Don’t Bet on the NSAID Shuffle


Synopsis: Response to treatment with nonsteroidal drugs from two different chemical families is highly correlated in subjects with osteoarthritis and rheumatoid arthritis.

Source: Walker JS, et al. Arthritis Rheum 1997;40: 1944-1954.

Some believe that there are "responders" and "non-responders" to anti-rheumatic drug therapy. If it is the case that there are patients who will respond to one drug, or one class of drugs, and not to another, there is the possibility that one could determine before initiating treatment which drug would be best for an individual patient. In doing so, one could avoid giving a potentially toxic drug to someone who will not benefit from it. While the discussion about responder status usually arises in the context of disease-modifying drug therapy for rheumatoid arthritis, Walker and colleagues chose to study nonsteroidal anti-inflammatory drugs (NSAIDs). They set out to answer two important questions about patients with osteoarthritis (OA) and rheumatoid arthritis (RA). The first was: Is "response" to NSAIDs an inherent characteristic of an individual patient? The second was: If response is an intrinsic patient characteristic, can one predict from pretreatment measurements whether a patient will respond to NSAID therapy? Their cleverly designed study tried to wring the answers to these two questions from a remarkably small study sample of 20 patients, nine with RA and 11 with OA. The study was designed to compare the response to each of two different NSAIDs given in a randomized (but balanced) order for four periods of four weeks each. This allowed a subject’s outcome during each period to be analyzed independently. The hypothesis that was tested was that a patient’s response to one NSAID during one period would be closely correlated with the response during other periods. The authors selected a clinically significant (30%) improvement in five of seven clinical variables (pain, joint tenderness, joint swelling, patient assessment, physician assessment, disability, and acute phase protein levels) during the last two weeks of each four-week period to be the criterion for "response."

The two groups of patients (OA vs RA) had marked differences at baseline and had to be analyzed separately. For the nine subjects with RA, a "response" was noted during 15 of 36 patient-treatment periods. For subjects with OA, a "response" was noted in 18 of 44 patient-treatment periods. When subjects were categorized into groups according to the number of treatment periods during which they responded, the results were as noted in the figures. The observed distributions are very different than those that would be predicted if "response" status were not a characteristic of the patient and was determined for each treatment period by chance.

Figure 1

Osteoarthritis distribution of response periods

Figure 2

Rheumatoid arthritis distribution of responses

The authors compare the observed distribution for RA and OA patients (black bars in the figures) to the "expected" distribution, which is predicted if treatment periods were independent of patient and were determined only by chance (white bars). They report that there is a statistically significant difference between the two distributions for both RA (P = 0.01) and for OA (P < 0.01). This is convincing evidence that "response" is indeed a characteristic of a patient (subject), and, in this study of two different NSAIDs (ketorolac and piroxicam) from two different chemical classes, response to one NSAID is correlated with response to another.

After validating the concept of "responder" to NSAID treatment, the next step was to perform a multiple logistic regression analysis to see whether any of the clinical or laboratory variables measured at entry into the study predicted response to NSAID treatment. For patients with RA, only the pretreatment white blood cell count (WBC) and the erythrocyte sedimentation rate (ESR) were significantly correlated with subjects’ response status. Lower WBC and lower ESR correlated with an increase in number of periods during which a response to NSAIDs was observed. Unfortunately, the range of WBC for four response periods vs. no response periods was between 6000 per mm3 to about 9500 per mm3, that is, well within the normal range for WBC. The ESR ranged from near normal for subjects with 3-4 periods of response and was elevated in the 50-60 mm/hr range, on average, for subjects who did not meet response criteria during any of the experimental periods. It is notable that other biological markers that were measured at the beginning and during the study (interleukin-1 alpha and beta, interleukin-1 receptor antagonist, interleukin-6, throboxane-B2, and tumor necrosis factor alpha) had no predictive value. In patients with OA, none of the clinical or laboratory pretreatment variables was correlated with response to treatment.


If one is going to try to create a model that can be used to predict a patient’s response to a medication, it is worth the effort to be sure the thing being predicted, "response," is a real and reproducible phenomenon and not a statistical fluke. The reported statistical significance of the results confirms the common sense notion that a subject’s response to one NSAID would be reasonably reproducible from one time period to another and the somewhat less intuitive notion that response to one NSAID would be correlated with response to another. Because the subjects met the response criteria in less than half of the study periods, I was concerned that this could result in the statistical difference that was seen. The unexpectedly high numbers of subjects at the extremes of the distributions argues strongly that the observed difference is not due to selecting criteria that resulted in skewing the distribution. In order to confirm this, a model that used the actual fraction of response periods (15/36 for RA and 18/44 for OA), rather than the 1/2 figure used by the authors, leads to P values that are similar, namely P = 0.02 for RA and P less than 0.001 for OA.

The results of the attempt to predict response using baseline variables were disappointing in that none of the variables or combinations of measures correlated with response in subjects with OA, and only ESR and WBC were correlated with response in those with RA. The trial included a very small number of patients, however, and the authors noted that several variables showed trends that might have achieved statistical significance had there been a larger number of subjects enrolled.

It is a matter of faith with some rheumatologists that one patient’s response to an NSAID may be dramatically different from that to another NSAID. This article suggests that, at least for ketoprofen and piroxicam, the responses are likely to be closely correlated. The practical importance of this report may be a dampening of enthusiasm for a prolonged "NSAID shuffle." For patients with RA, this suggests that if symptoms and signs of inflammation are not promptly controlled with a 4-6 week course of one’s NSAID of choice, some disease-modifying drug (methotrexate, gold, hydroxychloroquine, sulfasalazine, cyclosporin, azathioprine, etc.) should be started rather than trials of a succession of NSAIDs. For patients with OA, lack of response to an NSAID may mean a switch to an analgesic such as tramadol or a narcotic analgesic or, depending upon the severity of the symptoms, referral for total joint arthroplasty.

Gemfibrozil treatment of post-CABG patients with isolated low HDL has been associated with marked reduction in subsequent clinical events despite lack of any demonstrated angiographic benefit.

Which combination of white blood cell count (WBC) and erythrocyte sedimentation rate (ESR) is correlated with the best response to nonsteroidal anti-inflammatory drugs in patients with rheumatoid arthritis?

a. Lower WBC and Higher ESR

b. Higher WBC and Lower ESR

c. Higher WBC and Higher ESR

d. Lower WBC and Lower ESR