METHODS

Exploratory approach

We recruited 19 Dutch rheumatologists working in different regions of the Netherlands, and representing both university hospital (including tertiary referral centres) and general hospital working environments. They were asked to prioritise various items reflecting demographic data, socioeconomic status, disease specific comorbidity, and reports by both patients and physicians on specific characteristics of ankylosing spondylitis for their importance in determining whether or not a patient with this disease should be treated with TNF blocking drugs. None of the rheumatologists had been involved in any clinical trial with these agents in ankylosing spondylitis, and their experience was only anecdotal.

The rheumatologists were asked to make their decision under the following assumptions:

• TNF blocking drugs are available on demand;

• the cost of TNF blocking drugs is completely reimbursed;

• patients may expect a 50% decrease in BASDAI (the Bath ankylosing spondylitis disease activity index) in more than 50% of the patients (based on the results of two phase 2 clinical trials)^4,5;

• long term toxicity is unknown.

Prioritisation was done by rank ordering the variables by perceived importance, from 1 (most important) to 5 (least important), and by scoring the importance per item on a nominal four point scale (0, not important to +++, very important).

The items were as follows: age; sex; disease duration; occupation (“white collar” (office job) or “blue collar” (physical job)); loss of job during the past 12 months; permanent disability pension (after a period of 12 months of continuous sick leave, a full or partial work disability can be granted in the Netherlands); history or the presence of extraspinal features (such as hip arthritis, shoulder arthritis, or arthritis in other peripheral joints, enthesitis, inflammatory bowel disease, or acute anterior uveitis); limitation of spinal function (lumbar spine, thoracic spine, and cervical spine were inquired about separately); grade of syndesmophyte bridging (from no bridging to complete bridging); rate of development of functional impairment (global impression); rate of development of structural radiographic damage (global impression); increased acute phase reactants (C reactive protein and erythrocyte sedimentation rate); NSAID tolerance and efficacy; tolerance and efficacy of previous DMARDs; efficacy of current DMARDs; physician’s global assessment of current disease activity; physician’s global assessment of cumulative disease activity; and physician’s global assessment of cumulative disease severity.

The rheumatologists were further encouraged to mention any other factors related to their decision to start TNF blocking therapy, and rank them accordingly.

The analysis was descriptive. Variables were rank ordered by the percentage of rheumatologists mentioning this variable, by a weighted score of importance (not important, 0 to very important, 4); and by “top five” rank order by every rheumatologist.

Individualised approach: cohort analysis

For the analysis on how physician’s anti-TNF treatment decisions for individual patients can best be captured in measurable constructs, use was made of the cohort from the OASIS study (outcome in ankylosing spondylitis international study). OASIS is an ongoing prospective study of consecutive patients with ankylosing spondylitis from three different countries (Belgium, France, and the Netherlands).^6,7 Patients were first recruited in 1997. Recruitment was complete within three months, and patients were followed consecutively by at least annual assessments. The assessment profile included all instruments selected by the Ankylosing Spondylitis Assessment Working Group as a core set to assess outcome in ankylosing spondylitis,^8,9 plus other widely used instruments, such as BASDAI, as well as extensive information on the socioeconomic situation, comorbidity, and disease severity.

The study described here was limited to the Dutch patients only. The data from the last available follow up visit were used for further analysis. These data were collected independently without reference to the treating physicians, and the findings were not communicated with them. As an example, radiography of the spine will be mentioned here. The treating rheumatologist was not aware of the damage score that was assigned by us to that patient (for example, 70 modified SASSS (Stoke ankylosing spondylitis spinal score) units—a score which was used in the analyses). But it is possible, though not obligatory, that the rheumatologist had insight in the radiologist’s report that “there are signs of spondylitis anteriorly, with small syndesmophytes at different levels.”

The treating physician of each patient was asked by one of us (BR) to use their clinical insight to judge whether or not a patient should be treated with TNF blocking drugs, based on the same assumptions as for the exploratory approach above. The treating physician was allowed, but not obliged, to study the medical dossier of that particular patient in order to make his decision in case of any doubt.

Data analysis

The primary analysis was descriptive: means (SD), medians (centiles), and frequencies, stratified by TNF blocking therapy (yes v no) were calculated as appropriate. Univariate between-group differences were tested statistically by two sample t tests, the Mann–Whitney two sample test, or χ² tests (Fisher’s exact test) as appropriate.

To compare between-group differences across variables, the mean between-group difference for each variable was standardised by dividing it by the standard deviation of the mean between-group difference. Skewed data were logarithmically transformed before standardising. Standardised mean differences were rank ordered, so that the variables that were ranked highest showed the greatest between-group contrast.

Backward selection logistic regression analysis was done to investigate which variables contributed independently to the decision on whether a patient was an appropriate candidate for TNF blocking therapy.

RESULTS

Exploratory approach

Nineteen rheumatologists completed the questionnaire on the variables potentially contributing to the decision to start TNF blocking therapy. These rheumatologists estimated that a median of 10% (range 5% to 25%) of the patients with ankylosing spondylitis under their care were appropriate candidates for TNF blocking therapy. The results of the question on prioritising the variables according to their contribution to treatment decision making are shown in table 1.

Rheumatologists considered a rapid decline in functional capacity as well as rapid radiographic progression to be the most important indicators for starting treatment with TNF blocking drugs in patients with ankylosing spondylitis. Fewer rheumatologists reported current and cumulative disease activity as assessed by the physician as important, but the rheumatologists who did give these variables a relatively high weight ranked them 2 and 3 in the weighted scores. Extraspinal disease manifestations—such as hip, shoulder, and other kinds of peripheral arthritis—were also ranked high. Although patient derived variables such as general pain and night pain were mentioned by 75% of the rheumatologists in their top 5, their importance was rated as modest (rank 6/10), and the patient’s overall assessment of disease activity was not considered a very important determinant in the decision to start TNF blocking drugs (mentioned by 51% of rheumatologists only).

The cohort analysis

A judgment by their treating physicians (n = 9) on the appropriateness of TNF blocking therapy could be obtained for 79 patients. Of these, 24 (30%) were considered a candidate for this kind of treatment. Tables 2 to 6 show how these two groups (candidates v non-candidates) differed with respect to demographic data and comorbidity (table 2), patient derived assessments (table 3), assessments of spinal mobility (table 4), physician derived assessments (table 5), and radiology (table 6).

Patients with indications for TNF blocking therapy tended more often to have a history of uveitis and were more often HLA-B27 carriers than patients without such indications, according to their treating physicians. Physicians did not base their decision on whether or not the patient was in work. All patient-derived assessments, including BASDAI (disease activity) and BASFI (function), scored higher on average, and the difference was statistically significant in many cases. Spinal mobility, as assessed by several different tests, was worse in the patients with indications for TNF blocking treatment, as were the radiological appearances (both status score and progression score). Patient derived assessments—including joint counts and acute phase reactants—did not differ between the two subgroups.

Table 7 shows that the modified SASSS (radiology) status score and the BASFI (function), as single variables, discriminated best between patients with and without a physician reported indication for TNF blocker treatment, followed by lumbar lateral flexion (spinal mobility), BASG (in the previous week), and BASG (in the previous six months) (patient’s global assessments of wellbeing).

Table 8 shows the results of the multivariate logistic regression analysis in which all variables presented in tables 2 to 6 were investigated for their independent potential to explain the physician’s judgment about the appropriateness of prescribing TNF blocking therapy. Physicians appeared to assign TNF blocking therapy preferentially to male rather than to female patients, and independently to those characterised by a relatively high level of radiological damage and those with impaired cervical spinal mobility. None of the other variables contributed independently to explaining the variation in assignment.

DISCUSSION

Confronted with the question as to what kind of patients with ankylosing spondylitis should be treated with TNF blocking therapy, rheumatologists—if asked outside their office, that is, if not actually facing such a situation with one of their patients—say that they base their decision on factors such as rapid functional deterioration, rapid radiographic progression, a high level of current and previous disease activity, and the presence of peripheral arthritis. When tested with actual patient records, however, they appear to prefer TNF blocking therapy for patients with advanced disease, reflected by high scores on the assessments for spinal mobility and radiographic damage, rather than active disease, with or without extraspinal involvement. Apart from that, the multivariate analysis showed that they would assign a higher priority to male patients, perhaps reflecting the general appreciation that male patients with ankylosing spondylitis have a worse prognosis

We found a clear discrepancy between what physicians think they should do and what they actually do if confronted with the patient’s records. The rationale for assigning TNF blocking treatment to patients with a high rate of functional deterioration and radiographic progression may be that deterioration can be halted by appropriate treatment. The rationale of starting this type of treatment preferentially in patients with advanced disease is more questionable—there is no indication that TNF blocking therapy can undo established radiographic damage, and trial results have shown that impairment of spinal mobility is only partly reversible. A more likely explanation for the phenomenon that physicians prioritise patients with advanced disease may be that measures reflecting severity (radiographic damage, spinal mobility) adequately capture other domains, among which are disease activity and function, both from the patient’s perspective and from the physician’s perspective; in other words, patients with more advanced, severe disease have by definition greater disease activity, on the assumption that a high level of cumulative disease activity has led to the present situation. There are two arguments underscoring this concept. First, the assessments that actually reflect disease activity scored significantly higher in the TNF assigned group. Second, radiographic damage was revealed as the most important discriminating factor, though the physicians were not aware of the actual radiographic score (because these data were not available in the case records), so they must have based their decision on other constructs directly related to radiological damage. An example might be that physicians could have an intuitive impression about spinal mobility in an individual patient, and spinal mobility is associated with radiographic damage.

Another discrepancy is that physicians mentioned that they placed high value on the presence of extraspinal disease (arthritis of the peripheral joints). We did not encounter this tendency in the cohort analysis, on either univariate or multivariate analysis. Again, the explanation may be that peripheral arthritis is already captured by other domains. We have shown recently that patients with peripheral arthritis have generally greater disease activity than patients without arthritis.¹⁰ Another explanation may be that the physicians simply did not recall whether or not there was, or had been, peripheral arthritis.

It is also interesting to investigate the factors that did not contribute to the decision to assign patients to TNF blocking therapy. Patient derived measures of disease activity played a minor role in physicians’ perception, though the global assessment of wellbeing made by the patient appeared to be contributory. An important limitation of this study is that physicians were asked to make their judgment without consulting the patient. Results may have been biased towards physician derived (and physician remembered) assessments, and this could explain the relative dearth of patient derived assessments in the set of contributory factors. Another factor that we thought beforehand would be contributory—whether or not the patient was in work—turned out to be unimportant with respect to the decision to assign TNF blocking drugs in either type of analysis. Labour force participation and sick leave are strongly subjected to national legislation, and the interpretation of these results should therefore be cautious.

The context of this study does not reflect reality. In the real world, the decision to start TNF blocking drugs will be the result of communication between patient and physician. Both parties will bring in their own conditions and priorities, and the net result of this negotiating process will be a more balanced judgment about the appropriateness of starting TNF blocking drugs, as opposed to the findings in this exploratory study which emphasise physicians’ opinions. Nevertheless, these results—and those of the “international start TNF blocker study ankylosing spondylitis” (ISSAS), which are expected in the second part of 2003—may shed light on how physicians view the need to start TNF blocking drugs, and can therefore be an indication for recommendations and guidelines for TNF blocking therapy in the future.