Five-year Pain Intensity and Treatment Trajectories of Post-9/11 Veterans with Mild Traumatic Brain Injury

Abstract

Pain is a pervasive problem that affects nearly half of the U.S. Veterans deployed in support of the Global War on Terror (Post-9/11 Veterans) and over half of the Post-9/11 Veterans with diagnosed traumatic brain injury (TBI). The goal of the current study was to identify pain phenotypes based on distinct longitudinal patterns of pain scores in light of pain treatment among Post-9/11 Veterans over five years of care using latent growth mixture analysis stratified by TBI status. Five pain phenotypes emerged: (1) simple low impact stable pain, (2) complex low impact stable pain, (3) complex low impact worsening pain, (4) complex moderate impact worsening pain, and (5) complex high impact stable pain. Baseline pain scores and slopes were significantly higher in Veterans with mild TBI for some phenotypes. The mild TBI cohort was younger, had more men, more whites, less blacks, less education, more unmarried, more Marines and Army, more active duty in comparison to the no TBI cohort. Distinct trajectories in pain treatment were apparent among the pain intensity subgroups.

INTRODUCTION

Patterns of multimorbidity exist in Post-9/11 Veterans deployed in support of the Global War on Terror.³⁰ In 2009, approximately 40% of Post-9/11 Veterans were diagnosed with some type of pain, but only half of those individuals had pain without traumatic brain injury (TBI) or posttraumatic stress disorder (PTSD).⁷ Among those with diagnosed TBI, over 60% had one or more comorbid pain diagnoses demonstrating the co-occurrence of pain and TBI.⁷ Pain in the context of TBI can be due to physical trauma or overuse injuries and may be driven by musculoskeletal, neuropathic, and central mechanisms. Pain when co-morbid with TBI is accompanied by a vast array of prolonged and often overlapping neurocognitive, psychiatric, and somatic symptoms. Various medications are prescribed to manage post-concussive symptoms; however, the use of these medications is based on their effectiveness in non-TBI cohorts. Treatment for pain conditions relevant to this cohort traditionally include central nervous system (CNS) medications which cause cognitive slowing and place patients at risk for complications such as overdose and abuse.^3,12,27,32 Thus, pain is a pervasive problem that has a significant impact on the large number of Post-9/11 Veterans with TBI.^18,25

Studies estimate that 7–20% of Post-9/11 Veterans have suffered a TBI, and 1.7 million people in the general population sustain a TBI annually with the majority being of mild severity.^14,21 Moreover, chronic pain is a significant problem that affects 40–50% of patients with mTBI.²⁵ Despite studies connecting TBI to chronic pain both directly (i.e., through dysfunctional pain modulatory system¹¹) and indirectly (i.e., through psychiatric comorbidities like PTSD and depression³¹), there is limited information available on the longitudinal course of treatment utilization and outcomes in this complex patient population. This lack of research is fueled by limitations in data and over-reliance on traditional linear models. A better understanding of pain trajectories and treatment utilization in Veterans with TBI is vital to identify gaps in care and subpopulations that need intensive pain care plans.⁴

Pain scores are one of the most commonly collected health metrics in electronic health records (EHR) and are potentially a rich source of longitudinal data for healthcare research.²⁴ In the Veterans Health Administration (VHA), pain scores have been routinely collected during outpatient encounters since 1998 using the 0–10 Numeric Rating Scale (NRS). Despite the universal use of the NRS pain screening, there have been few attempts to capitalize on commonly gathered pain data to define the meaning of these scores over time or in conjunction with treatment. Establishing the utility of low burden health informatics like EHR-based pain ratings is a high priority in improvement of pain care pathways for Post-9/11 Veterans with complex health needs, because pain ratings can predict poor response to rehabilitation and increased socioeconomic costs without introducing the burden of implementing new assessment tools.^8,23 Likewise, information on treatment patterns is sparse. Understanding treatment patterns associated with changes in pain scores over time will provide insights into the (in)effectiveness of current treatments.

The goal of this study was to identify distinct pain phenotypes using pain intensity and treatment patterns in a cohort of Post-9/11 Veterans with mTBI over five years of VA care. We hypothesized that distinctive classes of pain intensity would be associated with differential likelihoods of receiving pharmacological and non-pharmacological treatments for pain, and that, consistent with our findings for comorbidity phenotypes for mTBI and no TBI cohorts, that there would be some distinct pain phenotypes for individuals with mTBI, and some shared phenotypes for mTBI and no TBI cohorts.²⁹ TBI itself is associated with worse pain outcomes when compared to patients without TBI;¹³ therefore, we also expected that subgroups with TBI would have higher pain scores.

METHODS

Data Preparation, Descriptive Analysis and Measures

Data preparation consisted of sample refinement, variable construction, and descriptive analysis of the study cohort.

Study Cohort

Using data from a larger study of health outcomes among Post-9/11 Veterans (Trajectories of Resilience and Complex Comorbidities),³⁰ we identified individuals who entered VA care in fiscal year (FY) 2007, 2008, or 2009 and received care (inpatient, outpatient, medications) at least once a year for five consecutive years. We chose the 5 year study length to capture as much longitudinal data without overlapping with VA policy changes regarding pain treatment such as the Opioid Safety Initiative. Each year of care was considered an observation period for pain scores and healthcare utilization. We restricted our analysis to those with at least one outpatient pain score per year during years one through five of VHA care, those diagnosed with mild TBI (mTBI) based on validated ICD-9-CM coding algorithms,^10,20 and those with no diagnosis of TBI. Because pain is a concern among individuals with mTBI, we stratified analyses comparing patterns for individuals with mTBI to those without TBI. We excluded Veterans with cancer, hospice/palliative care, and hospital length of stay greater than 30 days. Specifics on the derivation of the final analytic cohort are shown in figure 1. The analytic sample included 43,697 Post-9/11 Veterans of which 32.5% had mild TBI.

Figure 1.

Derivation of the analytic cohort.

Data on sociodemographic (age, sex, race/ethnicity, highest education received) and military characteristics (branch of service, component of service, military rank, and deployment history) were obtained from the roster identifying Post-9/11 deployed Veterans which was provided by the Department of Veterans Affairs Epidemiology program (OEF/OIF/OND Roster). We identified individuals with multiple deployments as those whose dates of first and last deployment differed.

VA outpatient and inpatient health records were obtained from the national repository in Austin, TX. Outpatient prescriptions dispensed by the VA were from the Pharmacy Benefits Management database. Data for this retrospective, population-based cohort study were derived from multiple databases and linked using an encrypted identifier to examine longitudinal patterns of pain management. The study was approved by the University of Texas Health Science Center at San Antonio, the University of Utah, and the Edith Nourse Rogers Memorial Veterans Hospital institutional review boards.

Latent Class Indicators

Pain Scores

In the VHA health care system, an 11-point Numerical Rating Scale (NRS) is administered to measure the Veterans pain intensity at the time of care. The patient can report a number between 0 and 10, with 0 representing no pain and 10 representing worst possible pain. We created a mean pain score for each year based on all scores reported in a fiscal year. The NRS is widely recognized as the most reliable and valid measure of pain intensity (compared to other assessment modalities like visual analog ratings and verbal rating scales¹⁹), but there is some disagreement about how to interpret NRS scores. Until recently, NRS scores have been classified into three categories with ratings of 0–4 representing “mild” pain, 5–6 representing “moderate” pain, and ratings from 7–10 representing “severe” pain.³³ In 2016, the National Pain Strategy Population Research Workgroup (PRW) published findings from a study of 770 adults showing that these traditional NRS pain score categories do not align well with assessments of how pain impacts life activities. Individuals with low impact pain reported NRS scores between 1 and 4, moderate impact reported NRS between 2 and 6, and high impact reported NRS between 4 and 8.³⁶ Thus, for the purposes of this study, we chose to adopt NRS scores of 4 or more as an indication of high-impact pain.

NRS pain scores are routinely obtained during outpatient clinical encounters at the VA. These ratings are obtained by the provider within the outpatient clinic in which the patient is attending an appointment. We convened a meeting of several pain and measurement/assessment experts to determine how to define the cohort under study. Research on pain ratings has shown that multiple ratings increase validity over single ratings; therefore, we only included Veterans with multiple ratings over a 5-year period. Although higher frequency of ratings may be beneficial, there is little literature guiding the extent to which rating frequency improves psychometrics of pain ratings beyond benefits provided by multiple ratings. Before deciding on yearly pain scores, we looked at the number of Veterans with pain scores every 3 months (n=100) and every 6 months (n=4014). Ultimately, our team of experts chose to rely on the largest cohort of Veterans with multiple pain ratings over a 5-year period, which included Veterans who received at least one pain assessment per year. This approach reduces information bias that would occur by including only the sickest Veterans who receive frequent VHA care, and more adequately represents Veterans with chronic pain and TBI who are seeking care in the VHA.

Pain Treatment

Similarly, we created variables for each pain treatment modality beginning with extant guidelines followed by grouping medications based on expert consensus of our clinical team (16 medication groups and 6 non-pharmacological approaches) each year based on a multidisciplinary work group including clinical pain management, pharmacy, and military polytrauma expertise. The workgroup reviewed a comprehensive list of treatments and grouped them based on mechanism of action and side effect profile. Medication groups that are commonly used to treat pain or pain-related comorbidity such as insomnia or anxiety were identified using pharmacy data and included topical analgesics, acetaminophen, cox-1 and cox-2 nonsteroidal anti-inflammatory drugs (NSAIDs), antimigraine agents, skeletal muscle relaxants, antidepressants, anticonvulsants, opioids (codeine, tramadol, hydrocodone, oxycodone single, and methadone-hydromorphone-fentanyl patches), and sedative-hypnotics (z-drugs, benzodiazepines). Non-pharmacological modalities were identified using clinic and procedure codes which included interventional pain management, Complementary and Integrative Health (CIH), physical and occupational therapy, and inpatient medicine, psychiatry and surgical stays. See Supplemental Tables 1 and 2 for specifics on included medications and coding used to construct the non-pharmacological treatment measures.

Statistical Analysis

We began with computing descriptive statistics stratified by TBI status. Latent growth mixture modelling (LGMM)^26,28 stratified by TBI status was used to identify distinct pain score trajectories for each TBI group. To understand trajectories of pain intensity in light of pain treatments, we further conducted general latent variable mixture modelling (LVMM) of pain scores with pain treatments jointly.

The distinct pain intensity trajectory classes determined by LGMM were based on the similarity of the patterns of the yearly mean pain scores and the associated within-subject and within-class variability collected over five years, that is, each class shared the same multivariate normal distribution. LGMM analyses require specification of the number of latent classes and the pattern (i.e., growth parameters) of the repeated measures for each class. For each given class number, both quadratic and piecewise linear trajectories were considered (see Supplemental Figure 1). We also allowed the variability of the pain scores to vary by class.

The trajectory classes associated with pain scores and the treatments jointly were estimated by LVMM based on the similarity of the patterns of the yearly mean pain scores and the likelihoods of treatments received, where the repeated measures of pain scores were modelled by LGMM as described above, and the indicators of treatment received each year were modeled by the latent class model.

Each LGMM and LVMM was calculated using 20 starting parameter values with 10,000 Estimation-Maximization (EM) iterations to ensure model convergence. We tested models ranging from 2 to 4 classes for both mTBI and no TBI groups. The best fitting LGMM and LVMM was determined based on model convergence, Akaike information criterion (AIC), Bayesian information criterion (BIC), sample size adjusted BIC and Vuong-Lo-Mendell- Rubin Likelihood Ratio Test as well as clinical logic.

Smaller AIC and BIC values indicating better model fit, and significant Vuong-Lo-Mendell- Rubin Likelihood Ratio Test of a fitted LVMM suggests a better fit of the current model compared to the LVMM with one less number of classes. The decision about piecewise linear versus quadratic pain trajectories was primarily based on clinical utility. After the best fit LGMM was identified, a subject’s unbiased pain trajectory class membership was estimated using the pseudoclass technique by drawing a random sample from the multinomial distribution with probabilities being the propensity score (or posterior probability) of being in each class given the subject’s observed pain scores, and the fitted LGMM analysis result.^2,37 Similarly, a subject’s trajectory class associated with pain score and treatment jointly was estimated by drawing a random sample from the multinomial distribution with probabilities being the posterior probability of being in each class given the subject’s observed pain scores and treatments, and the fitted LVMM analysis result. For example, in a model with two latent classes, propensity scores of 0.9 and 0.1 suggest that he or she is likely to belong to the first class with a probability of 0.9 (90%) and to the second class with a probability of 0.1 (10%). Our team evaluated the pain and treatment trajectories with the best fit and provided descriptive labels to these phenotypes of pain. Then the chi-square test and 2-sample t-test (based on variable distributions) were used to compare the difference in baseline characteristics and trajectory parameters between the mTBI and no TBI groups among those with similar trajectory patterns as well as between trajectory classes conditioned on TBI status. We used a stepwise method based on the pseudoclass theory which provided unbiased estimates of class-specific quantities.^2,37 The strength of this approach includes its theoretical base and avoidance of the bias due to using the most likely class.

Descriptive statistics were calculated using SAS 9.2. LGMM and LVMM analyses were conducted using Mplus Version 7.

RESULTS

Sociodemographic characteristics of the sample are presented by TBI status in Table 1. A majority of the Veterans with mTBI were 29 years and younger, men, White, enlisted, active duty, in the Army, and had high school education. In comparison to the Veterans with mild TBI, the no TBI cohort was slightly older, had more women, had fewer White, more Black, had slighter higher educational attainment, fewer enlisted, less active duty, fewer members in the Army, and more members in the Navy/Coast Guard and Air Force. See Tables 2, 3 and 4 for details.

Table 1.

Baseline characteristics of study cohort by TBI status.

	Mild TBI	No TBI
	10,717	32,980	p-value
Age
Mean ± SD	28.4 ± 7.8	32.0 ± 9.2	<0.0001
29 years and under	65.2%	50.0%
30–39 years	20.5%	21.6%
40–49 years	11.9%	21.7%
50 + years	2.4%	6.6%
Sex
Men	93.5%	80.2%	<0.0001
Race/ethnicity
White	70.0%	62.7%	<0.0001
Black	12.8%	21.2%
Hispanic	12.9%	11.4%
Asian	2.1%	2.7%
Native American/Pacific Islander	1.8%	1.2%
Unknown	0.5%	0.9%
Education
Less than high school	1.2%	1.1%	<0.0001
High school	85.3%	75.9%
Some college	6.9%	11.1%
College + Higher education	5.2%	10.6%
Unknown	1.5%	1.4%
Marital status
Married	40.7%	46.3%	<0.0001
Branch of service
Army	70.2%	54.0%	<0.0001
Air Force	3.9%	14.3%
Navy/ Coast Guard	7.1%	19.6%
Marines	18.8%	12.0%
Military component
Active	71.0%	64.6%	<0.0001
Military rank
Enlisted	97.3%	93.3%	<0.0001
Deployment history
Multiple deployments	49.8%	49.2%	0.3197

Table 2.

Complex low impact, stable trajectory baseline demographics for Post-9/11 Veterans with mTBI and no TBI.

	Mild TBI	No TBI
Class count, n	3,454	4,447
Class proportion	32.2%	13.5%	p-value
Age
Mean ± SD	28.4 ± 7.8	32.0 ± 9.2	<0.0001
29 years and under	71.8%	53.3%
30–39 years	15.9%	23.4%
40–49 years	9.9%	18.2%
50 + years	2.5%	5.1%
Sex
Men	93.4%	76.4%	<0.0001
Race/ethnicity
White	70.4%	68.0%	<0.0001
Black	12.7%	16.8%
Hispanic	12.5%	11.0%
Asian	2.4%	2.6%
Native American/Pacific Islander	1.3%	1.1%
Unknown	0.7%	0.6%
Education
Less than high school	1.3%	1.2%	<0.0001
High school	85.6%	77.3%
Some college	6.2%	10.2%
College + Higher education	5.4%	10.0%
Unknown	1.6%	1.3%
Marital status
Married	33.2%	45.0%	<0.0001
Branch of service
Army	64.8%	58.4%	<0.0001
Air Force	3.8%	14.1%
Navy/ Coast Guard	7.0%	15.7%
Marines	24.5%	11.9%
Military component
Active	71.9%	64.0%	<0.0001
Military rank
Enlisted	96.9%	93.2%	<0.0001
Deployment history
Multiple deployments	55.2%	49.3%	<0.0001

Table 3.

Complex moderate impact, worsening trajectory baseline demographics for Post-9/11 Veterans with mild TBI and no TBI

	Mild TBI	No TBI
Class count, n	3,393	10,020
Class proportion	31.7%	30.4%	p-value
Age
Mean ± SD	29.9 ± 8.2	33.5 ± 9.6	<0.0001
29 years and under	63.0%	46.6%
30–39 years	20.6%	22.4%
40–49 years	13.8%	24.8%
50 + years	2.6%	6.3%
Sex
Men	93.8%	81.2%	<0.0001
Race/ethnicity
White	67.0%	55.7%	<0.0001
Black	15.6%	26.9%
Hispanic	13.1%	12.8%
Asian	2.0%	2.6%
Native American/Pacific Islander	1.8%	1.2%
Unknown	0.4%	0.9%
Education
Less than high school	1.0%	1.2%	<0.0001
High school	86.2%	77.5%
Some college	6.7%	10.9%
College + Higher education	4.8%	8.9%
Unknown	1.3%	1.5%
Marital status
Married	43.8%	50.1%	<0.0001
Branch of service
Army	71.7%	54.3%	<0.0001
Air Force	4.1%	13.0%
Navy/ Coast Guard	7.2%	21.3%
Marines	17.1%	11.4%
Military component
Active	71.4%	67.5%	<0.0001
Military rank
Enlisted	97.9%	94.7%	<0.0001
Deployment history
Multiple deployments	48.0%	47.3%	0.5343

Table 4.

Complex high impact, stable trajectory baseline demographics for Post-9/11 Veterans with mild TBI and no TBI.

	Mild TBI	No TBI
Class count, n	1,915	3,853
Class proportion	17.9%	11.7%	p-value
Age
Mean ± SD	30.5 ± 7.8	33.2 ± 9.0	<0.0001
29 years and under	56.8%	45.3%
30–39 years	27.0%	28.2%
40–49 years	14.1%	21.0%
50 + years	2.1%	5.5%
Sex
Men	92.9%	79.3%	<0.0001
Race/ethnicity
White	72.7%	66.9%	<0.0001
Black	10.4%	18.6%
Hispanic	12.3%	11.0%
Asian	1.6%	1.6%
Native American/Pacific Islander	2.4%	1.5%
Unknown	0.6%	0.5%
Education
Less than high school	1.1%	1.2%	<0.0001
High school	84.9%	79.8%
Some college	7.8%	11.0%
College + Higher education	4.5%	6.8%
Unknown	1.7%	1.2%
Marital status
Married	49.8%	53.6%	<0.0001
Branch of service
Army	75.4%	60.2%	<0.0001
Air Force	4.5%	14.3%
Navy/ Coast Guard	7.0%	16.6%
Marines	13.2%	8.9%
Military component
Active	70.0%	66.0%	<0.0001
Military rank
Enlisted	97.8%	96.2%	<0.0001
Deployment history
Multiple deployments	43.0%	44.4%	0.3391

Pain Phenotypes

Latent variable mixture models found that five distinct pain trajectories (hereafter pain phenotypes) emerged. Based on the model fit indices (Supplemental Table 3), both mTBI and no TBI cohorts each had four pain phenotypes. Both cohorts had three common phenotypes and each cohort had one unique phenotype. Descriptive names were provided to each pain phenotype based on the diversity of treatment approaches (simple, complex) and the pattern of pain scores (stable, improving, worsening) during years 1 through 5 of VA care.

The pain phenotype with the lowest pain scores (simple low impact, stable pain) represented the largest group among Veterans with no TBI (see figure 2). Treatment was comprised of NSAIDs, with little change in treatment probabilities between years 1 and 5.

Figure 2.

Pain treatment during year 1 and 5 of VA care among the simple low impact, stable pain phenotype.

The second pain phenotype, complex low impact, stable pain (figure 3), found in both mTBI and no TBI cohorts accounted for 32.2% of mTBI and 13.5% of the no TBI cohort. Pain scores did not differ between mTBI and no TBI cohorts, but the treatment patterns were wildly divergent. The no TBI cohort was significantly more likely to receive all types of treatments, with markedly different patterns for opioids, antidepressants, benzodiazepines, z-drugs, interventional pain, and other non-pharmacological approaches with marked and statistically significant increases in those treatments between years 1 and 5. There were smaller, but statistically significant increases in use of NSAIDs, muscle relaxants and opioids between years 1 and 5 for the mTBI cohort. The mTBI cohort was younger, and was more likely to include men, Whites, Marine and Army Veterans, active duty, enlisted, and individuals who were unmarried, had multiple deployments and less education compared to the no TBI cohort.

Figure 3.

Pain treatment during year 1 and 5 among the complex low impact, stable pain phenotype.

The third pain phenotype, complex low impact, worsening pain (figure 4), was found only in the mTBI cohort. Mean pain scores year 1 were 2.64 and progressed to 2.99 by year 5. Treatment patterns for this phenotype were similar to the stable low impact pain no TBI phenotype with high probabilities of all treatment modalities and extremely high probabilities of z-drugs and benzodiazepines in year 1 and marked increases by year 5.

Figure 4.

Pain treatment during year 1 and 5 of VA care among the complex low impact, worsening pain phenotype.

The fourth pain phenotype, complex moderate impact, worsening pain (figure 5), was found in both mTBI and no TBI cohorts in similar proportions. Baseline pain scores were significantly higher among Veterans with mTBI; however, the rate of change in pain scores were similar between both cohorts. Patterns of treatment were similar across TBI strata, however probabilities of treatment with many of the modalities (e.g., antimigraine, anticonvulsants, hydrocodone, benzodiazepines, interventional pain) were significantly higher at all time points for the mTBI cohort. Probabilities for use of many modalities (e.g., NSAIDs, antidepressants, anticonvulsants, opioids, interventional pain) increased significantly between years 1 and 5. The mTBI phenotype was younger, and more likely to include men, Whites, Marines and Army Veterans, active duty, enlisted, and individuals with less education and those who were unmarried in comparison to the no TBI cohort.

Figure 5.

Pain treatment during year 1 and 5 of VA care among the complex moderate impact, worsening pain phenotype.

The fifth pain phenotype, complex high impact, stable pain (figure 6), was found in both mTBI and no TBI cohorts, but was more common in the mTBI cohort (17.9% vs. 11.7%). While mean pain scores were stable across time, they were significantly higher in the mTBI versus no TBI cohort (4.86 vs. 4.41, p<.0001). Treatment patterns were similar for mTBI and no TBI, with both cohorts having high use of all treatment modalities and statistically significant increases in the probability of use of many treatment modalities, notably opioids, benzodiazepines, and z-drugs, over time. Individuals in this trajectory also had the highest probability of receiving interventional pain treatment in years 1 and 5. The mTBI cohort was younger, was more likely to include men, Whites, unmarried Veterans, individual with lower education, active duty, Marines and Army in comparison to the no TBI cohort.

Figure 6.

Pain treatment during year 1 and 5 of VA care among the complex high impact, stable pain phenotype.

DISCUSSION

This study occurs in the context of growing interest in the use of health informatics to assess pain management and outcomes. We used real-world data collected during clinical care to characterize pain and the treatment of pain in Post-9/11 Veterans. While differences in pain score trajectories among phenotypes were subtle (means ranged from 1.31–4.86), treatment trajectories were distinct and characterized by treatment propensity of different modalities. As expected, Veterans with mTBI demonstrated some different pain trajectories than those without TBI. While Veterans with mTBI reported statistically significant (higher) levels of pain intensity compared to Veterans with no TBI, pain intensity only varied by one NRS point. Despite modest differences in pain ratings, Veterans with mTBI had more intensive patterns of treatment use suggesting a greater need for care to address their pain. The present study focused on NRS pain scores over other measures (e.g., disability, pain interference) as NRS ratings are readily available in the EHR compared to functional measures. Repeated pain scores, even at low frequency, identified meaningful trajectories that (a) differentiate Veterans with chronic pain and comorbid mTBI from those without and (b) correspond with clinically meaningful differences in pain management.

Of note, we did not find a simple low impact, stable pain phenotype in the mTBI group. Rather, the lowest impact group for mTBI was the complex low impact, stable pain phenotype where relatively low pain scores were accompanied by modest probability of treatment with NSAID, opioids, OT/PT and benzodiazepines. There was, however, a strikingly different pattern of treatment complexity for the no TBI counterpart with dramatically higher probabilities of opioids, psychotropics, and interventional pain treatment. The more complex treatment pattern that mirrored the pain score trajectory of the complex low impact pain, stable no TBI phenotype was found in the mTBI complex low impact pain with worsening phenotype, which did not have a no TBI counterpart. Mean pain scores started higher in the mTBI worsening phenotype compared with the no TBI stable phenotype (mean 2.61 vs. 2.12) and gradually increased despite dramatic increases in diverse treatment approaches. These findings suggest that mental health comorbidity in mTBI is associated with higher pain scores than in individuals with no TBI. Alternatively, the significantly higher probability of opioid and interventional pain treatment in those with no TBI with lower pain scores suggests clinicians may treat individuals without TBI differently and possibly more aggressively at lower reported pain scores.

In contrast, those with mTBI had significantly higher probabilities than those with no TBI for use of psychotropics, opioids, interventional pain and CIH in the complex moderate impact pain with worsening and complex high impact pain phenotypes. This may be due to the significantly higher pain scores in those with mTBI. Assuming providers are utilizing similar treatments for common pain syndromes in the mTBI and no TBI groups, the higher baseline pain scores in the mTBI group may reflect the presence of unmanaged co-morbid conditions and central sensitization of pain. This suggests those who have TBI experience pain and pain-related symptoms differently. In addition, having long-term mTBI related symptoms and multimorbidity may necessitate the use of CIH. To fully address this possibility, future studies will need to examine differences in treatments between those with and without mTBI diagnosis, while controlling for pain conditions being addressed.

In all phenotypes except simple low impact pain, we found significantly higher probabilities of psychotropic, opioid, anticonvulsant, interventional pain, PT/OT and CIH use in light of either stable or slightly increasing pain scores. Moreover, the data suggest that individuals with the highest reported pain scores receive multimodal treatments as recommended by VA-DoD guidelines, with high rates of interventional pain treatment, and the highest rates of CIH. These intensive treatments over time were not associated with reductions in pain scores; rather, pain remained stable or increased slightly. Thus, it appears that intensive pain management was required to maintain relatively stable pain—or perhaps a level of functionality that allowed Veterans to maintain work, family and social roles.

Strengths and Limitations

Our study cohort required at least five years of continuous care; therefore, those who met inclusion criteria may be complex patients with higher disability rates. Such a possibility remains to be verified by generalizability analysis that accounts for missing data mechanisms. In addition, individual pain ratings derived from EHR may not directly reflect other measures of pain administered at a clinical encounter.¹⁶ However, using multiple pain ratings to establish a central tendency of pain measurement does improve the validity of the NRS.¹⁷ As we examined healthcare use as it occurs in routine clinical care, observations were limited to VA encounters and medications on the VA formulary. However, the majority of pain pharmacotherapy was captured as we focused on medications that are recommended for chronic pain management.^5,6,34,35 Finally, given the decent entropies found in our LGMM analyses (0.873 and 0.887 for mTBI and no TBI groups), we applied the pseudoclass approach for the comparison of baseline characteristics between the mTBI and no TBI groups among those with similar trajectory patterns or between trajectory classes conditioned on TBI status. According to a recent simulation study by Asparouhov and Muthen² further examination of predictors for pain trajectory classes using the Bolck-Croon-Hagenaars approach²⁸ may further verify the results derived from the pseudoclass approach.

Implications for research and clinical practice

Given the increasing availability of administrative data developed during clinical care, methods used in this project can be translated for use in other clinical settings. Our study found that even a single pain rating per year can identify trajectories of pain and treatment, so VA care providers would likely benefit from electronic health data that show trajectories of pain ratings across the Veterans care history. Baseline measures are essential to making judgments about the overall impact of treatment for pain, but adding trajectories of pain over time, even as a simple line graph in the EHR, may expose a clinically meaningful trajectory of pain that could highlight symptom complexity and/or the need for more comprehensive care.

We found differences among the pain phenotypes which may have a significant impact on the success of pain interventions. As such we believe the trajectories identified in this study could provide a meaningful conceptual foundation for future studies on the role of pain function in light of treatment for individuals with mental health comorbidities (e.g., PTSD, depression). Stratifying results along characteristics relevant to our trajectories may identify key subgroups that may or may not benefit from specific treatments. Effect sizes associated with pain trajectory classes and TBI status provided in Supplemental Tables 4 and 5 could help design future studies. Targeted follow-up survey and interviews of individuals from each of the pain phenotypes may provide insight into how individual differences (e.g., anxiety, pain catastrophizing, pain avoidance), activity levels, and perceptions of treatment may lead to specific patterns, and the impact of those patterns on patient lives.

We found the majority of pain treatments increased over 5 years with little or no impact on pain scores. However, use of non-pharmacological therapies was relatively low over the study period. This study period was prior to the advent of Whole Health, so availability of non-pharmacological pain management options may be greater now, or may reflect lack of service availability based on location of care. The finding that those with mTBI had higher levels of comprehensive care than their no TBI counterparts may reflect either symptom complexity and/or access to more comprehensive pain management care through the polytrauma system of care. Furthermore, geographic location may influence availability of non-pharmacological therapies. Systematic study of variation in care is needed to better understand targets for intervention to improve pain management across the VHA and other health systems.

Patterns of treatment also revealed potential safety issues. High NSAID use was observed in all pain score trajectories, especially by year five of VA care. Similar to opioid use precautions, serious risks exist with high dose and long-term use of NSAIDs.^9,15, In addition to the gastrointestinal (GI) bleeding and ulcers, long-term NSAID use has unintended consequences on pharmacotherapy options.^1,6 Patients’ on NSAIDs long-term may not tolerate antidepressants due to the GI side effects associated with both medications. Periodically assessing over-the-counter NSAID use may help prevent GI distress associated with chronic use and increase the likelihood of antidepressant use as depression and anxiety commonly occur with pain.

Treatment patterns also suggest a possibility for unintentional overdose in those phenotypes with multiple CNS medications. A systematic evaluation of the association of pain phenotypes on suicide related behaviors and overdose is needed to identify subgroups of Post-9/11 Veterans who may be ideal candidates for increased monitoring and intensive counseling. Additional analysis linking Department of Defense data prior to entering VA and emergence of comorbid conditions over time will further clarify the longitudinal path of pain, pain treatment and the role of TBI and co-occuring conditions.

CONCLUSION

In our 5-year longitudinal analysis of pain scores and pain treatment, we found five distinct trajectories of pain intensity, with three relatively stable and two showing slight worsening. We also found clear differences in pain treatment between those with and without mild TBI. As such, we propose tailored chronic pain interventions for Post-9/11 Veterans with mTBI. Classification of Post-9/11 Veterans into these trajectories may advance the management of chronic pain and provide direction for future studies among Veterans with mTBI.

Perspective:

The complexity of pain in patients with mTBI is categorically different than those with no TBI. Pain in patients with mTBI is heterogeneous with distinct phenotypes which may explain poor outcomes in this group. Identification of the individual differences may have a significant impact on the success of interventions.