Mycophenolate Mofetil Versus Cyclophosphamide for Remission Induction in Childhood Polyarteritis Nodosa: An Open‐Label, Randomized, Bayesian Noninferiority Trial

Cyclophosphamide (CYC) is used in clinical practice off‐label for the induction of remission in childhood polyarteritis nodosa (PAN). Mycophenolate mofetil (MMF) might offer a less toxic alternative. This study was undertaken to explore the relative effectiveness of CYC and MMF treatment in a randomized controlled trial (RCT).

If left untreated, the mortality rate of childhood PAN was historically close to 100% within months of disease onset (12,13); with aggressive immunosuppression, the mortality rate is as low as 4% (4). Cyclophosphamide (CYC) has been used off-label for over 40 years in the treatment of PAN (14)(15)(16)(17) and is still recommended for induction of remission in childhood PAN, though this has never been studied in a pediatric randomized controlled trial (RCT) (12). If alternative treatment exists, it is desirable that CYC treatment in children be avoided, since adverse reactions associated with CYC include infertility and malignancy (18).
Mycophenolate mofetil (MMF) is an alternative immunosuppressant with lymphocyte selective suppressive effects, which is associated with remission rates similar to those observed with CYC in the treatment of lupus nephritis (19) and antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV) (20). MMF is not associated with urothelial malignancy or infertility and is used off-label in pediatric patients.
We hypothesized that MMF may be noninferior to CYC for induction of remission in childhood PAN and may be a less toxic alternative. Therefore, the purpose of the present study was to investigate the relative effectiveness of MMF and CYC for remission induction in childhood PAN. It is infeasible to conduct a conventional, definitive phase III study of childhood PAN due to its rarity. We therefore opted for a Bayesian approach to assess the relative efficacy of MMF and CYC. This was a 2-stage process. Stage 1 consisted of a robust 2-day elicitation process conducted to quantify clinical opinion in light of results from a trial in adults (mycophenolate mofetil versus cyclophosphamide for remission induction in ANCA-associated vasculitis [MYCYC]) (20). The results from this trial were previously published (21). Stage 2 was a multicenter, open-label RCT of mycophenolate mofetil versus cyclophosphamide for the induction of remission of childhood PAN (the MYPAN trial; http://www.mypan.org.uk), and these data were used to further quantify the relative effectiveness of each treatment for remission induction in patients with newly diagnosed childhood PAN.

PATIENTS AND METHODS
Study design and patients. MYPAN was an international multicenter, open-label, randomized controlled prospective trial comparing MMF treatment with intravenous (IV) CYC treatment for the remission of childhood PAN (Figure 1). The trial was sponsored by University College London, and the trial was coordinated and data were stored by the Liverpool Clinical Trials Centre (LCTC) at the University of Liverpool. Centers were identified among members of the Paediatric Rheumatology International Trials Organisation (PRINTO) (www.printo.it) (22).The trial was conducted using a Bayesian noninferiority design (noninferiority margin 10%). Children were randomized 1:1 to receive either MMF (1,200 mg/m 2 /day, maximum 1 gm twice daily) (12,23) or a standard IV CYC regimen (12). Randomization was achieved using a secure web-based tool generated centrally by the LCTC. Minimization variables for randomization were planned additional therapy with methylprednisolone >15 mg/kg at trial entry (yes/no) and plasma exchange at trial entry (yes/no). Treatment allocation prior to randomization was concealed from recruiting clinicians. Both trial groups received the same glucocorticoid treatment regimen per study protocol. Children were excluded if they did not meet the classification criteria for childhood PAN, if they received alternative diagnoses, if they had chronic infection, if they experienced previous reactions to one of the study medications, or if they had immunodeficiency or malignancy. Genetic screening for ADA-2 was administered as part of the routine evaluation of the patients (i.e., outside the protocol).
Ethics approval. The protocol was approved by the Multicentre Research Ethics Committee in the UK and from relevant ethics committees for each participating center internationally. The study was conducted in accordance with the ethics principles of the Declaration of Helsinki. Patient and public involvement informed the design of the protocol and patient-facing trial documents. All participants provided written informed consent.

| 1675
Treatments. Participants received oral MMF or IV CYC treatment for 18 months, which comprised 3-6 months of induction therapy (1:1 randomization); followed by 12-15 months of oral azathioprine (AZA) maintenance therapy ( Figure 1). Both groups in the trial received tapering glucocorticoids (see below). Unless participants had an allergy, prophylaxis with trimethoprim/sulfamethoxazole was required until week 24. Trial treatment ended after 18 months.
CYC was administered at weeks 0, 2, and 4 and then every 3 weeks until remission was achieved (maximum 10 IV doses and minimum 6 doses) (Supplementary Table 2 CYC could be discontinued after a minimum of 6 doses if disease was in remission. Patients were administered oral AZA (2 mg/kg/ day, maximum 200 mg/day) (12,26) the day following discontinuation of MMF treatment or 10-14 days after the last dose of IV CYC.
All patients received prednisolone starting at 1 mg/kg/day (maximum 80 mg dose), which was decreased to 0.1 mg/kg/day by 6 months and to 0.05-0.075 mg/kg by 9 months (Supplementary Table 3, available on the Arthritis & Rheumatology website at http://onlin elibr ary.wiley.com/doi/10.1002/art.41730/ abstract). Intravenous methylprednisolone could also be administered at trial entry (maximum 30 mg/kg for 3 doses or 3 gm total) at the investigator's discretion.
Principal investigators recorded medications received by the patient on a medication clinical research form during protocol face-to-face follow-up trial visits as specified below. In addition, patients completed a diary listing the medications taken as an outpatient, which allowed careful cross-checking of the accuracy of medications taken on a daily basis.

Assessments.
Assessments were performed at weeks 0, 4, 10, 16, and 24 when the primary end point of remission was evaluated. Thereafter, assessments occurred at weeks 36, 48, 60, and 72. A final follow-up visit also took place on the date of the last patient's last visit, which varied considerably among the patients. Therefore, only results up to and including week 72 are reported here.
Disease activity was determined using the PVAS (25). Briefly, the PVAS ranges from 0 to 63, with higher scores denoting active clinical disease activity within 9 organ systems and a score of 0 indicating the absence of disease activity. Safety events were coded using The Medical Dictionary for Regulatory Activities version 19.
Primary outcome measure. The primary outcome measure was remission within 6 months, which was defined as the absence of disease activity (PVAS 0 [of a maximum 63]) on 2 consecutive visits at least 1 month apart, with adherence to glucocorticoid taper protocol (20,21,25). The primary end point was assessed at 6 months because this reflects the typical time point to assess the effectiveness of remission induction in routine clinical practice (12) and is therefore used in most vasculitis trials (20). Secondary end points assessed over the full 18-month trial were as follows: remission within 6 months regardless of glucocorticoid taper; time to remission; pediatric vasculitis damage index (PVDI) score (27,28); mycophenolic acid 12-hour trough levels; the cross-culturally adapted and validated version of the Childhood Health Assessment Questionnaire (C-HAQ) for disability and the Child Health Questionnaire (CHQ) for quality of life (29); costeffectiveness using the UK NHS costs and quality-adjusted lifeyears (QALYs) measurement based on the Child Health Utility-9D (CHU-9D) (30) and EuroQol-5D-3L (EQ-5D-3L) questionnaires (31); cumulative glucocorticoid dose; growth; disease relapse within 18 months; adverse events; withdrawal from trial due to drug intolerance; and mortality. Statistical analysis. Sample size. A maximum target sample size of 40 was chosen pragmatically, as this was the largest number feasible to recruit among PRINTO sites. A Bayesian approach is not restricted by small sample sizes and allows data to be combined with existing evidence. The larger the recruitment, the greater the contribution of trial results to the totality of evidence, post-MYPAN trial. Bayesian sample-size calculations suggested that this would yield a power of 62% to ascertain noninferiority of the primary end point (32). The MYPAN trial was conducted using a Bayesian design due to the challenge of low participant numbers, given that childhood PAN is extremely rare. Bayesian power was therefore also calculated for smaller sample sizes (8 patients [41% power], 10 patients [52% power], and 12 patients [53% power]).
Data analyses. Per the recommendation in the Consolidated Standards of Reporting Trials statement (33), reported results are from the intent-to-treat (ITT) population. Missing data were not imputed. The date at which the primary outcome was achieved was the first of the 2 consecutive visits in which the PVAS was 0. The primary outcome was examined using a Bayesian analysis.
Bayes theorem was used to combine expert prior opinion with the MYPAN data to obtain posterior distributions for remission rates with CYC treatment (pC), remission rates with MMF treatment (pM), and the log odds ratio of remission with MMF compared to CYC (θ).
Full details of the primary outcome measure analysis methods were previously published (21,32). Briefly, noninferiority of MMF was defined as a Bayesian posterior probability of obtaining remission within 6 months, within 10% (absolute difference) of CYC. Quantities of interest were pC, pM, and θ. Bayesian prior distributions for pC and pM were established during a prior elicitation workshop in September 2013 (before recruitment for the MYPAN trial began), using expert opinion and evidence presented from the MYCYC trial (20,21,32). The posterior distributions for pC, pM, and θ were calculated and summarized by their modes, which reflected the most likely values for these quantities, and by 90% credibility intervals (90% CrIs), which quantified our certainty. We also calculated 2 posterior probabilities, i.e., that the 6-month remission rate among patients taking MMF is noninferior to the 6-month remission rate among patients taking CYC (pM ≥ pC -0.10) and that achievement of remission within 6 months is more likely to occur among patients taking MMF than among those taking CYC (pM > pC). All secondary outcome measures were analyzed descriptively | 1677 using frequentist statistics (i.e., number, median, and interquartile range [IQR]), unless otherwise stated. Results were summarized graphically using Kaplan-Meier curves, patient profile plots, and radar plots.
For each participant, total NHS costs associated with primary care, secondary care, and community care services, and medication use were measured over 18 months. This was based on resource use questionnaires completed by trial participants or their parents or guardians during clinic appointments, and via information from case report forms. Unit costs were obtained from standard NHS sources (https://impro vement.nhs.uk/resou rces/ natio nal-tarif f/). The estimation of preference weights for each health state was generated from patient responses to CHU-9D questionnaires (30). QALYs were then computed by applying the trapezium rule to estimate the area under the curve. Second-year costs and QALYs were discounted at 3.5%. Differences between the MMF and CYC treatment groups in costs and QALYs were estimated by linear regression, with the per-patient cost (or per QALY) as the dependent variable and the treatment group as the only independent predictor. A nonparametric bootstrap analysis using 10,000 replicates was performed to assess the joint uncertainty in mean costs and QALYs. The probability of each treatment being cost-effective was determined at the threshold of £20,000 per QALY, which operates within the NHS (34) and in accordance with the National Institute for Health and Care Excellence guidance (https://www.nice.org.uk/proce ss/pmg9/). Analyses were performed using R version 3.6.1 or SAS version 9.4.  Figure 2. Six patients were randomized to receive MMF, and 5 patients were randomized to receive CYC. All 11 patients received their allocated treatment and were retained for the primary analysis; 1 patient withdrew from follow-up at 26 weeks. Baseline characteristics of the patients are provided in Table 1    Findings for the primary outcome measure. Remission within 6 months of randomization occurred in 4 of 6 patients (67%) in the MMF group and 4 of 5 patients (80%) in the CYC group. The Bayesian posterior distributions for remission rates (modes) were 71% (90% CrI 51, 83) for MMF and 75% (90% CrI 57, 86) for CYC, and the odds ratio of remission within 6 months with MMF compared to CYC was 0.81 (90% CrI 0.40, 1.65) (Figures 3 and 4). The posterior probability that MMF is noninferior to CYC was 0.76, indicating that noninferiority is likely. Also, the posterior probability that the 6-month remission rate is higher in MMF than in CYC was 0.31, indicating that MMF superiority is unlikely (Figure 3). Vasculitis damage, glucocorticoid exposure, and mortality. The median PVDI score was 0 (of a maximum 72) for both groups at trial entry. The median PVDI score at trial end (18 months) was 0 (IQR 0, 1) in the MMF group and 2 (IQR 0, 3) in the CYC group (Supplementary Table 5 11,175]) and was associated with fewer QALYs (0.047 [−0.5749, 0.4798]) compared to CYC. The probability of MMF being cost-effective at a threshold of £20,000 per QALY was 0.32, which was evaluated in the 7 patients from the UK only.

Safety outcome measures.
Total adverse events were similar between the 2 groups. Thirty-eight events (63% mild severity and 37% moderate severity) occurred in 5 of 6 patients treated with MMF, and 31 events (97% mild and 3% moderate) occurred in all 5 patients treated with CYC ( Table 2 and Supplementary  Table 11, available on the Arthritis & Rheumatology website at http://onlin elibr ary.wiley.com/doi/10.1002/art.41730/ abstract). A total of 4 serious adverse events, including 2 infections, occurred in 3 of 6 patients in the MMF group. One patient had abdominal pain (deemed not related to MMF) and a lower respiratory tract infection (possibly related to MMF), which fully resolved with treatment. One patient had colitis (deemed not related to MMF) ongoing at trial end, and 1 patient had herpes zoster (possibly related to MMF), which fully resolved with treatment. No serious adverse events were observed in the CYC group (Supplementary Table 12

DISCUSSION
A major challenge in the study of rare diseases is conducting a clinical trial with sufficient power to inform best clinical practice when the anticipated sample size is small. Historically, this has been an insurmountable barrier in the study of rare pediatric autoimmune diseases, and explains why a clinical trial of childhood PAN had never been undertaken until now (21). We adopted a Bayesian clinical trial design with the objective of quantifying disease remission rates with CYC and MMF treatment, combining a robust elicitation of prior opinion and evidence with our trial data. Six-month remission rates observed in the MYPAN trial were consistent with prior opinions, and since we could only recruit 11 patients, the totality of evidence is heavily influenced by those prior distributions. We calculated a Bayesian posterior probability of 76% for noninferiority of MMF compared to CYC for remission within 6 months. This observation, while not definitive, is still clinically useful, particularly since conducting a confirmatory frequentist trial is impossible (21). Further clinical face validity for the noninferiority of MMF compared to CYC is suggested by the fact that glucocorticoid use could be successfully tapered in all patients and all patients had nearly identical cumulative glucocorticoid exposure. Therefore, our results suggest that MMF might represent a viable alternative to CYC for remission induction in childhood PAN. Moreover, these results will inform prior opinions for any future trials in childhood PAN (e.g., the Biologics in Refractory Vasculitis Study [https://www. isrctn.com/ISRCT N1650 2655]).
The MYPAN data are consistent with data from studies in AAV, most notably the MYCYC study (20) (though not completely independently, as the MYCYC data helped inform the prior opinion used in MYPAN), which included adults and children and showed that MMF was noninferior to CYC for inducing remission. Following remission, all patients in our trial received AZA and glucocorticoid maintenance therapy, with no relapses. This observation contrasts with findings in the MYCYC trial which showed that relapses occurred earlier and more frequently among patients in the MMF group (33%) compared to among those in the CYC group (19%) (20). Thus, the previous suggestion that relapses in childhood PAN are less common than in childhood AAV is supported by our results (1).
Other secondary end points are also potentially clinically relevant, though the results are purely descriptive. Remission was exhibited at a median of 7.1 weeks among patients in the MMF group, compared to a median of 17.6 weeks in the CYC group. PVDI scores were lower in the MMF group, implying less damage, although our trial was not powered to demonstrate statistical significance of this observation. PVDI scores (and in adults, VDI scores) are not weighted; hence, overall low numerical scores can still indicate severe damage in patients. Therefore, future studies are needed to further examine the potential clinical importance of this preliminary observation.
C-HAQ disability scores and pain scores at trial end were comparable among patients in both groups, though scores were numerically lower among patients in the MMF group. While we must be careful not to overinterpret this purely descriptive observation, a possible obvious explanation is that the C-HAQ score reflected a more rapid resolution of disease activity among patients in the MMF group, resulting in faster resolution of disability and pain. Similarly, and in accordance with this suggestion, quality of life improved more rapidly and to a greater extent among patients in the MMF group compared to the CYC group, particularly in regard to the physical summary score. The health economic analysis in the UK suggested that MMF may generate fewer QALYs and may be more expensive than CYC, though with a significant element of uncertainty. No patients died in either trial group. Lastly, remission was achieved in all the patients in the MMF group who completed follow-up. There were no new safety signals for MMF or CYC. Notably, 2 infections were considered to be possibly linked to MMF. Improved short-term safety with MMF was thus not demonstrated. However, long-term safety issues are probably of more importance and are not captured in our trial. The use of MMF along with a standard dose of glucocorticoids offers clear advantages over CYC in terms of fertility preservation in younger patients, and potentially lower malignancy rates later in life, which is of particular concern among pediatric patients (18,20).
Our trial has several notable strengths. To our knowledge, it is the first randomized trial in childhood PAN. Patients were recruited from regional tertiary centers; thus, the trial cohort was fully representative of the spectrum of disease in childhood PAN, as indicated by the extent of organ involvement observed. The study also included the use of standardized tools developed specifically for children with vasculitis to allow accurate classification of childhood PAN (35), and of disease activity and remission (using the PVAS) (24). Our study was also the first to record vasculitis damage prospectively using the PVDI, which to date has been only preliminarily used in retrospective studies (27,28).
The strengths of this trial should be viewed against its limitations, notably, that the clinical trial evidence is based on a small sample size, augmented by a distillation of clinical experience in the form of prior distributions. However, the fact that the posterior distributions we observed are largely consistent with prior expert opinions adds important clinical face validity to the conclusion, which must be based on the final Bayesian posterior distributions and may provide the prior distributions for future cumulative research of childhood PAN. In addition, MYPAN was not blinded, for purely practical reasons. Although glucocorticoid exposure was documented, glucocorticoid toxicity was not systematically captured using the glucocorticoid toxicity index (36). Only 7 of 11 patients were screened for DADA2 as part of their routine evaluation, which might have important implications for determining the efficacy of both MMF and CYC in childhood PAN (10,11). Health economic analyses were based on UK costs, and therefore may not apply uniformly in other countries (e.g., in Turkey, where MMF is more expensive than CYC). Our trial also did not address the possibility that higher doses of MMF might be even more efficacious. Regulatory approval for dose escalation was initially requested in the MYPAN trial, but not granted by competent authority since it was suggested that adverse effects might also increase with higher doses. Finally, generalizability to other ethnic groups was limited as 10 of 11 patients in our trial were White.
In summary, MMF is probably noninferior to CYC for induction of remission in childhood PAN when combined with glucocorticoids. MMF may also be associated with better quality of life compared to CYC.