METHODS

The STURE database (Stockholm TNFα follow up registry) collects efficacy and safety data for all patients starting biological treatments at the major hospitals in Stockholm, as part of the nationwide registry of antirheumatic therapies in Sweden (ARTIS). The assessments are done at 0, 3, 6, and 12 months and annually thereafter, and include the American College of Rheumatology (ACR) core outcomes (the 28 swollen and tender joint count, visual analogue scales for global health and pain, the health assessment questionnaire disability index (HAQ), erythrocyte sedimentation rate, C reactive protein, and physician’s global assessment of disease activity); the 28 joint count based disease activity score (DAS28)^4,5; records of concurrent drug treatments; employment status; and side effects. In the study reported here, we analysed data on those patients whose infliximab dose was increased to 5–7 mg/kg per infusion. At our institution, during the period of this study, all infusions were given at eight week intervals. Based on these data, the following comparisons were made:

• Comparison of disease activity by DAS28, swollen joint count, and ACR-N (numeric ACR) response⁶ at the evaluation before a dose increase (that is, at the time when the decision to increase the dose was made) with the values after the dose increase. The dose increases were implemented at the next scheduled infusion, at a median of three weeks after the decision was made, and the first assessment thereafter would occur according to the schedule of visits for the registry.

• Comparison of the best DAS28, swollen joint count, and ACR-N response seen in these patients after the dose increase, and comparison of these values with each patient’s best result at any time point before the dose increase. These clinical responses were also analysed using established EULAR (European league against rheumatism)⁷ and ACR⁸ response criteria.

• Comparison with two control groups, one formed of patients who were treated with infliximab but without dose increase and one of patients who were treated with etanercept (dose increases of etanercept are not employed at our centre). Analysis of the control groups was based on the hypothesis that infliximab dose increases are “triggered” by deteriorating clinical results. Therefore in control patients the analysis of clinical results was anchored at the time of the first increase in DAS28 after treatment started (the reference time point). We then made the following comparisons: first, between the best values for DAS28, swollen joint count, and ACR-N at any time before the dose increase or the reference time point; second, between the values at the time of dose increase or the reference time point; and finally, of the best values after the dose increase or the reference time point.

While the STURE database is part of the ARTIS national biologicals safety registry, this study involved only data collected and analysed at the Karolinska Hospital.

Statistical analyses were done using StatView 5.0.1 for PC (SAS Institute Inc, Cary, North Carolina, USA).

RESULTS

Disease activity after infliximab dose increase

We evaluated 44 patients whose infliximab doses were increased from 3 to 5–7 mg/kg/infusion. The higher dosage is given as a range to reflect the fact that, in our practice, infliximab as prescribed is routinely rounded upward to the nearest 100 mg increment (ampoule size). Our dataset does not record the exactly calculated amount by actual body weight at the time of each infusion. At our institution during the period of this study, all infusions were given at eight week intervals throughout. Comparison of DAS28 values at the last evaluation before dose increase (that is, at the time when the decision to increase the dose was made) with the values after dose escalation showed a significant improvement at the first assessment following the dose increase 4.0 (0.18) months later (mean (SD); median 3.0), as shown in fig 1A. When we also compared the best DAS28 result seen in these patients at any time point after the dose increase, the improvement was maintained during further follow up. The best DAS28 results in this group of patients were achieved after a mean of 7.2 (0.56) months (range 3 to 18), and because infliximab infusions were always given at eight week intervals, this corresponded to a median of three infusions at the higher dose. The ACR28 swollen joint count following dose increase showed a significant improvement compared with the values at the last visit before the dose increase, and additional improvement thereafter (fig 1B).

For all outcomes, the quantitative degree of improvement was rather modest. For instance, the median improvement after dose increase in DAS28 was 0.6, which is just at the lower limit for the EULAR criterion of “moderate clinical responder.” In an additional analysis, the clinical “response” to dose escalation was also analysed using established response criteria, and achieved a moderate to good EULAR response in 16 patients (36%) and moderate to good ACR20 response in eight patients (18%).

Infliximab dose increases may recapture a previous response

To assess the nature of the improvements seen with increasing doses, further comparisons were made by including each patient’s best result at any time point before the dose increase. The patient’s best DAS28 values at any time before the dose increases were significantly lower (that is, better) than the levels immediately preceding the dose increase (3.46 (0.17) v 4.04 (0.19), p<0.0001), and similar highly significant differences were seen for the other outcomes (not shown). This simply reflects the fact that the decision to increase the infliximab dose was made when disease activity had worsened. On average, the time from the best result on conventional dose infliximab until dose escalation was 3.7 (0.53) months (range 3 to 15).

As seen in fig 2, the best DAS28 and best swollen joint count obtained after the dose increase—while being better than the results immediately before the dose increase—were similar to the best results obtained at any previous time point before the dose increase. Comparison of ACR-N before and after the dose increase showed a similar pattern. One tempting interpretation of these findings is that the dose increase served to “recapture” a response to infliximab that had diminished with time. However, this pattern of response could also be explained by a regression-like effect.

Of these 44 patients, eight experienced an infusion reaction during at least one infusion. This is similar to the frequency in all infliximab treated patients in our unit. There did not appear to be an increase in such infusion reactions around the time of the dose increases, and none of these infusion reactions was regarded as severe or resulted in discontinuation of the treatment.

Improvement after infliximab dose increase: comparison with controls

To assess the observed improvement after infliximab dose escalation, we analysed two control groups—one group of patients treated with infliximab but in whom the dose was not increased, and one group treated with etanercept. As indicated above (Methods), the analysis of the control groups was based on the hypothesis that infliximab dose increases are “triggered” by worsening clinical results. We made the following comparisons: first, of the best values for DAS28, swollen joint count, and ACR-N at any time before the dose increase or the reference time point; second, of the values at the time of dose increase or the reference time point; and third, of the best values after the dose increase or the reference time point.

The time intervals between the best previous result and the reference time point in these control groups were 4.2 (0.39) months (range 3 to 6) for the infliximab controls, and 4.5 (1.83) months (range 3 to 9) for the etanercept controls.

Because these three groups of patients were not randomised, we analysed whether changes in other drug treatments could have influenced our results. At the last visit before the dose increase or the reference time point, around half the patients in each group were taking daily oral glucocorticoids (prednisolone), at a mean dose of 5.4 (0.36) mg/day. At the first visit following that time point, the mean daily dose was 4.6 (0.32) mg/day—a small but statistically significant decrease. This was seen in all three groups, the changes in dose being similar and not statistically different in each group (not shown). However, these data had a non-normal distribution, a few patients having large reductions to account for the overall reduced dosage. The median change in prednisolone dose for each group was zero. With respect to other DMARDs, most of the patients were taking concomitant methotrexate. There were no dose increases for either methotrexate or other DMARDs in conjunction with the infliximab dose escalations.

As can be seen in fig 3A, DAS28 values in the three groups showed very similar patterns. For the cases, the values at the last visit before the dose increase were worse than previous values, and after the dose increase they become better again (“recapturing the response”). However, in the controls, the pattern was similar—that is to say, the values at the reference time point were worse than previously (this following logically from the manner in which the reference time point was defined), but also showing an “improvement” after the reference time point, resulting in similar values to those obtained before. The time from reference time point to the best DAS28 result was 10.3 (0.70) months (range 6 to 18) for the etanercept controls, and 7.2 (0.56) months (range 0 to 18) for the infliximab controls, similar to the infliximab cases. Swollen joint counts and ACR-N values for this analysis are shown in fig 3B and 3C, and reveal similar patterns, but with the difference that in the “cases” the ACR-N responses at each time point were lower than in the controls, while they showed the same pattern of worsening followed by improvement. Thus the changes in outcome in the three groups were numerically similar and not statistically different.

DISCUSSION

We used a structured longitudinal follow up registry of patients treated with biological agents to study the effects in clinical practice of increasing the dose of infliximab above 3 mg/kg/infusion. Such dose increases occur quite often and are usually occasioned by insufficient clinical response, although insufficient duration of response (that is, less than eight weeks) is also commonly given as the reason. This may be surprising in that it makes intuitive sense to increase the frequency rather than the dose in such an event. However, until 2001 no data were available to indicate which course of action would be best, and increasing the frequency of infusions was discouraged owing to the limited availability of nursing resources for intravenous treatments. A recently published pharmacokinetic modelling study⁹ suggested that increasing the frequency of infusions was more effective for such patients than increasing the dose, and this is likely to affect these types of decisions in the future.

Thus for most patients the dose increase was prompted by an insufficient clinical response. Our analysis shows that following dose increases, better results were seen, such that the average DAS28, swollen joint count, and ACR-N improved and 18–36% of patients met formal response criteria. On the other hand, the improvements were rather modest in magnitude, and the values after dose escalation were similar to previously achieved best results; they might therefore be interpreted as showing that a previous response to infliximab treatment that had diminished over time could be “recaptured” by the dose increase. However, in our case–control analysis, we were led to conclude that a more frugal explanation for these findings is a regression-like effect—that is, inasmuch as dose increases are prescribed at a time when the patients are doing relatively poorly, it is more than likely that they will do better again at time points thereafter. Thus in the control groups we chose a reference time point to mirror the time at which a dose increase could have been prescribed—namely, those times at which a worsening of clinical outcomes had occurred. The two control groups showed a similar pattern of worsening and “recapture” to the cases. The time to achieve the best result was similar for the three groups (the mean for all patients was 8.6 months, with a range from 3 to 18 months), which would appear long enough to allow maximum responses to dose escalation to occur. Of note, the ACR-N values were overall somewhat worse for the cases, suggesting that this is a patient group with a less favourable clinical course, which may indeed be the reason why the doses were in fact increased in these patients and not in the infliximab controls.

There were no statistically significant differences between the changes in the outcome measures for the cases and the controls at any of the time points. However, it could be argued that a small but true difference might be missed in this analysis of just over 100 patients (type 2 error). We would counter that any true difference in outcomes that went unnoticed in this study would have to be of such small magnitude that its practical relevance would be extremely limited.

The improvements seen in our patients following a period of worsening disease activity, irrespective of treatment, is typical of the waxing and waning course of rheumatoid arthritis and many other chronic diseases. The appearance of an improvement when a subset of patients is analysed, starting at a time of worsened disease, causes an error in interpretation known in statistics as “regression to the mean.” It has been suggested that the sometimes impressive responses seen in the placebo groups in clinical trials are explained to a much greater extent by regression to the mean than by the actual placebo effect.^10,11

One could speculate that infliximab dosage increases might be prompted by decreasing efficacy as a result of anti-infliximab antibody formation. We cannot at this time test for such antibodies. However, the lack of a measurable increase in infusion reactions in these patients around the time of the dose increase would argue indirectly against this possibility.

The cost of infliximab is very high, and the typical dose increase doubles the cost of the treatment. In addition, some risks may be greater with higher doses. As indicated above, in patients with severe pre-existing congestive heart failure, increased mortality was seen with the higher infliximab dosage.² In the ATTRACT clinical trial, while the total number of malignancies was not higher than expected, the distribution of malignancies within the treatment groups was skewed towards the higher dosages.^1,12 And in previous studies with another TNFα blocking agent, D2E7 (adalimumab, Humira), a risk for tuberculosis was observed, especially at high dosages.¹³ Thus a burden is imposed on physicians who elect to prescribe higher infliximab dosages to assess the results of this intervention. When informally reviewing the cases in this study we found, rather soberingly, that the treating physicians tended to assess the results as more favourable than was warranted by the actual outcomes, and that even when the results were assessed as being no better than before, the dose increases were rarely reversed.

We conclude that dosage increases of infliximab may have less clinical benefit than is expected or perceived in clinical practice, and that much of the perceived benefit is, in fact, a regression effect. We recommend that formal studies of infliximab dose escalation be undertaken and that individual physicians wishing to increase infliximab dosage do so under very clearly prespecified conditions of follow up. We also conclude that use of formal surveillance systems similar to the one used in this study should be encouraged when introducing new drugs, in order to permit optimisation of their use in clinical practice and to avoid substantial costs for perceived but scientifically unverifiable patient benefits.