Veteran suicide and associated community characteristics in Oregon.

Developing a public health approach to suicide prevention among United States (US) military veterans requires additional data and guidance on where, when, for whom, and what prevention resources should be deployed. This study examines veteran suicide mortality across one US state (Oregon) to identify county-level "hotspots" for veteran suicide, identify community characteristics associated with increased suicide among veterans, and examine excess spatial risk after accounting for space, time, and community characteristics. We linked Oregon mortality data with VA databases to identify veterans who had resided in Oregon and died by suicide between January 1, 2009 and December 31, 2018 (n = 1727). Community characteristic data were gathered at the county level from publicly available datasets on social determinants of health known to be associated with poor health outcomes, including suicide risk. We estimated spatial generalized linear mixed models for the full 10-year period and for each 5-year period using integrated nested Laplace approximation with county as the higher hierarchy. Smoothed standardized mortality ratios were used to identify counties with higher risk of veteran suicide. We found a small clustering of counties in the southwestern corner of Oregon that held the highest risk for veteran suicide across the ten years studied. In multivariable models, higher prevalence of unmarried persons was the only community measure significantly associated with increased veteran suicide risk. However, social contextual factors as a group, along with geographic space, explained most risk for suicide among veterans at the population level.

Evaluating the temporal association between the recency of prescribed controlled substance acquisition and fatal and non-fatal opioid overdose.

BACKGROUND: Previous studies have explored psychosocial effects as possible triggers of opioid overdose (OOD). However, little is known about the temporal association between OOD and prescribed controlled substance (CS) acquisition. OBJECTIVE: The objective of this study was to evaluate the temporal relationship between OOD and acquiring prescribed CSs prior to OOD. METHODS: This study is an exploratory descriptive analysis using Arkansas Prescription Drug Monitoring Program (AR-PDMP) data linked to death certificate and statewide inpatient discharge records. All persons with ≥1 AR-PDMP prescription fill(s) between 1 January 2014 and 31 December 2017 were included (n = 1,946,686). For persons that experienced OOD and had ≥1 PDMP record(s), the difference in days between OOD and the most recent AR-PDMP prescription filled prior to an OOD was recorded. To account for censoring, a sensitivity analysis was conducted restricting the study group to "New AR-PDMP Entrants" that had at least a 180-day gap between consecutive AR-PDMP fill dates. RESULTS: 28,998,307 AR-PDMP records were analyzed for 1,946,686 individuals. 7195 persons experienced 9223 OODs and 414 (4.49%) of those were fatal. Of these, 6236 experienced ≥1 OOD and acquired prescribed CSs prior to or on the day of the first OOD. Of those that experienced ≥1 OOD(s), 2201 (30.59%) had an AR-PDMP record in the 0- to 5-day period prior to their overdose and 497 (6.91%) had an AR-PDMP record the day prior to their overdose. Among New AR-PDMP Entrants that experienced ≥1 OOD(s), 408 (27.38%) had an AR-PDMP record in the 0- to 5-day period prior to their overdose. CONCLUSION: Though the vast majority of persons accessing CSs in Arkansas did not experience an OOD, a sizable proportion of persons that experience an OOD(s) obtained prescribed CSs immediately prior.

Using data science to improve outcomes for persons with opioid use disorder.

Medication treatment for opioid use disorder (MOUD) is an effective evidence-based therapy for decreasing opioid-related adverse outcomes. Effective strategies for retaining persons on MOUD, an essential step to improving outcomes, are needed as roughly half of all persons initiating MOUD discontinue within a year. Data science may be valuable and promising for improving MOUD retention by using "big data" (e.g., electronic health record data, claims data mobile/sensor data, social media data) and specific machine learning techniques (e.g., predictive modeling, natural language processing, reinforcement learning) to individualize patient care. Maximizing the utility of data science to improve MOUD retention requires a three-pronged approach: (1) increasing funding for data science research for OUD, (2) integrating data from multiple sources including treatment for OUD and general medical care as well as data not specific to medical care (e.g., mobile, sensor, and social media data), and (3) applying multiple data science approaches with integrated big data to provide insights and optimize advances in the OUD and overall addiction fields.

Adjustment to the COVID-19 pandemic: associations with global and situational meaning.

COVID-19 has created pervasive upheaval and uncertainty in communities around the world. This investigation evaluated associations between discrete dimensions of personal meaning and psychological adjustment to the pandemic among community residents in a southern US state. In this cross-sectional study, 544 respondents were assessed during a period of reopening but accelerating infection rates. Validated measures were used to evaluate theoretically distinct dimensions of perceived global meaning (Meaning-in-Life Questionnaire) and pandemic-specific meaning (Meaning in Illness Scale). Adjustment outcomes included perceived stress, pandemic-related helplessness, and acceptance of the pandemic. In multivariate models that controlled for demographic and pandemic-related factors, stronger attained global meaning (i.e., perceptions that life is generally meaningful) and attained situational meaning (i.e., perceptions that the pandemic experience was comprehensible) were related to better adjustment on all three outcomes (all p's < .001). In contrast, seeking situational meaning (i.e., ongoing efforts to find coherence in the situation) was associated with poorer adjustment on all indices (all p's < .001). Results offer novel information regarding theoretically salient dimensions of meaning, which may have direct relevance for understanding how community residents adapt to the challenges of a major public health crisis.

Enrollee Experience with Providers in the Arkansas Medicaid Expansion Program.

BACKGROUND: Patient ratings of their healthcare experience as a quality measure have become critically important since the implementation of the Affordable Care Act (ACA). The ACA enabled states to expand Medicaid eligibility to reduce uninsurance nationally. Arkansas gained approval to use Medicaid funds to purchase a qualified health plan (QHP) through the ACA marketplace for newly eligible beneficiaries. OBJECTIVE: We compare patient-reported satisfaction between fee-for-service Medicaid and QHP participants. DESIGN: The Consumer Assessment of Healthcare Providers and Systems (CAHPS) was used to identify differences in Medicaid and QHP enrollee healthcare experiences. Data were analyzed using a regression discontinuity design. PARTICIPANTS: Newly eligible Medicaid expansion participants enrolled in Medicaid during 2013 completed the Consumer Assessment of Health Providers and Systems (CAHPS) survey in 2014. Survey data was analyzed for 3156 participants (n = 1759 QHP/1397 Medicaid). MEASURES: Measures included rating of personal and specialist provider, rating of all healthcare received, and whether the provider offered to communicate electronically. Demographic and clinical characteristics of the enrollees were controlled for in the analyses. METHODS: Regression-discontinuity analysis was used to evaluate differential program effects on positive ratings as measured by the CAHPS survey while controlling for demographic and health characteristics of participants. KEY RESULTS: Adjusted logistic regression models for overall healthcare (OR = 0.71, 95%CI = 0.56-0.90, p = 0.004) and personal doctor (OR = 0.68, 95%CI = 0.53-0.87, p = 0.002) predicted greater satisfaction among QHP versus Medicaid participants. Results were not significant for specialists or for use of electronic communication with provider. CONCLUSIONS: Using a quasi-experimental statistical approach, we were able to control for observed and unobserved heterogeneity showing that among participants with similar characteristics, including income, QHP participants rated their personal providers and healthcare higher than those enrolled in Medicaid. Access to care, utilization of care, and healthcare and health insurance literacy may be contributing factors to these results.

Improving quality of life assessments for high-need adult Medicaid service users with mental health conditions.

PURPOSE: There is a lack of consensus on how to evaluate health and social service programs for people with mental health (MH) conditions. Having service users be the primary decision makers in selecting outcome measures can inform a meaningful evaluation strategy. We sought to identify the quality of life (QoL) survey preferences of high-need adult service users with MH conditions. METHODS: A systematic review identified generic, self-reported QoL surveys with evidence of validity in MH populations of interest. An advisory panel selected the most promising surveys to assess the success of programs like Medicaid for MH service users. Three groups of high-need, adult service users with MH conditions and one group of direct care staff ranked the surveys from the advisory panel, and generated and ranked characteristics that were desirable or undesirable in a QoL survey. RESULTS: Twenty-two surveys met the inclusion criteria. Of the six surveys selected by the advisory panel, groups of service users and direct care staff most preferred the Warwick-Edinburg Mental Well-being Scale (WEMWBS). The WEMWBS best embodied the features prioritized by the groups: to have a user-friendly format and positive focus, to be clearly worded and brief, and to avoid presumptive or unrealistic items. Service user groups appreciated survey topics most amenable to self-report, such as satisfaction with relationships. CONCLUSION: Using QoL surveys that service users prefer can reduce the chance that deteriorating QoL is going unchecked, and increase the chance that decisions based on survey findings are meaningful to service users.

Increased Vascular Pathology in Older Veterans With a Purple Heart Commendation or Chronic Post-Traumatic Stress Disorder.

The goal of this retrospective cohort study was to determine whether stressors related to military service, determined by a diagnosis of chronic post-traumatic stress disorder (cPTSD) or receiving a Purple Heart (PH), are associated with an increased risk of vascular risk factors and disease, which are of great concern for veterans, who constitute a significant portion of the aging US population. The Veterans Integrated Service Network (VISN) 16 administrative database was searched for individuals 65 years or older between October 1, 1997 to September 30, 1999 who either received a PH but did not have cPTSD (PH+/cPTSD-; n = 1499), had cPTSD without a PH (PH-/cPTSD+; n = 3593), had neither (PH-/cPTSD-; n = 5010), or had both (PH+/cPTSD+; n = 153). In comparison to the control group (PH-/cPTSD-), the PH+/cPTSD- group had increased odds ratios for incidence and prevalence of diabetes mellitus, hypertension, and hyperlipidemia. The PH-/cPTSD+ group had increased odds ratios for prevalence of diabetes mellitus and for the incidence and prevalence of hyperlipidemia. The PH-/cPTSD+ and PH+/cPTSD- groups were associated with ischemic heart disease and cerebrovascular disease, but not independently of the other risk factors. The PH+/cPTSD+ group was associated only with an increase in the incidence and prevalence of hyperlipidemia, though this group's much smaller sample size may limit the reliability of this finding. We conclude that certain physical and psychological stressors related to military service are associated with a greater incidence of several vascular risk factors in veterans aged 65 years or older, which in turn are associated with greater rates of ischemic heart disease and cerebrovascular disease.

Prevalence of Chronic Kidney Disease, Thrombotic Cardiovascular Events, and Use of Oral P2Y12 Inhibitors among Veterans.

BACKGROUND: Contemporary prevalence of chronic kidney disease (CKD) and thrombotic cardiovascular (CV) events remains unclear in Veterans enrolled in the Veterans Affairs Health Care System (VA) care. Although oral P2Y12 inhibitors (P2Y12i) are increasingly being prescribed to this patient population, the overall prescription trend for P2Y12i remains unclear. METHODS: Using national VA corporate warehouse data, we used International Classification of Diseases-9 codes to identify Veterans with CKD, dialysis-dependent CKD, and CV events. VA pharmacy data were used to count P2Y12i prescriptions for the federal fiscal years (FY) 2011 through 2015. RESULTS: The period prevalence of Veterans with CKD was 378,233 (6.1%). The point prevalence of CKD increased by 49% from 132,979 (2.30%) in FY11 to 213,444 (3.42%) in FY15. The period prevalence of Veterans with dialysis-dependent CKD was 150,298 (2.4%). In all, 128,703 (56.7%) CV events occurred in Veterans with CKD. Veterans with CKD were given 50.1% of prescriptions for clopidogrel, 49.3% for prasugrel, and 60.4% for ticagrelor. In this patient population, year-to-year increases in P2Y12i prescriptions were observed with a dramatic increase in ticagrelor prescriptions. CONCLUSION: CKD is common among Veterans and its true prevalence is likely being underestimated. The prevalence of dialysis-dependent CKD is higher among Veterans than the non-Veteran US population. CV events are widely co-prevalent and these patients are commonly prescribed P2Y12i. The recent increase in ticagrelor prescriptions in this patient population and large cost differences between the 3 P2Y12i underline the need for future studies to identify the preferred P2Y12i for these patients.

Factors Associated with Opioid Initiation in OEF/OIF/OND Veterans with Traumatic Brain Injury.

Objective: These analyses examined opioid initiation and chronic use among Iraq (OIF) and Afghanistan (OEF/OND) veterans with a new diagnosis of traumatic brain injury (TBI) in the Veterans Health Administration (VHA). Methods: Data were obtained from national VHA data repositories. Analyses included OEF/OIF/OND veterans with a new TBI diagnosis in 2010-2012 who used the VHA at least twice, had not received a VHA opioid prescription in the 365 days before diagnosis, and had at least 365 days of data available after TBI diagnosis. Results: Analyses included 35,621 veterans. Twenty-one percent initiated opioids; among new initiators, 23% used chronically. The mean dose was 24.0 mg morphine equivalent dose (MED) daily (SD = 24.26); mean days supplied was 60.52 (SD = 74.69). Initiation was significantly associated with age 36-45 years (odds ratio [OR] = 1.09, 95% CI = 1.01-1.17, P = 0.04), female gender (OR = 1.22, P < 0.001), having back pain (OR = 1.38, P < 0.0001), arthritis/joint pain (OR = 1.24, P < 0.0001), or neuropathic pain (OR = 1.415, P < 0.02). In veterans age 36-45 years, those living in small rural areas had higher odds of chronic opioid use (OR = 1.31, P < 0.0001, and OR = 1.33, P = 0.006, respectively) and back pain (OR = 1.36, P = 0.003). Headache/migraine pain was associated with decreased odds of chronic opioid use (OR = 0.639, P = 0.003). Conclusions: Prevalence of opioid use is relatively low among OEF/OIF/OND veterans with newly diagnosed TBI who are using VHA. Among those who initiated opioids, about 25% use them chronically. Prescribing was mostly limited to moderate doses, with most veterans using opioids for approximately two months of the 12-month study period.

Correction to: Frequency and predictors of health services use by native Hawaiians and Pacific islanders: evidence from the U.S. National Health Interview Survey.

Following publication of the original article [1], the author reported the following errors in Table 1 and Table 2.

Frequency and predictors of health services use by Native Hawaiians and Pacific Islanders: evidence from the U.S. National Health Interview Survey.

BACKGROUND: Native Hawaiians and Pacific Islanders (NHPIs) are one of the fasting growing racial groups in the United States (US). NHPIs have a significantly higher disease burden than the US population as a whole, yet they remain underrepresented in research. The purpose of this study is to examine factors associated with health care utilization among NHPIs. METHODS: Drawing from the 2014 NHPI-National Health Interview Survey, we used stereotype logistic regressions to examine utilization of emergency department (ED) and outpatient services among 2172 individuals aged 18 and older. RESULTS: NHPIs with chronic diseases were twice as likely to be multiple ED users and nearly four times as likely to be frequent-users of outpatient services. Social support played a protective role in preventing multiple use of ED. Having a usual source of care made it more than eight times as likely to be a frequent-user of outpatient services. Use of eHealth information increased the odds of using ED and outpatient services. Ability to afford health care increased the odds of using outpatient services. There was no association between health insurance coverage and use of ED and outpatient services among NHPIs. CONCLUSIONS: This research provides the first available national estimates of health services use by NHPIs. Efforts to improve appropriate use of health services should consider leveraging the protective factors of social support to reduce the odds of frequent ED use, and having a usual source of care to increase use of outpatient services.

"Availability of healthcare providers for rural veterans eligible for purchased care under the veterans choice act".

BACKGROUND: Military Veterans in the United States are more likely than the general population to live in rural areas, and often have limited geographic access to Veterans Health Administration (VHA) facilities. In an effort to improve access for Veterans living far from VHA facilities, the recently-enacted Veterans Choice Act directed VHA to purchase care from non-VHA providers for Veterans who live more than 40 miles from the nearest VHA facility. To explore potential impacts of these reforms on Veterans and healthcare providers, we identified VHA-users who were eligible for purchased care based on distance to VHA facilities, and quantified the availability of various types of non-VHA healthcare providers in counties where these Veterans lived. METHODS: We combined 2013 administrative data on VHA-users with county-level data on rurality, non-VHA provider availability, population, household income, and population health status. RESULTS: Most (77.9%) of the 416,338 VHA-users who were eligible for purchased care based on distance lived in rural counties. Approximately 16% of these Veterans lived in primary care shortage areas, while the majority (70.2%) lived in mental health care shortage areas. Most lived in counties that lacked specialized health care providers (e.g. cardiologists, pulmonologists, and neurologists). Counterintuitively, VHA played a greater role in delivering healthcare for the overall adult population in counties that were farther from VHA facilities (30.7 VHA-users / 1000 adults in counties over 40 miles from VHA facilities, vs. 22.4 VHA-users / 1000 adults in counties within 20 miles of VHA facilities, p < 0.01). CONCLUSIONS: Initiatives to purchase care for Veterans living more than 40 miles from VHA facilities may not significantly improve their access to care, as these areas are underserved by non-VHA providers. Non-VHA providers in the predominantly rural areas more than 40 miles from VHA facilities may be asked to assume care for relatively large numbers of Veterans, because VHA has recently cared for a greater proportion of the population in these areas, and these Veterans are now eligible for purchased care.

Impact of the INBRE summer student mentored research program on undergraduate students in Arkansas.

The Institutional Development Award (IDeA) program, housed within the National Institute for General Medical Sciences, administers the Networks of Biomedical Research Excellence (INBRE) as a strategic mission to broaden the geographic distribution of National Institutes of Health (NIH) funding within the United States. Undergraduate summer student mentored research programs (SSMRP) are a common feature of INBRE programs and are designed to increase undergraduate student interest in research careers in the biomedical sciences. Little information is available about student perspectives on how these programs impact their choices relative to education and careers. Therefore, we conducted qualitative interviews with 20 participants from the Arkansas INBRE SSMRP in the years 2002-2012. Each telephone interview lasted 30-45 min. An interview guide with a broad "grand tour" question was used to elicit student perspectives on SSMRP participation. Interviews were digitally recorded, then transcribed verbatim, and the transcript checked for accuracy. Content analysis and constant comparison were used to identify nine themes that were grouped into three temporal categories: before, during, and after the SSMRP experience. Students viewed the experience as positive and felt it impacted their career choices. They emphasized the value of mentoring in the program, and some reported maintaining a relationship with the mentor after the summer experience ended. Students also valued learning new laboratory and presentation skills and felt their research experience was enhanced by meeting students and scientists with a wide range of career interests. These data suggest that the Arkansas INBRE and the NIH IDeA program are successfully meeting the goal of increasing interest in research among undergraduates.

Identifying and Understanding Barriers and Facilitators to Medication Adherence Among Marshallese Adults in Arkansas.

Background: Significant health disparities are present in Marshallese adults residing in the United States, most notably a high incidence of type 2 diabetes and other chronic conditions. There is limited research on medication adherence in the Marshallese population. Objective: This study explored perceptions of and experiences with medication adherence among Marshallese adults residing in Arkansas, with the aim of identifying and better understanding barriers and facilitators to medication adherence. Methods: Eligible participants were Marshallese adults taking at least one medication for a chronic health condition. Each participant completed a brief survey and semistructured interview conducted in Marshallese by a bilingual Marshallese staff member. Interviews were recorded, transcribed, and translated from Marshallese to English. Qualitative data were coded for a priori and emergent themes. Results: A total of 40 participants were included in the study. The most common contributing factor for nonadherence was forgetting to take medication (82%). A majority of participants (70%) reported difficulty paying for medicine, 45% reported at least one form of cost-related nonadherence, and 40% engaged in more than one cost-related nonadherence practice. Family support and medication pill boxes were identified as facilitators for medication adherence. The majority of the participants (76.9%) stated that they understood the role of a pharmacist. Participants consistently desired more education on their medications from pharmacy providers. Conclusion: This is the first study to explore barriers and facilitators to medication adherence among Marshallese patients. The findings can be used to develop methods to improve medication adherence among Marshallese.

The influence of propoxyphene withdrawal on opioid use in veterans.

PURPOSE: Our aim is to determine if propoxyphene withdrawal from the US market was associated with opioid continuation, continued chronic opioid use, and secondary propoxyphene-related adverse events (emergency department visits, opioid-related events, and acetaminophen toxicity). METHODS: Medical service use and pharmacy data from 19/11/08 to 19/11/11 were collected from the national Veterans Healthcare Administration healthcare databases. A quasi-experimental pre-post retrospective cohort design utilizing a historical comparison group provided the study framework. Logistic regression controlling for baseline covariates was used to estimate the effect of propoxyphene withdrawal. RESULTS: There were 24,328 subjects (policy affected n = 10,747; comparison n = 13,581) meeting inclusion criteria. In the policy-affected cohort, 10.6% of users ceased using opioids, and 26.6% stopped chronic opioid use compared with 3.8% and 13.5% in the historical comparison cohort, respectively. Those in the policy-affected cohort were 2.7 (95%CI: 2.5-2.8) and 3.2 (95%CI: 2.9-3.6) times more likely than those in the historical comparison cohort to discontinue chronic opioid and any opioid use, respectively. Changes in adverse events and Emergency Department (ED) visits were not different between policy-affected and historical comparison cohorts (p > 0.05). CONCLUSIONS: The withdrawal of propoxyphene-containing products resulted in rapid and virtually complete elimination in propoxyphene prescribing in the veterans population; however, nearly 90% of regular users of propoxyphene switched to an alternate opioid, and three quarters continued to use opioids chronically.

Predictors of Initiation and Engagement of Cognitive Processing Therapy Among Veterans With PTSD Enrolled in Collaborative Care.

Collaborative care (CC) increases access to evidence-based pharmacotherapy and psychotherapy. The study aim was to identify the characteristics of rural veterans receiving a telemedicine-based CC intervention for posttraumatic stress disorder (PTSD) who initiated and engaged in cognitive processing therapy (CPT) delivered via interactive video. Veterans diagnosed with PTSD were recruited from 11 community-based outpatient clinics (N = 133). Chart abstraction identified all mental health encounters received during the 12-month study. General linear mixed models were used to identify characteristics that predicted CPT initiation and engagement (attendance at 8 or more sessions). For initiation, higher PTSD severity according to the Clinician Administered PTSD Scale (d = -0.39, p = .038) and opt-out recruitment (vs. self-referral; d = -0.49, p = .010) were negative predictors. For engagement, major depression (d = -1.32, p = .006) was a negative predictor whereas a pending claim for military service connected disability (d = 2.02, p = .008) was a positive predictor. In general, veterans enrolled in CC initiated and engaged in CPT at higher rates than usual care. Those with more severe symptoms and comorbidity, however, were at risk of not starting or completing CPT.

Social network analysis of biomedical research collaboration networks in a CTSA institution.

BACKGROUND: The popularity of social networks has triggered a number of research efforts on network analyses of research collaborations in the Clinical and Translational Science Award (CTSA) community. Those studies mainly focus on the general understanding of collaboration networks by measuring common network metrics. More fundamental questions about collaborations still remain unanswered such as recognizing "influential" nodes and identifying potential new collaborations that are most rewarding. METHODS: We analyzed biomedical research collaboration networks (RCNs) constructed from a dataset of research grants collected at a CTSA institution (i.e., University of Arkansas for Medical Sciences (UAMS)) in a comprehensive and systematic manner. First, our analysis covers the full spectrum of a RCN study: from network modeling to network characteristics measurement, from key nodes recognition to potential links (collaborations) suggestion. Second, our analysis employs non-conventional model and techniques including a weighted network model for representing collaboration strength, rank aggregation for detecting important nodes, and Random Walk with Restart (RWR) for suggesting new research collaborations. RESULTS: By applying our models and techniques to RCNs at UAMS prior to and after the CTSA, we have gained valuable insights that not only reveal the temporal evolution of the network dynamics but also assess the effectiveness of the CTSA and its impact on a research institution. We find that collaboration networks at UAMS are not scale-free but small-world. Quantitative measures have been obtained to evident that the RCNs at UAMS are moving towards favoring multidisciplinary research. Moreover, our link prediction model creates the basis of collaboration recommendations with an impressive accuracy (AUC: 0.990, MAP@3: 1.48 and MAP@5: 1.522). Last but not least, an open-source visual analytical tool for RCNs is being developed and released through Github. CONCLUSIONS: Through this study, we have developed a set of techniques and tools for analyzing research collaboration networks and conducted a comprehensive case study focusing on a CTSA institution. Our findings demonstrate the promising future of these techniques and tools in understanding the generative mechanisms of research collaborations and helping identify beneficial collaborations to members in the research community.

Development and validation of machine-learning algorithms predicting retention, overdoses, and all-cause mortality among US military veterans treated with buprenorphine for opioid use disorder.

BACKGROUND: Buprenorphine for opioid use disorder (B-MOUD) is essential to improving patient outcomes; however, retention is essential. OBJECTIVE: To develop and validate machine-learning algorithms predicting retention, overdoses, and all-cause mortality among US military veterans initiating B-MOUD. METHODS: Veterans initiating B-MOUD from fiscal years 2006-2020 were identified. Veterans' B-MOUD episodes were randomly divided into training (80%;n = 45,238) and testing samples (20%;n = 11,309). Candidate algorithms [multiple logistic regression, least absolute shrinkage and selection operator regression, random forest (RF), gradient boosting machine (GBM), and deep neural network (DNN)] were used to build and validate classification models to predict six binary outcomes: 1) B-MOUD retention, 2) any overdose, 3) opioid-related overdose, 4) overdose death, 5) opioid overdose death, and 6) all-cause mortality. Model performance was assessed using standard classification statistics [e.g., area under the receiver operating characteristic curve (AUC-ROC)]. RESULTS: Episodes in the training sample were 93.0% male, 78.0% White, 72.3% unemployed, and 48.3% had a concurrent drug use disorder. The GBM model slightly outperformed others in predicting B-MOUD retention (AUC-ROC = 0.72). RF models outperformed others in predicting any overdose (AUC-ROC = 0.77) and opioid overdose (AUC-ROC = 0.77). RF and GBM outperformed other models for overdose death (AUC-ROC = 0.74 for both), and RF and DNN outperformed other models for opioid overdose death (RF AUC-ROC = 0.79; DNN AUC-ROC = 0.78). RF and GBM also outperformed other models for all-cause mortality (AUC-ROC = 0.76 for both). No single predictor accounted for >3% of the model's variance. CONCLUSIONS: Machine-learning algorithms can accurately predict OUD-related outcomes with moderate predictive performance; however, prediction of these outcomes is driven by many characteristics.

Impact of transitioning from long-term to intermittent opioid therapy on the development of opioid-related adverse outcomes: A retrospective cohort study.

BACKGROUND: Increasing pressures exist to reduce or discontinue opioid use among patients currently on long-term opioid therapy (LTOT). It is essential to understand the potential effects of opioid reduction. METHODS: This retrospective cohort study was conducted among veterans with chronic pain and on LTOT. Using 1:1 propensity score-matched samples of veterans switching to intermittent opioid therapy and those continuing LTOT, we examined the development of subsequent substance use disorders (SUD composite; individual SUD types: opioid, non-opioid drug, and alcohol use disorders) and opioid-related adverse outcomes (ORAO composite; individual ORAO types: accidents resulting in wounds/injuries, opioid-related and alcohol/non-opioid medication-related accidents and overdoses, self-inflicted and violence-related injuries). Sensitivity analyses were conducted using logistic regression with stabilized inverse probability of treatment weighting (SIPTW) and instrumental variable (IV) models. RESULTS: A total of 29,293 veterans switching to intermittent therapy were matched to veterans continuing LTOT. With matched samples, no differences were found in composite SUDs and ORAOs between the groups. With SIPTW, veterans switching to intermittent opioid therapy had higher odds of composite SUDs and ORAOs (SUDs aOR=1.12, 95%CI: 1.07,1.17; ORAOs aOR=1.05, 95%CI:1.00,1.09). IV models found lower risks for composite SUDs and ORAOs among veterans switching to intermittent opioid therapy (SUDs: ß = -0.38, 95%CI:-0.63,-0.13; ORAOs: ß = -0.27, 95%CI:-0.50,-0.04). CONCLUSIONS: There were no consistent associations between transitioning patients from LTOT to intermittent opioid therapy and the risk of SUDs and ORAOs.