An examination of the impact of treatment on later risk for a substance use disorder in young people with major depressive disorder and bipolar disorder

Abstract

Objective

We examined the impact of treatment of young people with major depressive disorder (MDD) and bipolar disorder (BPD) on risk for subsequent substance use disorder (SUD).

Methods

Patients aged 16–30 years seen at Mass General Brigham between 02/16/15 and 07/14/23 with MDD or BPD and no history of SUD were included. Diagnoses were determined using electronic health record data (billing codes, problem lists, patient-reported outcome measures, and other proxies of diagnosis(es)). Patients were treated or untreated based on pharmacological and/or psychosocial treatment following the onset of MDD/BPD and prior to SUD onset. Groups were matched on propensity scores (1:1) and compared on the development of SUD using a Cox regression model.

Results

A total of 3601 and 796 patients (mean age 22.6 ± 4.3 years) were identified with no SUD and MDD or BPD, respectively. Final analysis of matched patients with and without treatment included 1666 with MDD and 314 with BPD. The most common treatment was pharmacologic only for MDD (89.3 %) and BPD (94.3 %). Overall, there was no significant difference in the development of SUD between those treated and not treated for MDD (HR=1.06 [0.84, 1.32], p = 0.63) nor those treated and not treated for BPD (HR=0.80 [0.49, 1.30], p = 0.37). However, longer duration of treatment for both MDD and BPD was associated with a significant reduction in risk for SUD. There was a 2.1 % risk reduction for each additional cumulative month of treatment for MDD (HR=0. 979 [0.976, 0.981], p < 0.001) and a 2.6 % risk reduction for each additional cumulative month of treatment for BPD (HR=0.974 [0.968, 0.979], p < 0.001).

Conclusion

Longer treatment duration for young people with MDD and BPD is associated with decreased SUD risk.

Highlights

• •Overall, treatment for depressed youth did not decrease risk for substance use disorder (SUD).
• •Overall, treatment for bipolar youth did not decrease risk for SUD.
• •However, longer treatment duration for both depressed and bipolar youth decreased SUD risk.

1. Introduction

The prevalence of psychiatric disorders remains high with 23 % of adults in the United States endorsing a past-year mental illness in 2023 [1]. Mood disorders, including major depressive disorder (MDD) and bipolar disorder (BPD), are among the most common disorders with lifetime prevalences of 21 % and 1.5 % respectively [2], [3], [4]. Both disorders are associated with reduced quality of life and functional disability [5] and are highly comorbid with other psychiatric disorders including substance use disorder (SUD). For example, SUD co-occurs among 5.8–25 % of individuals with MDD [6], [7] and approximately 60 % of individuals with BPD [8], [9].

Individuals with SUD co-occurring with MDD or BPD have worse outcomes than individuals with either of these disorders alone [10], [11], [12]. Given the high rate of comorbidity between MDD/BPD and SUD, and the substantial impairment resulting from their co-occurrence, it is important to further understand opportunities for SUD prevention and early intervention. This is particularly true in young people with mood disorders, who are at higher risk to develop a SUD during late adolescence to early adulthood [13].

Clinically, practitioners may assume that treating young people with mood disorders will decrease risk for subsequent adverse outcomes such as SUDs. However, a 2022 systematic review exploring this assumption found that only 3 of 29 studies examining the impact of pharmacotherapy on SUD risk in young people with psychopathology focused on MDD or BPD [14]. One of these studies, a randomized controlled trial of 192 young people with MDD that used structured interviews to assess mental health and SUD diagnoses, reported that those who responded to MDD treatment were less likely to subsequently develop a drug use disorder when compared to treatment non-responders [15]. Similarly, a retrospective cohort study that utilized survey data from 121,576 adolescents with MDD found that those who did not report treatment manifested a higher likelihood of nonmedical use of prescription pain medication [16]. In contrast, a prospective 6-year longitudinal follow-up study that used structured interviews of 167 young people, who ranged from age 12–17-years at baseline, reported mixed findings regarding the impact of BPD treatment on substance use outcomes [17]. Although treatment with lithium was associated with a decreased risk of SUD, antidepressants and other psychotropic drugs were associated with an increased risk of SUD [17].

With a limited literature and mixed findings, the high co-occurrence of mood disorders and SUD, and the additional morbidity associated with the co-occurring disorders, further research is needed to examine the impact of treatment and treatment duration for young people with mood disorders on risk for SUD. To this end, the current study uses the electronic health record (EHR) to evaluate the impact of treatment of MDD and BPD in young people separately on the risk for subsequent SUD. We hypothesized that treatment of young people with MDD and BPD would decrease the risk for the subsequent development of SUD.

2. Methods

2.1. Patient inclusion

This study included patients aged 16–30 years seen at Mass General Brigham (MGB) facilities between 02/16/2015 and 07/14/2023 with MDD or BPD who did not have a co-occurring non-nicotine SUD at baseline. The study start date corresponds to the implementation of a standardized EHR, EPIC, within the MgB system. MGB is a large not-for-profit, integrated healthcare system in the Northeast which provides care for over 1.5 million patients annually. It encompasses primary care and specialty services at hospitals and community health centers across New England.

Patient characteristics were extracted from the Epic EHR through review of records from primary care adult and pediatric medicine, behavioral health, and other community and hospital-based practices. Individuals with MDD or BPD with psychotic features were included, but patients with a diagnosis of a primary psychotic disorder (e.g., schizophrenia or schizoaffective disorder, ICD-10 codes: F20.*-F29.*) were excluded. Since we were interested in the subsequent development of non-nicotine SUD after MDD/BPD onset, we also excluded patients who met criteria for a non-nicotine SUD prior to or during an initial MDD or BPD episode.

2.2. MDD/BPD ascertainment

A multi-layered approach was utilized to identify individuals with MDD or BPD, which combined billing codes, problem lists, patient reported outcome measures (PROMs), and prescription to establish proxy diagnoses [18]. Briefly, the presence of MDD was determined using a validated methodology for the identification of diagnosis through EHR data collation which exhibited 100 % positive predictive value for MDD [18]. Patients were considered to have MDD if they met one of the following criteria: (1) a diagnostic International Classification of Diseases, Tenth Revision (ICD-10) billing code for MDD linked to visit encounters or patient problem lists (F32.*-F39.*); (2) positive responses for MDD criteria on PROMs; or (3) receipt of medication for MDD treatment documented in the patient chart with subsequent evidence of MDD. For patients who reached multiple criteria, the MDD index date, or MDD “start” date, was the date at which the earliest criterion was reached.

In alignment with MDD, patients were identified as having BPD if their charts contained (1) relevant ICD-10 codes (F30.*-F31.*) and/or (2) receipt of medication for BPD with a later diagnosis of BPD.

2.3. SUD ascertainment

Young people with a non-nicotine SUD diagnosis were identified based on (1) diagnostic billing codes (F10.*-F16.* and F18.*-F19.*), (2) PROMs responses, and (3) receipt of FDA-approved medications for SUD treatment. These diagnostic criteria are detailed in the Supplemental Materials and have been validated in a sample of patients with MDD and SUD exhibiting a sensitivity and specificity of 92 % [18].

2.4. Design

MDD and BPD samples were analyzed separately, using equivalent methodologies. For each sample, patients were categorized as either ‘treated’ or ‘untreated’ based on evidence of psychosocial and/or pharmacological therapy documented in the patient chart following the onset of MDD/BPD and prior to any non-nicotine SUD onset. Regarding pharmacotherapy, patients were considered to be treated if at any point in the study period they received a relevant medication at a minimum effective dose for greater than or equal to 4 weeks [19], [20]. Patients with documented receipt of psychosocial therapy were considered treated if they attended 6 sessions of therapy within a 6-month period [21].

2.5. Covariates

Propensity score matching (1:1) was conducted for treated and untreated groups. Key covariates for propensity score matching were derived from the EHR. Covariates were selected due to known associations with SUD risk [22] and included patient age at index of MDD/BPD, sex at birth, median income in zip code of residence at index, insurance at index, prior encounters with a provider in behavioral health, and other historical or current psychiatric comorbidities. Data were also collected on utilization of any within-system medical/psychosocial healthcare following MDD/BPD index and included psychiatric and non-psychiatric emergency department visits and psychiatric and non-psychiatric hospitalizations.

2.6. Statistical methods

Within each sample, the primary analysis examined the risk of developing a non-nicotine SUD in the propensity score matched sample of treated versus untreated patients using a Cox proportional hazards model. To assess the robustness of the findings in the primary analysis, two sensitivity analyses were conducted. The first examined the risk of developing a non-nicotine SUD in the full sample of treated versus untreated patients using a Cox proportional hazards model that controlled for propensity score. The second examined the risk of developing a non-nicotine SUD in the full sample of patients using a Cox proportional hazards model with a time-varying covariate for treatment and controlling for the sociodemographic and clinical characteristics used to calculate propensity scores. In this second sensitivity analysis, patients were not categorized as treated versus untreated, but instead, their follow-up time was categorized into treated versus untreated periods and the risk of developing a non-nicotine SUD was compared between times with treatment versus times without treatment.

Additionally, within each sample, we applied the primary analysis to development of specific SUDs, examining the risk for AUD versus no SUD and non-AUD SUD versus no SUD. We also used a Cox proportional hazards model to examine the relationship between treatment duration (cumulative number of months) and the risk for developing a non-nicotine SUD in the full sample, controlling for baseline characteristics.

To examine who was likely to develop a SUD in our two samples, we also investigated the demographic and clinical characteristics of those who did and did not develop a non-nicotine SUD. We analyzed our data using a training and test set approach which allowed us to build a model to predict development of non-nicotine SUD in one dataset and evaluate how well the model worked in a separate, unbiased dataset. Detailed methods can be found in the Supplemental Materials.

All tests were two-tailed and performed at the 0.05 alpha level using Stata 18.0 (StataCorp, 2023).

3. Results

3.1. Major depressive disorder analysis

3.1.1. Sample derivation

A total of 3601 patients with MDD were included in our analyses (Fig. 1A). These patients were followed within the healthcare system for an average of 1158.4 ± 808.1 days. Stratification by depression treatment status identified 833 untreated patients and 2768 treated patients. One-to-one propensity score matching restricted our sample to 1666 patients including 833 untreated patients and 833 treated patients. Twenty-two percent (n = 783/3601) of patients in the full sample and 19 % (n = 316/1666) of patients in the matched sample developed a non-nicotine SUD.

Fig. 1.

CONSORT Diagram. 1A. Depression Analysis. 1B. Bipolar Analysis.

Among the 2768 treated patients in the full sample, 89 % (n = 2459) received pharmacologic therapy only, 3 % (n = 77) received psychosocial therapy only, and 8 % (n = 231) received both pharmacologic and psychosocial therapy. Among the 833 treated patients in the matched sample, 90 % (n = 753) received pharmacologic therapy only, 4 % (n = 28) received psychosocial therapy only, and 6 % (n = 52) received both pharmacologic and psychosocial therapy.

3.1.2. Matched sample characteristics

In the matched sample, patients had a mean age of 22.9 years, were 64 % female, and 69 % white (Table 1A). The average median income was $104,143.30 and 50 % of patients had private insurance. The most common psychiatric comorbidity was anxiety disorders (55 %) and the least common was disruptive/conduct disorders (<1 %) (Table 2A). Emergency department (ED) visits and hospitalizations for non-psychiatric reasons (31 % and 11 %, respectively) were more common than ED visits and hospitalizations for psychiatric reasons (11 % and 3 %, respectively). Thirty-two percent of patients had a previous visit with a psychiatric provider prior to the date of their depression diagnosis. Sociodemographic and clinical characteristics for the whole sample can be found in the Supplementary Materials (Tables S2 and S3).

Table 1.

Sociodemographic characteristics in the matched depression and matched bipolar samples.

	A. Matched Depression Sample N = 1666	B. Matched Bipolar Sample N = 314
	Mean ± SD	Mean ± SD
Age at index diagnosis (years)	22.9 ± 4.3	23.6 ± 4.2
Length of follow-up (days)	1052.1 ± 788.5	1099.6 ± 859.1
Median Income (dollars)	104,143.3 ± 37,359.3	103,027.8 ± 34,531.9
	N (%)	N (%)
Sex
Male	599 (36)	106 (34)
Female	1067 (64)	208 (66)
Race
American Indian or Alaska Native	5 (<1)	1 (<1)
Asian	151 (9)	17 (5)
Black or African American	137 (8)	6 (2)
white	1157 (69)	269 (86)
Other/Declined/Unavailable	216 (13)	21 (7)
Ethnicity
Hispanic	229 (14)	33 (11)
Not Hispanic	1349 (81)	261 (83)
Declined/Unavailable	88 (5)	20 (6)
Health Insurance
Public	181 (11)	21 (7)
Private	830 (50)	156 (50)
Not Available	655 (39)	137 (44)

Table 2.

Psychiatric comorbidity (diagnosed prior to or within 6 months after index diagnosis) and hospital encounters (prior to or within 6 months after index diagnosis) in the matched depression and matched bipolar samples.

	A. Matched Depression Sample N = 1666	B. Matched Bipolar Sample N = 314
	Mean ± SD	Mean ± SD
Total number of comorbidities (excluding index diagnosis)	0.9 ± 0.8	1.4 ± 1.3
	N (%)	N (%)
Psychiatric Comorbidity
Nicotine use disorder	16 (1)	6 (2)
Attention deficit hyperactivity disorder	249 (15)	65 (21)
Autism Spectrum Disorder	37 (2)	19 (6)
Anxiety disorders	920 (55)	147 (47)
Obsessive compulsive disorder	139 (8)	22 (7)
Disruptive/Conduct disorders	5 (<1)	1 (<1)
Post traumatic stress disorder	95 (6)	32 (10)
Eating disorders	78 (5)	28 (9)
Hospital Utilization
Emergency Department (ED) visit: psych related	185 (11)	47 (15)
ED visit: non-psych related	524 (31)	90 (29)
Hospitalization: psych related	44 (3)	27 (9)
Hospitalization: non-psych related	179 (11)	18 (6)
Previous visit with psychiatric provider	537 (32)	139 (44)

3.1.3. Primary analysis on matched sample

Overall, we failed to find clinically or statistically significantly meaningful differences in non-nicotine SUD outcomes between groups. In the matched sample, the Cox proportional hazards model revealed no significant difference in the risk for developing a non-nicotine SUD in patients who received any treatment versus patients who received no treatment (HR=0.99, 95 % CI=0.79, 1.24; p = 0.91) (Table 3A).

Table 3.

Comparison of results from different statistical models. Outcome for all models is development of a substance use disorder. Results reported for the A) depression analysis and B) bipolar analysis.

Statistical Model	Comparator Groups	Total N	HR (95 % CI)	P-value
A. Depression Analysis
Cox Model with Sample Matched on Propensity Score	Any treatment vs. no treatment	1666	0.99 (0.79, 1.24)	0.91
Cox Model Controlling for Propensity Score	Any treatment vs. no treatment	3601	1.01 (0.84, 1.23)	0.90
Cox Model with Time-varying Covariate for Treatment and Controlling for Demographic and Clinical Covariates	Treated vs. Untreated Periods	3601	1.69 (1.45, 1.96)	<0.001
B. Bipolar Analysis
Cox Model with Sample Matched on Propensity Score	Any treatment vs. no treatment	314	0.80 (0.49, 1.30)	0.37
Cox Model Controlling for Propensity Score	Any treatment vs. no treatment	796	0.81 (0.54, 1.20)	0.29
Cox Model with Time-varying Covariate for Treatment and Controlling for Demographic and Clinical Covariates	Treated vs. Untreated Periods	796	1.08 (0.79, 1.48)	0.63

Demographic and clinical covariates controlled for in models and used to calculate propensity scores: psychiatric emergency department (ED) visits, non-psychiatric ED visits, psychiatric hospitalizations, non-psychiatric hospitalizations, prior visit with psychiatric provider, number of psychiatric comorbidities, age at index diagnosis, sex, race, median income, insurance at index diagnosis.

3.1.4. Sensitivity analyses

The Cox proportional hazards model controlling for propensity score in the full sample of 3601 patients yielded the same results as the primary analysis, showing no significant difference in the risk of developing a non-nicotine SUD in patients with depression who received any treatment versus no treatment (HR=1.01, 95 % CI=0.84, 1.23; p = 0.90) (Table 3A). Conversely, the Cox proportional hazards model with a time-varying covariate for treatment and controlling for demographic and clinical covariates showed significantly increased risk for developing a non-nicotine SUD during treated periods versus during untreated periods (HR=1.69, 95 % CI=1.45, 1.96; p < 0.001) (Table 3A).

3.1.5. Analysis of SUD Subtypes

When analyzing SUD subtypes, we examined development of AUD versus no SUD and non-AUD SUD versus no SUD. In the matched sample, the Cox proportional hazards model revealed no significant differences between patients who received any treatment versus patients who received no treatment in the risks for developing an AUD (N = 1528; HR=1.00, 95 % CI=0.74, 1.36; p = 0.98) or a non-AUD SUD (N = 1523; HR=1.00, 95 % CI=0.74, 1.35; p = 0.99). A breakdown of the non-nicotine SUD subtypes is presented in Table 4A.

Table 4.

Prevalence of alcohol use disorder, cannabis use disorder, and all other substance use disorders (SUDs).

Substance Use Disorder (SUD)	A. Patients with SUD in the Matched Depression Sample (N = 316)	B. Patients with SUD in the Matched Bipolar Sample (N = 67)
	N (%)	N (%)
Alcohol	178 (56)	34 (51)
Alcohol Only	143 (80)	28 (82)
Alcohol + Additional SUD(s)	35 (20)	6 (18)
Cannabis	127 (40)	29 (43)
Cannabis Only	87 (69)	24 (83)
Cannabis + Additional SUD(s)	40 (31)	5 (17)
Other SUD	54 (17)	11 (16)
One “Other” SUD only	41 (76)	7 (64)
2 + “Other” SUDs or “Other” SUD(s) + Alcohol and Cannabis	13 (24)	4 (36)

3.1.6. Treatment duration and development of a non-nicotine SUD

A Cox proportional hazards model revealed a significant relationship between treatment duration and development of a non-nicotine SUD. Controlling for baseline characteristics, the risk of developing a non-nicotine SUD decreases by 2.1 % for each additional cumulative month of treatment for MDD (HR=0.979, 95 % CI=0.976, 0.981, p < 0.001).

3.1.7. Predicting development of a non-nicotine SUD

No significant sociodemographic or clinical characteristics predicted the development of non-nicotine SUD among patients with depression. Models in the training and test sets resulted in AUC scores of 0.61 and 0.60, respectively, demonstrating that the predictive value of these variables was weak. Additional details of the analysis can be found in Supplemental Materials.

3.2. Bipolar disorder analysis

3.2.1. Sample derivation

A total of 796 patients with BPD were included in our analyses (Fig. 1B). These patients were followed within the healthcare system for an average of 1216.3 ± 871.6 days. Stratification by BPD treatment status identified 159 untreated patients and 637 treated patients. One-to-one propensity score matching narrowed our sample down to 314 patients, including 157 untreated patients and 157 treated patients. There were two untreated patients who did not have matches, resulting in 159 total untreated patients but only 157 in the matched sample. Twenty-three percent (n = 180/796) of patients in the full sample and 21 % (n = 67/314) of patients in the matched sample developed a non-nicotine SUD. Among the 637 treated patients in the full sample, 92 % (n = 588) received pharmacologic therapy only, 0.6 % (n = 4) received psychosocial therapy only, and 7 % (n = 45) received both pharmacologic and psychosocial therapy. Among the 157 treated patients in the matched sample, 94 % (n = 148) received pharmacologic therapy only, 0 % (n = 0) received psychosocial therapy only, and 6 % (n = 9) received both pharmacologic and psychosocial therapy.

3.2.2. Matched sample characteristics

In the matched sample, patients had a mean age of 23.6 years, were 66 % female, and 86 % white (Table 1B). The average median income was $103,027.90, and 50 % of patients had private insurance. The most common psychiatric comorbidity was anxiety disorders (47 %) and the least common was disruptive/conduct disorders (<1 %) (Table 2B). ED visits for non-psychiatric reasons were more common than ED visits for psychiatric reasons (29 % and 15 %, respectively), while hospitalizations for psychiatric reasons were slightly more common than hospitalizations for non-psychiatric reasons (9 % and 6 %, respectively). Forty-four percent of patients had a previous visit with a psychiatric provider prior to the date of their BPD diagnosis. Sociodemographic and clinical characteristics for the whole sample can be found in the Supplementary Materials (Tables S2 and S3).

3.2.3. Primary analysis on matched sample

In the matched sample, the Cox proportional hazards model revealed a clinically significant qualitative reduction in risk but no statistically significant difference in the risk for developing a non-nicotine SUD in patients with BPD who received any treatment versus patients who received no treatment (HR=0.80, 95 % CI=0.49, 1.30; p = 0.37) (Table 3B).

3.2.4. Sensitivity analyses

The Cox proportional hazards model controlling for propensity score in the full sample of 796 patients yielded the same results as the primary analysis, showing a notable clinical trend but a statistically nonsignificant differences in the risk of developing a non-nicotine SUD in patients with BPD who received any treatment versus no treatment (HR=0.81, 95 % CI=0.54, 1.20; p = 0.29) (Table 3B). Similarly, the Cox proportional hazards model with a time-varying covariate for treatment and controlling for demographic and clinical covariates showed no significant difference in the risk for developing a non-nicotine SUD during treated periods versus during untreated periods (HR=1.08, 95 % CI=0.79, 1.48; p = 0.63) (Table 3B).

3.2.5. Analysis of SUD subtypes

As in the MDD analysis, when analyzing SUD subtypes, we examined development of AUD versus no SUD and non-AUD SUD versus no SUD. In the matched sample, the Cox proportional hazards model revealed no significant differences between patients who received any treatment versus patients who received no treatment in the risks for developing an AUD (N = 281; HR=1.21, 95 % CI=0.59, 2.50; p = 0.61) or a non-AUD SUD (N = 286; HR=0.65, 95 % CI=0.34, 1.22; p = 0.18). A breakdown of the non-nicotine SUD subtypes is presented in Table 4B.

3.2.6. Treatment duration and development of a non-nicotine SUD

Similar to findings with depression, a Cox proportional hazards model revealed a significant relationship between treatment duration and development of a non-nicotine SUD. Controlling for baseline characteristics, the risk of developing a non-nicotine SUD decreases by 2.6 % for each additional cumulative month of treatment for BPD (HR=0.974, 95 % CI=0.968, 0.979, p < 0.001).

3.2.7. Predicting development of non-nicotine SUD

There were no significant sociodemographic or clinical predictors of development of a non-nicotine SUD in patients with BPD. Models in both the training and test sets yielded AUC statistics of 0.56 and 0.53, respectively, indicating that the ability of the included characteristics to predict development of a non-nicotine SUD was weak and barely better than chance. Specific details of the analysis can be found in the Supplemental Materials.

4. Discussion

The results of our analysis of EHR-derived naturalistic data partially supported our hypothesis. Counter to our hypothesis we found that treatment overall of MDD or BPD in young people aged 16–30 years old did not mitigate the risk for the subsequent development of SUD. Despite using quasi-experimental methods with propensity matching, our results failed to show that in and of themselves medications, psychotherapy, or a combination of the two were protective against SUD. However, when we examined duration of treatment, we found longer duration of treatment was associated with a clinically and statistically significant reduced risk for SUD in both MDD and BPD groups.

Our finding that MDD treatment for young people did not decrease risk for SUD is consistent with one study in the literature. Research in younger populations have indicated that MDD increases the risk for later alcohol use disorder [23], without significant attenuation by treatment [15]. Specifically, Curry and colleagues found a positive response to MDD treatment over 12 weeks (psychotherapy, pharmacotherapy, or the combination) in adolescents aged 17–23 did not impact the likelihood of developing an AUD, but was associated with a decreased risk of drug use disorder over a follow-up period of 63 months [15]. Although Carra and colleagues only examined non-medical use of prescription pain medication and did not examine for a drug use disorder, they also found young people who did not report treatment for depression were at increased risk. Neither study examined the effect of the duration of depression treatment. Given mixed findings regarding the impact of treatment of MDD in young people on later SUD risk additional research is needed that further examines the timing and duration of treatment for MDD in young people as well as risk associated with different types of SUD.

For BPD, we found a notable clinically significant but statistically nonsignificant reduction in SUD in young people with treated BPD, and a reduction in risk for SUD associated with longer duration of treatment. Our findings are reminiscent of Goldstein et al. who, in a controlled longitudinal study of adolescents and young adults with BPD, showed some reduced risk of subsequent SUD associated with lithium, but not other medications used for BPD [17]. In fact, higher rates of SUD were associated with other mood stabilizing compounds.

Our findings highlight the importance of treatment duration for MDD and BPD and are consistent with research on the impact of treatment of attention deficit hyperactivity disorder (ADHD). Longer duration of ADHD treatment for children and adolescents is associated with decreased risk for SUD in young people [24], [25]. This research has also found that earlier initiation of ADHD treatment is associated with decreased risk for SUD [24], [25] and raises the question of timing of treatment for MDD and BPD.

When MDD and BPD treatment overall is examined it may be that it needs to be initiated earlier to decrease risk for later substance use. While our sample consisted of young people aged 16–30 years, previous reviews emphasize the benefits of even earlier intervention to decrease substance use risk for youth in general, particularly among adolescents aged 13–19 [26]. Consistent with this, Wittchen et al. found that prevention efforts targeting substance use, especially for nicotine and alcohol, are most effective if implemented before age 14 years [27]. Additionally, studies indicate that the adolescent-onset of BPD (in individuals ages 13 and over) is associated with a higher risk of SUD compared to childhood-onset BPD [28]. This corroborates findings by Rao et al. showing a heightened risk of new-onset SUD and recurrence during the critical transition to adulthood [29]. These findings demonstrate that earlier interventions in mood disorders may be necessary for SUD reduction, and our sample age range of 16–30 years may have been beyond the optimal window for achieving significant improvements in SUD outcomes. Lastly, these aggregate data also indicate that with shorter duration- and presumed later onset treatment, ancillary measures directly addressing substance use and SUD such as ongoing and systematic screening and brief referral to treatment [30], [31] are necessary, particularly given the highly elevated risk of SUD in mood disorders in young people.

Our findings are limited by substantial methodological limitations. Our sample was derived from naturalistic treatment data harnessed from coded fields within the EHR, making it difficult to manage the confound of treatment versus severity of illness predicting SUD. For example, our sensitivity analyses revealed a statistically significant association between treatment for depression and increased SUD risk. To better account for severity of illness, we utilized propensity score matching. However, by only using coded fields within the EHR we were not able to account for other demographic and clinical variables that influence SUD onset including for example educational level, employment, and family history of SUD. Our sample predominantly had private insurance and higher-than-average incomes compared to other areas of the United States, limiting generalizability to different socioeconomic groups. Additionally, the sample was not evenly distributed by sex and race, with a majority consisting of white female individuals. Although our proxy diagnoses for BPD, MDD, and SUD were previously validated, these proxy diagnoses are not as robust as structured interviews. Further, although our sample size was relatively large, it was substantially reduced when matching based on propensity scores, limiting our power to detect smaller differences. Lastly, as our sample was derived from the EHR of a large healthcare network, we were unable to verify whether members of our sample were engaging in therapy outside of our system, impacting the distribution of treatment modalities within our sample. Likewise, we were also unable to verify adherence to pharmacologic treatment with data available in the EHR record. However, the average duration of treatment for individuals on pharmacologic treatment was 23.3 months for those with MDD, and 24.2 months for those with BPD, indicating refills of medication beyond the one month threshold used to define “treated.” It will be important for future research to also examine treatment modality and intensity when examining the impact of treatment on SUD risk in young people with mood disorders.

Despite these limitations, our findings suggest that although treatment of MDD and BPD in young people aged 16–30 years in and of itself may be insufficient in mitigating subsequent risk of SUD, longer duration of treatment for both disorders is associated with decreased risk. Additional studies examining treatment duration as well as the age of onset of treatment initiation and other preventive interventions for SUD are necessary in this high-risk mood disordered group.

Funding

This research was supported by the National Institute on Drug Abuse [UH3DA050252-02, 2020].

Declaration of Competing Interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Timothy E. Wilens discloses that he has research funding from the NIH and from Ironshore for investigator initiated trials. Additionally, he receives royalties from intellectual property agreements to Ironshore, Guilford Press, and Cambridge University Press. He is also a consultant to the Gavin Foundation, Bay Cove Human Services, US National Football League (ERM Associates), US Minor/Major League Baseball, and White Rhino/3D. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

The following authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper (Amy M. Yule, Ann L. Kim, Katy Burns, Maura DiSalvo, Vinod Rao, Mira Stone, and Sylvia Lanni).

Appendix A. Supplementary material

Supplementary material