Results

THIN–HES-linked and THIN primary care data sets descriptions

During the study period, 219,390 deliveries were identified in the THIN–HES-linked data set. We excluded 32,808 delivery records, as they did not have a baby with the same family number as the mother. We then matched a baby for each of the remaining 186,582 deliveries. We excluded a further 72,807 records, either because they were linked to more than one potential mother or because the baby was not registered within the first year from birth. This left us with a total of 113,775 valid deliveries linked to a baby identified in the THIN–HES-linked data set.

In the THIN primary care data set, we looked for incident deliveries (VD or CS) from 1 January 2006 to 31 December 2018 in women aged 13–50 years at the time of delivery who were already registered with the practice. We found a total of 441,540 delivery records in the database, of which 54,939 were excluded, as there was no baby born in the same month and year of delivery with the same family number as the mother. We further excluded 135,318 records of babies who were either linked to more than one potential mother or not registered within a year from birth. This left a total of 251,283 mother–baby pairs identified using Read codes in the THIN primary care database. Finally, we extracted information from records of the babies who had ‘born by vaginal delivery’ or ‘born by caesarean section’ Read codes from the THIN primary care database and found 46,593 records. All of these babies, however, were already present in our linked primary care data set derived from the delivery codes noted in the mother’s medical record and, therefore, did not add any new babies that could be linked to a mother.

Combined, there were a total of 365,058 mother–baby pairs from THIN–HES-linked and THIN primary care data sets. A total of 57,317 of these pairs were identified in both the THIN primary care and THIN–HES-linked databases (i.e. duplicates) and, therefore, we removed those records from the THIN primary care database. This left us with a final total of 307,741 unique mother–baby pairs from the THIN primary care and THIN–HES-linked databases (Figure 2).

FIGURE 2 Study population selection, including mother–baby data linkage in THIN data sets. (Figure not reproduced here; see the original article.)

CPRD–HES-linked and CPRD primary care data sets descriptions

The CPRD mother–child-linked database contained 646,967 mother–baby pairs in total. When examining CPRD data, our main goal was to identify linked mothers and babies who had a recorded mode of delivery (VD/CS).

We used the same criteria to define patient start date and patient end date for the CPRD database as we did for THIN. There were 444,823 mother–baby pairs in the CPRD–HES-linked data set. To retain only the valid mother–baby pairs in the CPRD–HES-linked data, we excluded (1) 11,172 records because information on babies was not present in the primary care practice data and (2) 171,269 records because either the mother or baby did not meet the study inclusion criteria (78,229 babies were excluded because they were not registered within 1 year from date of birth, 61,553 mothers were excluded because their delivery date was not in between the patient start date and the patient end date or their age at delivery did not match study inclusion criteria, and 31,487 mother–baby pairs were excluded because both the mother and the baby did not meet the study inclusion criteria). This left us with 262,382 mother–baby pairs that had an OPCS-4 or ICD-10 delivery code identifying the mode of delivery and met our study inclusion criteria (as explained in Chapter 2).

In the CPRD primary care data set, there were 452,353 mother–baby pairs, from which we excluded 121,598 records as they did not meet the study inclusion criteria, leaving us with 330,755 mother–baby pairs with a recorded mode of delivery in the CPRD primary care database.

In total, we had 593,137 mother–baby pairs from both CPRD data sets, of which we excluded 183,230 duplicates (valid mother–baby pairs found independently in both the CPRD primary care and CPRD–HES data sets), removing them from the CPRD primary care data set. This left us with 409,907 valid mother–baby pairs across the two CPRD data sets (Figure 3).

FIGURE 3 Study population selection in CPRD data sets. (Figure not reproduced here; see the original article.)

Final THIN–CPRD data set description

Across all data sets, we identified 717,648 valid deliveries. We then excluded 201,703 (28.1%) duplicates (163,408 from THIN data sets and 38,295 from the CPRD data sets). This resulted in a final total of 515,945 mother–baby pairs for inclusion in our analysis [144,333 (28.0%) from THIN data sets combined and 371,612 (72.0%) from CPRD data sets combined] (Figure 4).

FIGURE 4 Flow diagram of the final THIN–CPRD data set. (Figure not reproduced here; see the original article.)

Out of the total of 515,945 births, 144,861 (28.1%) were births by CS and the remaining 371,084 (71.9%) were by VD. The number of actual linked mother–baby pairs is smaller than the estimate in our sample size justification. This was mainly because we used THIN data from the year 2010 to derive our estimates and, since 2010, many practices have left the Vision system and in turn the number of practices contributing to THIN and CPRD GOLD has substantially reduced over the years, resulting in decreasing numbers of deliveries in the THIN–CPRD data set in the more recent years (Figure 5). At the time of producing our sample size estimates, we overlooked this phenomenon and did not account for it and, hence, ended up with an overestimate of the number of deliveries in this combined data set.

FIGURE 5 The number of births by delivery mode and year in the final THIN–CPRD data set. (Figure not reproduced here; see the original article.)

Overall, 57.2% (295,079/515,945) of the births in the final THIN–CPRD data set were from THIN–HES- or CPRD–HES-linked data sets (Figure 6). The proportion of births from HES-linked data sets decreased over the years, with the majority of births from 2015 coming from primary care databases, and HES-linked data were not available after the beginning of 2018. As the recording of delivery mode is less complete in primary care, the decrease of births from HES-linked data sets explains the larger proportions of CS births in the later years of the study in the final THIN–CPRD data set. Over one-quarter [26.0% (76,819/295,079)] of births identified in HES-linked data sets were delivered by CS.

FIGURE 6 The number of births by delivery mode and year from the THIN–CPRD data set, with linked HES data used for sensitivity analysis. (Figure not reproduced here; see the original article.)

HES data set description

The derivation of the HES analysis set covering the whole of England, which was produced from the matching of delivery episodes and baby records, is summarised in Figure 7.

FIGURE 7 Flow diagram for derivation of the mother–baby-linked HES data set covering the whole of England and linkage to the national survey of the hospital prophylactic antibiotic administration policies. (Figure not reproduced here; see the original article.)

After randomly selecting one baby in cases of multiple births from 7,222,967 successfully matched mother–baby pairs, the mother-baby-linked HES data set covering the whole of England included 7,147,884 mother–baby pairs, of which 25.4% (1,813,799/7,147,884) were CS births. The data set linked to the responses to the national survey of hospital prophylactic antibiotic administration policies contained records of 3,945,351 mother–baby pairs (with one baby in the data set in cases of multiple births), 25.4% (n = 1,001,598) of which were CS births. Figure 8 shows the absolute numbers of CS and VD births over time in the full HES data set and the HES data set linked to the survey data. The sharp drop in the number in the final year is because it includes births only until the end of March 2017.

FIGURE 8 The number of births by delivery mode and year in (a) the HES data set covering the whole of England; and (b) the HES data set linked to the data from the national survey of hospital prophylactic antibiotic administration policies used in the main analysis. (Figure not reproduced here; see the original article.)

Results of the national survey of hospital prophylactic antibiotic administration policies

An overall survey response rate of 75.7% (143/189) was achieved.

The vast majority of hospitals [95.8% (137/143)] reported that their unit’s policy was to provide antibiotic prophylaxis before skin incision and five (or fewer) hospitals had this policy for either pre-labour or in-labour CS only.

Of the hospitals with a pre-incision antibiotic policy, 86.9% (119/137) indicated in which year they had implemented this policy change. The increase in the proportion of hospitals with a pre-incision antibiotics policy was gradual (Figure 9) and there was relatively little variation between different countries in the UK and regions within England (Appendix 1). No hospitals reported that they had implemented a pre-incision antibiotic policy before 2008. By 2013, over half of all hospitals had a policy to administer antibiotic prophylaxis before skin incision.

FIGURE 9 Cumulative percentage of hospitals over time with policies to administer antibiotic prophylaxis before skin incision for women undergoing CS in the whole of the UK and England separately. (Figure not reproduced here; see the original article.)

All of the hospitals that responded to the question about the type of prophylactic antibiotics currently administered used broad-spectrum antibiotics. The most common regimens were co-amoxiclav (40.6%, 54/133), cefuroxime (29.3%, 39/133) and cefuroxime together with metronidazole (26.3%, 35/133).

Of the 125 hospitals that had changed the practice to pre-incision antibiotics and responded to whether or not they used the same antibiotics before the policy change, 72.0% (90/125) had not changed their antibiotic regimen. The vast majority (≥ 95%) of those hospitals that had changed antibiotics after implementing pre-incision antibiotic policy had previously used co-amoxiclav.

Clindamycin was the most commonly used prophylactic antibiotic for women with a penicillin allergy (71.4%, 95/133), either administered on its own (53.4%, 71/133) or used along with gentamicin (12.8%, 17/133). Other regimens included gentamicin and teicoplanin and/or metronidazole in 12.8% (17/133) of hospitals. Metronidazole was used on its own or along with other antibiotics (such as cefuroxime or cefaclor) in 7.5% (10/133) of hospitals and teicoplanin alone was administered in 6.0% (8/133) of hospitals. Of the 123 hospitals that had changed their practice to pre-incision antibiotics and responded to whether or not they had used the same antibiotics before the policy change for women with penicillin allergy, 91.1% (112/123) used the same antibiotic regimen.

Eight hospitals with pre-incision antibiotic policies had conducted audits from 2014 onwards regarding the proportion of women given prophylactic antibiotics before skin incision, and reported that 70–100% of women had received pre-incision antibiotics.

Results of covariate analysis

Covariate analysis graphs are included in Appendix 2. There were no marked differential changes over time between the CS and VD groups in terms of maternal, household and child characteristics.

Primary outcome results

Main analysis

Outcomes recorded in primary care

Table 6 shows the numbers of children with a first-time diagnosis of asthma and eczema, the total years of follow-up and the overall rate per 1000 person-years (note that these rates are not directly comparable to the rates in the total 0- to 4-year-old population, as rates differ by age and the maximum age to which the children born in 2015, 2016, 2017 and 2018 were followed up to was 4 years, 3 years, 2 years and 1 year, respectively). As the rates of outcomes are compared according to the probability that each mother had received pre-incision antibiotics based on the year of delivery and the national uptake of the pre-incision antibiotic policy for CS in that year, it is not possible to present this information in a table by exposure group. Therefore, we have plotted the rates of outcomes by delivery mode, year of birth and age in Appendix 3.

There was a 9% decrease in diagnosis of asthma in children exposed to pre-incision antibiotics compared with those with no such exposure (IRR 0.91, 95% CI 0.78 to 1.05). There was a 2% decrease in diagnosis of eczema in children with pre-incisional antibiotics compared with those with no such exposure (IRR 0.98, 95% CI 0.94 to 1.03). Together, these show that there was not enough evidence of an association between timing of prophylactic antibiotics administration and asthma and eczema diagnoses. The minimum increase in risk we estimated that we would be able to detect during our power calculations was not within the width of the CIs.

As part of the analysis model, a term was included to allow for delivery mode (CS or VD). For each study outcome, we have calculated an IRR comparing CS with VD to ascertain the difference in risk based on delivery mode (see Appendix 4).

Outcomes recorded in secondary care

Table 7 shows the numbers of children with a first-time hospital admission for asthma and eczema, the total years of follow-up (calculated assuming a follow-up of 5 years for children born before 2015 and follow-up of 4 years, 3 years, 2 years and 1 year for the children born in 2015, 2016, 2017 and 2018, respectively) and the overall rate per 1000 person-years. Pre-incision antibiotics were not associated with statistically significant increases in the risk of a childhood asthma or eczema resulting in hospital admission.

There was a 5% increase in first-time hospital admissions for asthma in children exposed to pre-incision antibiotics compared with those with no such exposure (IRR 1.05, 95% CI 0.99 to 1.11). There was a 4% decrease in first-time hospital admissions for eczema in children exposed to pre-incision antibiotics compared with those with no such exposure (IRR 0.96, 95% CI 0.71 to 1.29). Together, these findings show that there was not enough evidence of an association between timing of prophylactic antibiotics administration and asthma and eczema resulting in hospital admission, although, owing to a small number of hospital admissions, the CIs for eczema were very wide.

Sensitivity analyses

The results of sensitivity analyses for asthma diagnosis in primary care (Table 8) were fairly similar to the results of the main analysis, with generally slightly lower IRRs than in the main analysis, including for moderate and severe asthma, based on prescribing information. We performed several additional sensitivity analyses for asthma outcome definition and these IRRs were also similar to the IRRs reported in the main analysis.

The sensitivity analysis of HES data for more severe asthma resulting in hospital admission produced very similar results to the main analysis, with no indication of significantly increased or reduced risk with introduction of pre-incision antibiotics (Table 9).

The results of sensitivity analyses for eczema were similar to the main results, suggesting no significant impact of pre-incision antibiotics on the risk of eczema, including for moderate and severe eczema, defined based on prescribing information (Table 10) and admission to hospital (Table 11).

A Poisson model has the potential for overdispersion, although we included robust standard errors to allow for the misspecification of the variance. A comparison of the results from the Poisson model and the results from a negative binomial model (with the same covariates) showed very similar results.

Secondary outcome results

Outcomes in children

Outcomes recorded in primary care

We found no evidence of a significant risk difference for any of the other allergic, allergy-related, autoimmune diseases, infections and less specific measures of child health (i.e. colic and failure to thrive) recorded in primary care in relation to the timing of prophylactic antibiotic administration (Table 12). The absolute numbers of children with autoimmune and neurodevelopmental diseases were small in this data set, resulting in a high level of uncertainty around the estimates. Given that multiple tests were undertaken and many of the included outcomes were exploratory because of a poor evidence base for an association between microbiota, early antibiotic exposure and these outcomes, a higher degree of caution is required when considering these results where the strength of the association is not high. Some of the autoimmune conditions and immune system-related diseases were extremely rare in children aged < 5 years [i.e. juvenile idiopathic arthritis, systemic sclerosis and scleroderma, inflammatory myopathies, systemic lupus erythematosus, autoimmune (idiopathic) thrombocytopenic purpura, juvenile pernicious (megaloblastic) anaemia, inflammatory bowel disease and leukaemia], preventing any meaningful calculation of risk.

We observed a very small (3%) increase in the likelihood of being prescribed antibiotics in early childhood (IRR 1.03, 95% CI 1.00 to 1.06), which was statistically significant because of the overall very high rates of antibiotic prescribing in young children. Similarly, there was a very small (2%) relative increase in the total number of consultations for any reason recorded in primary care (IRR 1.02, 95% CI 1.01 to 1.02).

Outcomes recorded in secondary care

We did not find evidence of a significant increase in the relative risk associated with pre-incision antibiotics for any of the following more severe secondary outcomes of interest resulting in a hospital admission [i.e. anaphylaxis, type 1 diabetes, juvenile idiopathic arthritis, autoimmune (idiopathic) thrombocytopenic purpura, lower respiratory tract infections and bronchiolitis, urinary tract infection, necrotising enterocolitis and leukaemia] (Table 13). The numbers of children with a diagnosis of coeliac disease, systemic sclerosis and scleroderma, inflammatory myopathies, systemic lupus erythematosus, juvenile pernicious (megaloblastic) anaemia and inflammatory bowel disease were very small in the study population and we could not estimate IRRs for these outcomes in secondary care. There was a very small (2%) increase in the risk of being admitted to hospital for gastroenteritis (IRR 1.02, 95% CI 1.00 to 1.05) and a very small (1%) relative increase in the risk of having a hospital admission for any reason (IRR 1.01, 95% CI 1.01 to 1.02). These outcomes were exploratory and, although the differences in risk were statistically significant, they may not be clinically significant.

Having a pre-incision antibiotics policy was associated with a 25% relative reduction in early-onset neonatal sepsis (IRR 0.75, 95% CI 0.65 to 0.87) and a 12% reduction in late-onset sepsis (IRR 0.88, 95% CI 0.80 to 0.97), although the overall recorded incidence of neonatal sepsis has increased over the study period both in children born by CS and in those born by VD (see Appendix 3).

Maternal outcomes

Table 14 shows the overall numbers and percentages of women who developed maternal infections during the study period by delivery mode, the uncontrolled risk estimates comparing pre-incision antibiotics with antibiotics administered after cord clamping and the estimates adjusting for rates in the VD comparator group, as there were changes in the recorded incidence of different maternal post-partum infection indicators in routine data over the time (see Appendix 3).

We found a significant negative association between giving antibiotics before skin incision and the composite outcome of maternal infectious morbidity, with a relative reduction in infectious morbidity of 33% in the uncontrolled analysis in primary care (IRR 0.67, 95% CI 0.56 to 0.79) and 30% in the analysis controlling for infection rates in the VD group (IRR 0.70, 95% CI 0.63 to 0.77). The overall rates of maternal sepsis recorded in routine health-care records increased substantially over the study period (see Appendix 3). The observed reduction in the risk of the maternal infectious morbidity outcome was principally due to the marked decrease in wound infections in the CS group recorded in primary care over the study period, and differences in endometritis and maternal sepsis in the adjusted analyses were not statistically significant. In the controlled analysis, there was also no evidence of a significantly reduced risk of urinary tract infections. Pre-incision antibiotic policy was associated with a 6% (IRR 0.94, 95% CI 0.85 to 1.03) and an 11% reduction (IRR 0.89, 95% CI 0.85 to 0.92) in women being prescribed antibiotics in the post-natal period in primary care in the uncontrolled analysis and analysis adjusted for delivery mode, respectively.

Table 15 shows results for maternal infectious morbidity recorded in hospital data. Diagnoses were recorded after delivery, during the birth episode or as a new hospital admission in the post-partum period and, therefore, the incidence (apart for maternal sepsis) is lower than that in the primary care records.

The relative risk for the composite maternal infectious morbidity outcome was significantly reduced in the analysis adjusted for delivery mode only, showing a 16% reduction associated with pre-incision antibiotic policy (IRR 0.84, 95% CI 0.81 to 0.87). For individual outcomes, the risk of wound infection recorded in secondary care was reduced by 16% in the adjusted analysis (IRR 0.84, 95% CI 0.81 to 0.87); however, the risk of maternal sepsis was still increased, even after adjusting for increasing rates of maternal sepsis in women who delivered vaginally in the same locations where CSs were performed. After adjusting for delivery mode, there was no significant association between urinary tract infections and giving antibiotics before skin incision. The length of hospital stay was slightly reduced in the analysis adjusted for delivery mode.

Additional sensitivity analyses

The effect sizes in the sensitivity analyses for all childhood outcomes with unexpected statistically significant findings in primary care were reduced, compared with the main analysis (Table 16). Sensitivity analyses suggest a high degree of imprecision in the estimate of the risk of attention deficit hyperactivity disorder (ADHD) in relation to exposure to pre-incision antibiotics. Given the poor existing evidence base on whether or not antibiotics or microbiota changes are clearly associated with ADHD, these results may be spurious.

Overall, the results of the sensitivity analyses in HES data were similar to the results of the main analysis. They confirm the reduced risk for early-onset neonatal sepsis; however, there was more uncertainty around the estimates for late-onset sepsis. The associations between pre-incision antibiotics and hospital admission for gastroenteritis, as well as admission to hospital for any reason, remained statistically significant, although the size of the increased risk remains very small (Table 17).

Results of subgroup analyses

Type of caesarean section

The results of subgroup analyses comparing the effect of pre-incision antibiotics with post-cord clamping antibiotics in emergency and elective CS on asthma and eczema, adjusted for rates of VD, were similar to the main results (Tables 18 and 19). Although uncertainty was larger around the estimates, the results do not suggest an increased risk associated with a change from post-cord clamping to pre-incision antibiotics in either elective or emergency CS.

Antibiotic regimen

Table 20 shows results of analyses restricted to hospitals that were administering the same antibiotic regimen before and after policy change regarding the timing of prophylactic antibiotics. No particular antibiotic regimen was associated with a higher risk of being admitted to hospital because of asthma. Hospital admissions due to eczema were very rare and more evidence would be required to confirm if there may be a clinically significant decrease in the risk of eczema associated with pre-incision administration of co-amoxiclav. Overall, there was no evidence that a particular antibiotic regimen was associated with a different risk of necrotising enterocolitis when comparing pre-incision and post-clamping antibiotics.