We obtained data for this cohort study from three Swedish nationwide registers maintained by the National Board of Health and Welfare: the Swedish prescribed drug register, the medical birth register, and the national patient register. Data from the drug and birth registers included the period from 2005 to 2009, whereas data from the patient register included information from 1997 to 2009. The unique personal identification number assigned to all Swedish permanent residents allows for linkage of information across the three registers.15
The Swedish prescribed drug register contains information on all prescriptions filled in Sweden, including the Anatomical Therapeutic Chemical code of dispensed substances, amount, formulation, and date of prescribing and dispensing.16
It does not, however, cover drugs administered in hospitals. The medical birth register contains data on almost all births in Sweden.17
The information is obtained by midwives and attending doctors in connection with visits and admissions to hospital from the antenatal visit to the neonatal period. The data include maternal personal variables, tobacco use, height and weight in early pregnancy, complications during pregnancy, delivery, and data from the neonatal period (for example, Apgar score and neonatal hypoglycaemia). Furthermore, anthropometrics of the infants (birth weight, birth length, and head circumference) are recorded in the register. The initial visit for antenatal care occurs during the first trimester in more than 95% of the pregnancies.18
Gestational age is primarily based on prenatal ultrasound estimation of the last menstrual period, if present; otherwise it is estimated on the recorded date of the first day of the last menstrual period. Since 1990 ultrasonography for determination of gestational length has been offered to all pregnant women in Sweden (accepted by 95%).19
The national patient register comprises information on diagnoses from all specialised inpatient and outpatient care in Sweden (excluding primary care facilities with general practitioners). Patients with bipolar disorder are treated within specialised psychiatric care facilities in Sweden, and all information from these facilities is included in the national patient register. Since 1997 the diagnoses have been coded according to the International Classification of Diseases
, 10th revision (ICD-10). From the medical birth register we identified women who had an estimated last menstrual period anytime after 1 July 2005 and gave birth to a singleton anytime before the end of 31 December 2009. We excluded those with missing data on smoking, height, or cohabitation, as well as those giving birth to a stillborn infant.
Only women with at least two recorded diagnoses for bipolar disorder (ICD-10 code F30-31) in the national patient register were classified as having bipolar disorder. We defined use of a mood stabiliser as filling a prescription that supplied a quantity of the drug to cover intake during pregnancy according to the prescribed dosage. We assumed that the patient started taking the drug directly after the prescription was filled and followed the dosage directions. Mood stabilisers included lithium, antipsychotic drugs (ATC code N05A, except for prochlorperazine, levomepromazine, and melperone), and any of the anticonvulsants carbamazepine, lamotrigine, and valproate (ATC codes N03AF01, N03AX09, and N03AG01). Women who used mood stabilisers but had no record of a bipolar disorder were excluded from the analyses. We also obtained information about the number of women admitted to a psychiatric hospital during pregnancy and with no prescription filled for antipsychotics. Women with bipolar disorder were classified by use of mood stabilisers as either treated or untreated. The comparison (reference) group consisted of women without a bipolar disorder and their infants.
For congenital malformations, we considered only major ones. Thus we did not include accessory auricle, prominent ear, single umbilical artery, tongue tie, undescended testicle, unstable hip or dislocation/subluxation of the hip joint, polydactyly, syndactyly, or non-neoplastic nevus. We defined instrumental delivery as caesarean delivery or use of forceps or ventouse; gestational diabetes was defined as a recorded diagnosis (ICD code O24) during pregnancy; preterm birth was defined as being born before 37 weeks of gestation; and very preterm was defined as being born before 32 weeks of gestation. Being small for gestational age or large for gestational age on weight, length, and head circumference were defined as a measurement in the ≤2.3 centile or ≥97.7 centile (which corresponds to 2 standard deviations), respectively, of the total population by infant’s sex.20
A symmetrical small for gestational age infant was defined as being small for gestational age on both weight and length. Apgar scoring is done by midwives and is an assessment of the newborn infant’s heart rate, respiration, muscle tone, and reflex irritability on a scale from 0-10 points. A low Apgar score at five minutes postpartum was defined as a score of <7. Neonatal morbidity included jaundice (ICD-10 codes P57.8, P57.9, P58, and P59) and hypoglycaemia (ICD-10 code P70, except for P70.2).
Potential confounders were maternal country of origin, smoking, height, and cohabitation status at the first antenatal visit, together with maternal age when giving birth, birth order of the infant, and a diagnosis of maternal alcohol or other substance misuse disorder.
We compared women with bipolar disorder, either treated or not treated with mood stabilisers during pregnancy, as separate categories with the rest of the study population. All outcomes were analysed using univariate logistic regression models, except for congenital malformations in which we carried out Fisher’s exact test because of the low number of events. For the adverse pregnancy and birth outcomes as well as for the anthropometric outcomes, we carried out multivariable analyses after adjusting for the potential confounders of maternal age, height, country of origin, cohabitation, smoking, infants’ birth order, and a diagnosis of maternal alcohol or substance misuse disorder. When assessing the Apgar score, we included mode of delivery in the model. To adjust for the effect of more than one child of the same mother, we calculated estimates in all logistic regression models by the generalised estimating equation approach using the GENMOD procedure in SAS software, version 9.2. An overall analysis of the variation in outcome between the three usage groups was done by computing a P value for each analysis. Estimated risks are presented as odds ratios with 95% confidence intervals.