Stated Pain Levels, Opioid Prescription Volume, and Chronic Opioid Use Among United States Army Soldiers

Stated Pain Levels, Opioid Prescription Volume, and Chronic Opioid Use Among United States Army... Abstract Introduction The use of opioids has increased drastically over the past few years and decades. As a result, concerns have mounted over serious outcomes associated with chronic opioid use (COU), including dependency and death. A greater understanding of the factors that are associated with COU will be critical if prescribers are to navigate potentially competing objectives to provide compassionate care, while reducing the overall opioid use problem. In this study, we study pain levels and opioid prescription volumes and their effects on the risk of COU. This study leveraged passive data sources that support automated decision support systems (DSSs) currently employed in a large military population. The models presented compute monthly, person-specific, adjusted probability of subsequent COT and could potentially provide critical decision support for clinicians engaged in pain management. Materials and Methods The study population included all outpatient presentations at military medical facilities worldwide among active duty United States Army soldiers during July 2011 to September 2014 (17,664,006 encounters; population N = 552,193). We conducted a retrospective cohort study of this population and employed longitudinal data and a discrete time multivariable logistic regression model to compute COT probability scores. The contribution of pain scores and opioid prescription quantities to the probability of COT represented analytic foci. Results There were 13,891 subjects (2.5%) who experienced incident COT during the observed time period. Statistically significant interactions between pain scores and prescription quantity were present, in addition to effects of multiple other control variables. Counts of monthly opioid prescriptions and maximum stated pain scores per month were each positively associated with COT. A wide range in individual COT risk scores was evident. The effect of prescription volume on the COT risk was larger than the effect of the pain score, and the combined effect of larger pain scores and increased prescription quantity was moderated by the interaction term. Conclusions The results verified that passive data on the US Army can support a robust COT risk computation in this population. The individual, adjusted risk level requires statistical analyses to be fully understood. Because the same data sources drive current military DSSs, this work provides the potential basis for new, evidence-based decision support resources for military clinicians. The strong, independent impact of increasing opioid prescription counts on the COT risk reinforces the importance of exploring alternatives to opioids in pain management planning. It suggests that changing provider behavior through enhanced decision support could help reduce COT rates. INTRODUCTION The perspective of the medical community on opioids has changed substantially over the last several decades. Before the mid-1990s, opioid use was largely seen as appropriate in severe pain and serious cases of chronic pain.1 However, the use of agents such as oxycodone then accelerated, increasing up to eightfold over the next 15 yr.2 As a result, concerns have mounted over serious outcomes associated with chronic opioid use (COU), including dependency and death.3 Leaders today have emphasized that prescribers should again exercise restraint in beginning or continuing opioid therapy in the absence of serious indications for their use.4–6 In this new era, clinicians must therefore employ great selectivity when considering opioids to ensure that pain control is achieved, while considering potential adverse outcomes of therapy. A greater understanding of the factors that are associated with COU will be critical if prescribers are to navigate potentially competing objectives to provide compassionate and quality care, while reducing the overall opioid use problem.7 One goal of new research might be to quantify the evidence-based probability of a patient continuing into COU after an initial opioid prescription, based on the available data at those points in time. Because such patients will have no opioid use history, the likelihood of COU and its attendant problems may be difficult to ascertain. An evidence-based COU probability computation might enhance the personalized approach taken when considering each such new opioid prescription. There are at least two main areas that might be quantified and leveraged with respect to the risk of COU: the patient’s needs and the provider’s response. With respect to patient needs, his or her stated pain score is the main numeric indicator of the pain burden. However, this value is not a major factor in existing practice guidelines on opioids. Research on the impact of pain scores on opioid-related outcomes has been inconsistent.8,9 In general, related research has mostly focused on the accuracy of the pain score and changes thereof, especially in the nursing research literature.10–12 Regarding the clinician’s response to the patient’s stated pain, key metrics include whether opioids are prescribed, and if so, the strength and duration of the therapy provided. Recent research suggests that clinicians may vary widely in their opioid-prescribing practices and that the probability of long-term opioid use is increased among patients receiving opioids from providers who prescribe large quantities of these agents.13,14 These findings suggest that provider behavior is a non-trivial contributor to whether COU occurs, and this factor requires further study to determine its impact. The broad goals of this study were to provide evidence that might inform pain management choices and to create methods that might drive clinical decision support tools. Its specific aim was to compute the contribution of stated pain scores and opioid prescription parameters on the probability of COU using data commonly found in electronic records. The work was conducted using a retrospective, longitudinal data analysis of the total active duty US Army, a population in which prior research using similar data has produced decision support systems for military clinicians.15,16 As in the total US population, opioids represent an important concern for public health in the US military. Over 15% of combat-deployed US soldiers indicated, they had engaged in past-month opioid use in one 2014 study.17 The same study indicated that greater than 40% of these individuals stated mild pain or had no current pain. Patients with opioid abuse disorders constitute a subset of those engaged in COU. The rate of these disorders is 0.7% among US persons over 12 yr of age, yet 11% of American military service members state that they have misused prescription medications when surveyed.18 METHODS Study Population We conducted a retrospective, longitudinal cohort study of combined administrative and medical data on the total, active duty US Army. Non-military persons, National Guard and Reserve soldiers, and individuals serving in other US military branches such as the Navy were not available. The data included military duty status and demographic data as well as detailed clinical elements. A broad array of data sources was combined in order to generate as complete picture as possible. The data sources leveraged included the following: Defense Manpower Data Center (DMDC) Master File: Demographic and military service data DMDC Transaction File: Duty status changes, including discharges from service Military Health System Data Repository (MDR) Combined Ambulatory Professional Encounter Record: Details of outpatient care, including diagnoses MDR Clinical Data Repository Vitals File: Pain scores MDR Pharmacy Detail Transaction Service: Details of dispensed medications The dataset included all outpatient encounters by all active Army soldiers during January 2011 to September 2014 at military treatment facilities worldwide, per electronic health records. The data were condensed into a longitudinal, person-month format in order to support descriptive analyses and time-to-event regression modeling. We selected initially opioid-naïve subjects for study in order to avoid beginning observation during potential pain management episodes. For each eligible member, we defined the index observation month as the month within which the subject had an outpatient clinical presentation for any reason, after at least 6 mo without opioid prescriptions. As the data began in January 2011, the earliest possible index observation month was July 2011. Subsequent, eligible observations were person-months in which outpatient care was received at military treatment facilities. Observation for COU continued until either the occurrence of an outcome (three subsequent months of opioid prescriptions) or when less than 3 mo of time remained in which to observe the subject for COU. The total data on each subject concluded either through expiration of the dataset, discharge from the military, or death. Of 827,265 individuals who served during 2011–2014, the selection criteria provided 6.73 million person-months of observation on 552,193 individuals. Regression Analyses We applied discrete time multivariable logistic regression to derive odds ratios for the associations between the selected covariates and the dependent variable. In addition to estimating individual variable effects, we also employed interaction terms between selected factors to ascertain whether their effects on the odds of COU were independent or not. An interaction term between factors does not represent the multiplication of their associated odds, but the additional combined effects of these two factors. An odds ratio for an interaction term above 1 would indicate that the combination of the factors has a greater effect than their combined individual effects considered independently. If the odds ratio for the interaction term is below 1, the combined effect of the variables in the interaction is reduced compared with their individual effects. Dependent Variable Because of its aforementioned, associated risks, we specified the appearance of COU itself as the dependent variable. COU was defined as present in the first observed month of opioid use after which the subject experienced three or more consecutive months that included further opioid therapy.19,20 Opioid prescriptions were identified using official pharmacy records. Selected agents included any prescribed substance classified as an opiate or opioid under the American Hospital Formulary System (AHFS).21 Agents were included regardless of morphine equivalent dose or the quantity of medication per prescription as part of our goal to examine any chronic use of opioids. For each person-month in the dataset, the binary dependent variable took on a value of “1” if a subject was dispensed at least one outpatient opioid within each of the three subsequent consecutive months. Our dataset was structured at the person-month level, with all events and statuses reported as of the end of each month. The dependent variable definition therefore ensured that the independent factors always predated potential outcomes, supporting predictive or causal inference for effects discovered. Independent Variables – Medical Covariates The clinically focused, time-varying factors extracted from electronic health records and encounter systems and employed as independent variables were as follows: Maximum Stated Pain Each Month The highest pain score stated by the patient, if any, in each observed month was used to construct a categorical independent variable based on the Numeric Rating Scale (NRS-11).22 This methodology was employed after reviewing prior research approaches23–26 and examining frequency distributions for the scores in our data. Person-months in which the maximum stated pain score 3 or below were labeled as low, those with maximum stated pain between 4–6 were categorized as medium, and those with any pain score of 7 or greater were designated as high. Opioid Medication Use Opioid use was captured using two variables. One variable, current opioid use, represented the count of the gross number of opioid prescriptions dispensed in each observed person-month. A second variable, cumulative opioid use, counted the total number of prior opioid prescriptions received as of each month. For both of these variables, opioid agents were defined as previously described for the dependent variable. Interaction Variable Between Current Opioid Use and Maximum Stated Pain The multiplicative interaction term was further employed for the current opioid use variable and the maximum stated pain variable. Chronic Pain Diagnoses It appears clear that specific medical conditions and the acuity and chronicity of pain, not merely the reported pain severity, could drive the COU probability. We therefore conducted an analysis of the pain-related and other diagnoses assigned to subjects when stratified by their stated pain levels. This analysis was intended to identify possible diagnoses for which further control might be required. Exploring conditions in the “338” family of pain-related ICD-9-CM diagnoses resulted in using code 338.29 (other chronic pain) to create a categorical variable for the count of such diagnoses each person-month. Due to multicollinearity and/or lack of statistical significance, no other pain-related diagnoses were retained to define covariates for the final model. Volume of Prior Pain Encounters We recognized that pain is often intermittent and that “as-needed” use of opioids may be directed based on total, recent pain severity and chronicity. Accordingly, statements of pain at outpatient encounters in recent, prior months could have driven later opioid prescriptions. To address this concern, a further categorical covariate controlled for the total number of outpatient encounters involving stated pain in the 6 mo before each observed month. Total Monthly Outpatient Utilization Because any clinical encounter theoretically provides the opportunity for the recognition of pain and the associated receipt of prescriptions, the total number of encounters of any type each month was used as a covariate. Psychotropic Medication Use Concomitant opioids and other medications such as psychotropic agents may increase the risk of medication-related harm.27 We therefore accounted for prescriptions for top psychotropic medications including benzodiazepines and selective serotonin reuptake inhibitors (SSRIs). Tobacco Use Tobacco dependence predicts poorer outcomes among those treated for opioid dependence.28 We included the self-report of tobacco use at any prior or current health encounter as a binary variable. Independent Variables – Non-clinical Covariates Demographics We controlled for age, gender, and race as potential confounders. We further controlled for marital status as a categorical variable with three values: married, never married, and previously married (divorced/other). Military Service Factors Because of the potential variation in the total time of exposure to the military occupational environment, we included a covariate for active service time. Pay grade was included as a categorical variable for control due to the possible influence of socioeconomic status on COU.29 The seven pay grade categories were E-1 to E-3, E-4, E-5 to E-6, E-7 to E-9, W-1 to W-5, O-1 to O-3, and O-4 and above. We explored the running number of combat deployments as of each person-month, obtained from each soldier’s official records, as a potential, unique predictor in this population. This factor was not retained in final models due to low effect size and loss of statistical significance when controlling for the other factors. Location We included the subject’s geographic location to control for discrete patterns in the standard of medical care and other unknown variations. Locations were identified in terms of the subject’s work location among the largest 32 Army installations worldwide, including overseas locations in Europe and Korea. All other locations were combined into a single category. Year We introduced a categorical variable for the calendar year. This variable was intended to provide control for any population trends in opioid prescribing. Season We further controlled for the season of the year to provide visibility on potential recurrent seasonal patterns in opioid prescriptions. Time Since the Index Observation We directly parameterized the passage of time after the index observation as a continuous variable. When included in the regression, in combination with the person-month data structure and censoring after the outcome, this covariate produced odds ratios that were similarly interpretable as hazard ratios produced by Cox proportional hazards models.30 This research study was determined to be exempt by the Institutional Review Board at the University of Maryland at College Park. The study protocol underwent secondary review by the Defense Health Agency’s Human Research Protection Office. All statistical analyses were conducted using Stata 14 statistical software.31 Results Population Characteristics The 552,193 subjects included 13,891 (2.5%) individuals who experienced an incident COU outcome. Table I displays the summary demographic characteristics of the subjects with and without COU at the last person-month of observation. There were some notable differences between those with and without COU. Female, White, married, and formerly married subjects were modestly overrepresented among those with COU. The mean age among those with COU was higher than those without COU. Clinically related risk factors for COU, such as opioid and psychotropic prescriptions, tobacco use, and past pain complaints, demonstrated more substantial differences between the groups. There were non-trivial distribution differences across multiple demographic medical factors when comparing those with and without COU. All comparisons were statistically significant when assessed using chi-square tests. Table I. Summary of Subject Characteristics Comparinga Subjects with and Without the Chronic Opioid Use (COU) Outcome Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  aChi-square test p-values were < 0.001 for all comparisons of traits of those who did and did not experience COT. Table I. Summary of Subject Characteristics Comparinga Subjects with and Without the Chronic Opioid Use (COU) Outcome Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  aChi-square test p-values were < 0.001 for all comparisons of traits of those who did and did not experience COT. Regression Model Results The adjusted odds ratios associated with each independent variable as computed by the regression model are presented in Table II. The odds of impending COU were substantial and increased in a lockstep fashion with increasing total numbers of opioid prescriptions in the current month. Subjects with three or more opioid prescriptions were at 7.59 times the odds of COU in the following months compared with those with no prescriptions, when controlling for the other factors (95% confidence interval [CI]: 6.30–9.15). Notably, due to the presence of the model’s interaction term between opioid use and pain level, this was the independent effect of opioid prescriptions when no pain or a zero pain score was recorded at all encounters in the observation month. In a similar manner, with increasing recorded pain levels, the odds of COU increased in a monotonic manner. A high pain score (7–10 out of 10) was associated with a 2.53-fold increase in the odds of impending COU (95% CI: 2.40–2.68) when the subject received no opioids that month. Table II. Summary of Odds Ratios (ORs) and 95% Confidence Intervals (CIs) from Discrete Time Logistic Regression Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Note. Summary of odds ratios (ORs) and 95% confidence intervals (CIs) from multivariablea discrete time logistic regression, including all observed person-months with outpatient encounters in military medical facilities. The independent variable was the subsequent presence of chronic opioid use. aThe model further controlled for the subject’s active military service time, military pay grade, geographic location, the calendar year, season of year, and observed time in months. bStatistical significance: <0.05*; <0.01**; <0.001***. Table II. Summary of Odds Ratios (ORs) and 95% Confidence Intervals (CIs) from Discrete Time Logistic Regression Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Note. Summary of odds ratios (ORs) and 95% confidence intervals (CIs) from multivariablea discrete time logistic regression, including all observed person-months with outpatient encounters in military medical facilities. The independent variable was the subsequent presence of chronic opioid use. aThe model further controlled for the subject’s active military service time, military pay grade, geographic location, the calendar year, season of year, and observed time in months. bStatistical significance: <0.05*; <0.01**; <0.001***. There were substantial increases in the odds of COU associated with increasing numbers of psychotropic medication prescriptions and tobacco use. The counts of prior opioids, total outpatient encounters per month, prior encounters with pain statements, and chronic pain diagnoses were also each robustly associated with COU. The model revealed that women were at modestly decreased adjusted odds of COU compared with men, in contrast to the findings suggested by the simple analysis in Table I. We observed a monotonic increase in the COU odds with increasing age. Whites were at the greatest odds when compared with all other races. Single subjects were at reduced odds of COU compared with married subjects, but formerly married subjects did not differ in these odds. Because the odds ratios of the model’s interaction terms were all below 1, the effect of each of the variables in the interaction was reduced when the other variable increased, and vice versa. The interactions tell us that when we see an increase in the number of opioid prescriptions, we should expect to see a reduction in the effect of stated pain. Further, all the interaction terms except one were statistically significant. This indicates that the quantity of opioid prescriptions modified the association between pain scores and the risk of COU and that biased risk estimates would be obtained if the interaction term was excluded. Table III summarizes the combined odds ratios computed from the opioid volume, stated pain, and the interaction between the two factors. The combined multiplicative odds associated with three or more opioid prescriptions and a high maximum stated pain were 8.26. Table III. Combined Odds Ratios for Chronic Opioid Use Computed from Opioid Volume and Pain Scores Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Table III. Combined Odds Ratios for Chronic Opioid Use Computed from Opioid Volume and Pain Scores Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Predicted COU Risk Scores We computed the individual person-month COU risk by totaling the products of the regression model’s coefficients and the covariate values. These risk values represented the COU probability prediction for each person. The mean and median predicted values for subgroups defined by the number of opioid prescriptions received and the maximum pain stated each month, if any, are displayed in Table IV. Table IV. Mean (Median) Adjusted Risk Percentages for Chronic Opioid Use Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  Table IV. Mean (Median) Adjusted Risk Percentages for Chronic Opioid Use Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  This analysis revealed that when provided three or more opioid prescriptions within the same month, the average probability of impending COU among patients with no, low, medium, or high stated pain ranged from 2.22% among those without a pain score or a zero score to 3.40% for those with high pain. This was a statistically significant difference in mean probabilities (chi-square test: p-value <0.001). Large differences were observed for the maximum computed COU probabilities across subsets of the study population defined by pain scores and opioid volume, when adjusting for the other covariates. Individual person-month risk scores as high as 7.87% were found among patients with no stated pain and no opioid prescriptions, and up to 36.5% among those with a maximum of high pain and three or more prescriptions. Therefore, even though the average population risk is low, the models were able to identify members at notably high risk of COU. DISCUSSION Past research on opioids has targeted opioid abuse as an outcome of interest.8,32 However, patients engaging in opioid abuse constitute but a subset of the total group of those chronically receiving opioids. The growing consensus is to generally practice caution and an appropriate level of restraint when considering opioid prescriptions for patients.4–6 Accordingly, this project examined COU itself, a common form of therapy in clinical care, as a specific outcome of interest. In part, we focused on stated pain levels because prescribers may have little suspicion that COU will occur among patients with low pain. For example, the subset of patients demonstrating “drug-seeking” behavior may be expected to state high pain;33 thus, low pain may engender less concern over chronic use potential. Our results indicate that chronic opioid use may in fact occur at non-trivial rates among patients with relatively low stated pain levels at encounters before chronic use develops. This suggests that alerting clinicians to the potentially widely varying, person-specific risk could be useful, especially considering the relatively low level of COU risk in this population. Of the factors studied, it was the number of opioid prescriptions received that demonstrated the strongest association with impending chronicity, across all pain levels. For subjects with only low stated pain, the average of their predicted probabilities was 65.3% of that of those with high pain. This suggests that remain attentive no matter the patient’s stated pain level, as prescribing more opioids will independently increase the COU risk regardless of the pain stated. Importantly, these adjusted odds ratios describe the impact when controlling for and compared with the other important factors such as the presence of concomitant psychotropes, diagnosed chronic pain, and the recent pain score history. The practical use of this research may include providing COU risk information in the electronic health record. Such use may address another under-appreciated but nonetheless significant concern. Recent research indicates that providers find it difficult to start conversations about opioid use patterns with their patients.34 Although such discussions may be needed when weaning a patient from opioid therapy or when denying a frank request for these medications, they may not be pursued in the absence of evidence on individual risk. A COU risk indicator may provide the impetus to initiate these potentially difficult conversations. The strengths of this work include the very large population accessed and the detailed nature of the available data. However, this study was potentially limited by several considerations, notably including the reliance on data from formatted fields in health records. The narrative exchange with the patient in a given encounter might often reveal details that formatted fields cannot capture. For this reason, the absence of pain scores in months with outpatient care almost certainly represented missing data for many subjects who, in fact, suffered and reported pain. Similarly, when either a zero or a low pain score was the value recorded, the patient may have suffered intermittent severe pain that was not high at the time of the captured encounter(s). Therefore, the potential exists for underestimated pain level information in the health records in which it was captured. However, by controlling for precedent pain complaints over a 6-month window, we believe that we were able to adequately account for other care that might have prompted subsequent opioid prescribing. We also recognize the range of morphine equivalent dosing across the agents combined when defining COU in this study and plan to address this factor in greater detail in our ongoing research. Further, the data included any prescriptions dispensed by military facilities or paid for by the Army’s health care coverage for soldiers if dispensed by a civilian pharmacy. However, if an opioid prescription was obtained from a civilian pharmacy using self-purchased insurance or by paying cash, this was not captured in our data. As the rate of such behavior among soldiers is unknown, this limitation may have resulted in an underestimate of the opioids received. It is also conceivable that this work may have been limited by under-reporting of pain-related visits themselves due to our sole use of data from the electronic health record (EHR). We did not have access to information on care that may have been documented only on paper. The Army implemented its EHR in 200735, and the proportion of paper documentation used after that time is not known. However, by using every EHR-based pain encounter for the total population, we believe that the best possible use of the available data has occurred. It is additionally worth noting that the US Military health system’s policy during the time of these data was for an “appropriate assessment for acute and chronic pain at every medical encounter in patients seeking care at MTFs.”36 Even at brief outpatient visits, such as for immunizations or blood pressure checks, unmet pain control needs may have been recognized and new prescriptions obtained in order to comply with the pain assessment goal of the policy. For these reasons, we do not propose that the presence of no recorded pain or a low stated pain level at a single encounter was necessarily inconsistent with appropriately receiving opioids around that time. A general limitation of this research is that associations with other factors we did not explore may prove useful in predicting COU risk in this and other populations. Future studies may include analyses of factors such as military occupational specialty, among which opioid use may vary. Further, the increasing availability of clinical notes from electronic medical records may provide additional information that could potentially strengthen COU risk models. Our future research will target such data as potential means of improving risk projections. Finally, we recognize that this research in a military dataset may not fully generalize to the civilian population and that further research in such groups will be required to confirm or refute external validity. Our ongoing research will include more detailed examination of opioids prescribed in the current datasets and newly obtained data, in terms of quantity per prescription and the agents received. We further plan to explore inpatient care among the patients examined here. The derivation and validation of prediction rules based on the approach presented here also represents planned research, given the goal of providing decision support. We will need to conduct further research on persons with a range of opioid experience levels, as we do not yet know the impact of opioid tolerance on the adjusted COU risk. The data available to the research team are expected to support all of the planned work. In conclusion, we have identified clinically relevant risk factors for chronic opioid use and demonstrated the need to stratify the analysis by pain level. Our findings could guide clinical decision-making in pain management setting, and the growing availability of clinical data in digital form facilitates such development. Further research is needed to expand our knowledge in this critical area. Our ongoing work will include assessments of the relationships identified and further refinement of the methods presented here. Funding The University of Maryland-Leidos seed grant program partially supported this research. The grant program sponsors had no role in the conduct of the research or preparation of the manuscript. References 1 Hare BD, Lipman AG: Uses and misuses of medication in the management of chronic, noncancer pain. In: Problems in Anesthesia . Edited by Hare BD, Fine PG. Philadelphia, PA, JB Lippincott, 1990. 2 ARCOS, Retail Drug Summary Reports; US Department of Justice, Drug Enforcement Administration, Office of Diversion Control. 2006. Available at: https://www.deadiversion.usdoj.gov/arcos/retail_drug_summary/; accessed October 13, 2017. 3 Crofford L: Adverse effects of chronic opioid therapy for chronic musculoskeletal pain. Nat Rev Rheumatol  2010; 6( 4): 191– 7. Google Scholar CrossRef Search ADS PubMed  4 Kahan M, Mallis-Gagnon A, Wilson L, Srivastava A: Canadian guideline for safe and effective use of opioids for chronic non-cancer pain: clinical summary for family physicians. Can Fam Physician  2011; 57( 11): 1257– 66. Google Scholar PubMed  5 Franklin GM: Opioids for chronic noncancer pain: a position paper of the American Academy of Neurology. Neuro  2014; 83( 14): 1277– 84. Google Scholar CrossRef Search ADS   6 Qaseem A, Wilt TJ, McLean RM, Forciea: Noninvasive treatments for acute, subacute and chronic low back pain: a clinical practice guideline from the American College of Physicians. Ann Int Med  2017; 166( 7): 514– 30. Google Scholar CrossRef Search ADS PubMed  7 Sullivan T: Opioid epidemic must also be a call to arms for healthcare IT. In: The Hill. 2016. Available at: http://thehill.com/blogs/pundits-blog/healthcare/295483-opioid-epidemic-must-also-be-a-call-to-arms-for-healthcare-it; accessed October 13, 2017. 8 Ives TJ, Chelminski PR, Hammett-Stabler CA, et al.  : Predictors of opioid misuse in patients with chronic pain: a prospective cohort study. BMC Health Services Research 2006. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1513222/; accessed October 13, 2017. 9 Hah JM, Sharifzadeh Y, Wang BM, et al.  : Factors associated with opioid use in a cohort of patients presenting for surgery. Pain Res Treat 2015. Available at: http://www.hindawi.com/journals/prt/2015/829696/; accessed October 13, 2017. 10 Stites M: Observational pain scales in critically ill adults. Crit Care Nurse  2013; 33( 3): 68– 79. Google Scholar CrossRef Search ADS PubMed  11 Kelly AM: The minimum clinically significant difference in visual analogue scale pain does not differ with severity of pain. Emerg Med J  2001; 18: 205– 7. Google Scholar CrossRef Search ADS PubMed  12 Wiliamson A, Hoggart B: Pain: a review of three commonly used pain rating scales. J Clin Nurs  2005; 14( 7): 798– 804. Google Scholar CrossRef Search ADS PubMed  13 Barnett ML, Olenski AR, Jena AB: Opioid-prescribing patterns of emergency physicians and risk of long-term use. NEJM.  2017; 376: 663– 73. Google Scholar CrossRef Search ADS   14 Shah A, Hayes CJ, Martin BC: Characteristics of initial prescription episodes and likelihood of long-term opioid use – United States, 2006–2015. MMWR Morb Mortal Wkly Rep  2017; 66( 10): 265– 9. Google Scholar CrossRef Search ADS PubMed  15 Nelson D, Kurina L: Clinical prediction of musculoskeletal-related “Medically Not Ready” for combat duty statuses among active duty U.S. army soldiers. Mil Med  2013; 178( 12): 1365– 72. Google Scholar CrossRef Search ADS PubMed  16 Nelson DA, Wolcott VL, Kurina LM: Prediction of all-cause occupational disability among US Army soldiers. Occup Environ Med 2016. Online ahead of print. 17 Toblin R, Quartana P, Riviere L, et al.  : Chronic pain and opioid use in US soldiers after combat deployment. JAMA Intern Med  2014; 174( 8): 1400– 1. Google Scholar CrossRef Search ADS PubMed  18 Bray R, Pemberton M, Lane M, et al.  : Substance use and mental health trends among U.S. military active duty personnel: key findings from the 2008 DoD Health Behavior Survey. Mil Med  2010; 175( 6): 390– 9. Google Scholar CrossRef Search ADS PubMed  19 Chou R, Fanciullo G, Fine P, et al.  : Opioids for chronic noncancer pain: prediction and identification of aberrant drug-related behaviors: a review of the evidence for an American Pain Society and American Academy of Pain Medicine Clinical Practice Guideline. J Pain  2009; 10( 2): 131– 46. Google Scholar CrossRef Search ADS PubMed  20 Dunn K, Saunders K, Rutter C, et al.  : Opioid prescriptions for chronic pain and overdose: a cohort study. Ann Intern Med  2010; 152( 2): 85– 92. Google Scholar CrossRef Search ADS PubMed  21 American Society of Health-System Pharmacists. AHFS Pharmacologic-Therapeutic Classification. Available at: http://www.ahfsdruginformation.com/ahfs-pharmacologic-therapeutic-classification/; accessed October 13, 2017. 22 McCaffery M, Beebe A: Pain: Clinical Manual for Nursing Practice . St. Louis, MO, C.V. Mosby Company, 1989. 23 Forchheimer M, Richards J, Chiodo A, et al.  : Determination in the measurement of pain and its relationship to psychosocial and functional measures after traumatic spinal cord injury: a retrospective model spinal cord injury system analysis. Arch Phys Med Rehabil  2011; 92( 3): 419– 24. Google Scholar CrossRef Search ADS PubMed  24 Hanley M, Masedo A, Jensen M, et al.  : J. pain interference in persons with spinal cord injury: classification of mild, moderate, and severe pain. J Pain  2006; 7( 2): 129– 33. Google Scholar CrossRef Search ADS PubMed  25 Krebs E, Carey T, Weinberger M: Accuracy of the pain numeric rating scale as a screening test in primary care. J Gen Intern Med  2007; 22( 10): 1453– 8. Google Scholar CrossRef Search ADS PubMed  26 Zelman D, Dukes E, Brandenburg N, et al.  : Identification of cut-points for mild, moderate and severe pain due to diabetic peripheral neuropathy. Pain  2005; 115( 1–2): 29– 36. Google Scholar CrossRef Search ADS PubMed  27 Giummarra M, Gibson S, Allen A, et al.  : Polypharmacy and chronic pain: harm exposure is not all about the opioids. Pain Med  2014; 16( 3): 472– 9. Google Scholar CrossRef Search ADS PubMed  28 Mannelli P, Wu L, Peindl K, et al.  : Smoking and Opioid detoxification: behavioral changes and response to treatment. Nicotine Tob Res  2013; 15( 10): 1705– 13. Google Scholar CrossRef Search ADS PubMed  29 Maclean A, Edwards R: The pervasive role of rank in the health of U.S. veterans. Armed Forces Soc  2010; 36( 5): 765– 85. Google Scholar CrossRef Search ADS PubMed  30 Gibbons R, Duan N, Meltzer D, et al.  : Waiting for organ transplantation: results of an analysis by an Institute of Medicine Committee. Biostatistics  2003; 4( 2): 207– 22. Google Scholar CrossRef Search ADS PubMed  31 StataCorp, Stata 14 software, College Station, Texas, 2015. 32 Cochran BN, Flentje A, Heck NC, et al.  : Factors predicting development of opioid use disorders among individuals who receive an initial opioid prescription: mathematical modeling using a database of commercially-insured individuals. Drug Alcohol Depend  2014; 138: 202– 8. Google Scholar CrossRef Search ADS PubMed  33 Grover F: Quantifying drug-seeking behaviors in the emergency department. 2012. Available at: http://www.physiciansweekly.com/drug-seeking-behaviors-emergency-department/; accessed October 13, 2017. 34 Hagemeier NE, Gray JA, Pack RP: Prescription drug abuse: a comparison of prescriber and pharmacist perspectives. Subst Use Misuse  2015; 31( 9): 820– 9. 35 Woodson J: Memorandum for Assistant Secreatary of the Army (M&RA), Assistant Secreatary of the Navy (M&RA), Assistant Secreatary of the Airforce (M&RA), Director Joint Staff on Policy for Comprehensive Pain Management. 2011. Available at: http://www.health.mil/Policies/2011/04/05/Policy-for-Comprehensive-Pain-Management; accessed October 13, 2017. 36 Yoshida S, Bacon B: Contextual History and Visual Timeline of AHLTA and VISTA/CPRS Products. 2008. Available at: http://piim.newschool.edu/_media/pdfs/PIIM-RESEARCH_AHLTA_VISTA_History.pdf; accessed October 13, 2017. Author notes The views expressed in this article are those of the authors and do not necessarily reflect the views or official policies of the U.S. Government, Department of Defense, Defense Health Agency, Department of the Army, or the U.S. Army Medical Department. © Association of Military Surgeons of the United States 2018. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Military Medicine Oxford University Press

Stated Pain Levels, Opioid Prescription Volume, and Chronic Opioid Use Among United States Army Soldiers

Loading next page...
 
/lp/ou_press/stated-pain-levels-opioid-prescription-volume-and-chronic-opioid-use-tv8oLfxqWW
Publisher
Association of Military Surgeons of the United States
Copyright
© Association of Military Surgeons of the United States 2018. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
0026-4075
eISSN
1930-613X
D.O.I.
10.1093/milmed/usy026
Publisher site
See Article on Publisher Site

Abstract

Abstract Introduction The use of opioids has increased drastically over the past few years and decades. As a result, concerns have mounted over serious outcomes associated with chronic opioid use (COU), including dependency and death. A greater understanding of the factors that are associated with COU will be critical if prescribers are to navigate potentially competing objectives to provide compassionate care, while reducing the overall opioid use problem. In this study, we study pain levels and opioid prescription volumes and their effects on the risk of COU. This study leveraged passive data sources that support automated decision support systems (DSSs) currently employed in a large military population. The models presented compute monthly, person-specific, adjusted probability of subsequent COT and could potentially provide critical decision support for clinicians engaged in pain management. Materials and Methods The study population included all outpatient presentations at military medical facilities worldwide among active duty United States Army soldiers during July 2011 to September 2014 (17,664,006 encounters; population N = 552,193). We conducted a retrospective cohort study of this population and employed longitudinal data and a discrete time multivariable logistic regression model to compute COT probability scores. The contribution of pain scores and opioid prescription quantities to the probability of COT represented analytic foci. Results There were 13,891 subjects (2.5%) who experienced incident COT during the observed time period. Statistically significant interactions between pain scores and prescription quantity were present, in addition to effects of multiple other control variables. Counts of monthly opioid prescriptions and maximum stated pain scores per month were each positively associated with COT. A wide range in individual COT risk scores was evident. The effect of prescription volume on the COT risk was larger than the effect of the pain score, and the combined effect of larger pain scores and increased prescription quantity was moderated by the interaction term. Conclusions The results verified that passive data on the US Army can support a robust COT risk computation in this population. The individual, adjusted risk level requires statistical analyses to be fully understood. Because the same data sources drive current military DSSs, this work provides the potential basis for new, evidence-based decision support resources for military clinicians. The strong, independent impact of increasing opioid prescription counts on the COT risk reinforces the importance of exploring alternatives to opioids in pain management planning. It suggests that changing provider behavior through enhanced decision support could help reduce COT rates. INTRODUCTION The perspective of the medical community on opioids has changed substantially over the last several decades. Before the mid-1990s, opioid use was largely seen as appropriate in severe pain and serious cases of chronic pain.1 However, the use of agents such as oxycodone then accelerated, increasing up to eightfold over the next 15 yr.2 As a result, concerns have mounted over serious outcomes associated with chronic opioid use (COU), including dependency and death.3 Leaders today have emphasized that prescribers should again exercise restraint in beginning or continuing opioid therapy in the absence of serious indications for their use.4–6 In this new era, clinicians must therefore employ great selectivity when considering opioids to ensure that pain control is achieved, while considering potential adverse outcomes of therapy. A greater understanding of the factors that are associated with COU will be critical if prescribers are to navigate potentially competing objectives to provide compassionate and quality care, while reducing the overall opioid use problem.7 One goal of new research might be to quantify the evidence-based probability of a patient continuing into COU after an initial opioid prescription, based on the available data at those points in time. Because such patients will have no opioid use history, the likelihood of COU and its attendant problems may be difficult to ascertain. An evidence-based COU probability computation might enhance the personalized approach taken when considering each such new opioid prescription. There are at least two main areas that might be quantified and leveraged with respect to the risk of COU: the patient’s needs and the provider’s response. With respect to patient needs, his or her stated pain score is the main numeric indicator of the pain burden. However, this value is not a major factor in existing practice guidelines on opioids. Research on the impact of pain scores on opioid-related outcomes has been inconsistent.8,9 In general, related research has mostly focused on the accuracy of the pain score and changes thereof, especially in the nursing research literature.10–12 Regarding the clinician’s response to the patient’s stated pain, key metrics include whether opioids are prescribed, and if so, the strength and duration of the therapy provided. Recent research suggests that clinicians may vary widely in their opioid-prescribing practices and that the probability of long-term opioid use is increased among patients receiving opioids from providers who prescribe large quantities of these agents.13,14 These findings suggest that provider behavior is a non-trivial contributor to whether COU occurs, and this factor requires further study to determine its impact. The broad goals of this study were to provide evidence that might inform pain management choices and to create methods that might drive clinical decision support tools. Its specific aim was to compute the contribution of stated pain scores and opioid prescription parameters on the probability of COU using data commonly found in electronic records. The work was conducted using a retrospective, longitudinal data analysis of the total active duty US Army, a population in which prior research using similar data has produced decision support systems for military clinicians.15,16 As in the total US population, opioids represent an important concern for public health in the US military. Over 15% of combat-deployed US soldiers indicated, they had engaged in past-month opioid use in one 2014 study.17 The same study indicated that greater than 40% of these individuals stated mild pain or had no current pain. Patients with opioid abuse disorders constitute a subset of those engaged in COU. The rate of these disorders is 0.7% among US persons over 12 yr of age, yet 11% of American military service members state that they have misused prescription medications when surveyed.18 METHODS Study Population We conducted a retrospective, longitudinal cohort study of combined administrative and medical data on the total, active duty US Army. Non-military persons, National Guard and Reserve soldiers, and individuals serving in other US military branches such as the Navy were not available. The data included military duty status and demographic data as well as detailed clinical elements. A broad array of data sources was combined in order to generate as complete picture as possible. The data sources leveraged included the following: Defense Manpower Data Center (DMDC) Master File: Demographic and military service data DMDC Transaction File: Duty status changes, including discharges from service Military Health System Data Repository (MDR) Combined Ambulatory Professional Encounter Record: Details of outpatient care, including diagnoses MDR Clinical Data Repository Vitals File: Pain scores MDR Pharmacy Detail Transaction Service: Details of dispensed medications The dataset included all outpatient encounters by all active Army soldiers during January 2011 to September 2014 at military treatment facilities worldwide, per electronic health records. The data were condensed into a longitudinal, person-month format in order to support descriptive analyses and time-to-event regression modeling. We selected initially opioid-naïve subjects for study in order to avoid beginning observation during potential pain management episodes. For each eligible member, we defined the index observation month as the month within which the subject had an outpatient clinical presentation for any reason, after at least 6 mo without opioid prescriptions. As the data began in January 2011, the earliest possible index observation month was July 2011. Subsequent, eligible observations were person-months in which outpatient care was received at military treatment facilities. Observation for COU continued until either the occurrence of an outcome (three subsequent months of opioid prescriptions) or when less than 3 mo of time remained in which to observe the subject for COU. The total data on each subject concluded either through expiration of the dataset, discharge from the military, or death. Of 827,265 individuals who served during 2011–2014, the selection criteria provided 6.73 million person-months of observation on 552,193 individuals. Regression Analyses We applied discrete time multivariable logistic regression to derive odds ratios for the associations between the selected covariates and the dependent variable. In addition to estimating individual variable effects, we also employed interaction terms between selected factors to ascertain whether their effects on the odds of COU were independent or not. An interaction term between factors does not represent the multiplication of their associated odds, but the additional combined effects of these two factors. An odds ratio for an interaction term above 1 would indicate that the combination of the factors has a greater effect than their combined individual effects considered independently. If the odds ratio for the interaction term is below 1, the combined effect of the variables in the interaction is reduced compared with their individual effects. Dependent Variable Because of its aforementioned, associated risks, we specified the appearance of COU itself as the dependent variable. COU was defined as present in the first observed month of opioid use after which the subject experienced three or more consecutive months that included further opioid therapy.19,20 Opioid prescriptions were identified using official pharmacy records. Selected agents included any prescribed substance classified as an opiate or opioid under the American Hospital Formulary System (AHFS).21 Agents were included regardless of morphine equivalent dose or the quantity of medication per prescription as part of our goal to examine any chronic use of opioids. For each person-month in the dataset, the binary dependent variable took on a value of “1” if a subject was dispensed at least one outpatient opioid within each of the three subsequent consecutive months. Our dataset was structured at the person-month level, with all events and statuses reported as of the end of each month. The dependent variable definition therefore ensured that the independent factors always predated potential outcomes, supporting predictive or causal inference for effects discovered. Independent Variables – Medical Covariates The clinically focused, time-varying factors extracted from electronic health records and encounter systems and employed as independent variables were as follows: Maximum Stated Pain Each Month The highest pain score stated by the patient, if any, in each observed month was used to construct a categorical independent variable based on the Numeric Rating Scale (NRS-11).22 This methodology was employed after reviewing prior research approaches23–26 and examining frequency distributions for the scores in our data. Person-months in which the maximum stated pain score 3 or below were labeled as low, those with maximum stated pain between 4–6 were categorized as medium, and those with any pain score of 7 or greater were designated as high. Opioid Medication Use Opioid use was captured using two variables. One variable, current opioid use, represented the count of the gross number of opioid prescriptions dispensed in each observed person-month. A second variable, cumulative opioid use, counted the total number of prior opioid prescriptions received as of each month. For both of these variables, opioid agents were defined as previously described for the dependent variable. Interaction Variable Between Current Opioid Use and Maximum Stated Pain The multiplicative interaction term was further employed for the current opioid use variable and the maximum stated pain variable. Chronic Pain Diagnoses It appears clear that specific medical conditions and the acuity and chronicity of pain, not merely the reported pain severity, could drive the COU probability. We therefore conducted an analysis of the pain-related and other diagnoses assigned to subjects when stratified by their stated pain levels. This analysis was intended to identify possible diagnoses for which further control might be required. Exploring conditions in the “338” family of pain-related ICD-9-CM diagnoses resulted in using code 338.29 (other chronic pain) to create a categorical variable for the count of such diagnoses each person-month. Due to multicollinearity and/or lack of statistical significance, no other pain-related diagnoses were retained to define covariates for the final model. Volume of Prior Pain Encounters We recognized that pain is often intermittent and that “as-needed” use of opioids may be directed based on total, recent pain severity and chronicity. Accordingly, statements of pain at outpatient encounters in recent, prior months could have driven later opioid prescriptions. To address this concern, a further categorical covariate controlled for the total number of outpatient encounters involving stated pain in the 6 mo before each observed month. Total Monthly Outpatient Utilization Because any clinical encounter theoretically provides the opportunity for the recognition of pain and the associated receipt of prescriptions, the total number of encounters of any type each month was used as a covariate. Psychotropic Medication Use Concomitant opioids and other medications such as psychotropic agents may increase the risk of medication-related harm.27 We therefore accounted for prescriptions for top psychotropic medications including benzodiazepines and selective serotonin reuptake inhibitors (SSRIs). Tobacco Use Tobacco dependence predicts poorer outcomes among those treated for opioid dependence.28 We included the self-report of tobacco use at any prior or current health encounter as a binary variable. Independent Variables – Non-clinical Covariates Demographics We controlled for age, gender, and race as potential confounders. We further controlled for marital status as a categorical variable with three values: married, never married, and previously married (divorced/other). Military Service Factors Because of the potential variation in the total time of exposure to the military occupational environment, we included a covariate for active service time. Pay grade was included as a categorical variable for control due to the possible influence of socioeconomic status on COU.29 The seven pay grade categories were E-1 to E-3, E-4, E-5 to E-6, E-7 to E-9, W-1 to W-5, O-1 to O-3, and O-4 and above. We explored the running number of combat deployments as of each person-month, obtained from each soldier’s official records, as a potential, unique predictor in this population. This factor was not retained in final models due to low effect size and loss of statistical significance when controlling for the other factors. Location We included the subject’s geographic location to control for discrete patterns in the standard of medical care and other unknown variations. Locations were identified in terms of the subject’s work location among the largest 32 Army installations worldwide, including overseas locations in Europe and Korea. All other locations were combined into a single category. Year We introduced a categorical variable for the calendar year. This variable was intended to provide control for any population trends in opioid prescribing. Season We further controlled for the season of the year to provide visibility on potential recurrent seasonal patterns in opioid prescriptions. Time Since the Index Observation We directly parameterized the passage of time after the index observation as a continuous variable. When included in the regression, in combination with the person-month data structure and censoring after the outcome, this covariate produced odds ratios that were similarly interpretable as hazard ratios produced by Cox proportional hazards models.30 This research study was determined to be exempt by the Institutional Review Board at the University of Maryland at College Park. The study protocol underwent secondary review by the Defense Health Agency’s Human Research Protection Office. All statistical analyses were conducted using Stata 14 statistical software.31 Results Population Characteristics The 552,193 subjects included 13,891 (2.5%) individuals who experienced an incident COU outcome. Table I displays the summary demographic characteristics of the subjects with and without COU at the last person-month of observation. There were some notable differences between those with and without COU. Female, White, married, and formerly married subjects were modestly overrepresented among those with COU. The mean age among those with COU was higher than those without COU. Clinically related risk factors for COU, such as opioid and psychotropic prescriptions, tobacco use, and past pain complaints, demonstrated more substantial differences between the groups. There were non-trivial distribution differences across multiple demographic medical factors when comparing those with and without COU. All comparisons were statistically significant when assessed using chi-square tests. Table I. Summary of Subject Characteristics Comparinga Subjects with and Without the Chronic Opioid Use (COU) Outcome Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  aChi-square test p-values were < 0.001 for all comparisons of traits of those who did and did not experience COT. Table I. Summary of Subject Characteristics Comparinga Subjects with and Without the Chronic Opioid Use (COU) Outcome Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  Independent Variable  With COU Outcome N = 13,891  Without COU Outcome N = 538,302  Female gender; n (% column total)  2096 (15.1)  69,839 (13.0)  Age, yr, in last observed month; mean (SD)  30.58 (7.55)  28.95 (7.75)  Race, n (% column total)   White  10,460 (75.3)  372,945 (69.3)   Black  2236 (16.7)  109,689 (20.4)   Asian  366 (2.6)  21,975 (4.1)   Other  739 (5.3)  33,693 (6.3)  Marital status in last observed month, n (% column total)   Married  9810 (70.6)  317,953 (59.1)   Never married  3001 (21.6)  188,417 (35.0)   Divorced/other  1080 (7.8)  31,932 (5.9)   Psychotropic prescriptions received per month; mean (SD)  0.23 (0.40)  0.07 (0.21)   Had any past tobacco use, as of last observed month; n (% column N)  8964 (64.5)  266,846 (49.6)   Opioid prescriptions received per month; mean (SD)  0.26 (0.54)  0.05 (0.12)   Maximum pain score per month; mean (SD)  2.35 (1.83)  0.99 (1.16)   Number of chronic pain diagnoses per month; mean (SD)  0.03 (0.14)  0.007 (0.06)   Number of encounters with a recorded pain score per month; mean (SD)  0.77 (0.86)  0.29 (0.42)   Number of outpatient encounters per month; mean (SD)  4.01 (4.40)  2.70 (5.53)  aChi-square test p-values were < 0.001 for all comparisons of traits of those who did and did not experience COT. Regression Model Results The adjusted odds ratios associated with each independent variable as computed by the regression model are presented in Table II. The odds of impending COU were substantial and increased in a lockstep fashion with increasing total numbers of opioid prescriptions in the current month. Subjects with three or more opioid prescriptions were at 7.59 times the odds of COU in the following months compared with those with no prescriptions, when controlling for the other factors (95% confidence interval [CI]: 6.30–9.15). Notably, due to the presence of the model’s interaction term between opioid use and pain level, this was the independent effect of opioid prescriptions when no pain or a zero pain score was recorded at all encounters in the observation month. In a similar manner, with increasing recorded pain levels, the odds of COU increased in a monotonic manner. A high pain score (7–10 out of 10) was associated with a 2.53-fold increase in the odds of impending COU (95% CI: 2.40–2.68) when the subject received no opioids that month. Table II. Summary of Odds Ratios (ORs) and 95% Confidence Intervals (CIs) from Discrete Time Logistic Regression Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Note. Summary of odds ratios (ORs) and 95% confidence intervals (CIs) from multivariablea discrete time logistic regression, including all observed person-months with outpatient encounters in military medical facilities. The independent variable was the subsequent presence of chronic opioid use. aThe model further controlled for the subject’s active military service time, military pay grade, geographic location, the calendar year, season of year, and observed time in months. bStatistical significance: <0.05*; <0.01**; <0.001***. Table II. Summary of Odds Ratios (ORs) and 95% Confidence Intervals (CIs) from Discrete Time Logistic Regression Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Variable or Variable Categorya  Model 1: All Person-Months with Outpatient Care in Military Medical Facilities (6,730,825 mo; 552,193 Individuals)    ORb  95% CI  Female gender  0.83***  0.79–0.87  Age, yr   <24  1.00  Referent   24–28  1.29***  1.22–1.37   29–35  1.52***  1.43–1.63   >35  1.62***  1.50–1.75  Race   White  1.00  Referent   Black  0.66***  0.63–0.69   Asian  0.60***  0.5–0.66   Other  0.72***  0.67–0.78  Marital status   Married  1.00  Referent   Never married  0.79***  0.75–0.83   Divorced/other  0.98  0.92–1.05  Psychotropic prescriptions received in current month   None  1.00  Referent   1  1.47***  1.39–1.56   2+  1.66***  1.56–1.76   Past tobacco use  1.52***  1.46–1.57  Opioid prescriptions received in current month   None  1.00  Referent   1  1.32***  1.19–1.46   2  3.35***  2.88–3.90   3+  7.59***  6.30–9.15  Maximum pain score recorded in current month   None recorded, or 0  1.00  Referent   1–3  1.13***  1.06–1.21   4–6  1.88***  1.79–1.97   7–10  2.53***  2.40–2.68  Interaction terms for opioids received ("Rxs") and maximum pain   1 Rx, low pain  0.76*  0.60–0.97   1 Rx, medium pain  0.55***  0.47–0.65   1 Rx, high pain  0.47***  0.40–0.56   2 Rxs, low pain  0.68*  0.47–0.97   2 Rxs, medium pain  0.42***  0.33–0.54   2 Rxs, high pain  0.35***  0.28–0.45   3+ Rxs, low pain  0.72  0.47–1.10   3+ Rxs, medium pain  0.44***  0.33–0.58   3+ Rxs, high pain  0.43***  0.34–0.56  Past total number of opioid prescriptions received   None  1.00  Referent   1  1.57***  1.50–1.65   2  2.22***  2.10–2.36   3+  4.01***  3.82–4.21  Number of chronic pain diagnoses made in current month   None  1.00  Referent   1  1.71***  1.53–1.91   2+  1.87***  1.57–2.24  Number of encounters in prior 6 mo with a recorded pain score   None  1.00  Referent   1  1.37***  1.30–1.45   2–3  1.82***  1.723–1.91   4+  2.37***  2.25–2.49  Number of outpatient encounters overall in current month   1  1.00  Referent   2–3  1.25***  1.19–1.32   4+  1.57***  1.50–1.64  Note. Summary of odds ratios (ORs) and 95% confidence intervals (CIs) from multivariablea discrete time logistic regression, including all observed person-months with outpatient encounters in military medical facilities. The independent variable was the subsequent presence of chronic opioid use. aThe model further controlled for the subject’s active military service time, military pay grade, geographic location, the calendar year, season of year, and observed time in months. bStatistical significance: <0.05*; <0.01**; <0.001***. There were substantial increases in the odds of COU associated with increasing numbers of psychotropic medication prescriptions and tobacco use. The counts of prior opioids, total outpatient encounters per month, prior encounters with pain statements, and chronic pain diagnoses were also each robustly associated with COU. The model revealed that women were at modestly decreased adjusted odds of COU compared with men, in contrast to the findings suggested by the simple analysis in Table I. We observed a monotonic increase in the COU odds with increasing age. Whites were at the greatest odds when compared with all other races. Single subjects were at reduced odds of COU compared with married subjects, but formerly married subjects did not differ in these odds. Because the odds ratios of the model’s interaction terms were all below 1, the effect of each of the variables in the interaction was reduced when the other variable increased, and vice versa. The interactions tell us that when we see an increase in the number of opioid prescriptions, we should expect to see a reduction in the effect of stated pain. Further, all the interaction terms except one were statistically significant. This indicates that the quantity of opioid prescriptions modified the association between pain scores and the risk of COU and that biased risk estimates would be obtained if the interaction term was excluded. Table III summarizes the combined odds ratios computed from the opioid volume, stated pain, and the interaction between the two factors. The combined multiplicative odds associated with three or more opioid prescriptions and a high maximum stated pain were 8.26. Table III. Combined Odds Ratios for Chronic Opioid Use Computed from Opioid Volume and Pain Scores Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Table III. Combined Odds Ratios for Chronic Opioid Use Computed from Opioid Volume and Pain Scores Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Opioid Prescriptions Received that Month  No Pain Score or a Zero Score  A Maximum Pain Score of 1–3  A Maximum Pain Score of 4–6  A Maximum Pain Score of 7–10  None  1.00 (referent)  1.13  1.88  2.53  1  1.32  1.13  1.36  1.57  2  3.35  2.57  2.65  3.00  3+  7.59  6.18  6.28  8.26  Predicted COU Risk Scores We computed the individual person-month COU risk by totaling the products of the regression model’s coefficients and the covariate values. These risk values represented the COU probability prediction for each person. The mean and median predicted values for subgroups defined by the number of opioid prescriptions received and the maximum pain stated each month, if any, are displayed in Table IV. Table IV. Mean (Median) Adjusted Risk Percentages for Chronic Opioid Use Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  Table IV. Mean (Median) Adjusted Risk Percentages for Chronic Opioid Use Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  Number of Opioid Prescriptions Received  No or a Zero Pain Score  Low Pain (Score 1–3)  Medium Pain (Score 4–6)  High Pain (Score 7–10)  None  0.13 (0.08)  0.23 (0.14)  0.43 (0.25)  0.59 (0.34)  1  0.27 (0.16)  0.33 (0.10)  0.46 (0.27)  0.50 (0.29)  2  0.82 (0.50)  0.87 (0.53)  1.03 (0.62)  1.12 (0.67)  3+  2.22 (1.37)  2.29 (1.39)  2.63 (1.59)  3.40 (2.10)  This analysis revealed that when provided three or more opioid prescriptions within the same month, the average probability of impending COU among patients with no, low, medium, or high stated pain ranged from 2.22% among those without a pain score or a zero score to 3.40% for those with high pain. This was a statistically significant difference in mean probabilities (chi-square test: p-value <0.001). Large differences were observed for the maximum computed COU probabilities across subsets of the study population defined by pain scores and opioid volume, when adjusting for the other covariates. Individual person-month risk scores as high as 7.87% were found among patients with no stated pain and no opioid prescriptions, and up to 36.5% among those with a maximum of high pain and three or more prescriptions. Therefore, even though the average population risk is low, the models were able to identify members at notably high risk of COU. DISCUSSION Past research on opioids has targeted opioid abuse as an outcome of interest.8,32 However, patients engaging in opioid abuse constitute but a subset of the total group of those chronically receiving opioids. The growing consensus is to generally practice caution and an appropriate level of restraint when considering opioid prescriptions for patients.4–6 Accordingly, this project examined COU itself, a common form of therapy in clinical care, as a specific outcome of interest. In part, we focused on stated pain levels because prescribers may have little suspicion that COU will occur among patients with low pain. For example, the subset of patients demonstrating “drug-seeking” behavior may be expected to state high pain;33 thus, low pain may engender less concern over chronic use potential. Our results indicate that chronic opioid use may in fact occur at non-trivial rates among patients with relatively low stated pain levels at encounters before chronic use develops. This suggests that alerting clinicians to the potentially widely varying, person-specific risk could be useful, especially considering the relatively low level of COU risk in this population. Of the factors studied, it was the number of opioid prescriptions received that demonstrated the strongest association with impending chronicity, across all pain levels. For subjects with only low stated pain, the average of their predicted probabilities was 65.3% of that of those with high pain. This suggests that remain attentive no matter the patient’s stated pain level, as prescribing more opioids will independently increase the COU risk regardless of the pain stated. Importantly, these adjusted odds ratios describe the impact when controlling for and compared with the other important factors such as the presence of concomitant psychotropes, diagnosed chronic pain, and the recent pain score history. The practical use of this research may include providing COU risk information in the electronic health record. Such use may address another under-appreciated but nonetheless significant concern. Recent research indicates that providers find it difficult to start conversations about opioid use patterns with their patients.34 Although such discussions may be needed when weaning a patient from opioid therapy or when denying a frank request for these medications, they may not be pursued in the absence of evidence on individual risk. A COU risk indicator may provide the impetus to initiate these potentially difficult conversations. The strengths of this work include the very large population accessed and the detailed nature of the available data. However, this study was potentially limited by several considerations, notably including the reliance on data from formatted fields in health records. The narrative exchange with the patient in a given encounter might often reveal details that formatted fields cannot capture. For this reason, the absence of pain scores in months with outpatient care almost certainly represented missing data for many subjects who, in fact, suffered and reported pain. Similarly, when either a zero or a low pain score was the value recorded, the patient may have suffered intermittent severe pain that was not high at the time of the captured encounter(s). Therefore, the potential exists for underestimated pain level information in the health records in which it was captured. However, by controlling for precedent pain complaints over a 6-month window, we believe that we were able to adequately account for other care that might have prompted subsequent opioid prescribing. We also recognize the range of morphine equivalent dosing across the agents combined when defining COU in this study and plan to address this factor in greater detail in our ongoing research. Further, the data included any prescriptions dispensed by military facilities or paid for by the Army’s health care coverage for soldiers if dispensed by a civilian pharmacy. However, if an opioid prescription was obtained from a civilian pharmacy using self-purchased insurance or by paying cash, this was not captured in our data. As the rate of such behavior among soldiers is unknown, this limitation may have resulted in an underestimate of the opioids received. It is also conceivable that this work may have been limited by under-reporting of pain-related visits themselves due to our sole use of data from the electronic health record (EHR). We did not have access to information on care that may have been documented only on paper. The Army implemented its EHR in 200735, and the proportion of paper documentation used after that time is not known. However, by using every EHR-based pain encounter for the total population, we believe that the best possible use of the available data has occurred. It is additionally worth noting that the US Military health system’s policy during the time of these data was for an “appropriate assessment for acute and chronic pain at every medical encounter in patients seeking care at MTFs.”36 Even at brief outpatient visits, such as for immunizations or blood pressure checks, unmet pain control needs may have been recognized and new prescriptions obtained in order to comply with the pain assessment goal of the policy. For these reasons, we do not propose that the presence of no recorded pain or a low stated pain level at a single encounter was necessarily inconsistent with appropriately receiving opioids around that time. A general limitation of this research is that associations with other factors we did not explore may prove useful in predicting COU risk in this and other populations. Future studies may include analyses of factors such as military occupational specialty, among which opioid use may vary. Further, the increasing availability of clinical notes from electronic medical records may provide additional information that could potentially strengthen COU risk models. Our future research will target such data as potential means of improving risk projections. Finally, we recognize that this research in a military dataset may not fully generalize to the civilian population and that further research in such groups will be required to confirm or refute external validity. Our ongoing research will include more detailed examination of opioids prescribed in the current datasets and newly obtained data, in terms of quantity per prescription and the agents received. We further plan to explore inpatient care among the patients examined here. The derivation and validation of prediction rules based on the approach presented here also represents planned research, given the goal of providing decision support. We will need to conduct further research on persons with a range of opioid experience levels, as we do not yet know the impact of opioid tolerance on the adjusted COU risk. The data available to the research team are expected to support all of the planned work. In conclusion, we have identified clinically relevant risk factors for chronic opioid use and demonstrated the need to stratify the analysis by pain level. Our findings could guide clinical decision-making in pain management setting, and the growing availability of clinical data in digital form facilitates such development. Further research is needed to expand our knowledge in this critical area. Our ongoing work will include assessments of the relationships identified and further refinement of the methods presented here. Funding The University of Maryland-Leidos seed grant program partially supported this research. The grant program sponsors had no role in the conduct of the research or preparation of the manuscript. References 1 Hare BD, Lipman AG: Uses and misuses of medication in the management of chronic, noncancer pain. In: Problems in Anesthesia . Edited by Hare BD, Fine PG. Philadelphia, PA, JB Lippincott, 1990. 2 ARCOS, Retail Drug Summary Reports; US Department of Justice, Drug Enforcement Administration, Office of Diversion Control. 2006. Available at: https://www.deadiversion.usdoj.gov/arcos/retail_drug_summary/; accessed October 13, 2017. 3 Crofford L: Adverse effects of chronic opioid therapy for chronic musculoskeletal pain. Nat Rev Rheumatol  2010; 6( 4): 191– 7. Google Scholar CrossRef Search ADS PubMed  4 Kahan M, Mallis-Gagnon A, Wilson L, Srivastava A: Canadian guideline for safe and effective use of opioids for chronic non-cancer pain: clinical summary for family physicians. Can Fam Physician  2011; 57( 11): 1257– 66. Google Scholar PubMed  5 Franklin GM: Opioids for chronic noncancer pain: a position paper of the American Academy of Neurology. Neuro  2014; 83( 14): 1277– 84. Google Scholar CrossRef Search ADS   6 Qaseem A, Wilt TJ, McLean RM, Forciea: Noninvasive treatments for acute, subacute and chronic low back pain: a clinical practice guideline from the American College of Physicians. Ann Int Med  2017; 166( 7): 514– 30. Google Scholar CrossRef Search ADS PubMed  7 Sullivan T: Opioid epidemic must also be a call to arms for healthcare IT. In: The Hill. 2016. Available at: http://thehill.com/blogs/pundits-blog/healthcare/295483-opioid-epidemic-must-also-be-a-call-to-arms-for-healthcare-it; accessed October 13, 2017. 8 Ives TJ, Chelminski PR, Hammett-Stabler CA, et al.  : Predictors of opioid misuse in patients with chronic pain: a prospective cohort study. BMC Health Services Research 2006. Available at: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1513222/; accessed October 13, 2017. 9 Hah JM, Sharifzadeh Y, Wang BM, et al.  : Factors associated with opioid use in a cohort of patients presenting for surgery. Pain Res Treat 2015. Available at: http://www.hindawi.com/journals/prt/2015/829696/; accessed October 13, 2017. 10 Stites M: Observational pain scales in critically ill adults. Crit Care Nurse  2013; 33( 3): 68– 79. Google Scholar CrossRef Search ADS PubMed  11 Kelly AM: The minimum clinically significant difference in visual analogue scale pain does not differ with severity of pain. Emerg Med J  2001; 18: 205– 7. Google Scholar CrossRef Search ADS PubMed  12 Wiliamson A, Hoggart B: Pain: a review of three commonly used pain rating scales. J Clin Nurs  2005; 14( 7): 798– 804. Google Scholar CrossRef Search ADS PubMed  13 Barnett ML, Olenski AR, Jena AB: Opioid-prescribing patterns of emergency physicians and risk of long-term use. NEJM.  2017; 376: 663– 73. Google Scholar CrossRef Search ADS   14 Shah A, Hayes CJ, Martin BC: Characteristics of initial prescription episodes and likelihood of long-term opioid use – United States, 2006–2015. MMWR Morb Mortal Wkly Rep  2017; 66( 10): 265– 9. Google Scholar CrossRef Search ADS PubMed  15 Nelson D, Kurina L: Clinical prediction of musculoskeletal-related “Medically Not Ready” for combat duty statuses among active duty U.S. army soldiers. Mil Med  2013; 178( 12): 1365– 72. Google Scholar CrossRef Search ADS PubMed  16 Nelson DA, Wolcott VL, Kurina LM: Prediction of all-cause occupational disability among US Army soldiers. Occup Environ Med 2016. Online ahead of print. 17 Toblin R, Quartana P, Riviere L, et al.  : Chronic pain and opioid use in US soldiers after combat deployment. JAMA Intern Med  2014; 174( 8): 1400– 1. Google Scholar CrossRef Search ADS PubMed  18 Bray R, Pemberton M, Lane M, et al.  : Substance use and mental health trends among U.S. military active duty personnel: key findings from the 2008 DoD Health Behavior Survey. Mil Med  2010; 175( 6): 390– 9. Google Scholar CrossRef Search ADS PubMed  19 Chou R, Fanciullo G, Fine P, et al.  : Opioids for chronic noncancer pain: prediction and identification of aberrant drug-related behaviors: a review of the evidence for an American Pain Society and American Academy of Pain Medicine Clinical Practice Guideline. J Pain  2009; 10( 2): 131– 46. Google Scholar CrossRef Search ADS PubMed  20 Dunn K, Saunders K, Rutter C, et al.  : Opioid prescriptions for chronic pain and overdose: a cohort study. Ann Intern Med  2010; 152( 2): 85– 92. Google Scholar CrossRef Search ADS PubMed  21 American Society of Health-System Pharmacists. AHFS Pharmacologic-Therapeutic Classification. Available at: http://www.ahfsdruginformation.com/ahfs-pharmacologic-therapeutic-classification/; accessed October 13, 2017. 22 McCaffery M, Beebe A: Pain: Clinical Manual for Nursing Practice . St. Louis, MO, C.V. Mosby Company, 1989. 23 Forchheimer M, Richards J, Chiodo A, et al.  : Determination in the measurement of pain and its relationship to psychosocial and functional measures after traumatic spinal cord injury: a retrospective model spinal cord injury system analysis. Arch Phys Med Rehabil  2011; 92( 3): 419– 24. Google Scholar CrossRef Search ADS PubMed  24 Hanley M, Masedo A, Jensen M, et al.  : J. pain interference in persons with spinal cord injury: classification of mild, moderate, and severe pain. J Pain  2006; 7( 2): 129– 33. Google Scholar CrossRef Search ADS PubMed  25 Krebs E, Carey T, Weinberger M: Accuracy of the pain numeric rating scale as a screening test in primary care. J Gen Intern Med  2007; 22( 10): 1453– 8. Google Scholar CrossRef Search ADS PubMed  26 Zelman D, Dukes E, Brandenburg N, et al.  : Identification of cut-points for mild, moderate and severe pain due to diabetic peripheral neuropathy. Pain  2005; 115( 1–2): 29– 36. Google Scholar CrossRef Search ADS PubMed  27 Giummarra M, Gibson S, Allen A, et al.  : Polypharmacy and chronic pain: harm exposure is not all about the opioids. Pain Med  2014; 16( 3): 472– 9. Google Scholar CrossRef Search ADS PubMed  28 Mannelli P, Wu L, Peindl K, et al.  : Smoking and Opioid detoxification: behavioral changes and response to treatment. Nicotine Tob Res  2013; 15( 10): 1705– 13. Google Scholar CrossRef Search ADS PubMed  29 Maclean A, Edwards R: The pervasive role of rank in the health of U.S. veterans. Armed Forces Soc  2010; 36( 5): 765– 85. Google Scholar CrossRef Search ADS PubMed  30 Gibbons R, Duan N, Meltzer D, et al.  : Waiting for organ transplantation: results of an analysis by an Institute of Medicine Committee. Biostatistics  2003; 4( 2): 207– 22. Google Scholar CrossRef Search ADS PubMed  31 StataCorp, Stata 14 software, College Station, Texas, 2015. 32 Cochran BN, Flentje A, Heck NC, et al.  : Factors predicting development of opioid use disorders among individuals who receive an initial opioid prescription: mathematical modeling using a database of commercially-insured individuals. Drug Alcohol Depend  2014; 138: 202– 8. Google Scholar CrossRef Search ADS PubMed  33 Grover F: Quantifying drug-seeking behaviors in the emergency department. 2012. Available at: http://www.physiciansweekly.com/drug-seeking-behaviors-emergency-department/; accessed October 13, 2017. 34 Hagemeier NE, Gray JA, Pack RP: Prescription drug abuse: a comparison of prescriber and pharmacist perspectives. Subst Use Misuse  2015; 31( 9): 820– 9. 35 Woodson J: Memorandum for Assistant Secreatary of the Army (M&RA), Assistant Secreatary of the Navy (M&RA), Assistant Secreatary of the Airforce (M&RA), Director Joint Staff on Policy for Comprehensive Pain Management. 2011. Available at: http://www.health.mil/Policies/2011/04/05/Policy-for-Comprehensive-Pain-Management; accessed October 13, 2017. 36 Yoshida S, Bacon B: Contextual History and Visual Timeline of AHLTA and VISTA/CPRS Products. 2008. Available at: http://piim.newschool.edu/_media/pdfs/PIIM-RESEARCH_AHLTA_VISTA_History.pdf; accessed October 13, 2017. Author notes The views expressed in this article are those of the authors and do not necessarily reflect the views or official policies of the U.S. Government, Department of Defense, Defense Health Agency, Department of the Army, or the U.S. Army Medical Department. © Association of Military Surgeons of the United States 2018. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Military MedicineOxford University Press

Published: Mar 26, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off