Large randomized controlled trials are the best way to solve important medical questions about therapy, but they are difficult, time-consuming, and expensive. Meta-analyses attempt to combine the results from previously published studies to enhance the statistical power of an analysis in lieu of recruiting the large number of patients that would otherwise be necessary.

We did a meta-analysis of all the published, randomized, double-blind, placebo-controlled trials of azathioprine or 6-mercaptopurine therapy in patients with active or quiescent Crohn disease.

Methods

Top Methods Results Discussion Author & Article Info References

We did a literature search using the MEDLINE database (1966 to May 1994) with the Medical Subject Headings anti-metabolites, azathioprine, or mercaptopurine and Crohn disease or inflammatory bowel disease. Citations were limited to those published in English and French. We also did a manual search using references from these articles, review articles, and proceedings from major gastrointestinal meetings. Double-blind, placebo-controlled trials of oral azathioprine or 6-mercaptopurine therapy in patients older than 18 years who had active or quiescent Crohn disease were selected. Active disease was defined as the presence of serious or increasing symptoms from intestinal or extraintestinal Crohn disease. Quiescent disease, or disease in remission, was defined as the presence of mild or no symptoms before the patient entered the study, regardless of the use of prophylactic medication.

Each study was independently reviewed by three authors, and methodologic criteria and the results of each study were recorded. We collated data according to the definition of success used in each study because different instruments were used to measure disease activity and no single definition of response was used in all the studies. Although some of the instruments have not been externally validated, they all appeared to have provided a reasonable measurement of clinically active and inactive disease. For studies of therapy in active disease, we recorded the number of patients entering clinical remission as defined by the study. For studies of maintenance therapy, we recorded the number of patients maintaining clinical remission as defined in the study. We also extracted data to measure the effect of dose and duration of azathioprine or 6-mercaptopurine therapy; the effect of concurrent therapy with sulfasalazine, 5-aminosalicylic acid, and steroids; the number of fistulae that improved or healed; and the number of patients who withdrew from the study because of toxicity. In crossover studies, we only incorporated data from the first portion of the study to avoid possible carryover effects of medication into the second part of the study and to make these studies more comparable to those that did not have a crossover design. For all studies, we enumerated adverse effects severe enough to require the withdrawal of a patient from a trial. All results were tabulated on an intention-to-treat basis.

Two authors independently did a quality analysis of each study using criteria derived from those of Chalmers and colleagues [15]. The quality analysis was divided into three parts. Part one addressed the study protocol: patient selection, differences in drug and placebo regimens, randomization and blinding techniques, and a priori estimates of sample size. Part two addressed the analysis: details of the statistical significance of major end points, the presence of a posterior ß-error estimate for negative trials, use of the intention-to-treat rule, and analysis of adverse effects. Part three addressed the presentation of results: tests for equivalence of treatment and control groups, tabulation of remissions and relapses, and presentation of data on the timing of relapse and remission events. The complete quality analysis criteria may be obtained from the authors on request. Discrepancies between reviewers about retrieved data or quality assessment scores were settled by consensus.

Statistical Analysis

We separately analyzed active and quiescent disease therapy groups and used commercially available statistical software to calculate odds ratios and 95% CIs for individual studies [16, 17]. For the analysis of disease response and the effect of drug dose, duration of therapy, and cumulative dose, we calculated an estimate of the common odds ratio using a maximum likelihood estimate to produce a logistic regression analysis based on a binomial model [18]. For each model, the studies and treatment assignment were entered as indicator variables [19]. If the residual deviance was judged significant by a chi-square test after the model was fitted, then the model did not fully explain the difference between treatment and placebo groups after adjustment for study differences. This is an indication of the heterogeneity of the odds ratios, and it prompted a search for differences in the methods used by the studies, such as dose and duration of therapy, that could account for the heterogeneity. We then constructed a new model that incorporated these terms. A model was considered satisfactory if the residual deviance was not statistically significant (P > 0.05).

In the primary analysis, the studies were weighted according to the total number of participants. We assessed the effect of quality scores by repeating the analysis using the quality scores to weight the studies according to the techniques of Detsky and colleagues [20]. We also repeated the analysis using the studies that scored at or above the median quality score, weighted by the number of study participants.

This type of logistic regression analysis reproduces the traditional Mantel-Haenszel analysis [20-23] but also allows the modeling of more complex situations by including strata variables and other terms to measure the effect of variables such as drug dose, duration of therapy, and cumulative dose. We graphed the individual trial odds ratios and 95% CIs to confirm that a substantial overlap existed. We also did a sensitivity analysis by excluding each study in turn to ensure that no single study was solely responsible for a significant result.

For the secondary analysis, we evaluated fistula response, steroid-sparing effect, and adverse effects. Data from the placebo and therapy groups were totaled, and pooled odds ratios and 95% CIs were calculated. Note that if an odds ratio exceeds 1.0, a beneficial effect of therapy is suggested; if the CI for an odds ratio includes unity, there is no statistical evidence for the superiority of treatment over placebo.

Results

Top Methods Results Discussion Author & Article Info References

Of the 222 studies we reviewed, 9 trials met the inclusion criteria [13, 14, 24-30]. Some of the randomized, double-blind, placebo-controlled trials were reported more than once, usually as an abstract at a symposium before formal publication. Results of one trial involving 11 patients [31] were published only in abstract form in 1974; this report contained insufficient information for analysis and was therefore excluded from further consideration. Results of another trial [30] were published in 1994 as an abstract, but the full-length manuscript is currently in preparation (Wright JP. Personal communication). The results of this trial as detailed in the abstract were included in the analysis, but we could not assign a quality score because the full-length report was not available.

The nine selected trials had various designs (Tables 1 and 2): Four were exclusively concerned with therapy for active disease, two addressed maintenance of remission, and three had separate arms for patients with active and quiescent disease. Because the trials were done over 25 years on three continents, it is unlikely that many patients participated in more than one trial. We considered active disease and disease in remission separately so that patients participating in both these groups would not bias the results. There is a reasonable chance that the trials of O'Donoghue and colleagues [29] and Willoughby and associates [14], in which azathioprine was used for maintenance of remission, may have enrolled some of the same patients; both these trials were done at St. Bartholomew's Hospital in London, United Kingdom, and their results were reported in 1978 and 1971, respectively. The maximum potential number of shared patients is 10, which represents 3% of all patients included in the analysis of maintenance therapy. This small number and the fact that the odds ratios of both studies are similar to those in other studies suggests that no substantial bias from that source occurred.

Only Present and colleagues [25] used 6-mercaptopurine in their trial; all other investigators used azathioprine. Azathioprine is rapidly converted to 6-mercaptopurine in vivo; because the clinical indications of these two drugs are similar, combining these trials in the analysis is appropriate.

The investigators of the National Cooperative Crohn's Disease Study (NCCDS) [27] divided their study into three parts. Part I, phase 1, lasted 17 weeks and dealt with patients with active disease; part I, phase 2, examined the ability of the patients who achieved remission in phase 1 to maintain their remission while receiving azathioprine or placebo for an additional 35 weeks. Patients studied in part II entered the trial while in remission, and, although patients in this arm were observed for 2 years, we chose to censor the results at 1 year because the other studies in this analysis were of similar duration. The investigators of the NCCDS did not compare results of part II with those of part I, phase 2, because patients in the latter started part I, phase 1, with active disease, whereas those in the former were already in remission. For our meta-analysis, we used both groups because all the patients were in remission at the start of the trial arms.

We also divided the study of Candy and colleagues [30] into two parts. In the first part, active disease was treated with a 3-month tapering course of prednisolone plus either azathioprine or placebo. Patients who had successfully completed part 1 were included in the second part, and their ability to maintain remission while receiving study medication alone for an additional 12 months was examined.

Quality scores (maximum, 100) ranged from 42 to 89 (median, 52) for studies of active disease Table 1 and from 46 to 89 (median, 51) for studies of quiescent disease (Table 2). The mean absolute difference in the inter-evaluator score for each study was 7 points (95% CI, 3 to 10 points).

Azathioprine or 6-Mercaptopurine Therapy in Active Disease

Four studies [13, 24-26] exclusively addressed active disease, and three (NCCDS [27], Willoughby and colleagues [14], and Candy and colleagues [30]) were active-disease therapy arms from multiarm studies. Three hundred sixty-seven patients were enrolled; 177 received azathioprine or 6-mercaptopurine and 190 received placebo. All patients were symptomatic: Some had an acute exacerbation of previously quiescent disease [24], others had chronic disease unresponsive to other therapy [14, 26], and some [13, 25, 27, 30] were evaluated only on the severity of their symptoms at the time of randomization. In three studies [25, 27, 30], the Crohn's disease activity index (CDAI) [32-34] was used to assess disease activity. In the other studies, various similar schemes were used that evaluated a combination of the patient's subjective sense of well-being and objective symptom score, a physician's blinded clinical assessment, and results of basic laboratory tests such as a complete blood count and tests for erythrocyte sedimentation rate and albumin level.

The definition of response differed among the studies (Table 1). The steroid-sparing effect of therapy was a primary end point for some studies [14, 24-26], whereas others [13, 30] were concerned only with clinical response and did not consider the effect of concurrent medications. Investigators of the NCCDS [27] measured only clinical response and, by design, excluded concurrent therapy, even though 37% of patients in each of the control and azathioprine groups had discontinued steroid therapy within 2 weeks of entering the trial.

In studies with patients already receiving sulfasalazine [13, 14, 24-26] or 5-aminosalicylic acid [24], the use of these drugs was continued. Some studies did not allow dose adjustment [14], but others did [13, 25, 26]. Investigators of one study [24] did not indicate whether changes in dose were allowed, and those of another study [30] did not provide adequate information for evaluation of concurrent therapy. When such data were reported, patients receiving 5-aminosalicylic acid or sulfasalazine were equally distributed between control and treatment groups at randomization, but changes in concurrent drug use were not measured as an end point in any of the studies.

The response rates for the control and treatment groups of each study are shown in Table 1 and Figure 1. One hundred of 177 patients receiving treatment responded (56% [CI, 49% to 64%]) compared with 60 of 190 patients receiving placebo (32% [CI, 25% to 38%]). The estimated common odds ratio for response to azathioprine or 6-mercaptopurine was 3.09 (CI, 2.45 to 3.91). When only active drug compared with placebo was considered, significant residual deviance was observed (P = 0.001), which indicates that this model did not explain all the difference between treatment and placebo groups. However, when the data were reanalyzed with incorporation of a term for the duration of therapy, the residual deviance was not significant (P = 0.09); this finding indicates that differences in duration of therapy were responsible for the poor fit of the first model. In addition, the coefficients associated with each study did not significantly differ from 0, which indicates that all the studies conformed to this model. After analyzing the data by excluding the results of each study in turn, we still observed a statistically significant effect for antimetabolite therapy. This observation suggests that no single study substantially altered the result. When we excluded the study of Present and colleagues [25], in which 6-mercaptopurine was used, and combined the trials in which azathioprine was used, the estimated common odds ratio was 1.45 (CI, 1.12 to 1.87).

View larger version (22K): [in this window] [in a new window]   Figure 1. Odds ratio of response in randomized controlled studies of azathioprine and 6-mercaptopurine for active disease. The odds ratio (log scale) for each study is represented by the filled square, and the 95% CIs are represented by the horizontal bars. Values greater than 1.0 indicate a therapeutic benefit compared with placebo. The estimated common odds ratio for response was 3.09 (CI, 2.45 to 3.91). Study reference numbers are in parentheses. NCCDS = National Cooperative Crohn's Disease Study.

Azathioprine Therapy in Quiescent Disease

Two studies [28, 29] exclusively addressed quiescent disease, and four (part I, phase 2, and part II of NCCDS [27]; Willoughby and colleagues [14]; and Candy and associates [30]) used separate arms for quiescent disease (Table 2). Three hundred nineteen patients were studied; 136 received azathioprine and 183 received placebo. 6-Mercaptopurine was not used in any trial. Two studies [14, 28] enrolled patients who had become dependent on steroids to maintain remission and attempted to withdraw steroids after azathioprine or placebo was added. Investigators of another study [29] identified a group of patients whose conditions stabilized while they were receiving azathioprine, replaced azathioprine with placebo in half of these patients, and followed them for a year. Patients who had achieved remission from active disease in a preceding portion of the trial were enrolled in part I, phase 2, of NCCDS and in part 2 of Candy and colleagues' study [30]. In part II of NCCDS, patients already in remission were enrolled, although some of these had recently had bowel resections.

Table 2 and Figure 2 show the response rates for the control and treatment groups of each study. Overall, 91 of 136 patients receiving treatment responded (67% [CI, 59% to 75%]) compared with 96 of 183 patients receiving placebo (53% [CI, 45% to 60%]). The estimated common odds ratio for response to azathioprine was 2.27 (CI, 1.76 to 2.93). Although there was no significant residual deviance (P = 0.12), two studies (part I, phase 2, of NCCDS and the study by Candy and colleagues [30]) did not conform to this model. In the former study, both the placebo and azathioprine groups had a high rate of maintaining remission: 75% and 84%, respectively, at 35 weeks. These patients originally presented with active disease, and, in part I, phase 1, of the trial, had successfully responded within 17 weeks to either placebo or azathioprine. Thus, the placebo group of phase 2 consisted of patients who had spontaneously entered remission and who may have had less severe disease. The azathioprine group may represent a select population of azathioprine responders who had received 17 weeks of therapy before entering into phase 2, which thereby gave them a head start on therapy. A high rate of maintenance of remission and a low odds ratio might therefore be expected. In part 2 of the study by Candy and colleagues, patients who had achieved remission with azathioprine or placebo and a tapering course of prednisolone were studied. In part 1, the placebo group had a high rate of remission (67%) because they received prednisolone for the entire 13 weeks of the trial. The discontinuation of prednisolone therapy in part 2 of the study exacerbated disease in patients with active disease; this resulted in an unusually low remission rate in the placebo group (10%). The patients who continued to receive azathioprine had a response rate similar to that of the treatment groups of other studies (56%), resulting in a high odds ratio of response. A sensitivity analysis done by excluding the results of each study in turn still showed a statistically significant therapeutic effect for antimetabolite therapy; this confirms that no single study altered the conclusions.

View larger version (23K): [in this window] [in a new window]   Figure 2. Odds ratio of response in randomized controlled studies of azathioprine for maintenance therapy. The odds ratio (log scale) for each study is represented by the filled square, and the 95% CIs are represented by the horizontal bars. Values greater than 1.0 indicate a therapeutic benefit compared with placebo. The estimated common odds ratio for response was 2.27 (CI, 1.76 to 2.93). Study reference numbers are in parentheses. NCCDS = National Cooperative Crohn's Disease Study.

Effect of Dose, Duration, and Cumulative Dose

Azathioprine and 6-mercaptopurine have no established relative bioactivity. The dose-response calculation for treatment of active disease assumed equal absorption and complete in vivo conversion of azathioprine to 6-mercaptopurine [35]; the calculation yielded a relative potency of 1.82 for 6-mercaptopurine compared with azathioprine based on their molecular weights. This calculation predicts that 6-mercaptopurine, 1.5 mg/kg body weight per day, as used by Present and colleagues, is equivalent to azathioprine, 2.7 mg/kg per day.

When we analyzed the data on active disease therapy for the effect of drug dose (range, 2.0 to 3.0 mg of azathioprine/kg per day) and duration of therapy (range, 2 to 12 months), each was found to be statistically significant by itself (P < 0.0001 for each), but only duration of therapy was significant when the two were considered simultaneously (P = 0.0003). The estimated common odds ratio for response increased by 1.25 (CI, 1.20 to 1.30) with each additional month of therapy; residual deviance was not significant (P = 0.13). This effect can be seen in Table 3, which shows the odds ratios of response in active disease with less than 17 weeks, 17 weeks, and more than 17 weeks of therapy. The odds ratio reached statistical significance at 17 weeks and increased thereafter (P = 0.03 for trend).

When we analyzed the maintenance therapy data for the effect of azathioprine dose (range, 1.0 to 2.5 mg/kg per day) and duration of therapy (range, 5.5 to 12 months), each was significant by itself (P = 0.0001 and P = 0.002, respectively), but only dose was significant when the two were considered together (P = 0.008). The estimated common odds ratio for response increased by 1.75 (CI, 1.51 to 2.03) with each additional mg/kg per day. Residual deviance was not significant (P = 0.55).

One way to simultaneously consider the effect of dose and duration is to calculate a cumulative dose for each trial (mg/kg per day multiplied by the number of days in the trial) and to use this dose in the regression analysis. For treatment of active disease, this model predicted an increase in odds ratio of 1.32 (CI, 1.25 to 1.38) per 100 mg/kg; residual deviance was not significant (P = 0.10) (Figure 3). With maintenance therapy, the predicted odds ratio increased by 1.19 (CI, 1.13 to 1.24) per 100 mg/kg; residual deviance was not significant (P = 0.52) (Figure 4).

View larger version (15K): [in this window] [in a new window]   Figure 3. Effect of cumulative dose in the therapy for active disease. The odds ratio for each study is represented by the filled square, and the 95% CIs are represented by the vertical bars. Numbers in parentheses are the reference numbers for each study. The odds ratio of response is increased by 1.32 (CI, 1.25 to 1.38) with each 100 mg/kg body weight increase in cumulative dose.

View larger version (13K): [in this window] [in a new window]   Figure 4. Effect of cumulative dose in maintenance therapy. The odds ratio for each study is represented by the filled square, and the 95% CIs are represented by the vertical bars. The numbers in parentheses are the reference numbers for each study. The odds ratio of response is increased by 1.19 (CI, 1.13 to 1.24) with each 100 mg/kg body weight increase in cumulative dose.

Effect of Quality Scores

The quality score attempts to quantify the integrity of a study on the basis of the features of the protocol, the statistical analysis, and the presentation of results in the belief that a study with a high score gives a more accurate estimate of the true effect of therapy. The use of the scores is controversial because they are not a direct measure of this effect [36]. Detsky and colleagues [20] suggested that these scores be used as weights or to determine a threshold value such as the median for studies to include in an analysis. When we used the quality scores in this way, the estimated common odds ratios changed little, but the CIs widened (Table 4). When we used the scores as weights, we did not observe a statistically significant effect for therapy in active or quiescent disease. When we repeated the analysis using the studies at or above the median quality score, the odds ratio for active disease was 3.66 (CI, 2.11 to 6.34), whereas the odds ratio for quiescent disease was 1.59 (CI, 0.9 to 2.8). The best use of a quality scoring system may be to exclude from an analysis studies that have a less rigorous design and execution because such studies may be more susceptible to bias.

Response of Fistulae to Azathioprine and 6-Mercaptopurine

Fistula responses were reported in five studies [13, 14, 25, 26, 28]. In two trials [24, 27], no statistically significant difference in healing rates between placebo and therapy groups was reported. However, because these studies did not include numeric data, we could not incorporate their results. Two trials [29, 30] did not report data on fistula. We defined fistula response as complete healing or decreased discharge. If a fistula developed during the study, it was included as an "unhealed" fistula. The results from the study by Present and colleagues [25] include data from both arms of the crossover because data from the individual arms were not reported. All other data from crossover studies were obtained before crossover. Twenty-two of 41 patients receiving therapy responded (54% [CI, 37% to 69%]) compared with 6 of 29 patients receiving placebo (21% [CI, 8% to 40%]). This resulted in a pooled odds ratio of 4.44 (CI, 1.50 to 13.20) favoring fistula healing.

Steroid-Sparing Effect

The ability to reduce the prednisone or prednisolone dose was an important outcome measure for some studies. It was assessed variously as the ability to follow a predefined steroid-tapering regimen and as a reduction in the steroid dose required to maintain remission in patients already receiving steroids. Because of the variation in data presentation, we could not determine a percentage change in steroid use in patients receiving antimetabolites. We defined reduction in steroid use as successful completion of a tapering regimen or reduction in the prednisone or prednisolone dose to less than 10 mg/d with control of symptoms. In the five studies reporting steroid data [13, 14, 24, 25, 30], steroid consumption was decreased in 76 of 117 patients with active disease who received antimetabolites (65% [CI, 56% to 74%]) compared with 39 of 109 patients receiving placebo (36% [CI, 27% to 45%]). The pooled odds ratio of 3.69 (CI, 2.12 to 6.42) indicates a significant steroid-sparing effect. In the two studies reporting steroid data [14, 28], steroid consumption was decreased in 13 of 15 patients receiving maintenance therapy with azathioprine (87% [CI, 60% to 98%]) compared with 8 of 15 patients receiving placebo (53% [CI, 27% to 79%]). The pooled odds ratio was 4.64 (CI, 1.00 to 21.54).

Other Effects of Therapy

There were insufficient data to examine the relation between therapy and the duration of disease before patients entered a study, the location of disease, the presence of extra-intestinal manifestations of Crohn disease, or the effect of concurrent therapy with 5-aminosalicylic acid or sulfasalazine.

Adverse Effects of Therapy

Adverse events severe enough to cause withdrawal from a trial occurred in 27 of 302 patients receiving azathioprine or 6-mercaptopurine (8.9% [CI, 6.0% to 12.7%]) and in 6 of 353 patients receiving placebo (1.7% [CI, 0.6% to 3.7%]); the pooled odds ratio was 5.26 (CI, 2.20 to 12.60). When specified, the most common adverse effects were allergic reactions (2.0%) consisting of fever or rash, or both, and arthritis; leukopenia (1.7%); pancreatitis (1.3%); and nausea (1.3%) (Table 5). No malignancies were reported. One patient died [29]; this patient had had a long history of azathioprine use, developed persistent leukopenia while receiving azathioprine, and subsequently died of infection. There was a trend toward an increasing incidence of adverse effects with greater cumulative dose of antimetabolites.

Discussion

Top Methods Results Discussion Author & Article Info References

In active disease, azathioprine and 6-mercaptopurine had a estimated common odds ratio of response of 3.09 (CI, 2.45 to 3.91). Although there was a statistically significant increase in response with increasing duration of therapy, the antimetabolite dose did not have an independent effect beyond 2.0 to 3.0 mg/kg per day. The delayed response to antimetabolite therapy is a recurring theme in the literature. The time of peak response in the studies of active disease that reported it ranged from 9 weeks [27] to more than 26 weeks [25]. In our analysis, the response odds ratio reached 1.95 (CI, 1.10 to 3.46) at 17 weeks of therapy and 19.2 (CI, 6.27 to 58.8) at greater than 17 weeks of therapy. We noted a trend for increased response in studies of greater duration (P = 0.03). Seventeen weeks seems to be the minimum period for an adequate trial of antimetabolite therapy. The absence of a statistically significant effect of increasing dose in therapy of active disease is somewhat paradoxical but may be due to the narrow range of doses in these studies. To capture the effect of both dose and duration of therapy and to aid in the comparison of different antimetabolite therapy protocols, we propose the concept of cumulative dose.

The role of combined antimetabolite and steroid therapy in active disease has been controversial. The NCCDS did not show a statistically significant effect for azathioprine monotherapy and has been criticized for not allowing concurrent steroid therapy during the delay period before azathioprine could act and for not allowing sufficient time to assess the response to azathioprine [40]. Willoughby and colleagues [14], Ewe and colleagues [24], and Candy and colleagues [30] all used azathioprine or placebo and a tapering course of prednisolone. Ewe and associates and Willoughby and coworkers both concluded that the treatment group responded more rapidly and completely, whereas Candy and colleagues did not find a statistically significant effect for therapy because of high response rates in the placebo group. Combination therapy with azathioprine and steroids may lead to a higher response rate with less steroid use. If one considers only the studies of active disease lasting 17 weeks or longer that allowed concurrent steroid use [14, 24-26, 30] the estimated common odds ratio of response increases to 4.34 (CI, 2.44 to 7.72).

Although azathioprine and 6-mercaptopurine are used for similar indications in Crohn disease, little information is available about their relative potency and differences in their therapeutic effects. The choice of drug for an individual patient seems to depend more on local experience and personal convictions than on objective data. The study by Present and colleagues [25] was the only randomized, double-blind trial to use 6-mercaptopurine; however, only active disease was studied. Present and colleagues' results strongly favored therapy over placebo and thus substantially influenced the estimated common odds ratio. This result could be attributed to substantive differences between the medications, but it may also be due to factors in the study design. The investigators used therapy for 52 weeks and allowed concurrent steroid use; both these factors may increase the probability of achieving remission in active disease.

Among patients in remission, azathioprine has a benefit for maintaining remission (estimated common odds ratio, 2.27 [CI, 1.76 to 2.93]). Because the trials were of relatively short duration, the long-term effectiveness of azathioprine is also unclear. The clinical significance of this finding is not clear, but it would be important to patients who have had severe or persistent disease or complications from medical or surgical therapy. We observed an increased response with increasing azathioprine dose (P = 0.008) but not for increased duration of therapy when the effect of dose was simultaneously considered. Increased cumulative dose was also significantly associated with maintenance of remission (P = 0.01). Part II of NCCDS did not show an effect for azathioprine at 1.0 mg/kg per day, but in light of the above analysis, a beneficial effect might have been observed if a larger dose had been used.

Although there may be a subgroup of patients who respond particularly well to azathioprine [13, 26, 28], it has not been possible to identify them before initiating therapy. One of the difficulties in comparing the studies of maintenance of remission is the difference in clinical end points. Two studies [14, 28] measured the dose of prednisone required to maintain symptom control; another [29] withdrew azathioprine from a highly selected cohort of patients who had already responded to it; and others [27, 30] did not allow concurrent steroid therapy.

Our analysis supports the concept that azathioprine is useful as adjunctive therapy for reducing steroid use among patients with either chronically active disease or an acute exacerbation. We observed a steroid-sparing effect for active disease (pooled odds ratio, 3.69 [CI, 2.12 to 6.42]) and a similar trend for quiescent disease (pooled odds ratio, 4.64 [CI, 1.00 to 21.54]). The use of a potentially toxic drug such as azathioprine in patients with an acute flare of otherwise mild disease would seem overly aggressive, but in patients with extensive disease, chronic symptoms, or frequent exacerbations, this approach would seem reasonable.

This raises the question of what clinical end points are relevant. A disease activity score gauges the severity of illness, but not the cost at which this result is achieved. Patients who may particularly benefit from antimetabolite therapy include those who are dependent on steroids, refuse steroids, have steroid-associated side effects, or have contraindications to steroid use. Patients with severe disease or complications that are not surgically remediable may also benefit, but the use of steroids in this situation may be based as much on desperation as on hard data. To further address this question, inclusion of a quality-of-life score as an end point in future studies would be useful.

Although incorporating quality scores into the analysis did not greatly alter the estimated common odds ratios, it did widen the CIs (Table 4). This effect is most pronounced when the quality scores are used as weights because using them in this way ignores the number of patients in a study. The scores may be best applied to determine a threshold value for a satisfactory study. The analysis can then proceed using weighting by the number of study participants. The quality scores are arbitrary and do not address potential sources of bias outside of the methods.

The use of azathioprine for refractory fistulae was originally based on uncontrolled studies [3-58, 41]. For the five placebo-controlled trials that reported numeric results, the pooled response odds ratio of response was 4.44 (CI, 1.50 to 13.20), favoring healing or decreased discharge. Because most patients (46 of 70; 66%) came from a single study, that of Present and colleagues [25], these results must be interpreted with a degree of caution, particularly because data from two other studies that reported no significant effect [24, 27] could not be included in the analysis. It would be useful if future studies reported numeric results so that data could be pooled for meta-analysis.

The adverse effects of antimetabolite therapy are well recognized [42-45]. Because patients with Crohn disease may have similar symptoms, a comparison with a control group is useful to estimate the incidence of adverse effects attributable to therapy. The studies had a combined observation period of approximately 428 patient-years (236 patient-years for the placebo groups and 192 patient-years for the therapy groups). Patients receiving therapy had an 8.9% incidence of adverse effects, with a pooled odds ratio of 5.26 (CI, 2.20 to 12.60) for development of an adverse effect severe enough to require withdrawal from a trial. These data are similar to those of Present and associates [46], who reviewed their extensive experience with 6-mercaptopurine in 396 patients with inflammatory bowel disease and approximately 1800 patient-years of follow-up. They reported a higher incidence of adverse reactions, presumably because of the longer duration of therapy and observation: significant infection (7.4%), pancreatitis (3.3%), neoplasm (3.1%), bone marrow suppression (2.0%), allergy (2.0%), and drug hepatitis (0.3%). O'Brien and coworkers [11] reported an overall 10% incidence of side effects severe enough to justify stopping azathioprine or 6-mercaptopurine therapy in their uncontrolled series of 78 patients with Crohn disease.

Meta-analysis is susceptible to publication bias because it is possible that reports of trials with negative results have not been published; the more likely reason for studies not being published, however, is small sample size. Because antimetabolite therapy for Crohn disease is controversial, a study would probably be published regardless of its outcome; in fact, 9 of the 13 data groups analyzed could be considered "negative" because the CIs include unity. We discussed this project with physicians active in the inflammatory bowel disease field who were unaware of additional studies of antimetabolite therapy. Our literature search uncovered only one abstract that was not published as a full article [31].

Future clinical investigation of antimetabolite therapy in Crohn disease should address the effects of greater cumulative doses (for example, >500 mg/kg) because few controlled data are currently available at these levels. Unfortunately, assessment of these effects would require trials of long duration, which would be difficult and expensive.

Another potentially fruitful area of investigation would be to try to individualize therapy. 6-Mercaptopurine is inactivated by methylation, which is catalyzed by thiopurine methyltransferase. The toxicity and effectiveness of 6-mercaptopurine therapy in childhood acute lymphocytic leukemia is inversely correlated with the activity of this enzyme [47]. Evaluation of thiopurine methyltransferase activity in patients with Crohn disease may allow individualized dosing to produce an optimal therapeutic effect while minimizing the potential for leukopenia. Colonna and Korelitz [48] have advocated using the presence of mild leukopenia as an indicator of the appropriate therapeutic dose of 6-mercaptopurine. In their recent retrospective analysis of 6-mercaptopurine therapy in 98 patients with refractory Crohn's disease, they showed a correlation between leukopenia and the achievement and maintenance of remission.

Meta-analysis is a powerful technique for pooling data from multiple sources to increase the statistical ability to determine associations. However, there is a risk in combining studies with different therapeutic protocols and end points because they may not be measuring the same effect, a process akin to making punch from apples and oranges. The studies under consideration use different clinical instruments to evaluate disease activity. Each instrument appears to provide a clinically relevant definition of active disease and disease in remission. In addition, variation in drug dose and duration of therapy provides an opportunity to quantify the effect of these variables. We conclude that antimetabolite therapy has a role in a group of patients with active or quiescent Crohn disease. Its use can be most strongly advocated for patients with chronically active disease who require steroid therapy. It has a steroid-sparing effect and may be of use to patients who have contraindications to steroid therapy or complications resulting from steroid use. Azathioprine appears to be useful in maintaining remission in patients with stable or inactive disease.

It is the nature of Crohn disease to wax and wane, which makes assessment of response difficult. If antimetabolite therapy is to be used, our analysis suggests that therapy be continued for at least 17 weeks and preferably for 26 to 52 weeks. Patients with active disease should receive a tapering course of steroids during this delay. Ultimately, the choice to use antimetabolite therapy should be individualized, and therapy should be undertaken only after the risks and benefits are carefully discussed with the patient.