Article Text

Download PDFPDF

Extended report
Disability in rheumatoid arthritis in the era of biological treatments
  1. Eswar Krishnan,
  2. Bharathi Lingala,
  3. Bonnie Bruce,
  4. James F Fries
  1. Division of Rheumatology and Immunology, Stanford University Department of Medicine, Palo Alto, California, USA
  1. Correspondence to Dr Eswar Krishnan, Division of Rheumatology and Immunology, Stanford University Department of Medicine, Suite 203, 1000 Welch Road, Palo Alto, CA 94304, USA;{at}


Objective Rheumatoid arthritis (RA) is a disabling disease. The authors studied the impact of new, expensive and occasionally toxic biological treatments on disability outcomes in real-world populations of patients with RA.

Methods The authors analysed Health Assessment Questionnaire Disability Index data on 4651 adult patients with RA collected prospectively from 1983 to 2006. They studied trends in disability using multilevel mixed-effects multivariable linear regression (mixed) models that adjusted for the effects of time trends in gender, ethnicity, age, smoking behaviour and disease duration.

Results Overall, the patients were predominantly female (76%), were predominantly white (88%), had 13 years of education and have had RA for 13 years, on average. The time period from 1983 to 2006 saw major increases in the use of disease-modifying agents and biological agents, and a decrease in smoking. After adjustments, the disability rates declined at annual rates of 1.7% (1.5–1.8%) overall and 2.7% (2.4–3.1%) among men. The annual rate of disability declines in the biological era was greater than that in the preceding period, suggesting accelerated improvement. These declines were documented in all patient subgroups such as men, women, African–Americans, obese, older age groups and early disease (p<0.001), but not among the 1401 patients (where disability remained stable) who died on follow-up.

Conclusion Aggressive use of traditional disease-modifying agents and introduction of biological agents were associated with substantial gains in disability outcomes. Our finding supports the prevailing notion that ‘tight inflammation control’ is a desirable therapeutic strategy.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


In the USA and other developed countries from 1982 to 2005, population-based rates of disability have declined at a rate of 1–2% annually.1 2 The causes of this decline have not been confirmed, although they likely include factors related to prevention, treatment, the environment and declines in specific disability-causing chronic illnesses and their rates of progression.

Over the past three decades and up to the late 1990s, longitudinal studies have documented that disability in rheumatoid arthritis (RA) has also declined by about 2% a year.3,,5 Explanations for the decline in RA may be more accurately estimated than in the general population and may be related to temporal associations with major advances in treatment. Several studies have suggested that major changes in the RA treatment paradigm, with reversal of the ‘treatment pyramid’6 and use of aggressive methotrexate-based disease-modifying strategies, have contributed to improvements in functional disability.7 8 Disease-modifying anti-rheumatic drugs (DMARDs) have been shown to improve function and/or to slow radiological progression. This change in treatment philosophy was driven, in part, by increasing concerns about non-steroidal anti-inflammatory drug (NSAID)-associated gastropathy and, in contrast, by reassuring safety trends for DMARDs, specifically methotrexate.9,,11

Furthermore, since 1998, new powerful biological agents, such as etanercept, infliximab and adalimumab (biological DMARDs), have become available. These medications, often in combination with methotrexate, have demonstrated greater therapeutic effectiveness than non-biological DMARDs alone in multiple clinical trials. In head-to-head trials, the effects of biological DMARDs on standard end points, including functional disability as measured by the Health Assessment Questionnaire Disability Index (HAQ or HAQ-DI), have been substantially better than those of non-biological DMARDs alone.12

The increasing use of biological agents suggests that the rate of disability in RA could decline continuously, or even that the slope of that improvement could steepen. Alternatively, further improvement could become more difficult as the overall penetration of DMARD treatment approaches universality. Combined methotrexate–leflunomide use and total DMARD use also increased markedly over this period. Furthermore, recent evidence suggests a strong association between cigarette smoking and the incidence and severity of RA,13,,15 and there has been a rapid decline in smoking behaviours in the USA.

Thus, historical trends in RA treatment suggest an ‘NSAID’ era prior to 1992, where NSAIDs formed the base of the therapeutic pyramid; a ‘methotrexate’ era from 1992 to 1998, where the use of DMARD, principally methotrexate, greatly increased; and a ‘biological DMARD’ period from 1999 through the present, in which tumour necrosis factor inhibitors joined the DMARD mix.

We performed a cohort analysis to examine whether disability in patients with RA has continued to improve since the availability of newer biological agents, whether that rate of improvement was greater or less than that in prebiological years and what other factors may be important. Our study outcome was the time trend in disability in patients with RA.

Materials and methods


We conducted a prospective longitudinal analysis of patients with RA drawn from the Arthritis, Rheumatism and Aging Medical Information System (ARAMIS).16 ARAMIS is a national chronic disease databank system consisting of longitudinal data sets from diverse locales in the USA and Canada with open-rolling patient enrolment. A rolling-cohort design utilising consecutive patients enables replenishment of patients who die or withdraw from the study and enables a 25-year study with standardised outcomes. ARAMIS cohorts recruited by other methods, such as national samples of patients with RA in their first year of disease, were not included. This study was approved by the Stanford University Human Subjects Protection Program, and all subjects gave written informed consent.


The patient pool included patients from population-based cohorts and cohorts from academic medical centres. Details of these cohorts have been previously published.16 17 To be included in the present analyses, patients must have had RA that met the American College of Rheumatology criteria,18 had to be 17 years of age or older and must have completed one or more HAQs over the observation period.


A detailed description of measures and ARAMIS patient follow-up protocols has been published. All patients were mailed a HAQ at 6-month intervals (available at The questionnaire requested information on disability, symptom onset, disease duration, hospitalisations, medications and adverse effects. As well, patients completed validated pain and patient global health visual analogue scales.16 Non-responders were followed up with additional mailings and telephone calls at prescribed intervals based on the standard ARAMIS protocol.

The HAQ or HAQ-DI assesses disability on a scale of 0–3 (3=worst disability) in eight areas: dressing and grooming, arising, eating, walking, hygiene, reach, grip and activities. The HAQ is scored by summing the highest score in each category and then dividing by 8.19 The HAQ score was adjusted for the use of aids and devices in all the years. Pain and patient global health were assessed using a visual analogue scale scored as 0–3 and 0–100, with higher scores being worst.

Statistical analyses

Our primary outcome (dependent variable) was the trend (slope) of the HAQ-DI score over time. In multivariable regression models, our adjustments included gender, ethnicity and baseline measures of age, disease duration and smoking. We also took into account the patient's calendar year of entry into the ARAMIS cohort to allow for the possibility of bias due to time trends in recruitment. We used STATA 11 (College Station, Texas, USA) for our analyses.

The ARAMIS data were ‘unbalanced’ with gaps and unequal numbers of observations among patients. Furthermore, the data were collected by different centres over time. Analyses of these data were performed by using multilevel mixed-effects regression models (mixed models) utilising maximum likelihood methods that would address within-centre variations in patient enrolment. We fitted several parallel multivariable models with varying statistical assumptions and within subgroups (such as those with early RA, and age, gender and baseline disability subgroups). Bivariate and multivariable sensitivity analyses were performed to examine undue influences of attrition, patient deaths and databank centre. Missing values were treated as missing in all analyses. None of these techniques improved insights into the data, and none suggested results differing from those descriptive results presented here. We graphically examined time trends by fitting ordinary least-squares regression models to the time periods of interest.


Patient characteristics

We had 4651 patients available for analyses. Table 1 describes selected characteristics by calendar year (1983–2006). Overall, the patients were predominantly female (76%), were predominantly white (88%), and had 13 years of education. There were relatively fewer patients who had incident cases of RA in the biological era (>1998). More non-Caucasians were enrolled in later years. The earliest cohorts, as expected, had the longest disease duration at baseline (the date entering ARAMIS) and on follow-up. On average, the patients completed 13 semiannual HAQs. Patients with a baseline disease duration of 5 years or less had completed a maximum of 9–10 questionnaires. The average non-death attrition rate in this cohort was about 3.5% per questionnaire administration cycle.17

Table 1

Selected patient characteristics by year

Table 1 documents time trends in some of the non-treatment variables that could potentially influence time trends in disability. The numbers of subjects remained relatively constant over time, with enrollees replacing most drop-outs and deaths. The female:male ratios were constant at about 4:1. African–American patients doubled in frequency from 2% to 4% with enhanced enrolment effort. HAQ scores improved from 1.3 in the 1980s to 1.2 in the 1990s to 1.1 or lower in the 2000s. Smoking prevalence declined from 27% to 5% from 1992 to 2006.

Figure 1 shows the trends in mean HAQ score from 1982 to 2006 with least-squares linear regression over the period 1982–1990, the period 1991–2000 and the period after 2000. All curves, except for the first, show statistically significant improvement, with negative (declining) slopes. Trends were similar in men and women.

Figure 1

Trends in disability in RA from ARAMIS cohorts over the period 1982–1990 (NSAID-based era), the period 1991–2000 (methotrexate DMARD period) and the period after 2000 (methotrexate and biological DMARD period). The trend lines were fitted using ordinary least-squares regression. The Y-axis shows HAQ-DI (0=least disability; 3=worst disability), and the X-axis shows calendar year. Disability remained relatively constant over the NSAID-based period (prior to 1990), then declined significantly (p=0.001) in the non-biological (methotrexate) era (1992–1997) and in the biological and non-biological era (p<0.01). ARAMIS, Arthritis, Rheumatism and Aging Medical Information System; DMARD, disease-modifying anti-rheumatic drug; HAQ-DI, Health Assessment Questionnaire Disability Index; NSAID, non-steroidal anti-inflammatory drug; RA, rheumatoid arthritis.

Table 2 shows the changing utilisation of different drugs over time. Methotrexate use rose steadily until it plateaued in 1999. Leflunomide became available in 1999 and rose quickly to a 10% share, for a total of about 67% share for methotrexate and leflunomide (Arava). Hydroxychloroquine (Plaquenil) use rose from 12% in the early periods to 20% in later periods. Use of biological DMARD began sequentially in this database in 1999, with etanercept (Enbrel), infliximab (Remicade) and adalimumab (Humira) added, for a total share rising to 37%, with more than half of these patients using etanercept. With a combined use of non-biological DMARDs and biological DMARDs, the percentage rose from 50% in 1983 to 77% in 2006, with more combination treatment in later years. Prednisone treatment (and dose) changed little over the observation period.

Table 2

Percentage of subjects on different drugs over time

Figure 2 shows these major trends graphically. The use of non-biological DMARDs rose and then plateaued. Biological DMARDs appeared in 1999 and achieved a 37% share quite rapidly, thus increasing the total DMARD percentage, although less than expected (13% vs 37%), because these data include combination DMARD treatment, particularly biological and methotrexate combinations.

Figure 2

Trends in treatment utilisation among RA patients in ARAMIS. Non-biological and biological DMARDs have been introduced into RA treatment over specific time periods. Methotrexate rose rapidly in prevalent use from 1983 to 1995, leflunomide rose rapidly in prevalent use from 1999 to 2006 and biological DMARDs rose rapidly in prevalent use from 1999 to the present. ARAMIS, Arthritis, Rheumatism and Aging Medical Information System; DMARDs, disease-modifying anti-rheumatic drugs; RA, rheumatoid arthritis.

Table 3 provides the results of multivariate analyses. Improving time trends were observed in all patient subgroups, except among those who subsequently died during the study period. Notable is the greater magnitude of β coefficients among patients observed in the biological era (−0.0196; −0.215 to −0.178) compared to the referent period (NSAID era) and the immediate prior era (−0.0900; −0.105 to −0.072), providing evidence that the rate of progression might have accelerated in the former era.

Table 3

Multivariable models for assessing the slope of disability over time, adjusting for the influence of age, gender, non-white ethnicity, baseline disease duration, body mass index and smoking status


Our data suggest, as hypothesised, an NSAID-based era with no decrease in disability, a methotrexate era of rapid decline and a biological era of continued decline. The composition of our cohort changed over time in a way that can expected to increase disability over time (increasing age, disease duration and increasing proportion of African–Americans). Thus, the estimated declines in RA disability in the biological era are likely to be an underestimate of the true declines. The results of this longitudinal observational study spanning three decades indicate that disability in RA has continued in a steady decline since 1992, including recent years since 1998 when biological agents began to be used. This confirms and extends other studies that have reported significant declines in RA disability in the USA and elsewhere.3 20,,28 These findings are also similar to the findings of the large National Database of the German Collaborative Arthritis Centres (n=38 723), which, from 1997 to 2007, documented dramatic declines in disease activity score,28 swollen and tender joints, and work disability, which are associated with an increased utilisation of non-biological and biological DMARD medications.20 Sokka et al21 described decreasing rates of radiographic damage progression over time in three historical cohorts of RA and suggested lower rates in more recent years.

There remain questions about the causes of the declines from 1993 to 1999 (‘methotrexate’ era) and from 1999 to 2006 (‘non-biological and biological DMARD’ era). Potential explanations include a trend towards milder initial disease, differences in patient selection and improved treatment, as evidenced by a more frequent use of disease-modifying agents.7 29 30 It is possible that RA itself is becoming milder over time, but the magnitude of such declines is too small to explain the ∼1.7% annual decline that we have documented. Differences in patient selection may be another plausible explanation; this could explain our findings if all patients enrolled in the later years were selectively milder cases. This is not a testable hypothesis, since all these patients were treated prior to entry and their disability status would reflect the net effect of treatments. This leaves the most obvious explanation: the dramatic change in prevalent treatment for RA. In the absence of randomised data, causality may be supported by deductive reasoning: if a treatment did not change over a period, then that treatment could not have caused a change in outcomes. Methotrexate use in these patients rose from a 3% prevalence in 1983 to a 45% prevalence in 1998—a very large change. Over this period in these patients, there was essentially no change in the use of hydroxychloroquine, sulfasalazine or other non-biological DMARDs, and leflunomide was not yet available. Prednisone use remained constant at about 40%. Thus, the only plausible explanatory treatment change from 1993 to 1999 was increasing use of methotrexate.

In contrast, from 1999 to 2006, methotrexate use did not increase, and considerable amounts of methotrexate were used in combination with biologicals. Methotrexate dosage may have increased, but this is difficult to interpret because supplementation of folic acid in cereals and breads also began in 1998. The use of biological DMARDs in these patients rose from 3% to 37%, with much of the use including etanercept. Leflunomide prevalence increased rapidly to a 10% ‘share’, and hydroxychloroquine use increased, somewhat lifting the non-biological DMARD ‘share’ to two-thirds of patients. Thus, the treatment trends consistent with improvement from 1999 to 2006 are dominated by the use of biological DMARDs, perhaps abetted by leflunomide use, hydroxychloroquine use and escalation in methotrexate dose. Another relevant trend has been the setting of more stringent targets for RA treatment success, such as ‘remission’ or ‘low disease activity’, which perhaps leads to an earlier escalation to more aggressive treatment regimes. Such a strategic change, we hypothesise, could be credited for the gains, as much as the efficacy, of individual agents.

There are also non-treatment influences that could partly explain disability declines over these periods. As discussed, there is an ongoing national secular decline in disability. This could result in declines in comorbid contributions to RA disability, including heart disease, chronic obstructive pulmonary disease and stroke. The national secular decline in smoking behaviour,31 seen here also in our patients with RA, could contribute to less severe incident disease and slower RA progression, although this mechanism is likely to be relatively minor given the declines within smokers and non-smokers.

Furthermore, our results may have been attenuated by regression to the mean, which causes patients with flaring disease to preferentially tend to receive more treatment that is aggressive. This could result in a reduction in the impact of biological treatments to a magnitude lower than that observed in placebo-controlled trials.

The use of biological treatments remains selective. The utilisation of these newer drugs in RA is limited by cost, contraindication, patient preference and perceived efficacy. In Germany, as of 2007, only 23% of patients were prescribed any disease-modifying agents, and only 16% of patients were prescribed biological agents.20 Data from the Thomson Healthcare MarketScan Research database in the USA suggest, as do our own data, that the utilisation of biologicals in the USA in 2006 was higher at 26%.32 Corresponding figures in Medicaid beneficiaries are 12–16%.33 34 These figures may increase with increases in the use of newer biological treatments.

Thus, the penetration of biological treatments into ‘market share’—at about one-fourth to one-third of patients with RA—remains relatively small, and positive effects may be diluted by the inclusion of those patients who were previously on efficacious but less powerful treatments. Moreover, the use of the strongest traditional DMARDs (methotrexate and leflunomide) has also increased over this period, as has the total percentage of patients on DMARDs. A conservative interpretation of these data therefore includes congratulating scientists, cliniciansand patients on the continued decline, and suggesting that this is consistent with continued emphasis on early and consistent DMARD use and the incorporation of biologicals into our therapeutic repertoire.

Our data sets are perhaps uniquely qualified to compare disability outcomes in RA across three decades and three treatment paradigms. The strengths of this study include the large number of diverse patients available for analyses, which includes patients from the general population and from academic medical centres. In contrast to our earlier paper that documented improvement in RA disability by year of symptom onset,3 these analyses take the perspective of ‘treatment adoption’, with direct linkages between treatment changes by year and disability reductions in the same year. The approaches are complementary, and analyses of these data from the symptom onset perspective yielded confirmatory results. However, this perspective did not reveal the relatively sharp changes in trend lines. This likely occurs because a treatment (eg, methotrexate) is introduced at a time point and then given to patients in all symptom onset cohorts, all of whom tend to benefit. Thus, the treatment–year effect is larger than the first symptom–year effect. The limitations of our study stem from the relatively smaller number of biologically treated patients and the shorter follow-up period.

On the basis of drug treatments employed over time and our observed differences in yearly disability levels, we constructed three hypothetical ‘eras’, which are arbitrary in that major treatment changes evolved rather than changed abruptly. Thus, there are treatment overlaps between ‘eras’. Nevertheless, the treatment changes occurred by plan rather than by accident, as the NSAID-based treatment era gave way to a methotrexate-dominated era, which then embraced biological DMARDs as they became available. These data strongly suggest that the first of these eras did not work, that the second did and that improvement trends have been continued to the present.

Improving disability is a dominant—but not the only—goal in the treatment of RA. Aggressive use of medications, including biological agents, can also improve health outcomes (such as need for arthroplasty and job disability), although there is an accompanying increased risk for drug toxicity (such as in infections). The net effect is likely to be favourable, as shown in a meta-analysis of clinical trials, but measuring the magnitude in the real world would be critical.35 Nevertheless, the gains we have demonstrated may not be available to parts of the world where the management of RA is less aggressive, as pointed out by Sokka et al.36


View Abstract


  • Funding The study was funded by Centocor Ortho-Biotech (Horsham, Pennsylvania).

  • Competing interests The authors declare that they have no competing interests. The present analysis was performed using an investigator-initiated funding mechanism from Centocor Ortho-Biotech, whose products include biological treatments for RA.

  • Ethics approval The study was approved by the Stanford Institutional Review Board.

  • Provenance and peer review Not commissioned; externally peer reviewed.