Polygenic Risk and Progression to Bipolar or Psychotic Disorders Among Individuals Diagnosed With Unipolar Depression in Early Life

Data were obtained from the iPSYCH Danish case-cohort study (iPSYCH2012). This sample includes all individuals born in Denmark between 1981 and 2005 who received a diagnosis of affective disorder, schizophrenia, autism, attention deficit hyperactivity disorder (ADHD), or anorexia nervosa in a publicly funded psychiatric hospital through December 31, 2012, as well as a random sample of 30,000 individuals drawn from the Danish population born between 1981 and 2005 who survived to their first birthday and had known mothers (24). Cases were identified from the Danish Psychiatric Central Research Register (DPCRR) (25), which includes all psychiatric diagnoses given in inpatient settings at Danish psychiatric hospitals from 1969 to 1994, as well as diagnoses given in inpatient, outpatient, and emergency department settings from 1995 onward. Diagnoses in the DPCRR are assigned at discharge by a treating psychiatrist on the basis of ICD-8 criteria from 1969 to 1993 and ICD-10 criteria from 1994 onward (26). Cases in the iPSYCH2012 sample were selected from a version of the DPCRR that was complete through 2012 (24); however, information on diagnoses through 2016 is now available through a register update.

This study was approved by the Danish Data Protection Agency and the Danish Health Data Authority. We did not obtain informed consent from participants, because it is not required for register-based studies, in accordance with Danish law.

Genotyping of members of the iPSYCH2012 case-cohort sample was done from blood spots collected at birth as part of routine clinical practice and stored in the Danish Newborn Screening Biobank (27). Blood spots were located for 93.3% of the original sample (N=80,422), and 90% of the original sample passed quality-control measures (N=77,639) (24). Genetic data for members of the iPSYCH2012 sample can be linked with information stored in Danish national registers, including the DPCRR, using the unique personal identification number assigned to all individuals born or residing legally in Denmark (28).

A flowchart of the sample selection process is presented in Figure 1. We defined a cohort of all individuals from the iPSYCH2012 sample who had received a primary, secondary, or underlying-cause diagnosis of depression (ICD-10 codes F32–F33) in a Danish psychiatric hospital at age ≥10; who were successfully genotyped and passed quality-control measures; who were of European ancestry as determined by principal component analysis; and who had no prior diagnoses of bipolar disorder (codes F30–F31) or psychotic disorders (codes F20–F29, excluding code F24) in the DPCRR. Finally, we removed at random one member from each pair of related individuals (pi-hat score >0.20 or second-degree relatedness or closer). The final study sample comprised 16,949 individuals.

To ensure that our study sample was representative of individuals treated for depression in hospital-based settings in Denmark, only individuals who were selected for inclusion in the iPSYCH2012 cohort on the basis of a depression diagnosis were included. Individuals selected for inclusion in iPSYCH2012 on the basis of a different psychiatric diagnosis (e.g., ADHD) who received a depression diagnosis after 2012 were not included. However, members of the iPSYCH2012 subcohort (i.e., the 30,000 individuals randomly selected from the Danish population regardless of case status) were included in our study sample regardless of when they received their depression diagnosis.

The main outcomes were a diagnosis of bipolar disorder (ICD-10 codes F30 and F31) or psychotic disorders (codes F20–F29, excluding code F24). As secondary outcomes, we examined conversion to schizophrenia (code F20) and psychotic depression (codes F32.3 and F33.3), as well as two composite categories: any affective psychotic disorder, which included both psychotic depression (codes F32.3 and F33.3) and psychotic bipolar disorder (codes F30.2, F31.2, and F31.5), and any disorder with psychotic features, a composite of all outcomes with psychotic features. For the analyses of progression to affective psychotic disorders, we removed individuals with psychotic depression as their first depression diagnosis (N=458). A list of specific ICD-10 codes included in each outcome category is presented in Table S1 in the online supplement.

The main exposures were PRS for major depression, PRS for bipolar disorder, and PRS for schizophrenia. PRSs were created using the LDpred method (infinitesimal model) (29) based on the most recent summary statistics from the Psychiatric Genomics Consortium and 23andMe (not including the iPSYCH2012 sample) (19, 30, 31). PRSs were standardized according to their means and standard deviations in the study population. Additionally, we examined the effects of parental history of bipolar disorder (ICD-8 codes 296.19, 296.39, and 298.19; ICD-10 codes F30 and F31) and psychotic disorders (ICD-8 codes 295.x, 297.x, 296.89, 298.29, 298.39, 298.89, 298.99, 299.04, 299.05, 299.09, and 301.83; ICD-10 codes F20–F29, excluding F24) and, for comparison, unipolar depression (ICD-8 codes 296.09, 296.29, 298.09, and 300.49; ICD-10 codes F32 and F33). Parental history was assessed by linking maternal and paternal personal identification numbers to the DPCRR. A proband was considered to have a parental history of one of these disorders if either the mother or the father received the diagnosis in a psychiatric hospital on or before the date of the proband’s first depression diagnosis. Parental history was defined as a mutually exclusive, hierarchical variable such that parents were categorized according to the most severe diagnosis they received before the date of the proband’s first depression diagnosis (that is, psychotic disorders took precedence over bipolar disorder, which took precedence over unipolar depression).

Hazard ratios for the association between the PRSs for bipolar disorder, schizophrenia, and major depression and progression to bipolar disorder or psychotic disorders were estimated using Cox proportional hazards models, with days since the first depression diagnosis as the time metric. Models were adjusted for sex and the first five principal components and stratified by genotype wave (24) to control for batch effects. Because genotype wave correlates strongly with birth year, this stratification also partially controls for potential cohort effects. In addition, because previous research has shown substantial overlap in the underlying genetic liabilities for bipolar disorder, schizophrenia, and depression (17, 32), we obtained jointly estimated hazard ratios by including all three scores in the same regression model. Finally, we tested for interaction between PRS variables by fitting models with cross-product interaction terms. Absolute risks were estimated using cumulative hazards obtained from Cox regression models. Statistical significance was assessed at a Bonferroni-corrected alpha of 0.017 to account for the fact that effects were tested for three PRS scores. Analyses were conducted in SAS, version 9.4.

We conducted four separate sensitivity analyses to evaluate the extent to which our sampling choices affected the effects of PRS on progression to bipolar disorder or psychotic disorders. First, we examined progression among only those patients with inpatient or outpatient hospital contacts. Second, we examined progression among only those patients with depression as their main diagnosis. Third, we examined progression in the full sample of depressed patients including close relatives, and, finally, we examined progression in the full sample including depressed patients of non-European genetic ancestry. We also ran models for each primary outcome with death as a competing event to ensure that our results were not biased by higher mortality rates among patients who would eventually progress to a diagnosis of bipolar disorder or psychotic disorders. Finally, some individuals (N=131) received diagnoses of both bipolar disorder and a psychotic disorder during follow-up assessment. These patients were treated as case subjects in the main analyses for both bipolar disorder and psychotic disorders; however, we verified whether the potential for multiple outcome states affected the PRS associations by fitting multistate models using the mstate package in R.

The demographic and clinical characteristics of the study sample are summarized in Table 1. Patients were predominantly young (80% were diagnosed before age 25) and female (69%). The majority of patients (61%) were treated in outpatient settings, and the most common severity specification (44%) was moderate depression. More than 80% were diagnosed as having an ICD-10 code F32 single depressive episode, and 84% had depression as their main diagnosis. Participants were followed for a maximum of 21.1 years, with a median follow-up time of 7.3 years (interquartile range, 5.4–10.0 years) for bipolar disorder and 7.1 years (interquartile range, 5.1–9.8 years) for psychotic disorders.

The PRSs for bipolar disorder and schizophrenia were moderately correlated (Pearson’s r=0.40, p<0.0001). There were weaker but significant correlations between the PRSs for major depression and for bipolar disorder (r=0.14, p<0.0001) and between the PRSs for major depression and for schizophrenia (r=0.13, p<0.0001). The mean PRSs among patients who progressed to bipolar disorder or psychotic disorders are listed in Table S2 in the online supplement, and hazard ratios are listed in Table S3.

PRSs for both bipolar disorder and schizophrenia were associated with progression to bipolar disorder in the unadjusted models; however, after mutual adjustment, only the PRS for bipolar disorder was significantly associated with progression to bipolar disorder (progression to bipolar disorder for each one-standard-deviation increase in PRS score for bipolar disorder, adjusted hazard ratio=1.11, 95% CI=1.03, 1.21, p=0.009). Only the PRS for schizophrenia was significantly associated with progression to psychotic disorders (adjusted hazard ratio=1.10, 95% CI=1.04, 1.16, p=0.0004) (Figure 2; see also Table S3 in the online supplement). There were no interactions between PRS variables.

The effect of the PRS for schizophrenia on the hazard of progression to schizophrenia was slightly smaller than its effect on progression to psychotic disorders more generally (adjusted hazard ratio=1.08, 95% CI=1.00, 1.17, p=0.05). The PRS for schizophrenia had the strongest association with progression to any disorder with psychotic features (adjusted hazard ratio=1.08, 95% CI=1.02, 1.14, p=0.005) (see Figure S1 in the online supplement), but this is relatively uninformative given that psychotic disorders constituted more than 80% of the diagnoses in this category (see Table S1 in the online supplement).

There was a statistically significant interaction (β=0.18, p<0.0001) between the PRSs for bipolar disorder and for schizophrenia as risk factors for progression to affective psychotic disorders (both as a composite category and for psychotic depression alone). Depressed patients with high PRSs for both bipolar disorder and schizophrenia were at increased risk for progression to affective psychosis; however, among patients with low liability for either bipolar disorder or schizophrenia, higher liability for the other was associated with decreased risk for affective psychotic disorders (see Figure S2 in the online supplement).

The effects of parental history on progression to bipolar disorder and psychotic disorders are shown in Figure 3. There was a significant effect of parental history on the hazard of progression to psychotic disorders and, in particular, to bipolar disorder, such that patients with a parental history of bipolar disorder were more than five times as likely to progress to bipolar disorder compared with patients with no parental history, and patients with a parental history of psychotic disorders were 63% more likely to progress to psychotic disorders compared with individuals with no parental history. The mean PRSs for bipolar disorder and schizophrenia were higher among patients with a parental history of bipolar disorder and psychotic disorders, respectively, compared with individuals with no parental history (see Figure S3 in the online supplement). However, the associations between parental history and progression were only slightly attenuated after controlling for PRS variables (Figure 3). For progression to secondary outcomes, only parental history of psychotic disorders was associated with progression to schizophrenia or any psychotic diagnosis (see Figure S4 in the online supplement).

Overall, absolute risk of progression to bipolar disorder and psychotic disorders was 7.3% (95% CI=6.4, 8.3; N=712) and 13.8% (95% CI=12.2, 15.5; N=1,640), respectively (see Table S4 in the online supplement). The absolute risks of progression to bipolar disorder and psychotic disorders by PRS quartile are summarized in Table S5 in the online supplement. The estimated absolute risk of progression to bipolar disorder increased by approximately 0.5% per quartile of PRSs for bipolar disorder, and the estimated absolute risk of progression to psychotic disorders increased by approximately 1% per PRS quartile. In the top 1% of PRSs for bipolar disorder, absolute risk of progression to bipolar disorder was 9.1% (95% CI=7.6, 10.8) compared with 5.7% (95% CI=4.7, 6.9) in the bottom 1%. For PRSs for schizophrenia, absolute risk among the top 1% was 16.2% (95% CI=14.0, 18.7) compared with 11.6% (95% CI=10.0, 13.5) for the bottom 1%.

We estimated the absolute risk of progression to bipolar disorder and psychotic disorders separately among individuals with and without a parental history (Figure 4). Individuals who ranked in the top 1% of PRSs for bipolar disorder who also had a parental history of bipolar disorder had a 40.8% estimated absolute risk of progressing to bipolar disorder, compared with a 26.2% risk among individuals in the bottom quartile with a parental history of bipolar disorder and a 5.6% risk among individuals in the bottom quartile with no parental history. Individuals who ranked in the top 1% of PRSs for schizophrenia who also had a parental history of psychotic disorders had a 25.9% absolute risk of converting to a psychotic disorder, compared with 19.1% among individuals in the bottom quartile with a parental history of psychotic disorders and an 11.6% risk among individuals in the bottom quartile with no parental history.

The pattern of results for the primary outcomes was similar across all sensitivity analyses, suggesting that our results were fairly robust to sampling choices (see Tables S6–S9 in the online supplement). The same held true for the secondary outcomes, except that the effects of PRS for bipolar disorder on psychotic depression (adjusted hazard ratio=1.15, 95% CI=1.01, 1.31, p=0.04) and affective psychotic disorders (adjusted hazard ratio=1.21, 95% CI=1.07, 1.37, p=0.002) were stronger when only main diagnoses were considered (see Table S7 in the online supplement). Effect estimates from the multistate and competing risk models were virtually identical to the estimates from the main analyses (see Table S10 in the online supplement). The multistate models also illustrate that while the risk of psychotic disorders was elevated among depressed patients who progressed first to bipolar disorder, the risk of bipolar disorder was only slightly elevated among depressed patients who progressed first to psychotic disorders, which is in line with the ICD-10 diagnostic hierarchy (see Figure S5 in the online supplement).

Several important limitations should be taken into consideration when interpreting these results. First, the study sample did not include individuals treated for depression by general practitioners or private-practice psychiatrists, because diagnostic data for these services are not reported to the DPCRR. Only around 25% of individuals who are medically treated for depression in Denmark receive hospital-based psychiatric care within 5 years of their first antidepressant prescription (36); therefore, our results pertain predominantly to the most severe depression cases. Second, progression was measured using hospital-based contacts, and therefore we were not able to evaluate the presence of subclinical hypomanic, manic, or psychotic symptoms. Thus, our results likely underestimated the true associations between PRS and progression to bipolar and psychotic disorders. Third, the oldest iPSYCH2012 participants were only 35 years old in 2016, and the majority of participants were much younger. Thus, there may be individuals among those classified as not progressing who will receive a diagnosis of bipolar disorder or psychotic disorders in the future. This may also have biased our results toward the null. Fourth, the PRSs for bipolar disorder and schizophrenia used in this study only accounted for a limited fraction of the phenotypic variance of these disorders (30, 31). As results from larger genome-wide association studies become available, prediction of diagnostic progression will likely improve. Finally, to avoid population stratification, the sample was limited to individuals of European ancestry. Consequently, these results may not generalize outside of a Danish or European context. Genetics research as a whole is biased toward discoveries that stand to benefit individuals of European ancestry disproportionately compared with those of other ancestral backgrounds, which has important ethical implications for the field (37).