Long term impact of prophylactic antibiotic use before incision versus after cord clamping on children born by caesarean section: longitudinal study of UK electronic health records

Abstract Objective To investigate the impact on child health up to age 5 years of a policy to use antibiotic prophylaxis for caesarean section before incision compared with after cord clamping. Design Observational controlled interrupted time series study. Setting UK primary and secondary care. Participants 515 945 children born in 2006-18 with linked maternal records and registered with general practices contributing to two UK primary care databases (The Health Improvement Network and Clinical Practice Research Datalink), and 7 147 884 children with linked maternal records in the Hospital Episode Statistics database covering England, of which 3 945 351 were linked to hospitals that reported the year of policy change to administer prophylactic antibiotics for caesarean section before incision rather than after cord clamping. Intervention Fetal exposure to antibiotics shortly before birth (using pre-incision antibiotic policy as proxy) compared with no exposure. Main outcome measures The primary outcomes were incidence rate ratios of asthma and eczema in children born by caesarean section when pre-incision prophylactic antibiotics were recommended compared with those born when antibiotics were administered post-cord clamping, adjusted for temporal changes in the incidence rates in children born vaginally. Results Prophylactic antibiotics administered before incision for caesarean section compared with after cord clamping were not associated with a significantly higher risk of asthma (incidence rate ratio 0.91, 95% confidence interval 0.78 to 1.05) or eczema (0.98, 0.94 to 1.03), including asthma and eczema resulting in hospital admission (1.05, 0.99 to 1.11 and 0.96, 0.71 to 1.29, respectively), up to age 5 years. Conclusions This study found no evidence of an association between pre-incision prophylactic antibiotic use and risk of asthma and eczema in early childhood in children born by caesarean section.


Further details on healthcare data sources and linkage
Mother and baby records in the CPRD GOLD dataset provided to us by the CPRD Centre of the Medicines and Healthcare products Regulatory Agency (MHRA) were already linked. 1 Linkage of anonymised records is possible due to unique family identification numbers based on residence, child's birth year and month, delivery, pregnancy related and postnatal information. 2, 3 We adapted a similar strategy for linking mother and baby records in THIN. For each delivery during the study period in the mother's healthcare record, we checked if there was a baby and a mother with the same family identification number and if the delivery's year and month matched the baby's year and month of birth. In both THIN and CPRD, children had to be registered with the mother's general practice within the first year of life to be included in the study dataset.
Some general practices contribute data to both THIN and CPRD GOLD databases. To identify patient records from these practices, we adapted previously published algorithms using practice-related, patient registration, demographic and medical information to ensure that duplicate records were not included in our final 5 As there is no common identifier to link maternal and baby birth records in HES, we adapted a previously validated algorithm for linking maternal and baby data in this dataset. 6 We first deterministically linked maternal and baby records if they matched on the code for the general practice, maternal age, birthweight, gestation, birth order and baby's sex. We allowed for missingness in up to two of these five variables but disallowed the match if any of the five variables, which were not empty, did not agree. The remaining maternal delivery records were linked probabilistically by comparing them to birth records on 21 variables. Each variable was given a positive match weight, if there was a match between the mother's and baby's record, and a negative weight if there was a mismatch. We applied previously published weights and matching threshold to decide which matches were probable. 6

Further details on statistical analyses
Statistical analyses for outcomes recorded in primary care In the THIN-CPRD dataset analysis, the primary outcomes and dichotomous secondary outcomes in children are summarised using the number of children with a record of the outcome of interest and the length of time at risk, which was estimated as the time from birth to the earliest of: the child's fifth birthday, the date when the child was diagnosed with the outcome of interest for the first time, the date the child was lost to follow-up (for example, due to moving from the general practice), or the study end date (31 December 2018). The rate of each outcome was calculated as the ratio of the number of events and exposure time in person years, multiplied by 1,000. Secondary outcomes that were counts are summarised using the total number of events and the person years at risk. Maternal outcomes are summarised as the percentage of total births stratified by delivery mode. A Poisson regression model was fitted to the data with an offset for the follow-up time. The model using THIN-CPRD data included terms for the year of birth, age (year of life), their interaction, delivery mode (to allow for differential risk of the outcome in those who had CS compared to VD), and antibiotic timing (to allow for differential risk in those receiving preincision rather than post-cord clamping prophylactic antibiotics). Whereas in a simple before-after time series design this variable would be a zero-one variable, we used the probability of antibiotic use before skin incision in each year assessed through our national survey on the timing of antibiotic prophylaxis for CS, adjusting for misclassification. We present the incident rate ratios (IRRs) and the 95% confidence intervals as the rate of outcomes in those who would have been exposed to pre-incision prophylactic antibiotics to the rate of outcomes in those in the unexposed group (receiving antibiotic prophylaxis after cord clamping). For outcomes that were very rare in our study population (with fewer than 200 individuals diagnosed over the whole study period), we present summary statistics only.

Sensitivity and subgroup analyses
We conducted a series of different sensitivity analyses to assess the robustness of the results: sensitivity to outcome definition, definition of timing of the prophylactic antibiotic policy change, data recording quality and random effect. Sensitivity analyses were done for the two primary outcomes and secondary outcomes where results suggested a statistically significant association with the timing of antibiotic prophylaxis and the association was unexpected.
In the sensitivity analysis to outcome definition, we investigated the impact of prophylactic antibiotic timing on severity of asthma and eczema based on prescribing information indicative of moderate or severe asthma and eczema. We also explored several alternative definitions of asthma and eczema, including asthma diagnosed in children between the third and the fifth year of life, asthma definition based on Read codes in the Quality and Outcomes Framework, 7 eczema diagnosis based on Read codes in children between the first and the fifth birthday and at least two related prescriptions within 90 days before or 365 days after the eczema Read code record, and eczema diagnosis based on Read codes only.
In the sensitivity analysis for the definition of timing of the prophylactic antibiotic policy change, we compared births (between 2006 and 2010) before the NICE guidance recommendation to administer antibiotic prophylaxis for CS prior to skin incision to the births in the years when over half of all hospitals had introduced the pre-incision antibiotic policy (between 2013 and 2018). Children delivered by CS were classified as either not having received pre-incision antibiotics (between 2006 and 2010) or exposed to pre-incision antibiotics (between 2013 and 2018).
In the sensitivity analysis evaluating the impact of data recording quality, THIN-CPRD database was restricted to mother-baby pairs with linked secondary care data as the most accurate source for mode of birth.
In the sensitivity analysis examining non-independence of observations, we fitted a mixedeffects Poisson model including a random effect for general practice.
Finally, for the primary outcomes we also undertook exploratory subgroup analysis by the type of CS (elective and emergency). For this analysis, we report the IRR for prophylactic antibiotic timing in the elective CS group, emergency CS group and the p-value for the interaction between them indicating the strength of evidence of a differential effect of antibiotic prophylaxis timing based on CS type.

Statistical analyses for outcomes recorded in HES
In the HES dataset analysis, most primary and dichotomous secondary outcomes are summarised using the number of children admitted to hospital at least once and the length of time at risk, which was estimated as the time from birth to the earliest of: the child's fifth birthday or the study end date (31 March 2019). The rate of each outcome was calculated as the ratio of the number of events and exposure time in person years, multiplied by 1,000. For secondary outcomes with a very short exposure time (neonatal sepsis and necrotising enterocolitis), we present the outcome rate per 1,000 births. Maternal outcomes are summarised the same way as in the THIN-CPRD dataset, apart from hospital length of stay, which was not normally distributed and therefore the median and the inter-quartile range (IQR) is presented.
Like in the THIN-CPRD analysis, a Poisson regression model was fitted to the HES data for all dichotomous outcomes, with an offset for exposure time. It included terms for the year of birth, mode of birth (VD or CS), and prophylactic antibiotic timing, which was categorised as before policy change, year of policy change, and after policy change, for each birth based on the year of policy change in each hospital that responded to the national survey on antibiotic prophylaxis for CS. The comparison between births after the year of the policy change (assumed to have been exposed to pre-incision antibiotics) and births before the year of the policy change (assumed having no exposure to pre-incision antibiotics), gives an estimate of the effect of the policy change in prophylactic antibiotic timing. For continuous outcomes (maternal length of hospital stay), we fitted a linear regression model and present the difference in mean length of stay between deliveries when pre-incision antibiotic policy was in place and deliveries during post-cord clamping policy.

Sensitivity and subgroup analyses
We conducted several sensitivity analyses: sensitivity to definition of timing of the prophylactic antibiotic policy change, random effect, and exploratory discordant sibling analysis.
In sensitivity to definition of timing of antibiotic policy change, we used the full mother-baby linked HES dataset, not limiting it to the births that could be linked to the hospitals who had reported the year when they implemented pre-incision antibiotic policy, and included the overall probability of pre-incision antibiotic use in each year estimated from our national survey on the timing of antibiotic prophylaxis using responses from hospitals in England.
In the sensitivity analysis examining non-independence of observations, we fitted a mixedeffects Poisson model including a random effect for hospital.
We also conducted an exploratory discordant sibling analysis attempting to control further for family-related genetic and environmental factors. For this analysis, we used a subset of data which included mothers who gave birth at least twice during the study period using the same delivery mode (CS or VD), including having one birth before and one birth after the year of the hospital prophylactic antibiotic policy change.
In addition to the exploratory subgroup analysis by CS type, for outcomes recorded in HES we also performed subgroup analysis by the most common antibiotic regimen type used (co-amoxiclav alone, cefuroxime alone, cefuroxime and metronidazole). Antibiotic regimen analysis was restricted to data from those hospitals who reported the use of the same antibiotics before and after the change in the timing of antibiotic prophylaxis. A Poisson model was fitted separately for each antibiotic regimen. This exploratory analysis was undertaken for the primary outcomes and necrotising enterocolitis due to previously reported increased risk of this disease in newborns of mothers who had received coamoxiclav. 8

Further information on the national survey of hospital prophylactic antibiotic administration policies
The survey on prophylactic antibiotic administration policies during CS was designed using REDCap in 2017 and a survey link was sent to all clinical directors for maternity units performing caesarean sections in the UK. Non-responders were followed up using e-mail, telephone and at professional conferences, and were also offered an option to complete the survey in MS Word.
An overall survey response rate of 75.7% (143/189) was achieved with hospitals responding whether they had implemented the pre-incision antibiotic policy in their hospital. A small proportion, 4.2% (6/143) of the hospitals that responded to the survey had not implemented the pre-incision antibiotics policy. Of the hospitals that had implemented the policy, 86.9% (119/137) indicated in which year they had implemented the pre-incision antibiotic policy. Of these respondents, 9.2% (11/119) did not specify the exact year of the policy change, but had entered a free text comment saying, for example, before which year the policy was changed, such as "prior to 2010". This year was used in the analysis as the year of the policy change. Few hospitals (6) stated that they had implemented the policy before 2010. In the survey, we also asked if the hospital had undertaken a clinical audit regarding the use of prophylactic antibiotics at the time of CS within the last few years and if so, what proportion of women were given prophylactic antibiotics before skin incision. In the seven hospitals who had audited the compliance with their pre-incision antibiotic policy, the proportion of CS deliveries where antibiotics were administered before skin incision varied between 70 to 100%.  Yes -Antibiotic prescribing* Yes -Length of hospital stay* Yes * Exploratory outcome. † Pelvic abscess was part of the composite infectious morbidity recorded in primary care; there was no specific ICD-10 code to identify pelvic abscess separately in HES dataset. ‡ Only if there was also a record of a diagnosis of any of the health conditions included in the composite infectious morbidity outcome in the post-partum period. Table S2. Number of children with asthma recorded in primary care (THIN-CPRD) database, overall rate per 1,000 person years and the relative risk of pre-incision compared to postcord clamping prophylactic antibiotics (sensitivity analyses)       * After adjustment for: year of delivery, mode of birth. † Due to these conditions occurring in the neonatal period, rates were calculated per 1,000 births. ‡ Analysis of the full HES dataset using the estimated probability of introduction of pre-incision antibiotics according to calendar year. § Random effect model for hospital for this outcome did not converge and so no results were obtained.   The graphs below display the rates of different outcomes over the study period. In each graph, the rates are presented separately for CS and VD. For outcomes investigated in the THIN-CPRD dataset, we present graphs for each year of life individually. No confidence intervals have been included in the graphs, but to aid the interpretation of the precision of these rates, we provide some example confidence intervals for the rate of the outcome (per 1,000 person-years) for a series of event rates based on the numbers of events and follow-up time found in THIN-CPRD data ( Table S11) and HES data ( Table S12).

Child outcomes in HES dataset
As HES captures all hospital admissions in England, all children were assumed to be followed up until the age of five years if they were born before 2015. For the birth years of 2015, 2016 and 2017, children were followed up only up to the age of four, three and two years respectively, therefore for these years the incidence rate reflects the incidence of the outcome of interest in these age groups.