Associations between macrolide antibiotics prescribing during pregnancy and adverse child outcomes in the UK: population based cohort study

Abstract Objective To assess the association between macrolide antibiotics prescribing during pregnancy and major malformations, cerebral palsy, epilepsy, attention deficit hyperactivity disorder, and autism spectrum disorder in children. Design Population based cohort study. Setting The UK Clinical Practice Research Datalink. Participants The study cohort included 104 605 children born from 1990 to 2016 whose mothers were prescribed one macrolide monotherapy (erythromycin, clarithromycin, or azithromycin) or one penicillin monotherapy from the fourth gestational week to delivery. Two negative control cohorts consisted of 82 314 children whose mothers were prescribed macrolides or penicillins before conception, and 53 735 children who were siblings of the children in the study cohort. Main outcome measures Risks of any major malformations and system specific major malformations (nervous, cardiovascular, gastrointestinal, genital, and urinary) after macrolide or penicillin prescribing during the first trimester (four to 13 gestational weeks), second to third trimester (14 gestational weeks to birth), or any trimester of pregnancy. Additionally, risks of cerebral palsy, epilepsy, attention deficit hyperactivity disorder, and autism spectrum disorder. Results Major malformations were recorded in 186 of 8632 children (21.55 per 1000) whose mothers were prescribed macrolides and 1666 of 95 973 children (17.36 per 1000) whose mothers were prescribed penicillins during pregnancy. Macrolide prescribing during the first trimester was associated with an increased risk of any major malformation compared with penicillin (27.65 v 17.65 per 1000, adjusted risk ratio 1.55, 95% confidence interval 1.19 to 2.03) and specifically cardiovascular malformations (10.60 v 6.61 per 1000, 1.62, 1.05 to 2.51). Macrolide prescribing in any trimester was associated with an increased risk of genital malformations (4.75 v 3.07 per 1000, 1.58, 1.14 to 2.19, mainly hypospadias). Erythromycin in the first trimester was associated with an increased risk of any major malformation (27.39 v 17.65 per 1000, 1.50, 1.13 to 1.99). No statistically significant associations were found for other system specific malformations or for neurodevelopmental disorders. Findings were robust to sensitivity analyses. Conclusions Prescribing macrolide antibiotics during the first trimester of pregnancy was associated with an increased risk of any major malformation and specifically cardiovascular malformations compared with penicillin antibiotics. Macrolide prescribing in any trimester was associated with an increased risk of genital malformations. These findings show that macrolides should be used with caution during pregnancy and if feasible alternative antibiotics should be prescribed until further research is available. Trial registration ClinicalTrials.gov NCT03948620

. Unadjusted and propensity-score-adjusted baseline characteristics (N [%]) of children whose mother were prescribed macrolides or penicillins from 14 gestation weeks to delivery ("the second to third trimester"). .  Table S6. Unadjusted and propensity-score-adjusted baseline characteristics (N [%]) of children whose mother were prescribed macrolides or penicillins from 4 gestation weeks to delivery ("in any trimester"). .  Table S7. Unadjusted and propensity-score-adjusted baseline characteristics (N [%]) of children whose mother were prescribed macrolides or penicillins 10 to 50 weeks before pregnancy. .  Table S11. Sensitivity analysis on the association between adverse child outcomes and macrolides versus penicillins prescribed during pregnancy: restricting to mothers whose antibiotics were prescribed to respiratory tract infections. .    Descriptive data 14 (a) Give characteristics of study participants (e.g., demographic, clinical, social) and information on exposures and potential confounders (b) Indicate the number of participants with missing data for each variable of interest (c) Cohort study -summarise follow-up time (e.g., average and total amount)  Table 2 Main results 16 (a) Give unadjusted estimates and, if applicable, confounder-adjusted estimates and their precision (e.g., 95% confidence interval). Make clear which confounders Text S2. Outcome identification.
The main outcomes of this study were major (any and five system-specific) malformations and four neurodevelopmental disorders.
Eligible outcomes for this study include those could potentially result from short-term fetal hypoxia.
We therefore included major malformations (any and system-specific malformations) and neurodevelopmental disorders. Malformations with specific known causes such as malformation resulted from maternal infections, fetal alcohol syndrome, Valproate syndrome and chromosomal malformations were not included. Twelve system-specific malformations were defined according to the European Surveillance of Congenital Anomalies (EUROCAT). (5) We then excluded 1) the musculoskeletal malformation (e.g. club foot, knock-knee and hip dislocation) as a system-specific malformation and as "any major malformation", because they are not reliably recorded in GP records (6); and 2) system-specific malformations that we had insufficient power to detect a 2-fold relative risk increase at 80% power (5% α level). Five out of the eleven systemspecific malformations from the EUROCAT classification fulfilled the power criterion according to its prevalence table and were analysed as system-specific malformations, including nervous system malformation, cardiovascular malformation, gastrointestinal malformation, genital malformation and urinary malformation (details were described in our protocol on www.clinicaltrials.gov [NCT03948620]). (7) Any of the eleven system-specific malformations (except for musculoskeletal malformations) was evaluated as "any major malformation", and identified from child GP records by 3 years old using Read codes which were mapped to the tenth edition of the International Classification of Diseases (ICD-10) code lists provided by EUROCAT. (5) Neurodevelopmental disorders (cerebral palsy, epilepsy, ADHD and ASD) were defined as the time to the first diagnostic or treatment code indicating the outcome by 14 years old. We identified potential cerebral palsy cases based on informative prescription or Read codes using the Random Forest approach, as we have previously described. (8) The potential cerebral palsy cases were then validated by a paediatric-neurologist (FC) blinded to the prenatal antibiotics exposure. Other neurodevelopmental disorders (epilepsy, ADHD and ASD) were identified using previously validated criteria using diagnostic codes and/or prescriptions (Supplementary Table S2).(9-11) Table S2. Codes for outcome identification.

Outcome Case identification
Any major congenital malformation Any major system specific malformation according to the EUROCAT classification. We use Read code lists mapped to ICD 10 codes Chapter Q. Exclude: 1) minor anomalies post-2005 * ; 2) malformations caused by known chromosomal abnormalities and teratogens (i.e. Teratogenic syndromes with malformations, Fetal alcohol syndrome, Valproate syndrome, Maternal infections resulting in malformations, Genetic syndromes + microdeletions, Chromosomal malformations); and 3) musculoskeletal malformations.

Cerebral palsy
Besides cases identified by ≥ 1 diagnostic code, we identified cerebral palsy cases from informative prescription or Read codes using the Random Forest approach and were validated by a paediatric-neurologist (FC) blinded to the prenatal antibiotics exposure. (8) Epilepsy Two prescriptions of antiepileptic drug (AED, identified based on British National Formula Chapter 4.8) within four months or ≥ 1 diagnosis (11) Attention deficit hyperactivity disorder (ADHD) ≥ 2 occurrence of prescriptions for ADHD (identified based on British National Formula Chapter 4.4) or diagnoses (attention deficit hyperactivity disorder, hyperkinetic disorders, hyperkinetic syndrome, hyperkinetic reaction of childhood or adolescence, overactive child syndrome and disturbance of activity and attention) within 4 month (9) Autism spectrum disorder (ASD) At least one diagnostic code ((infantile or childhood) autism, Asperger's syndrome, Rett's syndrome, Heller's syndrome, Autistic spectrum disorder, disintegrative disorder, and other pervasive developmental disorders) (10) *The mapping from ICD 10 code to Read code was performed using R package "CALIBERcodelists". EUROCAT revised its list of minor anomalies at 2005, and we applied the updated "Excluded minor anomalies post-2005" list in this study. GA: gestational age. PG03.11 Q750 8 *The frequencies were calculated for Read codes (not diagnosis) without de-duplication. In accordance with the confidentiality preserving policy of CPRD, we suppressed the information where the frequency cell contains <5 events (noted as "<5"). -Parity -Categorised as "0", and "≥ 1" Number of times that the women has given live-birth which were captured in the CPRD Mother Baby Link before the current pregnancy. Multiple births -"Singleton", and "(One of the) Twin, triplets, or quadruplets captured in the database". Alcohol misuse Most recent measurement from 10 years before pregnancy to the end of pregnancy.
"Yes" and "No" Alcohol misuse was defined as ≥ 14 units of alcohol per week, including moderate or severe drinker. Self-reported alcohol consumption was collected prospectively and coded by general practitioners or practice nurses on the consultation date in CPRD. The most recent alcohol consumption record was used to classify participants drinking behaviour, and "exdrinker" was categorised as not alcohol misuser if there was no evidence of alcohol withdraw before pregnancy start. Alcohol misuse was defined using: 1) One of the Read codes indicating alcohol consumption; or, 2) A prescription for disulfarim or acamprosate; or, 3) Self-reported average weekly alcohol intakes >= 14 units in the "Additional Clinical Details". We applied the code list of alcohol consumption developed by Bell at al. (12) Illicit drug use Most recent measurement from 10 years before pregnancy to the end of pregnancy.
"Yes" and "No" Illicit drug use was defined using: 1) One of the Read codes indicating drug use, addiction, and overdose; or 2) A prescription for methadone treatment. We assume that although a mother may stop using illicit drugs, the underlying behaviour was unlikely to vary significantly over time.

Obesity
Most recent measurement from 3 years before pregnancy till the end of the first trimester.
"Yes" and "No" Mothers who were obese prior to the 2 nd trimester of pregnancy were identified from the Read codes for obesity (or a BMI of ≥30 kg/m 2 -either directly entered or calculated from the most recent height measurement and median pre-pregnancy weight after excluding outliers. I.e. height outside the range 1-2m and weight outside the range 35-300kg, were removed). It was assumed that once a mother reached clinical obesity, the chance of her returning to a normal BMI in three years was minimal.
14 Tobacco use Most recent measurement from 3 years before pregnancy to the end of pregnancy.
"Yes" and "No" Tobacco use was defined as daily cigarette consumption of 1-100 cigarettes per day or other tobacco use. The most recent tobacco consumption record was used to classify participants drinking behaviour, and "ex-smoker" was categorised as non-recent smoker. Tobacco use was defined using: 1) One of the Read codes indicating tobacco consumption; or, 2) A prescription for smoking cessation aid; or, 3) Self-reported daily cigarette consumption of 1-100 cigarettes per day in the "Additional Clinical Details". Hypertension 50 weeks prior to delivery "Yes" and "No" Mothers with hypertension during pregnancy were identified based on 1) Systolic and diastolic blood pressure was above 140mmHg and 90mmHg, respectively, or, 2) One of the Read code for hypertension and associated diagnoses (including preeclampsia, eclampsia and HELLP syndrome), or, 3) One prescription for hypertension drugs from sections 2.

Genitourinary tract infection
During pregnancy "Yes" and "No" Common terms categorised as "Genitourinary tract infection" include urinary tract infection, cystitis, vaginitis and the prescription of Nitrofurantoin. Sexually Transmitted Infection During pregnancy "Yes" and "No" Common terms categorised as "Sexually Transmitted Infection" include chlamydia infection, trachoma, "TORCH" (Toxoplasmosis, Other agents such as HIV, Rubella, Cytomegalovirus and Herpes simplex) and other sexually transmitted infections (STIs). *When the key codes indicating a binary condition were not identified in the medical history of a subject, we classified the subject as absence of the condition. There were no missing for multi-categorical covariates in this study ("Age at delivery" and "Calendar year of delivery").
16 Table S5. Unadjusted and propensity-score-adjusted baseline characteristics (N [%]) of children whose mother were prescribed macrolides or penicillins from 14 gestation weeks to delivery ("the second to third trimester"). *Exposure propensity scores were measured as the predicted probability of receiving macrolides versus penicillins, conditional on the maternal and pregnancy related characteristics included in this table. 50 Strata were created based on the distribution of the propensity score of macrolides group. Weights for the penicillins group were calculated according to the distribution of the macrolides group among the strata and were used to estimate adjusted baseline characteristics. A meaningful between-group imbalance was assessed by an absolute standardised difference (Std.diff, the difference in means in units of standard deviation) of more than 0.1. Numbers in adjusted penicillins group were non-integer, because they were weighted based on the distribution of propensity score of macrolides group.
17 Table S6. Unadjusted and propensity-score-adjusted baseline characteristics (N [%]) of children whose mother were prescribed macrolides or penicillins from 4 gestation weeks to delivery ("in any trimester"). *Exposure propensity scores were measured as the predicted probability of receiving macrolides versus penicillins, conditional on the maternal and pregnancy related characteristics included in this table. 50 Strata were created based on the distribution of the propensity score of macrolides group. Weights for the penicillins group were calculated according to the distribution of the macrolides group among the strata and were used to estimate adjusted baseline characteristics. A meaningful between-group imbalance was assessed by an absolute standardised difference (St.diff, the difference in means in units of standard deviation) of more than 0.1. Numbers in adjusted penicillins group were non-integer, because they were weighted based on the distribution of propensity score of macrolides group.
18 Table S7. Unadjusted and propensity-score-adjusted baseline characteristics (N [%]) of children whose mother were prescribed macrolides or penicillins 10 to 50 weeks before pregnancy. *Exposure propensity scores were measured as the predicted probability of receiving macrolides versus penicillins, conditional on the maternal and pregnancy related characteristics included in this table. 50 Strata were created based on the distribution of the propensity score of macrolides group. Weights for the penicillins group were calculated according to the distribution of the macrolides group among the strata and were used to estimate adjusted baseline characteristics. A meaningful between-group imbalance was assessed by an absolute standardised difference (St.diff, the difference in means in units of standard deviation) of more than 0.1. Numbers in adjusted penicillins group were non-integer, because they were weighted based on the distribution of propensity score of macrolides group. 0.015 *The macrolides group included 7987 (clarithromycin), 494 (clarithromycin) and 151 (azithromycin) children. In accordance with the confidentiality preserving policy of CPRD, we suppressed the information where the frequency cell contains <5 events (noted as "<5") and where necessary to avoid deduction. For clarithromycin, we only analysed any major malformation due to the limited number of events of other adverse child outcomes (there were six events of the four neurodevelopmental disorders in total in children prenatally prescribed clarithromycin). 151 azithromycin were prescribed during the whole pregnancy with <5 events of malformation, which precluded the analyses. ADHD: attention-deficit/hyperactivity disorder; ASD: autism spectrum disorder; CI: confidence interval; RR: risk ratio; HR: hazard ratio. .5%) children in the study cohort were with non-missing duration of treatment. The macrolides group included 456 (<7 days, 1st trimester), 1376 (≥7 days, 1st trimester), 1634 (<7 days, 2nd -3rd trimester) and 4058 (≥7 days, 2nd -3rd trimester) children. The penicillins group included 8683 (<7 days, 1st trimester), 12592 (≥7 days, 1st trimester), 28314 (<7 days, 2nd -3rd trimester) and 40659 (≥7 days, 2nd -3rd trimester) children. In accordance with the confidentiality preserving policy of CPRD, we suppressed the information where the frequency cell contains <5 events (noted as "<5") and where necessary to avoid deduction. Within macrolides prescription during the 1st trimester, 95% prescriptions less than 7 days were of 5-6 days, and 93% prescriptions >= 7 days were of 7 days. Overall, 94.7% macrolides or penicillins prescriptions were of 5 to 7 days. ADHD: attention-deficit/hyperactivity disorder; ASD: autism spectrum disorder; CI: confidence interval; RR: risk ratio; HR: hazard ratio. , 3334 (macrolides, 2 nd -3 rd trimester), and 37592 (penicillins, 2 nd -3 rd trimester) children were included in the analyses. In accordance with the confidentiality preserving policy of CPRD, we suppressed the information where the frequency cell contains <5 events (noted as "<5") and where necessary to avoid deduction. Higher risks for genital malformation were observed for the both groups in the sibling cohort for unknown reason. ADHD: attention-deficit/hyperactivity disorder; ASD: autism spectrum disorder; CI: confidence interval; RR: risk ratio; HR: hazard ratio. .368 *In accordance with the confidentiality preserving policy of CPRD, we suppressed the information where the frequency cell contains <5 events (noted as "<5") and where necessary to avoid deduction. ADHD: attentiondeficit/hyperactivity disorder; ASD: autism spectrum disorder; CI: confidence interval; RR: risk ratio; HR: hazard ratio.

Text S3. Probabilistic multiple bias analysis on outcome misclassification and live-birth bias
The Clinical Practice Research Datalink (CPRD) has been used increasingly widely in pharmacoepidemiology studies within academic, regulatory, and pharmaceutical organisations to inform treatment guidelines and clinical practice guidance.(13) However, outcome measurements derived from administrative databases such as CPRD are not perfect and misclassification bias may exist. As CPRD data were prospectively collected as part of routine healthcare, it is reasonable to assume that measurement errors of outcomes were non-differential between macrolides and penicillins groups. This non-differential outcome misclassification is likely to bias the relative risk (RR) estimates towards the null. (14) Besides, we included only pregnancies that resulted in live-born children, thus some severe adverse outcomes (e.g. nervous system, cardiovascular and gastrointestinal malformations) that result in fetal deaths were missed. This depletion of affected fetuses may occur more often among women exposed to macrolides (versus penicillins), as shown in our systematic review (15). Therefore, risk ratio of these outcomes measured only in live births would be subject to selection (live-birth) bias with unknown direction.
We thus conducted probabilistic multiple bias analyses to quantify the bias due to outcome misclassification as well as jointly with live-birth bias to facilitate interpretation. Specifically, we estimated adjusted RR (95% CI) for each adverse child outcome for first-trimester macrolides (versus penicillins) prescribing using bias parameters stemming from both previous studies and educated guess.
Multiple bias analyses (which provided bias-adjusted RR estimates using standard 2x2 tables) were described in detail elsewhere (16). Briefly, frequencies in the tables were adjusted by a set of bias parameters, i.e. sensitivity and specificity for outcome misclassification, and probability of live birth for selection bias. These parameters were randomly sampled from given probability distributions (e.g. 5,000 iterations from triangular distributions in this study). In each iteration, we adjusted for misclassification bias and selection bias by sampling and adjusting the frequencies sequentially, incorporated with a random error to obtain the adjusted estimates with 95% limits. The analyses were performed using RStudio version 3.5.1 and R package "episensr". (17) The bias parameters used and bias-adjusted results were presented in Text S3. Table-1 and Text S3. Table-2, respectively.
Results show that given the assumptions described above, adjustment for the outcome misclassification and live-birth bias resulted in elevated RRs for malformations. The RR increased from 1.62 to 1.78 for cardiovascular malformations, and slightly from 1.55 to 1.58 for any major malformation. RRs for the nervous system and genital malformations increased and became statistically significant with wide 95% limits. The adjustment for outcome misclassification did not alter our findings for neurodevelopmental disorders. 26 Text S3. Table-1

Parameters
Evidence on bias parameters Distributions of bias parameters Outcome misclassification Sensitivity Major malformations: The CPRD primary care database was considered a more complete source to investigate major malformation compared with national malformation registry, because primary care follow up records for registered patients. In contrast, malformation registry data is based on voluntary reports and active follow-up which is subject to attrition. (18)(19)(20)(21) Based on our data, the prevalence of major malformation and major cardiovascular malformation were 17.0 and 6.3 per 1000 by the age of 3, respectively. These prevalence rates were slightly higher than those reported by the European Surveillance of Congenital Anomalies (EUROCAT) UK estimates (15. 3 and 4.3 per 1000). The prevalence of major cardiovascular malformation in our data was also consistent with other reports using CPRD, of 5.1 to 8.3 per 1000 from ages 1 to age 6 in CPRD. (19) Considering there would be a small portion of malformations diagnosed after age 3 years,(21) we hence assume a not perfect but high sensitivity of malformation in our study, e.g. 0.95, with the range from 0.90 to 1.

ADHD:
The prevalence estimates vary widely across studies. While the prevalence in screening studies using the Development and Well-Being Assessment (DAWBA) was 36 per 1000 boys and 9 per 1000 girls, studies based on CPRD reported much lower prevalence rates of ADHD ranging from 4.4 to 8.7 per 1000 boys, and 0.5 to 1.2 per 1000 girls. (9,24,25) We observed a prevalence of 7.5 per 1,000 boys and 1.4 per 1,000 girls in this study, comparable to other CPRD studies. The lower prevalence captured in primary care databases is not surprising though, as ADHD is believed to be an underdiagnosed and undertreated condition, with only 43.7%-54.1% children with current ADHD receiving medications in the US and UK. (26,27) We assumed a sensitivity from 0.50 to 0.90, with a mode of 0.70. Triangular (0.50, 0.70, 0.90)

ASD:
The prevalence is about 10 per 1000 for the whole population in the UK.(28) We observed a prevalence of 7.7 per 1,000 live births till age 14, and thus assumed a sensitivity from 0.77 to 1, with a mode of 0.89. Triangular (0.77, 0.89, 1)

Specificity
Specificity is not commonly measured for rarer outcomes in CPRD. However, a high specificity for all outcomes was expected in this study, due to both the low prevalence and the high positive predictive value (PPV). The high PPV of diagnosis in CPRD has been addressed by a number of studies. The PPV for major malformations, including cardiovascular malformations and hypospadias, has been reported to be 93% to 96% (31)(32)(33). The identification criteria we used for neurodevelopmental disorders have also been validated by previous researches in UK's primary care databases. (9,10,29) We thus assume a PPV of 95% for all outcomes in general population. Based on the definition of specificity, Triangular (0.997, 0.999, 1) We then assume a specificity for all outcomes from 0.997 to 1, with a mode of 0.999. Live-birth bias for the association between first trimester macrolides prescribing and severe malformations (i.e. nervous system malformation, cardiovascular malformation and gastrointestinal malformation) Probability of live-birth (selection) P (live-birth|(non-malformed, penicillin)): 0.83. Around 17% pregnancies were terminated with non-clinical indication. (30) We thus assumed that the probability of live birth in penicillins group without malformation was with a mode of 0.83, and a range of 10%. Triangular (0.78, 0.83, 0.88) P (live-birth|(malformed, penicillin)): 0.63, 0.73 and 0.78 for nervous system malformation, cardiovascular malformation and gastrointestinal malformation respectively. Based on estimated risk of termination, stillbirth, and first day neonatal death among cases with specific malformations, we assume 20%, 10% and 5% of cases with nervous system malformation, cardiovascular malformation and gastrointestinal malformation were dead before registration with the general practice.(31) Therefore, the probability of live birth is estimated to be 1-17%-(20%, 10% or 5%)=63%, 73% or 78% for cases with these three malformations, respectively. We estimated a range of 10%. P (live-birth|(non-malformed, macrolides)) = P (live birth|(non-malformed, penicillin))-10%=0.73. Based on our previous system review, where the pooled odds ratio for miscarriage between macrolides and penicillins was 1.82, we assumed that first trimester macrolides exposure would decrease the probability of live birth by up to 10% (based on a probability of miscarriage of 12% in penicillin group*82%), compared to penicillins in fetuses with or without malformation. (15) Triangular (0.68, 0.73, 0.78) P (live-birth|(malformed, macrolides)) = P (live birth|(malformed, penicillin))-10%=0.53, 0.63 and 0.68 for nervous system malformation, cardiovascular malformation and gastrointestinal malformation respectively.  The numbers of event in penicillins group were weighted based on the distribution of propensity score of macrolides group, which were used to calculate the adjusted risk/hazard ratio in the main analyses. b: Because the risk ratios for cerebral palsy, epilepsy, ADHD and ASD were comparable with the reported hazard ratios, we measured their risk ratios for simplicity. c: In accordance with the confidentiality preserving policy of CPRD, we suppressed the information where the frequency cell contains <5 events (noted as "<5") and where necessary to avoid deduction. CI: confidence interval; RR: risk ratio; HR: hazard ratio. ADHD: attention-deficit/hyperactivity disorder; ASD: autism spectrum disorder Table S12. Post-hoc analyses on the association between common specific malformation and macrolides versus penicillins prescribed during pregnancy. *Calculated in male babies. In accordance with the confidentiality preserving policy of CPRD, we only analyses outcomes where there were at least 5 cases in 1 st trimester or 2 nd to 3 rd trimester, macrolides group. We suppressed the information where the frequency cell contains <5 events (noted as "<5") and where necessary to avoid deduction. CI: confidence interval; RR: risk ratio; HR: hazard ratio.