Romano, Patrick S. MD, MPH; Yasmeen, Shagufta MD, MRCOG; Schembri, Michael E.; Keyzer, Janet M. MPA, RNC; Gilbert, William M. MD
The growing interest in health care quality has stimulated efforts to measure hospital outcomes by using secondary data sources such as birth certificates, hospital discharge abstracts, and insurance claims. For example, the Joint Commission on Accreditation of Health Care Organizations (JCAHO) has identified 3 core measures of pregnancy-related hospital care: vaginal birth after cesarean, inpatient neonatal mortality, and third- or fourth-degree laceration.1 All of these measures are designed for use with hospital administrative data. More recently, the Agency for Healthcare Research and Quality (AHRQ), through its Quality Indicators project, has promoted the use of International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) coded data to ascertain obstetric trauma.2,3 Laceration rates have also been used to monitor and improve quality at the local level.4 HealthGrades, a proprietary health care rating company, publishes hospital-specific rates of “major complications” after cesarean and vaginal singleton deliveries, and after “patient choice” cesarean deliveries.5 Other vendors, such as Solucient (Solucient Risk-Adjusted Outcomes Software 5.2. Delivery Models. Solucient LLC, Evanston, IL) and HealthShare,6 offer similar products. The validity of all of these approaches depends upon the accuracy of the available data.
Multiple previous studies have evaluated the accuracy of complication codes in Medicare claims,7–10 the Veterans Health Administration Patient Treatment File,11–14 the California Patient Discharge Data Set,15 and similar data sets from single hospitals16,17 or other countries.18 In general, these studies have demonstrated substantial disagreement between ICD-9-CM–coded complications and medical records, with great variation in coding across hospitals,19 leading to concerns that administrative data should not be used to compare provider complication rates.20 However, these studies have focused almost entirely on medical-surgical patients and have not evaluated the coding of obstetric records.
The current study was conducted to validate the ICD-9-CM coding of obstetric complications in the nation's largest statewide patient discharge data program. We examined the sensitivity and positive predictive value of maternal hospital discharge abstracts, using the complete inpatient medical record as the gold standard. We hypothesized that clearly defined, clinically meaningful complications that require physician intervention, such as perineal lacerations and cesarean-related injuries, would be accurately coded in hospital discharge data, whereas more ambiguous complications, such as postpartum hemorrhage and endometritis, would be poorly coded.
MATERIALS AND METHODS
This study was undertaken as part of the California Hospital Outcomes Program, a legislatively mandated effort to improve the quality of hospital care and respond to the needs of purchasers and consumers by publishing risk-adjusted outcomes reports. These reports are based on the Patient Discharge Data Set, which includes an abstract of every discharge from every non-federal, licensed hospital in California. Each abstract lists the patient's birth date, sex, race, ZIP code, encrypted social security number, source and type of admission, discharge disposition, expected principal source of payment, total charges, principal diagnosis and up to 24 secondary diagnoses, up to 21 procedures, and up to 5 external causes of injury. The state government has responded to concerns about the validity of its outcomes reports by supporting studies to evaluate the accuracy of hospital-reported ICD-9-CM codes and the implications of data quality problems.
After delivery was selected as an appropriate condition for public reporting on hospital outcomes, we searched the MEDLINE database (1985–2000) to identify relevant clinical trials and case series. Additional papers were identified by an advisory panel (which included 4 obstetricians and/or perinatologists, 2 family physicians, 1 obstetric nurse specialist, and 1 health information professional) and by reviewing reference lists in obstetrics texts and meta-analyses. We excluded papers without abstracts or in languages other than English, studies from developing countries, and studies limited to patients with unusual procedures or risk factors. We then reviewed abstracts to locate studies with at least 250 patients that reported on postpartum maternal complications and/or readmissions. After discussing these findings with our advisory panel, we developed a comprehensive list of maternal complications, which we mapped to ICD-9-CM with the assistance of 2 coding professionals. We also implemented ICD-9-CM mappings of maternal complications that are used by JCAHO,1 AHRQ,3 HealthGrades,5 and Solucient (Solucient LLC).
Our sampling frame consisted of women between 10 and 55 years of age who were discharged from a nonfederal, licensed acute care hospital in California, after giving birth between January 1, 1992, and November 19, 1993. These dates were selected to permit validation of a published report on risk-adjusted outcomes. We defined delivery based on a principal or secondary diagnosis of 640–676, with a fifth digit of 1 or 2, or 650. We excluded cases with a principal diagnosis of postpartum care (V24.x), hydatidiform mole (630), other abnormal product of conception (631), or ectopic pregnancy (633.xx). We also excluded cases with a principal or secondary diagnosis of malignancy (141.x–172.x, 174.x–208.xx), missed abortion (632), or pregnancy with abortive outcome (634.xx–639.x). Finally, we excluded cases with any diagnosis of significant trauma or injury-related fetal death (codes available upon request). Cesarean delivery was defined to include all 74.xx codes, except 74.3 (removal of extratubal ectopic pregnancy) and 74.91 (hysterotomy to terminate pregnancy, if not associated with a live birth [V27.0, V27.2–V27.3, V27.5–V27.6]). Vaginal deliveries were defined by default as all other deliveries, except that those with any diagnosis of “cesarean delivery without mention of indication” (669.7x) or any procedure suggesting surgical delivery (74.3, 74.91) were excluded.
We linked delivery records with both postpartum (within 6 weeks after delivery) and antepartum hospitalizations (within 273 days before delivery), by using the patient's social security number and date of birth (precise algorithm available upon request). If 2 records had the same social security number but different dates of birth, both were discarded to minimize the risk of false linkage. About 20.1% of deliveries were excluded because of missing or invalid social security numbers. The date of delivery was assigned as the date of the earliest delivery-associated procedure (72.xx, 73.5x–74.xx) or the date of admission (in the absence of any valid delivery-associated procedure dates).
Through this linkage process, we identified and reconciled cases that appeared to have 2 or more deliveries within 182 days, delivery records sharing the same admission date, records with admission dates 1–7 days apart with identical procedures and diagnoses, and records with overlapping admission and discharge dates. Most cases of the first anomaly were corrected by searching for specific coding errors (eg, reporting a delivery diagnosis without a delivery procedure or outcome of delivery); uncorrectable cases were discarded. The same correction algorithm was applied to women with reported interpartum intervals of 182–223 days, except that uncorrectable cases were retained. Deliveries that occurred within 6 months after a molar (630–631), ectopic (633.x), or aborted (634.xx–637.xx, 639.x) pregnancy were excluded, unless retention of at least one viable fetus (651.3x–651.6x) was documented. The last 3 anomalies were resolved by manually selecting the more complete record of the same hospitalization or by randomly selecting from among identical records. Paired records that could not be “unduplicated” were discarded.
To ensure adequate representation, we identified and oversampled hospitals with fewer or more readmissions than expected. After excluding 15 hospitals with no licensed perinatal beds and 60 hospitals that performed too few deliveries (< 679) to be flagged as outliers, we used multivariable probit regression21 to estimate each patient's risk of experiencing a postpartum readmission. Only 1.94% of deliveries during the study period occurred at the excluded hospitals. We stratified the 267 remaining hospitals according to the number of postpartum readmissions: significantly (P < .01) or marginally (.01 < P < .10) more than expected, significantly or marginally fewer than expected, and neither. Within 3 of these 5 strata, we substratified northern California Kaiser hospitals for a separate collaborative project. We randomly sampled 46 hospitals from the 8 resulting strata (eg, 6 per stratum). Five hospitals declined to participate and were replaced by 10 randomly selected alternates. The resulting sample of 52 hospitals is representative of all acute care hospitals with active obstetric services in California: 3 city/county, 7 district, 10 Kaiser, 23 other nonprofit, 1 private university, and 8 for-profit hospitals.
Next, we randomly sampled eligible patients within each sampled hospital. Records with one or more readmissions and cesarean deliveries were oversampled to boost the number of patients with adverse outcomes and thereby improve efficiency. Stratified random sampling increases the reliability of estimates for patients of interest but allows the researcher to generate unbiased population estimates using sampling weights, as described below. To achieve 80% power to detect a 20% absolute interstratum difference in sensitivity or positive predictive value for a high-prevalence condition (eg, 14%), with type 1 error of 5%, we drew a sample of 1,662 deliveries, in which the number of cases contributed by each hospital within a stratum was proportional to its volume.
We asked each participating hospital to photocopy each sampled record in its entirety, including associated prenatal records if available. Each record was reviewed by 1 of 4 experienced accredited record technicians or certified coding specialists, who recoded the ICD-9-CM diagnosis and procedure codes, as well as maternal demographic and prenatal data, blinded to the original discharge abstract. A regional coding authority tested these individuals before they were hired, trained and supervised them, and verified at least 10% of their records (throughout the study) to ensure at least 95% accuracy. Discrepancies were resolved through collective review of appropriate coding references. Because JCAHO and AHRQ have endorsed perineal laceration rates as a quality indicator, differences between hospital-reported and recoded data on this outcome were carefully evaluated by 2 authors (J.M.K., P.S.R.).
Because of the complex sample structure, all analyses were weighted (unless otherwise noted) by the inverse of the sampling probability, which was calculated by multiplying the probability of sampling a specific hospital by the probability of sampling an individual within that hospital. These weights were adjusted to reflect both nonsubmitted records and records that were later classified as ineligible.
In this paper, the accuracy of hospital-reported ICD-9-CM complication codes was measured in terms of sensitivity and positive predictive value, using our recoding as the gold standard. We defined sensitivity as the percentage of patients with a complication identified through recoding for whom the same complication was reported on the hospital's original discharge abstract. We defined positive predictive value as the percentage of patients with a complication reported on the hospital's original discharge abstract for whom the same complication was independently found through recoding. We do not report specificity, or the percentage of patients without a complication (according to recoding) who were correctly reported as not having it, because this parameter was never below 97%, and nearly always exceeded 99%. We estimated confidence intervals using the SVYTAB procedure in STATA 7.0 (StataCorp LP, College Station, TX), which takes into account both oversampling of cesarean deliveries and clustering of observations within hospitals. The study protocol was approved by the appropriate committees at the University of California, Davis and the California Health and Human Services Agency.
We received 1,614 of the 1,662 records that we requested from participating hospitals (97.1%). Three of these records did not actually represent deliveries; 1,611 records were abstracted (30.3% primary cesarean, 18.9% repeat cesarean, 51.0% vaginal). This cohort had a weighted mean (standard deviation) age of 28.0 (0.52) years and a racial/ethnic composition similar to the target population (55% white, 8% African American, 8% Asian, 29% Hispanic, 0.1% Native American, and 0.8% “other”).
Tables 1 and 2 show the validated frequency, sensitivity, and positive predictive value of ICD-9-CM coded complications of obstetric care on California hospital discharge abstracts. Unweighted sensitivities and positive predictive values permit computation of the number of false-positive and false-negative cases. Weighted values permit extrapolation to the entire population of women who were delivered at acute nonfederal hospitals with active obstetric services in California during the study period. However, weighted estimates are less stable because women who were readmitted had a higher sampling probability than women who were not. Therefore, the weighted estimate should be interpreted cautiously when it differs greatly from the corresponding unweighted estimate. Unless otherwise stated, weighted estimates are cited below.
Table 1 shows that both third and fourth degree perineal lacerations were reported accurately, with sensitivities exceeding 90% and positive predictive values exceeding 65% (or 85% unweighted). A thorough review of 9 of the 12 discrepant cases by the lead nurse abstractor and first author confirmed 4 of the 5 false negatives, but 3 of the 4 apparent false positives were determined to be true positives. Unfortunately, records for 3 of the discrepant cases could not be retrieved when this review was undertaken. Reallocation of the 3 cases mislabeled as false positives would increase the unweighted positive predictive values of third-, fourth-, and either third- or fourth-degree lacerations to 93%, 100%, and 95%, respectively. Repair of a third- or fourth-degree laceration was also accurately reported, but adding this code to the definition of third- or fourth- degree laceration affected only 3 cases. Other intrapartum surgical complications were less frequent than perineal lacerations, so the corresponding estimates in Table 1 are less reliable. Only 2 of the 4 cases of urinary tract injury during cesarean delivery were properly reported. Pelvic hematomas and accidental pelvic injuries were poorly reported, with sensitivities of 45% and 41%, and positive predictive values of 69% and 99.7%, respectively.
Table 3 shows the highest reported degree of perineal laceration compared with the highest degree found on recoding. The data presented in boldface type, representing agreement between the 2 data sources, include 91% of vaginal deliveries. These data indicate the numbers or percentages of women who were confirmed as having the same degree of perineal laceration as reported on their discharge abstracts. Reallocation of cases mislabeled as false positives (described above) would increase the unweighted percentage of correctly reported third- and fourth-degree lacerations from 85% and 94% to 93% and 100%, respectively. Of the 72 disagreements in the table, 36 related to whether the patient suffered no injury or a first-degree laceration, and 11 related to whether the patient suffered a first- or second-degree laceration.
Table 2 describes hospitals' reporting of postpartum complications. Although most of these complications were reported with at least 80% positive predictive value, all suffered from substantial underreporting, with sensitivities of 68% or less. At the extreme, none of 4 thromboembolic complications was reported on the hospital discharge abstract.
Table 4 summarizes the estimated performance of several complex algorithms for identifying obstetric complications, including AHRQ's Patient Safety Indicators,2 experimental indicators, and legacy indicators from the Healthcare Cost and Utilization Project (used by HealthShare);6 HealthGrades' measures of major complications after vaginal and cesarean delivery,5 and Solucient's (Solucient LLC) measures of vaginal and cesarean complications. To avoid discarding informative data, we applied these algorithms without vendor-recommended denominator exclusions (eg, HealthGrades focuses on single liveborn deliveries). With the exception of 2 of AHRQ's Patient Safety Indicators (obstetric trauma with instrumented vaginal delivery, obstetric wound complications with vaginal deliveries), these algorithms suffer from similar underreporting as the complications we defined in Table 2.
The current study represents a comprehensive analysis of the accuracy of obstetric complication codes on hospital discharge abstracts. Merging California's database with information recoded from the medical records of patients admitted for delivery at 52 hospitals, we found very high levels of agreement (ie, sensitivities and positive predictive values of at least 85%) for third- and fourth-degree perineal lacerations, regardless of whether we used both diagnosis and procedure codes or diagnosis codes alone to ascertain these injuries. The quality of coding for other intrapartum injuries varied, but these findings are harder to interpret because of small numbers. Most postpartum complications and public-sector and commercial algorithms based on complication codes were subject to substantial underreporting, but relatively little overreporting.
These findings are not surprising because hospitals are only required to report “conditions that affect patient care in terms of requiring: clinical evaluation; or therapeutic treatment; or diagnostic procedures; or extended length of hospital stay; or increased nursing care and/or monitoring.”22 These criteria encompass any complication that requires surgical correction, such as a third- or fourth-degree perineal laceration or an iatrogenic injury to the cervix or urinary tract. However, these criteria exclude most pelvic hematomas, many superficial wound infections, and many cases of atelectasis and similar complications.
Health information professionals are further instructed “never [to] code a diagnosis as a complication unless it is stated as such and documented in the medical record by the attending physician.”23 They are warned not to ‘‘reach into the medical record to code other conditions for the sake of coding...[instead] contact the physician.'’24 The emphasis in these guidelines is for coders to rely upon physician documentation, which may be difficult to interpret, incomplete, or even misleading. In the absence of specific clinical criteria for such codes as “complications of the administration of anesthetic or other sedation in labor and delivery” (668.xx), coders at different hospitals may apply these codes very differently. Even for some straightforward codes, such as accidental puncture or laceration (998.2), ascertainment may vary because coders are instructed that “when a tear is documented in the operative report...the surgeon should be queried as to whether [it] was an incidental occurrence inherent in the surgical procedure or whether the tear should be considered...a complication.”25 Anecdotal evidence from coders indicates that such queries rarely occur, are often unanswered, and may evoke medicolegal concerns.
Our results are generally consistent with prior research on the Complications Screening Program (CSP), a complex algorithm for using administrative data to screen for potential complications, which demonstrated that 31% of surgical patients with an ICD-9-CM–coded complication did not have supporting clinical evidence, and 19% did not even have a supporting physician note.9 However, this study did not include obstetric patients. To our knowledge, none of the organizations promoting algorithms for ascertaining obstetric complications based on ICD-9-CM codes has published validation findings, although JCAHO has been involved in such efforts. One study focusing on uterine rupture during trial of labor found that the ICD-9-CM codes traditionally used to identify this complication (665.0x–665.1x)26 had a sensitivity of 64% or less and a positive predictive value of 51%.27
The most important limitation of this research is that our recoded data do not provide an ideal gold standard. However, we selected coders with experience in obstetrics, trained them thoroughly using specific written guidelines, monitored them carefully, and gave them unlimited time to code each record. When cases with discrepant data were thoroughly reviewed by the lead nurse abstractor and the first author, nearly every false negative was confirmed but some false positives were classified as recoding errors. Second, ICD-9-CM coding of obstetric complications may have improved in the past decade, especially with more widespread use of electronic record systems. The mean number of diagnoses reported on California discharge abstracts increased from 4.50 in 1997 to 5.46 in 2003,28 although comparable statistics are not available for obstetric abstracts. Finally, our exclusion of low-volume hospitals from the sampling frame may limit the generalizability of our findings, although fewer than 2% of women were affected by this exclusion.
Despite these limitations, our findings have important implications for current methods of evaluating the quality of obstetric care. The hospital discharge database serves as an excellent source of data on third- and fourth-degree perineal lacerations, confirming the criterion validity of JCAHO's Core Measure and AHRQ's Patient Safety Indicator on this topic. This new evidence complements evidence from prior studies that third- and fourth-degree perineal lacerations are associated with the use of episiotomy29–31 and have long-term effects on anal sphincter function.32 Although our findings do not rule out undercoding at individual hospitals, they suggest that hospitals, accrediting organizations, purchasers of hospital care, and public agencies can generally use administrative data to monitor the perineal laceration rate. If this rate is significantly higher than expected based on patient characteristics, providers should reevaluate their use of episiotomy and other components of intrapartum management.
Administrative data seem less useful for monitoring other types of early postpartum complications, such as wound infections, endometritis, and anesthesia-related complications. Our results raise serious questions, for example, about the validity of the complications measures promoted by HealthGrades and Solucient in their online hospital report cards. Similar problems are likely to affect other composite measures of postpartum complications33,34 and the “labor and delivery” measures proposed by the California Maternal Quality of Care Working Group,35 which are mostly based on the same ICD-9-CM codes. Given that the mean hospital stay for a vaginal delivery is only 2.1 days,36 it seems almost fruitless to search discharge abstracts from delivery hospitalizations for these complications, which typically do not become apparent for several days. Readmission records or outpatient claims may be more promising sources of data on postpartum complications that require medical attention.37
2. Romano PS, Geppert JJ, Davies S, Miller MR, Elixhauser A, McDonald KM. A national profile of patient safety in U.S. hospitals based on administrative data. Health Aff 2003;22:154–66.
4. Main EK, Bloomfield L, Hunt G, for the Sutter Health First Pregnancy and Delivery Clinical Initiative Committee. Development of a large-scale obstetric quality improvement program that focused on the nulliparous patient at term. Am J Obstet Gynecol 2004;190:1747–58.
7. Hartz AJ, Kuhn EM. Comparing hospitals that perform coronary artery bypass surgery: the effect of outcome measures and data sources. Am J Public Health 1994;84:1609–14.
8. Lawthers AG, McCarthy EP, Davis RB, Peterson LE, Palmer RH, Iezzoni LI. Identification of in-hospital complications from claims data. Is it valid? Med Care 2000;38:785–95.
9. Rosen AK, Geraci JM, Ash AS, McNeff KJ, Moskowitz MA. Predicting postoperative adverse events of common surgical procedures in the Medicare population. Prepared for Health Care Financing Administration. Springfield (VA): National Technical Information Service; 1991.
10. Keeler EB, Kahn KL, Bentow SS. Assessing quality of care for hospitalized Medicare patients with hip fracture using coded diagnoses from the Medicare Provider Analysis and Review files. Prepared for the Health Care Financing Administration. Santa Monica (CA): Rand Corporation; 1992.
11. Geraci JM, Ashton CM, Kuykendall DH, Johnson ML, Wu L. International classification of Diseases, 9th Revision, Clinical Modifications codes in discharge abstracts are poor measures of complication occurrence in medical inpatients. Med Care 1997;35:589–602.
12. Berlowitz DR, Brand HK, Perkins C. Geriatric syndromes as outcome measures of hospital care: can administrative data be used? J Am Geriatr Soc 1999;47:692–6.
13. Best WR, Khuri SF, Phelan M, Hur K, Henderson WG, Demakis JG, et al.. Identifying patient preoperative risk factors and postoperative adverse events in administrative databases: Results from the Department of Veterans Affairs National Surgical Quality Improvement Program. J Am Coll Surg 2002;194:257–66.
14. Barbour GL. Usefulness of a discharge diagnosis of sepsis in detecting iatrogenic infection and quality of care problems. Am J Med Qual 1993;8:2–5.
15. Romano PS, Schembri ME, Rainwater JA. Can administrative data be used to ascertain clinically significant postoperative complications? Am J Med Qual 2002;17:145–54.
16. Faciszewski T, Johnson L, Noren C, Smith MD. Administrative databases' complication coding in anterior spinal fusion procedures. What does it mean? Spine 1995;20:1783–8.
17. Massanari RM, Wilkerson K, Streed SA, Hierholzer WJ Jr. Reliability of reporting nosocomial infections in the discharge abstract and implications for receipt of revenues under prospective reimbursement. Am J Public Health 1987;77:561–4.
18. Hawker GA, Coyte PC, Wright JG, Paul JE, Bombardier C. Accuracy of administrative data for assessing outcomes after knee replacement surgery. J Clin Epidemiol 1997;50:265–73.
19. Romano PS, Chan BK, Schembri ME, Rainwater JA. Can administrative data be used to compare postoperative complication rates across hospitals? Med Care 2002;40:856–67.
20. Geraci JM. The demise of comparative provider complication rates derived from ICD-9-CM code diagnoses. Med Care 2002;40:847–50.
21. Romano PS, Luft HS, Remy LL, Rainwater J, Rennie DJ, Zhou H, et al. Report of the California Hospital Outcomes Project: Maternal outcomes following delivery, risk-adjustment methodology and preliminary findings. Sacramento (CA): Office of Statewide Health Planning and Development; 1996.
22. Guidelines for reporting other (additional) diagnoses. Coding Clinic for ICD-9-CM 1990;7(2):12–18.
23. Ask the editor. Coding Clinic for ICD-9-CM 1994;11(5):5–14.
24. Principal and other diagnoses delineated. Coding Clinic for ICD-9-CM 1984;1(1):9–14.
25. Coding questions submitted by the Maryland Medical Record Association. Coding Clinic for ICD-9-CM 1990;7(3):16–18.
26. Lydon-Rochelle M, Holt VL, Easterling TR, Martin DP. Risk of uterine rupture during labor among women with a prior cesarean delivery. N Engl J Med 2001;345:3–8.
27. Use of hospital discharge data to monitor uterine rupture–Massachusetts, 1990-1997. MMWR Morb Mortal Wkly Rep 2000;49:245–8.
29. Woolley RJ. Benefits and risks of episiotomy: a review of the English-language literature since 1980. Part I. Obstet Gynecol Surv 1995;50:806–20.
30. Myers-Helfgott MG, Helfgott AW. Routine use of episiotomy in modern obstetrics. Should it be performed? Obstet Gynecol Clin North Am 1999;26:305–25.
31. Hartmann K, Viswanathan M, Palmieri R, Gartlehner G, Thorp J Jr., Lohr KN. Outcomes of routine episiotomy: a systematic review. JAMA 2005;293:2141–8.
32. Sultan AH, Kamm MA, Hudson CN, Bartram CI. Third degree obstetric anal sphincter tears: risk factors and outcome of primary repair. BMJ 1994;308:887–91.
33. Koroukian SM. Relative risk of postpartum complications in the Ohio Medicaid population: vaginal versus cesarean delivery. Med Care Res Rev 2004;61:203–24.
34. 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.
35. Gregory KD, Hobel C, Korst KM, Lu M, Reyes C. A framework for developing maternal quality of care indicators. Los Angeles (CA): Cedars-Sinai Medical Center and UCLA School of Medicine; 2001.
36. U.S. Agency for Healthcare Research and Quality. Care of women in U.S. hospitals, 2000. HCUP Fact Book No. 3. AHRQ Publication No. 02-0044, October 2002. Available at: http://www.ahrq.gov/data/hcup/factbk3/factbk3.htm
. Retrieved July 22, 2005.
37. Lydon-Rochelle M, Holt VL, Martin DP, Easterling TR. Association between method of delivery and maternal rehospitalization. JAMA 2000;283:2411–6.
© 2005 by The American College of Obstetricians and Gynecologists.