Risk of miscarriage in women with psychiatric disorders.

We identified all registered pregnancies in Norway with an estimated date of conception between January 1 st 2010 and December 31 st 2016, using three national health registries: the Medical Birth Registry of Norway, the Norwegian Patient Registry and the general practice database ( 9 ). The patient registry comprises visits to medical specialists and hospitals, with diagnoses coded according to International Classification of Diseases version 10 (ICD-10). Visits to general practitioners are coded according to the International Classification of Primary Care version 2 (ICPC-2). The birth registry includes information on pregnancies ending after 12 gestational weeks (live births, stillbirths, late miscarriages and late induced abortions). The patient registry and the general practice database provides information on all women in contact with health care services during pregnancy, including contacts during first trimester. We linked information from the three health registries (birth registry, specialist care and the general practice database) using unique personal identification numbers. We identified live births and fetal deaths after 12 gestational weeks from the birth registry. A fetal death at 20 gestational weeks or later or with a birthweight of 500 grams or more was considered a stillbirth, while a fetal death before 20 gestational weeks with a birthweight less than 500 grams was considered a miscarriage. This distinction of stillbirth is in accordance with the American National Stillbirth Society and the National Institutes of Health. We included the following ICD-10 codes to capture miscarriages in the patient registry: hydatidiform mole (O01); blighted ovum and non-hydatidiform mole (O02.0); missed abortion (O02.1); other specified abnormal products of conception (O02.8); abnormal product of conception, unspecified (O02.9); spontaneous abortion (O03); and threatened abortion (O02.0). Induced abortion was defined by the following ICD-10 codes in the patient registry: medical abortion (O04), other abortion (O05) and unspecified abortion (O06). We have expanded our definition of miscarriage used in a previous study to include events reported in the general practitioner database ( 2 ). Bleeding in pregnancy (ICPC-2 code W03) and spontaneous abortion (W82) were defined as first trimester miscarriages if they were not followed by a registration in the birth registry or the patient registry. Pregnancies in the patient registry and general practitioner database are not registered with unique pregnancy IDs, which could lead to inadvertent counting of the same pregnancy multiple times. We minimized this error by requiring a minimum of 6 weeks (42 days) between two successive records in the patient registry, and a minimum of 3 months (90 days) between two successive records in the general practitioner database. These cut-offs were chosen by inspecting the distribution of times between registrations in the two registers. We used a longer interval in the general practitioner database because women have more follow-up visits with their general practitioners after a miscarriage. We also required that a record of a miscarriage or induced abortion in the patient registry or general practitioner database be at least 6 weeks (patient registry) or 3 months (general practitioner database) after a registered delivery to the woman in the birth registry. Finally, we excluded any miscarriage or induced abortion that occurred within the gestational period of a registered pregnancy to the same woman in the birth registry. In the case of multiple fetuses, the outcome was regarded as a live birth if all deliveries resulted in live births, as a miscarriage if there was at least one miscarriage but no stillbirth, and as a stillbirth if at least one of the deliveries resulted in a stillbirth but none in a miscarriage. A multiple birth could result in both a miscarriage and a stillbirth if there was a discrepancy in the birthweight between the fetuses. We identified diagnostic codes for 12 psychiatric disorder groups using information available in the patient registry and general practice database. The 12 diagnostic groups were schizophrenia spectrum disorders, bipolar spectrum disorders, depressive disorders, anxiety disorders, somatoform disorders, eating disorders, intellectual disability, autism spectrum disorders, attention-deficit/hyperactivity disorder (ADHD), conduct disorder, personality disorders, and unspecified psychiatric disorders. The diagnostic codes to classify the various conditions are shown in Supplementary Table 1 . We required a minimum of two registrations of the administrative code(s) used to define the particular psychiatric diagnosis to classify women as diagnosed with the psychiatric disorder to avoid coding errors. To ensure that woman had been diagnosed with the condition prior to pregnancy, we required that at least one registration should be before the estimated date of the last menstrual period, defined as the date of delivery minus the gestational age in days. For miscarriages and induced abortions identified in the patient registry and general practice database (in which gestational week was not available), we assumed a gestational age of 12 weeks to ensure that the condition would have been diagnosed prior to pregnancy, assuming that the pregnancy would have been registered in the birth registry if the gestational age was more than 12 weeks. We had information on maternal age at the time of conception, gravidity, year of conception, substance use disorders, and diagnoses of somatic conditions. Diagnoses of substance use disorders and somatic diseases were obtained from the patient and general practitioner databases. We included autoimmune diseases (systemic lupus erythematosus, type 1 diabetes, celiac disease, multiple sclerosis, rheumatoid arthritis/ankylosing spondylitis, ulcerative colitis, psoriasis, Addison disease, Crohn´s disease, haemolytic anemia), endocrinological diseases (Cushing syndrome, hypothyroidism, hyperthyroidism, hyperaldosteronism, hyperparathyroidism), cardiometabolic diseases (atherosclerosis, hypertensive disorders, type 2 diabetes), allergic diseases (asthma, allergic rhinitis, atopic dermatitis), neurological diseases (epilepsy, migraine) and reproductive diseases (polycystic ovary syndrome (PCOS), and endometriosis). The various ICD-10 codes (specialist health care services) and ICPC-2 codes (primary health care services) used to capture these disorders are listed in Supplementary Table 2 . We used logistic regression to calculate odds ratios (OR) of miscarriage according to psychiatric disorders. We attempted to estimate relative risks using log-binomial regression; however, the model failed to converge for several of the psychiatric diagnoses. We therefore present results using logistic regression for all disorders for consistency. We also estimated risk differences (RD) as an absolute measure of the difference in the risk of miscarriage, using linear regression. Because some women experienced more than one pregnancy during the time period, we used cluster variance estimation to calculate 95% confidence intervals. We have previously shown that miscarriage risk increases in a non-linear fashion with maternal age ( 2 ), so we adjusted for woman’s age at the start of pregnancy by including age as both a linear and squared term. We adjusted for the competing risk of induced abortion by adding 20% of induced abortions to the comparison group in the analysis, together with all live and stillbirths. Our basis for this adjustment is described in detail in the supplementary methods (SA1) in the appendix. We conducted stratified analyses according to whether the woman had previously been pregnant, and whether the miscarriage was registered in the specialist health-care services or only in primary health-care services. We also conducted sensitivity analyses further adjusting for year of conception, restricting to psychiatric disorders registered in the specialist health-care services, and excluding hydatidiform mole from the definition of miscarriage. To examine the independent relationship of the various diagnoses with miscarriage risk, we included all 12 psychiatric diagnoses simultaneously in the model. As an indicator of psychiatric multimorbidity, we also examined the risk of miscarriage according to the total number of different psychiatric diagnoses a woman had been diagnosed with prior to pregnancy, categorized as none (reference), one, two, and three or more diagnoses. This did not include additional registrations of administrative codes used to define the same diagnosis. All analyses were conducted using STATA version 15 (Statacorp, Texas). The study was approved by the Committee for Medical and Health Research Ethics of South/East Norway. The use of the national health registries for health-related research does not require consent according to Norwegian legislation.

We identified a total of 593 009 pregnancies with last menstrual period between January 1 st 2010 and December 31 st 2016 ( Figure 1 ). Of these pregnancies, 409 952 (69.1%) ended in a live birth, 1 740 (0.3%) in stillbirth, 95 641 (16.1%) in induced abortion, and 85 676 (14.4%) in miscarriage. Miscarriage risk estimated as a proportion of miscarriages plus live and stillbirths was 17.2%, which was reduced to 16.6% after adjustment for the competing risk of induced abortions. Twenty-two percent of women had a diagnosis of one or more psychiatric conditions, with depressive disorders (12.1%) and anxiety disorders (8.3%) being by far the most common ( Table 1 ). The proportion of pregnant women diagnosed with other psychiatric disorders ranged from 0.02% for autism spectrum disorders to 1.5% for ADHD ( Table 1 ). Women with psychiatric disorders had an increased risk of miscarriage ( Figure 2 ). The absolute increase in miscarriage risk ranged from 2% to 5% on a baseline rate of nearly 17% ( Supplementary Figure 1 ). Odds ratios for specific diagnostic groups were 1.35 for bipolar disorders (95% confidence interval 1.26-1.44); 1.32 for personality disorders (1.12-1.55); 1.27 for ADHD (1.21-1.33); 1.25 for anxiety disorders (1.23-1.28); 1.25 for depressive disorders (1.23-1.27); 1.18 for somatoform disorders (1.07-1.31); and 1.14 for eating disorders (1.08-1.22) ( Figure 2 ). Women’s age was an important confounder, as shown by the stronger associations in unadjusted analysis ( Supplementary Figure 2 ). Further adjustment for year of conception, to account for the amount of follow-up information available, did not changes the associations ( Supplementary Figure 3 ). Five percent of all pregnancies were to women with more than one psychiatric diagnosis. The most common combination was anxiety and depression diagnoses (2% of all pregnancies). When we adjusted the risk of miscarriage for each diagnosis for the co-occurrence of other psychiatric diagnoses, the risks of miscarriage were slightly attenuated ( Figure 3 ). The only exception was schizophrenia spectrum disorders, where we observed a strengthening of the association, with an adjusted OR of 1.22 (1.03-1.44). We also looked at the risk of miscarriage according to the number of psychiatric diagnoses as an indicator of psychiatric multimorbidity. The adjusted OR of miscarriage with one was 1.27 (1.25-1.30); with two, 1.45 (1.40-1.51); and with three or more diagnoses, 1.51 (1.31-1.73). A past miscarriage might contribute to psychiatric conditions, so we conducted a sensitivity analysis restricting to first pregnancies ( Supplementary Figure 4 ). The main associations persisted. Excluding hydatidiform mole from the definition of miscarriage did not change results. About 25% of miscarriages were identified in the general practitioner database with no subsequent record in the patient registry or the birth registry. When we analyzed the general practitioner and specialist health-care sources of data separately, results were similar overall, although risks were more consistently elevated among women whose miscarriage was referred to specialist health-care services ( Supplementary Figure 5 ). Associations were slightly stronger when restricting the evaluation to psychiatric disorders diagnosed in specialist care services ( Supplementary Figure 6 ). We also conducted an evaluation of the role of somatic chronic conditions. The proportion of women with the different chronic somatic conditions is shown in Supplementary Table 3 . Further adjustment for substance use disorders and all the somatic conditions did not explain the increased risk of miscarriage seen among women with psychiatric disorders in the main analysis ( Supplementary Figure 7 ).

In this large registry-based study, women with a broad range of psychiatric disorders had an increased risk of miscarriage. Risks were most prominently elevated with bipolar disorders, personality disorders, ADHD, anxiety disorders, depressive disorders, and somatoform disorders. We also observed a weak evidence of an increased risk of miscarriage among women with schizophrenia, which was strengthened after mutual adjustment for co-occurring psychiatric disorders. The risk of miscarriage increased with multiple psychiatric diagnoses. Our results confirm specific associations with risk of miscarriage reported in previous studies of women with eating disorders ( 3 ), bipolar disorder ( 6 , 7 , 10 ), and anxiety or depression ( 4 , 5 ). A systematic review found evidence for increased miscarriage risk specifically among women with anorexia nervosa ( 3 ), and a few small-scale studies support an increased risk with anxiety or depression ( 4 , 5 ). Some previous studies obtained information on pre-existing psychiatric disorders retrospectively, after the end of pregnancy, which can leave open the possibility of recall bias following a miscarriage. Women with schizophrenia have an increased risk of pregnancy complications ( 11 , 12 ). The modest increased risk of miscarriage among women with schizophrenia spectrum disorders identified in this study is novel and should be further explored. It is well documented that women who experience miscarriage (especially recurrent miscarriage) have an increased risk of psychiatric disorders ( 13 , 14 ). We therefore conducted a sensitivity analysis restricted to women who were pregnant for the first time to address the possibility of reverse causation. The results were overall very similar to our main findings including all pregnancies. While it is plausible that miscarriage can increase the risk of psychiatric illness, this sensitivity analysis shows that such an effect does not contribute to the associations we observe. A direct effect of psychiatric illness on miscarriage risk is plausible. One potential biological mechanisms is the link between peptides and proteins synthesized in the brain and placental development ( 15 ). It has been hypothesized that women with psychiatric conditions such as depression, bipolar disorder, and schizophrenia experience an increased risk of adverse obstetric outcomes through changes in these neurotrophin factors in the brain ( 16 , 17 ). Another possibility is an effect of stress hormones, which can influence both placental development and the intrauterine environment, perhaps increasing the risk of miscarriage ( 5 ). These biological mechanisms may be shared across diverse psychiatric disorders. More specific mechanisms may be at work in conditions such as anorexia nervosa, in which nutritional deficiency may impair the woman’s ability to sustain a pregnancy ( 18 ). Prenatal depression has also been associated with placental barrier dysfunction ( 5 ). There can also be other direct or indirect explanations for the relationship between psychiatric disorders and risk of miscarriage. This can include both socio-economic pathways and lifestyle factors. The risk of psychiatric disorders is strongly related to socio-economic factors ( 19 , 20 ), but the extent to which such factors might influence miscarriage risk is not well established. We used the E-value ( 21 ) to estimate that an unmeasured confounder would have to be associated with a 1.7 fold increased risk of both miscarriage and bipolar disorders (the strongest association in our analysis) to completely explain the relationship. Lifestyle characteristics that are strongly linked to psychiatric disorders (for example smoking and higher body-mass index) ( 22 , 23 ) may also contribute to explain an increased risk of miscarriage. Differences in lifestyle factors between women with and without psychiatric disorders could therefore be one explanatory pathway for an increased risk of miscarriage. Our prospective data on diagnosis of both miscarriage and psychiatric conditions provide a strong basis for assessing associations free of self-reporting bias. Coverage across the whole population of Norway provides power and protects against selection bias that might arise within self-selected study populations. Perhaps the most important limitation of our analysis is the lack of data on medications used for psychiatric conditions. We did not have this information available due to the strict regulations for linkage with the prescription registry. An increased risk of miscarriage has been reported with use of lithium for bipolar disorders ( 6 , 7 ), various antidepressants and antianxiety medications ( 24 , 25 ), and medications for ADHD ( 8 , 26 ). However, if psychiatric illness has a causal effect on risk of miscarriage, it is also possible that treatment reduces the risk of miscarriage. A better understanding of underlying mechanisms would allow clinicians to properly weigh the potential risks of treatment against the risks of untreated disease. We are likely to have underestimated the proportion of women with a psychiatric condition prior to pregnancy, as our databases capture diagnoses only among women who seek care from the health-care system ( 27 , 28 ). If this is the case, out estimates of miscarriage risk may be biased towards the null. Furthermore, we did not have information available on the age of diagnosis of the psychiatric disorder, and could therefore not account for the potential chronicity of the condition. In conclusion, a wide range of psychiatric diagnoses were associated with increased risk of miscarriage. While these associations are biologically plausible, we cannot assess the extent to which treatment for the illness might influence the observed risk. Nevertheless, the findings highlight the importance of identifying and adequately treating psychiatric disorders among women of reproductive age.