Risk-adjusted cesarean section rates for the assessment of physician performance in Taiwan: a population-based study

The rise in cesarean section (CS) delivery rates in Western nations and many developing countries has become an issue of major public health concern. Over the past decade, about one-third of all births nationwide in Taiwan were delivered by CS, with the proportion having fluctuated between 32.39% and 34.47% [1]; indeed, cesarean deliveries have now become the most common of all major surgical procedures currently being undertaken under the country's National Health Insurance (NHI) program. Although considerable controversy remains as to whether elective CS can provide any benefits, either to the newborn child or to the parturient [2, 3], the medical costs associated with the high CS rate in Taiwan have nevertheless become a tremendous financial burden on the island's finite healthcare resources.

In order to ensure the optimum level of quality care for pregnant women and their infants, the Bureau of NHI (BNHI) has placed the CS rates of all providers under close scrutiny, adopting actual quarterly average rates by region and/or hospital accreditation levels, as a means of evaluating hospital/physician obstetric practice patterns [4]. However, numerous studies have consistently provided evidence to show that differences in case mixes among patient populations of different healthcare providers in a wide range of settings may lead to bias in the judgment of whether a provider truly has a high CS rate. Those studies suggest that prior to any comparison being undertaken between different providers, all CS rates should be adjusted for risks [5–9]. A pioneering study in Taiwan by Hsu et al. [10] pointed out the unfairness of using crude CS rates as a means of comparing hospital performance within the Taipei Municipal Hospital System, and also the inappropriateness of drawing conclusions based on crude rates on the link between hospital practice patterns and quality of care.

Prior research into risk-adjusted CS rates tended to focus on profiles of hospital CS rates; thus, to the best of our knowledge, no study has yet been carried out on case mix adjustment based upon profiles of individual physicians' CS rates. This clearly hinders the efforts by policymakers to develop effective interventions aimed at lowering the CS rates at the physician level. This study therefore took advantage of a unique dataset which merges birth certificate data with the NHI claims data, to determine the impact that adjustment for patient-specific risk factors has on CS rates among different physicians in Taiwan. The findings of this study will not only facilitate more-accurate comparisons of CS rates among physicians, but can also help policymakers target their interventions at specific physicians.

In total, 258 physicians delivered 40,973 babies in the first quartile, 257 physicians delivered 54,464 babies in the second quartile, 258 physicians delivered 45,456 babies in the third quartile, and 258 physicians delivered 31,618 babies in the fourth quartile.

Similarly, the χ² test in Table 2 also indicated that with the notable exception of those cases involving a prolapsed cord, on presentation of each of the risk factors, the decision to perform a surgery significantly increased from the lower quartile to the higher quartiles.

All of the risk factors included within the univariate analysis were retained in the stepwise regression analysis for adjusting the CS rates. The results for model 1 in Table 3 indicate that after controlling for all the risk factors within the model, the adjusted odds ratio for most of the obstetric- and pregnancy-related complications were of considerably greater magnitude compared to the unadjusted risks, while those of premature labor and maternal age were reduced. However, with the exceptions of premature rupture of the membrane and parity, the direction of the relationship between the CS rate and all other variables remained unchanged. Nulliparous women and women with premature rupture of the membrane were no longer protective against CS, but instead, incurred increased risk.

The results of the stepwise logistic regression, using parturients delivered by physicians in quartiles 1, 2, and 3, are presented in model 2 of Table 3. Within the model, with the exception of cervical incompetence, the direction of the relationship between the CS rate and all other risk factors remained unchanged, with the adjusted odds ratio for most of the obstetric- and pregnancy-related complications being of similar magnitudes when compared to the results of model 1. Cervical incompetence was excluded from the prediction of the CS rate in model 2.

Both the likelihood ratio and the Hosmer-Lemeshow test were significant in models 1 and 2, at p < 0.0001, indicating that while the risk factors considered in both models were significant, in terms of predicting the probability of a CS, the fit was nevertheless imperfect. However, the number of concordant pairs was very high, with the respective c-statistic for models 1 and 2 respectively being 0.870 and 0.872, thereby indicating that the predictive accuracy of the logistic model was good in both models.

The overall results from model 1 show that of 1031 physicians, 448 physicians (43.45%) had CS rates which were above the predicted CI; as expected, the proportion of physicians whose CS rates were above the predicted 95% CI increased for the higher quartiles. In the lowest quartile, no physician's actual rate was above the predicted CI, while in the highest quartile, the actual rates of as many as 218 physicians (84.85%) were above the predicted CI.

Of the 448 physicians whose actual rates were greater than the predicted CI, 230 physicians (51.34%) were not in the highest quartile of actual rates; on the other hand, of the 446 physicians whose actual rates were below the predicted CI, 212 physicians (47.53%) were not in the lowest quartile.

When a stricter standard was applied to the assessment of physician performance, the results from model 2 in Table 4 indicated that compared to the results of model 1, there were increases in the numbers of physicians whose CS rates were above the predicted CI in each quartile. Overall, as many as 60% of the physicians were found to have higher CS rates than the predicted CI.

In addition, within model 2, CS rates of 24 physicians in the lowest quartile were now higher than the predicted CI. Moreover, the actual rates of all the physicians in the highest quartile were above the predicted CI, and indeed, the CS rates of no physicians in either quartile 3 or quartile 4 were below the predicted CI.

CS rates have been used for a decade or so as a means of comparing obstetric practices across different providers, with many studies throughout this period having documented the need for risk adjustment for fairer comparisons among providers [5–10]. In 2000, the American College of Obstetricians and Gynecologists (ACOG) recommended that all CS rates should be risk-adjusted (case-mix adjusted) prior to any comparisons being made [15]; however, while this may be desirable, in many areas or countries, such adjustment is usually not feasible, largely due to the lack of adequate data sources.

The CS rate in Taiwan has hovered between 32% and 34% since implementation of the NHI program in 1995. Although lower than the respective CS rates of 36% and 40% found in Brazil and Chile, the CS rate in Taiwan is nevertheless much higher than the 15% limit recommended by the World Health Organization [16].

Inappropriate CS procedures increase maternal and neonatal morbidity and healthcare costs; however, the BNHI has continually adopted crude CS rates as its measure of quality in obstetrics management, and this can clearly lead to unfair assessments of those providers who are serving high-risk populations, since these providers will obviously have higher CS rates, and will therefore appear to be providing a lower quality of care.

By examining all 172,511 births in Taiwan in 2003, the present study took advantage of a unique population-based dataset which merges birth certificate data with the NHI claims data of this population to show how adjustments of patient risk factors can significantly change the overall assessment of a physician's obstetrics performance.

Our data specifically demonstrate that more than half of all physicians with inappropriately high rates were not in the highest quartile, and that these would be overlooked if the assessment was only targeted at those physicians in the highest quartile of unadjusted rates. Conversely, we also found that some physicians (40 of 258 physicians) in the highest quartile were actually providing obstetric care that was appropriate to the risk. However, our data also demonstrate that when stricter standards were applied to the evaluation of physician performance, a discernibly higher proportion of physicians were seen to be producing inappropriately high CS rates. In particular, we suggest that all of the physicians in quartile 4 should be placed under close scrutiny.

The CS prediction model adopted for this study was the logistic regression method, which is very similar to that used in analyses undertaken in the US and Europe [5–9].

In general terms, the factors associated with CS are largely similar between various regions or countries; however, due to differences in the data sources or specific patient populations under examination, the odds ratios associated with each of the variables may vary.

In contrast to many prior studies [6–8], our results show that prior to adjustment, multiparous women are at higher risk of cesarean delivery, mainly attributable to their prior history of a CS. Our data show that of the 87,777 births to parous women, 31,747 (36.17%) were delivered by CS, and 17,594 (55.42%) of these were repeat CSs. The data also reveal that of the 84,734 nulliparous women, 28,332 (33.44%) underwent a CS.

There are limitations to this study which should be noted. First, the possibility may exist for differential misclassification in obstetric/pregnancy risks for CS and vaginal delivery, since women who deliver vaginally tend to underreport complications much more than do those who undergo a CS. Second, the analysis in this study relied upon linked administrative data; however, some of the prior studies evaluating the use of a CS indicated that the use of administrative data may yield similar or higher discrimination than models based upon medical records [17, 18]. It has been noted that across different medical institutions, medical record coders may have inconsistent coding practices, while clinicians may have very different views on diagnosis. Third, we do not have access to information on whether an obstetrician collaborated with midwives or family physicians who themselves do not have training/privileges to perform a CS. Such collaboration may have inflated an obstetrician' s apparent CS rate and elevated him/her into the uppermost percentile. Finally, our study was based on a dataset of linked birth certificate and NHI claims data for a single country, which may not be representative of other regions or countries. Thus, we recommend caution when interpreting these results, or attempting to apply them to other regions or countries.

Our study arrived at the same general conclusions reported in the literature in both the US and Europe: that is, comparisons of unadjusted CS rates might not validly reflect the quality of obstetric care being provided, and may ultimately lead to biased judgments by health care purchasers. It is worth noting that the same issues which apply to obstetrics care in the US and Europe also apply to non-Western cultures, such as Taiwan.

While it is clearly important to be able to identify the risk factors associated with CS, in order to optimize the utilization of both resources and personnel, risk adjustment using average performance as a benchmark is not necessarily a true measure of quality [19]. Our study results also show that when we changed the standard of quality assessment, the evaluation results changed.

This study represents the first attempt within a developing country to use a population-based dataset as a means of calculating risk-adjusted CS rates in order to identify those physicians who are above or below the average practice in a high CS rate setting, which prevails in Taiwan. We believe that this represents a cornerstone with regard to the design of effective strategies for the monitoring of the 'epidemic' CS rates currently existing in Taiwan.