Contingency management for stimulant use disorder and association with mortality: a cohort study

Abstract

Objective.

While opioid overdose has begun to decrease, stimulant overdose has continued to increase in recent years and has not been adequately addressed. Unlike opioid use disorder, there are no FDA-approved medications to treat stimulant use disorder (StUD). The most effective treatment is contingency management (CM), a behavioral intervention that provides tangible rewards to reinforce target behaviors, such as biochemically-verified abstinence. Despite the effectiveness of CM on near-term substance use behaviors, the long-term impacts on key outcomes such as mortality are unclear. The objective of this work is to examine whether patients with StUD who receive CM have a decreased risk of mortality.

Methods.

This is a retrospective cohort study of those with StUD who received or did not receive CM using linked electronic health records and death records in the largest integrated health system in the U.S., the Veterans Health Administration (VHA), from July 2018 through December 2020. The primary outcome is mortality in the year following the index CM visit. All-cause mortality data was obtained from the National Death Index and linked to electronic health record data. Adjusted hazard ratios (AHR) were estimated using stratified Cox proportional hazards models.

Results.

A total of 1,481 patients with StUD who received CM were included alongside 1,481 matched controls. Over the 1-year follow-up period, those who received CM were 40% less likely to die (AHR=0.60, 95% CI: 0.37, 0.97) than those who did not receive CM.

Conclusion.

This study provided the first evidence that CM use in real-world healthcare settings is associated with reduced risk of mortality among patients with StUD.

Introduction

Although opioid overdose deaths have started to show modest reductions, stimulant overdose deaths continue to increase, with more than a 5-fold increase in the past decade. (1, 2) Stimulants are the fastest-growing substance use category in overdose deaths and were implicated in 50% of overdose deaths in the US in 2021,(3, 4) While there have been many efforts to reduce opioid use in recent years, stimulant use has not been adequately addressed.(5, 6) In addition to overdose risk, due to both stimulant toxicity and also because stimulants are a vector of exposure to opioids,(7) people with stimulant use disorder (StUD) are at increased risk for substantial psychosocial harms, such as psychosis and depression,(8) and medical conditions, such as cardiovascular conditions and infections, (e.g., HIV, hepatitis C),(9) and finally, increased risk of psychiatric hospitalizations compared to other substance use disorders (SUDs), and mortality.(7, 10, 11)

To counter these enormous impacts, widespread access to effective treatment for StUD is critical. Clear and consistent evidence supports contingency management (CM) as the most effective treatment for StUD.(12-14) Because of this, and in light of the surging opioid and stimulant overdose crisis in the US, the US Department of Health and Human Services recommends CM as a first-line treatment.(15) However, CM implementation has been slow, especially in comparison to treatments for opioid use disorder.(16) Real-world evidence on the impacts of CM on preventing mortality would buttress the case for widespread implementation.

The Department of Veteran Affairs (VA) has undertaken the largest-scale and longest-standing CM implementation effort across the US. The VA CM initiative, implementation strategies, and initial evaluation findings have been reported elsewhere.(17-19) The VA CM program was rolled out in 2011, and standardized documentation of CM visits using a note template that generates implementation and outcome data was integrated into the VA nationwide electronic health record system in 2018.(17) As such, VA’s CM implementation is a natural testbed to examine the impacts of CM for StUD in real-world settings. The current report looks at the real-world impact of receiving CM on all-cause mortality compared to those who did not receive CM among patients with StUD. We hypothesized that CM would reduce mortality compared to those who did not receive CM.(20, 21) In addition, as an exploratory outcome, we looked at the rate of psychiatric hospitalizations in the year following CM initiation among those who received CM compared to those who did not.

Methods

Data source

Patient data from July 2018 through Dec 2020 were obtained from the Corporate Data Warehouse (CDW), a repository of electronic health records from the VA, the largest single provider of addiction care in the country. Patient data was linked to National Death Index (NDI) data extending through Dec 2021 obtained from the VA Suicide Data Repository, which provides cause of death for Veterans.(22) This timeframe was chosen to capitalize on the most up-to-date mortality data available through the VA Suicide Data Repository, and also captures CM delivery during its peak prior to service utilization drops during the COVID-19 pandemic.(23)

Study design

This study uses a retrospective cohort design, examining the association between CM and mortality, as well as psychiatric hospitalizations. This project was designated as quality improvement and received a non-research determination from the VHA Office of Mental Health and Suicide Prevention; thus, it did not require institutional review board oversight. This report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.

Study population

We focused on adult patients with a diagnosis of StUD in the past year. Study participants were identified using the International Classification of Diseases (ICD) 10th Revision Codes (see Appendix). We identified 138,280 patients diagnosed with StUD between July 2018 and December 2020.

Exposure

Among the population with StUD, 1,698 patients received CM for StUD between July 2018 to Dec 2020, and 136,582 did not receive any CM care. We identified those who received CM care based on electronic health record note templates (i.e., health factor data) used specifically to document CM receipt.

Matching

We applied risk-set matching to minimize bias.(24, 25) For every patient who started CM, we searched for the entire cohort to identify a matching control patient among those with StUD, who did not initiate CM prior to the date the treated patient initiated CM (index date) plus 30 days to mimic intent-to-treat procedures in clinical trials. As such, this approach will estimate the potential effect of starting the CM intervention and is expected to provide a conservative estimate of the intervention effect, if the intervention effect is positive. The matching criteria were based on prior literature of variables associated with exposure to CM and mortality risks. (26) In addition to being at risk of CM exposure during the same calendar time (i.e., when the treated patient initiated CM), matching criteria included the same VA facility, race, gender, age (+/− 2 years), past-year opioid use disorder diagnosis, past-year hospitalization (yes/no), and past-year SUD clinic visit for StUD (yes/no). This method is used to match controls who had a similar likelihood of initiating CM at the same time as the treated patient. We did 1:1 exact matching with replacement, where a control patient meeting the matching criteria was allowed to be matched to more than one treated patient.

Covariates

Covariates included key patient characteristics and measures of patient complexity, including number of mental and physical comorbidities in the past year (0 to 1, 2, 3 or more comorbid conditions) based on the Elixhauser Comorbidity Index (27-29), geographic locality (urban, rural, other/unknown) defined based on the rural-urban commuting area codes,(30) homeless status, mental health conditions including serious mental illness (SMI; bipolar disorder, psychosis, schizophrenia), other mental illness (non-SMI; PTSD, anxiety, depression), cardiovascular disease, past year hospitalizations (0, 1, 2 or more), number of past year SUD clinic visits (0,1 to 5, more than 5 visits). See Appendix for more information on covariate definitions.

Outcome variables

The primary outcome was time to all-cause mortality within one year from index date. All-cause mortality serves as a comprehensive and objective measure, capturing the total mortality burden associated with SUDs.(31, 32) This approach addresses inconsistencies and underreporting common in cause-specific mortality data due to varying medical examiner practices and errors in death certification.(33-35) Mortality data were obtained through the VA Suicide Data Repository and linked to patient clinical records.(22) In addition to all-cause mortality, we also descriptively examined stimulant involvement across all-cause and overdose deaths. Given overlap of the observational period with COVID-19, we also report on number of COVID-19 deaths. As an exploratory outcome, we looked at time to first psychiatric hospitalization within one year from index date. Hospitalization was determined based on patient clinical records, and we differentiated between medical, surgical, or psychiatric hospitalizations in the VA.

Statistical analyses

Baseline patient characteristics, including those used in matching and other covariates, were summarized by the treatment groups (CM vs. no CM). Crude descriptive summaries of outcomes, rate of death and of hospitalization were reported by group. Stratified Cox proportional hazards regression was used to compare the risks of all-cause mortality and psychiatric hospitalization. Stratification was based on matched pairs, treating the pairs as separate stratum, to estimate the effect of CM on mortality and on psychiatric hospitalization while accounting for matching. For the all-cause mortality outcome, hospitalization during the follow-up period was included as a time-varying covariate (yes/no) in the model. This approach allowed the occurrence of hospitalizations over time to account for mortality risk. When psychiatric hospitalization was considered as the outcome, since death is a potential competing risk, we used the competing risk model by Fine-Gray proportional subdistribution hazard model in Cox model.(36) Potential confounders not used in matching, including the number of co-morbid conditions, rurality, homeless status, cardiovascular disease, mental health status, distribution of past-year hospitalization, and distribution of past-year SUD clinical treatment, were included as covariates in the regression models. Results are summarized using covariate-adjusted hazard ratios (AHR).

Results:

Demographic and clinical characteristics

We identified 138,280 patients with past-year StUD diagnosis between July 2018 and Dec 2020 in the VA. Most (92.9%) were male, 53.7% were between the ages of 45 and 64 years old, 43.5% identified as white and 41.9% identified as Black. Many (41.0%) had experienced homelessness or housing instability in the prior year. Other SUDs and mental health conditions were common, with 42.7% having a past-year alcohol use disorder and 44.7% having a past-year depressive disorder (Table 1). Table 1 also presents demographic and clinical information for the two matched groups. In the matching process, one control patient matched with a CM-treated patient more than once. In each matched group (N=1,481), most (73.4%) had engaged in past-year SUD clinic treatment, many (40.7%) were hospitalized in the prior 12 months, and about a quarter (26.2%) had a past-year opioid use disorder.

Table 1.

Patient characteristics among patients diagnosed with stimulant use disorder from July 2018 to December 2020.

Patient Characteristics	Patients with StUD diagnosis N=138,280		CM treated N=1,481		Matched control N=1,481
	N	%	N	%	N	%
Variables used in matching
Age in years	Mean:53.7	Sd:12.9	Mean:52.5	Sd:11.5	Mean:52.5	Sd:11.5
Race/Ethnicity
White, non-Hispanic	60,128	43.5	639	43.2	639	43.2
Black, non-Hispanic	57,867	41.9	716	48.4	716	48.4
Hispanic	9,806	7.1	64	4.3	64	4.3
Other/Unknown, non-Hispanic	10,479	7.6	62	4.2	62	4.2
Male	128,435	92.7	1403	94.7	1403	94.7
Prior year Opioid use disorder (yes)	22,182	16.0	388	26.2	388	26.2
Prior year Hospitalization (yes)	n/a	n/a	602	40.7	602	40.7
Prior year SUD clinic visit for StUD (yes)	n/a	n/a	1,087	73.4	1,087	73.4
Variables not used in matching
Rurality
Urban	123,072	89.0	1,365	92.2	1,330	89.8
Rural	11,094	8.0	85	5.7	112	7.6
Other/Unknown	4,114	3.0	31	2.1	39	2.6
Homelessness	56,648	41.0	865	58.4	802	54.2
Behavioral Health Disorders
Alcohol use disorder	59,002	42.7	836	56.5	793	53.5
Tobacco use disorder	58,870	42.6	762	51.5	787	53.1
Other SUD	26,756	19.4	477	32.2	447	30.2
Bipolar disorder	18,208	13.2	218	14.7	248	16.8
Depressive disorder	61,798	44.7	788	53.2	805	54.4
Cannabis	18,101	13.1	281	19.0	265	17.9
PTSD	46,788	33.8	539	36.4	550	37.1
Anxiety disorder	37,249	26.9	480	32.4	476	32.1
Other psychological disorder	17,865	12.9	245	16.5	232	15.7
Number of comorbidities
0	7,644	5.5	85	5.7	69	4.7
1	17,419	12.6	162	10.9	178	12.0
2	25,709	18.6	269	18.2	215	14.5
3 or more	87,508	63.3	965	65.2	1,019	68.8
Facility-level information
Number of facilities	129		100		100

Abbreviations: sd=standard deviation, SUD=substance use disorder, StUD=stimulant use disorder

Mortality

In the CM cohort, 27 patients died in the 12 months following CM initiation, while 46 patients died in the control cohort (unadjusted HR=0.58, 95% CI: 0.36, 0.94, p=0.026), resulting in crude incidence rates per 100,000 person-days of 4.8 deaths in the CM group and 8.2 in the control group. Overdose deaths occurred in 10 patients in the CM cohort and 15 patients in the control cohort. Seven deaths in the CM cohort and 12 deaths in the control cohort had documented stimulant involvement (see Table 2). After adjusting for covariates including time-varying hospitalization, reduced mortality risk associated with CM was 41% (AHR=0.59, 95% CI: 0.36,0.95, p=0.030). Patients with more complex underlying conditions (i.e., those with 3 or more comorbidities) were more likely to die during the follow-up period (AHR=2.11, 95% CI: 1.02, 4.39, p=0.045; see Table 3).

Table 2.

Unadjusted mortality and hospitalization outcomes between CM treated and 1:1 matched controls during the 12-month follow-up.

	CM treated N=1,481		Matched control N=1,481		Hazard ratio	95% CI	P value
	N	%	N	%
All-cause mortality	27	1.82	46	3.11	0.58	0.36, 0.94	0.026
Crude mortality rate (per 100,000 person days)	4.76		8.17		n/a	n/a	n/a
Overdose deaths	10	0.7	15	1.01	0.67	0.30, 1.48	0.321
Stimulant-involved deaths	7	0.5	12	0.8	0.58	0.23, 1.48	0.257
Hospitalization (yes/no)	679	45.9	539	36.4	1.37	1.22, 1.54	<.0001
# of hospitalizations/patient	Mean:1.11	Sd:2.02	Mean:0.99	Sd:2.01	n/a		n/a
Hospitalization type (% of all hospitalizations)
Psychiatric		42.12%		36.74%	1.44	1.22, 1.69	<.0001
Surgical		5.52%		5.79%	1.27	0.89, 1.79	0.183
Medical		24.48%		35.09%	0.96	0.80, 1.15	0.635

Note: Hazard ratio compares the hazard (risk) of an event occurring between the two groups and thus cannot be calculated for continuous variables.

Table 3.

Adjusted outcome models for people with stimulant use disorder, comparing those who received contingency management (CM) vs. not over 12-month follow-up.

Outcome	Hazard ratio	95% CI	P value
All-cause mortality
CM vs. no CM	0.59	0.36, 0.95	0.030
Number of comorbidities (0-1 as reference)
2	1.54	0.69, 3.43	0.288
3 or more	2.11	1.02, 4.39	0.045
Rurality (urban as reference)
Rural	0.64	0.34, 1.21	0.169
Homelessness (housed as reference)	1.06	0.77, 1.47	0.720
Cardiovascular disease	1.13	0.76, 1.68	0.562
Mental health (no mental health condition as reference)
Serious mental illness	0.93	0.57, 1.52	0.770
Non-serious mental illness	0.88	0.55, 1.39	0.574
Other substance use disorder	0.80	0.54, 1.19	0.274
Number of past-year hospitalizations (0 as reference)
1 hospitalization	1.32	0.88, 1.98	0.180
2 or more hospitalizations	0.81	0.52, 1.25	0.339
Number of past-year SUD clinic visits
1 to 6 visits	1.08	0.70, 1.67	0.728
More than 6 visits	0.94	0.56, 1.55	0.798
Time-varying hospitalizations	1.28	0.92, 1.77	0.144
Psychiatric Hospitalizations
CM vs. no CM	1.48	1.25, 1.75	<.0001
Number of comorbidities (0-1 as reference)
2	0.75	0.52, 1.07	0.110
3 or more	0.87	0.63, 1.21	0.421
Rurality (urban as reference)
Rural	0.85	0.59, 1.22	0.375
Homelessness (housed as reference)	1.17	0.97, 1.41	0.100
Cardiovascular disease	0.89	0.74, 1.06	0.189
Mental health (no mental health condition as reference)
Serious mental illness	1.86	1.38, 2.51	<.0001
Non-serious mental illness	1.44	1.09, 1.89	0.010
Other substance use disorder	1.20	0.94, 1.52	0.142
Number of past-year hospitalizations (0 as reference)
1 hospitalization	2.22	1.79, 2.75	<.0001
2 or more hospitalizations	3.58	2.81, 4.55	<.0001
Number of past-year SUD clinic visits
1 to 6 visits	0.88	0.70, 1.10	0.258
More than 6 visits	0.83	0.67, 1.02	0.309

Given the observational period overlayed with COVID-19, we also note that 0 deaths in the CM case cohort and 3 deaths in the control cohort, corresponding to crude incidence rates per 100,000 person-days of 0 in the CM cohort and 0.53 in the control cohort. In a sensitivity analysis, we excluded people who died from COVID-19-related causes (i.e., removing the controls and their matched cases; see Appendix). The association between CM receipt and mortality remains similar to the primary analysis (AHR=0.61, 95% CI: 0.38, 0.98, p=0.043).

Hospitalization

In the CM group, 45.9% of patients were hospitalized in the follow-up period versus 36.4% of people in the control group. The most common type of hospitalization was psychiatric with 42.1% and 36.7% of all admissions in the CM and control cohorts, respectively, being psychiatric in nature. Medical admissions made up 24.5% and 35.1% of admissions in the CM and control cohorts, respectively. Surgical admissions made up 5.5% and 5.8% of admissions in the CM and control cohort. In the multivariable model, CM was associated with higher risk of psychiatric hospitalization (AHR=1.48, 95% CI: 1.25, 1.75, p<0.001). Those with serious mental illness (AHR=1.86, 95% CI: 1.23, 2.51, p<0.001), and those with non-serious mental (AHR=1.44, 95% CI: 1.09, 1.89, p=0.010), and those with one prior year hospitalization (AHR=2.22, 95% CI: 1.79, 2.75, p<0.001), and those with two or more prior year hospitalization (AHR=3.58, 95% CI: 2.81, 4.55, p<0.001) were more likely to be psychiatric hospitalized (see Table 3).

Discussion:

This paper provides critical insights about the impact of CM on mortality. Mortality rates were high in this cohort study of people with StUD, with a death rate >8% compared to <4% among veterans in general in the VA.(37) People who received CM had about 40% lower rate of all-cause mortality in the year following CM initiation, after adjusting for all included covariates, compared to the match control cohort of those who have StUD but did not receive CM. The magnitude of the decreased risk for death among those who received CM compared to those with StUD in the control group is clinically important and is similar to the reduction in all-cause mortality from treating opioid use disorder with buprenorphine (AHR=0.63, 95% CI: 0.46, 0.87).(32). Moreover, the reduction in mortality cannot be accounted for by differences in hospitalizations between the two groups.

Hospitalizations were also high across the board, with ~40% of people in each cohort being hospitalized. This is likely reflective of higher levels of medical and psychiatric comorbidity in these patients with StUD. Indeed, about three-quarters of patients across both conditions had multiple comorbidities within the previous year, including elevated prevalence of other SUDs, psychiatric disorders, and medical conditions. Furthermore, adverse social determinants of health, including low income, unemployment, and social isolation, elevate the risk for hospitalization.(38-40) In this study, elevated hospitalizations in the CM cohort were driven by psychiatric hospitalizations. Psychiatric hospitalizations were more likely for those who received CM in the year following CM. The elevated rate of hospitalizations among those in the CM cohort may be linked to overall higher utilization or identification of symptoms in outpatient CM care that lead to psychiatric hospitalization, as found in other studies of outpatient SUD care.(41, 42)

This study confirms what has long been thought anecdotally, first, that CM treatment is associated with increased behavioral health treatment utilization(43) shown here as increased psychiatric hospitalizations. The higher rate of hospitalizations may, in part, be protective against mortality. By increasing treatment utilization, such as hospitalizations, patients may experience improved health and reduce mortality. Most importantly, our data confirms that widescale access to CM for those with StUD can save lives, both reducing all-cause morality as well as preventing overdose deaths. This is particularly important given these findings are from real-world clinical care as opposed to clinical trials focused on efficacy, lending support to the assertion that widescale uptake of CM across health systems and states could potentially help reduce deaths.(3)

In order for CM to have broad uptake, we must increase penetration of CM programs both within the VA and throughout communities that are facing high consequences from stimulant use disorder, as well as other substance use disorders. A major step toward increasing CM availability in real-world settings is state-wide demonstrations for Medicaid beneficiaries. For example, there are currently CM demonstrations ongoing in the states of CA and MI, with others forthcoming in WA and MT.(44, 45) These demonstrations are a major step in expanding access to CM. However, several barriers still stand in the way of improving the reach of CM in real-world settings. Primary among these are incentive caps, which hold incentive distributions to <$600 per calendar year due to tax reporting requirements,(46) despite empirical evidence and calls from policy groups to consider incentives as part of medical treatment, not as taxable income. Moreover, it is widely believed that this limit is actually only $75, which further suppresses CM dissemination.

Limitations

This is an observational study that may be subject to selection bias, though we document differences in baseline characteristics between the cohorts and adjust for these differences using multivariable analyses. Future work, including prospective studies to further clarify causal relationships, is needed. Second, though we adjust for characteristics between cohorts, there is always the potential for unmeasured differences to impact the outcomes differentially. For example, we report on a number of comorbidities between groups, but the severity of comorbidities is an additional dimension that is not straightforward to extract from medical records, and thus we leave it to be addressed in future prospective work on this topic. Third, COVID-19 overlaps with the observational period. Though we see similar COVID-related mortality rates between the CM and non-CM cohorts, the case may be that structural or other unmeasured impacts of COVID-19 may influence study findings. Fourth, this report uses the longest-standing and largest-scale CM implementation in the US, which was implemented in the VA. The VA is a large, national health system; however, it may limit generalizability of findings. As other large-scale demonstrations, such as the MediCali 1115 Recovery Incentives pilot, have more time in the field, similar evaluations are needed.

Conclusion

In conclusion, this study provides the strongest real-world evidence to date that CM for StUD is associated with a significant reduction in all-cause mortality. Patients who received CM were approximately 40% less likely to die in the year following treatment initiation, even after adjusting for clinical complexity and other risk factors. These findings, drawn from the largest integrated health care system in the U.S., underscore the urgent need to expand access to CM as a life-saving intervention. Broader adoption of CM across public and private systems could meaningfully reduce preventable deaths among people with StUD and help close critical gaps in the nation’s overdose response.