Trajectories of depression among patients in treatment for opioid use disorder: A growth mixture model secondary analysis of the XBOT trial

Abstract

Background and Objectives:

To inform clinical practice we identified subgroups of adults based on levels of depression symptomatology over time during opioid use disorder (OUD) treatment.

Methods:

Participants were 474 adults in a 24-week treatment trial for OUD. Depression symptoms were measured using the 17-item Hamilton Depression Rating Scale (HAM-D) at nine time points. This was a secondary analyses of the Clinical Trials Network Extended-Release Naltrexone vs. Buprenorphine for Opioid Treatment (XBOT) trial using a growth mixture model.

Results:

Three distinct depression trajectories were identified: Class 1 High Recurring - 10% with high HAM-D with initial partial reductions (of HAM-D across time), Class 2 Persistently High- 5% with persistently high HAM-D, and Class 3 Low Declining - 85% of the participants, with low HAM-D with early sustained reductions. The majority (low declining) had levels of depression that improved in the first 4 weeks and then stabilized across the treatment period. In contrast, 15% (high recurring and persistently high) had high initial levels that were more variable across time. The persistently high class had higher rates of opioid relapse.

Discussion and Conclusion:

In this OUD sample, most depressive symptomatology was mild, and improved after medication treatment for opioid use disorder (MOUD). Smaller subgroups had higher depressive symptoms that persisted or recurred after initiation of MOUD. Depressive symptoms should be followed in patients initiating treatment for OUD, and when persistent should prompt further evaluation and consideration of antidepressant treatment.

Scientific Significance:

This study is the first to identify three distinct depression trajectories among a large clinical sample of individuals in MOUD treatment.

Introduction

Recent increases in drug overdose deaths attributable to opioids ¹ has strengthened calls to improve treatment approaches for opioid use disorder (OUD). Efforts to prioritize evaluation of secondary outcomes in OUD trials² have been identified as an important next step for the field so that treatments may be better personalized ³. Depression is one such secondary outcome which warrants special attention among individuals with OUD ⁴ because it is common in patients entering treatment ^5,6, related to higher healthcare costs ⁷, and has been linked to increased rates of overdose ⁸. Depressive disorders co-occur in 15–25% of individuals with OUD and 27–61% of people who misuse opioids ^9–11. Patients with co-occurring OUD and major depressive disorder (MDD) tend to have more difficulty in treatment and incur less favorable outcomes than patients without depression ¹², and there is a bi-directional relationship in that pre-existing OUD increases the risk of developing MDD and MDD is a vulnerability to subsequent OUD ¹³. Although co-occurring depression is a well-known vulnerability factor for individuals with OUD that influences the clinical approach, there are significant knowledge gaps in the trajectories of depression symptoms over time for individuals receiving treatment for OUD. Identifying patterns of depression trajectories during OUD treatment may provide evidence for targeted interventions among specific subgroups of patients.

In general, baseline depression symptoms among most patients with substance use disorder (SUD) improve with SUD treatment ^14,15. Similarly, mean depression symptoms in OUD improve during MOUD treatment ^12,16 including methadone ¹⁷, buprenorphine ^18,19, and naltrexone ¹⁹. Findings on the relationship between depression symptoms at baseline and OUD outcomes have been mixed, with some studies suggesting worse prognosis ²⁰ and other studies better prognosis ²¹. Other studies have attempted to identify depressive disorders (in OUD populations) as a target for antidepressant treatment with positive results ^22,23. Fewer studies have examined patterns of depression over time in patients with OUD treated with medication²⁴. Several studies show high rates of placebo response in patients initiating methadone selected for high depression scores ²⁵, highlighting the difficulty in distinguishing patients whose depression will persist from baseline symptoms alone, and also illustrating that much depressive symptomatology resolves with MOUD. However, mean depression scores give an incomplete picture, missing clinically important heterogeneity. What characteristics distinguish patients whose depression does not improve with MOUD? Are there different patterns of depressive symptoms over time that might guide treatment? How do depression outcomes relate to OUD outcomes?

In a latent class analysis of pain and depression trajectories using data from the Prescription Opioid Addiction Treatment Study (POATS; CTN-0030), researchers found that during a 12-week stabilization on buprenorphine, depression symptoms decreased rapidly over the first four weeks and then leveled off for 71% of the patient sample ²⁶. However, a vulnerable patient group (20% of the sample) had extremely high levels of depression symptoms that remained persistent over the 12-week stabilization period. A previous secondary analysis of the XBOT study (CTN-0051) (Na, 2021) examined the Hamilton Depression Rating Scale (HAM-D) across the first 4 weeks of treatment with buprenorphine or extended-release naltrexone. Two thirds of patients with at least mild depression (HAM-D > 7) improved by week 4. Those with moderate/severe depression at baseline (8 ≤ HAM-D ≤ 16) were less likely to remit (HAM-D ≤ 7) by week 4 (52.8%) compared to those with mild depression (76%) (OR=0.43, 95% CI=[0.21–0.89], p=0.02), and failure to remit trended toward an association with greater risk of relapse to opioids compared to those who remitted (OR=1.82, 95% CI=[0.93–3.57], p=0.08). Another secondary analysis of XBOT ²⁷ examined the course of mean HAM-D scores over the entire 36 weeks of follow-up, finding steady decreases from baseline though week 12, then modest increases through the end of treatment at week 24 and post-treatment follow-up at week 36. Mean HAM-D was associated with self-report lifetime history of depression and anxiety.

The present project extends previous work by using growth mixture modeling (GMM) to identify distinct subgroups of patients in OUD treatment who exhibit different trajectories of depression symptoms over time. We hypothesized that there would be varying patterns (trajectories) of depression symptoms, with the majority group(s) of participants either having low and/or declining depression symptoms, but an important smaller sub-group(s) having high and/or persistent depression symptoms over time. We also hypothesized that membership in a higher severity, vulnerable class(es) would be associated with characteristics typical of depression disorders, as well as worse opioid-related outcomes.

Method

Participants

We performed a secondary data analysis of the National Institute on Drug Abuse Clinical Trials Network 0051 study, commonly known as the XBOT trial ²⁸. The purpose of the parent study was to compare the effectiveness of two types of pharmacological treatment for OUD, daily buprenorphine/naloxone (BUP-NX) and monthly injectable extended-release naltrexone (XR-NTX) in a 24-week randomized treatment trial. Although both medications had previously demonstrated efficacy for OUD, no study had conducted a head-to-head effectiveness trial. Briefly, participants were ≧18 years of age and seeking treatment for OUD from one of eight sites across the US from 2014–2016. The parent trial recruited and randomized 570 subjects, but this secondary analysis included only individuals who started their assigned medication (N=474). The analysis was limited to this per protocol sample because we were most interested in depressive symptom trajectories of individuals who were actually treated with medication for OUD (and therefore remained in the trial long enough to collect follow up data on depressive symptoms).

Measures

The HAM-D 17-item measure was used to assess participant depression symptoms at baseline and nine time points during the study intervention ²⁹. The HAM-D is a widely used measure of depression that has been validated and is reliable in clinical samples ³⁰. The HAM-D has clinical cut off scores to indicate depression symptomatology that rises to the threshold of a psychiatric disorder, but for this study, data were analyzed dimensionally using a total symptom score. Depression scores did not differ significantly between study treatment arms in the parent study ²⁸. Self-reported substance use was collected with the Timeline Follow Back ³¹. Urine toxicology was done on weekly urine samples that were tested for opioids. Opioid relapse was operationalized as four or more consecutive weeks of any non-study opioid use (by urine toxicology, or self-report, or failure to provide a urine sample) or seven or more consecutive days of self-reported non-study opioid use²⁸. Consistent with the parent study, relapse was determined only after week 3 to account for the lingering presence in urine tests of opioid medications used for detoxification and that participants were considered likely to “test” medication effects with sporadic opioid use early in treatment.

Baseline characteristics including demographics, substance use, mental health, functional status, and other clinical measures of interest were selected as potential factors that might profile or differentiate the various classes established by the GMM analysis. The EuroQol administered at baseline was used to examine concurrent quality of life indicators ³². The measure includes five subscales higher scores relating to more severe levels of difficulty with mobility, self-care, usual activities, pain, and mental health. Baseline characteristics and other outcomes included in the results were pre-identified based on clinical experience and previous literature. Characteristics not reported in the results include other psychiatric history variables, additional substance use variables, medication preference, living with other persons with SUD, and justice involvement. Details of these measures have been previously reported in the primary XBOT paper ²⁸.

Statistical analyses

The group based finite mixture analyses (i.e., GMM) were computed using MPlus version 8. The purpose of the analyses was to distinguish specific subgroups based on different responding patterns (trajectories) of Hamilton depression score over time after initiating treatment. In the GMM we modeled Hamilton depression scores across the 24 weeks of treatment, including intercept, slope, and quadratic parameters as the latent group indicators for the depression trajectories. We implemented model building techniques outlined in Nagin (2005) ³³. Typically, model selection is determined by a combination of factors including global fit indices, clinical relevance of the groups (i.e., substantial worsening or improvement across time), and group size ³⁴. Informing our model selection was the AIC, BIC, entropy, and the LMR-LRT goodness-of-fit indices. AIC and BIC scores that are lower indicate better fit. Entropy scores range between 0 and 1 with scores closer to 1 indicating better fit. Models with entropy scores above .80 are acceptable. Adjusted LMR-LRT scores indicate that the chosen model fits the data better than a model with one fewer (k-1) classes. Missing data were handled using a maximum likelihood estimator with robust standard errors. Posterior probability of class membership was produced for all participants in the data set, and individuals were grouped into the class for which they had the greatest posterior probability. Group differences based on posterior probabilities of most likely class membership were computed using ANOVA for continuous measures, chi-square for categorical variables, and log-rank test (with contrast) for opioid relapse-free survival. Type I error was set at 5% for all analyses. As this was an exploratory analysis, no adjustment was made for multiple comparisons. All post-GMM analyses were completed with SAS^® version 9.4.

Results

Model fit

Goodness-of-fit indices for models of HAM-D trajectories with 1–4 classes we computed and considered (see Table 1). Though up to 5 classes were deliberated, the 5-class solution was discarded due to no participants being assigned membership to one of the classes, based on posterior probability. Considering model fit indices, class sizes, and clinical relevance, the 3-class solution was deemed the most parsimonious. Characteristics of the 3-class solution were a BIC of 19587, entropy of .85, and LMR-LRT approaching significance at p = .062. More than 85% of subjects had Class 3 for most likely membership. Most subjects had clear membership delineation (>90% probability). Fifty-two subjects had <80% probability for their most likely class.

Table 1.

Growth mixture model fit indices and most likely class membership sizes.

Model	AIC	BIC	Δ BIC	Class Size	Entropy	LMR-LRT
1 Class	19708	19788	---	100%
2 Class	19569	19665	123	83%, 17%	.81	p=0.071
3 Class	19474	19587	78	85%, 10%, 5%	.85	p=0.062
4 Class	19422	19551	36	82%. 11%%, 5%, 3%	.85	p=0.172

Note: BIC = Bayesian Information Criterion, AIC = Akaike Information Criterion, LMR-LRT = Lo-Mendel-Rubin Likelihood Ratio Test of Significance, Par. = Parameters in model. Bold typeface indicates model chosen for best overall fit and interpretability.

HAM-D trajectories of the classes

Figure 1 displays mean HAM-D score trajectories across 24 weeks, and proportional membership for each of the classes in the 3-class solution.

Figure 1.

Mean levels of depression (HAM-D) by class across the 24-week treatment period. Note: Class 1 – Improved then Recurring (n=45, 10%), Class 2 – Persistently High (n=25, 5%), and Class 3 – Low Declining (n=404, 85%).

Table 2 summarizes the mean HAM-D scores and depression severity categorization for the three classes at baseline, at week 4, and averaged across time. There were no significant differences in class assignment between the BUP-NX and the XR-NTX treatment arms.

Table 2.

HAM-D scores for the 3 classes at baseline, at week 4, and averaged across time (including baseline).

	Class 1	Class 2	Class 3
Baseline
Mean HDRS (SD)	13.44 (7.31)	14.40 (8.97)	8.20 (5.98)
0–7	28.9%	28.0%	54.8%
8–16	37.8%	44.0%	34.0%
>17	33.3%	28.0%	11.2%
Week 4
Mean HDRS (SD)	7.03 (4.95)	16.33 (8.34)	3.83 (3.89)
0–7	62.5%	14.3%	86.3%
8–16	35.0%	42.9%	12.5%
>17	2.5%	42.9%	1.3%
Avg across 24 weeks
Mean HDRS (SD)	10.91 (4.29)	14.78 (5.74)	4.97 (4.00)
0–7	24.4%	4.0%	77.7%
8–16	66.7%	72.0%	20.3%
>17	8.9%	24.0%	2.0%

Class 1 – Improved then Recurring, comprising 10% of the sample, was characterized by higher initial HAM-D scores, with baseline mean HAM-D =13.4, and 70% in the moderate / severe range. This class also showed a rapid decline in HAM-D scores, with week 4 mean HAM-D almost halved to 7.0, and 38% in the moderate / severe (HAM-D≧8) range. But the declining trajectory of this class had a variable course, stabilizing at low levels temporarily, then increasing slowly again from week 12 on, back to baseline levels. Mean HAM-D across 24 weeks was 10.9. We have summarized depression symptoms in this class as “temporarily improved, then recurring.”

Class 2 -Persistently High, comprising 5% of the sample, was characterized by the highest initial HAM-D scores, with baseline mean HAM-D =14.4, and 72% in the moderate / severe range. This class had an early increase in scores, with week 4 mean HAM-D 16.3, and 86% in the moderate / severe (HAM-D≧8) range. This class also had notable persistence of high scores for most of the study, despite a partial decline from weeks 16–20. Mean HAM-D across 24 weeks was 14.8, with 96% in the moderate/severe range. We have summarized depression symptoms in this class as “persistently high.”

Class 3 – Low Declining, comprising 85% of the sample, was characterized by low initial HAM-D scores with baseline mean HAM-D= 8.2, and 45% falling in the moderate / severe (HAM-D≧8) range. This class also showed rapid and steady decline in HAM-D scores over the course of the study period, with week 4 mean HAM-D halved to 3.8, and only 14% in the moderate / severe (HAM-D≧8) range. Mean HAM-D across 24 weeks was 5.0. We have summarized depression symptoms in this class as “initially low and declining.”

Baseline characteristics of the classes

Table 3 provides baseline descriptors for each of the classes, including demographics and clinical characteristics.

Table 3.

Demographics and concurrent outcomes of class membership.

	Overall (N=474)		Class 1 (N=45, 10%)		Class 2 (N=25, 5%)		Class 3 (N=404, 85%)
Variable	N	% or M (SD)	N	% or M (SD)	N	% or M (SD)	N	% or M (SD)	Prob
Sex									0.4635
Male	331	69.8%	30	66.7%	15	60.0%	286	70.8%
Female	143	30.2%	15	33.3%	10	40.0%	118	29.2%
AgeatRandomization - Mean(SD)	474	33.7 (9.6)	45	33.4 (9.2)	25	35.6 (10.9)	404	33.6 (9.6)	0.5933
Categorized Age									0.6312
25 or Younger	98	20.7%	7	15.6%	6	24.0%	85	21.0%
Older than 25	376	79.2%	38	84.4%	19	76.0%	319	79.0%
Hispanic ethnicity									0.8389
Yes	80	16.9%	9	20.0%	4	16.0%	67	16.6%
Race									0.4835
Caucasian	358	75.5%	31	68.9%	20	80.0%	307	76.0%
African American	47	9.9%	4	8.9%	1	4.0%	42	10.4%
Other races	69	14.6%	10	22.2%	4	16.0%	55	13.6%
Education level									0.7014
<HS	114	24.1%	13	28.9%	8	32.0%	93	23.0%
HS/GED	156	32.9%	12	26.7%	7	28.0%	137	33.9%
>HS	204	43.0%	20	44.4%	10	40.0%	174	43.1%
Employed									0.5479
Yes	177	37.3%	20	44.4%	10	40.0%	147	36.4%
Homeless									0.7272
Yes	125	26.4%	13	28.9%	8	32.0%	104	25.7%
Majordepressivedisorderhistory									0.0057 b
Yes	153	32.2%	23	51.1%	11	44.0%	119	29.5%
Anxietydisorderhistory									0.0008 b c
Yes	220	46.5%	30	66.7%	17	68.0%	173	42.8%
Psychoticepisodeshist									0.0002 b
Yes	13	2.7%	5	11.1%	2	8.0%	6	1.5%
Sexualabuse(lifetime)									0.0082 b c
Yes	133	28.2%	19	42.2%	13	52.0%	101	25.0%
EuroQoL Mobility
none	416	87.8%	35	77.8%	19	76.0%	362	89.6%	0.0131 b
mild-moderate	58	12.2%	10	22.2%	6	24.0%	42	10.4%
Self-care
none	462	97.5%	43	95.6%	23	92.0%	396	98.0%	0.1228
mild-moderate	12	2.5%	2	4.4%	2	8.0%	8	2.0%
Usualactivities
none	395	83.3%	32	71.1%	18	72.0%	345	85.4%	0.0151 b
mild-moderate	79	16.7%	12	28.9%	6	28.0%	55	14.6%
Pain/discomfort
none	197	41.6%	14	31.1%	4	16.0%	179	44.3%	0.0067 c
mild-moderate	277	58.4%	31	68.9%	21	84.0%	235	55.7%
Anxiety/depression
none	145	30.6%	7	15.6%	7	28.0%	131	32.4%	0.0636
mild-moderate	329	69.4%	38	84.4%	18	72.0%	279	67.6%

Note: Prob = Chi-square or analysis of variance differences test p-value. Bolded values are significant at the .05 level. Pairwise group comparisons:

^b=significant difference between Classes 1 & 3

^c=significant difference between Classes 2 & 3

There were no demographic differences by class. Reports of lifetime sexual abuse at baseline, reported history of major depressive disorder diagnosis, reported history of prior anxiety disorder diagnosis, and reported history of prior psychotic episode were significantly different between classes. Lastly, there were significant differences indicated on levels of pain, ability to perform usual activities, and mobility from the EuroQol.

Differences in class relapse rates

Figure 2 shows an opioid relapse-free survival curve (with contrast) by HAM-D class. There was a significant difference (p=.0007) in time to relapse between the classes using a log-rank test, with the persistently high class time to relapse (median = 42 days) significantly shorter compared to both the improved and recurring class (median = 142 days, p=.0083) and the low declining class (median = 136 days, p=.0002) in pair-wise comparisons. Time to relapse was not significantly different between the low and declining class and the improved and recuring class (p=.3900) in a pair-wise comparison.

Figure 2.

Relapse free survival by class. Note: Significant difference in time-to-relapse between GMM classes (p=0.0007)

Discussion

In line with our stated hypotheses, we identified three trajectory classes, each with unique patterns of depression symptoms across time. While most patients’ depression scores improved with MOUD treatment, we identified a small, but key group with persistently high depressive symptoms. Though it is difficult to generalize beyond this specific sample, the results of our study suggest that individuals who report high depressive symptoms lasting beyond the first several weeks of MOUD treatment are at heightened risk for poor outcomes including opioid relapse and persistently high depressive symptoms for at least six months.

The majority of participants had relatively low and persistently declining depression symptoms from baseline and throughout the study (low declining class). And among those with higher levels of baseline depression symptoms, some (improved then recurring) had a more variable course with substantial decline followed by increase, while a minority had persistent elevations (the persistently high class). This is consistent with previous findings from the GMM secondary analysis of CTN-0030 among patients with OUD specific to prescription opioid use ²⁶, which found persistent declines in depression symptoms in the majority trajectory class over 12 weeks of buprenorphine treatment, but persistently high levels in a minority (20%) class. Similarly, a recent national study of (N=3016) patients entering OUD treatment with a positive screen for depression found that a three-class model was most parsimonious³⁵. Likewise, the classes followed a similar pattern of depression trajectories across 4 weeks of OUD treatment to our trajectories across 24 weeks. However, class size was smaller for the low declining class in that study, though this may be expected considering that everyone in the study screened positive for depression. It is also consistent with the previous secondary analysis of the XBOT study (Na et al., 2021) which found that of the half of participants with either mild or moderate depression, 2/3 had either remitted (HAM-D≤7) or responded (HAM-D reduced ≥50%) at week 4, leaving 1/6 who had mild or moderate depression and had not remitted or responded. It is also consistent with the preponderance of evidence that shows treatment with MOUD is associated with improved depression for many patients ^{12,16,18,19,25}.

While very common, depression symptoms can be difficult to interpret in early treatment, and clinicians face a significant challenge in determining the cause of elevated depression symptoms seen at the outset of OUD treatment. Such elevations may be due to life circumstances associated with OUD (e.g., demoralization, ruptured relationships, employment problems), the physiological effects of habitual opioid use or opioid withdrawal, or an independently co-occurring psychiatric illness such as MDD. Indeed, in this study, about half of the participants reported at least moderate depression scores at the outset of OUD treatment.

Furthermore, subsequent improvements may be due to myriad factors including the beneficial effects of stopping or substantially reducing opioid use, crisis stabilization, the installation of hope, therapeutic alliance, fostering a social support network, case management, psychosocial interventions that may have included some non-specific components that addressed depression symptoms, or spontaneous remission. It is important to note that the HAM-D may also be picking up non-specific symptoms related to chronic opioid use and to withdrawal (such as protracted insomnia, reduced appetite, fatigue) that often resolve over several weeks ³⁶. However, for some OUD patients, depression symptoms do not remit naturally over time with MOUD treatment. We found that such participants had higher rates of various baseline characteristics that might be expected to correlate with more severe depression and the presence of a co-occurring syndromal depressive disorder rather than what has been conceptualized as substance-induced depression, such as prior history of diagnosis and/or treatment for depression or other psychiatric disorders, current problems with depression/anxiety, history of sexual abuse, etc..

We also found that those in the higher depression classes had higher rates of opioid relapse. The majority class, with low and declining HAM-D had normative rates of relapse (51.2%). The persistently high class members with persistently high HAM-D had much greater rates of relapse (84.0%), and the improved then recurring class members also had higher rates of relapse (62.2%) although this difference did not reach statistical significance. This exploratory finding of an association between high and persistent (or recurring) depression symptoms and relapse is consistent with the findings of Na et al which found a trend towards greater relapse in those who had moderate / severe depression symptoms at baseline, and then did not show depression remission at week 4.

These results highlight a subgroup (persistently high class), which although representing a small minority of participants (5%), have persistently high HAM-D scores despite MOUD treatment, and presence of various indicators and correlates of depressive disorders. Membership in this class seems to be associated with a very malignant course of OUD, and poor responsiveness/adherence to standard MOUD treatment ¹⁷, with only 16% achieving non-relapse survival at 24 weeks. The high recurring class also seemed to have vulnerability to depression and trended toward higher rates of relapse, intermediate between the other 2 classes. It’s also notable that this class showed a steepening of the decline in non-relapse survival at around week 20, during the period of recurrent HAM-D increases from week 12.

Our study results offer some potentially helpful clues on how clinicians should identify those OUD patients with high depression risk. Those with higher initial depression scores were more likely to fall into classes 1 or 2. But this was a crude predictor, as 2/3 of those with baseline moderate/ severe depressive scores had later response or remission (see Na et al., 2021). More helpful is an assessment of depression symptoms after several weeks (weeks 3–6) of MOUD. Those who remain in the severe range are more likely to be members of the worst prognosis class. Visual inspection of Fig 1 suggests that sustained depression remission, with HAM-D<7 and not just relative reduction from baseline (response), should be the target. Serial assessment over time rather than any one cross-sectional assessment is more likely to be helpful in determining trajectory. Presence of other clinical indicators of major depression may be suggestive of vulnerability as well.

There has been little recent direct research exploration of specific approaches to the treatment of major depression in OUD. Nevertheless, the strategy of augmenting MOUD with antidepressant medication in patients with persistent depression would certainly be supported by broad clinical consensus (The ASAM National Practice Guideline for the Treatment of Opioid Use Disorder: 2020 Focused Update). Additional interventions could include modifications to increase the effectiveness of the OUD treatment, such as more frequent monitoring, intensification of psychosocial treatment, exercise (Murri et al, 2019)³⁸, medication dose adjustments, medication adherence enhancements, assertive outreach, and involvement of family supports.

The study has limitations. First, we examined only participants from the XBOT trial that had successfully inducted on to the study medications (per protocol). This strategy was chosen because we were particularly interested in the course of depression with MOUD treatment. Additionally, most of those who did not initiate MOUD relapsed early and were lost to follow up, without HAM-D data. Second, there is no gold standard for model selection in latent class and growth mixture modeling. As such, our selection of the 3-class model may not be identified as the most parsimonious by other researchers. However, acceptable entropy, class sizes, and clinical relevance, strengthened our 3-class model interpretation. The parent study did not include clinical assessments or structured interviews that might have helped identify and differentiate independent depressive disorders. Third, the HAM-D may be picking up parallel depression-like markers including indicators associated with subacute opioid withdrawal. Fourth, there may have been key baseline characteristics not included in the parent study which may have predicted class membership in a meaningful manner. A final limitation of this study is that, although depression symptoms were not a target of the study, and very few participants received additional treatment for depression a small minority may have been receiving treatment for depression at various points during the study, which limits our ability to draw conclusions about the specific effect of treatment with MOUD on depression symptoms. It is difficult to know how the results of this study may generalize to individuals seeking non-pharmacological treatment for OUD or treatment for other SUD. We speculate that there may be a similar persistently high severity cohort for these populations as well, and that the clinical guidance to repeatedly assess depressive symptoms in early recovery to determine likely trajectory may generalize to other groups.

In summary, a major strength of this study is the use of GMM to examine trajectories of depression across time during treatment as opposed to characterizing levels of depression at a specific time point (traditional latent class analysis). For clinicians, our findings drive home the importance of providing depression screening and assessment in OUD treatment to identify individuals with persistent depression and introduce early intervention to reduce the risk of relapse. Furthermore, the identification of distinct OUD patient subgroups based on depression symptoms may provide important information about expected trajectories of both vulnerabilities to co-occurring psychiatric conditions and their relation to OUD treatment outcomes that would inform intervention targets. While the majority of individuals receiving MOUD treatment have low and/or persistently declining depression symptoms, small but critical subgroups identified in our latent class analysis have different, more persistent depression symptom trajectories (either persistently high or improving then recurring). Persistently high depression is associated with higher opioid relapse rates. MOUD treatment is the preferred and sufficient strategy for the majority of OUD patient with depressive symptoms but patients in these poorer-prognosis subgroups may have particular vulnerability to depression, may have major depressive disorders, and should be targeted for additional tailored interventions including closer monitoring, augmentation of standard MOUD treatment, and antidepressant medication treatment. Future research should include deliberate examination of OUD populations with co-occurring depression, including trials of specific combined treatment approaches.

1. DATA AVAILABILITY STATEMENT

Supplementary materials that display the conceptual model of the statistical analysis for the growth mixture model and the model fit indices for each class are available upon request from the corresponding author.