Longitudinal patterns of analgesic and central acting drug use and associated effectiveness in fibromyalgia


  • Funding sources

    No funding support was received for this study.

  • Conflicts of interest

    Drs. Wolfe, Michaud and Katz declare no conflicts of interest. Dr. Lee has stock in Merck, Novartis and Elan Corporations, and has a research grant from Forest Laboratories. Dr. Häuser received honoraria for education lectures from Eli Lilly, Pfizer and Janssen-Cilag, and a congress travel grant by Eli-Lilly.


Frederick Wolfe

E-mail: fwolfe@arthritis-research.org



We describe the changing pattern of analgesic and new central acting drug (NCAD) use (pregabalin, duloxetine, milnacipran) in fibromyalgia and measure NCAD effectiveness in clinical practice.


About 3123 US adult patients with fibromyalgia participated in an 11-year longitudinal study of fibromyalgia outcomes. We assessed severity-adjusted treatment prevalence and measured the effect of any use of NCAD on pain and fatigue, and functional status using the Health Assessment Questionnaire (HAQ) disability index.


In 2010, 46.7% of patients used opioids, including 12.5% who used strong opioids. During the 11 years, severity-adjusted strong opioid use increased from 6.3% to 11.7% and any opioid use from 40.0% to 46.6%. Nonsteroidal anti-inflammatory drug (NSAID) use decreased from 74% to 44%. Tricyclic use dropped in half, from 27% to 15%, while NCAD use increased from less than 10% to 39%. The estimated 25th and 50th percentiles for NCAD discontinuation time were 1 and 2.5 years. Overall pain, fatigue and HAQ scores were unchanged over the 11 years. For patients treated with NCAD, pain scores were reduced significantly by 0.17 (0.03, 0.30) units following the start of NCAD, an improvement of 2.8%. Some sensitivity analyses showed improvements of up to 4.3%. There was no significant improvement in fatigue or functional status.


There is a changing pattern of drug treatment in fibromyalgia, consisting mostly of decreased NSAID and amitriptyline use and an increase in NCAD. Drug costs are substantially higher because of NCAD use, but we found no evidence of clinical benefit for NCAD compared with prior therapy.

What's already known about this topic?

  • Analgesic and centrally acting drug use is common among fibromyalgia patients.

What does this study add?

  • With time there has been a switch from NSAIDs and tricyclic antidepressants to newer and more expensive centrally acting drugs. We are unable to detect benefit from newer therapies as they are used in clinical practice.

From the hundreds of articles and medical expert group recommendations, a variable consensus among treatment experts has emerged regarding treatment of fibromyalgia. Briefly, there is an agreement that fibromyalgia is best treated by multimodal therapy that includes exercise, cognitive behavioural therapy (CBT) and pharmacotherapy (Hauser et al., 2010b). In the United States, except for self-therapies such as exercise and weight loss, however, the main treatment for fibromyalgia falls to pharmacotherapy, because it is ordinarily difficult or impossible to obtain insurance company approval for CBT and for therapies that include CBT, balneotherapy, hypnotherapy, bio-feedback, massage, acupuncture or psychotherapy.

Therapeutic recommendations can be summed up as often being contradictory, as shown in Häuser's systematic review of the recommendations of the European League against Rheumatism (EULAR), the American Pain Society (APS) and the Association of Medical Societies of Germany (AWMF) (Hauser et al., 2010b). For example, all groups recommend against the use of nonsteroidal anti-inflammatory drugs (NSAIDs). EULAR and AWMF did not recommend strong opioids, and APS found uncertain evidence for their efficacy. Tramadol received an ‘A’ recommendation by EULAR and a ‘C’ recommendation by AWMF. By contrast, drugs that are ‘centrally acting’ were generally recommended. These agents include tricyclic antidepressants (TCAs), anticonvulsants such as pregabalin, and norepinephrine–serotonin reuptake inhibitors such as duloxetine and milnacipran. These new centrally acting drugs (NCADs) have generally been shown to be superior to placebo. Meta-analyses indicate that there is generally ‘strong evidence’ for efficacy for NCAD (Hauser et al., 2010a). However, the degree of efficacy is small, with effect sizes for pain in the neighbourhood of 0.35; and there is almost no effect on fatigue, a central and key symptom of fibromyalgia (Hauser et al., 2010a). While the US Food and Drug Administration (FDA) has approved pregabalin, duloxetine and milnacipran for the treatment of fibromyalgia, the European Medicines Agency (EMEA) refused to approve them based on the ratio of benefit to adverse effects (Briley, 2009; Häuser et al., 2011).

Even so, there is virtually no evidence about effectiveness of fibromyalgia treatment in the community. In the report below, we briefly describe current therapies and changes in therapies that have occurred over the last 10 years, during which time pregabalin, duloxetine and milnacipran were approved by the FDA and intensively marketed to physicians and directly to patients (Barker, 2011). We measure the effectiveness of centrally acting drug therapies by comparing pain, fatigue and functional status prior to the use of these drugs with values obtained during and after the use of the drugs.

1. Methods

We studied 3123 adult patients with fibromyalgia who participated in the National Data Bank for Rheumatic Diseases (NDB) longitudinal study of fibromyalgia outcomes (Wolfe and Michaud, 2010b). Participants were volunteers recruited from the practices of US rheumatologists (75%), or by self-referral (25%) who completed mailed or Internet questionnaires about their health at 6-month intervals between 2000 and 2010. Patients were not compensated for their participation. Diagnosis was made or verified by the patient's rheumatologist. The NDB utilizes an open cohort design in which patients are enrolled continuously. The average date of entry into the study was the first quarter of 2004.

At each assessment, in addition to demographic data, we collected self-report data on all prescription and non-prescription pharmacology treatments as well as severity measurements. In the current report, we concentrate on three key outcome variables (Wolfe et al., 2011), visual analogue scales (VASs) for pain and fatigue, and functional status as measured by the Health Assessment Questionnaire (HAQ) disability index.

Patients were categorized as using one of the study drugs if they used it at any time during a 6-month reporting period. We categorized strong opioids as including hydromorphone, dihydromorphinone, oxymorphone, butorphanol, methadone, morphine, oxycodone, meperidine and fentanyl. Weak opioids included hydrocodone, tramadol, codeine, pentazocine and propoxyphene. NSAIDs were any of more than 20 drugs that were approved and marketed in the United States during the study years, including COX-II inhibitors. Non-opioid drugs included acetaminophen and aspirin. Aspirin was not included in the NSAID group. Anticonvulsants included pregabalin and gabapentin, the latter drug having been used widely in fibromyalgia, although not approved for use by FDA. Selective serotonin–norepinephrine reuptake inhibitors (SNRIs) included in this study were duloxetine and milnacipran. To primarily study FDA-approved drugs, we combined pregabalin, duloxetine and milnacipran in a category of NCADs, and we added gabapentin to this grouping because of its similarity to pregabalin.

1.1 Statistical methods

Of the 3123 patients, an average of 1740 participated in each of the 21 semi-annual assessments, reflecting new enrolments over the entire course of the study and dropouts (withdrawal of consent). The average number of semi-annual questionnaires completed by each patient was 6.2, and the total number of questionnaires studied was 19,201.

To adjust for differences in patients over time, we estimated treatment prevalence (‘severity-adjusted’ prevalence) and severity data (pain, fatigue and HAQ) based on longitudinal generalized estimating equation (GEE) analyses that used all observations and controlled for age, sex, ethnicity, education, smoking status, self-referral, body mass index, marital status, total income, VAS pain, VAS fatigue and HAQ. Predictions of severity data omitted VAS pain, fatigue and HAQ as independent predictors. The annual predicted values represent the average marginal effect at each study year. For clarity, the confidence intervals (CIs) at each predicted year are omitted from the graphs, but are available on request from the authors.

To estimate the benefit of NCAD, we studied patients who were not receiving NCAD at study admission and subsequently started such therapy. Using GEE analyses, we compared pre- and post-NCAD observations for pain, fatigue and HAQ. Post-NCAD observations were any observation after the start of any NCAD whether or not the patient was still on an NCAD. In sensitivity analyses, we restricted analysis to patients who had received NCAD in at least two 6-month periods. In another analysis, we restricted analysis to post-NCAD observation during which NCAD was used. We also analysed the data after removing all patients treated only with gabapentin, as gabapentin was not an FDA-approved therapy.

Data were analysed using Stata version 12.0 (Corporation, 2011).

This study was approved by the Via Christi Institutional Review Board of Wichita, Kansas, USA.

2. Results

The mean age [standard deviation (SD)] of the fibromyalgia participants was 54.1 (12.6) years at entry, and 95.4% were women. The mean (SD) duration of fibromyalgia symptoms was 13.8 (10.7).

2.1 Non-central acting analgesics

In 2010, 46.7% of patients used opioids, including 12.5% who used strong opioids (Table 1). During the 11-year period of the study, severity-adjusted strong opioid use increased from 6.3% to 11.7%, weak opioid use from 35.6% to 39.6% (Fig. 1), and any opioid use from 40.0% to 46.6%. The most commonly used weak opioids in 2010 were an acetaminophen–hydrocodone combination (19.2%) followed by tramadol (18.3%). Holding all covariates at their means, the annual probability of strong opioid use increased by 0.6% (0.4, 0.9), weak opioid use increased by 0.6% (0.2, 1.0), and any opioid use increased by 0.9% (0.5, 1.3). In 2010, NSAID use was 46.7%. On average, NSAID use decreased by 3.0% (2.6, 3.5) annually.

Figure 1.

Severity-adjusted percentages of patients using analgesic drugs in 2000–2010. NSAID, nonsteroidal anti-inflammatory drug.

Table 1. Percent use of analgesics and centrally acting drugs, and severity scores in 2010
VariablePercent use in (2010)Pain (0–10)Fatigue (0–10)HAQ (0–3)
Mean (SD)Mean (SD)Mean (SD)
  1. HAQ, Health Assessment Questionnaire; NSAIDs, nonsteroidal anti-inflammatory drugs; SD, standard deviation; SNRIs, serotonin–norepinephrine reuptake inhibitors.
Strong opioids (%)12.56.6 (2.1)7.2 (2.2)1.4 (0.6)
Weak opioids (%)38.76.1 (2.3)6.7 (2.6)1.3 (0.7)
Any opioid (%)46.76.2 (2.3)6.7 (2.6)1.3 (0.6)
Any non-opioid (%)38.45.3 (2.6)6.2 (2.9)1.1 (0.7)
Any non-NSAID analgesic (%)67.55.7 (2.5)6.4 (2.8)1.2 (0.7)
Any NSAID (%)46.75.4 (2.5)6.1 (2.8)1.1 (0.7)
Any analgesic (%)82.85.5 (2.6)6.2 (2.8)1.2 (0.7)
Central acting drugs    
Tricyclic antidepressants (%)15.25.4 (2.7)6.1 (2.8)1.1 (0.6)
Anticonvulsants: gabapentin and pregabalin) (%)29.86.1 (2.4)6.7 (2.5)1.3 (0.7)
SNRIs – duloxetine and milnacipran (%)22.36.0 (2.3)6.8 (2.6)1.2 (0.6)
Anticonvulsants and SNRIs (%)43.76.0 (2.4)6.7 (2.6)1.3 (0.7)
All central acting drugs (%)52.85.8 (2.5)6.5 (2.7)1.2 (0.7)

2.2 Centrally acting agents

Overall, the severity-adjusted use of centrally acting agents increased from 33.8% in 2000 to 48.8% in 2010 (Fig. 2), despite a decrease in TCA use from 26.2% in 2000 to 15.4% in 2010. The increase in use of central acting agents was driven by SNRIs that increased from 0% in 2003 to 20.4% in 2010, and by anticonvulsants that increased from a baseline of 8.7% in 2000 to 26.4% in 2010. By 2010, use of anticonvulsants (primarily pregabalin) and SNRIs (duloxetine and milnacipran) together had increased to 39.5%. There was an overlap in the observed use of these drugs in 2010. Of the 548 patients who had used one or more of the drugs in this period, 82.7% had used one of the three approved NCADs, and 28.8% had used a TCA.

Figure 2.

Severity-adjusted percentages of patients using centrally acting drugs in 2000–2010. CNS, central nervous system; NCAD, new central acting drug; SNRI, serotonin–norepinephrine reuptake inhibitor.

2.3 Clinical severity and its change

In 2010, the mean (SD) of the three outcome variables was: pain 5.4 (2.6), patient global 5.0 (2.4), and HAQ 1.11 (0.67). Their changes over time are shown in Fig. 3. Adjusted for covariates, on average, pain decreased by 0.03 (0.01, 0.05) units annually, or approximately 0.3 units per 10 years. Similarly, the covariate-adjusted annual decrease in patient global was 0.03 (0.01, 0.05). HAQ scores were not significantly changed, with an annual change of 0.00 (−0.01, 0.00).

Figure 3.

Adjusted mean pain, fatigue and Health Assessment Questionnaire (HAQ) scores of fibromyalgia study patients in 2000–2010.

2.4 The effect of NCAD on clinical status

For patients who began NCAD therapy, the estimated 25th percentile for drug discontinuation was 1 (95% CI 1, 1) year and the 50th percentile was 2.5 (2, 3.5) years. The reported median daily doses were: duloxetine 60 mg, gabapentin 900 mg, and milnacipran 100 mg. To determine the effect of NCAD on outcomes, we studied patients who had received NCAD and divided their observations into two groups: observations prior to NCAD and observations after the start of NCAD. For all patients treated with NCAD for any length of time (‘with >0 NCAD’), Table 2 shows that pain scores were reduced significantly by 0.17 (−0.03, −0.30) units following the start of NCAD. There was no significant change in fatigue or HAQ scores.

Table 2. Effect of treatment with NCAD on fibromyalgia outcomes
VariablenGroupPre NCADaOn + Post CADaDifference (95% CI)p-value
Mean (95% CI)Mean (95% CI)
  1. NCAD: new centrally acting drug – pregabalin, duloxetine, milnacipran. ‘With >0 NCAD’ refers to patients who used NCAD in at least one 6-month period. ‘With >1 NCAD’ refers to patients who used NCAD in at least two 6-month periods. CAD, central acting drug; CI, confidence interval; HAQ, Health Assessment Questionnaire.
  2. Means are average adjusted values for study variable before and following the start of CAD therapies.
  3. aPre NCAD = data from observations prior to start of NCAD. On + Post CAD = observations after the start of NCAD whether or not the patient was still using NCAD.
Pain508With >0 NCAD6.21 (6.09, 6.32)6.04 (5.90, 6.18)−0.17 (−0.30, −0.03)0.014
344With >1 NCAD6.22 (6.08, 6.36)5.99 (5.83, 6.15)−0.23 (−0.39, −0.08)0.002
Fatigue508With >0 NCAD6.64 (6.52, 6.76)6.61 (6.46, 6.76)−0.03 (−0.17, 0.10)0.635
344With >1 NCAD6.34 (6.49, 6.78)6.57 (6.39, 6.74)−0.07 (−0.22, 0.08)0.370
HAQ508With >0 NCAD1.28 (1.26, 1.30)1.30 (1.27, 1.34)0.02 (−0.01, 0.05)0.126
344With >1 NCAD1.28 (1.26, 1.31)1.29 (1.25, 1.33)0.01 (−0.02, 0.04)0.609

We considered that one 6-month period might not be sufficient to fully express potential benefit, so we repeated the analyses after excluding patients who received NCAD for only one 6-month period. Table 2 (‘with NCAD >1’) shows that fatigue and HAQ continue to be non-significantly related to NCAD, but that the pain effect increased slightly to a treatment difference of −0.23 (−0.08, −0.39). In additional analyses, we further restricted the ‘with NCAD >0’ group by excluding all observations following the start of NCAD, in which NCAD was not taken. In effect, this compares all not-on-NCAD observations with all-on-NCAD observations. The changes in scores were −0.14 (−0.28, 0.01) for pain, −0.03 (−0.18, 0.12) for fatigue, and 0.03 (−0.00, 0.06) for HAQ.

Finally, we ran the (‘with NCAD >1’) analyses after excluding all patients ever treated with gabapentin, in effect restricting analyses just to FDA-approved therapies. The differences for pain, fatigue and HAQ were as follows: pain −0.24 (−0.39, −0.82), fatigue −0.13 (−0.29, −0.04), and HAQ 0.01 (−0.3, 0.04). With the ‘NCAD >1’ the results were: pain −0.26 (−0.44, −0.09), function −0.15 (−0.34, 0.05), and HAQ −0.00 (−0.04, 0.03).

3. Discussion

The high use of psychotropic and analgesic medications by fibromyalgia patients in 2002–2005 has been extensively documented using the large Pharmetrics database (Berger et al., 2007). Use of medications in fibromyalgia differs little from use in other pain conditions, with the exception that only 5% of the European patients with other pain conditions used what we are calling NCAD (Breivik et al., 2006). The contribution of the current report is to demonstrate the changing patterns of fibromyalgia pharmacotherapy, the unchanging patterns of fibromyalgia pain, fatigue and functional status, and the effect of NCAD therapy on fibromyalgia outcomes. We found that, during the period of 2000–2010, tricyclic use dropped in half from 27% to 15%, while heavily promoted and FDA-approved anticonvulsant and SNRI antidepressant use increased from less than 10% (accounted for by off-label use of gabapentin) in 2000 to 44% in 2010 (Fig. 2).

We also found a decrease in NSAID use (from 74% to 44%) and a slight overall increase in opioid use, ending at 47% in 2010. Berger et al. (2007) noted 38% of opioid use in the Pharmetrics study. Opioid users, and SNRI and anticonvulsant users in our study had pain scores ≥6, slightly increased from non-opioid and NSAID groups, as shown in Table 1.

A key finding of this study was the maintenance of high levels of pain, fatigue and functional status, as well as the absence of clinically important change in these variables in fibromyalgia patients over the 11 years of observation period. We also found evidence that NCAD therapy did not improve the outcome of fibromyalgia patients in a clinically meaningful way, although statistically significant improvement was seen in pain scores, but not in fatigue and HAQ scores. The primary level of improvement for pain was 0.17 units (Table 2), although in some sensitivity analyses this improvement was as great as 0.26. Thus, improvement between 2.8% and 4.3% was noted with NCAD therapy. In SD or effect size units, the effect size-like change was 0.07 and 0.10.

Meta-analyses of clinical trials with respect to central acting drugs indicate that for duloxetine, milnacipran and pregabalin compared with placebo effect sizes were: pain 0.33, 0.19, 0.27; fatigue −0.10, 0.13, 0.16 (Hauser et al., 2010a). Meta-analysis of other classes of drugs indicates that for the tricyclic amitriptyline the effect sizes for pain and fatigue were: amitriptyline 1.64 and 1.12; selective serotonin re-uptake inhibitors (SSRIs) fluoxetine and paroxetine 0.39 and 0.17. Cohen's categories as to the magnitude of the effect size indicate that values >0.2 indicate a small effect size, ≥0.5 a medium effect, and ≥0.8 a large effect (Cohen, 1988). The meta-analysis data are derived from clinical trials that are inherently short term. However, there are no reliable long-term data as to efficacy. Although some open-label extension data exist (Chappell et al., 2009; Mease et al., 2009; Murphy et al., 2009), open-label studies are inherently biased as to efficacy (Taylor and Wainwright, 2005; Wolfe and Michaud, 2010a).

The current data are consistent with the meta-analytic data in showing much less effect for fatigue than for pain. It is not surprising that the clinical effect was much less than found in clinical trials, because patients were switching from one active treatment to another active treatment, not from placebo to active treatment, as is the case with randomized controlled trials. Even so, the failure to demonstrate meaningful improvement should give pause.

Combined with the generally small effect sizes noted in meta-analyses of clinical trials and the recommendations against approval by EMEA, our results not only raise the question as to whether the new fibromyalgia therapies are truly effective, but also how one should determine effectiveness in clinical practice, and whether expensive treatments with minimal benefit should be used. In the United States in 2011, pregabalin at 300 mg/day costs around $3000 per year, duloxetine approximately $3760 per year, and milnacipran approximately $1400 per year. By contrast, amitriptyline 75 mg daily has an annual cost of $65.

There are a number of limitations to our study. One such limitation is non-random selection wherein patients with the worst prognosis (see Table 1) are the ones most likely to be prescribed with an NCAD. In addition, because patients switch among therapies, it is difficult to have a clear picture of individual drug performance. Our study is an effectiveness study, not an efficacy study, and explores results of drug treatment in the community. In such settings, patients may not follow the strict prescription guidelines of RCTs. In this study, we treated all NCADs similarly, but it is possible that one drug is better than another, something we were unable to see with our analyses and in analyses just limited to single drugs. We performed a number of sensitivity analyses to see if other methods of analysis yielded better results. But in general, no matter how we analysed the data there was limited benefit, and that benefit was only seen with pain, not fatigue or functional status. Data such as these underscore what may be considered the futility of applying RCT analysis results to chronic therapies in the community. Is it the case that EMEA was wiser than the FDA in not approving NCAD for use in fibromyalgia?

In summary, we found that a general movement towards the use of NCAD occurred over the last 10 years, but that switching to NCAD resulted in substantially increased drug costs, but not in clinically meaningful benefit.

Author contributions

F.W. and K.D.M. collected and analysed the data, and participated in the writing of the manuscript. Y.C.L. and W.H. participated in the design of the study, the writing of the manuscript, and the analysis of the study data. R.S.K. and W.H. participated in the writing of the manuscript and the analysis of the study data. All authors discussed the results and commented on the manuscript.