Healthcare Costs Associated With Chronic Opioid Use and Fibromayalgia Syndrome

, , ,
The American Journal of Pharmacy Benefits, November/December 2014, Volume 6, Issue 6

While fibromyalgia is associated with a small increase in healthcare costs compared with matched controls, chronic opioid use in these patients results in a large increase.

Fibromyalgia (FM) is an idiopathic, functional syndrome characterized by chronic widespread pain and diffuse tenderness.1 This disorder affects more than 6 million patients in the United States, the majority being female, and is associated with significant clinical and economic burdens to patients, the healthcare system, and society as a whole.2,3 FM patients show elevated healthcare costs compared to the general population, but similar costs to patients suffering from ankylosing spondylitis and rheumatoid arthritis.4-7 Additionally, a study published by Palacio et al found FM costs were elevated compared with a control group matched on age, sex, and a comorbidity index.8 While that study utilized the comorbidity index to more accurately capture a measure of cost differences, it is unlikely this measure alone was sufficient to control for the difference. Another recent study showed that FM treatment patterns vary widely based on geographic location.9 Other potential sources of variability that have not been controlled previously in the FM literature include concomitant medications and diagnosing provider.

Current treatment recommendations suggest using a multimodal approach to medication management, including both drugs approved by the FDA to treat FM (eg, pregabalin, duloxetine, minalcipran) and older medications not approved but that show evidence of efficacy (eg, tricyclic antidepressants, selective serotonin reuptake inhibitors, serotonin and norepinephrine reuptake inhibitors).10-12 However, FM treatment guidelines caution against chronic opioid medication use, citing a lack of evidence to support this practice and the harmful effects commonly associated with this type of use.13 Furthermore, this lack of efficacy is coupled with the unique psychological and physiological characteristics of FM patients14,15 that put them at increased risk of abuse16 and increased likelihood of adverse effects.17 This combination of factors makes chronic use of opioid medications for the treatment of FM troubling. Moreover, chronic opioid use may actually be associated with worse health outcomes in these patients.

The aim of this study is to estimate the cost associated with a FM diagnosis and the impact of chronic opioid use on the healthcare costs of FM patients. Using a large retrospective cohort of commercially insured patients, we first estimated the healthcare costs associated with a FM diagnosis using a propensity score—matched group of controls similar across age, sex, comobidity burden, diagnosing provider, and geographic location. Next, we examined the impact of chronic opioid use on FM patients’ healthcare costs using a stratified sample across various patient groups. Finally, we analyzed the interaction effect of chronic opioid use in FM patients. Our results provide the most methodologically accurate estimate to date of the financial effects of chronic opioid use in FM patients.

METHODSData Source

Our research team obtained a license to use deidentified patient health claims information from a large commercially insured population for the period January 1, 2007, to December 31, 2009. The University of Kentucky Institutional Review Board provided approval for this study. The data set was a nationally representative sample of employed, commercially insured individuals with dependents, and included 15 million patients annually across the United States. Data were collected at the patient level and linked across administrative and health records, including: administrative data (plan type, gender, age, eligibility date spans); pharmacy claims data (national drug code, strength, quantity and date dispensed, days supplied, pricing); physician and facility claims (physician or facility code, procedure codes, diagnosis codes, revenue codes, diagnosis related group, service dates, pricing); and lab results (logical observation identifiers names and codes, lab test name, and result). For the purposes of this study the entire 3-year data slice was considered as a single cross-section.

Study Population and Characteristics

The data set was queried for patients with FM as identified by the International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) code 729.1 (myalgia and myositis, unspecified). Patients with at least 1 claim between January 1, 2007, and December 31, 2009, were included in the sample. Only patients with at least 1 year of continuous eligibility between ages 18 and 64 years were considered for this study; patients with malignancies or missing data were excluded. A control group was selected using the same inclusion/exclusion criteria used for the cases, except where FM was identified, and controls were identified based on a list of related disease states (

Table 1

).3 These pain conditions and conditions with ill-defined symptoms were chosen because each has been shown to be strongly associated with a FM diagnosis.3 To divide cases and controls, chronic opioid use was defined as receiving a supply of opioid therapy greater than 183 days during a 12-month period of eligibility.18

Two outcome variables of interest were collected: medical costs and prescription costs, based on medical and prescription charges from the claims data. Medical costs included all charges (from the payer and patient) for inpatient and outpatient claims during the eligibility period. Each cost type was summed at the patient level, then annualized to control for differing eligibility spans across the sample. The top and bottom 1% of annualized medical and annualized prescription charges were dropped to control for outliers. Patient-level characteristics included age, sex, state of residence, comorbid conditions, and concurrent medication use by class. Initial diagnosing provider type for FM (eg, internal medicine, rheumatology, nurse practitioner) and the diagnosing provider type for the first related condition diagnosed in the data for controls were also included.

Statistical Analysis

Descriptive statistics including age, sex, comorbid conditions, concurrent medications, diagnosing provider type, and geographic location were calculated for all patients before and after propensity score matching. (The values for all covariates before and after matching can be seen in the

eAppendix

.) Three analyses were carried out using these data.

Analysis 1 examined the statistical significance of cost differences between FM patients and controls. Controls were selected from the larger cohort by using a first-stage logistic regression that identified a patient’s propensity for being diagnosed with FM based on all the characteristics described above. Using a greedy, nearest-neighbor algorithm, patients were then matched based on this propensity.

Analysis 2 compared costs between FM patients receiving chronic opioid therapy with those using nonopioid medication(s) or interventions. This analysis constructed a similar propensity score algorithm to the one previously described, with the only difference being the propensity to receive chronic opioid therapy rather than the propensity to have a diagnosis of FM. This analysis used propensity stratification rather than matching.

Analysis 3 examined the cost differences between controls and FM patients receiving chronic opioid therapy. Controls were selected from the full cohort and matched using a first-stage logistic regression that obtained the propensity for receiving chronic opioid therapy as a FM patient based on the same set of characteristics as the other analyses. Finally, we performed a linear regression with the data from the first and third analyses using cost as the dependent variable. Multiple models were used to ascertain the contribution of FM diagnosis, chronic opioid therapy, and the interaction of the 2 terms to costs. Data extraction was completed using Oracle SQL Developer and SAS v9.2, the matching process was carried out using Fortran, and all statistical analyses were completed in STATA v12.

RESULTSCosts Associated With FM

The first analysis matched 494,396 patients with FM to 1,919,409 patients based on their propensity to have a FM diagnosis as determined by demographics, comorbidities, medication usage, diagnosing provider types, and geographic location. The resulting comparison groups each contained 445,912 (90.2% of cases) patients. The coefficient of variation for patient characteristics was reduced from 20% to 4% based on this one-to-one matching algorithm. While some variables remained statistically different, the overall groups were balanced based on propensity to receive a FM diagnosis. (eAppendix)

Each group was an average age of 44 years, and two-thirds were female. As expected in a population with FM, pain comorbidities including back pain, neuropathic pain disorders, fatigue, and sleep disorders were present at elevated rates compared with the general population.19-22 Furthermore, medication classes typically used to treat these disorders—including benzodiazepines, nonsteroidal anti-inflammatory drugs (NSAIDs), hypnotics, and muscle relaxants—saw increased use. The diagnosing provider varied, but chiropractors and primary care physicians were most common. Overall, medical and prescription costs were higher for FM patients before and after the matching process (as shown in

Figure 1

). The matching algorithm resulted in a decrease in the difference of medical and prescription costs of 90% and 80%, respectively. Differences of just 1% for medical costs between FM patients and controls ($13,941 vs $13,802; P <.01) and 5% for prescription costs ($2463 vs $1974; P <.01) remained for the matched groups. As can be seen in

Table 2

, the effect of a FM diagnosis was negative (—$83.54; 95% CI, –152.55 to –16.53) on medical costs and small on prescription costs ($120.31; 95% CI, 109.98-130.62) after controlling for all other factors.

Costs Associated With Chronic Opioid Use in FM

The second analysis examined the effect of chronic opioid use only in patients with FM. The sample included 494,396 FM patients, of which 10% were chronic opioid users. Chronic opioid users were significantly older (46.8 years vs 43.9 years; P <.01) and more likely to be female (73.8% vs 65.4%; P <.01). They were also less likely to be diagnosed by a chiropractor and more likely to be diagnosed by a specialist—especially a pain management specialist. In order to control for the differences seen in these 2 groups, patients were stratified into deciles based on their propensity to receive chronic opioid therapy. Figure 2 shows the results of this stratification on medical and prescription cost ratios between chronic opioid users and nonusers. The medical costs for FM patients receiving chronic opioid treatment ranged from 1.4 to 2.4 times more than those not using chronic opioid medication(s). FM patients with the lowest propensity to receive opioids showed the highest cost disparity. Prescription costs followed a similar, albeit more pronounced pattern, varying from 2.2 at the high propensity end to 7.2 at the lowest.

Effect of the Interaction of Chronic Opioid Use and FM on Costs

The final analysis used the total population of 2,413,805 FM and control patients; of these 50,159 were FM patients receiving chronic opioid therapy. Using the same matching algorithm from the first analysis, patients were matched one-to-one based on their propensity to be a FM patient receiving chronic opioid therapy. This resulted in 2 groups, each comprised of 48,333 (96.4% of cases) patients. The matching algorithm reduced variation between the groups by 90%. While some variables were still significanlty different, overall, the groups were balanced across observed covariates. (eAppendix)

Each group was an average of 47 years-old, and twothirds were female. As seen in the first analysis, pain comorbidities, including back pain, neuropathic pain disorders, fatigue, and sleep disorders, were present at rates elevated from the general population.19-22 Again, medication classes typically used to treat these disorders saw increased use (ie, benzodiazepines, NSAIDs, hypnotics, and muscle relaxants). Provider differences were similar to the first analysis as well.

Figure 3

illustrates the differences in medical and prescription costs between the 2 groups before matching. The matching process was similarly effective at reducing the differences seen in medical and prescription costs by 90% and 80%, respectively. However, even after matching based on propensity to receive chronic opioid therapy as a FM patient, medical ($28,209 vs $24,471) and prescription ($7012 vs $4861) costs were elevated by 15% and 44%, respectively.

Using linear regression, Table 2 shows the effect of a FM diagnosis and chronic opioid use when controlling for all other factors associated with the propensity to be a chronic opioid user with FM. In this case, the FM diagnosis was negatively associated with medical costs (—$233.63; 95% CI, –298.14 to –169.12) and had a small positive association with prescription costs ($133.43; 95% CI, 124.38-142.48). However, chronic opioid use strongly affected both medical ($9094.05; 95% CI, 8924.79-9263.31) and prescription ($3391.81; 95% CI, 3368.84-3414.79) costs.

DISCUSSION

Our study provides further explanation concerning healthcare costs in FM patients compared with controls. Palacio et al8 published the most methodologically advanced analysis of these costs through the use of an age, sex, and comorbidity index—matched control group. Their findings agreed with past studies showing an increase in healthcare costs of many types.3,7 Previous studies showed greater differences in costs associated with FM compared with Palacio et al,8 which Palacio et al attributed to their use of a comorbidity index. Similarly, our study shows that with more robust matching, taking into account not only comorbidities but also concurrent medications, location, and diagnosing provider, the difference attributable to the diagnosis of FM is further reduced. While an unmatched comparison of medical and prescription costs showed a very large increase, matching decreased this difference by a great deal.

One concerning thread throughout the FM literature is the increased use of opioid medications to treat pain in patients with FM.8,23 Our study used 2 approaches to examine the extent to which opioid medication(s) affect the outcomes of FM patients. With the first approach, we stratified FM patients according to their likelihood to receive chronic opioid therapy (ie, the propensity attributed to these patients based on the observed, included variables). Using the stratification schema described above, we showed a significant relationship between patient propensity to receive chronic opioid therapy and the nondesirable economic consequences that were associated with this treatment decision. Patients with FM, but without other contributing factors (eg, comorbid pain conditions) that received chronic opioid medication(s) had medical costs 150% higher than those that did not receive this treatment. However, FM patients with other contributing factors (eg, comobidities, concurrent medication use) only saw a 50% increase in costs associated with chronic opioid medication use. The difference in the highest stratum and the lowest stratum is even more pronounced for prescription costs.

In order to better control for the differences seen in the first 2 analyses and to better attribute them to either the disease or the treatment choice, a difference-in-difference approach was utilized. Using the entire sample and matching one-to-one based on the interaction of FM and chronic opioid therapy, we calculated the differences in costs attributable to chronic opioid use in FM. As in the first analysis, matching reduced the difference seen by nearly a factor of 10. Despite this reduction, however, there remained statistically and clinically significant increases associated with chronic opioid use in FM for both medical and prescripiton costs. It is important to note that these differences were not attributable to a diagnosis of FM or to the chronic use of opioids in general, as each of these characteristics were factored into the analysis. The differences seen were associated with the interaction of these terms (ie, the specific effect seen only when FM patients received chronic opioid therapy).

Finally, to monetize the contributions a FM diagnosis and chronic opioid use had on healthcare costs, we used a 2-stage analysis. This analysis found that for 2 patients who were similar in age, gender, comorbidities, concurrent medications, and geographic location, a FM diagnosis did not contribute much to overall healthcare costs. The diagnosis was associated with lower medical costs (1%) but higher prescription costs (17%). Similarly, when controlling for the propensity to be a chronic opioid user with FM, medical costs slightly decreased (2%) while prescription costs increased (9.2%). In contrast, when controlling for these same characteristics, opioid use was associated with a large increase in both medical (84%) and prescription costs (230%).

Our findings have important implications for providers managing patients with FM. They emphasize the need to use nonopioid pharmacologic and nonpharmacologic strategies to manage this disorder.24 While economic outcomes are an imperfect proxy for the quality of clinical care, our study clearly demonstrates 3 findings: 1) the increased costs associated with a FM diagnosis in the literature to date is at least partially due to the use of dissimilar control groups; 2) the economic costs associated with chronic opioid use in FM patients is substantial; and 3) the less complicated a FM patient’s clinical situation (eg fewer comorbidities, less concurrent medication use), the greater the increase in costs associated with chronic opioid use. These findings signal the adverse economic impact chronic opioid use has on FM patients. Moreover, when considering these economic findings as a representation of a patient’s clinical outcomes, the effect of chronic opioid medications on FM patients is disconcerting. There are many factors that contribute to prescribing opioids for patients with chronic pain (eg, discipline-specific practices, familiarity with or availability of nonpharmacologic strategies in a particular geographic region, culturally defined acceptance of a certain level of pain). However, our data clearly demonstrated that the use of opioids in FM, after controlling for all of the measured variables available in our data set, increased medical costs and possibly worsened health outcomes.

Limitations

In addition to the limitations generally applicable to secondary database research—such as the possibility of miscoding, the limited number of coded fields, and the use of charges as a proxy for actual healthcare costs—a few other limitations are worth noting. First, the validity of ICD-9 -CM code 729.1 as an identifier of FM within secondary databases has not been tested; however, the decision to use this code was consistent with previous research in this area. Second, because our analysis was limited to commercially insured patients between the ages of 18 and 64 years, the generalization of the findings was limited to this population. Third, the cross-sectional nature of the study did not allow us to assess the temporary nature of the exposure and the outcome. Also, 10% of the FM cases initially identified were lost during the matching process; these cases were generally sicker than those that were matched. This loss resulted in a conservative estimation of the differences between cases and controls. Although the matching process drastically reduced the variation between groups, minor differences in the matched groups were seen for individual variables. Finally, there was a possibility of “overmatching” due to the lengthy list of variables controlled for in the propensity score. Each group of variables was shown to vary significantly among FM patients in our analysis and in other published work, however.3,9 Given this variation and the large number of matched pairs available for analysis, overmatching was not a significant concern.

While differences seen in FM patients and controls were marginal, those attributed to chronic opioid use in these same patients were considerably higher. The utilization of chronic opioid therapy in the treatment of FM is a practice based not on evidence available to practitioners, but on other variables, both observed and unobserved. Given the profound lack of evidence supporting the use of opioids in FM, their prevalence as a treatment option remains a mystery. Coupled with the increasing armamentarium that has shown evidence of safety and efficacy and the clear societal and personal adverse effects of chronic opioid use, the prevalence of opioid medication use in FM is very troubling. Beyond all of these concerns—which are common to the treatment of most chronic, nonmalignant pain conditions—the psychology and physiology of FM combined with this new evidence of poorer outcomes further supports recommendations against opioid use in this patient population.