Abstract
Importance:
Antidepressants are recommended as the initial choice of treatment for moderate and severe depression, but the choice of antidepressant class can be challenging, with the possibility of serious adverse events such as increased suicidality with antidepressant initiation.
Objective:
To evaluate the utility of polygenic scores in predicting heterogeneous effects of antidepressants on suicidal thoughts.
Design:
Using genetic and clinical data on both children and adults with major depressive disorder from the All of Us Research Program, we used a target trial emulation framework to evaluate the heterogeneous effects of antidepressants on suicidal thoughts across genetic characteristics.
Setting:
Longitudinal cohort from the All of Us Research Program.
Participants:
More than 7,000 patients, both children and adults, with major depressive disorder.
Interventions:
The initiation of selective serotonin reuptake inhibitors (SSRIs), compared to the initiation of serotonin and norepinephrine reuptake inhibitors (SNRIs) and bupropion.
Main Outcomes and Measures:
The effects of antidepressants on suicidal thoughts, as well as subgroup effects across polygenic scores (PGSs) of various psychiatric disorders.
Results:
Patients with higher PGS for psychiatric disorders, particularly for attention deficit-hyperactivity disorder (ADHD), were more likely than those with lower scores to experience suicidal thoughts with the initiation of SSRIs, relative to bupropion (hazard ratio 1.66, 95%CI, 1.30–2.12, for participants with ADHD PGS higher than median vs. hazard ratio 1.06, 95%CI, 0.81–1.38, for participants with ADHD PGS lower than median; P-for-interaction, 0.01). We observed a similar trend for schizophrenia, although the interaction was not statistically significant (hazard ratio 1.52, 95%CI, 1.19–1.93, for participants with schizophrenia PGS higher than median vs. hazard ratio 1.1.17, 95%CI, 0.89–1.53, for participants with ADHD PGS lower than median; P-for-interaction, 0.16). We did not observe differences in the effects of SSRIs relative to SNRIs across PGSs.
Conclusions and Relevance:
The genetic predisposition to psychiatric disorders may, at least partially, underlie the heterogeneity in the risk of suicidal thoughts with antidepressant use. In the personalized medicine framework, PGSs for various psychiatric disorders may help tailor antidepressants to each patient to avoid serious adverse effects such as suicidal thoughts.
Introduction
Depression is a psychiatric disorder that continues to put large burden worldwide [1]. Several evidence-based treatments exist for the disorder, including pharmacotherapy, psychotherapy, electroconvulsive therapy, and transcranial magnetic stimulation. Antidepressants are recommended as the initial choice of treatment for moderate and severe depression among adults. Because the efficacies of commonly used antidepressants are generally comparable between classes, the choice of antidepressants will largely be based on the adverse effect profile [2]. However, the challenge lies in predicting the adverse effects, which is essential in determining which patients to prescribe the most commonly-used selective serotonin reuptake inhibitors (SSRIs) or to prescribe alternative medications such as serotonin and norepinephrine reuptake inhibitors (SNRIs) or bupropion. This is a question that has come up repeatedly in clinical psychiatry, but one without a clear answer [3]. Currently, finding the appropriate medication for each patient usually requires trial-and-error, and the delayed therapeutic effects of antidepressants makes this even more challenging [4].
Among the adverse effect profiles of antidepressants, perhaps one that has received the most attention is the possibility of increased suicidality with antidepressant initiation, the so called "black-box" warning of antidepressants by the Food and Drug Administration [5, 6]. As a result, clinicians have sought to predict suicidal behavior among antidepressant users, first using patients’ clinical and sociodemographic characteristics. Though several potential predictors have been found, conclusive evidence is still lacking [3, 7]. One exception may be younger age, which has been reported to be associated with increased suicidal behavior with antidepressant initiation [8], but this has sparked considerable debate, with some calls to remove the "black-box" warning amidst declining antidepressant prescription rates that led to the Food and Drug Administration’s modification of the warning [5]. As such, better predictors of antidepressants’ adverse events are needed to better inform practice.
Researchers have thus turned to patients’ genetics, including polygenic scores (PGSs), to better understand factors that may help predict antidepressant response. This may be a promising avenue given the growing evidence that psychiatric disorders, including depression, are typically complex traits, with variants associated with these diseases dispersed across the allelic spectrum [9]. There have been many studies on the link between the effectiveness of antidepressants and PGSs for depression. For instance, studies have pointed to associations between a higher PGS for major depressive disorder and a prescription of more than two antidepressants [10] as well as changes in depression scores with SSRI use [11].
There have also been attempts, with mixed success, to guide antidepressant choice with PGSs for other psychiatric disorders, namely schizophrenia, bipolar disorder, and attention-deficit hyperactivity disorder (ADHD), each of which shares genetic factors with depression [12, 13]. For instance, PGS for schizophrenia was not associated with changes in depression scores after treatment with SSRIs, though the study may have been underpowered [14]. Though some suggest that non-response to antidepressants may be due to undiagnosed bipolar disorder, bipolar disorder PGS was not associated with antidepressant response [15]. Finally, studies on PGSs for ADHD, a common comorbidity of major depressive disorder [16], was associated with higher odds of prescription of multiple antidepressants [10].
These findings, though valuable, have mostly focused on antidepressant response and less on the adverse effects of antidepressants. However, clinicians often base their choice of antidepressants on the possibility of serious adverse events such as suicidal behavior, especially when prescribing to young people [5]. As such, we evaluate the antidepressants’ effects on suicidal ideation, which is of utmost clinical importance both as a primary outcome (as one of the main symptoms of depression) and as a potential adverse event. We aimed to provide evidence on the effects of different types of antidepressants with the goal of advancing precision/genomic medicine and guiding clinical practice. We focus on antidepressants that are primarily used: SSRIs, SNRIs, and bupropion [2]. Importantly, SNRIs and bupropion are viable alternatives to SSRIs, which are often considered to be first-line antidepressants; thus, this study could inform for whom these medications should be used in place of SSRIs.
Methods
Data source
We used data from the All of Us Research Program, a large National Institutes of Health-led longitudinal cohort of a diverse sample of the US population [17]. The All of Us Research Program combines participant-derived information from surveys, electronic health records, and biospecimens (including whole genome sequences of blood samples), among other sources, and launched its national recruitment in May 2018. As of 2024, more than 835,000 participants have been enrolled, of whom more than 588,000 had provided biospecimens. Short-read whole genome sequences were available for more than 245,000 of these participants. Enrollment centers are from various settings, including health provider organizations and community partners, with emphasis on recruiting participants from groups historically underrepresented in biomedical research.
All analyses were conducted on the All of Us Researcher Workbench, a cloud-based platform that allows researchers to perform data extraction, curation, and analyses. No individual-level data were downloaded onto a local computer.
Target trial emulation
We applied the target trial emulation framework [18] to evaluate the heterogeneity in the effects of antidepressants on suicidal thoughts among patients with major depressive disorder across PGSs for several psychiatric disorders. We used an active-comparator design for drug effect comparison [19] to align the timings of eligibility evaluation, intervention, and the initiation of the follow-up period. The follow-up period started at treatment assignment and ended at the occurrence of the outcome or at five years from treatment assignment, whichever occurred first. All individuals with a diagnosis of major depressive disorder and with available outcome, treatment, and covariate data, including genomic data, available were eligible for matching.
Each participant in the treatment group was matched without replacement to the control group at a 1-to-1 ratio. We used propensity scores to match the participants, with a caliper of 0.10. Propensity scores were calculated by constructing logistic regression models of the treatment using all covariates, and the matched treatment and control samples were compared by calculating the standardized mean differences (SMDs) for each covariate. We included a square term of age in the propensity score model to obtain adequate balance between the covariates. Data on all participants meeting the inclusion criteria were used to construct the models, meaning some participants in the propensity score models were not included in the final matched sample. Matching was performed sequentially in chronological order of the month of antidepressant initiation and was done separately for each month. The order of matching of participants with the same month of antidepressant initiation was chosen at random.
Variable extraction
Data for all variables were extracted from the All of Us Researcher Workbench. The All of Us Research Program records different diseases and drugs based on the codes in each participating hospital. The codes are thus not uniform: one patient may have a disease coded in International Classification of Disease (ICD)-10, while another may have a disease coded in ICD-9. Thus, the All of Us Research Program uses the Systematized Nomenclature of Medicine (SNOMED) [20] to match up these disease codes. Similarly, the All of Us Research Program records drugs using RxNorm [21], which codes all medications available on the US market. We manually surveyed all SNOMED and RxNorm keywords related to the diseases and drugs of interest in the All of Us Research Program, and chose relevant SNOMED and RxNorm keywords for each disease or drug, as well as the dates they were recorded. The keywords we used are available in eMethods. All other variables, including sociodemographic variables, were recorded via a survey by the All of Us Research Program.
This study sought to emulate a RCT that performs comparisons of two pairs of antidepressants: SSRIs vs. bupropion, and SSRIs vs. SNRIs. In both comparisons, treatment effects were estimated with SSRIs as the treatment and the other drug as the control.
Participants were allocated into the treatment or control group based on whichever drug was initiated first. For instance, if an individual was started on SSRIs and was later given SNRIs, the individual was classified into the SSRI group. Thus, all analyses of treatment effects were intention-to-treat analyses. If an individual was given both drugs on the same date, they were excluded from the analyses.
The outcome of interest was suicidal thoughts, as coded in electronic health records. As suicidal ideation is most common with the initiation of antidepressants, we did not evaluate the long-term effects of antidepressants, and the outcome was censored at five years.
We included the following covariates obtained from self-reported surveys in the matching process: gender (female, male, or other), age, race and ethnicity (Non-Hispanic Black, Non-Hispanic White, Hispanic, or other), education level (Whether they attended college), annual income (less than $25,000, $25,000 to $50,000, $50,000 to $100,000, or more than $100,000), health insurance coverage, birthplace (US or other), and smoking status (at least 100 cigarettes in one’s lifetime or not). All survey variables were recorded at the time of recruitment into the All of Us Research Program, regardless of the timing of drug initiation.
We also included the previous diagnoses of suicidal thoughts, generalized anxiety disorder, alcohol abuse, schizophrenia, eating disorders, post-traumatic stress disorder (PTSD), and ADHD, as well as the previous use of tricyclic antidepressants. Here, a previous diagnosis or previous use of drug was defined as the coding of the disease or drug in the electronic health record any time before the initiation of the drug of interest. We did not include the previous diagnosis of any other mood disorders.
We included PGSs for depression, self-injurious behavior, bipolar disorder, schizophrenia, anxiety, and attention deficit-hyperactivity disorder (ADHD) as covariates. These were chosen based on previous literature on potential associations between PGSs and treatment effect heterogeneity of antidepressants and psychiatric disorders that have genetic overlap with depression [10, 12, 14, 15], as well as the availability of each PGS on PGS catalog. Of note, the PGS for anxiety included not only generalized anxiety disorder but also panic disorder (episodic paroxysmal anxiety), mixed anxiety and depressive disorder, other mixed anxiety disorders, other specified anxiety disorders, and anxiety disorder, unspecified.
We obtained whole genome sequences from each participant and calculated their PGS using weights from selected PGSs from the PGS catalog. Each PGS was standardized across the full All of Us Research Program sample prior to analyses. For each PGS, we calculated the Nagelkerke’s R-squared value and area under the ROC (AUROC), for the full All of Us Research Program sample and by race and ethnicity. Evaluation of PGSs were performed using all available participant data; i.e., all participants with available PGSs and outcomes, even if they did not meet the inclusion criteria for the evaluation of treatment effects, were included.
Statistical analysis
First, we first compared the risks of suicidal thoughts between SSRI users and bupropion users and between SSRI users and SNRI users by plotting a Kaplan-Meier curve. In addition, we fit Cox proportional hazard models to estimate hazard ratios and 95% confidence intervals (CIs) for the outcome events.
Second, we conducted subgroup analyses to evaluate the heterogeneity in the risks of suicidal thoughts associated with antidepressant use across PGSs and several important demographic and clinical variables (gender, age, smoking status). Gender was dichotomized as the number of individuals reporting to be neither male nor female were small. Age was trichotomized to group the participants into roughly same subsamples and to capture the possibility that antidepressants respond differently among young populations. PGSs were dichotomized at the median for better interpretability. P-values were calculated for the coefficients of the interaction terms between the treatment and covariates of interest in the Cox proportional hazard models.
Ethical considerations
The All of Us Research Program has been approved by the All of Us Institutional Review Board. As the Program follows a "passport model" that grants broad access to the research dataset approved by the program institutional review board instead of the institutional review board at researchers’ affiliations, the present study was exempt from obtaining ethical approval from authors’ institutions. All researchers who access the All of Us Research Program data are authorized and approved via a process that includes registration, affiliation with an institution that has completed a Data Use and Registration Agreement, identity verification, completion of ethics training, and attestation to a data use agreement. Results reported comply with the All of Us Data and Statistics Dissemination Policy that prohibits disclosure of group counts under 20 to protect participant privacy.
Results
A total of 7,848 participants were matched for SSRIs vs. bupropion and 9,316 participants were matched for SSRIs vs. SNRIs, and both samples included children and adults. In the SSRIs vs. bupropion sample, 68.0% were female, the mean age was 47.8 years (standard deviation (SD), 14.4), and 11.3% identified as Hispanic and 14.4% as Non-Hispanic Black. Among the previous diagnoses surveyed, anxiety was the most prevalent at 9.4%. In the SSRIs vs. SNRIs sample, 71.1% were female, the mean age was 49.6 (standard deviation (SD), 14.3), and 12.2% identified as Hispanic and 15.1% as Non-Hispanic Black. Among the previous diagnoses surveyed, anxiety was the most prevalent at 9.0%. Both the SSRIs vs. bupropion and SSRIs vs. SNRIs samples achieved good balance of covariate distributions, including previous diagnoses of psychiatric disorders and PGSs (Tables 1 and 2).
Table 1:
Comparison of matched bupropion and SSRI samplesa
| Bupropion n = 3924 |
SSRI n = 3924 |
SMD | |
|---|---|---|---|
|
| |||
| Gender (%) | 0.048 | ||
| Female | 2631 (67.0) | 2704 (68.9) | |
| Male | 1267 (32.3) | 1187 (30.2) | |
| Other | 26 ( 0.7) | 33 ( 0.8) | |
| Age (mean (SD)) | 47.93 (14.08) | 47.58 (14.76) | 0.024 |
| Race and ethnicity (%) | 0.099 | ||
| Hispanic | 391 (10.0) | 497 (12.7) | |
| Non-Hispanic Black | 542 (13.8) | 590 (15.0) | |
| Non-Hispanic White | 2814 (71.7) | 2659 (67.8) | |
| Other | 177 ( 4.5) | 178 ( 4.5) | |
| College education or higher (%) | 3030 (77.2) | 2906 (74.1) | 0.074 |
| Annual income (%) | 0.077 | ||
| -$25,000 | 1315 (33.5) | 1420 (36.2) | |
| $25,000-$50,000 | 785 (20.0) | 830 (21.2) | |
| $50,000-$100,000 | 987 (25.2) | 909 (23.2) | |
| $100,000- | 837 (21.3) | 765 (19.5) | |
| Has health insurance (%) | 3823 (97.4) | 3765 (95.9) | 0.083 |
| Born in the US (%) | 3621 (92.3) | 3555 (90.6) | 0.060 |
| Ever-smoker (%) | 2180 (55.6) | 1975 (50.3) | 0.105 |
| Previous diagnosis of suicidal thoughts (%) | 148 (3.8) | 210 (5.4) | 0.076 |
| Previous diagnosis of anxiety (%) | 338 (8.6) | 396 (10.1) | 0.051 |
| Previous diagnosis of alcohol abuse (%) | 218 (5.6) | 227 (5.8) | 0.010 |
| Previous diagnosis of substance abuse (%) | 183 (4.7) | 198 (5.0) | 0.018 |
| Previous diagnosis of schizophrenia (%) | 54 (1.4) | 52 (1.3) | 0.004 |
| Previous diagnosis of eating disorders (%) | 12 (0.3) | 12 (0.3) | <0.001 |
| Previous diagnosis of PTSD (%) | 244 (6.2) | 254 (6.5) | 0.010 |
| Previous diagnosis of ADHD (%) | 194 (4.9) | 138 (3.5) | 0.071 |
| Previous tricyclic antidepressant use (%) | 351 (8.9) | 339 (8.6) | 0.011 |
| PGS for depression (mean (SD)) | 0.00 (1.03) | 0.02 (0.98) | 0.018 |
| PGS for suicidal thoughts (mean (SD)) | 0.05 (0.99) | 0.05 (1.01) | 0.001 |
| PGS for anxiety (mean (SD)) | −0.01 (1.01) | −0.01 (1.01) | 0.005 |
| PGS for bipolar disorder (mean (SD)) | −0.19 (0.98) | −0.16 (0.98) | 0.029 |
| PGS for schizophrenia (mean (SD)) | −0.29 (0.92) | −0.23 (0.94) | 0.067 |
| PGS for ADHD (mean (SD)) | −0.14 (0.96) | −0.10 (0.97) | 0.037 |
Each PGS was standardized across the full All of Us Research Program sample prior to analyses. SSRI, serotonin reuptake inhibitor; SMD, standardized mean difference; PTSD, post-traumatic stress disorder; ADHD, attention deficit-hyperactivity disorder; PGS, polygenic score.
Table 2:
Comparison of matched SNRI and SSRI samplesb
| SNRI n = 4658 |
SSRI n = 4658 |
SMD | |
|---|---|---|---|
|
| |||
| Gender (%) | 0.063 | ||
| Female | 3376 (72.5) | 3248 (69.7) | |
| Male | 1240 (26.6) | 1371 (29.4) | |
| Other | 42 ( 0.9) | 39 ( 0.8) | |
| Age (mean (SD)) | 50.47 (13.80) | 48.69 (14.70) | 0.125 |
| Race and ethnicity (%) | 0.070 | ||
| Hispanic | 528 (11.3) | 608 (13.1) | |
| Non-Hispanic Black | 683 (14.7) | 725 (15.6) | |
| Non-Hispanic White | 3248 (69.7) | 3104 (66.6) | |
| Other | 199 ( 4.3) | 221 ( 4.7) | |
| College education or higher (%) | 3515 (75.5) | 3404 (73.1) | 0.055 |
| Annual income (%) | 0.048 | ||
| -$25,000 | 1885 (40.5) | 1877 (40.3) | |
| $25,000-$50,000 | 991 (21.3) | 927 (19.9) | |
| $50,000-$100,000 | 1018 (21.9) | 1019 (21.9) | |
| $100,000- | 764 (16.4) | 835 (17.9) | |
| Has health insurance (%) | 4526 (97.2) | 4490 (96.4) | 0.044 |
| Born in the US (%) | 4288 (92.1) | 4236 (90.9) | 0.040 |
| Ever-smoker (%) | 2380 (51.1) | 2286 (49.1) | 0.040 |
| Previous diagnosis of suicidal thoughts (%) | 208 (4.5) | 271 (5.8) | 0.061 |
| Previous diagnosis of anxiety (%) | 384 (8.2) | 459 (9.9) | 0.056 |
| Previous diagnosis of alcohol abuse (%) | 233 (5.0) | 285 (6.1) | 0.049 |
| Previous diagnosis of substance abuse (%) | 189 (4.1) | 236 (5.1) | 0.048 |
| Previous diagnosis of schizophrenia (%) | 53 (1.1) | 71 (1.5) | 0.034 |
| Previous diagnosis of eating disorders (%) | 7 (0.2) | 13 (0.3) | 0.028 |
| Previous diagnosis of PTSD (%) | 340 (7.3) | 353 (7.6) | 0.011 |
| Previous diagnosis of ADHD (%) | 140 (3.0) | 172 (3.7) | 0.038 |
| Previous tricyclic antidepressant use (%) | 572 (12.3) | 399 (8.6) | 0.122 |
| PGS for depression (mean (SD)) | 0.02 (1.04) | 0.03 (1.00) | 0.015 |
| PGS for suicidal thoughts (mean (SD)) | 0.06 (1.02) | 0.05 (1.01) | 0.016 |
| PGS for anxiety (mean (SD)) | 0.00 (1.02) | 0.00 (1.01) | 0.004 |
| PGS for bipolar disorder (mean (SD)) | −0.18 (0.99) | −0.15 (0.99) | 0.026 |
| PGS for schizophrenia (mean (SD)) | −0.26 (0.93) | −0.21 (0.95) | 0.050 |
| PGS for ADHD (mean (SD)) | −0.11 (0.96) | −0.08 (0.97) | 0.033 |
Each PGS was standardized across the full All of Us Research Program sample prior to analyses. SSRI, serotonin reuptake inhibitor; SNRI, serotonin and norepinephrine reuptake inhibitor; SMD, standardized mean difference; PTSD, post-traumatic stress disorder; ADHD, attention deficit-hyperactivity disorder; PGS, polygenic score.
The sources of the selected PGSs and links in the PGS catalog are available in eTable 1. In general, the PGSs achieved modest performance in predicting the psychiatric disorders (AUROCs up to 0.67), though in some cases the AUROCs were close to 0.5. They tended to perform worse in groups other than non-Hispanic Whites (eFigures 1–6).
SSRIs vs. bupropion
SSRIs had a higher risk of suicidal thoughts relative to bupropion (hazard ratio, 1.36; 95% confidence interval (CI), 1.14 to 1.63; P-value < 0.001; Figure 1a). Participants generally had similar hazard ratios across nongenetic characteristics (Figure 2). Participants with PGS for ADHD higher than the median had a hazard ratio of 1.66 (95%CI, 1.30 to 2.12), while participants with PGS for ADHD lower than the median had a hazard ratio of 1.06 (95%CI, 0.81 to 1.38; P-for-interaction, 0.01). We observed a similar trend for schizophrenia: participants with PGS for schizophrenia higher than the median had a hazard ratio of 1.52 (95%CI, 1.19 to 1.93), while participants with PGS for schizophrenia lower than the median had a hazard ratio of 1.17 (95%CI, 0.89 to 1.53; P-for-interaction, 0.16; Figure 2).
Figure 1:
Kaplan-Meier curves of suicidal thoughts.
Figure 2:
Subgroup analyses for SSRIs vs. bupropion.
SSRIs vs. SNRIs
SSRIs and SNRIs had comparable risks of suicidal thoughts (hazard ratio of SSRIs relative to SNRIs, 1.05; 95%CI, 0.91 to 1.22; P-value, 0.50; Figure 1b). Males had a hazard ratio of 1.31 (95%CI, 1.04 to 1.65), while other genders had a hazard ratio of 0.86 (95%CI, 0.70 to 1.05; P-for-interaction, 0.01). Hazard ratios were similar across PGSs, with all PGS subgroups having treatment effects close to the null (Figure 3).
Figure 3:
Subgroup analyses for SSRIs vs. SNRIs.
Discussion
We find evidence that bupropion may be less likely to elicit suicidal thoughts compared to SSRIs among individuals with high genetic predisposition to psychiatric disorders, particularly for ADHD. Notably, conventional subgroup analyses across nongenetic characteristics did not reveal heterogeneity in the risk of suicidal thoughts. These findings suggest that considering PGSs in the personalized medicine framework may help clinicians decide which antidepressant should be prescribed for each patient to avoid serious adverse effects such as suicidal thoughts.
We find that bupropion may be less likely to elicit suicidal thoughts among individuals with high PGSs for ADHD. This may be associated with the fact that bupropion improves ADHD symptoms [22], which includes impulsivity and hyperactivity, and supports the role of impulsivity in suicidal ideation and behavior [23]. Note that our results suggest that the genetic predisposition to ADHD alone, regardless of whether it has clinically manifested, could be valuable in tailoring antidepressant treatment.
We did not find heterogeneity in the effects of SSRIs relative to SNRIs across PGSs. Instead, we found that males may have lower risk of increased suicidal thoughts with SNRI (relative to SSRI) use compared to other genders. This is generally in agreement with previous RCTs, which suggest that females may respond better to antidepressants than males and that this trend is more evident for SSRIs than SNRIs [24], though direct comparisons between SSRIs and SNRIs are not well-documented. Thus, this gender difference may be a novel finding that warrants further investigation.
Thus far, it was unclear whether genetic information could guide personalized care for patients with depression. Though there have been associational studies on PGSs for various psychiatric diseases, decisionmaking in psychiatric care was, for the most part, based primarily on patients’ clinical and sociodemographic characteristics and not on their genetic information. Furthermore, even when genetic information is available, few large-scale databases have both clinical and genetic data that can evaluate its utility in predicting the adverse events of antidepressants. Combining clinical and genetic information on a large sample of antidepressant users from the All of Us Research Program, the present study filled this knowledge gap on the utility of genetic information in personalized psychiatric care: at least for minimizing suicidal thoughts, PGSs are likely useful in personalizing pharmacotherapy for depression. Investigating this further may provide a better picture of the pathophysiology and subtyping of depression and comorbidities, with which we can further advance personalized psychiatry.
Our study should be interpreted in light of several limitations. First, the PGSs had limited performance in predicting each psychiatric disorder, especially for minoritized populations (though they were comparable to their performance described in their respective original studies). Although this is beyond the scope of the exploratory nature of our study (in addition to the fact that the development of PGSs was not an objective of this study), attempts to improve PGS performance among the general and minoritized populations, such as the one by Lennon and others [25], may be warranted. Second, there may have been misclassification, especially for self-reported variables such as education level, annual income, health insurance coverage, birthplace, and smoking status. Third, participants in the All of Us Research Program may not be generalizable to the general population, as the program obtains data only from patients in registered facilities. Fourth, we only consider the first antidepressant that was assigned to individuals, and do not consider the duration of treatment or antidepressants that may have been subsequently initiated. As such, our study does not comprehensively capture the full spectrum of treatment regimens observed in clinical practice. Fifth, though bupropion is an alternative to SSRIs, it is also used for smoking cessation. We sought to minimize the bias that may have arisen from this by accounting for smoking status in the matching process as well as limiting the sample to patients with major depressive disorder. Finally, there may have been unmeasured confounding by factors not evaluated in the All of Us Research Program, such as individuals’ clinical, sociodemographic, or molecular profiles. Our findings should be validated in other cohorts, as more cohorts with comprehensive genetic and clinical data become available in the future.
In conclusion, among individuals with major depressive disorder, we detect heterogeneity in the risk of suicidal thoughts with antidepressant use across polygenic scores for psychiatric disorders. Considering PGSs for various psychiatric disorders may help clinicians tailor antidepressants to each patient to avoid serious adverse effects such as suicidal thoughts.
Supplementary Material
Key points.
Question: How much do polygenic scores of psychiatric diseases predict the heterogeneity in the effects of antidepressants on suicidal thoughts?
Findings: The genetic predisposition to psychiatric disorders may, at least partially, underlie the heterogeneity in the risk of suicidal thoughts with antidepressant use.
Meaning: Polygenic scores for psychiatric disorders may help tailor antidepressants to each patient to avoid suicidal thoughts with their initiation, contributing to personalized psychiatry.
Acknowledgements
We gratefully acknowledge All of Us participants for their contributions, without whom this research would not have been possible. We also thank the National Institutes of Health’s All of Us Research Program for making available the participant data and/or samples examined in this study.
Funding
R.G. receives funding from Knight-Hennessy Scholars, the Japan Foundation for Pediatric Research (22–001), and Chernobyl-Fukushima Medical Foundation for other work not related to this study. K.I. receives research support from the Japan Society for the Promotion of Science (23KK0240), the Japan Agency for Medical Research and Development (AMED; JP22rea522107), the Japan Science and Technology (JST PRESTO; JPMJPR23R2), and the Program for the Development of Next-generation Leading Scientists with Global Insight (L-INSIGHT) sponsored by the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan. The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
The All of Us Research Program is supported by the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026553; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA #: AOD 16037; Federally Qualified Health Centers: HHSN 263201600085U; Data and Research Center: 5 U2C OD023196; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 U24 OD023163; Communications and Engagement: 3 OT2 OD023205; 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276.
Data and code availability
As per All of Us Research Program Data Use Policies, individuals are prohibited from removing participant-level data from the Researcher Workbench. Data analysis code that does not contain participant-level data are available through https://github.com/ryunosukegoto/PGS-HTE.
References
- (1).Liu Q.; He H.; Yang J.; Feng X.; Zhao F.; Lyu J. Journal of Psychiatric Research 2020, 126, 134–140. [DOI] [PubMed] [Google Scholar]
- (2).Gelenberg A. J.; Freeman M.; Markowitz J.; Rosenbaum J.; Thase M.; Trivedi M.; Van Rhoads R. American Journal of Psychiatry 2010, 167, 9–118. [Google Scholar]
- (3).Meerman J. J.; ter Hark S. E.; Janzing J. G.; Coenen M. J. Journal of Affective Disorders 2022, 304, 1–11. [DOI] [PubMed] [Google Scholar]
- (4).Frazer A; Benmansour S Molecular Psychiatry 2002, 7, S23–S28. [DOI] [PubMed] [Google Scholar]
- (5).Friedman R. A. New England Journal of Medicine 2014, 371, 1666–1668. [DOI] [PubMed] [Google Scholar]
- (6).Hengartner M. P.; Amendola S.; Kaminski J. A.; Kindler S.; Bschor T.; Plöderl M. Journal of Epidemiology and Community Health 2021, 75, 523–530. [DOI] [PubMed] [Google Scholar]
- (7).Iniesta R. et al. Journal of Psychiatric Research 2016, 78, 94–102. [DOI] [PubMed] [Google Scholar]
- (8).Hammad T. A.; Laughren T.; Racoosin J. Archives of General Psychiatry 2006, 63, 332–339. [DOI] [PubMed] [Google Scholar]
- (9).Sullivan P. F.; Daly M. J.; O’donovan M. Nature Reviews Genetics 2012, 13, 537–551. [DOI] [PMC free article] [PubMed] [Google Scholar]
- (10).Fabbri C.; Hagenaars S. P.; John C.; Williams A. T.; Shrine N.; Moles L.; Hanscombe K. B.; Serretti A.; Shepherd D. J.; Free R. C., et al. Molecular Psychiatry 2021, 26, 3363–3373. [DOI] [PMC free article] [PubMed] [Google Scholar]
- (11).Ward J.; Graham N.; Strawbridge R. J.; Ferguson A.; Jenkins G.; Chen W.; Hodgson K.; Frye M.; Weinshilboum R.; Uher R., et al. PloS One 2018, 13, e0203896. [DOI] [PMC free article] [PubMed] [Google Scholar]
- (12).Levey D. F.; Stein M. B.; Wendt F. R.; Pathak G. A.; Zhou H.; Aslan M.; Quaden R.; Harrington K. M.; Nuñez Y. Z.; Overstreet C., et al. Nature Neuroscience 2021, 24, 954–963. [DOI] [PMC free article] [PubMed] [Google Scholar]
- (13).Cross-Disorder Group of the Psychiatric Genomics Consortium The Lancet 2013, 381, 1371–1379. [DOI] [PMC free article] [PubMed] [Google Scholar]
- (14).García-González J.; Tansey K. E.; Hauser J.; Henigsberg N.; Maier W.; Mors O.; Placentino A.; Rietschel M.; Souery D.; Žagar T., et al. Progress in Neuro-Psychopharmacology and Biological Psychiatry 2017, 75, 128–134. [DOI] [PubMed] [Google Scholar]
- (15).Tansey K. E.; Guipponi M.; Domenici E.; Lewis G.; Malafosse A.; O’Donovan M.; Wendland J. R.; Lewis C. M.; McGuffin P.; Uher R. American Journal of Medical Genetics Part B: Neuropsychiatric Genetics 2014, 165, 77–83. [DOI] [PubMed] [Google Scholar]
- (16).Bron T. I.; Bijlenga D.; Verduijn J.; Penninx B. W.; Beekman A. T.; Kooij J. S. Journal of Affective Disorders 2016, 197, 29–35. [DOI] [PubMed] [Google Scholar]
- (17).Ramirez A. H. et al. Patterns 2022, 3, 100570. [DOI] [PMC free article] [PubMed] [Google Scholar]
- (18).Hernán M. A.; Dahabreh I. J.; Dickerman B. A.; Swanson S. A. Annals of Internal Medicine 2025, 178, 402–407. [DOI] [PMC free article] [PubMed] [Google Scholar]
- (19).Suissa S.; Moodie E. E.; Dell’Aniello S. Pharmacoepidemiology and Drug Safety 2017, 26, 459–468. [DOI] [PubMed] [Google Scholar]
- (20).Cote R. A.; Robboy S. JAMA 1980, 243, 756–762. [DOI] [PubMed] [Google Scholar]
- (21).Liu S.; Ma W.; Moore R.; Ganesan V.; Nelson S. IT Professional 2005, 7, 17–23. [Google Scholar]
- (22).Wilens T. E.; Spencer T. J.; Biederman J.; Girard K.; Doyle R.; Prince J.; Polisner D.; Solhkhah R.; Comeau S.; Monuteaux M. C., et al. American Journal of Psychiatry 2001, 158, 282–288. [DOI] [PubMed] [Google Scholar]
- (23).Bruno S.; Anconetani G.; Rogier G.; Del Casale A.; Pompili M.; Velotti P. Journal of Affective Disorders 2023, 339, 571–583. [DOI] [PubMed] [Google Scholar]
- (24).Khan A.; Brodhead A. E.; Schwartz K. A.; Kolts R. L.; Brown W. A. Journal of Clinical Psychopharmacology 2005, 25, 318–324. [DOI] [PubMed] [Google Scholar]
- (25).Lennon N. J.; Kottyan L. C.; Kachulis C.; Abul-Husn N. S.; Arias J.; Belbin G.; Below J. E.; Berndt S. I.; Chung W. K.; Cimino J. J., et al. Nature Medicine 2024, 30, 480–487. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
As per All of Us Research Program Data Use Policies, individuals are prohibited from removing participant-level data from the Researcher Workbench. Data analysis code that does not contain participant-level data are available through https://github.com/ryunosukegoto/PGS-HTE.