ABSTRACT
Introduction
Early/unplanned military separation in Active Component U.S. service members can result in reduced readiness during periods of high-tempo combat and increased demand for health care services within the Military Health System and Veterans Administration. Although current assessment tools leverage prescription data to determine deployment-limiting medication receipt and the need for interventions or waivers, there is a lack of understanding regarding opioid prescription patterns and subsequent early/unplanned military separation after return from deployment. As such, understanding these relationships could support future tool development and strategic resourcing. Therefore, the goal of the present study was to identify unique 12-month opioid prescription patterns and evaluate their relationship with early/unplanned military separation in Active Component service members who returned from deployment.
Materials and Methods
This retrospective, IRB-approved cohort study included data from 137,654 Active Component Army service members who returned from deployment between 2007 and 2013, received a post-deployment (index) opioid prescription, and had at least 1 year of Active Component service post-opioid initiation. A k-means clustering analysis identified clusters using opioid prescription frequency, median dose, median days supply, and prescription breaks (≥30 days) over the 12-month post-initiation (monitoring) period. A generalized additive model examined whether cluster membership and additional covariates were associated with early/unplanned separation.
Results
In addition to the single opioid prescription (38%), the cluster analysis identified five clusters: brief/moderate dose (25%), recurrent breaks (16%), brief/high dose (11%), long/few prescriptions (8%), and high prescription frequency (2%). In the generalized additive model, the probability of early/unplanned military separation was higher for the high prescription frequency cluster (74%), followed by recurrent breaks (45%), long/few prescriptions (37%), brief/moderate dose (30%), and brief/high dose (29%) clusters, relative to the single prescription (21%) cluster. The probability of early/unplanned separation was significantly higher for service members with documented substance use disorders, mental health conditions, or traumatic brain injuries during the monitoring periods. Service members assigned male were more likely to have an early/unplanned separation relative to service members assigned female. Latinx service members and service members whose race was listed as Other were less likely to experience early/unplanned separation relative to white service members. Relative to Junior Officers, Junior Enlisted and Senior Enlisted service members were more likely to experience early/unplanned separation, but Senior Officers were less likely.
Conclusions
Further evaluation to support the integration of longitudinal opioid prescription patterns into existing tools (e.g., a screening tool for deployment-limiting prescriptions) may enable more timely intervention and support service delivery to mitigate the probability and impact of early/unplanned separation.
INTRODUCTION
U.S. military service members have a high prevalence of acute and chronic pain and a higher burden of injuries sustained during training and deployment.1,2 To manage pain, clinical guidelines specify opioid-related thresholds that warrant caution including elevated morphine equivalents daily doses (e.g., daily milligram morphine equivalence (MME) ≥ 50 mg) and long-term opioid therapy (defined as “the use of opioids on most days for >3 months”)3,4 because of the probability of continued use5 and adverse events (e.g., overdose and misuse).6 Prescription opioids have been the subject of Defense Health Agency–Procedural Instructions7,8 and multiple DoD Inspector General reports.9,10 Yet, there is a lack of information identifying opioid prescription patterns and their relationships with early/unplanned military separation.
In addition to daily MME ≥ 50 mg daily doses and long-term opioid therapy, some opioid prescription patterns may be associated with adverse occupational outcomes in U.S. service members (e.g., early/unplanned military separation) but are not described explicitly in the Defense Health Agency–Procedural Instructions or clinical practice guidelines. For example, research indicates that cumulative opioid prescription dose, days supply, and prescription frequency are associated with an elevated risk for adverse events.11–13 A richer understanding of opioid prescription patterns and their relationships with unplanned/early military separation could enable clinicians to address unmet health care needs or ensure effective transition services, as well as support DoD and Veterans Administration (VA) leaders in determining future resource allocation and programming. For example, a Defense Health Agency tool enables health care personnel to screen service members for potentially deployment-limiting medications, identify the need for and provide medical intervention, and complete medical waivers if indicated.14 Such a tool could be adapted or extended to include historical patterns in opioid prescribing to flag when further assessment and potential intervention are needed to achieve readiness.
To address these gaps in knowledge, the overall goal of this study was to identify longitudinal opioid prescription patterns over a 12-month period following opioid initiation in Army service members who returned from Afghanistan/Iraq deployments and to examine whether such patterns were associated with early/unplanned military separation. We hypothesized that there will be several opioid prescription pattern clusters characterized by combinations of opioid parameters (e.g., more prescriptions with elevated doses). We further hypothesize that some opioid prescription clusters will be associated with increased risk for early/unplanned military separation because of medical, conduct, or performance reasons.
METHODS
The dataset for this study, the Substance Use and Psychological Injury Combat (SUPIC) study, is described in detail in previous publications.15 The present sample included Active Component Army Soldiers in the SUPIC database who returned from deployment between 2007 and 2013. The SUPIC database is a population-based dataset of all Army Soldiers who returned from an Afghanistan/Iraq deployment between October 2007 and July 2013. This time frame was selected for the SUPIC longitudinal study to capture Soldiers returning from deployments during a period of intense military operations during the Afghanistan/Iraq conflicts and to allow for the examination of health conditions that emerged and health services received during these critical reintegration periods. Study results capture the period of peak combat deployments during these conflicts16 to improve post-deployment health care service capabilities in future conflicts.
Index deployment data were obtained from the DoD’s Contingency Tracking System maintained by the Defense Manpower Data Center to identify the index Afghanistan/ Iraq deployment (i.e., the first deployment associated with the Afghanistan/Iraq conflicts ending within the study window). Deployments to non-contingency operations (i.e., peace-keeping missions) were not included in these data. The Contingency Tracking System includes the administrative dates of deployment start and end. Demographic information was obtained from the DoD’s Defense Enrollment Eligibility Records System. Additional data from the SUPIC study include all outpatient and inpatient health care from all ambulatory claims and admissions in military “direct care” settings and civilian “purchased care” settings covered by TRICARE, as well as pharmacy data from direct and purchased care from the Pharmacy Data Transaction Service. Separation data were provided by the Defense Manpower Data Center for separation dates occurring through September 2015.
Of the 479,423 Active Component service members who had at least one opioid prescription dispense record in the SUPIC database, 420,807 had an index opioid prescription (the first opioid prescription) dispensed after return from their index deployment. If there was more than one deployment, the first recorded deployment was selected as the index deployment. Index opioid prescriptions occurred from October 2007 to July 2013, thereby enabling each service member to have a 1-year opioid monitoring period (e.g., October 2008 to July 2014) and at least a year period in which they could receive an early/unplanned separation (e.g., November 2008 to September 2015). Therefore, service members whose index opioid prescription occurred after July 31, 2013 (n = 99,112) were excluded. Because deployment start and end dates are not always precise in administrative records, it was unclear whether service members with an opioid prescription occurring before their deployment end date received the prescription while deployed. To reduce variability because of unmeasured confounds, including the lack of prescription information from in-theater and while deployed, service members were excluded if they had an opioid prescription claim within 180 days before the index opioid prescription (e.g., before the deployment end date) and, therefore, were not opioid naïve at the time of the index opioid prescription (n = 9,867). Service members were also excluded if they were not Active Component for any month during the 12-month opioid monitoring period (n = 173,850). Service members with opioid prescription missing values or values (e.g., daily MME, days supply) >99th percentile were removed, given that such prescription values were likely errors (n = 291). Lastly, service members with a prior listed unplanned/early separation code (e.g., temporary disability retirement) recorded in their Defense Manpower Data Center data were also excluded (n = 33).
The initial opioid prescription dispense record occurring after the index deployment was selected, followed by all dispensed opioid prescriptions and ambulatory care claims within 1 year of the initial opioid prescription were selected (i.e., the “opioid monitoring period”). Opioid prescriptions included buprenorphine, butorphanol, codeine, dextropropoxyphene, dihydrocodeine, fentanyl, hydrocodone, hydromorphone, meperidine, methadone (excluding the liquid form), morphine, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol prescriptions.1 Opioid parameters included long- vs. short-acting type, days supply, quantity, and dispense date. Daily MME of each prescription was created by multiplying the quantity by the medication-specific morphine conversion factor as described by the CDC17 and dividing the prescription’s MME by the days supply. Opioid restarts were defined as a new opioid prescription dispensed >30 days after the end of the prior opioid prescription’s days supply.
Service information included the end of the index deployment year, the number of months served before the index deployment, the number of prior deployments, and rank at the end of the index deployment. Demographic information included age at time returned from deployment, assigned sex, race, and ethnicity from the enrollment record. Additional variables included the presence of mental health conditions, substance use disorders, or traumatic brain injury (e.g., per ICD-9 codes on ambulatory claims) documented during the opioid monitoring period. Lastly, the primary occupational outcome was the occurrence of an early/unplanned military separation because of medical reasons (e.g., disability with severance pay and permanent or temporary disability retirement), misconduct and poor performance (e.g., commission of a serious offense, drug use, court martial, unsatisfactory performance, and failure to meet minimum qualifications for retention), or death occurring after the end of the opioid monitoring period at any point. For sensitivity analyses, the primary outcome was divided into two separate outcomes: Presence of separation because of (1) medical reasons (not including death) and (2) misconduct or poor performance.
Analyses were completed in R. Supplementary Figure S1 provides an overview of the analytic periods and variables of interest. Descriptive statistics (median and interquartile ranges [IQR], frequencies, and proportions) described sample characteristics. Next, variables were constructed using the R package dplyr18 to obtain the total number of opioid prescriptions, the median days supply, the median daily MME, and the number of opioid prescription restarts that occurred within the 12-month opioid monitoring period.
Each patient’s total number of opioid prescriptions, median days supply, median daily MME, and number of opioid prescription restarts were centered and standardized. Then, the cluster R package19 was used to construct multifaceted k-means clusters using standardized values corresponding to the number of opioid prescriptions, median days supply, median daily MME, and number of opioid prescription restarts occurring during the 12-month opioid monitoring period. Only patient records with more than one documented opioid prescription were included in the cluster analysis. The different cluster solutions were compared using several fit statistics (e.g., within-cluster sum of squared error, percent of the variance explained, and Akaike information criterion) and descriptive evaluation. The descriptive evaluation was completed using visualizations built with the factoextra R package20 and descriptive statistics, such as cross-tabulations of cluster membership for different cluster solutions. If differences between a lower number and higher number cluster solution were attributed to the proportion of service members in the smallest cluster being divided into two or more smaller clusters, then the lower number cluster solution was selected.
Once the optimal number of clusters was selected, bivariate comparisons between clusters, and service members with only one documented opioid prescription who were not included in the cluster analysis (Single Prescription Group), were conducted using the tableone R package.21 Lastly, a multivariable generalized additive model (GAM) examined the primary outcome of early/unplanned military separation. Here, the Single Prescription Group was included as the reference cluster. GAM was selected because of the nonlinear relationships that may occur between continuous covariates and outcomes. In a GAM, covariates can be fixed or smooth terms, the latter of which uses splines to model nonlinear relationships. The GAM was completed using the mgcv R package.22 Covariates included factors that may be related to separation, such as age, race and ethnicity, rank, assigned sex, year of index opioid prescription, traumatic brain injury, substance use disorder, and mental health conditions. The ggeffects R package23 was used to extract marginal means (95% CI), and the ggplot2 R package24 was used to create visualizations. Sensitivity analyses examined separation because of disability and misconduct or poor performance, separately, in two additional GAMs.
RESULTS
Of the full sample of service members who had an index opioid prescription following return from their index deployment and met the study criteria (N = 137,654), more than half (n = 97,860; 62%) had at least one additional opioid prescription after opioid initiation and were included in the cluster analysis. The clusters were based on the number of opioid prescriptions, median days supply, median daily MME, and number of opioid prescription restarts during the 12-month opioid monitoring period. The within-cluster sum of squared error, percent of the variance explained, and Akaike information criterion improved as the number of clusters increased. However, there appeared to be a diminishing return on increasing the number of clusters after a five-cluster solution, as centroids were less differentiated. Therefore, the five-cluster solution was selected. Given the clustering process did not include service members who received a single opioid prescription in the 12-month opioid monitoring period (n = 51,170, 37%), the single-prescription cohort was added as an additional comparison group. The sample sizes of each cluster were 37,732 (brief/moderate dose, 25%), 11,170 (long/few prescriptions, 8%), 23,285 (recurrent breaks, 16%), 2,687 (high prescription frequency, 2%), and 15,466 (brief/high dose, 11%).
The mean (95% CI) opioid prescription parameters by cluster are visualized in Figure 1. The five clusters varied across the number of opioid prescriptions, median days supply, median daily MME, and number of opioid prescription restarts. The brief/moderate dose cluster was characterized by having the lowest values for all parameters, aside from a median daily MME, which was higher than the long/few prescriptions cluster. The long/few prescriptions cluster had the highest median days supply, the lowest MME doses, and the highest proportion of service members whose index prescription was tramadol. The recurrent breaks cluster was marked by the highest number of restarts and the second-highest number of prescriptions. The high prescription frequency cluster was characterized by a high number of opioid prescriptions, the highest proportion of any tramadol prescription or long-acting opioid prescription receipt, and the second-highest proportion of tramadol prescrition receipts at index prescription. The brief/high dose cluster was characterized by the highest median daily MME. Descriptive statistics for each cluster are reported in Table I, and between-cluster standardized mean differences are reported in Supplementary Table S1.
FIGURE 1.
Differences in opioid-related parameters between clusters. A = brief/moderate dose cluster, B = long/few prescriptions cluster, C = recurrent breaks cluster, D = high prescription frequency cluster, and E = brief/high dose cluster.
TABLE I.
Cluster Information and Overall SMDs
| Overall | Single prescription | Brief/moderate dose | Long/Few prescriptions | Recurrent breaks | High prescription frequency | Brief/high dose | P | SMD | |
|---|---|---|---|---|---|---|---|---|---|
| n (%) | 137,654 | 51,170 (37.2) | 34,732 (25.2) | 11,170 (8.1) | 22,429 (16.3) | 2,687 (2.0) | 15,466 (11.2) | ||
| Patient characteristics | |||||||||
| Rank, n (%) | <.001 | 0.16 | |||||||
| Junior Enlisted | 70,557 (51.3) | 24,450 (47.8) | 18,714 (53.9) | 5,476 (49.0) | 12,513 (55.8) | 1,656 (61.6) | 7,748 (50.1) | ||
| Senior Enlisted | 55,737 (40.5) | 21,475 (42.0) | 13,504 (38.9) | 4789 (42.9) | 8,600 (38.3) | 942 (35.1) | 6,427 (41.6) | ||
| Junior Officer | 5,891 (4.3) | 2,828 (5.5) | 1,281 (3.7) | 394 (3.5) | 673 (3.0) | 50 (1.9) | 665 (4.3) | ||
| Senior Officer | 2,988 (2.2) | 1,340 (2.6) | 677 (1.9) | 287 (2.6) | 322 (1.4) | 19 (0.7) | 343 (2.2) | ||
| Warrant Officer | 2,481 (1.8) | 1,077 (2.1) | 556 (1.6) | 224 (2.0) | 321 (1.4) | 20 (0.7) | 283 (1.8) | ||
| Assigned male, n (%) | 120,825 (87.8) | 45,628 (89.2) | 30,305 (87.3) | 9,884 (88.5) | 18,804 (83.8) | 2,538 (94.5) | 13,666 (88.4) | <.001 | 0.13 |
| Race and ethnicity, n (%) | <.001 | 0.17 | |||||||
| White | 77,657 (56.4) | 28,214 (55.1) | 19,703 (56.7) | 6,351 (56.9) | 12,638 (56.3) | 1,878 (69.9) | 8,873 (57.4) | ||
| American Indian and Alaska Native | 1,352 (1.0) | 505 (1.0) | 321 (0.9) | 130 (1.2) | 214 (1.0) | 31 (1.2) | 151 (1.0) | ||
| Asian and Pacific Islander | 16,740 (12.2) | 6,226 (12.2) | 4,095 (11.8) | 1,377 (12.3) | 2,770 (12.4) | 426 (15.9) | 1,846 (11.9) | ||
| Black | 25,291 (18.4) | 9,794 (19.1) | 6,436 (18.5) | 1,974 (17.7) | 4,196 (18.7) | 165 (6.1) | 2,726 (17.6) | ||
| Latinx/e | 14,917 (10.8) | 5,749 (11.2) | 3,772 (10.9) | 1,199 (10.7) | 2,329 (10.4) | 169 (6.3) | 1,699 (11.0) | ||
| Other | 1,697 (1.2) | 682 (1.3) | 405 (1.2) | 139 (1.2) | 282 (1.3) | 18 (0.7) | 171 (1.1) | ||
| Age, median (IQR) | 26 (23, 33) | 26 (23, 32) | 26 (22, 32) | 27 (23, 34) | 27 (23, 33) | 26 (23, 30) | 26 (23, 33) | <.001 | 0.09 |
| Return year, n (%) | <.001 | 0.11 | |||||||
| 2007 | 7,614 (5.5) | 2,608 (5.1) | 1,742 (5.0) | 765 (6.8) | 1329 (5.9) | 220 (8.2) | 950 (6.1) | ||
| 2008 | 28,392 (20.6) | 10,747 (21.0) | 6,835 (19.7) | 2,641 (23.6) | 4,435 (19.8) | 601 (22.4) | 3,133 (20.3) | ||
| 2009 | 40,620 (29.5) | 15,249 (29.8) | 10,434 (30.0) | 3,045 (27.3) | 6,564 (29.3) | 803 (29.9) | 4,525 (29.3) | ||
| 2010 | 32,114 (23.3) | 11,802 (23.1) | 7,885 (22.7) | 2,816 (25.2) | 5,351 (23.9) | 601 (22.4) | 3,659 (23.7) | ||
| 2011 | 19,649 (14.3) | 7,229 (14.1) | 5,358 (15.4) | 1,291 (11.6) | 3,334 (14.9) | 325 (12.1) | 2,112 (13.7) | ||
| 2012 | 8,100 (5.9) | 3,046 (6.0) | 2,168 (6.2) | 537 (4.8) | 1,238 (5.5) | 126 (4.7) | 985 (6.4) | ||
| 2013 | 1,165 (0.8) | 489 (1.0) | 310 (0.9) | 75 (0.7) | 178 (0.8) | 11 (0.4) | 102 (0.7) | ||
| Remained in Active Component ≥2 years post-index opioid, n (%) | 102,624 (74.6) | 40,657 (79.5) | 25,984 (74.8) | 7,920 (70.9) | 15,106 (67.4) | 1,458 (54.3) | 11,499 (74.4) | <.001 | 0.22 |
| Opioid-related information | |||||||||
| Any tramadol prescription, n (%) | 42,789 (31.1) | 7,406 (14.5) | 11,497 (33.1) | 5,651 (50.6) | 13,002 (58.0) | 2,026 (75.4) | 3,207 (20.7) | <.001 | 0.66 |
| Tramadol at index, n (%) | 24,165 (17.6) | 7,406 (14.5) | 5,785 (16.7) | 3,409 (30.5) | 5,238 (23.4) | 757 (28.2) | 1,570 (10.2) | <.001 | 0.25 |
| Any long-acting prescription, n (%) | 4,973 (3.6) | 42 (0.1) | 970 (2.8) | 670 (6.0) | 1,320 (5.9) | 1,194 (44.4) | 777 (5.0) | <.001 | 0.48 |
| Total days supply, median (IQR) | 10 (5, 32) | 4 (3, 6) | 12 (8, 23) | 54 (34, 97) | 37 (19, 64) | 268 (181, 392) | 10 (6, 18) | <.001 | 1.40 |
| Median days between prescriptions, median (IQR) | 4.5 (0, 64) | 0 (0, 0) | 46 (6, 153) | 27 (1, 115.9) | 45 (21, 86) | 2 (0, 4.5) | 37 (6, 134.5) | <.001 | 0.86 |
| Median prescription days supply, median (IQR) | 4.50 (3, 7) | 4 (3, 6) | 4 (3, 6) | 19 (17, 30) | 5 (3, 7) | 8 (5, 14) | 4 (3, 5) | <.001 | 1.27 |
| Median morphine equivalent dose, median (IQR) | 150 (100, 225) | 135 (75, 180) | 125 (90, 175) | 180 (113, 300) | 150 (110, 225) | 263 (179, 420) | 225 (150, 300) | <.001 | 0.49 |
| Median morphine equivalent daily dose, median (IQR) | 33 (23, 47) | 31.25 (20, 50) | 31.07 (25, 38) | 16 (10, 24) | 32 (25, 41) | 33 (29, 45) | 61 (54, 75) | <.001 | 1.15 |
| Days from deployment end to index prescription, median (IQR) | 233 (105, 486) | 285 (126, 560) | 234 (107, 487) | 209 (96, 439) | 173 (81, 373) | 136 (56, 331) | 221 (108, 460) | <.001 | 0.21 |
| Number of prescriptions, median (IQR) | 2 (1, 4) | 1 (1, 1) | 2 (2, 3) | 3 (2, 5) | 5 (4, 8) | 22 (19, 29) | 3 (2, 4) | <.001 | 1.70 |
| Number of prescription breaks >30 days, median (IQR) | 1 (0, 1) | 0 (0, 0) | 1 (0, 1) | 1 (0, 1) | 2 (2, 3) | 1 (1, 2) | 1 (0, 1) | <.001 | 1.60 |
| Additional covariates and outcomes | |||||||||
| Traumatic brain injury, n (%) | 22,079 (16.0) | 6,066 (11.9) | 5,617 (16.2) | 2,024 (18.1) | 4,790 (21.4) | 1,019 (37.9) | 2,563 (16.6) | <.001 | 0.24 |
| Substance use disorder, n (%) | 57,993 (42.1) | 18,355 (35.9) | 15,146 (43.6) | 4,846 (43.4) | 11,239 (50.1) | 1,770 (65.9) | 6,637 (42.9) | <.001 | 0.24 |
| Mental health condition, n (%) | 56,489 (41.0) | 17,272 (33.8) | 14,303 (41.2) | 5,167 (46.3) | 11,729 (52.3) | 1,894 (70.5) | 6,124 (39.6) | <.001 | 0.32 |
| Separation, n (%) | 42,204 (30.7) | 10,962 (21.4) | 10,406 (30.0) | 4,194 (37.5) | 10,048 (44.8) | 2,015 (75.0) | 4,579 (29.6) | <.001 | 0.48 |
| Disability-related separation, n (%) | 28,379 (20.6) | 6,171 (12.1) | 6,556 (18.9) | 3,122 (27.9) | 7,707 (34.4) | 1,680 (62.5) | 3,143 (20.3) | <.001 | 0.47 |
| Separation because of misconduct or poor performance, n (%) | 13,380 (9.7) | 4,657 (9.1) | 3,749 (10.8) | 1,032 (9.2) | 2,252 (10.0) | 309 (11.5) | 1,381 (8.9) | <.001 | 0.04 |
Abbreviation: SMD = standardized mean difference.
All prescriptions refer to opioid prescriptions.
Results of the GAM are reported in Table II. Relative to service members who received a single opioid prescription, the probability of early/unplanned separation was significantly higher for each cluster. As indicated in Figure 2, the mean (95% CI) probability of early/unplanned separation was the highest for the high prescription frequency cluster, followed by the recurrent breaks cluster and the long/few prescriptions cluster. In sensitivity analyses, this pattern remained consistent for separation because of disability, but not misconduct or poor performance. Instead, the probability of separation because of misconduct or poor performance was lower in all clusters, aside from the brief/moderate dose cluster, relative to the service members who received a single prescription.
TABLE II.
Results From the Main GAM and Two Sensitivity GAMs Predicting Separation
| All reasons | Disability | Misconduct and poor performance | ||||
|---|---|---|---|---|---|---|
| Fixed term | Estimate (95% CI) | P | Estimate (95% CI) | P | Estimate (95% CI) | P |
| Cluster (Ref: Single prescription) | ||||||
| Brief/moderate dose | 0.32 (0.29, 0.35) | <.001 | 0.41 (0.37, 0.45) | <.001 | 0.07 (0.02, 0.12) | <.001 |
| Long/few prescriptions | 0.65 (0.60, 0.70) | <.001 | 0.87 (0.82, 0.92) | <.001 | −0.08 (−0.15, −0.01) | .04 |
| Recurrent breaks | 0.86 (0.82, 0.9) | <.001 | 1.09 (1.05, 1.13) | <.001 | −0.07 (−0.13, −0.01) | .01 |
| High prescription frequency | 1.86 (1.76, 1.96) | <.001 | 1.95 (1.86, 2.04) | <.001 | −0.17 (−0.3, −0.04) | .01 |
| Brief/high dose | 0.33 (0.29, 0.37) | <.001 | 0.52 (0.47, 0.57) | <.001 | −0.10 (−0.17, −0.03) | <.001 |
| Deployment Return year | −0.19 (−0.20, −0.18) | <.001 | −0.15 (−0.16, −0.14) | <.001 | −0.15 (−0.16, −0.14) | <.001 |
| Rank (Ref: Junior Officer) | ||||||
| Junior Enlisted | 1.24 (1.16, 1.32) | <.001 | 1.24 (1.14, 1.34) | <.001 | 0.75 (0.61, 0.89) | <.001 |
| Senior Enlisted | 0.61 (0.53, 0.69) | <.001 | 0.83 (0.73, 0.93) | <.001 | 0.01 (−0.13, 0.15) | .88 |
| Senior Officer | −0.93 (−1.13, −0.73) | <.001 | −0.83 (−1.05, −0.61) | <.001 | −1.10 (−1.60, −0.60) | <.001 |
| Warrant Officer | −0.08 (−0.23, 0.07) | .29 | 0.02 (−0.15, 0.19) | .79 | −0.20 (−0.50, 0.10) | .18 |
| Assigned male (Ref: Female) | 0.47 (0.43, 0.51) | <.001 | 0.17 (0.13, 0.21) | <.001 | 0.81 (0.74, 0.88) | <.001 |
| Race and ethnicity (Ref: white) | ||||||
| American Indian and Alaska Native | −0.01 (−0.14, 0.12) | .85 | −0.18 (−0.32, −0.04) | .02 | 0.24 (0.07, 0.41) | .01 |
| Asian and Pacific Islander | −0.01 (−0.05, 0.03) | .78 | −0.03 (−0.07, 0.01) | .18 | 0.04 (−0.02, 0.10) | .22 |
| Black | −0.02 (−0.05, 0.01) | .24 | −0.25 (−0.29, −0.21) | <.001 | 0.38 (0.33, 0.43) | <.001 |
| Latinx | −0.08 (−0.12, −0.04) | <.001 | −0.13 (−0.18, −0.08) | <.001 | 0.05 (−0.01, 0.11) | .13 |
| Other | −0.23 (−0.36, −0.10) | <.001 | −0.32 (−0.46, −0.18) | <.001 | 0.01 (−0.24, 0.26) | .96 |
| Traumatic brain injury | 0.39 (0.36, 0.42) | <.001 | 0.52 (0.49, 0.55) | <.001 | −0.18 (−0.23, −0.13) | <.001 |
| Substance use disorder | 0.30 (0.27, 0.33) | <.001 | 0.09 (0.06, 0.12) | <.001 | 0.47 (0.43, 0.51) | <.001 |
| Mental health condition | 0.83 (0.80, 0.86) | <.001 | 0.81 (0.78, 0.84) | <.001 | 0.41 (0.37, 0.45) | <.001 |
Abbreviations: GAM = generalized additive model; Ref = reference group
The smooth term for age was also included and was significant (all P < .001) across all three models.
FIGURE 2.
The probability of separation for all reasons, disability, and misconduct or poor performance across each cluster per generalized additive models. Single = single opioid prescription, A = brief/moderate dose cluster, B = long/few prescriptions cluster, C = recurrent breaks cluster, D = high prescription frequency cluster, and E = brief/high dose cluster.
There were several other covariates significantly associated with the probability of early/unplanned separation. As indicated in Supplementary Figure S2, the mean (95%) probability of early/unplanned separation was significantly higher for service members with documented substance use disorders, mental health conditions, or traumatic brain injuries during the monitoring. Service members assigned male were more likely to experience early/unplanned separation, relative to service members assigned female. Latinx service members and service members whose race was listed as Other were less likely to experience early/unplanned separation relative to white service members. Relative to Junior Officers, Junior Enlisted and Senior Enlisted service members were more likely to experience early/unplanned separation, but Senior Officers were less likely. Lastly, likely correlated with rank, age (not shown in Supplementary Figure S2) was significantly associated with the probability of early/unplanned separation. The pattern was such that the probability increased from ages 18 to 28 years and then decreased after age 29 years.
DISCUSSION
In the present analyses, there were five distinct clusters of opioid prescription patterns, as indicated by the cluster analysis, in addition to the group of service members who received a single opioid prescription (38% of the sample). Clusters with elevated risk of separation were those in which service members were more likely to receive a median days supply ≥5 days per prescription, a total opioid days supply >23 days, and ≥3 opioid prescriptions within a 12-month period. Additionally, clusters with any tramadol or long-acting opioid prescription had an elevated probability of early/unplanned separation. Per sensitivity analyses, the relationships between cluster membership and early/unplanned separation may only be applicable for service members who left for disability reasons, not misconduct or poor performance.
Current Defense Health Agency tools enable prescribers and health care personnel to screen service members for deployment-limiting prescriptions14 and evaluate their own opioid prescribing practices (e.g., Opioid Prescriber Monthly Trend Report).7 However, the screening tool does not include historical opioid prescription patterns (e.g., those occurring over the course of a year, without an active prescription in the past month) that are indicative of further assessment and resource needs; the Opioid Prescriber Monthly Trend Report is limited to opioid and naloxone prescription frequencies, including prescriptions to patients with elevated MME, receiving long-term opioid therapy, and/or co-prescribed a benzodiazepine. Given the present findings included a post-deployment sample, future health services research could be leveraged by the Defense Health Agency and the Services to enhance the scope of prescription-related screening tools and deliver proactive interventions and transition support.
Service members whose race was Other, Latinx service members, and those assigned female had a lower probability of early/unplanned separation, relative to white service members and those assigned male, respectively. In sensitivity analyses, Black patients (−19%), Latinx (−10%), and service members whose race was Other (−24%) were less likely to receive a disability-related separation, relative to white service members. However, Black (+43%) and American Indian and Alaska Native (+27%) service members were more likely to experience a separation because of misconduct or poor performance, relative to white service members. Such findings are consistent with military investigations identifying systemic racism in military justice, discipline, and administrative discharges.25,26 Reports released by the U.S. Government Accountability Office have identified racialized inequities in VA disability claims approval.27 However, it is unclear the extent to which disability-related processes and determinations are inequitable within the U.S. Military Health System, as service members may receive disability ratings from the DoD, VA, or both. Future work, similar to findings by the Air Force and Government Accountability Office, is required to ensure equitable access to disability evaluations and benefits.
Rank and documented medical conditions were also associated with the probability of early/unplanned separation. The probability of early/unplanned separation was the highest in Junior Enlisted service members, followed by Senior Enlisted, Junior and Warrant Officers, and Senior Officers. Although traumatic brain injury and substance use disorders were each separately associated with a ∼5% greater probability of early/unplanned separation, service members with mental health conditions were almost twice as likely (∼28%) to have an early/unplanned separation relative to those without such conditions (∼15%). Taken together, any future development of screening tools must account for data biases resulting from such inequities. Without doing so, any screening tool could compound systemic racism, sexism, and classism.28,29
What remains unclear is whether receipt or dose of additional non-opioid pharmacological and non-pharmacological therapies could alter subsequent cluster memberships and early/unplanned separation. Although we did not examine pain diagnoses or pain experience, prescription opioids are routinely used to treat acute and chronic pain in service members.2 Engagement in non-pharmacological therapies, as well as their timing, could mitigate the likelihood of negative opioid-related or other adverse outcomes in service members prescribed opioids.15,30 However, less is known regarding the dose-dependent relationships of both opioids and non-opioid therapies, or unique combinations of pain management treatments, as they relate to outcomes. Similarly, it is unclear the potential role and utility of tramadol. Tramadol prescribing is not specifically described in clinical practice guidelines, but almost a third of the sample received at least one tramadol prescription and 18% received tramadol as their index opioid prescription. Although tramadol has a much lower MME than other opioid medications, evidence indicates that it may not provide superior pain management, relative to non-opioid pain medications.31,32
The present analysis was limited in its scope and analytic approach; results are by no means ready for adoption within existing or new screening tools. The observational nature and limits of the electronic health record data mean that many factors that could have been associated with both opioid prescription receipt and outcomes were not included in our models. We did not evaluate the role of underlying pain conditions, (poly)traumatic injuries, or pain management interventions, both pharmacological and non-pharmacological. Therefore, it is unclear the extent to which cluster memberships and their relationships early/unplanned separation could alter as a function of care receipt; this gap is a priority for future study. The data include health care claims data processed by TRICARE. Therefore, opioid prescriptions and diagnoses provided by health care providers outside the Military Health System that were paid for out-of-pocket or recreational opioid use were not captured in this study. The k-means clustering approach and the narrow sample inclusion (e.g., Active Component Soldiers after a deployment to Iraq or Afghanistan with at least 1 year of service after initial opioid prescription) limit the generalizability of the study. Given the reduction in combat deployments, it is unclear whether such clusters would replicate in the full and current Active Component. Despite these limitations, to our knowledge this is the first study to characterize multiple parameters of opioid prescription, which may be associated with early/unplanned separation. Future studies would benefit from expanding evaluation to the full Active Component, their entire medical records, and Periodic Health Assessments, as well as including additional potentially deployment-limiting prescriptions (e.g., psychotropic medications) and health care patterns within clustering to create optimal screening and support tools.
Supplementary Material
ACKNOWLEDGMENTS
None declared.
Contributor Information
Vi T Nghiem, Department of Anesthesiology, Brooke Army Medical Center, Fort Sam Houston, TX 78234, USA.
Mary Jo Larson, Heller School for Social Policy and Management, Institute for Behavioral Health, Brandeis University, Waltham, MA 02453, USA.
Rachel Sayko Adams, Heller School for Social Policy and Management, Institute for Behavioral Health, Brandeis University, Waltham, MA 02453, USA; Department of Health Law, Policy & Management, Boston University School of Public Health, Boston, MA 02118, USA; Veterans Health Administration, Rocky Mountain Mental Illness Research Education and Clinical Center, Aurora, CO 80045, USA.
Natalie Moresco, Heller School for Social Policy and Management, Institute for Behavioral Health, Brandeis University, Waltham, MA 02453, USA.
Krista B Highland, Department of Anesthesiology, Uniformed Services University, Bethesda, MD 201814, USA.
SUPPLEMENTARY MATERIAL
SUPPLEMENTARY MATERIAL is available at Military Medicine online.
FUNDING
This study was supported by funding from the National Center for Complementary and Integrative Health (R01AT008404). Funding to support cohort development and measure specification was from the National Institute on Drug Abuse (R01DA030150) and the National Institute of Mental Health (R01MH120122).
CONFLICT OF INTEREST STATEMENT
The authors do not have relevant conflicts of interest to report.
DATA AVAILABILITY
The data are sourced from the DoD and cannot be made available directly to the public or individual researchers by the authors. Any request for data would need to go through the appropriate procedures (e.g., data sharing agreements) and authorities (Defense Health Agency).
CLINICAL TRIAL REGISTRATION
Not applicable.
INSTITUTIONAL REVIEW BOARD (HUMAN SUBJECTS)
This retrospective longitudinal cohort study was approved by the Brandeis University (IRB# 14153) and USU IRBs.
INSTITUTIONAL ANIMAL CARE AND USE COMMITTEE (IACUC)
Not applicable.
INSTITUTIONAL CLEARANCE
Institutional clearance approved.
INDIVIDUAL AUTHOR CONTRIBUTION STATEMENT
Each author provided substantial contributions to the conception and design of the study, literature search, the interpretation of the data, draft of this article and its revisions, and final approval of the submitted version. Additionally, R.S.A. and M.J.L. contributed to the data acquisition, N.M. contributed to the data management, and K.B.H. conducted the analyses.
REFERENCES
- 1. Adams RS, Thomas CP, Ritter GA, et al. : Predictors of postdeployment prescription opioid receipt and long-term prescription opioid utilization among army active duty soldiers. Mil Med 2019; 184(1–2): e101–9.doi: 10.1093/milmed/usy162. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Hepner KA, Roth CP, Sherry TB, McBain RK, Ruder T, Engel CC: Assessing the quality of outpatient pain care and opioid prescribing in the Military Health System. Rand Health Q 2022; 9(4): 59–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Dowell D, Haegerich TM, Chou R: CDC guideline for prescribing opioids for chronic pain—United States, 2016. JAMA 2016; 315(15): 1624–45.doi: 10.1001/jama.2016.1464. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. US Department of Veterans Affairs , Department of Defense. VA/DoD clinical practice guidelines for management of opioid therapy for chronic pain. Available at https://www.healthquality.va.gov/guidelines/Pain/cot/VADoDOTCPG022717.pdf; accessed May 15, 2018.
- 5. Deyo RA, Hallvik SE, Hildebran C, et al. : Association of prescription drug monitoring program use with opioid prescribing and health outcomes: a comparison of program users and nonusers. J Pain 2017; 19(2): 166–77.doi: 10.1016/j.jpain.2017.10.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Chou R, Deyo R, Devine B, et al. : The effectiveness and risks of long-term opioid treatment of chronic pain. Evid Rep Technol Assess (Full Rep) 2014; 2014(218): 1–219.doi: 10.23970/ahrqepcerta218. [DOI] [PubMed] [Google Scholar]
- 7. Defense Health Agency. Procedural Instruction 6025.08: Pain management and opioid safety in the Military Health System (MHS). 2023. Available at https://www.health.mil/Reference-Center/DHA-Publications/2023/02/08/DHA-AI-6025-08; accessed March 13, 2023.
- 8. Defense Health Agency. Procedural Instruction 6025.07: Naloxone in the MTFs. 2018. Available at https://www.health.mil/-/media/Files/MHS/Policy-Files/SIGNED--DHAPI-602507--Naloxone-in-the-MTFs.ashx; accessed March 13, 2023.
- 9. Report No . DODIG-2020-048 audit of controls over opioid prescriptions at selected DoD military treatment facilities (Dodig-2020-048). US Federal Government. 2020.
- 10. Department of Defense Office of Inspector General . Evaluation of the DoD’s management of opioid use disorder for military health system beneficiaries DODIG-2019-091. US Federal Government. 2019.
- 11. Mack KA, Zhang K, Paulozzi L, Jones C: Prescription practices involving opioid analgesics among Americans with Medicaid, 2010. J Health Care Poor Underserved 2015; 26(1): 182–98.doi: 10.1353/hpu.2015.0009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Shah A, Hayes CJ, Martin BC: Factors influencing long-term opioid use among opioid naive patients: an examination of initial prescription characteristics and pain etiologies. J Pain 2017; 18(11): 1374–83.doi: 10.1016/j.jpain.2017.06.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Gold LS, Hansen RN, Avins AL, et al. : Associations of early opioid use with patient-reported outcomes and health care utilization among older adults with low back pain. Clin J Pain 2018; 34(4): 297–305.doi: 10.1097/ajp.0000000000000557. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Cha J, Filan E, Stapolsky G, et al. : A multidisciplinary approach to screen deployment-limiting health conditions. Mil Med 2021; 188(3–4): 653–7.doi: 10.1093/milmed/usab502. [DOI] [PubMed] [Google Scholar]
- 15. Larson MJ, Adams RS, Ritter GA, et al. : Associations of early treatments for low-back pain with military readiness outcomes. J Altern Complementary Med 2018; 24(7): 666–76.doi: 10.1089/acm.2017.0290. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Peters HM, Plagakis S: Department of Defense Contractor and Troop Levels in Afghanistan and Iraq: 2007-2018. Washington, DC: Congressional Research Service; 2019.
- 17. Prescription Drug Monitoring Program Technical Assistance Center . Technical Assistance Guide No. 01-13: calculating daily morphine milligram equivalents. Available at http://www.pdmpassist.org/pdf/BJA_performance_measure_aid_MME_conversion.pdf; accessed April 23, 2014.
- 18. Wickham H, Averick M, Bryan J, et al. : Welcome to the Tidyverse. J Open Source Softw 2019; 4(43): 1–6. doi: 10.21105/joss.01686. [DOI] [Google Scholar]
- 19. Maechler M, Rousseeuw P, Struyf A, Hubert M: cluster: “Finding Groups in Data”: Cluster Analysis Extended. 2022. Available at https://cran.r-project.org/web/packages/cluster/index.html; accessed March 13, 2023.
- 20. Kassambara A, Mundt F: Factoextra R Package: Easy Multivariate Data Analyses and Elegant Visualization. 2020. Available at https://cran.r-project.org/web/packages/factoextra/index.html; accessed May 16, 2023.
- 21. Panos A, Mavridis D: Tableone: an online web application and R package for summarising and visualising data. Evid Based Ment Health 2020; 23(3): 127–30.doi: 10.1136/ebmental-2020-300162. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Pedersen EJ, Miller DL, Simpson GL, Ross N: Hierarchical generalized additive models in ecology: an introduction with mgcv. PeerJ 2019; 7: e6876–6.doi: 10.7717/peerj.6876. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Ludecke D: ggeffects: Tidy data frames of marginal effects from regression models. J Open Source Softw 2018; 3(26): 1–5.doi: 10.21105/joss.00772. [DOI] [Google Scholar]
- 24. Wickham H: ggplot2: Elegent Graphics for Data Analysis. Springer-Verlag; 2016. [Google Scholar]
- 25. Department of The Air Force. Report of inquiry (S8918p): disparity review. US Federal Government. Washington, DC. 2021. Available at https://www.af.mil/Portals/1/documents/2021SAF/09_Sept/DR_ROI_Baseline.pdf; accessed March 15, 2023.
- 26. Department of The Air Force. Report of inquiry (S8918p): disparity review. US Federal Government. Washington, DC. 2020. Available at https://www.airandspaceforces.com/app/uploads/2020/12/IRDR.pdf; accessed March 15, 2023.
- 27. US Government Accountability Office. VA disability benefits: actions needed to further examine racial and ethnic disparities in compensation. US Federal Government. 2023. Available at https://www.gao.gov/products/gao-23-106097; accessed July 29, 2023.
- 28. Obermeyer Z, Powers B, Vogeli C, Mullainathan S: Dissecting racial bias in an algorithm used to manage the health of populations. Science 2019; 366(6464): 447–53.doi: 10.1126/science.aax2342. [DOI] [PubMed] [Google Scholar]
- 29. Wang HE, Landers M, Adams R, et al. : A bias evaluation checklist for predictive models and its pilot application for 30-day hospital readmission models. J Am Med Inform Assoc 2022; 29(8): 1323–33.doi: 10.1093/jamia/ocac065. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Meerwijk EL, Larson MJ, Schmidt EM, et al. : Nonpharmacological treatment of army service members with chronic pain is associated with fewer adverse outcomes after transition to the Veterans Health Administration. J Gen Intern Med 2020; 35(3): 775–83.doi: 10.1007/s11606-019-05450-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Schiphorst Preuper HR, Geertzen JH, van Wijhe M, et al. : Do analgesics improve functioning in patients with chronic low back pain? An explorative triple-blinded RCT. Eur Spine J 2014; 23(4): 800–6.doi: 10.1007/s00586-014-3229-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Chung JW, Zeng Y, Wong TK: Drug therapy for the treatment of chronic nonspecific low back pain: systematic review and meta-analysis. Pain Physician 2013; 16(6): E685–704.doi: 10.36076/ppj.2013/16/E685. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data are sourced from the DoD and cannot be made available directly to the public or individual researchers by the authors. Any request for data would need to go through the appropriate procedures (e.g., data sharing agreements) and authorities (Defense Health Agency).