Uterine rupture is often cited as the most catastrophic event associated with attempting labor after a previous cesarean delivery and past research has focused on predicting uterine rupture and the risks associated with rupture for mother and neonate.1–4 In this article, we explore whether the occurrence of rupture in one woman affects the obstetric management of labor and delivery in other women cared for at the same hospital. Such an effect may be present, because psychology researchers have found that recent events can affect decision-making.5
In this article, our objective was to identify the extent to which a hospital's vaginal birth after cesarean delivery (VBAC) rate, trial of labor after cesarean delivery (TOLAC) rate, or trial of labor success rate decreased in the months after a uterine rupture. We hypothesized that the occurrence of a uterine rupture might alter health care providers' perception of risk or decrease their risk tolerance, leading to decreases in TOLAC, and trial of labor success. A reduction in either of these rates would lead to an increase in the hospital-level rate of repeat cesarean delivery.
MATERIALS AND METHODS
The study population was drawn from hospital deliveries in the Nationwide Inpatient Sample, Healthcare Cost and Utilization Project, Agency for Healthcare Research and Quality6 between 1998 and 2010, inclusive. The Nationwide Inpatient Sample is the largest all-payer inpatient care database that is publicly available in the Unites States and is part of the Healthcare Cost and Utilization Project. The data are reviewed for completeness, undergo logic checks, and are compared with other national data sources of hospital care such as the National Hospital Discharge Survey to maintain database quality. The data have been used extensively for research and quality assurance projects.7,8 Each year contains data from approximately 8 million hospital stays in 1,000 hospitals sampled to approximate a 20% stratified sample of U.S. hospitals.6 As a result of the nature of the random sampling of hospitals in the sample, hospitals can contribute data across multiple calendar years, but these years may be scattered across the study period. Because the number of years each hospital was sampled for inclusion into the data set differs, we discuss hospital-years throughout this article, where one hospital-year encapsulates all deliveries to a particular hospital within a given calendar year. The institutional review board at the McGill University Faculty of Medicine approved this study.
International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes were used to identify the study population, cases of severe uterine rupture, and the delivery outcomes. The specific codes are noted in the Appendix, available online at http://links.lww.com/AOG/A572: Additional Information. Delivery admissions were identified using the algorithm described by Kuklina et al,9 and from these, we extracted women with a previous cesarean delivery as our study population. The ICD-9-CM definition for uterine rupture was established before increased concern about rupture and includes milder complications such as lacerations of the uterus and obstetric trauma not elsewhere classifiable.10 We created a composite outcome for severe uterine rupture by restricting to ruptures accompanied by hysterectomy, postpartum haemorrhage, blood transfusion, embolization, or stillbirth or a combination of these outcomes. We were unable to identify severe ruptures that resulted in neonatal morbidity or mortality because the data set does not link maternal and neonatal records. We decided a priori to only include ruptures associated with severe maternal morbidity or stillbirth, because these ruptures were most likely to affect practice and decision-making. Restricting to severe events may have also increased the validity of our variable, because a validation study of another administrative database found that severe events tend to be more accurately coded than less severe events.11 Multiple ruptures at a hospital were used if they occurred in different years. Outcomes were the hospital-level VBAC rate (number of vaginal births to women with a previous cesarean delivery divided by number of women with a previous cesarean delivery), TOLAC rate (number of women entering labor with a previous cesarean delivery divided by number of women with a previous cesarean delivery), and trial of labor success rate (number of vaginal births to women with a previous cesarean delivery divided by number of women entering labor after a previous cesarean delivery). Because there is no ICD code for labor, it was identified using the algorithm described by Uddin and Simon.12
The difference-in-differences methodology was used to estimate changes in rates of the outcomes after uterine rupture that are above and beyond changes experienced by hospitals that do not have ruptures.13 This methodology is a type of controlled prepost design and is often superior to the more commonly used prepost design that can suffer from confounding as a result of underlying time trends in the outcomes. The design uses all hospitals to estimate a common time trend in the outcome that is subtracted from the change in the outcome experienced by hospitals with uterine rupture (Fig. 1). The design controls directly for preuterine rupture differences in TOLAC rates and delivery volume between hospitals with and without ruptures and any other time-fixed characteristics of hospitals that differ between hospitals with and without ruptures.
We used a conditional linear probability model to estimate changes in each of our outcomes (VBAC, TOLAC, and trial of labor success rates) at the time of uterine rupture within hospitals that are over and above changes experienced by hospitals without uterine rupture. Twelve indicator variables denoted the number of months after each rupture's occurrence, where the first indicated deliveries in the same calendar month as the rupture, the second indicated deliveries in the calendar month after the rupture, and the last indicated deliveries in the 11th calendar month after the rupture. Estimating an effect for each month separately allowed us to investigate the presence of the lag of the effect, the duration (in months) that any effect appears to last, and allowed the magnitude of the effect to vary. In addition, the intercept term from each model can be interpreted as the average preuterine rupture rate of the outcome at a hospital.14 A construction of the model starting from a simpler framework is given in the Appendix (http://links.lww.com/AOG/A572). All statistical analyses were conducted using Stata 12.115 and both Stata and R 3.0.216 were used to create the figures. Statistical code is provided in the Appendix (http://links.lww.com/AOG/A572). We used the sampling weights provided with the data to calculate nationally representative trends in the outcomes.17
The design assumes that hospitals with and without uterine ruptures do not differentially experience other changes to the outcome rate at the time of the rupture such as a change in hospital protocol only at hospitals with uterine rupture in the month of the rupture. To test this, we conducted a negative control test,18 where a failure to “pass” the test would provide an indication that this model assumption is invalid. To conduct the test we fit the same model but used an outcome that could not plausibly be affected by the occurrence of rupture. We chose diabetes (both preexisting and gestational diabetes; see the Appendix for ICD-9-CM codes [http://links.lww.com/AOG/A572]) as the placebo outcome. Because the proportion of women with a previous cesarean delivery who also have diabetes should not change with the occurrence of a rupture, any nonnull effect estimate of rupture on the portion of women with diabetes would reveal a violation of the assumption and imply the existence of other hospital-level changes at the time of the rupture and a potential bias in the estimated effects from the primary analyses. The model also assumes that the underlying time trends in the VBAC, TOLAC, and the trial of labor success rates are parallel in hospitals with and without uterine rupture. To assess the validity of this assumption, we repeated the primary analyses estimating the common secular trend using only hospitals that had a uterine rupture occurrence.
We identified 10,888,501 deliveries (between 741,787 and 930,086 each year) in 3,128 hospitals, of which 1,534,755 (14%) had a previous cesarean delivery in 2,986 hospitals. In 1998, 88,171 women (12% of the obstetric population) had a previous cesarean delivery, and by 2010, 128,664 (17% of the obstetric population) had a previous cesarean delivery. Two coding errors were detected and resulted in the exclusion of one hospital-year of deliveries and the recoding of the ICD-9-CM code for cesarean delivery in a subset of hospitals in 2000 (details in the Appendix, http://links.lww.com/AOG/A572). Eight percent of deliveries were missing information on delivery month and were excluded from the analysis. The vast majority of these exclusions were deliveries in Florida, which did not provide admission month in the data set. California, New York, and West Virginia each had some records with missing admission month: 0.01, 0.13, and 11.34% of deliveries in each state were missing admission month, respectively.
After exclusions, there were 1,414,134 deliveries to women with a previous cesarean delivery across 2,859 hospitals and 600 severe uterine rupture events. These hospitals were observed for a total of 8,485 hospital-years, implying that the average hospital contributed data in the sample across 3 years. Of the 600 severe uterine ruptures, 510 were the first uterine rupture to occur at a hospital in a given calendar year and it is these ruptures that are used in the analysis. There were 211,850 deliveries in 510 hospital-years containing uterine ruptures and 1,202,284 deliveries in 7,975 hospital-years that did not have ruptures.
The VBAC rate decreased from 35% in 1998 to 10% in 2006 and plateaued thereafter (Fig. 2). In 1998, the TOLAC rate was 50%. By 2006, it decreased to 19%. The trial of labor success rate decreased from 70% in 1998 to 51% in 2010 in these women.
In 1998, the risk of severe uterine rupture was 6.1 ruptures per 10,000 women with a previous cesarean delivery, and this decreased to approximately 4.2 severe ruptures per 10,000 in 2010.
Figure 3 depicts the risk difference estimates (also referred to as “excess risk”) for changes in the VBAC, TOLAC, and trial of labor success rates, where each estimate can be interpreted as the change in the absolute risk of the outcome (at the specified time point compared with prerupture months) experienced by hospitals that had a uterine rupture, above any secular change in the outcomes experienced by hospitals with no rupture. Although the overall VBAC rate decreased in time (Fig. 2), the VBAC rate in hospitals with uterine rupture was estimated to experience an additional decrease, especially in the month immediately after the rupture (Fig. 3A). In this first postrupture month, for every 1,000 women with a previous cesarean delivery, 17 fewer women had a vaginal delivery (estimate: −17/1,000; 95% confidence interval [CI] −4/1,000 to −31/1,000, P=.01). In subsequent months, the direction of the effect continued to be negative, but the estimates were not statistically significant.
Figure 3B illustrates the estimated risk differences for the TOLAC rate. For deliveries in the same month, the estimate is consistent with no difference in TOLAC rate, because the estimated excess risk in TOLAC was 14 per 1,000 (95% CI −1/1,000 to 29/1,000, P=.06). In the 11 months after the rupture, all of the point estimates are negative and every CI includes the null value except for the fifth month.
Before a uterine rupture occurrence, the average rate of trial of labor success was estimated to be 60 successful vaginal deliveries per 100 labors in women with a previous cesarean delivery, as estimated by the intercept term of the model. In the month of the uterine rupture, the trial of labor success rate was significantly lower compared with previous months (estimate −55/1,000; 95% CI −70/1,000 to −40/1,000, P<.001) after accounting for the secular time trend (Fig. 3C). The trial of labor success rate was still significantly lower in the month directly after the uterine rupture (estimate −25/1,000; 95% CI −44/1,000 to −6/1,000, P=.01). By the third month, however, there appeared to be no difference between the exposed and unexposed hospitals, suggesting that the effect of a uterine rupture on the trial of labor success was transient.
To investigate the model assumptions, we examined the effect of uterine rupture on a negative control outcome: the portion of women in the study population who had diabetes. Our results support the notion that there was no effect of uterine rupture occurrence on the portion of women with diabetes; all of the effect estimates include the null value and there was no discernible trend in the effect estimates across time (Fig. 4). Second, we repeated the main analyses using only hospital-years that contained a uterine rupture occurrence. Results supported the model assumption that the underlying trends in the outcomes would not differ between those hospitals that experienced a uterine rupture and those that did not in the absence of a uterine rupture (Fig. 5).
Our results show that there was a decrease in the rate of successful trials of labor in women with a previous cesarean delivery and a lower VBAC rate in the month after a severe uterine rupture despite a stable TOLAC rate. This suggests that health care providers may alter conceptualization of the intrapartum risks of trials of labor after cesarean delivery and subsequently alter their labor management of such patients in the month after severe rupture.
Rupture may lead health care providers to increase their estimate of the underlying risk of rupture and lead them to favor intrapartum cesarean delivery more readily. However, because ruptures will occur even under appropriate care, the rupture itself does not provide additional information to the clinician regarding other women's risk. This cognitive bias is termed the “availability heuristic.”19 Alternatively, health care providers may become more risk-averse after a severe rupture, known as “regret aversion.”20 Finally, health care providers may focus on the similarities between the woman who had the rupture and other women entering labor thereafter. If they neglect the fact that the baseline risk of uterine rupture is extremely low, they may mistakenly conclude that subsequent women's chances of rupture are high because they are clinically similar. This bias is termed the “representativeness heuristic.”21
Because clinical decision-making directly influences patient outcomes, it is important to optimize this process and reduce the effects of any cognitive bias.22 Here, the occurrence of uterine rupture may have led to more repeat cesarean deliveries, likely unnecessarily. By adhering closely to clinical guidelines, including guidelines on the management of labor in women with a previous cesarean delivery, such an effect should not persist. However, cognitive biases are pervasive and difficult to avoid and may warrant additional attention to minimize their effects.5 Educating decision-makers about such biases may be effective at reducing their effects on decision-making.22
Our study had several limitations. First, our definition of severe uterine rupture could not identify ruptures resulting in neonatal morbidity or mortality because this information was not included on the maternal record. Arguably, ruptures resulting in neonatal injury or death would have an even larger effect on practice and decision-making. If so, the effect we estimated in our study may be an underestimate of the true effect.
We identified labor using a previously published algorithm of ICD-9-CM codes. It is possible that some labors were missed, which would lead to an underestimate of the TOLAC rate and an overestimate of the trial of labor success rate. Any bias in the measurement of labor should have no effect on the effect estimate because such bias would have affected deliveries occurring both prerupture and postrupture in all hospitals. Furthermore, by measuring labor occurrence rather than labor intention, the labor rate will also include a small proportion of women who entered labor but had intended a repeat cesarean delivery. However, this subgroup's size should not vary at a hospital prerupture and postrupture and therefore should not affect our findings.
We chose to conduct our analysis on the hospital rather than the clinician level, because it is impossible to know whether the clinician coded on a delivery record had managed all of a woman's care, solely performed an emergent cesarean delivery, or been involved in some other role.
Only admission month is coded, implying that deliveries in the month of the rupture cannot be temporally ordered around the event. Thus, changes occurring only in the very short term may be undetected in our analysis. For the TOLAC rate, the additional concern is of reverse causality in the first month, because a higher TOLAC rate may have led to a uterine rupture. Thus, we focus our interpretation on the months after the uterine rupture.
Our study allowed us to study medical decision-making in situ. We applied a rigorous design that allowed better confounding control than statistical adjustment methods. Here, confounding can persist only if there are forces that affect the hospitals that have uterine rupture at the same time as the rupture and do not also affect the hospitals without the rupture, which is improbable. The other threat of confounding is if hospitals with and without uterine rupture had different preuterine rupture rates of the outcomes. We investigated this assumption and found no evidence suggesting a violation.
Our results suggest that recent adverse events may affect medical decision-making and increase health care providers' hesitancy to prolong labor in women with a previous cesarean delivery. By recognizing how adverse events can affect risk evaluation, health care providers can increase their awareness of these cognitive biases and move toward optimal decision-making in situations with high uncertainty.
1. Lydon-Rochelle M, Holt VL, Easterling TR, Martin DP. Risk of uterine rupture during labour among women with a prior cesarean delivery. N Engl J Med 2001;345:3–8.
2. Guise JM, McDonagh MS, Osterweil P, Nygren P, Chan BK, Helfand M. Systematic review of the incidence and consequences of uterine rupture in women with previous caesarean section. BMJ 2004;329:19–25.
3. Macones GA, Cahill AG, Stamilio DM, Odibo A, Peipert J, Stevens EJ. Can uterine rupture in patients attempting vaginal birth after cesarean delivery be predicted? Am J Obstet Gynecol 2006;195:1148–52.
4. Grobman WA, Lai Y, Landon MB, Spong CY, Leveno KJ, Rouse DJ, et al.. Prediction of uterine rupture associated with attempted vaginal birth after cesarean delivery. Am J Obstet Gynecol 2008;199:30.e1–5.
5. Croskerry P. The importance of cognitive errors in diagnosis and strategies to minimize them. Acad Med 2003;78:775–80.
6. Agency for Healthcare Research and Quality. HCUP Nationwide Inpatient Sample (NIS). Healthcare cost and utilization project (HCUP). 1998–2010. Agency for Healthcare Research and Quality. Available at: http://www.hcup-us.ahrq.gov/nisoverview.jsp
7. Healthcare Cost and Utilization Project. HCUP quality control procedures. 2014. Available at: www.hcup-us.ahrq.gov/db/quality.jsp
. Retrieved August 15, 2014.
8. Healthcare Cost and Utilization Project. HCUP NIS related reports. 2014. Available at: www.hcup-us.ahrq.gov/db/nation/nis/nisrelatedreports.jsp
. Retrieved August 15, 2014.
9. 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.
10. Centers for Disease Control and Prevention (CDC). Use of hospital discharge data to monitor uterine rupture—Massachusetts, 1990–1997. Atlanta (GA): 2000. p. 245–8. Available at: http://www.cdc.gov/mmwr/pdf/wk/mm4912.pdf
. Retrieved April 1, 2014.
11. Joseph KS, Fahey J; Canadian Perinatal Surveillance System. Validation of perinatal data in the discharge abstract database of the Canadian institute for health information. Chronic Dis Can 2009;29:96–100.
12. Uddin SG, Simon AE. Rates and success rates of trial of labor after cesarean delivery in the United States, 1990–2009. Matern Child Health J 2013;17:1309–14.
13. Angrist JD, Pischke JS. Parallel worlds: fixed effects, difference-in-differences and panel data. Mostly harmless econometrics. Princeton (NJ): Princeton University Press; 2009. p. 221–47.
14. Gould W. Interpreting the intercept in the fixed-effects model. StataCorp; 2013. Available at: http://www.stata.com/support/faqs/statistics/intercept-in-fixed-effects-model/
. Retrieved August 16, 2014.
15. StataCorp. Stata statistical software: release 12. College Station (TX): StataCorp LP; 2011.
16. R Core Team. R: a language and environment for statistical computing. Vienna (Austria): R Foundation for Statistical Computing; 2013.
17. Houchens R, Elixhuaser A. Final report on calculating Nationwide Inpatient Sample (NIS) variances, 2001. HCUP method series report #2003-02. 2005. Available at: http://www.hcup-us.ahrq.gov/reports/methods/CalculatingNISVariances200106092005.pdf
. Retrieved March 7, 2014.
18. Lipsitch M, Tchetgen ET, Cohen T. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology 2010;21:383–8.
19. Tversky A, Kahneman D. Availability: a heuristic for judging frequency and probability. Cogn Psychol 1973;5:207–32.
20. Zeelenberg M. Anticipated regret, expected feedback and behavioral decision making. J Behav Decis Mak 1999;12:93–106.
21. Tversky A, Kahneman D. Judgment under Uncertainty: Heuristics and Biases. Science 1974;185:1124–31.
22. Bornstein BH, Emler AC. Rationality in medical decision making: a review of the literature on doctors' decision-making biases. J Eval Clin Pract 2001;7:97–107.