Wealth-related inequality in caesarean section delivery and the contribution of recorded maternal and pregnancy risk markers in Ghana: analyses of three national surveys spanning 2014 to 2022

¹Akershus University Hospital, Nordbyhagen, Norway, ²Garden City University College, Kumasi, Ghana, ³Institute for Health Policy and Research (IHPR), Kumasi, Ghana, ⁴National Open University of Nigeria, Abuja, Nigeria

^&Corresponding author
Peter Kofi Taadi, Akershus University Hospital, Nordbyhagen, Norway

Introduction    

Caesarean section delivery can prevent maternal and perinatal morbidity and mortality when clinically indicated [1]. Without a clear indication, it has no established benefit for women or newborns and may expose them to avoidable short- and long-term risks [1]. An estimated 21.1% of births worldwide now occur by caesarean section delivery [2]. Yet some women who need the procedure receive it too late or not at all, while others undergo surgery from which they derive limited clinical benefit [3]. National rates therefore cannot show whether caesarean section delivery is distributed according to clinical need. In low- and middle-income countries, caesarean section delivery is almost five times more common among births in the richest wealth fifth than among those in the poorest fifth [4]. Rates are particularly high among women at low obstetric risk in some settings, including Brazil and China [4]. Pro-rich inequalities have been documented across diverse settings [5-7]. These may reflect barriers to surgery for poorer women with complications, more frequent use among wealthier women not explained by clinical need, or both. In Ghana, delivery fees were removed in four deprived regions in 2003 and nationwide in 2005 [8]. Free maternal healthcare was incorporated into the National Health Insurance Scheme in 2008 [9]. Nevertheless, women in northern Ghana reported payments for some services, medicines and diagnostic procedures [9]. Facility delivery increased after the 2008 policy [10], but skilled delivery remained concentrated among wealthier households [11,12]. The national caesarean section delivery rate rose from 4.4% in 2003 to approximately 20% in 2022, with substantial variation across regions [13,14]. Under the horizontal-equity principle, greater use among wealthier women with comparable need may indicate inequity [15,16]. We define legitimate need as maternal and pregnancy characteristics that alter the probability that caesarean section delivery is clinically required, while socioeconomic characteristics are treated as non-need factors [15,16].

However, household surveys record selected maternal and pregnancy risk markers rather than the clinical indications on which decisions about caesarean section delivery are based. Need-standardised estimates therefore describe inequality not accounted for by recorded risk markers and cannot establish whether an individual caesarean section delivery was clinically appropriate or whether all remaining inequality represents inequity. Previous Ghanaian studies found that caesarean section delivery is more common among wealthier women [17-19] and that wealth remained associated with its use after adjustment for maternal and pregnancy characteristics [18,19]. However, adjusted associations do not quantify how much of the observed inequality reflects the socioeconomic distribution of recorded maternal and pregnancy risk markers. Studies assuming that obstetric need is equally distributed across socioeconomic groups may overlook socioeconomic differences in need [20]. Hospital-based analyses also exclude women who do not reach care [21]. Need standardisation has been applied to maternal healthcare in India [22], whereas a Ghanaian analysis of facility delivery assumed equal need [23]. We are not aware of a nationally representative study using need standardisation to examine wealth-related inequality in caesarean section delivery in Ghana. We therefore estimated how much of the wealth-related inequality in caesarean section delivery in the 2022 Ghana Demographic and Health Survey (GDHS) was accounted for by recorded maternal and pregnancy risk markers. We assessed reproduction of the principal finding in the 2014 survey and used the 2017 Ghana Maternal Health Survey (GMHS) to examine whether adding self-reported complications around delivery changed the estimates.

Methods     

Study design: we conducted separate cross-sectional analyses of the 2022 Ghana Demographic and Health Survey (GDHS; primary analysis), the 2014 GDHS (reproduction in an earlier independent round), and the 2017 Ghana Maternal Health Survey (GMHS; supplementary analysis) [24-26].

Data sources: the GDHS used stratified two-stage cluster sampling. The 2022 GDHS interviewed 15,014 women aged 15-49 years in 618 clusters between October 2022 and January 2023, with a 98.0% response rate [24]. The 2014 GDHS interviewed 9,396 women in 427 clusters between September and December 2014, with a 97.3% response rate [25]. Births in both analyses occurred after free maternal healthcare was incorporated into the National Health Insurance Scheme in 2008 [9]. The 2003 and 2008 rounds were excluded because they contained only 103 and 138 caesarean section deliveries, including 13 and 12 among women in the poorest wealth fifth, respectively (unweighted), too few for precise need-standardized estimates, and the 2003 GDHS lacked health insurance data. The GMHS interviewed 25,062 women aged 15-49 years about their most recent live birth or stillbirth since 2012 [26]. It recorded self-reported complications around delivery, which the GDHS does not. We followed the STROBE statement for cross-sectional studies.

Study population: the index birth was each woman´s most recent live birth within the survey-specific reference period: three years preceding the 2022 interview and five years preceding the 2014 interview [24,25]. If a woman´s most recent pregnancy ended in miscarriage, abortion, or stillbirth, her most recent live birth within the relevant period was included. Selecting one birth per woman eliminated multiple observations per woman and reduced recall time. The analytical samples included 5,218 live births in 2022 and 4,294 in 2014. Mode of delivery was recorded for every index birth, and no eligible birth was excluded because of a missing outcome. Because the rounds differed in their reference periods and treatment of previous deliveries, they were analyzed separately and were not used to estimate temporal change. The supplementary GMHS analysis included the most recent eligible delivery of 12,077 women.

Measures

Outcome: the outcome was respondent-reported caesarean section delivery of the index birth. Births reported as occurring at home were treated as non-caesarean section deliveries. The primary estimand therefore covered inequality arising both before and after reaching a facility. The facility-birth analysis was interpreted conditionally because facility delivery may itself be influenced by socioeconomic position and obstetric risk.

Socioeconomic position: women were ranked using the survey-specific Demographic and Health Surveys (DHS) household wealth index, derived by principal component analysis of household assets and housing characteristics [24,25]. We used the survey-weighted fractional wealth rank for concentration indices, its fifths for description and the DHS wealth-quintile indicators in the standardisation models. A sensitivity analysis used an asset index built identically across the 2003-2022 rounds from 11 household items.

Recorded maternal and pregnancy risk markers: the surveys did not record the clinical indication for each caesarean section delivery. We therefore treated selected maternal and pregnancy characteristics associated with obstetric risk as markers of measured need. These were multiple pregnancy; first delivery, defined as no previous delivery; four or more previous deliveries, with one to three as the reference; maternal age below 18 years or at least 35 years at delivery, with 18-34 years as the reference; and a large or small baby. Previous deliveries included stillbirths in 2022 but only live births in 2014, so parity markers are not strictly comparable between surveys. Parity and plurality are components of the Robson classification [27], while adolescent and older maternal age are associated with obstetric complications [28,29]. Ghanaian facility-based studies identified a large baby as an indication for caesarean section delivery [30] and macrosomia as a predictor of the procedure [31]. These characteristics indicate an increased probability of complications but do not establish that an individual caesarean section delivery was clinically necessary. A baby was classified as large if reported as very large or weighing at least 4.0 kg and as small if reported as very small or weighing less than 2.5 kg. Neither size nor weight was available for 26 births in 2022 and seven in 2014. Eighteen births in 2022 and five in 2014 met both the large-baby and the small-baby definitions. Sensitivity analyses excluded births without size information and, separately, gave reported weight precedence over perceived size. Maternal age and parity may reflect both obstetric risk and socially patterned reproductive circumstances. Their classification as need-related characteristics may therefore affect the proportion of inequality attributed to measured need. A sensitivity analysis consequently restricted measured need to multiple pregnancy and baby size.

Previous caesarean section delivery: a uterine scar is clinically relevant to the management of a subsequent pregnancy, and previous caesarean section delivery was the leading indication for planned caesarean section delivery in a Ghanaian primary health facility [30]. However, it may also reflect earlier socially patterned access to surgery. The surveys recorded it only when the earlier birth occurred within the reference period. We therefore presented specifications without and with this marker and repeated the latter among women with a previous delivery.

Non-need characteristics: maternal education, urban or rural residence, health insurance membership, region and household wealth were treated as non-need characteristics, which may influence access to caesarean section delivery but do not indicate clinical need. Region may capture differences in case composition but also reflects service availability. Emergency obstetric and newborn care was unevenly distributed across Ghanaian districts in 2010 [32], while caesarean section delivery was consistently most common in Greater Accra between 1998 and 2014 [33]. Classifying region as measured need could therefore attribute inequality arising from service distribution to differences in need. Region was retained as a non-need characteristic in the primary analysis and reclassified as need-related in a sensitivity analysis. Each survey´s own regional categories were used.

Complications around delivery in the Ghana maternal health survey: the GMHS asked women about problems occurring immediately before, during or after delivery. Interviewers coded spontaneous responses into predefined categories; the responses were not provider-recorded diagnoses. The core complications were oedema or pre-eclampsia, convulsions or eclampsia, malpresentation described as the baby´s hands or feet coming first, prolonged labour and obstructed labour. These complications were not clinically verified, and some may have followed rather than prompted caesarean section delivery. The GMHS analysis therefore tested whether adding them changed the estimates. Whether the decision to operate occurred before or during labour was used descriptively.

Statistical analysis: each survey was analysed separately, accounting for survey weights, clustering and stratification. We used StataNow/MP 19.5 (StataCorp, College Station, Texas, USA). Within each wealth fifth, we estimated caesarean section delivery rates among women with and without a selected risk marker: multiple pregnancy, a large or small baby, or maternal age below 18 or at least 35 years. Parity was excluded from this descriptive grouping because first delivery and four or more previous deliveries have opposite associations with caesarean section delivery [13,34]. We measured wealth-related inequality using the Erreygers-corrected concentration index, which removes the dependence of the bounds on the mean for binary outcomes [35,36]. Positive values indicate concentration among wealthier women. We then applied indirect standardisation [15,16]. A survey-adjusted linear probability model related caesarean section delivery to the recorded risk markers and non-need characteristics. No wealth-by-marker interactions were included. Need-expected use was predicted for each woman from her risk markers while non-need characteristics were held at their survey-specific weighted means. Remaining inequality was calculated as the observed concentration index minus the concentration index of need-expected use. Although conventionally termed the horizontal inequity index [15], we report it as need-standardised remaining inequality because clinical need was incompletely measured. It does not provide definitive evidence of inequity. The proportion accounted for was the need-expected index divided by the observed index, multiplied by 100. A negative proportion indicates that recorded risk was concentrated in the direction opposite to the observed inequality. We used a linear probability model because it preserves the additive relationship among observed, need-expected and remaining inequality. In nonlinear models, holding non-need characteristics at their means does not fully remove their influence, although concentration indices of need-standardised use are relatively insensitive to model choice [16]. Because fitted values may fall outside 0-1, we reported their number and repeated the standardisation using logistic regression. Percentile 95% confidence intervals came from 1,000 bootstrap replications in which survey clusters were resampled within strata and the whole procedure was repeated.

Sensitivity analyses: each sensitivity analysis altered one element of the primary specification. Model checks used logistic prediction and allowed risk-marker associations to vary by wealth. Alternative specifications added maternal height in measured subsamples (2,709 women in 2022; 2,195 in 2014), included previous caesarean section delivery among women with a previous delivery, classified region as need-related, retained only multiple pregnancy and baby size, or, in 2022, used each woman´s most recent live birth or stillbirth with stillbirth as an additional marker. Population checks included facility births and all live births within the reference period (5,768 in 2022; 5,740 in 2014), with clusters resampled so that births to the same woman remainedtogether. Another analysis used the consistently constructed asset index. Baby-size analyses are reported in the text only.

Ethical considerations: the Ethical Review Committee of the Ghana Health Service and the Institutional Review Board of ICF approved the 2022 GDHS (GHS-ERC 004/08/22) and 2014 GDHS (GHS-ERC 18/03/14) protocols, and the GMHS approvals are reported in its documentation [26]. All respondents gave informed consent. We used anonymised data from the DHS Program.

Results     

Sample characteristics: in the 2022 GDHS, 908 of 5,218 index births (weighted 20.1%) were caesarean section deliveries. The rate increased from 10.2% in the poorest wealth fifth to 36.6% in the richest, while facility delivery increased from 69.6% to 96.2% (Table 1). First delivery and a large baby were more common among wealthier women. Four or more previous deliveries, maternal age below 18 years, and multiple pregnancy were more common among poorer women. In 2014, 488 of 4,294 index births (weighted 13.6%) were caesarean section deliveries, rising from 4.7% in the poorest fifth to 28.2% in the richest. Risk-marker distributions were generally similar to those in 2022, except that multiple pregnancy was most common in the richest fifth.

Caesarean section delivery by recorded risk and household wealth: in 2022, cesarean section delivery was more common among women with a selected risk marker than among those without one within every wealth fifth (Figure 1). Women in the poorest fifth with a marker had a lower caesarean section delivery rate than women in the richest fifth without one: 10.9% (95% CI: 8.3 to 13.6) versus 31.3% (95% CI: 25.9 to 36.7). The same ordering occurred in 2014: 5.4% (95% CI: 3.3 to 7.6) versus 20.0% (95% CI: 14.7 to 25.3). Among facility births in 2022, rates with and without a marker were 14.8% and 14.6% in the poorest fifth and 44.4% and 32.3% in the richest. Among multiple pregnancies in 2022, caesarean section delivery occurred in 33.3% of births in the two poorest fifths (95% CI: 20.5 to 46.0; unweighted n=86) and 68.0% in the two richest fifths (95% CI: 49.7 to 86.4; unweighted n=36). These subgroup estimates were imprecise.

Primary need-standardisation analysis: in 2022, the concentration index was 0.217 (95% CI: 0.179 to 0.256) for observed caesarean section delivery and 0.017 (95% CI: 0.006 to 0.028) for need-expected use (Table 2). Recorded risk markers accounted for 7.8% (95% CI: 2.8 to 13.1) of observed inequality, leaving the need-standardised remaining inequality of 0.200 (95% CI 0.164 to 0.235). Adding previous caesarean section delivery as a risk marker raised the need-expected index to 0.026 (95% CI: 0.013 to 0.040) and the proportion accounted for to 12.0% (95% CI: 6.3 to 17.8) and reduced remaining inequality to 0.191 (95% CI 0.156 to 0.225).

Reproduction in the 2014 Ghana demographic and health survey: in 2014, the observed concentration index was 0.189 (95% CI:0.158 to 0.223), and the need-expected index was 0.024 (95% CI: 0.011 to 0.036). Recorded risk markers accounted for 12.6% (95% CI: 6.1 to 18.8) of observed inequality, leaving remaining inequality of 0.165 (95% CI: 0.136 to 0.199). Adding previous caesarean section delivery as a risk marker increased the proportion accounted for to 26.0% (95% CI: 18.3 to 33.7) and reduced remaining inequality to 0.140 (95% CI: 0.110 to 0.174).

Sensitivity analyses: need-standardised remaining inequality persisted across all sensitivity analyses (Table 3). The proportion accounted for ranged from -0.4% to 14.7% in 2022 and from 2.6% to 26.8% in 2014. Classifying region as need-related produced the largest proportions, whereas restricting measured need to multiple pregnancy and baby size produced the smallest. Logistic models produced proportions of 8.1% in 2022 and 11.8% in 2014. Prioritising recorded birthweight gave 7.3% and 10.1%, and excluding births without size information 7.7% and 12.4%. The linear probability model produced fitted values outside the 0-1 range for 160 of 5,218 women in 2022 (3.1%) and 446 of 4,294 women in 2014 (10.4%). No need-expected predictions were outside this range in 2022, compared with seven in 2014 (0.2%).

Supplementary analysis of reported complications: among 12,077 women in the 2017 GMHS, the weighted caesarean section delivery rate was 12.6%, and the observed concentration index was 0.157. At least one core complication was reported by 2.9% of women in the poorest fifth and 7.0% in the richest. Adding the five reported complications increased the proportion accounted for from 0.5% (95% CI: 3.2 to 4.7) to 4.2% (95% CI: 0.0 to 8.9), leaving remaining inequality of 0.150 (95% CI: 0.130 to 0.172). Among women reporting a core complication, caesarean section delivery rates were 16.4% in the poorest fifth and 43.8% in the richest, against 23.4% among women in the richest fifth without one. The concentration index was 0.115 for caesarean section deliveries decided before labour and 0.042 for those decided during labour. Pre-labour decisions accounted for 2.2% of all deliveries in the poorest fifth and 17.1% in the richest.

Discussion     

In the 2022 GDHS, caesarean section delivery was concentrated among wealthier women, while recorded maternal and pregnancy risk markers accounted for only a small proportion of this inequality. The finding was reproduced in the 2014 GDHS and remained when self-reported complications around delivery were added using the 2017 GMHS. It held across alternative models, populations, and definitions of measured need, although the proportion accounted for varied when the previous caesarean section delivery or region was classified as need-related. Region, however, also reflects the distribution of surgical services, so treating it as need-related may attribute unequal service availability to need. Across all specifications, recorded risk markers did not account for most of the observed inequality. The substantial remaining inequality should not be interpreted as a direct measure of inequity or as evidence that Cesarean section deliveries among wealthier women were unnecessary. The surveys recorded selected risk markers rather than the clinical indications for caesarean section delivery. Conditions such as fetal distress, placenta praevia, placental abruption, cephalopelvic disproportion, and failed induction were either not captured or captured only through self-report. Standardization also relied on observed associations, which reflect both obstetric risk and the health system´s response to that risk. If poorer women with a recorded marker are less likely to reach surgical care, need-expected use may be underestimated. This concern is particularly relevant where rural facilities lack basic childbirth inputs [37]. Nevertheless, incomplete measurement alone is unlikely to explain the entire pattern. The finding recurred in two independent GDHS rounds, and adding five reported complications in the GMHS increased the proportion accounted for only modestly. An earlier analysis of the 2017 GMHS also found more frequent caesarean section delivery among wealthier women after adjustment for reported complications [38]. Moreover, recorded risk markers were only weakly concentrated among wealthier women. Poorer women may face barriers before reaching facilities capable of providing surgical care. Although formal delivery charges have been removed, women may still pay for transport, medicines, diagnostic procedures and other services [9]. Caesarean section delivery also requires surgical capacity, trained personnel, anaesthesia and blood [3]. Facilities providing these resources have been unevenly distributed across Ghanaian districts [32]. Differences may also arise after women reach a facility. In Ghana, surgery for absolute maternal indications increased with maternal education [39], while thresholds for recommending caesarean section delivery may vary across clinicians and institutions [40]. Evidence from low- and middle-income countries has also linked potentially unnecessary caesarean section delivery to non-clinical influences, including private financing [41]. Wealthier women may be more likely to use private facilities [4], where caesarean section delivery rates are often higher [4,42]. The present data cannot distinguish among these explanations.

Wealth-related inequality remained among facility births, but this conditional comparison is affected by selection into facility delivery. Births outside facilities accounted for 30.4% of births in the poorest wealth fifth and 3.8% in the richest. Poorer women reaching a facility may therefore have been more selected on complications than wealthier facility users. If so, measured need among poorer facility users may be understated, and the facility-only analysis may underestimate remaining inequality. Caesarean section delivery was more common among wealthier women without a reported complication than among poorer women with one. However, the complications were self-reported and unverified, and the absence of a reported complication does not establish the absence of a clinical indication. Decisions made before labour were also more strongly concentrated among wealthier women than decisions made during labour. This pattern may reflect planned management of recognised indications, differences in antenatal diagnosis, facility type, clinician practice or maternal preference. These findings extend Ghanaian studies that documented higher caesarean section delivery among wealthier women and persistent associations after adjustment for maternal and pregnancy characteristics [17-19] by quantifying how much of the aggregate inequality corresponded to the socioeconomic distribution of recorded risk markers. Unlike analyses that assumed equal obstetric need across socioeconomic groups [20,23], our approach allowed measured need to follow the observed distribution of these markers. The classification of need nevertheless matters. Treating socially patterned characteristics such as maternal age and parity as need-related can increase the proportion of inequality attributed to measured need. Conversely, constrained access may weaken the observed relationship between risk markers and caesarean section delivery, thereby reducing need-expected use.

Strengths and limitations: the study has several strengths. Nationally representative data including births outside facilities avoid the selection inherent in hospital-based studies [21]. Need standardisation quantified how much of the observed inequality corresponded to recorded risk, which adjusted associations cannot show. The principal finding was reproduced in an independent survey round, examined against reported complications from a third survey and held across all sensitivity analyses. Bootstrap confidence intervals reflected the complex survey design. These strengths should be weighed against several limitations. First, the surveys did not record clinical indications, so remaining inequality may include unmeasured need as well as differences in access, diagnosis, clinician practice and preference. Second, several measures were imperfect: baby size was reported retrospectively and sometimes conflicted with birthweight; complications were self-reported and unverified; previous caesarean section delivery was recorded only within the reference period and may reflect earlier access to surgery, so we reported specifications with and without it; and wealth, residence, region and insurance were measured at interview rather than at delivery. Third, two analytical samples were selective: the facility-only analysis was affected by selection into facility delivery, and restricting the analysis to each woman´s most recent live birth gives less weight to higher-order births, although results were similar with all live births. Fourth, the 2014 and 2022 surveys differed in reference periods and in how previous deliveries were defined, so the 2014 analysis is a reproduction rather than an estimate of change. Finally, the cross-sectional design cannot identify the mechanisms behind the remaining inequality or the appropriateness of individual caesarean section deliveries.

Implications for policy and research: the first priority is timely access to indicated caesarean section delivery: strengthening referral and emergency transport for rural women, because transport barriers delay care and weaken referral from community services [43], and upgrading facilities that lack only one or two emergency obstetric signal functions in underserved districts, where in 2010 only 89 of 1,159 maternity facilities provided all of them [32]. National monitoring should report caesarean section delivery by wealth: the national rate of about 20% in 2022 concealed 10.2% in the poorest fifth and 36.6% in the richest. The second priority is documentation and audit of indications. Public and private facilities should classify caesarean section deliveries by Robson group so that rates can be compared across case mixes [1,27], and surveys and facility records should capture the indication, not only the timing of the decision [44]. The findings do not support indiscriminate reductions in caesarean section delivery among wealthier women; the aim is equitable access to clinically appropriate care. Linked population and facility studies with verified indications are needed to explain why women with comparable needs receive different care.

Conclusion     

Recorded maternal and pregnancy risk markers accounted for only a small proportion of wealth-related inequality in caesarean section delivery in Ghana in 2022. This finding was reproduced in the 2014 GDHS and remained after self-reported complications were added using the 2017 GMHS. Because clinical indications were incompletely measured, remaining inequality cannot be interpreted as definitive evidence of inequity, unmet need or unnecessary surgery. Improving timely access to indicated caesarean section delivery should be accompanied by systematic documentation and audit of clinical indications and wealth-disaggregated monitoring.

Competing interests     

The authors declare that they have no competing interests.

Authors' contributions     

PKT conceived the study, designed the analysis, prepared and analysed the data and drafted the manuscript (conceptualisation, methodology, formal analysis, data curation, writing: original draft). HB contributed to the analytical approach and to revising the manuscript (methodology, writing: review and editing). JAA contributed to the analytical approach and to revising the manuscript (methodology, writing: review and editing). POA contributed to the design, critically revised the manuscript for methodological and intellectual content and supervised the work (methodology, writing: review and editing, supervision). All authors read and approved the final manuscript.

Acknowledgments     

We thank The DHS Program for access to the survey data.

Tables and figures     

Table 1: caesarean section delivery, facility delivery and recorded risk markers by household wealth fifth (weighted %), Ghana Demographic and health surveys 2022 and 2014

Table 2: wealth-related inequality in caesarean section delivery and the proportion accounted for by recorded risk markers (95% CI), Ghana Demographic and health surveys 2022 and 2014

Table 3: sensitivity analyses: need-standardised remaining inequality and proportion of observed inequality accounted for by recorded risk markers (95% CI), Ghana Demographic and health surveys 2022 and 2014

Figure 1: caesarean section delivery by household wealth fifth among births with and without a selected recorded risk marker, Ghana Demographic and Health Surveys 2022 and 2014

References