt, Karen E. MD, MPH; Jha, Ashish K. MD, MPH
More than 4 million women give birth in U.S. hospitals each year, and although outcomes for the majority are quite good, many women suffer complications including infections, hemorrhage and operative complications, and occasionally, life-threatening events.1 Previous studies have found large variations in obstetric outcomes across hospitals and providers, suggesting ample opportunity for improvement.2–4
To improve care, we need to better understand why variations exist: why certain hospitals and providers, caring for a similar patient population, have much lower complication rates than others. One possibility is that experience, as measured by hospital or provider volume, might be related to outcomes. Indeed, there is evidence that hospitals and surgeons who provide a higher volume of surgical procedures generally have better outcomes than those who provide a lower volume of services.5,6 In obstetrics, techniques of labor management can affect outcomes, and they vary widely between institutions and providers.7,8 However, the role of provider volume and hospital volume in determining differences in labor management or other practice differences that may lead to variation in obstetric outcomes is largely unknown.
Understanding whether volume affects obstetric outcomes has important implications. If greater volume is closely related to better outcomes, policymakers might choose to target the lowest-volume providers, who perform a smaller number of deliveries but may provide essential access to obstetric services, through further training, incentives, or credentialing requirements. Therefore, we sought to understand the range of hospital and provider obstetric volume in the United States, and to estimate the relationship across hospital volume, provider volume, and obstetric complication rate in U.S. hospitals.
MATERIALS AND METHODS
We used data from the 2007 Nationwide Inpatient Sample of the Healthcare Cost and Utilization Project, a federal-state-industry partnership sponsored by the Agency for Healthcare Research and Quality.9 All community hospitals from the participating states are stratified by rural and urban location, number of beds, region of the country, teaching status, and ownership; within each stratum, a systematic random 20% sample of hospitals (total approximately 1,000 hospitals) is drawn. The Nationwide Inpatient Sample includes all discharges from the sampled hospitals and can be used to produce nationwide estimates. Further details of the sampling strategy are described elsewhere.9
We linked Nationwide Inpatient Sample data to the American Hospital Association Annual Survey 2007 to obtain detailed hospital information.10 From the full 2007 dataset of 8.2 million admissions, we used the enhanced method11 to identify all 927,850 obstetric admissions. In this method, additional delivery-related International Classification of Diseases, 9th Revision and procedure codes are used to identify obstetric admissions, rather than the simpler method of including those with a V27 (maternal outcome of delivery) code. We defined an obstetric hospital as any hospital that performed 10 or more deliveries a year and self-identified as a hospital providing obstetric services in the American Hospital Association 2007 survey. This excluded 10 hospitals performing a total of 18 deliveries. Of the 40 states in the Nationwide Inpatient Sample dataset, 14 states did not report provider information, and an additional 10 states did not link to American Hospital Association survey information. After excluding deliveries from these states, 385,642 deliveries remained. Of these, 5,538 records (1.4%) were excluded because the provider identifiers were blank or there were multiple providers identified. The remaining 380,104 deliveries were included in the final analysis. These deliveries were weighted using Nationwide Inpatient Sample strata. Because each region was represented and weighted accordingly, the weights ensure that our results are nationally representative.9
Hospital obstetric volume was defined as the total number of deliveries occurring at that hospital in 2007; hospitals were divided into equal quartiles based on volume. Additional hospital characteristics such as region, teaching status, profit status, location (rural or urban), and presence of a neonatal intensive care unit (NICU) were obtained from the American Hospital Association 2007 survey.
We used the Healthcare Cost and Utilization Project encrypted provider identifiers in each record to assign deliveries to a provider. Provider obstetric volume was defined as the number of deliveries assigned to each provider in 2007; we divided providers into equal quartiles based on volume.
We selected patient characteristics primarily based on a previously well-described approach,12 further refined based on a review of earlier literature and clinical judgment. For each patient, we identified the presence or absence of the following risk factors: maternal age greater than 35 years, weekend admission, and 34 maternal comorbidities including previous cesarean delivery, fetal malpresentation, severe hypertension, multiple gestation, antepartum bleeding, herpes, macrosomia, unengaged head, maternal soft tissue disorder, preterm labor, congenital anomalies, oligohydramnios, and hydramnios. Table 1 in the Appendix, available online at http://links.lww.com/AOG/A255, lists specific International Classification of Diseases, 9th Revision codes used. Patients with one or more of these risk factors were defined as “high risk”; the remainder as “low risk.”
The principal study outcome was a binary indicator for delivery complication using a previously described approach.12 The outcome was measured at the patient level. Diagnosis codes used to identify maternal complications are shown in Table 1 in the Appendix (http://links.lww.com/AOG/A255). Complications were divided into the following categories: lacerations (third- and fourth-degree, cervical, severe vulvar), hemorrhage (immediate postpartum hemorrhage, delayed postpartum hemorrhage, disseminated intravascular coagulation), infections (postpartum sepsis or infection with severe complication with postpartum onset), thromboses (deep venous thrombosis or other vascular complication occurring postpartum), and other (postpartum shock, acute renal failure, acute pulmonary edema, retained foreign body), and for cesarean deliveries, operative complications (intraoperative pelvic organ injury, intestinal or bladder injury, return to operating room).
We summarized the clinical characteristics of the women cared for by each quartile of hospital and provider volume, as well as characteristics of the hospitals and providers in each quartile, and compared these across quartiles using χ2 tests. We also performed a Pearson correlation to describe the overall correlation between hospital volume and provider volume.
We built logistic regression models with quartile of hospital volume as our primary predictor, and maternal complication as our primary outcome, fully adjusting for patient characteristics. We next added hospital factors (proportion of patients covered by Medicaid, hospital region, hospital location, teaching status, and profit status) to our model. Because complication rates differed between vaginal deliveries and cesarean deliveries, we examined models with and without adjustment for a provider's cesarean delivery rate. All analyses were clustered by hospital and weighted using Nationwide Inpatient Sample weighing schema. To do this, we used Proc Surveymeans and Surveylogistic procedures within the SAS package and used a cluster statement to account for correlation between patients in hospitals. We repeated each of these analyses with quartile of provider volume as our primary predictor, again first performing analyses that adjusted for patient characteristics only, and subsequently adding hospital-level factors and mode of delivery to the model.
We carried out additional sensitivity analyses, including an estimation of the relationship of provider volume with outcomes within each stratum of hospital volume. To determine whether there was a plateau effect of provider volume on outcomes, we created risk-adjusted nonparametric curves using the Loess locally weighted scatterplot smoothing method. The Loess method uses locally weighted regression to plot a smooth curve through a set of data points without an a priori assumption of a model to fit all of the data.13 Visual inspection of the Loess curve showed a steep linear decline until a provider volume of 10 deliveries per year, after which a plateau was apparent (Fig. 1 in the Appendix, available online at http://links.lww.com/AOG/A255). Because this curve did not suggest a linear relationship between our predictors (hospital and provider volumes) and outcomes (complication rates), we felt more comfortable in our a priori choice of examining our predictors based on quartiles rather than as a continuous variable. Finally, we examined the characteristics of the hospitals where the lowest-volume providers practice, including their location, average provider volume within their hospitals, and complication rates.
This research using de-identified publicly available data were judged exempt from review by the Harvard School of Public Health Institutional Review Board. All analyses were performed using SAS 9.2.
There were 1,365 hospitals included in the analysis. Hospital volume ranged from 10 to 13,890 deliveries per year, a range consistent with reported obstetric hospital volumes in the United States.14 Quartile 1, the lowest-volume quartile, consisted of hospitals with fewer than 255 deliveries per year; quartile 2, 255–643 deliveries per year; quartile 3, 644–1,699 deliveries per year; and quartile 4, the highest-volume quartile, 1,700 or more deliveries per year.
Hospitals in the highest quartile of obstetric volume were more likely to care for women with medical and obstetric risk factors (Table 1). Low-volume hospitals were more likely to be rural and to be located in the West and Midwest, whereas high-volume hospitals were more likely to be teaching hospitals, to have a NICU, and to be located in the South and Northeast. Cesarean delivery rates were slightly higher at the highest-volume hospitals.
We found that aside from infections, which were more common in larger hospitals, those hospitals in the lowest-volume quartile had marginally higher odds of several individual types of complications, most of which did not reach statistical significance. In our composite outcome, the lowest-volume hospitals had a 10% higher odds of complications than the highest-volume providers, another difference that did not reach statistical significance (OR 1.1, 95% confidence interval [CI] [0.9–1.4], P=.117, Table 2). When we further adjusted for cesarean delivery rate as well as key hospital characteristics, the relationship between hospital volume and complication rates was neither monotonic nor statistically significant (Table 2, last row).
There were 6,963 providers included in the analysis. Provider volume ranged from 1 to 1,731 deliveries per year. Quartile 1, the lowest-volume quartile, consisted of providers with fewer than 7 deliveries per year; quartile 2, 7–31 deliveries per year; quartile 3, 32–89 deliveries per year; and quartile 4, the highest-volume quartile, 90 or more deliveries per year. Low-volume providers were more likely to be located in rural areas. The highest-volume providers cared for a group of women with a slightly higher burden of risk factors, and a higher proportion of Medicaid coverage (Table 3). The correlation between hospital volume and provider volume was 0.35.
We found a consistent relationship between provider volume and risk-adjusted odds of complications. Providers in the lowest volume quartile (those providing seven or fewer deliveries annually) had higher rates of each type of complication, and a 50% higher risk-adjusted odds of complications overall than providers in the highest-volume quartile (lowest quartile compared with highest quartile OR 1.5, 95% CI [1.3–1.7], Table 4). Women in the highest-volume group had, on average, a risk-adjusted complication rate of 12.7% whereas women in the lowest-volume group had, on average, a 17.8% risk-adjusted complication rate (Table 2 in the Appendix, http://links.lww.com/AOG/A255). Further adjusting these analyses for cesarean delivery rate as well as characteristics of the hospital where these providers practiced had only a moderate effect: women who had their delivery by low-volume providers still had a 30% higher odds of developing a complication compared with women who had their delivery by a high-volume provider (lowest quartile compared with highest quartile OR 1.3, 95% CI [1.1–1.5], Table 4). The results were similar when comparing adjusted rates of individual complications (Table 3 in the Appendix, http://links.lww.com/AOG/A255).
In sensitivity analyses, we found that provider volume was associated with lower risk-adjusted rates of complications within each strata of hospital volume (Fig. 1), a finding that extended to individual types of complications (Table 3 in the Appendix, http://links.lww.com/AOG/A255). Our results were similar when restricting our analysis to high-risk women: there was no relationship between hospital volume and complication rate, but high-volume providers had a lower complication rate after adjusting for patient factors and hospital characteristics (Table 4 in the Appendix, http://links.lww.com/AOG/A255). As another sensitivity analysis, we repeated our evaluation of the relationship of provider volume and outcome excluding the 1% of providers with the highest volume (those classified as performing 400 or more deliveries per year), which did not change our findings.
When we examined characteristics of the hospitals where the lowest-volume providers (those with fewer than seven cases) practice, we found that 95% of these hospitals performed at least 100 deliveries per year. After adjusting for provider volume, we found that the hospitals where at least one low-volume provider practices did not have significantly higher risk-adjusted odds of complications (OR 1.1, 95% CI 0.9–1.4).
We found that the lowest-volume obstetric providers, who perform fewer than seven deliveries per year, have 50% higher odds of complications and a 5% higher adjusted rate of complications than providers who perform more than 90 deliveries annually. This translates to a number needed to treat of 20: if there is in fact a causal link between volume and outcomes, there is one extra complication for every 20 women who receive care with a low-volume provider (those delivering seven or fewer neonates a year) compared with a high-volume provider. In contrast, hospital volume seems to have little effect on outcomes, suggesting that practitioner expertise may be a more important predictor of how women fare than the broader expertise in that hospital.
We were surprised to find that hospital volume was not associated with complication rates. This may be because management of normal labor likely varies widely between different practitioners even at the same hospital.15,16 Decisions during labor management, including timing of admission, frequency of examinations, and duration of pushing, all affect the risk of complications.7 It is possible that high-volume providers maintain enough experience in managing labor to reduce the risk of these complications. Understanding the mechanism by which high-volume providers achieve better outcomes is critically important.
We found that a quarter of all providers who deliver children in the United States perform seven or fewer deliveries a year. These providers have the greatest complication rates, and most are at hospitals with higher volumes. In considering policy approaches, it is important to recognize that in this observational study, we cannot establish a causal link between volume and outcome. For example, high-quality providers may draw high referral rates from selected populations, leading to lower measured complication rates. Low-volume providers may not have undergone as much specialty training in obstetrics. Our data do not allow us to identify the background of the providers, beyond the fact that they are all physicians. Regardless of the mechanism, our study shows a wide gap in outcomes between different providers. We believe it is important to better understand why low-volume providers have higher complication rates and what we might do to help them improve. One solution may be that hospitals or specialty boards might ensure that providers either sustain a minimum number of deliveries or participate in simulation exercises.
Previous studies of hospital volume and obstetric outcomes have been consistent with our results on hospital volume.4,17–19 The earlier data on the volume of obstetric providers and outcomes is less consistent: LeFevre and Klein et al both found no difference in adverse outcomes between high- and low-volume obstetric providers.20,21 However, our analyses of national data had a much larger range of provider volumes and obstetric outcomes.
Our study has limitations. We used a large dataset based on administrative data, which limited our ability to capture maternal comorbidities and complications. However, our approach to both risk adjustment and classification of complications employed widely used methods. Some of these complications have been previously validated.22 Our findings were consistent between those complications that have been previously well validated (ie, third- or fourth-degree lacerations) and those that have been less so.
As in any large national database study, systematic differences in coding practices could have influenced our results. It is possible that hospitals and providers differ systematically in their coding of obstetric complications and maternal comorbidities. Because professional coders usually complete the billing forms that feed into our administrative data, one would have to suspect that there are systematic variations in documentation between high- and low-volume providers (with high-volume providers documenting fewer complications). Whereas differential documentation may be present for some types of complications, it is less likely to explain complications like urinary tract injury or fourth-degree laceration (which we suspect are both documented consistently whenever they are known to occur).22 However, because we are unaware of any data that address the issue of differential documentation, we cannot be sure to what extent this contributes to our findings. In addition, we could not adequately assess the severity of complications, because diagnostic codes usually do not indicate severity. Finally, cesarean delivery rate varied across hospitals and cesarean rate doubtlessly has an effect on complication rates. Our approach was to compare complication rates independent of mode of delivery, and our findings did not change meaningfully whether we adjusted for cesarean delivery rate or not.
In summary, we examined the effect of hospital volume and provider volume on obstetric complication rates among women delivering in U.S. hospitals, and found that provider volume was closely associated with complication rates. In the United States, 24,000 women annually are delivered by providers who perform fewer than seven deliveries per year. These providers had significantly higher rates of maternal complications. These findings raise the possibility that maternal outcomes could be improved by selective referral to higher-volume providers, or by ensuring greater training, possibly tied to credentialing, for the lowest-volume providers.
1. Kuklina EV, Meikle SF, Jamieson DJ, Whiteman WK, Barfield WD, Hillis SD, et al.. Severe obstetric morbidity in the United States: 1998–2005. Obstet Gynecol 2009;113(2 pt 1):293–9.
2. Baicker K, Buckles KS, Chandra A. Geographic variation in the appropriate use of cesarean delivery. Health Aff (Millwood) 2006;25:w355–67.
3. Clark SL, Belfort MA, Hankins GD, Meyers JA, Houser FM. Variation in the rates of operative delivery in the United States. Am J Obstet Gynecol 2007;196:526.e1–5.
4. Garcia FA, Miller HB, Huggins GR, Gordon TA. Effect of academic affiliation and obstetric volume on clinical outcome and cost of childbirth. Obstet Gynecol 2001;97:567–76.
5. Birkmeyer JD, Siewers AE, Finlayson EV, Stukel TA, Lucas FL, Batista I, et al.. Hospital volume and surgical mortality in the United States. N Engl J Med 2002;346:1128–37.
6. Dudley RA, Johansen KL, Brand R, Rennie DJ, Milstein A. Selective referral to high-volume hospitals: estimating potentially avoidable deaths. JAMA 2000;283:1159–66.
7. Berghella V, Baxter JK, Chauhan SP. Evidence-based labor and delivery management. Am J Obstet Gynecol 2008;199:445–54.
8. Althabe F, Buekens P, Bergel E, Belizan JM, Campbell MK, Moss N, et al.. A behavioral intervention to improve obstetrical care. N Engl J Med 2008;358:1929–40.
9. Nationwide Inpatient Sample (NIS), Healthcare Cost and Utilization Project (HCUP). Rockville (MD): Agency for Healthcare Research and Quality; 2007.
11. Kuklina EV, Whiteman MK, Hillis SD, Jamieson DJ, Meikle SF, Posner SF, et al.. An enhanced method for identifying obstetric deliveries: implications for estimating maternal morbidity. Matern Child Health J 2008;12:469–77.
12. Asch DA, Nicholson S, Srinivas S, Herrin J, Epstein AJ. Evaluating obstetrical residency programs using patient outcomes. JAMA 2009;302:1277–83.
13. Efron B, Tibshirani R. Statistical data analysis in the computer age. Science 1991;253:390–5.
15. Goyert GL, Bottoms SF, Treadwell MC, Nehra PC. The physician factor in cesarean birth rates. N Engl J Med 1989;320:706–9.
16. Frigoletto FC Jr, Lieberman E, Lang JM, Cohen A, Barss V, Ringer S, et al.. A clinical trial of active management of labor. N Engl J Med 1995;333:745–50.
17. Mayfield JA, Rosenblatt RA, Baldwin LM, Chu J, Logerfo JP. The relation of obstetrical volume and nursery level to perinatal mortality. Am J Public Health 1990;80:819–23.
18. Nicholson WK, Witter F, Powe NR. Effect of hospital setting and volume on clinical outcomes in women with gestational and type 2 diabetes mellitus. J Womens Health (Larchmt) 2009;18:1567–76.
19. Chang JJ, Stamilio DM, Macones GA. Effect of hospital volume on maternal outcomes in women with prior cesarean delivery undergoing trial of labor. Am J Epidemiol 2008;167:711–8.
20. LeFevre ML. Physician volume and obstetric outcome. Med Care 1992;30:866–71.
21. Klein MC, Spence A, Kaczorowski J, Kelly A, Grzybowski S. Does delivery volume of family physicians predict maternal and newborn outcome? CMAJ 2002;166:1257–63.
22. Romano PS, Yasmeen S, Schembri ME, Keyzer JM, Gilbert WM. Coding of perineal lacerations and other complications of obstetric care in hospital discharge data. Obstet Gynecol 2005;106:717–25.
© 2011 by The American College of Obstetricians and Gynecologists.