Study population
This retrospective population-based cohort study obtained data through linkage of 5 Swedish national registers using unique personal identity numbers that are assigned to all Swedish residents. Linked registers included the Swedish Medical Birth Register, the Patient Register, the Prescribed Drug Register, Education Register and Total Population Register. The Medical Birth Register prospectively collects data of over 98% of all births in Sweden and includes demographic data, reproductive history, complications during pregnancy and delivery and neonatal outcomes. The Patient Register includes information on all inpatient care and outpatient clinical services, including psychiatry. All procedures and diagnoses are documented using International Classification of Disease (ICD) diagnostic codes. The Prescribing Drug Register contains data on all prescriptions dispensed in Sweden and includes information on dispensed item, date of dispensing, substance, formulation, package size, dispensed amount and dosage. Linkage between the registers is possible due to the individual identification number given to all citizens in Sweden.
All women giving birth to a liveborn or stillborn child at 22 gestational weeks or later in Sweden from January 2007 to December 2014 were included. Multiple pregnancies were excluded.
Exposure
Maternal lithium treatment prior and during pregnancy was obtained from the Prescribing Drug Register (ATC code N05AN). Lithium use during pregnancy was defined as a prescription dispensed during pregnancy or the 3 months prior to conception. To identify lithium exposure during this period, date of conception was calculated as date of birth − gestational age in days, with gestational age at birth obtained from the Medical Birth Register and estimated via ultrasound dating during the second trimester of pregnancy.
Pregnancy and neonatal outcomes
Primary and secondary outcomes of interest were prespecified and selected based on potential impacts of medication exposure during pregnancy and findings from previous studies. The primary outcomes were preeclampsia (ICD-10 codes O14, O15), spontaneous preterm birth before 37 weeks’ gestation and birth of a small for gestational age or large for gestational age infant. Information on onset of birth is routinely recorded in a standardized manner by the delivery ward midwife and is categorized as spontaneous, induced vaginal or caesarean section before onset of labour. Spontaneous onset of birth was defined as a registered spontaneous start or if a diagnosis of preterm premature rupture of the membranes (ICD code O42) was present. Small for gestational age and large for gestational age were defined as a birthweight of more than two standard deviations below or above the mean weight for gestational age according to the Swedish national reference curve [7]. Secondary outcomes were macrosomia (birthweight > 4000 g), hypoglycaemia (ICD code P70), Apgar < 7 at 5 min, perinatal death, any congenital malformation and cardiac malformations: ICD-10 code Q20 congenital malformations of cardiac chambers and connections, Q21 congenital malformations of cardiac septa, Q22 congenital malformations of pulmonary and tricuspid valves, Q23 congenital malformations of aortic and mitral valves, Q24 other congenital malformations of heart, Q25 congenital malformations of great arteries and Q26 congenital malformations of great veins.
Covariates
Information on maternal age at delivery (</≥35 years), height, weight, smoking at first antenatal visit (yes/no), parity (nulliparous or multiparous), use of assisted reproduction (yes/no) and gestational disorders was obtained from the medical birth register. Body mass index (BMI) was calculated as weight (kg)/height (m2) and categorized as (</≥ 30 kg/m2. Data on country of birth was obtained from the Total Population Register and categorized as Sweden, other European countries or the rest of the world. Educational attainment was retrieved from the Higher Education Registry and categorized as < 12 years, completion of high school or completion of university. Pre-existing maternal conditions were obtained by ICD codes from the Patient Register: bipolar disorder (F31), schizophrenia (F20–F29), psychosis (F29), pre-gestational hypertension (I10, O10) and diabetes (O240, O241, O244, O249, E10–14).
Subgroup analyses
To account for the impact of underlying psychiatric illness on perinatal outcomes, we performed two subgroup analyses. Both were planned a priori. First, we compared the same outcomes among women with a diagnosis of bipolar disorder, schizophrenia or psychosis who were either prescribed lithium or not.
Second, we compared women who had used lithium during pregnancy to those who had used lithium prior to pregnancy but not in the 3 months prior to conception or during pregnancy.
Statistical analysis
A prespecified statistical analysis plan was generated prior to analyses and agreed upon by the authors. Characteristics of the population were described according to lithium use during pregnancy. Lithium users and non-users were compared via bivariate analysis using Pearson’s Chi-squared test for categorical data and Student’s t test for continuous variables.
The average treatment effect for lithium-treated women (average treatment effect on the treated) on adverse pregnancy and neonatal outcomes was estimated using Stata’s ‘teffects’ command and presented as relative risk with 95% confidence intervals; within-mother clustering was accounted for in standard errors. Regression models were inverse probability weighted to achieve covariate balance between exposure groups. A propensity score for exposure was calculated for each individual using a logistic regression that included variables considered likely confounders based upon directed acyclic graphs and selected a priori for inclusion in the propensity score models. Included covariates were: maternal age (</> 35 years), body mass index (</> 30 kg/m2), smoking status, country of birth, education, parity, maternal psychiatric disorders (schizophrenia and psychosis, with an interaction term between these), other medical conditions (diabetes, hypertension) and antipsychotics (ATC code N05A), central stimulants (ATC-code N06BA), or lamotrigine (ATC code N03AX09) dispensed during or 3 months prior to pregnancy. Second, with the aim of achieving balance between exposed and unexposed groups, the estimated propensity scores were used to weight each subject, with a score of 1/propensity score assigned to the exposed individuals and 1/(1 − propensity score) to the unexposed. Propensity scores were also generated for subgroup analyses, with the following covariates included: maternal age (</> 35 years), body mass index (</> 30 kg/m2), smoking status, country of birth, education, parity, medical conditions (diabetes and hypertension) and antipsychotics, neuroleptics or lamotrigine dispensed during or 3 months prior to pregnancy. Propensity scores were checked for extreme values and overlap between exposure groups. The balance of individual covariates before and after inverse probability weighting was also assessed. Weighted standardized differences of less than 0.1 were indicative of covariate balance (raw and weighted standardized differences for included covariates and each outcome are shown in Additional File 1). Modelling only proceeded when there was sufficient overlap of propensity scores and covariate balance following weighting was achieved. Multiple imputation of exposure, outcome and covariate missing data was not performed and those with missing data were not included in the adjusted analysis.
Statistical analysis used StataMP® software (StataCorp. College Station, TX, USA).