{"paper_id":"318bb018-6005-468b-86fe-184104f18345","body_text":"1 \n \nTitle First trimester antidepressant use and miscarriage: a \ncomprehensive analysis in the UK Clinical Practice \nResearch Datalink \nAuthors’ \nnames \nFlorence Z. Martin (0000-0002-6804-008X), Paul Madley-Dowd \n(0000-0003-2932-9486), Viktor H. Ahlqvist (0000-0003-1383-\n3194), Gemma C. Sharp (0000-0003-2906-4035), Kayleigh E. \nEasey (0000-0002-3718-2915), Brian K. Lee (0000-0002-3635-\n8034), Abigail Merriel (0000-0003-0352-2106), Dheeraj Rai \n(0000-0002-7239-3523), Harriet Forbes (0000-0001-6888-2212)\n \nAuthors’ \ninformation  \nMRC Integrative Epidemiology Unit, Oakfield House, University \nof Bristol, BS8 2BN Florence Z. Martin PhD student MRC \nIntegrative Epidemiology Unit, University of Bristol, BS8 2BN \nPaul Madley-Dowd senior research associate Department of \nBiomedicine, Aarhus University, Høegh-Guldbergs Gade 10, \n8000 Aarhus, Denmark Viktor H. Ahlqvist postdoctoral fellow \nSchool of Psychology, University of Exeter, EX4 4QG Gemma \nC. Sharp associate professor of epidemiology School of \nPsychological Science, University of Bristol, BS8 1TU Kayleigh \nE. Easey senior lecturer Epidemiology and Biostatistics, Drexel \nUniversity Dornsife School of Public Health, PA 19104 Brian K. \nLee professor of epidemiology and biostatistics Centre for \nWomen’s Health Research, William Henry Duncan Building, 6 \nWest Derby Street, Liverpool, L7 8TX Abigail Merriel senior \nclinical lecturer in obstetrics Centre for Academic Mental Health, \nCanynge Hall, University of Bristol, BS8 2PN Dheeraj Rai \nprofessor of neurodevelopmental psychiatry Non-communicable \nDisease Epidemiology, London School of Hygiene and Tropical \nMedicine, Keppel Street, London, WC1E 7HT Harriet Forbes \nassistant professor  \nCorresponding \nauthor \nCorrespondence to: FZ Martin flo.martin@bristol.ac.uk \n \nFlorence Z. Martin 1,2, Paul Madley-Dowd 1,2,3, Viktor H. Ahlqvist 4,5, Gemma C. \nSharp1,6, Kayleigh E. Easey  1,7 , Brian K. Lee 8, Abi Merriel 9, Dheeraj Rai 1,2,3, Harriet \nForbes10 \n1 MRC Integrative Epidemiology Unit, University of Bristol, BS8 2BN \n2 Population Health Sciences, Bristol Medical School \n3 NIHR Bristol Biomedical Research Centre, University Hospitals Bristol and Weston \nNHS Foundation Trust and University of Bristol \n4 Department of Biomedicine, Aarhus University, Denmark \n5 Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden \n6 School of Psychology, University of Exeter, EX4 4QG \n7 School of Psychological Science, University of Bristol, BS8 1TU \n8 Epidemiology and Biostatistics, Drexel University Dornsife School of Public Health, \nPA 19104 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n  \n \n2 \n \n9 Centre for Womens Health Research, University of Liverpool \n10 Non-communicable Disease Epidemiology, London School of Hygiene and \nTropical Medicine \n \nABSTRACT  \nObjectives To investigate the risk of miscarriage associated with first trimester \nantidepressant use. \nDesign  Population-based cohort study. \nSetting  UK Clinical Practice Research Datalink (CPRD) GOLD. \nParticipants 661 825 individuals who had 1 021 384 pregnancies in CPRD GOLD \nbetween 1996 and 2018. \nMain outcome measures Miscarriage defined as a pregnancy loss prior to 24 \nweeks’ gestation. \nResults  Among the eligible pregnancies, 73 540 were prescribed \nantidepressants in trimester one (7.2%); 14.7% antidepressant prescribed \npregnancies ended in miscarriage, as opposed to 12.4% of those not prescribed \nantidepressants. Antidepressant use during trimester one was associated with \nmiscarriage in the unadjusted models (hazard ratio (HR) 1.21, 95% confidence \ninterval (CI) 1.19 to 1.23), which attenuated following adjustment for covariates (aHR \n1.04, 95% CI 1.02 to 1.06). These findings translated to an absolute risk adjusted for \nconfounders of 13.1% (95% CI 13.0 to 13.2) in the unexposed compared to 13.6% \n(95% CI 13.3 to 13.8) in the first trimester antidepressant exposed. The propensity \nscore matched model showed similar results (aHR 1.09, 95% CI 1.02 to 1.17, \nrespectively). In those with depression or anxiety in the 12 months before pregnancy, \nour estimate didn’t change (aHR 1.04, 95% CI 1.01 to 1.08).  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n3 \n \nConclusion  First trimester antidepressant use was associated with a small yet \nclinically insignificant increase in risk of miscarriage, with no evidence suggesting \ntaking antidepressants before pregnancy and into first trimester increases the risk of \nmiscarriage. The conclusions are less clear for ‘incident’ antidepressant use in \ntrimester one, however issues including gestational dating in early pregnancy and \nprobable residual confounding prohibit us from interpreting this observation as \ncausal. \n1 I NTRODUCTION  \nAntidepressant use during pregnancy is prevalent in many countries, with estimates \nsuggesting that upwards of 8% of pregnant people in the United Kingdom,\n1 Iceland,2 \nand the United States 3 use antidepressants at some point during pregnancy. \nAlthough most antidepressants are not contraindicated during pregnancy, they are \nprescribed with some caution,4 due to evidence suggesting small increases in risk of \nmiscarriage5-7 and other adverse outcomes, such as preterm delivery and \npostpartum haemorrhage.8 9  In the UK, the National Institute for Health and Care \nExcellence (NICE) updated its guidance in 2023 from severity-based advice to \npatient-centred decision-making when planning pregnancy or becoming pregnant on \nantidepressants, weighing up risks to both mum and baby on an individual basis.\n10-12 \nGlobally, the guidance around using antidepressants during pregnancy is mixed, 13 \nreflecting the uncertainty in the evidence base and in turn, challenges faced by \nprescribing clinicians. \nThe definition of miscarriage varies in different countries and time periods, but is \noften defined as a pregnancy loss before 20–24 weeks’ gestation. 14 15 A systematic \nreview and meta-analysis of 29 studies identified a modest increased risk of \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n4 \n \nmiscarriage following any antidepressant use during pregnancy (pooled odds ratio \n1.24, 95% confidence interval (CI) 1.18 to 1.31). 16 Biologically, it is plausible that \nantidepressants could causally increase the risk of miscarriage, due to their inhibition \nof platelets and subsequent association with increased bleeding events. 17 However, \nuntreated depression and anxiety during pregnancy are also associated with adverse \npregnancy outcomes, like preterm birth and low birthweight.\n18-21 Thus, it is plausible \nthat the link between antidepressant use during pregnancy and adverse outcomes \nlike miscarriage, could be explained by the underlying disease for which \nantidepressants are prescribed,\n1 22 rather than the drugs themselves; this concept is \nknown as confounding by indication. Given the use of general population, non-\nindicated controls in many of the included studies in the above systematic review, 16 \nand some studies omitting adjustment for underlying reason for prescribing, 23-29 it \nisn’t possible to conclude a causal relationship between antidepressants and \nmiscarriage from the present literature. \nIn this cohort study, we used Clinical Practice Research Datalink (CPRD) GOLD \ndata, with linked Hospital Episode Statistics (HES) data where available, to \ninvestigate the relationship between first trimester use of antidepressants and the \nrisk of miscarriage\n using a range of advanced methodological approaches, including \nan exposure discordant pregnancy design, propensity score matching, and stratified \nanalyses, to help account for confounding by and severity of the underlying disease. \n2 M ETHODS  \n2.1 D ATA SOURCES  \nCPRD GOLD is a UK-wide repository of anonymised general practice data and \nmakes up part of one of the largest resources of primary care data in the world. 30 It \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n5 \n \ncontains over 4 million active patients and covers ~7% of the UK’s population, \nrepresentative by age, sex, and ethnicity.30 The primary care data in CPRD GOLD is \nlinked to the Office for National Statistics death registration data and practice-level \nIndex of Multiple Deprivation (IMD) scores. For most English patients, CPRD GOLD \nis linked with Hospital Episode Statistics (HES) Admitted Patient Care (APC), \ncovering inpatient hospital episodes.\n30 \nThe CPRD GOLD Pregnancy Register has been described in detail elsewhere,31 but \nin short, it is a dataset that contains pregnancy episodes with affiliate estimated \npregnancy dates, outcome, and patient identifiers derived from the CPRD GOLD \nprimary care data. The data sources used in this study are detailed in Methods S2.1. \n2.2 S TUDY POPULATION  \nTo derive the study population, eligibility criteria were imposed on the entire CPRD \nGOLD population who had a record of at least one pregnancy episode between 1996 \nand 2018 in the Pregnancy Register. We cleaned the Pregnancy Register in \naccordance with recommendations from the authors of the Register algorithm, \nincluding removing conflicting and historical pregnancies.\n32 \nHES data were used to supplement pregnancy outcomes that were uncertain in the \nPregnancy Register (namely ‘unknown outcome’ and ‘unspecified loss’). Pregnancy \ndates in the Pregnancy Register were then amended using imputed values as \nimposed by the Pregnancy Register algorithm (Methods S2.2). Pregnancy episodes \nending in an ‘unknown outcome’ that were not recoverable using HES were \nexcluded. \nCPRD imposes an ‘up-to-standard’ (UTS) date on all enrolled practices, which \nrecords the date on which the practice began to contribute ‘high quality’ data to \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n6 \n \nCPRD, defined by several indicators. 30 We only included those who were registered \nwith a UTS practice and had adequate follow-up for at least a year prior to \npregnancy and up until the end of pregnancy; by extension, eligible individuals did \nnot transfer out of their practice or have a death date recorded prior to the end of \npregnancy. Individuals were also excluded if the last data collection date from their \npractice occurred before the end of their pregnancy.  \n2.3 E XPOSURE  \nAll antidepressants that are approved for treating depression in the UK were \nextracted from primary care prescriptions (Table S1). Briefly, quantity (total number of \ntablets prescribed) and daily dose (number of tablets taken per day) were used in \nconjunction with the prescription start date to estimate the prescription end date \n(Methods S2.3). These dates were compared with the pregnancy start date and the \nend date of trimester one to identify whether a prescription occurred within or \noverlapped with the first trimester to identify ‘exposed’.  \nThose with a prescription for antidepressants in the three months before pregnancy \nand during trimester one were defined as ‘prevalent’ users. Those without \nantidepressants in the three months prior to pregnancy but prescribed during \ntrimester one were defined as ‘incident’ users. \nWe identified antidepressant class prescription in trimester one: selective serotonin \nreuptake inhibitors (SSRI), tricyclic antidepressants (TCA), serotonin-noradrenaline \nreuptake inhibitors (SNRI), ‘other’ (Table S1), or multiple classes (i.e., those who \nswitched product class during trimester one or used multiple antidepressants from \ndifferent classes simultaneously). \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n7 \n \nFinally, we defined dose, standardised for each medication using the individual dose \ndistribution in milligrams and using percentiles to ascertain low ( ≤ 25th percentile), \nmedium, and high (>75th percentile) doses (Methods S2.3). In instances where \nmultiple doses were prescribed in trimester one, individuals were classified as the \nhighest dose they received in first trimester. \n2.4 O UTCOMES  \nThe outcome of each pregnancy episode was available in the Pregnancy Register. 31 \nMiscarriage from the Pregnancy Register was used as the outcome in this study.  \n2.5 C OVARIATES  \nConfounders were chosen based on subject matter knowledge of whether a \ncovariate could feasibly cause both the exposure and the outcome. The primary \nadjustment set contained age, year of pregnancy (‘96–’00, ‘01–’05, ‘06–’10, ‘11–’15, \n‘16–’18), IMD quintile, history of miscarriage and severe mental illness, smoking \n(non-, ex-, current smoker), parity (0, 1, 2, \n≥ 3), use of high dose folic acid, \nantipsychotics and anti-seizure medications in the 12 months before pregnancy, \nnumber of primary care consultations in the 12 months before pregnancy (0, 1–3, 4–\n10, >10), and depression and anxiety ever before the start of pregnancy; this is \ndescribed detail in Methods S2.5.  \nDepression and anxiety were identified using pre-defined, expert verified codelists in \nprimary care (Read codes) and HES APC (ICD-10 codes) (Methods S2.5).  \nEthnicity (White, South Asian, Black, Other, Mixed) 33 and body mass index (BMI; \n<18, 18–24.9, 25–29.9, >30kg/m 2) around the start of pregnancy contained >10% \nmissing data; thus, they were dropped from the primary adjustment set and included \nin a sensitivity analysis, described below.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n8 \n \n2.6 A NALYSIS  \nWe described baseline characteristics of the eligible sample by first trimester \nantidepressant use. We also compared the characteristics between eligible \npregnancies and those excluded having ended in ‘unknown outcome’. We ran the \nprimary and secondary analyses in complete records for covariates. \n2.6.1 P RIMARY ANALYSES  \n2.6.1.1 M ULTIVARIABLE COX MODEL  \nWe compared antidepressant exposed to unexposed, using crude and adjusted Cox \nmodels estimating hazard ratios (HR) and 95% CIs. Follow-up began on the \nestimated pregnancy start date; ‘incident’ users contributed unexposed time to the \nanalysis until the start of their antidepressant prescription. The end of follow-up was \nset to the first of either the outcome (miscarriage), other loss (Table S4), reaching \n168 days gestation, or study end (31\nst December 2018). We employed cluster-robust \nstandard errors (clustered by pregnant individual) to account for those who \ncontributed multiple pregnancies to the analysis.  \nTo enhance clinical interpretability, we estimated the absolute confounder-adjusted \nrisks (1-Survival) using Breslow's baseline estimator and integrated these with the \nhazard ratios through G-formula and bootstrapping for standard errors (1000 \nrepetitions). \nIn addition to adjusting for indication, we ran the model restricted to those with \nevidence of depression or anxiety in the 12 months prior to pregnancy. To further \ninvestigate severity, we restricted the model to those with ‘severe’ depression or \nanxiety, as defined by administered scale standardised scores (like PHQ-9, Methods \nS2.6) in the 12 months before pregnancy.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n9 \n \nWe additionally restricted the sample to those who were prescribed antidepressants \nin the three months prior to pregnancy; we compared those who continued \nantidepressants into trimester one to those who discontinued treatment prior to the \nstart of pregnancy. This assumed that those using antidepressants pre-pregnancy \nwere more characteristically similar, thus lessening residual confounding. \n2.6.1.2 E XPOSURE DISCORDANT PREGNANCIES  \nWe held genetic liability to miscarriage and variables that were not time-varying \nbetween pregnancies fixed by comparing pregnancies among the same individual in \nan exposure discordant design as an additional approach to manage confounding.\n34 \n35 We used a stratified Cox model adjusted for the primary adjustment set (except \nhistory of miscarriage), where each stratum in the model represented an individual \nwith ≥ 2 exposure discordant pregnancies (Methods S2.6).  \n2.6.1.3 P ROPENSITY SCORE MATCHING  \nWe performed propensity score matching, following the stepwise process laid out by \nDesai et al. 36 The propensity score included both putative confounders and \npredictors of the outcome (Table S2).37 \nThe sample in the propensity score matched analysis only included first pregnancies, \nto avoid individuals being matched to their own subsequent pregnancies. We used \nlogistic regression to estimate a propensity score, then 1:1 matched each \nantidepressant exposed pregnancy to an unexposed pregnancy without replacement \nusing a caliper of 0.2, and exact matching on number of CPRD consultations in the \n12 months before pregnancy (Methods S2.6).  \n2.6.2 S ECONDARY ANALYSES  \n2.6.2.1 ‘P REVALENT ’ AND ‘INCIDENT ’ ANALYSIS  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n10 \n \n‘Prevalent’ and ‘incident’ antidepressant users were compared to unexposed. We \nalso performed this analysis among those with any depression or anxiety and \n‘severe’ illness (Table S3) in the 12 months before pregnancy. \n2.6.2.2 C LASS AND DOSE ANALYSES  \nWe compared different antidepressant classes to unexposed. We also compared \nlow, medium, and high doses of antidepressant in trimester one to no use (Methods \nS2.3).  \n2.6.3 S ENSITIVITY ANALYSES  \nWe restricted the primary and secondary analyses to those with linked secondary \ncare data, due to pregnancy outcome modifications and high data availability in this \ngroup (Methods S2.2). We also performed the primary Cox model where exposure \nwas redefined as \n≥ 2 antidepressant prescriptions in trimester one to reduce potential \nexposure misclassification. \nWe investigated the association between pre-pregnancy depression, anxiety, \nantidepressant use and having a pregnancy that ended in an ‘unknown outcome’ to \nassess the potential for differential pregnancy exclusion from the sample. \nHaving dropped ethnicity and BMI from the adjustment set due to >10% missing \ndata, we included these covariates in sensitivity analysis. We also investigated the \nassociation between missingness in these variables and experiencing a miscarriage \nto assess the potential introduction of bias in the complete records analysis.\n38 \nTo account for the potential effect of behavioural changes between pregnancies \nwhere the outcome of one pregnancy influences care seeking and provision in the \nnext pregnancy, we restricted the primary Cox model to first pregnancies. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n11 \n \nTo test the impact of censoring pregnancies that ended in other types of early \npregnancy loss, including ectopic and molar pregnancies (Table S4), we absorbed \nthem into the miscarriage category in sensitivity analysis. \nAll analyses were performed in Stata 17 and R 4.3.1. This study was approved by \nCPRD’s Independent Scientific Advisory Committee (ISAC) in 2021 [ISAC number: \n21_000362]. \n2.7 P ATIENT AND PUBLIC INVOLVEMENT  \nNo patients were directly consulted regarding the definition of the research question, \nstudy design, analyses, or write-up. We shared our plans at public engagement \nevents, including the Pint of Science festival.39 We consulted with clinical colleagues \nand the British Pregnancy Advisory Service (BPAS) who are in regular discussion \nwith pregnant people concerned about the risk of miscarriage following first trimester \nantidepressant use. This provided sufficient motivation for the importance of the \npresent study to individuals of child-bearing age considering antidepressant \ntreatment. \n3 R ESULTS \n3.1 S TUDY POPULATION  \nThe CPRD GOLD Pregnancy Register contained 1 245 146 non-conflicting \npregnancies that began between 1996 and 2018 with sufficient follow-up. Upon the \nexclusion of ‘unknown outcome’ and multiple pregnancies, 1 021 384 pregnancies \n(among 661 825 individuals) were eligible (Figure 1). Pregnancy outcomes in the \neligible sample are summarized in Table S4. \n3.2 P OPULATION CHARACTERISTICS  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n12 \n \nFollowing exclusions, 73 540 were prescribed antidepressants during trimester one \n(7.2%). Those prescribed antidepressants during trimester one were slightly older \n(22.7% v 19.7% over 35) and were more likely to be obese (25.2% v 17.0%) than \nunexposed. Individuals prescribed antidepressants during trimester one were more \nlikely to be from the most deprived IMD quintile (30.7% v 26.9%) than unexposed. \nThose prescribed antidepressants were more likely to visit their doctor over 10 times \n(53.7% v 19.1%) and be using other medications (e.g., mood stabilisers, 4.1% v \n0.7%) in the 12 months prior to pregnancy, be multiparous, and be current smokers \nthan unexposed (Table 1). \nThose excluded due to an ‘unknown outcome’ pregnancy were broadly similar to the \neligible individuals, other than higher amounts of missing data in certain variables \nand on average more doctor visits before pregnancy (Table S5).  \n3.3 P RIMARY COX MODEL  \nAmong 967 925 complete record pregnancies, 71 460 were exposed to \nantidepressants in trimester one, with 14.6% ending in miscarriage, compared to \n12.3% of the 896 465 unexposed pregnancies (unadjusted HR 1.21, 95% CI 1.19 to \n1.23). Upon adjustment, the difference between groups decreased (adjusted HR \n(aHR) 1.04, 95% CI 1.02 to 1.06), with a standardized miscarriage risk of 13.6% \n(95% CI 13.3 to 13.8) in the exposed group and 13.1% (95% CI 13.0 to 13.2) in the \nunexposed group (Figure 2). This finding was consistent when we required \n≥ 2 \ndistinct antidepressant prescriptions in trimester one to be considered exposed (aHR \n1.02, 95% CI 1.00 to 1.05) (Table S6).  \nWhen restricting to those with depression or anxiety noted in the 12 months prior to \npregnancy (n=99 820) and those with “severe” depression (n=9170), we observed \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n13 \n \nsimilar results to the primary analysis (aHR 1.04, 95% CI 1.01 to 1.08 and aHR 1.02, \n95% CI 0.92 to 1.14, respectively) (Table S7).  \nWhen comparing those who continued antidepressants into trimester one (n=60 167) \nto those who discontinued in the three months before pregnancy (n=24 410), we \nobserved no difference in hazard of miscarriage (aHR 1.00, 95% CI 0.97 to 1.04) \n(Table S8).  \n3.4 E XPOSURE DISCORDANT PREGNANCY ANALYSIS  \nWhen comparing exposure discordant pregnancies within the same birthing parent \n(n=78 072), thereby accounting for all unobserved (e.g., genetics and many \nenvironmental factors that don’t change between pregnancies within an individual) \nand observed stable confounders, we saw an effect in line with the unadjusted \nprimary Cox model (aHR 1.20, 95% CI 1.16 to 1.25) (Figure 2). \nTo understand whether this may have been driven by order of pregnancy, we \ninvestigated the risk of miscarriage when the first pregnancy in the exposure \ndiscordant group of pregnancies was exposed and then when a subsequent \npregnancy in the group was exposed. This revealed that first trimester \nantidepressant use was only associated with miscarriage when the first pregnancy in \nthe group was exposed (aHR 1.98, 95% CI 1.82 to 2.16), not when a subsequent \npregnancy was exposed (aHR 0.97, 95% CI 0.93 to 1.02) (Figure 3, Table S10). The \nstark difference between these results suggests that pregnancy order may be driving \nthe result observed in the exposure discordant analysis as opposed to the \nmedication itself. \n3.5 P ROPENSITY SCORE MATCHING  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n14 \n \nWhen matching pregnancies on propensity score (n=25 026) where more measured \nconfounders were accounted for, our findings were consistent with those from the \nprimary Cox model (aHR 1.09, 95% CI 1.02 to 1.17) (Figure 2). \n3.6 ‘P REVALENT ’ AND ‘INCIDENT ’ ANALYSIS  \nIn the unadjusted models, both ‘prevalent’ and ‘incident’ use in trimester one was \nassociated with an increased hazard of miscarriage (HR 1.19, 95% CI 1.17 to 1.22 \nand HR 1.29, 95% CI 1.23 to 1.35, respectively). Interestingly, adjustment for \ncovariates only changed our conclusions for ‘prevalent’ use, not ‘incident’ use (aHR \n1.00, 95% CI 0.98 to 1.03 and aHR 1.24, 95% CI 1.19 to 1.30, respectively) (Figure \n2), despite similarity across their measured characteristics (Table S9). Our \nconclusions did not change when restricting to those with \n≥ 2 prescriptions in \ntrimester one or when depression or anxiety were noted in the 12 months before \npregnancy (Table S6, Table S7).  \n3.7 C LASS AND DOSE ANALYSIS  \nSSRI, SNRI, and ‘other’ antidepressant use during trimester one were associated \nwith a slight increase in risk of miscarriage as compared to no use (Figure 2). Low \nand medium dose were associated with miscarriage, where high dose attenuated to \nthe null as compared to unexposed following adjustment for covariates (Figure 2). \n3.8 S ENSITIVITY ANALYSES  \nOur results were consistent when restricting each analysis to those with linked data \n(Table S12). Having depression noted in the 12 months before pregnancy was \nmodestly associated with having an ‘unknown outcome’ pregnancy (Table S13). \nWhen adding ethnicity and BMI to the adjustment set for the primary Cox model, our \nestimates didn’t change (aHR 1.03, 95% CI 1.01 to 1.06) (Table S14). When \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n15 \n \nassessing whether the complete records analysis may have been biased, 38 those \nwho had a miscarriage were more likely to have missing data in ethnicity, BMI, and \nsmoking around the start of pregnancy (Table S15). \nSimilarly, when we restricted to first pregnancies, our estimates didn’t change \nsubstantially (aHR 1.07, 95% CI 1.03 to 1.10) (Table S16). When including ectopic \nand molar pregnancies into the definition of the outcome, the results were consistent \nwith the primary analysis (aHR 1.03, 95% CI 1.01 to 1.05) (Table S17). \n4 D ISCUSSION  \n4.1 P RINCIPAL FINDINGS  \nThis large population-based cohort study of nearly one million pregnancies in the UK \nfound no clear evidence that first trimester antidepressant use substantially \nincreases the risk of miscarriage, with no evidence suggesting taking \nantidepressants before pregnancy and into trimester one increases the risk of \nmiscarriage. The conclusions are less clear for ‘incident’ first trimester \nantidepressant use, however issues including gestational dating in early pregnancy \nand probable residual confounding prohibit us from interpreting this observation as \ncausal. The small observed increases in absolute risk, even if causal, are potentially \nclinically insignificant. The findings from the exposure discordant pregnancy analysis \npoint to the importance of pregnancy order. \n4.2 P REVIOUS LITERATURE  \nPrevious literature exploring antidepressant use during pregnancy has suggested a \nslight increased risk of miscarriage, as shown by a systematic review and meta-\nanalysis of 29 studies by Smith et al . which noted a number of methodological \nweaknesses in the previous literature.\n16 A large Danish study found an association \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n16 \n \nbetween SSRI use during pregnancy and trimester one miscarriage of a similar \nmagnitude to the findings presented here. 6 They conclude that unmeasured lifestyle \nfactors and confounding by indication are responsible for the small remaining \nassociation between SSRIs and miscarriage, given that they observed a complete \nattenuation of the effect to the null when comparing antidepressant exposure to \nunmedicated depression during pregnancy.\n6 Another study highlighted the \nchallenges faced in the field of pregnancy pharmacoepidemiology, particularly when \nattempting to deal with confounding by indication.\n40 It is plausible that those who with \n‘active’ depression (for example), have a higher baseline risk of miscarriage than \nthose who do not have depression. If this is not properly handled in analyses of \nantidepressants and miscarriage, there is likely to be residual confounding by \nseverity of indication.  \nStudies typically have accounted for indication by conducting additional analyses \ncomparing those on antidepressant treatment during pregnancy with those who have \nunmedicated depression; some have found a complete attenuation to the null,\n6 40  \nwhereas others have found a persistent risk of miscarriage following antidepressant \nuse.5 41-43  Some have compared medication classes to account for confounding by \nindication, whereby both the “exposure” and “comparator” groups are likely to have \nan indication for antidepressants because they’re all exposed to antidepressants. \nThree studies have leveraged the comparison between SNRIs and other \nantidepressants,\n44-46 whereby a pooled increased risk of miscarriage was observed \nfor those taking SNRIs. 16 However, SNRI antidepressants are not first-line \ntreatments in the UK, thus those prescribed them during pregnancy are likely to be \nmore unwell. Only four studies in the review included variables pertaining to \nindication in a multivariable model. 40 46 47  This highlights the persistent problem of \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n17 \n \nconfounding by severity of indication that is rarely eliminated even when accounting \nfor indication; furthermore, those that use an indication-based sample may be prone \nto bias amplification.48 It is interesting that the unadjusted estimate from the present \nstudy, HR 1.21 (1.19-1.23), is similar to the summary estimate observed in the above \nreview: pooled OR 1.24 (1.18 to 1.31).16 \nWe show some novel findings surrounding ‘incident’ use of antidepressants during \ntrimester one and miscarriage, where we observed a higher risk compared to no use \nand the confidence intervals do not overlap with the ‘prevalent’ use v no use. This \nintriguing association of ‘incident’ but not ‘prevalent’ use of antidepressants has been \nobserved previously for some neurodevelopmental outcomes.\n49 Although these \nfindings could be causal, whereby the introduction of a new drug substance into the \nbody could disrupt early fetal development and result in an early pregnancy loss, \nthere are several other plausible explanations that could explain the finding. As \ndiscussed above, residual confounding by severity of indication, health-seeking \nbehaviour, or data artefacts like the imputation of pregnancy length for most losses \nmight be partially driving the association. Antidepressant initiation symptoms such as \nheightened anxiety\n50 and the ongoing experience of symptoms during the time taken \nfor antidepressants to start working 51 may present alternative mechanisms for an \nincreased risk of miscarriage that should also be considered when interpreting the \nfinding for ‘incident’ users.  \n4.3 S TRENGTHS AND WEAKNESSES  \nThis study has several strengths. It is large, with over 600 000 individuals from a UK-\nrepresentative sample, 30 contributing nearly one million pregnancies over two \ndecades, improving precision of our estimates. It leverages multiple methods and \ncomparators to explore the role of confounding by indication and data issues \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n18 \n \nencountered when performing observational pregnancy pharmacoepidemiology \nstudies, particularly in early pregnancy loss. The use of the CPRD GOLD Pregnancy \nRegister allowed us to build on the systematic approach taken by Minassian et al. \nthat extracted pregnancy records from individuals who had been pregnant in the \nCPRD GOLD database.31 The use of cause-specific time-to-event models allowed us \nto retain pregnancies that were at risk of miscarriage while ongoing, but neither \nended in the outcome nor reached the end of follow up, i.e., had ended in a non-\nmiscarriage loss before week 24. It is important to consider the impact of their \ninclusion here; by keeping them in, we did not differentially deflate the denominator \nby exposure status and thus artificially inflate the proportion of pregnancies among \nthe exposed group that ended in miscarriage. This omission from previous studies \nmay have partially driven reports of an increased risk of miscarriage following use of \nantidepressants, even among studies that had adjusted for confounders.  \nThe study also has several limitations. Although the CPRD GOLD population is \nlarge, the application of eligibility criteria based on registration in a UTS practice and \nquality of patient data inevitably led to a smaller and more select sample of \nindividuals. We can be reassured that those excluded for having an ‘unknown \noutcome’ were similar characteristically to those included, but the findings likely only \ngeneralise to those that fulfil the criteria for this study, namely staying with the same \npractice for a year before and throughout pregnancy.  \nResidual confounding is likely present in these analyses despite our mixed \napproaches to accounting for it. Although we managed confounding by indication as \ncompletely as possible, like adjusting for depression and anxiety in the main analysis \nand restricting to those visiting the doctor for depression and/or anxiety or those \nhaving scored highly on depression and anxiety scales in the 12 months before \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n19 \n \npregnancy, it remains difficult to capture indication severity using CPRD. Thus, \nresidual confounding by underlying severity of indication for treatment surely \ncontributed to the results we observed, particularly for ‘incident’ use.  \nSystematic bias in these analyses cannot be ruled out. Differential exposure \nmisclassification was a concern in this study as pregnancies ending in miscarriage \nwere more likely to have an imputed gestational length than deliveries 31 and \ntherefore at higher risk of being misclassified as prescribed antidepressants in \ntrimester one. The results may have been biased in either direction if this type of bias \nwas present in these analyses.\n52 It is plausible that, given the imputation of \ngestational length for many losses, antidepressant prescriptions were sought having \nexperienced a miscarriage. Due to the derivation of pregnancy dates via a \npregnancy algorithm, the possibility for reverse causation may explain some of the \nmiscarriages observed in the ‘incident’ group. Finally, ascertainment bias is likely at \nplay here. Those seeking healthcare for depression, anxiety, or other indications \ntreated with antidepressants may be more likely to report pregnancies and early \npregnancy losses than those not engaging with healthcare for other reasons.  \nGiven that the presence and magnitude of each of these limitations cannot be easily \nquantified, it is reassuring that even if the finding from the main analysis was causal, \nit would translate to a modest increase in absolute risk from 13.1% in the unexposed \nto 13.6% in the exposed (i.e., a number need to harm of 200).  \n4.4 F UTURE WORK  \nWhere high quality miscarriage data are available, other causal inference \napproaches that aim to manage time-related biases like target trial emulation would \nbe useful to explore the finding for ‘incident’ use. Given the issue of time in these \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n20 \n \ndata, where most miscarriages have an imputed gestational length, CPRD GOLD \nmay not be appropriate for this. It is important to understand this finding to \nadequately inform individuals who may need to initiate antidepressants in the first \ntrimester. \n4.5 C ONCLUSIONS  \nWe found no clear evidence that antidepressant use during trimester one \nsubstantially increases the risk of miscarriage, with no evidence suggesting taking \nantidepressants before pregnancy and into trimester one increases the risk of \nmiscarriage. Although we observed a slight increased risk of miscarriage when \ncomparing ‘incident’ antidepressant use in trimester one to no use, the overall \nrelative risk translates to a modest increase in absolute risk and other biases cannot \nbe ruled out. Our findings suggest that antidepressants do not substantially increase \nthe risk of miscarriage for women on antidepressants when they become pregnant. \nWhat is already known on this topic \n- Antidepressant use during pregnancy was shown to increase the risk of \nmiscarriage according to a recent systematic review. \n- Confounding, including by indication, remains a pervasive problem in the \ninterpretation of the current evidence. \nWhat this study adds \n- A comprehensive analysis of first trimester antidepressant use and risk of \nmiscarriage in the CPRD GOLD Pregnancy Register, including multiple \napproaches to address confounding. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n21 \n \n- Estimates of standardised absolute risk of miscarriage among antidepressant \nexposed and unexposed to antidepressants in trimester one to aid in clinical \ninterpretability of the findings. \n- The results, particularly for ‘prevalent’ antidepressant use, are reassuring and \nsupport minimal risk of miscarriage following ongoing use of antidepressants \ninto trimester one from pre-pregnancy. \n5 E THICAL STATEMENT  \nThis study was approved by CPRD’s Independent Scientific Advisory Committee \n(ISAC) in 2021 [ISAC number: 21_000362]. \n6 D ATA AVAILABILITY STATEMENT  \n“Access to CPRD data, including UK Primary Care Data, and linked data such as \nHospital Episode Statistics, is subject to protocol approval” as per CPRD’s \nguidelines. Authors are unable to share the data in its raw form, however all  \nanalytical code and codelists are open-source and found via the following links: \nhttps://github.com/flozoemartin/Miscarriage\n and \nhttps://github.com/flozoemartin/codelists.  \n7 F UNDING \nFZM was supported by the Wellcome Trust (Grant ref: 218495/Z/19/Z). DR, BKL and \nHF acknowledge support from the NIH (1R01NS107607). GCS was supported by a \nMedical Research Council (MRC) grant (MR/S009310/1). The views expressed in \nthis publication are those of the author(s) and not necessarily those of the NHS, the \nNational Institute for Health Research, MRC, or the Wellcome Trust. FZM, PM-D, \nVHA, KEE, GCS, and DR are members of the UK MRC Integrative Epidemiology \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n22 \n \nUnit, which is funded by the MRC (MC_UU_00011/1, MC_UU_00011/3 and \nMC_UU_00011/7) and the University of Bristol.  \n8 R EFERENCES  \n1. Martin FZ, Sharp GC, Easey K, et al. Patterns of antidepressant prescribing in and \naround pregnancy: a descriptive analysis in the UK Clinical Practice Research \nDatalink. medRxiv 2024:2024.08.08.24311553. doi: \n10.1101/2024.08.08.24311553 \n2. Zoega H, Kieler H, Norgaard M, et al. Use of SSRI and SNRI Antidepressants \nduring Pregnancy: A Population-Based Study from Denmark, Iceland, Norway \nand Sweden. PLoS One  2015;10(12):e0144474. doi: \n10.1371/journal.pone.0144474 [published Online First: 20151214] \n3. Huybrechts KF, Palmsten K, Mogun H, et al. National trends in antidepressant \nmedication treatment among publicly insured pregnant women. Gen Hosp \nPsychiatry 2013;35(3):265-71. doi: 10.1016/j.genhosppsych.2012.12.010 \n[published Online First: 2013/02/05] \n4. Eakley R, Lyndon A. Antidepressant use During Pregnancy: Knowledge, Attitudes, \nand Decision-Making of Patients and Providers. J Midwifery Womens Health  \n2022;67(3):332-53. doi: 10.1111/jmwh.13366 [published Online First: \n20220510] \n5. Kitchin A, Huerta C, Llorente-Garcia A, et al. The role of prenatal exposure to \nantidepressants, anxiolytic, and hypnotics and its underlying illness on the risk \nof miscarriage using BIFAP database. Pharmacoepidemiol Drug Saf  \n2022;31(8):901-12. doi: 10.1002/pds.5488 [published Online First: \n2022/06/11] \n6. Johansen RL, Mortensen LH, Andersen AM, et al. Maternal use of selective \nserotonin reuptake inhibitors and risk of miscarriage - assessing potential \nbiases. Paediatr Perinat Epidemiol 2015;29(1):72-81. doi: 10. 1111/ppe.12160 \n[published Online First: 2014/11/11] \n7. Andersen JT, Andersen NL, Horwitz H, et al. Exposure to selective serotonin \nreuptake inhibitors in early pregnancy and the risk of miscarriage. Obstet \nGynecol 2014;124(4):655-61. doi: 10.1097/AOG.0000000000000447 \n[published Online First: 2014/09/10] \n8. Sujan AC, Rickert ME, Oberg AS, et al. Associations of Maternal Antidepressant \nUse During the First Trimester of Pregnancy With Preterm Birth, Small for \nGestational Age, Autism Spectrum Disorder, and Attention-\nDeficit/Hyperactivity Disorder in Offspring. JAMA  2017;317(15):1553-62. doi: \n10.1001/jama.2017.3413 [published Online First: 2017/04/19] \n9. Palmsten K, Hernandez-Diaz S, Huybrechts KF , et al. Use of antidepressants near \ndelivery and risk of postpartum hemorrhage: cohort study of low income \nwomen in the United States. BMJ 2013;347:f4877. doi: 10.1136/bmj.f4877 \n[published Online First: 2013/08/24] \n10. (NICE) NIfHaCE. Scenario: Planning a pregnancy on an antidepressant 2023 \n[updated November 2023. Available from: \nhttps://cks.nice.org.uk/topics/depression-antenatal-\npostnatal/management/planning-a-pregnancy-on-an-antidepressant/ \naccessed 29/05/2024 2024. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n23 \n \n11. (NICE) NIfHaCE. Senario: Unplanned pregnancy on an antidepressant 2023 \n[updated November 2023. Available from: \nhttps://cks.nice.org.uk/topics/depression-antenatal-\npostnatal/management/unplanned-pregnancy-on-an-antidepressant/ \naccessed 29/0/2024 2024. \n12. (NICE) NIfHaCE. Scenario: Pregnant women: new episode 2023 [updated \nNovember 2023. Available from: https://cks.nice.org.uk/topics/depression-\nantenatal-postnatal/management/pregnant-new-episode/ accessed \n29/05/2024 2024. \n13. Molenaar NM, Kamperman AM, Boyce P , et al. Guidelines on treatment of \nperinatal depression with antidepressants: An international review. Aust N Z J \nPsychiatry 2018;52(4):320-27. doi: 10.1177/0004867418762057 [published \nOnline First: 20180305] \n14. Service NH. Miscarriage  [Available from: \nhttps://www.nhs.uk/conditions/miscarriage/ accessed 03/10/2023 2023. \n15. Dugas C, Slane VH. Miscarriage. StatPearls. Treasure Island (FL)2022. \n16. Smith S, Martin F, Rai D, et al. Association between antidepressant use during \npregnancy and miscarriage: a systematic review and meta-analysis. BMJ \nOpen 2024;14(1):e074600. doi: 10.1136/bmjopen-2023-074600 \n17. Grech J, Chan MV, Ochin C, et al. Serotonin-Affecting Antidepressant Use in \nRelation to Platelet Reactivity. Clin Pharmacol Ther 2022;111(4):909-18. doi: \n10.1002/cpt.2517 [published Online First: 20220110] \n18. Jarde A, Morais M, Kingston D, et al. Neonatal Outcomes in Women With \nUntreated Antenatal Depression Compared With Women Without Depression: \nA Systematic Review and Meta-analysis. JAMA Psychiatry 2016;73(8):826-37. \ndoi: 10.1001/jamapsychiatry.2016.0934 [published Online First: 2016/06/09] \n19. Langham J, Gurol-Urganci I, Muller P, et al. Obstetric and neonatal outcomes in \npregnant women with and without a history of specialist mental health care: a \nnational population-based cohort study using linked routinely collected data in \nEngland. Lancet Psychiatry  2023;10(10):748-59. doi: 10.1016/S2215-\n0366(23)00200-6 [published Online First: 20230814] \n20. Shahhosseini Z, Abedian K, Azimi H. Role of Anxiety During Pregnancy in \nPreterm Delivery. Journal of Advances in Medical and Biomedical Research  \n2008;16(63):85-92. \n21. Broekman BF, Chan YH, Chong YS, et al. The influence of anxiety and \ndepressive symptoms during pregnancy on birth size. Paediatr Perinat \nEpidemiol 2014;28(2):116-26. doi: 10.1111/ppe.12096 [published Online First: \n20131124] \n22. Wong J, Motulsky A, Eguale T, et al. Treatment Indications for Antidepressants \nPrescribed in Primary Care in Quebec, Canada, 2006-2015. JAMA \n2016;315(20):2230-2. doi: 10.1001/jama.2016.3445 \n23. Chambers CD, Johnson KA, Dick LM, et al. Birth outcomes in pregnant women \ntaking fluoxetine. N Engl J Med  1996;335(14):1010-5. doi: \n10.1056/NEJM199610033351402 \n24. Kulin NA, Pastuszak A, Sage SR, et al. Pregnancy outcome following maternal \nuse of the new selective serotonin reuptake inhibitors: a prospective \ncontrolled multicenter study. JAMA 1998;279(8):609-10. doi: \n10.1001/jama.279.8.609 \n25. Einarson A, Choi J, Einarson TR, et al. Rates of spontaneous and therapeutic \nabortions following use of antidepressants in pregnancy: results from a large \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n24 \n \nprospective database. J Obstet Gynaecol Can 2009;31(5):452-6. doi: \n10.1016/s1701-2163(16)34177-9 \n26. Self-reported antidepressants use and spontaneous abortion. American Journal \nof Epidemiology; 2011. OXFORD UNIV PRESS INC JOURNALS DEPT, 2001 \nEVANS RD, CARY, NC 27513 USA. \n27. Abadie D, Hurault-Delarue C, Damase-Michel C, et al. Medication exposure and \nspontaneous abortion: a case-control study using a French medical database. \nClin Exp Obstet Gynecol 2015;42(4):431-6. \n28. Venlafaxine exposure in pregnancy, a multicenter ENTIS study. \nPharmacoepidemiology and Drug Safety; 2016. WILEY-BLACKWELL 111 \nRIVER ST, HOBOKEN 07030-5774, NJ USA. \n29. Wu P , Velez Edwards DR, Gorrindo P, et al. Association between First Trimester \nAntidepressant Use and Risk of Spontaneous Abortion. Pharmacotherapy \n2019;39(9):889-98. doi: 10.1002/phar.2308 [published Online First: 20190725] \n30. Herrett E, Gallagher AM, Bhaskaran K, et al. Data Resource Profile: Clinical \nPractice Research Datalink (CPRD). Int J Epidemiol  2015;44(3):827-36. doi: \n10.1093/ije/dyv098 [published Online First: 2015/06/08] \n31. Minassian C, Williams R, Meeraus WH, et al. Methods to generate and validate a \nPregnancy Register in the UK Clinical Practice Research Datalink primary \ncare database. Pharmacoepidemiol Drug Saf  2019;28(7):923-33. doi: \n10.1002/pds.4811 [published Online First: 2019/06/15] \n32. Campbell J, Bhaskaran K, Thomas S, et al. Investigating the optimal handling of \nuncertain pregnancy episodes in the CPRD GOLD Pregnancy Register: a \nmethodological study using UK primary care data. BMJ Open  \n2022;12(2):e055773. doi: 10.1136/bmjopen-2021-055773 [published Online \nFirst: 2022/02/24] \n33. Mathur R, Bhaskaran K, Chaturvedi N, et al. Completeness and usability of \nethnicity data in UK-based primary care and hospital databases. J Public \nHealth (Oxf) 2014;36(4):684-92. doi: 10.1093/pubmed/fdt116 [published \nOnline First: 20131208] \n34. Petersen AH, Lange T. What Is the Causal Interpretation of Sibling Comparison \nDesigns? Epidemiology 2020;31(1):75-81. doi: \n10.1097/EDE.0000000000001108 \n35. Sjolander A, Frisell T, Kuja-Halkola R, et al. Carryover Effects in Sibling \nComparison Designs. Epidemiology 2016;27(6):852-8. doi: \n10.1097/EDE.0000000000000541 \n36. Desai RJ, Franklin JM. Alternative approaches for confounding adjustment in \nobservational studies using weighting based on the propensity score: a primer \nfor practitioners. BMJ 2019;367:l5657. doi: 10.1136/bmj.l5657 [published \nOnline First: 20191023] \n37. Brookhart MA, Schneeweiss S, Rothman KJ, et al. Variable selection for \npropensity score models. Am J Epidemiol  2006;163(12):1149-56. doi: \n10.1093/aje/kwj149 [published Online First: 20060419] \n38. Hughes RA, Heron J, Sterne JAC, et al. Accounting for missing data in statistical \nanalyses: multiple imputation is not always the answer. Int J Epidemiol  \n2019;48(4):1294-304. doi: 10.1093/ije/dyz032 \n39. Martin FZ. Between a rock and a hard place: weighing up the evidence for \nantidepressant use during pregnancy. YouTube, 2022. \n40. Kjaersgaard MI, Parner ET, Vestergaard M, et al. Prenatal antidepressant \nexposure and risk of spontaneous abortion - a population-based study. PLoS \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n25 \n \nOne 2013;8(8):e72095. doi: 10.1371/journal.pone.0072095 [published Online \nFirst: 2013/09/10] \n41. Ban L, Tata LJ, West J, et al. Live and non-live pregnancy outcomes among \nwomen with depression and anxiety: a population-based study. PLoS One  \n2012;7(8):e43462. doi: 10.1371/journal.pone.0043462 [published Online First: \n20120824] \n42. Almeida ND, Basso O, Abrahamowicz M, et al. Risk of Miscarriage in Women \nReceiving Antidepressants in Early Pregnancy, Correcting for Induced \nAbortions. Epidemiology 2016;27(4):538-46. doi: \n10.1097/EDE.0000000000000484 \n43. Evans-Hoeker EA, Eisenberg E, Diamond MP , et al. Major depression, \nantidepressant use, and male and female fertility. Fertil Steril \n2018;109(5):879-87. doi: 10.1016/j.fertnstert.2018.01.029 \n44. Richardson JL, Martin F, Dunstan H, et al. Pregnancy outcomes following \nmaternal venlafaxine use: A prospective observational comparative cohort \nstudy. Reprod Toxicol  2019;84:108-13. doi: 10.1016/j.reprotox.2019.01.003 \n[published Online First: 20190111] \n45. Einarson A, Fatoye B, Sarkar M, et al. Pregnancy outcome following gestational \nexposure to venlafaxine: a multicenter prospective controlled study. Am J \nPsychiatry 2001;158(10):1728-30. doi: 10.1176/appi.ajp.158.10.1728 \n46. Ankarfeldt MZ, Petersen J, Andersen JT, et al. Duloxetine Exposure During \nPregnancy and the Risk of Spontaneous and Elective Abortion: A Danish \nNationwide Observational Study. Drugs Real World Outcomes 2021;8(3):289-\n99. doi: 10.1007/s40801-021-00252-9 [published Online First: 20210518] \n47. Nakhai-Pour HR, Broy P , Berard A. Use of antidepressants during pregnancy and \nthe risk of spontaneous abortion. CMAJ 2010;182(10):1031-7. doi: \n10.1503/cmaj.091208 [published Online First: 20100531] \n48. Ahlqvist VH, Madley-Dowd P, Ly A, et al. Bias amplification of unobserved \nconfounding in pharmacoepidemiological studies using indication-based \nsampling. Pharmacoepidemiol Drug Saf  2023;32(8):886-97. doi: \n10.1002/pds.5614 [published Online First: 20230405] \n49. Heuvelman H, Davies NM, Ben-Shlomo Y , et al. Antidepressants in pregnancy: \napplying causal epidemiological methods to understand service-use outcomes \nin women and long-term neurodevelopmental outcomes in exposed children. \nHealth Technol Assess 2023;27(15):1-83. doi: 10.3310/AQTF4490 \n50. Sinclair LI, Christmas DM, Hood SD, et al. Antidepressant-induced \njitteriness/anxiety syndrome: systematic review. Br J Psychiatry  \n2009;194(6):483-90. doi: 10.1192/bjp.bp.107.048371 [published Online First: \n2009/05/30] \n51. Gershon AA, Vishne T, Grunhaus L. Dopamine D2-like receptors and the \nantidepressant response. Biol Psychiatry  2007;61(2):145-53. doi: \n10.1016/j.biopsych.2006.05.031 [published Online First: 20060824] \n52. Kesmodel US. Information bias in epidemiological studies with a special focus on \nobstetrics and gynecology. Acta Obstet Gynecol Scand 2018;97(4):417-23. \ndoi: 10.1111/aogs.13330 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n26 \n \n9 F IGURES \nFigure 1 Sample selection and flow of pregnancy episodes through the study. \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n27 \n \nFigure 2 Findings from the primary and secondary analyses. \n \n* Adjusted Cox models included maternal age, pregnancy year, practice-level IMD quintile, history of miscarriage, smoking status around the start of pregnancy, parity at the start of pregnancy, \nuse of high dose folic acid, antipsychotics, or anti-seizure medication in the 12 months before pregnancy, number of primary care consultations in the 12 months before pregnancy, and severe \nmental illness, depression or anxiety ever before the start of pregnancy \n** Primary adjustment set minus history of miscarriage \n** Propensity score matched Cox models additionally included presence of linked data, area of residence, BMI and alcohol use around the start of pregnancy, illicit drug use in the 12 months \nbefore pregnancy, presence of diabetes, endometriosis, polycystic ovary syndrome, pre-pregnancy hypertension, eating disorders, pain disorders, migraine prophylaxis, tension-type headache \nor stress incontinence ever before the start of pregnancy, and use of potential teratogens in the 12 months before pregnancy \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n28 \n \nFigure 3 Exposure discordant pregnancy sensitivity analysis, restricting first to exposure discordant groups where the first pregnancy was \nantidepressant exposed and subsequent pregnancies in the group were not, then to groups where subsequent pregnancies were exposed to \nantidepressants but first pregnancies in the group were not. \n \n** Primary adjustment set minus history of miscarriage \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n29 \n \n10 TABLES \nTable 1 Characteristics of pregnant women eligible for inclusion. \n All (n=1,021,384) Exposed (n=73,493) Unexposed \n(n=947,891) \nAge (years)    \n<18 38,690 (3.8) 998 (1.4) 37,692 (4.0) \n18-24 232,914 (22.8) 17,605 (24.0) 215,309 (22.7) \n25-29 263,042 (25.8) 19,071 (25.9) 243,971 (25.7) \n30-34 283,073 (27.7) 19,133 (26.0) 263,940 (27.8) \n>=35 203,665 (19.9) 16,686 (22.7) 186,979 (19.7) \nYear of pregnancy    \n1996-2000 119,012 (11.7) 4,816 (6.6) 114,196 (12.0) \n2001-2005 242,286 (23.7) 14,689 (20.0) 227,597 (24.0) \n2006-2010 309,392 (30.3) 20,330 (27.7) 289,062 (30.5) \n2011-2015 256,246 (25.1) 22,564 (30.7) 233,682 (24.7) \n2016-2018 94,448 (9.2) 11,094 (15.1) 83,354 (8.8) \nPractice-level IMD \n(quintiles)    \n1 (least deprived) 161,493 (15.8) 9,546 (13.0) 151,947 (16.0) \n2 165,591 (16.2) 11,017 (15.0) 154,574 (16.3) \n3 187,170 (18.3) 13,319 (18.1) 173,851 (18.3) \n4 229,209 (22.4) 17,073 (23.2) 212,136 (22.4) \n5 (most deprived) 277,921 (27.2) 22,538 (30.7) 255,383 (26.9) \nEthnicity    \nWhite 631,614 (61.8) 47,635 (64.8) 583,979 (61.6) \nSouth Asian 31,494 (3.1) 864 (1.2) 30,630 (3.2) \nBlack 16,706 (1.6) 470 (0.6) 16,236 (1.7) \nOther 11,127 (1.1) 324 (0.4) 10,803 (1.1) \nMixed 6,589 (0.6) 387 (0.5) 6,202 (0.7) \nMissing 323,854 (31.7) 23,813 (32.4) 300,041 (31.7) \nBody mass index    \nUnderweight (<18.5 kg/m2) 33,616 (3.3) 2,697 (3.7) 30,919 (3.3) \nHealthy weight (18.5-24.9 \nkg/m2) 465,110 (45.5) 28,828 (39.2) 436,282 (46.0) \nOverweight (25.0-29.9 \nkg/m2) 238,249 (23.3) 17,393 (23.7) 220,856 (23.3) \nObese (>=30.0 kg/m2) 179,700 (17.6) 18,532 (25.2) 161,168 (17.0) \nMissing 104,709 (10.3) 6,043 (8.2) 98,666 (10.4) \nPrevious miscarriage    \nYes 160,994 (15.8) 14,406 (19.6) 146,588 (15.5) \nParity    \n0 485,775 (47.6) 27,208 (37.0) 458,567 (48.4) \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint \n\n  \n \n30 \n \n1 346,248 (33.9) 24,681 (33.6) 321,567 (33.9) \n2 131,356 (12.9) 13,808 (18.8) 117,548 (12.4) \n>=3 58,005 (5.7) 7,796 (10.6) 50,209 (5.3) \nMental health history1     \nDepression 252,356 (24.7) 56,267 (76.6) 196,089 (20.7) \nAnxiety 154,394 (15.1) 33,818 (46.0) 120,576 (12.7) \nSevere mental illness2 5,079 (0.5) 1,772 (2.4) 3,307 (0.3) \nPrimary care visits in the \n12 months before \npregnancy \n   \n0 119,052 (11.7) 4,974 (6.8) 114,078 (12.0) \n1-3 262,141 (25.7) 4,665 (6.3) 257,476 (27.2) \n4-10 419,751 (41.1) 24,392 (33.2) 395,359 (41.7) \n>10 220,440 (21.6) 39,462 (53.7) 180,978 (19.1) \nSmoking status around the \nstart of pregnancy    \nNon-smoker 414,763 (40.6) 19,910 (27.1) 394,853 (41.7) \nCurrent smoker 304,897 (29.9) 32,110 (43.7) 272,787 (28.8) \nEx-smoker 248,265 (24.3) 19,397 (26.4) 228,868 (24.1) \nMissing 53,459 (5.2) 2,076 (2.8) 51,383 (5.4) \nOther prescriptions 12 \nmonths before pregnancy    \nAntipsychotics 865 (0.1) 497 (0.7) 368 (0.0) \nMood stabilisers 9,912 (1.0) 3,000 (4.1) 6,912 (0.7) \nFolic acid (5mg) 58,830 (5.8) 7,169 (9.8) 51,661 (5.5) \n1 Identified using Read and ICD-10 codes from primary care data and HES data (for whom it was available), \nrespectively \n2 Bipolar disorder, psychosis, or schizophrenia \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 21, 2024. ; https://doi.org/10.1101/2024.10.19.24315779doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}