Socioeconomic differences in caesarean section – are they explained by medical need? An analysis of patient record data of a large Kenyan hospital

Our study was conducted in the public wing of a large academic referral hospital in Kenya. The maternity department consisted of an antepartum ward (29 beds), labor ward (18 beds), postpartum ward (35 beds), neonatology unit (max. 160 beds), and a hostel (45 beds, 24-h observation of post-partum women without complications). Maternity care in the hospital, as in the rest of Kenya, was officially free as of June 1st, 2013.

All women who gave birth in the public wing of the hospital between 1 January and 31 December 2014 were included in our study. Excluded were births of fetuses with an estimated weight below 650 g (21 deliveries) because C-sections were not performed in this hospital in these women. All women who gave birth at the hospital, were registered in an electronic Delivery Database after discharge. The Delivery Database contained a digitalized version of parts of the manual patient file that all women received, and included the following variables: patient number, admission date and time, discharge date, maternal age, parity, maternal occupation, ICD codes, multiple gestation, mode of birth, outcome mother, outcome infant, birth weight, name of ANC clinic the mother attended, referring facility, and reason for referral. Medical record officers digitalized the manual files and coded data according to the International Classification of Diseases and Related Health Problems 10th edition (ICD-10) and the International Classification of Procedures in Medicine [11, 12]. C-sections were also registered in a Surgery Database: a digitalized version of the Surgery Book, which contained all surgeries at the maternity department. We obtained anonymised versions of the Delivery Database and Surgery Database for our analyses. We linked the Delivery Database and Surgery Database on the basis of patient number, in combination with other variables in the databases where needed. To verify the accuracy of the electronic databases, we conducted a detailed review of a random selection of manual files (case notes). Hundred fifty women who gave birth (either vaginally or by CS) and 50 women who gave birth specifically by CS in MTRH in 2014 were randomly selected from the Delivery Database. For 131 (82 vaginal, 49 CS) out of the 200 women, we were able to retrieve the manual file. A detailed review of the manual files was done by LvdS and SS to assess whether or not women had a documented CS indication, using procedures described below.

The study outcome was defined mode of birth (C-section vs. vaginal birth). Records were included as C-section if the Delivery Database noted C-section as mode of birth and/or included an ICD code for C-section, and/or if the woman was registered in the Surgery Database as having delivered via C-section. All other records were included as vaginal births.

Socioeconomic position was defined on the basis of occupation of the woman giving birth. Occupation was registered in nine categories, which we summarized into four categories as follows: housewives (housewife, unemployed), small business (small business, casual laborer, farmer), professional (professional, government employee, private employee), and student (student/pupil). Maternal age was registered in years and categorised as follows: < 16, 16–20, 21–25, 26–30, 31–35, 36–40, and > 40 years.

The Robson classification has been developed to compare C-section rates across hospitals, and provides a starting point for accessing hospital-based C-section rates. To categorize women according to the Robson classification [13], we used information on parity, gestational age (in weeks), presentation (cephalic/ breech/ other non-cephalic presentation), number of fetuses, and previous C-section (yes/no) from the Delivery Database. As we had no information on spontaneous vs. induced labor, we used an adapted version of the Robson classification (see Table 6). 2845 women (23%) could not be classified into a Robson category because of missing information on gestational age. We developed two additional groups (Group 11: All nulliparous women, singleton cephalic, gestational age unknown; Group 12: All multipara, singleton cephalic, gestational age unknown) to address this problem.

Finally, parity was not always consistently recorded – sometimes as parity before birth, sometimes as parity after birth. As the proportion of women with parity recorded as zero led to an implausibly low estimate of the proportion of nulliparous women (5.5%), we included all women with parity recorded as zero or one as nulliparous, which might have led to an over-estimation of nulliparous women and a dilution of the effect of parity on mode of birth.

For each woman, we determined whether she had an indication for C-section according to the clinical guidelines of the hospital [14]. We obtained these guidelines from the maternity department and translated these into ICD codes and other information necessary to determine C-section indication. Table S1 provides a full overview of the hospital guidelines, information necessary and information available to determine clinical indication post-hoc. We used the ICD codes and other information in the Delivery and Surgery Databases to determine if a woman had a C-section indication according to the guidelines. The information in the databases was not always detailed enough to conclusively determine if a woman had a clinical indication. For example, fetal anomaly incompatible with spontaneous vertex birth (SVB) is a C-section indication according to the hospital guidelines. An ICD code for foetal abnormality exists, but this does not clarify whether the anomaly was incompatible with SVD. As another example, previous C-section is an indication for C-section according to the hospital guidelines in the case of two or more previous C-sections. The ICD codes contained whether a woman had previous C-section, but did not provide details on the number of previous C-sections. In such cases, where part of the information was missing, we used the precautionary principle and assumed that the woman had a C-section indication. For comparative purposes, we also determined for each woman if she had a C-section indication when using the Kenyan national guidelines [15, 16], the Dutch [17‐24] and English [25‐29] guidelines.

First, we calculated the C-section rate for the total population and by socioeconomic position and other background characteristics. Then, we calculated the percentage of women with C-section indication and examined the determinants of C-section indication using logistic regression analysis. Next, we calculated the C-section rate among women with and without C-section indication and the percentage of C-section deliveries without clinical indication. Then, we examined determinants of C-section using logistic regression analysis. Using multivariable logistic regression analyses, we examined if socioeconomic inequalities in the odds of C-section were explained by differences in medical need for a C-section (defined as C-section indication according to the hospital guidelines), previous C-section, maternal age and parity. We also divided the population into the Robson groups, and examined if there were socioeconomic inequalities in C-section rate within Robson groups. We analyzed the data using Stata 13 (Stata, College Station, TX, USA).

Our study shows that unnecessary primary C-sections and near universal repeat C-sections play an important role in explaining both the overall C-section rate and socioeconomic inequalities in C-section. Socioeconomic inequalities in C-section were moderate in the Kenyan referral hospital that we studied. These inequalities were foremost explained by a higher level of indicated C-sections -mostly related to previous C-section- among high SEP women. Nearly all women with a previous C-section had a repeat C-section for their subsequent pregnancy, and 3 in 10 C-sections had previous C-section as indication. But over-use of C-sections based on hospital guidelines was also substantial, and seen in all socioeconomic groups: over 4 in 10 primary C-sections had no documented indication. Higher over-use among high SEP women explained around one third of socioeconomic inequalities in primary C-sections. Socioeconomic differences in age and parity further contributed to explaining inequalities in indicated and unindicated C-sections. Our study suggests that prevention of unnecessary primary C-sections and promotion of safe trial of labor with close monitoring in women with a scarred uterus could help curb the C-section epidemic and help reduce socioeconomic differences in C-section.

Our analyses suffered from some problems. First, we used anonymised versions of the Delivery and Surgery Databases, which complicated patient identification and linking of the databases due to typos in patient numbers. 286 C-section records in the Delivery Database (2% of all deliveries, 13.2% of C-section deliveries) could not be matched to a Surgery Database record, and 148 C-section records in the Surgery Database (1% of all deliveries, 6.8% of C-sections) could not be matched to a Delivery Database record. To avoid over-estimating the C-section rate, we used the Delivery Database as basis for our analyses, rather than including all unlinked records. If we also had included the 148 unlinked records from the Surgery Database, the C-section rate would have been 17.1% instead of 16.5%.

Secondly, the analyses suffered from some uncertainty in determining clinical indication for C-section because only a limited set of variables was available in the electronic databases. Our use of the precautionary principle, as explained in the methods section, will probably have led to an overestimation of the proportion of caesarean deliveries with an indication. Importantly, multiple previous C-sections constituted a C-section indication according to the hospital guidelines, while information on the number of previous C-sections missed in the electronic records. Use of the precautionary principle led to the classification of all previous C-sections as indication, while many will have been first repeats. Also, we were not able to take into account clinical judgement not recorded in the electronic database. This may have led both to an under-estimation or over-estimation of the proportion of caesarean deliveries with a clinical indication. Detailed analysis of a random selection of the full manual files of C-section patients confirms our estimate of C-section over-use. Furthermore, there is no indication that an over- or underestimation of clinical indication for C-section would be differential by SEP.

Finally, maternal occupation as recorded in the patient files is a rough proxy for SEP, arguably with measurement error both in determining occupation itself and in classifying occupation into categories. There is no indication that such measurement error was systematic. Combined with the broad occupational categories used, random measurement error in occupational class will have led to an underestimation of socioeconomic differences in C-section rate.

Our findings pertain to an academic referral hospital and are not generalizable to Kenya at large, where nearly 40% of women have home births and, consequently, C-section rates at the population level are lower [4]. Socioeconomic differences in C-section rates are much larger in Kenya at large, as they partly capture socioeconomic differences in facility birth. Yet, the C-section rate in our study hospital was comparable to the institutional C-section rate in Kenya as a whole [4]. Given that the hospital draws on a broad catchment population, and that only a tiny proportion of women used the hospital as referral hospital, one might see our findings as a precursor of what may happen in the rest of Kenya -and arguably other low and middle income countries- when facility birth rates increase further, especially when repeat C-sections are highly common. At the same time, the C-section rate in our study hospital was still modest compared with those observed in some countries where population-level C-section rates reach up to 40–60% [30]. In such countries, the contribution of unnecessary primary C-sections to (inequalities in) the C-section rate will be much larger than in our study.

First, our study shows that a combination of criterion-based auditing and equity analysis can help gain a better understanding of drivers of C-section rates and inequalities in these rates – a first step to curb increasing over-use. Our study of over 12,000 births was only practically feasible because of the availability of electronic patient records. Electronic records can facilitate monitoring, and our study shows the potential for using hospital record data for improvements in health care delivery. At the same time, a more detailed documentation of decisions around mode of birth, including if C-section was on demand, is advisable for accountability purposes and to improve quality of care. Second, our study shows that socioeconomic differences in C-section rates, especially in contexts of moderate C-section rates and near universal repeat C-sections, can be largely explained by differences in medical indication (largely due to previous C-section), age and parity. This should be taken into account in future explanatory research on socioeconomic differences in C-section rates. Third, qualitative research on decisions around primary C-sections, both in the context of moderate C-section rates as in Kenya, as in the context of very high C-section rates such as for example Colombia, will be important to understand demand and supply side mechanisms that drive over-use. Finally, future research should address the paucity of evidence on how to safely and effectively reduce primary and repeat C-section rates in resource poor countries [31].

Our findings suggest that unnecessary primary C-sections, combined with a practice of near universal repeat C-sections fuel the C-section epidemic. Unnecessary primary C-sections cause needless maternal and infant morbidity [5‐8, 32]. The incidence of uterus rupture in women with a prior C-section, for example, is 1% in resource-poor countries [33]. Unnecessary primary C-sections combined with near universal repeat C-sections lead to a cascade of C-sections. Our finding that repeat C-sections substantially contribute to (inequalities in) the C-section rate correspond Vogel et al.’s conclusions that repeat C-section are an increasingly important driver of C-section rates in low- and middle-income countries [34]. We add that they are also an important driver of socioeconomic inequalities in C-section rates.

Little is known about how to effectively reduce unnecessary primary C-sections in low and middle income countries [35, 36]. Some evidence suggests that audit and feedback can reduce C-section rates [37] and that this is feasible in Sub-Saharan African contexts [38]. Changes in financial incentives for hospitals and doctors in combination with better pain relief and support during labor may also be effective [39]. Furthermore, investments in training and equipment for assisted vaginal birth, especially vacuum extraction, can reduce C-section rates in case of prolonged second stage labor or foetal distress [40]. While assisted vaginal birth is associated with reductions in morbidity and mortality, especially in resource-poor countries [40], such births remain rare in these settings [41]. Promoting safe trial of labor with close monitoring in women with a scarred uterus can reduce the prevalence of repeat C-sections, although high-quality evidence on the benefits and harms of vaginal birth after C-section remains scarce [42, 43]. Prevention of unnecessary primary C-sections and promoting safe trial of labor should be part of broader efforts to improve quality of maternity care, which should include shared decision making [44].