Ard.bmj.com



SUPPLEMENTARY MATERIAL

Treatment assignment and blinding

An independent, unblinded pharmacist, nurse, or physician was instructed by the investigator to perform the randomisation call to the interactive voice response system and receive the treatment assignment information. The unblinded pharmacist, nurse, or physician prepared the canakinumab or matching placebo solutions in syringes. Colchicine 0.5 mg or matching placebo capsules were supplied as blinded medication kits. Colchicine was from a single source and was provided centrally by Novartis. The unblinded pharmacist, nurse, or physician gave the study drugs, without revealing the identity of the drugs, to the investigator for dispensing to the patient. Randomisation data were kept strictly confidential and patients, investigator staff, individuals performing assessments, and data analysts (except those involved in the interim analyses), were blinded to the identity of the treatment from the time of randomisation until database lock.

Allopurinol dosing

Allopurinol dose was based on estimated creatinine clearance rate (CrCl): 100 mg/day for patients with a CrCl of 30–60 mL/min; 200 mg/day for patients with a CrCl of 60–90 mL/min, and 300 mg/day for patients with a CrCl of over 90 mL/min. Patients who had initiated allopurinol before study entry had to change to the specified maintenance dose before entering the study. The allopurinol maintenance dose was to remain constant up to week 16, unless there were safety concerns.

Rescue medication

Patients could take naproxen 500 mg twice daily as needed as rescue medication during a gouty arthritis flare for up to 7 days. If pain relief with naproxen was inadequate or patients were contraindicated to non-steroidal anti-inflammatory drugs (NSAIDs), patients could take prednisolone (up to a maximum daily dose of 30 mg) or prednisone, 30 mg/day for up to 5 days. If symptoms of the gouty arthritis flare persisted for more than 7 days, patients were to contact an investigator to ensure that they had no adverse events that could prevent further rescue medication use, and could then receive rescue medication for a further 7 days. Patient assessments of pain were to be performed before taking rescue medication.

Exclusion criteria

Exclusion criteria were: a gouty arthritis flare within 2 weeks of screening; history of allergy, contraindication, or intolerance to allopurinol or colchicine; absolute or relative contraindication to naproxen and oral prednisolone/prednisone; history of bone marrow suppression; use of prohibited medications before screening (chronic NSAID or systemic steroid use at least 1 week before screening; intra-articular or periarticular injection of corticosteroids within 4 weeks of baseline; any hypouricaemic therapy other than allopurinol within 1 month of baseline; any tumour necrosis factor inhibitor within 8 weeks of baseline; anakinra within 1 day of baseline; rilonacept within 1 week of baseline; any other biologic treatment within 8 weeks of baseline; intravenous immunoglobulin therapy within 8 weeks of baseline; leflunomide within 4 weeks of baseline; any immunosuppressive or immunomodulatory therapy within 4 weeks prior to baseline; or other investigational non-biologic drugs at the time of enrolment, or within 30 days or five half-lives before screening, whichever was longest); rheumatoid, infectious/septic, or other acute inflammatory arthritis; severe renal function impairment (estimated CrCl, 450 ms for men and >470 ms for women at screening or baseline; significant medical problems (e.g. uncontrolled hypertension, congestive heart disease, uncontrolled diabetes type 1 or type 2, thyroid disease; history of malignancy of any organ system within the preceding 5 years); history of organ transplantation; pregnant or nursing (lactating) women; women who are physiologically capable of becoming pregnant unless using an acceptable method of contraception; familial and social conditions rendering regular medical assessment impractical; and history of alcohol or drug abuse within 12 months of randomisation.

Other secondary outcomes

The following secondary outcomes were measured in the study but are not reported here: the efficacy of canakinumab compared with colchicine for patient’s global pain intensity on a 0-100 mm visual analogue scale (VAS) ranging from no pain (0) to unbearable pain (100) and a 5-point Likert scale (0=none, 1=mild, 2=moderate, 3=severe, 4=extreme) during gouty arthritis flares; physician’s global assessment of response to therapy on a 5-point Likert scale at 16 weeks; serum amyloid A levels; and the amount of rescue medication taken per gouty arthritis flare and taken in total up to 16 weeks post-randomisation.

Sample size determination

A sample size of 350 patients (50 patients in each of the 5 canakinumab single-dose groups and 100 patients in the colchicine group) was considered to be sufficient to give a 95% confidence interval (CI) of the target dose with reasonable precision, assuming: a target dose of 100 mg; use of a log-linear dose response model; and a mean number of flares within 16 weeks of 0.2 for the canakinumab 300 mg group and 0.6 for colchicine 0.5 mg. The inclusion of the two covariates, the stratification variable and body mass index (BMI), was not considered in the sample size calculation, on the assumption that their inclusion would decrease the variance of the residual error, and therefore increase the precision. The unconditional CI accounting for model selection was wider than the conditional CI that was based on the finally selected model. An accounting factor of 1.25 based on general simulations was taken into account. In order to increase the precision of the analysis the planned total sample size of 400 patients (including 50 patients in the multiple-dose canakinumab group) was increased by 10% to approximately 440 patients (55 patients in each canakinumab arm and 110 in the colchicine group).

Determination of the primary outcome

A pre-planned interim analysis was performed when 200 patients had completed 16 weeks. When all patients had completed 16 weeks the study was unblinded and a further efficacy analysis was performed. Efficacy and safety analyses were then performed when all patients had completed the study (24 weeks).

For analysis of the primary outcome, the number of flares per patient during 16 weeks was modelled using a negative binomial distribution and a common overdispersion (variance divided by expectation minus 1) for all treatment arms. The negative binomial distribution is a generalisation of the Poisson distribution. The poisson distribution can be used to model highly skewed discrete count data and assumes the variance equals the mean. However very often the variance exceeds the mean leading to overdispersion (underestimation of standard errors and overestimation of test statistics/significance). The negative binomial distribution however allows the variance to exceed the mean by including an additional parameter in the model. Therefore the negative binomial distribution can take account of the overdispersion and produce more reliable estimates.

The logarithm of the expectation of the number of flares during 16 weeks was described by four different candidate models (see below) for the dose-response relationship between the canakinumab (single) dose groups and a constant parameter for the comparator. For those patients who were observed for less than 16 weeks the negative binomial distribution was adjusted according to the observed time (i.e. both, the expectation and the overdispersion were divided by the observed time and multiplied by 16 weeks).

This model assumes that the number of flares for each patient follows a Poisson distribution with parameter λt, where t is the observation time and λ varies between patients according to a gamma distribution with a common scale parameter.

The following four candidate models and the corresponding mean response were used:

1. Emax model: θo + θ1 dose/(ED50 + dose)

2. linear in log-dose model: θo + θ1 log(dose)

3. logistic model: θo + θ1 /(1 + exp((ED50 - dose)/τ))

4. linear: θo + θ1 dose

The null hypothesis of a constant dose-response curve was tested at a significance level of 5% (adjusted for multiplicity) against the alternative hypothesis of a non-constant dose-response curve for each of the four candidate models. Among those models in which the null hypothesis was rejected, the model with the highest value of the AIC (Akaike information criterion) was selected for the estimation of the target dose. The target dose was estimated using maximum likelihood including a 95% CI. In order to account for the model uncertainty, the CI was determined by bootstrap methodology. Repeated bootstrap samples were generated within the treatment arms and the target dose estimation performed according to the above described procedure. The 95% CI was estimated as the 2.5% and 97.5% quartiles of the bootstrap estimates.

The variables allopurinol dose at baseline (categorical) and BMI at baseline (continuous) were included as covariates in the models.

Populations for analysis

All efficacy endpoints were analysed using the full analysis set (i.e. all randomised patients who had at least one post-baseline VAS assessment, analysed according to assigned treatment) and safety assessments were based on the safety analysis set (i.e. all randomised patients who received study drug and had at least one post-baseline safety assessment, analysed according to treatment received).

Analysis of secondary outcomes

The mean number of flares per patient at 16 weeks was initially analysed according to the pre-planned ANCOVA with treatment group, allopurinol dose at baseline, and BMI at baseline as covariates. (ANCOVA is a general linear model which tests whether a certain factor (normally treatment) has an effect on a continuous outcome variable after removing the effect of other quantitative predictors.) However, as the distribution of flares per patient was observed to be skewed, this analysis was considered inappropriate for these data. Therefore a post-hoc analysis using a negative binomial model with treatment group, allopurinol dose at baseline, and BMI at baseline as covariates and log (time on study) as an offset was performed. To adjust for multiplicity, a conservative Bonferroni approach was used. A p-value of less than 0.01/0.0083 (being 0.05 divided by 5/6 – being the number of treatment comparisons made) was considered statistically significantly superior to colchicine with respect to the number of flares.

The number of patients with flares within 16 weeks was analysed using logistical regression with treatment group, allopurinol dose at baseline, and BMI at baseline as covariates. An odds ratio >1 indicates that a gouty arthritis flare is more likely to occur in comparison with the colchicine 0.5 mg group.

The time to first flare was analysed using Cox’s proportional hazard regression model with treatment group, allopurinol dose at baseline, and BMI at baseline as covariates. The proportion of patients experiencing a flare at 16 weeks was calculated using the Kaplan–Meier formula; 95% CI was calculated using Greenwood’s formula.

The duration of gouty arthritis flares was analysed using ANCOVA with treatment group, allopurinol dose at baseline and BMI at baseline as covariates. The proportion of patients with serum urate levels ≥6 and ≥9 mg/dL were determined per treatment arm.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download