Variation in Clinical Characteristics and Longitudinal Outcomes in Individuals with Opioid Use Disorder Diagnosis Codes

Abstract

Background: Patterns of opioid use vary, including prescribed use without aberrancy, limited aberrant use, and potential opioid use disorder (OUD). In clinical practice, similar opioid-related International Classification of Disease (ICD) codes are applied across this spectrum, limiting understanding of how groups vary by sociodemographic factors, comorbidities, and long-term risks.

Objective: (1) Examine how Veterans assigned opioid abuse/dependence ICD codes vary at diagnosis and with respect to long-term risks. (2) Determine whether those with limited aberrant use share more similarities to likely OUD vs those using opioids as prescribed.

Design: Longitudinal observational cohort study.

Participants: National sample of Veterans categorized as having (1) likely OUD, (2) limited aberrant opioid use, or (3) prescribed, non-aberrant use based upon enhanced medical chart review.

Main measures: Comparison of sociodemographic and clinical factors at diagnosis and rates of age-adjusted mortality, non-fatal opioid overdose, and hospitalization after diagnosis. An exploratory machine learning analysis investigated how closely those with limited aberrant use resembled those with likely OUD.

Key results: Veterans (n = 483) were categorized as likely OUD (62.1%), limited aberrant use (17.8%), and prescribed, non-aberrant use (20.1%). Age, proportion experiencing homelessness, chronic pain, anxiety disorders, and non-opioid substance use disorders differed by group. All-cause mortality was high (44.2 per 1000 person-years (95% CI 33.9, 56.7)). Hospitalization rates per 1000 person-years were highest in the likely OUD group (831.5 (95% CI 771.0, 895.5)), compared to limited aberrant use (739.8 (95% CI 637.1, 854.4)) and prescribed, non-aberrant use (411.9 (95% CI 342.6, 490.4). The exploratory analysis reclassified 29.1% of those with limited aberrant use as having likely OUD with high confidence.

Conclusions: Veterans assigned opioid abuse/dependence ICD codes are heterogeneous and face variable long-term risks. Limited aberrant use confers increased risk compared to no aberrant use, and some may already have OUD. Findings warrant future investigation of this understudied population.

Keywords: chronic pain; long-term opioid therapy; opioid use disorder; risk.

Figure 1

Age-adjusted rates of outcomes per 1000 person-years. Individuals categorized as having likely OUD had the highest risk of all-cause mortality and hospitalization, followed by those with limited aberrant opioid use. Those categorized as having prescribed, non-aberrant opioid use had the lowest risk. Error bars represent 95% confidence intervals.

Figure 2

Estimated probability of individuals in the limited aberrant opioid use group (n = 86) belonging to the other two groups after model reassignment. The cutoff (probability = 0.5) indicates equal probability of reassignment to either group. Individuals assigned probabilities < 0.5 indicate model reassignment to the prescribed, non-aberrant opioid use group. Individuals assigned probabilities > 0.5 indicate model reassignment to the likely OUD group. Increasing distance from the cutoff indicates higher model confidence of class reassignment.