The Agency for Healthcare Research and Quality (AHRQ) published its Guide to Prevention Quality Indicators: Hospital Admission for Ambulatory Care Sensitive Conditions 1 in October 2001, suggesting that certain inpatient conditions, such as pyelonephritis, might be avoided through improved outpatient health care services. Although AHRQ did not examine pyelonephritis in pregnancy, it has been cited as a potential indicator of the quality of maternal care.2 The purpose of this investigation was to examine whether hospital-specific rates of pyelonephritis in pregnancy may be a potential “prevention” quality indicator, ie, not a measure of the quality of hospital services, but rather an indirect measure of the quality of ambulatory prenatal care.
There is a consensus that, among nonpregnant women, pyelonephritis largely can be avoided, because urinary tract infection can progress with inadequate treatment.1 Pregnant women are known to be at increased risk of pyelonephritis,3 and the current practice guideline is to screen for asymptomatic bacteriuria early in pregnancy.4 Asymptomatic bacteriuria and pyelonephritis have been associated with preterm birth,5–7 and there is little disagreement regarding the efficacy of urine screening during prenatal care.8–11 Although estimates that 1% to 7% of women may develop pyelonephritis during pregnancy were published in the 1980s, more recent, population-based data seem to estimate the incidence of gestational pyelonephritis at 1% to 2% of all women giving birth.3,12 The rate of recurrence during the same gestation is estimated to be approximately 20%,3 although appropriate acute and suppressive antibiotic therapy is key to prevention. It is believed that nearly all of these admissions might be avoided through adequate outpatient screening and treatment of asymptomatic bacteriuria. Upon diagnosis, most women are admitted as inpatients for treatment for 2–3 days, as recommended by the most recent educational bulletin from the American College of Obstetricians and Gynecologists.13 Some practitioners have suggested discharging low-risk women after 24 hours for follow-up at home3 or after intramuscular antibiotic injections in the emergency department with oral antibiotic follow-up,3,14,15 although this is not known to be a common practice.
Simon et al16 examined the effect of a hospital-based voluntary clinical policies program, documenting a drop in the pyelonephritis rate among pregnant women from 1.4% to 0.5% at a single hospital after the implementation of a urine screening and treating policy during prenatal care for both teaching and community physician services. Physician compliance at the outpatient sites illustrated a willingness to create the “provider partnership” required to achieve optimal maternal and fetal outcomes. With the continued pressure for physician-hospital affiliation, not only for shared risk contracting, but also for better health care coordination and quality, such a partnership would be in the best interests of patients, employers, insurers, and all maternal care providers.17
Given the large degree of evidence and consensus in the obstetric community with respect to the ability of prenatal care to prevent pyelonephritis and the reported successful use of hospital leverage to encourage compliance with urine screening and treatment among both private and teaching services, the objective of our study was to examine whether hospital-specific rates of gestational pyelonephritis may be a potential quality indicator for ambulatory maternal health care services.
MATERIALS AND METHODS
This study was reviewed and permitted by the Institutional Review Boards for the State of California and Cedars-Sinai Medical Center. The study data were derived from a linked database made up of data sets publicly available from the California Office of Statewide Health Planning and Development (OSHPD), Health Information Policy Division, for the calendar year 1997. The data sets linked included the infant vital statistics birth record, the infant hospital discharge record, the maternal hospital discharge record, and all maternal hospital discharge records from the gestational period. Relevant documents regarding the linkage performed by the State of California can be obtained from the published literature or OSHPD.18,19
The OSHPD data set contained 1 record per newborn. We transformed this data set using a deterministic linkage to match the records from multiple gestations. Starting with 507,592 birth records, we identified 494,762 singleton and 12,648 multiple gestation newborns, initially matching by maternal hospital number, admission date, county of residence, age (in years), payer source, hospital length of stay, and race. To validate the results, matched and unmatched records were further examined by maternal and newborn International Classification of Diseases, 9th Revision, Clinical Modification codes (ICD-9-CM), maternal social security number (scrambled), and hour of delivery for matched multiple gestation records. We removed from analysis 182 records with inconsistent data that may have resulted from erroneous coding. Antepartum inpatient admission records, ie, records of pregnant women discharged undelivered, were then linked with deterministic methods to the delivery data set described above. The resulting data set allowed us to examine procedures and clinical conditions associated with each delivery and associated antepartum hospital admissions. No outpatient records were examined.
Our methodologic approach was designed to be consistent with that suggested by AHRQ: We used administrative data, and we evaluated our results in light of the 6 criteria that AHRQ recommends be met by quality indicator candidates.1 To be included in this study, patients had to have first-trimester prenatal care (as determined from birth certificate data). Only women with access to first-trimester prenatal care are considered so that the issue of access to preventive care is mitigated. Hence, all women in the study population had optimal opportunity to be screened. For clarity, we used restrictions in defining the patient subgroup to isolate a relatively homogeneous and “low-risk” population. To strengthen any causal inference between prenatal care practice patterns and pyelonephritis, mothers who had 1 or more maternal, fetal, or placental complications, as listed in Appendix A and neonates who had any congenital anomaly (ICD-9-CM 740–759) or malignancy were excluded. Women with certain clinical conditions that were either potential risk factors for, or outcomes of, pyelonephritis were kept in the study population (Appendix B). For example, because preterm labor may be physiologically caused through the inflammatory process associated through pyelonephritis,8 women with preterm births were not excluded on that basis. Other potential outcomes of pyelonephritis were identified, and women with these conditions were not excluded: chorioamnionitis, prolonged ruptured membranes, intrauterine growth restriction, and oligohydramnios. Finally, women known to be at excess risk of gestational pyelonephritis, specifically those with sickle cell trait or disease, or renal disease remained in the study population. Hospitals with fewer than 200 deliveries during 1997 were excluded because of the difficulty in extrapolating results to practice patterns in such low volume hospitals, and deliveries to women with a previous delivery in California in 1997 were excluded. In summary, with the exception of several conditions reported to be risk factors for or outcomes of pyelonephritis, the study population consisted largely of low-risk women who delivered their infants in California hospitals in 1997 who had prenatal care in the first trimester of pregnancy.
A history of antepartum admission for gestational pyelonephritis was defined by the presence of ICD-9-CM codes 646.6 or 590.1 within the Diagnosis Related Group (DRG) of 383 and a date of admission 180 days or more before delivery. A history of admission for preterm labor was defined by ICD-9-CM code 644.0. For those women diagnosed with pyelonephritis, the interval from the first episode of pyelonephritis to the delivery was calculated.
Several outcomes were examined. The case definition for maternal infection, ie, endometritis, used code 670, uterine infection. Preterm birth was categorized through a combination of ICD-9-CM and DRG data from both the maternal and neonatal OSHPD data sets, and from birth certificate data, because of inconsistent information across data sources (code available by request). The occurrence of neonatal congenital infection was derived from ICD-9-CM codes 770.0 (congenital pneumonia), 038 (confirmed sepsis), and 771.8 (perinatal infection).
Other patient-level variables were constructed from the data set. Payer was categorized as private/managed care, Medicaid, staff model health maintenance organization (HMO), and other or unknown. Race or ethnicity was categorized as Asian, African-American, Hispanic, Native American, and white, and maternal age was classified as older (≥ 35 years) or younger. Parity was derived from birth certificate data, and women were classified as nulliparous, multiparous with previous cesarean delivery, and multiparous without previous cesarean delivery. The number of prenatal visits was obtained from birth certificate data.
Ownership of the hospital of birth was defined per the OSHPD data categories (staff model HMO, district, University of California, not-for-profit, for-profit [corporate], and government). The presence of a hospital teaching program for medical residents in obstetrics was ascertained from the Graduate Medical Education Book produced by the State of California.20
Rates of occurrence of pyelonephritis in this low-risk population are provided, along with χ2 tests with Yates continuity correction for 2 × 2 tables. Observed odds ratios (ORs) are reported with 95% confidence intervals (CIs).
A Bayesian hierarchical logistic regression model was used to identify significant risk factors and generate risk-adjusted hospital-specific rates (WinBUGS software V.1, Imperial College & MRC, Cambridge, UK). The hierarchical (2-level) model is necessary to account for the intracluster correlation resulting from the clustering of patients in hospitals.21 The application of such models is becoming more widespread.22,23 Estimates of hospital-specific parameters are partially pooled toward the overall mean, compromising between completely pooled (disregarding hospital clustering) and unpooled (modeling each hospital separately) estimates. Case-mix risk factors included age group, payer, race or ethnicity, and parity group. Hospital ownership and teaching status were used as hospital characteristics. Hospital-specific infection rates were calculated, including 95% credibility sets (Bayesian CI’s). Hospitals with adjusted rates and confidence intervals above the threshold of the upper quartile were labeled as “high outliers.”
We identified a total of 507,410 infants born to 500,984 women in the data set for calendar year 1997. These were associated with 44,187 antepartum discharges for 35,231 pregnancies, with mothers admitted up to 29 times for a single gestation.
The derivation of the study population is described in Figure 1. After creating the “low-risk” sample described above, the remaining 350,522 women were stratified on the basis of first-trimester prenatal care. Those women who did not begin prenatal care in the first trimester or who had unknown data with respect to the first prenatal care visit were 1.21 times more likely to have an antepartum admission for pyelonephritis (542/67,994 [0.80%] compared with 1,861/282,528 [0.66%], OR 1.21, 95% CI 1.10–1.33). These women were then excluded from the denominator. Of the remaining 282,528 women, the final study sample of 280,816 (56.0%) was derived after excluding deliveries at hospitals with fewer than 200 deliveries and second deliveries to the same woman. The analysis sample included 291 of the 322 California hospitals in the full data set.
The total number of women with at least 1 antepartum admission for pyelonephritis in the 180 days before delivery was 1,853 (0.66%). The maximal number of antepartum admissions for pyelonephritis was 7. The median interval from the first presentation with pyelonephritis to delivery was 92 days (range 1 to 180 days).
Birth outcomes in this study population were consistent with those reported in the literature (Table 1). Women who had an antepartum admission for pyelonephritis were more likely to have a preterm birth or infected newborn compared with women without such an admission.
Upon univariate analysis, the population most at risk for pyelonephritis seemed to be young, nulliparous, and African American (Table 2). Few patients had documentation of either sickle cell trait or sickle cell disease, and we were not able to evaluate the risk of pyelonephritis among these women. A small number of women, 165 (0.06%), had a history of a renal condition; of these, 10 (6.06%) had an admission for pyelonephritis (P < .001). Women delivering at government hospitals and with MediCal as their payer also seemed to be at increased risk.
Of note, with univariate analysis, staff model HMO patients not only had the lowest rate of pyelonephritis compared with all other categories of hospital ownership (0.38%), they also had the lowest percentage of patients completing at least 10 prenatal care visits (72%, compared with 85% for MediCal patients and 91% for all classes of private patients). Table 3 describes how the study patients’ payers are distributed across the different categories of ownership of the hospital of delivery and categories of race or ethnicity, because these patterns are unique to obstetric populations. For example, nearly one half of the patients delivering at for-profit hospitals have MediCal as their payer, and no specific race or ethnicity is uniformly associated with any one type of payer.
Upon adjustment of the OR through the multivariate hierarchical logistic regression model, most of the identified associations with pyelonephritis, eg, nulliparity and younger age, were minimized, and the predominant factors that affected the occurrence of pyelonephritis diminished to 2: having MediCal as the payer (OR 1.60, 95% CI 1.46–1.80), and being African American (OR 1.24, 95% CI 1.10–1.41). The effect of hospital ownership, an important driver of inpatient maternal quality measures in other studies,24–28 was minimal.
Hospital-specific rates of pyelonephritis were then calculated using the adjusted OR; this rate distribution is plotted in Figure 2. The median rate was 0.60% (range 0.22% to 2.64%), and the upper quartile of the distribution began at a rate of 0.76%. Fourteen hospitals (15% of the study hospitals) met our definition of being a "high outlier" because the lower confidence limit of the hospital pyelonephritis rate was greater than 0.76%.
The objective of this study was to examine whether hospital-specific rates of gestational pyelonephritis can function as a quality indicator for ambulatory maternal health care services. Our clinical findings were consistent with those reported in the obstetric literature. We found that rates of preterm birth and congenital neonatal infection were substantially increased among pregnant women with pyelonephritis, although we identified no increase in intrauterine infection. Furthermore, the rate of gestational pyelonephritis (0.66%) in this low-risk population also seemed consistent with previous reports, as did the range of adjusted hospital-specific rates (0.22% to 2.64%). Thus, although a normative or target hospital rate for gestational pyelonephritis remains to be established, these data demonstrate that, given its preventability and its ties to subsequent morbidity, gestational pyelonephritis has a role in the assessment of the quality of maternal care.
In an effort to account for the potential risk factors and variation due to differences in hospital populations, we used a Bayesian hierarchical logistic regression model. Two patient characteristics were outstanding: being African American (OR 1.24, 95% CI 1.10–1.41), and having MediCal as the payer for pregnancy services (OR 1.60, 95% CI 1.46–1.80). It is established that the presence of sickle cell disease or sickle cell trait, prevalent among African-American women, has been associated with increased risk for gestational pyelonephritis. Based on ICD-9-CM discharge coding, there were few cases of gestational pyelonephritis associated with sickle cell trait or disease. Nonetheless, it seems prudent to suggest that women with these conditions require close monitoring of their urine and surveillance for signs or symptoms of urinary tract infection after 20 weeks of gestation.12 Alternatively, the higher risk of gestational pyelonephritis among African-American women may reflect that they received a poorer quality of care. Previous work has shown that African-American women are less likely to receive recommended prenatal advice and procedures from their care providers than white women.29 We were unable to test this hypothesis with the available data.
Women on MediCal may be at risk either because 1) poverty may indirectly make them more physiologically vulnerable or 2) they receive suboptimal care. Multiple studies have reported that gestational pyelonephritis is associated with low socioeconomic status.3,12 Likewise, low socioeconomic status is associated with increased risk of other perinatal infections such as Group B β streptococcal infection, bacterial vaginosis, and sexually transmitted infections. However, a plausible physiologic hypothesis implicating poverty as causal in these conditions has not been established, although room exists for speculation that decreased nutritional status or immune function or another unmeasured physiologic factor may indeed play a role. We suggest that, particularly in light of results from Simon et al,16 a likely contributor to increased rates of gestational pyelonephritis may be suboptimal prenatal care. Examples of suboptimal care are that 1) patients may not undergo routine screening for asymptomatic bacteriuria, 2) patients may not be called back or be available for treatment after receiving a positive result, 3) different strategies for point of care testing may be used based on health system resources that may have less than ideal test accuracies,30,31 or 4) patients may not have received culture-specific agents or appropriate doses of these agents. We hypothesize that the clinical care practices of prenatal care providers who admit to hospitals with high rates of pyelonephritis differ from those of prenatal care providers who admit to hospitals with low rates.
These data suggest that MediCal patients are at higher risk for pyelonephritis regardless of whether they deliver at a for-profit or government hospital. Furthermore, MediCal patients comprise nearly half of the deliveries in California, and they deliver at hospitals in all ownership categories. Joint efforts between hospitals and prenatal care providers may be fruitful with respect to the prevention of acute pyelonephritis.
The AHRQ has laid out 6 criteria by which to examine potential candidates for quality indicators.1 The first criterion is face validity: are we measuring what we think we are measuring? The rates of pyelonephritis and associated complications and risk factors are consistent with previous reports. With routine surveillance, peer pressure may be sufficient to enforce good practice.
The second criterion to be met by a quality indicator is that of precision. Figure 2 indicates that there is a substantial amount of nonrandom variation among hospitals. After attempting to account for all potential reasons why hospital pyelonephritis rates might differ, hospital rates show wide variation, ranging from less than one-half percent to over 2%.
Criterion 3 is the requirement for minimum bias. We used a population-based cohort to be representative. Our efforts to derive and apply our findings to a “low-risk” group of women should assist in the interpretation of these rates. Nevertheless, we still found it necessary to adjust for multiple patient and hospital-level parameters to identify the 2 factors (African-American race and MediCal as payer) that put women at risk for pyelonephritis. Thus, the use of unadjusted hospital rates would not be recommended.
Construct validity is the fourth criterion. We are inferring that hospital rates of pyelonephritis reflect the quality of prenatal care. How valid is this inference? Linking the failure of prenatal providers to screen or treat patients with actual episodes of pyelonephritis would require primary data collection at the patient level, including both inpatient and outpatient records. Yet because a copy of prenatal records is usually entered in the patient’s medical chart, such data collection is feasible, and construct validity could be substantiated.
The fifth criterion is whether the indicator seems to be robust, or stable, when measured. These data have demonstrated that despite the low prevalence of pyelonephritis, “high outlier” hospitals can be identified, and that substantial practice variation exists.
The final criterion is feasibility of producing the measure. Most of the data required to identify patients with gestational pyelonephritis are routinely collected from inpatient discharge summaries that are required by the State of California. However, this is not a timely method, requiring a lag of 2–3 years to prepare the data. Interested hospitals could prepare the data themselves, linking antepartum admissions and prenatal records to the delivery and newborn admission. The potential exists for hospitals to find a method of pulling all of these data together if properly motivated.
As with other quality indicators derived by AHRQ, gestational pyelonephritis rates have been derived from administrative discharge data. Through experience across multiple disciplines, ICD-9-CM codes for patient conditions and procedures have often been shown to be highly specific, yet not always sensitive.32–34 However, as is true for many other quality indicators, the validity of the demographic and clinical data used here has not been well-substantiated. Until large scale studies can be performed, administrative data collection systems provide the most feasible option for research studies evaluating variation in clinical practices, healthcare outcomes, and quality of care.
In summary, these findings suggest that because of its preventability and its consequent related morbidity, gestational pyelonephritis meets clinical requirements as a quality indicator for ambulatory maternal care. Hospital-specific pyelonephritis rates also meet technical requirements for quality indicators. The use of such rates would provide an opportunity for hospitals to improve patient outcomes through partnership with obstetricians in the development of policies and procedures for women at risk.
1. Agency for Healthcare Research and Quality. Guide to prevention quality indicators: hospital admission for ambulatory care sensitive conditions, Revision 3. AHRQ Publication No. 02-R0203, January 9, 2004.
2. Korst LM, Gregory KD, Lu MC, Reyes C, Hobel CJ, Chavez G. A framework for the development of maternal quality of care indicators. Matern Child Health J. 2005;1–25 [Epub ahead of print].
3. Wing DA. Pyelonephritis. Clin Obstet Gynecol 1998;41:515–26.
4. American Academy of Pediatrics. Guidelines for perinatal care. 5th ed. Elk Grove Village (IL): American Academy of Pediatrics; Washington (DC): American College of Obstetricians and Gynecologists; 2002.
5. Millar LK, Cox SM. Urinary tract infections complicating pregnancy. Infect Dis Clin North Am 1997;11:13–26.
6. Connolly A, Thorp JM Jr. Urinary tract infections in pregnancy. Urol Clin North Am 1999;26:779–87.
7. Romero R, Oyarzun E, Mazor M, Sirtori M, Hobbins JC, Bracken M. Meta-analysis of the relationship between asymptomatic bacteriuria and preterm delivery/low birth weight. Obstet Gynecol 1989;73:576–82.
8. Vazquez JC, Villar J. Treatments for symptomatic urinary tract infections during pregnancy. The Cochrane Database Syst Rev 2003;(4):CD002256.
9. Smaill F. Antibiotics for asymptomatic bacteriuria in pregnancy (Cochrane Review). In: The Cochrane Library, Issue 3, 2002a. Oxford: Update Software.
10. Villar J, Bergsjo P. Scientific basis for the content of routine antenatal care. I. Philosophy, recent studies, and power to eliminate or alleviate adverse maternal outcomes. Acta Obstet Gynecol Scand 1997;76:1–14.
11. Rouse DJ, Andrews W, Goldenberg R, Owen J. Screening and treatment of asymptomatic bacteriuria of pregnancy to prevent pyelonephritis: a cost-effectiveness and cost-benefit analysis. Obstet Gynecol 1995;86:119–23.
12. Pastore LM, Savitz DA, Thorp JM Jr, Koch GG, Hertz-Picciotto I, Irwin DE. Predictors of symptomatic urinary tract infection after 20 weeks’ gestation. J Perinatol 1999;19(7):488–93.
13. American College of Obstetricians and Gynecologists. Antimicrobial Therapy for Obstetric Patients. ACOG Educational Bulletin 245. Washington, DC: ACOG; 1998.
14. Nicolle LE. Screening for asymptomatic bacteriuria in pregnancy. Canadian Guide to Clinical Preventive Health Care. Ottawa (Canada): Health Canada; 1994. p. 100–6.
15. Wing DA, Hendershott CM, Debuque L, Millar LK. Outpatient treatment of acute pyelonephritis in pregnancy after 24 weeks. Obstet Gynecol 1999;94:683–8.
16. Simon NV, Heaps KP, Chodroff CH. Improving the processes of care and outcomes in obstetrics/gynecology. Jt Comm J Qual Improv 1997;23:485–97.
17. Casalino L, Robinson JC. Alternative models of hospital-physician affiliation as the United States moves away from tight managed care. Milbank Q 2003;81:331–51.
18. Herrchen-Danielsen B, Gould JB. User Manual and technical report: linkage of vital statistics linked birth/infant death, infant, and maternal hospital discharge file. Berkeley (CA): UC Berkeley; 1996.
19. Herrchen B, Gould JB, Nesbitt TS. Vital statistics linked birth/infant death and hospital discharge record linkage for epidemiological studies. In: Computers and biomedical research. San Diego (CA): Academic Press; 1997. p. 290–305.
20. Office of Statewide Health Planning and Development (OSHPD). California graduate medical education programs, 1996–1997 Update. Sacramento (CA): OSHPD; 1998.
21. Spiegelhalter DJ, Thomas A, Best NG, Gilks WR, Lunn D. BUGS: Bayesian inference using Gibbs sampling. 2003. MRC Biostatistics Unit, Cambridge, United Kingdom. www.mrc-bsu.cam.ac.uk/bugs
22. Christiansen CL, Morris CN. Improving the statistical approach to health care provider profiling. Ann Intern Med 1997;127:764–8.
23. Hogan JW, Blazar AS. Hierarchical logistic regression models for clustered binary outcomes in studies of IVF-ET. Fertil Steril 2000;73:575–81.
24. Gregory KD, Korst LM, Platt LD. Variation in elective primary cesarean delivery by patient and hospital factors. Am J Obstet Gynecol 2001;184:1521–32.
25. Gregory KD, Korst LM, Gornbein JA, Platt LD. Using administrative data to identify indications for elective primary cesarean delivery. Health Serv Res 2002;37:1387–401.
26. Korst LM, Fridman M, Friedlich PS, Lu MC, Reyes C, Hobel CJ, et al. Hospital rates of maternal and neonatal infection in a low risk population. Matern Child Health J. 2005;1–10 [Epub ahead of print].
27. Lu MC, Fridman M, Korst LM, Reyes C, Hobel CJ, Chavez G. Postpartum hemorrhage rates among laboring women: A feasibility study of a potential indicator of hospitals’ quality of obstetrical care. Matern Child Health J 2005. [Epub ahead of print].
28. Korst LM, Gornbein JA, Gregory KD. Rethinking the cesarean delivery rate: how pregnancy complications may affect interhospital comparisons. Med Care 2005;43:237–45.
29. Kogan MD, Kotelchuck M, Alexander GR, Johnson WE. Racial disparity in reported prenatal care advice from health care providers. Am J Public Health 1994;84:82–8.
30. D’Souza Z, D’Souza D. Urinary tract infection during pregnancy—dipstick urinalysis vs. culture and sensitivity. J Obstet Gynaecol 2004;24:22–4.
31. Teppa RJ, Roberts JM. The uriscreen test to detect significant asymptomatic bacteriuria during pregnancy. J Soc Gynecol Investig 2005;12:50–3.
32. DiGiuseppe DL, Aron DC, Payne SM, Snow RJ, Dierker L, Rosenthal GE. Risk adjusting cesarean delivery rates: a comparison of hospital profiles based on medical record and birth certificate data. Health Serv Res 2001;36:959–77.
33. Iezzoni LI, Ash AS, Shwartz M, Daley J, Hughes JS, Mackiernan YD. Predicting who dies depends on how severity is measured: implications for evaluating patient outcomes. Ann Intern Med 1995;123:763–70.
34. Korst LM, Gregory KD, Gornbein JA. Elective primary cesarean delivery: accuracy of administrative data. Paediatr Perinat Epidemiol 2004;18:112–9.