(C) PLOS One This story was originally published by PLOS One and is unaltered. . . . . . . . . . . Crude and adjusted comparisons of cesarean delivery rates using the Robson classification: A population-based cohort study in Canada and Sweden, 2004 to 2016 [1] ['Giulia M. Muraca', 'Clinical Epidemiology Unit', 'Department Of Medicine', 'Solna', 'Karolinska University Hospital', 'Karolinska Institutet', 'Eugeniahemmet', 'Stockholm', 'Department Of Obstetrics', 'Gynaecology'] Date: 2022-08 Abstract Background The Robson classification has become a global standard for comparing and monitoring cesarean delivery (CD) rates across populations and over time; however, this classification does not account for differences in important maternal, fetal, and obstetric practice factors known to impact CD rates. The objectives of our study were to identify subgroups of women contributing to differences in the CD rate in Sweden and British Columbia (BC), Canada using the Robson classification and to estimate the contribution of maternal, fetal/infant, and obstetric practice factors to differences in CD rates between countries and over time. Methods and findings We conducted a population-based cohort study of deliveries in Sweden (January 1, 2004 to December 31, 2016; n = 1,392,779) and BC (March 1, 2004 to April 31, 2017; n = 559,205). Deliveries were stratified into Robson categories and the CD rate, relative size of each group and its contribution to the overall CD rate were compared between the Swedish and the Canadian cohorts. Poisson and log-binomial regression were used to assess the contribution of maternal, fetal, and obstetric practice factors to spatiotemporal differences in Robson group-specific CD rates between Sweden and BC. Nulliparous women comprised 44.8% of the study population, while women of advanced maternal age (≥35 years) and women with overweight/obesity (≥25 kg/m2) constituted 23.5% and 32.4% of the study population, respectively. The CD rate in Sweden was stable at approximately 17.0% from 2004 to 2016 (p for trend = 0.10), while the CD rate increased in BC from 29.4% to 33.9% (p for trend < 0.001). Differences in CD rates between Sweden and BC varied by Robson group, for example, in Group 1 (nullipara with a term, single, cephalic fetus with spontaneous labor), the CD rate was 8.1% in Sweden and 20.4% in BC (rate ratio [RR] for BC versus Sweden = 2.52, 95% confidence interval [CI] 2.49 to 2.56, p < 0.001) and in Group 2 (nullipara, single, cephalic fetus, term gestation with induction of labor or prelabor CD), the rate of CD was 37.3% in Sweden and 45.9% in BC (RR = 1.23, 95% CI 1.22 to 1.25, p < 0.001). The effect of adjustment for maternal characteristics (e.g., age, body mass index), maternal comorbidity (e.g., preeclampsia), fetal characteristics (e.g., head position), and obstetric practice factors (e.g., epidural) ranged from no effect (e.g., among breech deliveries; Groups 6 and 7) to explaining up to 5.2% of the absolute difference in the CD rate (Group 2: adjusted CD rate in BC 40.7%, adjusted RR = 1.09, 95% CI 1.08 to 1.12, p < 0.001). Adjustment also explained a substantial fraction of the temporal change in CD rates among some Robson groups in BC. Limitations of the study include a lack of information on intrapartum details, such as labor duration as well as maternal and perinatal outcomes associated with the observed differences in CD rates. Conclusions In this study, we found that several factors not included in the Robson classification explain a significant proportion of the spatiotemporal difference in CD rates in some Robson groups. These findings suggest that incorporating these factors into explanatory models using the Robson classification may be useful for ensuring that public health initiatives regarding CD rates are evidence informed. Author summary Why was this study done? The Robson classification system is a World Health Organization (WHO)-endorsed global standard for comparing and monitoring cesarean delivery (CD) rates. This classification does not include important maternal, fetal, and obstetric practice factors known to influence CD rates. The contribution of these characteristics to CD rate comparisons between and within populations has been identified as a key deficiency of this classification scheme and has not been comprehensively quantified. What did the researchers do and find? We conducted a population-based cohort study including 1,951,984 deliveries in Sweden and Canada between 2004 and 2016 to assess differences in CD rates among Robson groups, between countries and over time, with and without adjustment for maternal, fetal, and obstetric practice factors. The effect of adjustment between countries varied by Robson group from having no effect in some groups to explaining up to 61% of the variation in CDs in others. Adjustment for maternal, fetal, and obstetric practice factors explained a substantial fraction of the temporal change in CD rates among some Robson groups in Canada but had little impact on temporal changes in CD rates among Robson groups in Sweden. What do these findings mean? Public health initiatives based on Robson-classified CD rates may be misinformed without a comprehensive consideration of relevant maternal, obstetric practice, and fetal factors. Comprehensive and accurate perinatal data collection beyond the Robson criteria is necessary to ensure policies regarding CD rates are suitably evidence informed and prioritized. Future studies are warranted to evaluate the differences in CD rates in each Robson group in relation to maternal, fetal, and infant morbidity and mortality. Citation: Muraca GM, Joseph KS, Razaz N, Ladfors LV, Lisonkova S, Stephansson O (2022) Crude and adjusted comparisons of cesarean delivery rates using the Robson classification: A population-based cohort study in Canada and Sweden, 2004 to 2016. PLoS Med 19(8): e1004077. https://doi.org/10.1371/journal.pmed.1004077 Academic Editor: Gordon C. Smith, Cambridge University, UNITED KINGDOM Received: January 20, 2022; Accepted: July 15, 2022; Published: August 1, 2022 Copyright: © 2022 Muraca et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: The Swedish Medical Birth Registry is a national dataset and considered to be public property. However, access to the data is given only to researchers with permission from the Swedish Ethical Review Authority and after approval of the research plan by a Swedish National Board of Health and Welfare data manager. The data request may be sent to The Swedish National Board of Health and Welfare (https://www.socialstyrelsen.se/). The Canadian data were acquired from Perinatal Services British Columbia. Access to the Canadian data is available to researchers, health managers and decision-makers via a formal data request process administered by Perinatal Services British Columbia. More information on the data holdings and the request process is available at http://www.perinatalservicesbc.ca/our-services/data-surveillance/perinatal-data-registry/data-requests. All inferences, opinions, and conclusions drawn in this publication are those of the authors, and do not reflect the opinions or policies of Perinatal Services BC. Funding: GMM is the recipient of a Canadian Institutes of Health Research (CIHR) Postdoctoral Fellowship (www.cihr-irsc.gc.ca/); NR, LVL, and OS are supported by the Swedish Research Council for Health, Working Life and Welfare (Forskningsrådet om Hälsa, Arbetsliv och Välfärd; www.forte.se) NR: 4-2702/2019; LVL: 2019-01275; OS: 2013-09298). NR is also supported by the CIHR. OS is also supported by the Strategic Research Program in Epidemiology at Karolinska Institutet (www.ki.se/en/research/sfoepi-grants). SL is supported by a Scholar Award from the Michael Smith Foundation for Health Research (www.msfhr.org); and KSJ is supported by an Investigator award from the BC Children’s Hospital Research Institute (www.bcchr.ca). None of the authors have financial relationships with any organisations that might have an interest in the submitted work in the previous three years, nor do they have other relationships or activities that could appear to have influenced the submitted work. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Abbreviations: ARR, adjusted rate ratio; BC, British Columbia; BCPDR, British Columbia Perinatal Database Registry; BMI, body mass index; CD, cesarean delivery; CI, confidence interval; MBR, Medical Birth Register; OR, odds ratio; RR, rate ratio; WHO, World Health Organization Introduction In 2015, the World Health Organization (WHO) endorsed the Robson classification system as a global standard for comparing and monitoring cesarean delivery (CD) rates across populations and over time [1]. As a result, the use of the Robson classification in evaluating CD trends has expanded over the last decade and become a common tool to inform obstetric practice worldwide [2–7]. This classification scheme stratifies deliveries into 10 mutually exclusive and all-inclusive categories based on 6 obstetric characteristics: parity (nulliparous versus parous), previous CD (yes/no), plurality (single versus multiple fetuses), fetal presentation (cephalic, breech, transverse/oblique), labor onset (spontaneous, induced, prelabor CD), and gestational age (<37 versus ≥37 weeks) [8,9]. The Robson classification system represents an elegant method that uses stratification (to control for confounding) in order to isolate the effect of specific obstetric practices on CD rates. Nevertheless, there are important determinants of spatial and temporal variations in CD rates that are not integrated into the Robson analysis scheme [10,11]. For example, maternal characteristics such as age [12] and body mass index (BMI) [13] are 2 important determinants of CD rates that are not addressed by the Robson strata. Similarly, obstetric practice factors, such as the clinical management of postterm delivery [14], as well as fetal factors, such as position of the fetal head at delivery [15] and fetal size [16], have also been associated with CD but are not considered in the Robson scheme. Inferences about spatiotemporal differences in CD rates may be biased if they are based on patterns within the Robson groups but without appropriate consideration of these extraneous factors that influence CD rates [5]. Statistical adjustment using regression techniques can address the contribution of extraneous determinants (such as maternal age and BMI) so that the true association between obstetric practice and CD rate can be quantified. The few studies that have applied regression to the assessment of CD rates using the Robson framework have been limited by single-center or cross-sectional study designs [17,18], while prior population-based or longitudinal studies lacked information on important risk factors for CD (e.g., labor onset, early-pregnancy BMI, maternal comorbidity, fetal factors) [5–7,19–21]. As a result, the effect of maternal, fetal, and obstetric practice factors (not represented in the Robson classification) on international and temporal comparisons of CD rates using the Robson classification is unknown. We carried out a study to identify subpopulations responsible for differences in CD rates between Sweden and British Columbia (BC), Canada using the Robson classification system. Further, we sought to quantify the contribution of maternal characteristics, obstetric practice factors, and fetal characteristics to variations in the CD rate (a) between countries; and (b) within each country over a 13-year period. Methods This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 STROBE Checklist). The study methodology and analysis plan (S1 Text) were developed prior to data extraction and the actual analysis. We conducted a population-based, retrospective, cohort study including all deliveries (live births and stillbirths) ≥22 completed weeks of gestation in Sweden and BC. Data for the Swedish cohort of deliveries between January 1, 2004 and December 31, 2016 were obtained from the Swedish Medical Birth Register (MBR), a validated, nationwide health register containing more than 98% of all births in Sweden [22]. Data for the Canadian cohort were obtained from the British Columbia Perinatal Database Registry (BCPDR) for the fiscal years 2004/2005 to 2016/2017 (hereafter referred to as years 2004 to 2016). The BCPDR contains information on approximately 99% of births in the province of BC [23]. Validation studies show that the BCPDR is an accurate and comprehensive source of perinatal information [24,25]. Both the MBR and the BCPDR contain detailed demographic and clinical information on all mothers and babies including information on diagnoses and procedures classified using standardized codes [22,23]. The study was restricted to the years 2004 to 2016 when all diagnostic codes conformed to the 10th revision of the International Statistical Classification of Diseases and Related Health Problems in Sweden (ICD-10-SE) [26] and Canada (ICD-10-CA) [27]. Similarly, procedure codes were consistently coded over the study period with the Nordic Classification of Care Measures [28] in Sweden and the Canadian Classification of Health Interventions in Canada [29]. Analysis of CD rates using the Robson classification system Data were grouped into Robson categories and modified from the 10-group system to a 12-group system, used commonly [9] to differentiate women in Groups 2 and 4 between those who had an intrapartum CD after induction (defined as 2a and 4a) and those who had a CD before labor onset (2b and 4b; Table 1). Women with twin (or higher order) pregnancies were only counted once; the mode of delivery of the second twin was used in the analysis. The overall CD rate, the relative size of each Robson group, and the absolute contribution of each group to the overall CD rate were compared between the Canadian and Swedish cohorts using the Standard Robson Classification Report Table. We also assessed the data quality, population distributions, and Robson group-specific CD rates against WHO-recommended benchmarks using standardized criteria [1]. Temporal trends in CD rates were described by Robson group and assessed using the Cochran–Armitage test. Temporal trends in the relative contribution of each Robson group to the overall CD rate were also described. PPT PowerPoint slide PNG larger image TIFF original image Download: Table 1. Distribution of maternal, obstetric practice, and fetal/infant characteristics among deliveries in Sweden and BC, Canada, 2004–2016. https://doi.org/10.1371/journal.pmed.1004077.t001 Contribution of maternal, fetal, and obstetric practice factors to spatial differences in CD We categorized potential determinants of CD into 4 groups: maternal characteristics, maternal comorbidity, factors related to obstetric practice, and fetal factors. The maternal characteristics included age (<20, 20 to 24, 25 to 29, 30 to 34, 35 to 39, 40 to 44, ≥45 years), parity (in parous groups; 1, 2, 3 to 4, ≥5), early-pregnancy BMI classified in kg/m2 as normal (18.5 to 24.9), underweight (<18.5), overweight (25.0 to 29.9), obesity class 1 (30.0 to 34.9), class 2 (35.0 to 39.9), and class 3 (≥40), and maternal smoking during pregnancy. In Sweden and BC, maternal BMI in early pregnancy was calculated from self-reported height and weight or provider assessment at the first antenatal visit, which typically occurs within the first 14 weeks of gestation [22,23]. We also adjusted for maternal comorbidity, which included chronic hypertension, early-pregnancy diabetes, in vitro fertilization, and preeclampsia/eclampsia. Epidural anesthesia and postterm delivery (≥42 completed weeks gestation) were the obstetric practice factors identified as potential determinants of CD. We chose these 2 obstetric practice factors because (a) we hypothesized they would have substantial variability between the 2 countries and over time; and (b) this information was reliably captured in both the MBR and the BCPDR. Finally, we also adjusted for fetal/infant characteristics: position of the fetal head at delivery (in groups with a cephalic-presenting fetus; occiput posterior versus occiput anterior), birth weight, head circumference, and congenital anomaly. The distribution of these risk factors for CD were compared between Sweden and BC using frequencies and standardized differences [30]. Modified Poisson regression with robust standard errors was used to estimate the effect of maternal characteristics, comorbidity, fetal characteristics, and obstetric practice factors on the differences between CD rates in each Robson group in BC versus Sweden. These analyses were performed on individual pregnancies, which made it possible for a woman to contribute more than 1 pregnancy in the Robson groups that included multiparous women. Therefore, we used generalized estimating equations (with the mother’s identification as a cluster and assuming an exchangeable correlation structure), with adjustment for the possible correlation in outcome that could be introduced by subsequent births by the same mother. The difference between CD rates in the 2 countries that was explained by differences in risk factors was assessed by comparing the crude rate ratio (RR) and adjusted RR (ARR) in each Robson group and on the absolute scale, by comparing the crude and adjusted CD rate in BC versus Sweden. Missing BMI values were modeled using the missing-indicator method. Records with missing values for other variables were excluded as they did not exceed 3% of the total study population; thus, we conducted a complete-case analysis with respect to these variables (Fig 1). PPT PowerPoint slide PNG larger image TIFF original image Download: Fig 1. Flow diagram showing derivation of cohort. https://doi.org/10.1371/journal.pmed.1004077.g001 We calculated RRs instead of odds ratios (ORs) in our spatial analyses of Robson group-specific CD rates, as the RR is more readily interpretable and the OR does not correspond to the RR when the event rate is relatively high. However, the RR is artefactually constrained when the baseline CD rate is high (e.g., cannot exceed 2 if the baseline CD rate is 55%). The OR is not susceptible to this constraint, and therefore, we also calculated ORs and 95% confidence intervals (CIs) for Robson group-specific CD rates in BC versus Sweden. Contribution of maternal, fetal, and obstetric practice factors on temporal changes in CD We quantified the temporal trend in CD rates in Sweden and BC by year and also by period (2004 to 2007, 2008 to 2010, 2011 to 2013, and 2014 to 2016) in the 3 Robson groups with the largest contributions to the overall CD rate. The magnitude of the change in CD rates over time was estimated using RRs and 95% CIs. To evaluate the contribution of changes in maternal, fetal, and obstetric practice factors to temporal changes in CD rates in these Robson groups, we first compared the temporal trends in the risk factors listed above by period (2014 to 2016 versus 2004 to 2007) using RRs and 95% CIs (the linear trend by year was assessed using the Cochran–Armitage test). Secondly, since changes in maternal characteristics (e.g., increases in advanced maternal age) and maternal comorbidity (e.g., diabetes) can lead to changes in obstetric practice factors (e.g., increases in epidural use), we used sequential log-binomial models to identify the effect of each group of factors on the relationship between year or period and CD. The sequential approach was carried out by fitting a series of models with additional groups of factors added to each model in the sequence outlined above to quantify the contribution of each group of factors to CD trends over time. Temporal trends were estimated both for individual year and period through separate regression models. Sensitivity analyses Antenatal and maternity care in Sweden and Canada is universal and free; however, the model of maternity care differs substantially between countries. In Sweden, midwives are the primary care providers for women with low-risk pregnancies [31], whereas in BC, midwives are involved in maternity care for about 25.3% of deliveries [32]. To address differences in maternity care models in each country, our first sensitivity analysis compared deliveries in Sweden with those in BC restricted to women with midwifery care. We performed 2 additional sensitivity analyses to address missing information about early-pregnancy BMI (13.8% of the study population; 7.9% and 28.3% among deliveries in Sweden and BC, respectively). First, we conducted complete-case regression analysis by excluding deliveries without data on BMI from the regression models. Second, missing values on BMI were estimated using a Markov Chain Monte Carlo multiple imputation method. Results from 10 multiple imputation cycles were combined with the use of PROC MIANALYZE in SAS. The a priori level of statistical significance was set at a 2-sided p-value < 0.05 for all analyses. The Cochran–Armitage test was used to test for significance of linear trend, and the Wald chi-square test was used to test significance of differences in CD rates in the crude and adjusted regression models. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, North Carolina, United States of America). The study was approved by the Clinical Research Ethics Board at the University of British Columbia (H14-00674) and the Research Ethics Committee at Karolinska Institutet, Stockholm, Sweden (No. 2008/1182-31/4). Discussion We applied the Robson classification to 1,951,984 deliveries between 2004 and 2016 in Sweden and BC, Canada. The effect of controlling for maternal, fetal, and obstetric factors on the spatial comparisons differed by Robson group and ranged from no effect (e.g., among deliveries with breech presentation; Groups 6 and 7) to explaining up to 5.2% of the absolute 8.6% difference in CD rates in BC versus Sweden in Group 2 (nulliparous women with a term, cephalic fetus with induced labor, or a CD without labor). Nonetheless, wide spatial differences in CD rates persisted even after adjustment, especially in nulliparous women with a term, cephalic fetus with spontaneous labor, and women with a previous CD (Robson Groups 1 and 5, respectively). Adjustment for maternal, fetal, and obstetric practice factors explained a substantial fraction of the temporal change in CD rates in BC in Robson Groups 1 and 2a but had little impact on the temporal changes in CD rates observed among specific Robson groups in Sweden (such as Groups 2a and 5). Several studies of obstetric populations in high-income settings have found that Robson Groups 1, 2a, and 5 contribute the most to overall CD rates [3–6], and the distribution of deliveries across Robson groups in both Sweden and BC is consistent with expected relative group sizes [33]. To our knowledge, there are no studies that have quantified international variation in CD rates using the Robson classification with and without adjustment for differences in maternal, fetal, and obstetric practice factors. One analysis that compared CD rates across 4 Nordic countries using the Robson classification adjusted solely for maternal age and did not report the variation in crude and adjusted estimates of CD [5]. The few studies that have combined Robson stratification and regression to estimate temporal variation in CD rates have reported mixed results. Similar to our findings, attenuation in CD rates over time after case-mix adjustment was observed among women in Robson Groups 1 and 3 in analyses from northern Italy [17,18]. On the other hand, a population-based study in France found that an increase in CD in Robson Group 1 from 2003 to 2010 did not change after adjustment for risk factors for CD, although this may be explained by limitations in available information as the models did not include BMI, maternal comorbidity, or fetal characteristics [19]. The strengths of our study include the ability to examine a large number of deliveries owing to the use of 13 years of data from population-based birth registers in 2 countries. In addition, we were able to include important determinants of CD in our analyses, such as maternal comorbidity, maternal BMI, fetal macrosomia, and position of the fetal head at delivery that have been excluded in previous analyses. The data sources used have been shown to be accurate in validation studies [24,25], and the proportion of women left uncategorized by Robson classification due to missing data was low (<1.0%). The limitations of our study include potential data transcription errors and omissions in coding, which are inevitable in large health databases. Further, 13.8% of women (7.9% in Sweden and 28.3% in BC) in our study had missing values for early-pregnancy BMI and we included these women in our main models using a missing indicator approach. However, complete-case analyses (excluding cases with missing BMI values) and analyses including imputed values for missing BMI yielded similar results. Since we do not know if the BMI distribution was different among those with missing versus available BMI information, it is challenging to speculate on the direction and magnitude of bias introduced by missing data for BMI [34]. Confirmation of our findings in a cohort with complete BMI data would be valuable. The relatively large size of Group 4b (multiparous women with prelabor CD) in Sweden may indicate misclassification of women with a previous CD (i.e., they should have been assigned to Group 5); alternatively, this may indicate higher rates of CD on maternal request due to a variety of reasons, such as previously traumatic or prolonged labor. Finally, we were not able to account for important intrapartum details such as indication for CD, duration of labor, details regarding oxytocin augmentation, and fetal surveillance because our data sources lacked such information. We quantified the crude and adjusted temporal trend in CD in a commonly used subdivision of Robson Group 2 restricted to women who had an induction (Group 2a) to increase the homogeneity of this group before assessing the contribution of maternal, fetal, and obstetric practice characteristics on temporal trends in CD. However, it is important to note that the analysis of any subdivision alone may be misleading without a consideration of the other subdivisions that compose the complete group (Group 2b in this case). In our analyses, the exclusion of Group 2b is unlikely to have had a substantial effect on temporal changes in the CD rate in Group 2 since the relative size of Group 2b was small (1.1% and 1.2% in Sweden and BC, respectively) and remained small throughout the study period. The relative size and high rate of CD in Group 2a are the reasons for this group’s position as one of the 3 largest contributors to the overall CD rate in both Sweden and BC. While our study explained spatiotemporal variations in CD rates, it did not address the perinatal and maternal morbidity and mortality that was caused or prevented by the observed differences in CD rates. Future studies are required to evaluate the differences in CD rates in each Robson group within the context of maternal, fetal, and infant morbidity and mortality. Our findings reinforce that, although often misinterpreted, the Robson classification was not meant to be an endpoint of CD comparison, but rather a starting point or initial framework within which determinants and outcomes of CD can be analyzed [10]. These results also highlight the importance of comprehensive and accurate perinatal data collection to ensure public health initiatives regarding CD rates are suitably evidence informed and prioritized. The maternal, obstetric practice, and fetal/infant characteristics included in our analyses were not exhaustive, and it is possible that the inclusion of other factors not represented in the Robson classification scheme (e.g., duration of labor, indication for CD) could have even larger impacts on CD rates. The simplicity and efficiency of the Robson classification system has resulted in its endorsement by international health organizations and expansive uptake worldwide. However, our analyses show that maternal, fetal, and obstetric practice factors not included in the Robson classification explain a significant proportion of the spatiotemporal difference in CD rates in some Robson groups and should be incorporated into explanatory models evaluating CD rates in populations. Public health initiatives based on Robson-classified CD rates may lead to erroneous attributions of variation in CD rates to differences in obstetric practice without a comprehensive consideration of relevant maternal, obstetric, and fetal factors. [END] --- [1] Url: https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1004077 Published and (C) by PLOS One Content appears here under this condition or license: Creative Commons - Attribution BY 4.0. via Magical.Fish Gopher News Feeds: gopher://magical.fish/1/feeds/news/plosone/