Eastern Cooperative Group Trial of Interferon Gamma in Metastatic Melanoma: An Innovative Study Design

Received 4/19/95; revised 9/18/95; accepted 9/19/95. 1 Partially supported by the Veterans Administration. To whom requests for reprints should be addressed, at University of Wisconsin Comprehensive Cancer Center, 600 Highland Avenue, K4/666, Madison, WI 53792. Phone: (608) 263-5343; Fax: (608) 263-8613. jects with metastatic melanoma: ECOG Study E4987, submitted for publication), this biological activity does not translate into therapeutic activity in the metastatic disease setting in this trial. I N T R O D U C T I O N (IFNs) are a family of proteins with antiviral, immunomodulatory, and antiproliferative effects. IFN-oL, the first of these proteins to be identified, has been shown to have antitumor activity against a number of human neoplasms, including hairy cell leukemia, Kaposi's sarcoma, chronic myelocytic leukemia, indolent lymphomas, renal cell carcinoma, melanoma, and multiple myeloma (1, 2). Despite this activity, however, the majority of responses are partial, and the overall impact of these molecules on the treatment of human cancers is small. One possible explanation for these disappointing results is that the therapeutic doses and schedules for IFN have not been optimized for inducing immunomodulation. Although in vitro antiproliferative studies indicate that the growth inhibitory properties of IFN-oL are dose dependent, other preclinical studies suggest that high doses of biologicals may not be optimal for immune stimulation (3, 4). IFN-y is a cytokine, produced by T lymphocytes, which has broad potent immunoregulatory effects that distinguish it from IFN-a and -[3. In addition to a unique spectrum of antiproliferative properties, IFN-~/is a potent monocyte activator. It has been reported to augment monocyte tumoricidal activity (4), hydrogen peroxide generation (5), monocyte antibody-dependent cellular cytotoxicity, Fc receptor expression, and both class I and class II MHC (6). However, studies suggest that these effects may not be greatest at the maximally tolerated clinical doses. For example, peak effects on natural killer cell activation, T-cell subset distribution, and enzyme oligo-2'5'-adenylate synthetase were found to occur at doses from 300 to 1000 lxg/m2/ day, intermediate between the lowest and highest dosages tested and well below the maximally tolerated dose identified (7). In another Phase I trial, daily i.m. administrations of 0.25-0.5 mg/m2/day produced significant activation of tumoricidal activity in peripheral monocytes, whereas the daily administration of 1.0 mg/m 2 did not (4). A third Phase I trial evaluated biological effects in patients receiving daily IFN-y by i.m. injection, at doses ranging from 0.0001 to 0.25 mg/m 2 and concluded that the optimal dose for inducing immune modulation was 0.1 mg/m 2 (8). Preclinical studies with recombinant murine IFN-y also suggest that the therapeutic activity depends on the dose, schedule, and route of administration (9). Optimal antitumor activity in spontaneous and experimental B16-BL6 metastases was demonstrated when IFN-y was administered i.v. three times per week at 50,000 units/animal but not at 10,000 or 100,000 units/animal, suggesting a bell-shaped dose-response curve. No therapeutic activity was detected with twice weekly i.v. or daily i.m. administration compared with i.v. administration three times per week. Research. on January 6, 2018. © 1996 American Association for Cancer clincancerres.aacrjournals.org Downloaded from 30 IFN-y Trial in Metastatic Melanoma Table 1 Dose levels of IFN-'y Dose IFN-'y dose No. of No. level (mg/m z) patients evaluable 1 0.01mg/m 2 16 14 2 0.03mg/m 2 15 13 3 0.10mg/m 2 14 11 4 0.30mg/m 2 14 11 5 0.50mg/m 2 12 8 6 0.70mg/m z 11 8 7 0.90mg/m 2 13 12 Based on these preclinical and clinical results, the Eastern Cooperative Oncology Group conducted a Phase II/III doseseeking trial of IFN-y in patients with metastatic melanoma with nonbulky s.c., skin, or nodal disease. Using an innovative trial design, the objectives of this trial were, primarily, to define a therapeutic dose-response curve to IFN-y, secondarily, to delineate the immunomodulatory dose response curve further, and, finally, to correlate the two. We report here the clinical results of this trial. MATERIALS AND METHODS All patients were required to have measurable metastatic malignant melanoma with s.c., soft tissue, or nodal disease. Patients had to have adequate bone marrow (WBC, ->4,000/ mm3; granulocytes, >-2,000/m3; and platelets, ->100,000/mm3), renal (creatinine, ---2.0 mg/dl; protein, -<2+), and hepatic function (bilirubin, <1.5 mg/dl; aspartate aminotransferase -<3.0 times the upper limit of normal) and an European Cooperative Oncology Group performance status of 0 or 1 with an estimated life expectancy of at least 3 months. Patients may have received one prior cytotoxic chemotherapy regimen but no prior immunotherapy. Patients with visceral disease only, central nervous system metastases, history of seizures, cardiac disease, active infection, or second primary cancer or requiring therapy with steroidal or nonsteroidal anti-inflammatory drugs, antihistamines, barbiturates, palliative chemotherapy, radiotherapy, hormone therapy, immunotherapy, or any drugs required for the treatment of active cardiac disease were excluded from the study. All patients gave informed consent before enrolling in the study. For the first year of accrual, eligibility was limited to patients with "nonbulky" disease. Nonbulky disease was defined as no single tumor mass >3 cm in largest diameter and the liver <30% involved with tumor on computed tomography. Twenty-nine patients were entered with the nonbulky disease requirement. This entry criterion was dropped when it proved to be too restrictive for accrual purposes. Patients were randomized to one of seven dose levels of IFN-~/and were stratified by predominant site of disease (s.c., soft tissue, skin, or nodal disease versus other) and prior chemotherapy (none ver sus one regimen). The dose levels of IFN-y are detailed in Table 1. Doses ranged over a two-log range, from 0.01 to 1.0 mg/m 2, the MTD 2 of IFN-y as a 1-h i.v. infusion (10). 2 The abbreviations used are: MTD, maximum tolerated dose; RMSE, root mean squared error; BRR, best response ratio; HLA, human leukocyte antigen; NK, natural killer. IFN-~/was provided by Genentech (South San Francisco, CA). The drug was administered by 1-h i.v. infusion in 100 ml 5% dextrose in water containing 200 mg human serum albumin three times a week, every week, with each treatment separated by at least 47 h. The length of each cycle was 28 days. Toxicity was assessed on days 1 and 2 for all patients during the first 4 weeks of treatment. Patients entered at the two highest dose levels had additional assessments on days 3, 4, and 8. In subsequent courses, all patients were assessed by their physician every 28 days. The hematology survey, differential, platelet, aspartate aminotransferase, and creatinine laboratory tests were performed once a week for 4 weeks, then on days 1 and 15. Chemistry surveys were obtained every 28 days. Toxicities were graded according to the Common Toxicity Criteria. The acute flu-like syndrome, consisting of anorexia, fatigue or malaise, chills or rigors, myalgias, arthralgias, and other flu-like symptoms, which often occurs after the first several IFN treatments, was graded according to Table 2. Treatment was interrupted for all grade 3 toxicities until the parameter returned to grade 2 toxicity or less, and then treatment was resumed with the dose of IFN-y reduced by one dose level (or 50% for patients at dose level one). If the grade 3 toxicity recurred at the reduced dose, the patient was removed from the study. Patients were removed from study for grade 4 toxicity. Patients were treated for a minimum of 8 weeks unless serious toxicity or complications occurred; otherwise, treatment continued until progression of their disease. Patients who progressed during weeks 1-4 were continued on treatment weeks 5-8 unless there was documentation of continued deterioration. Patients were evaluated for response following 8 weeks of therapy and every 4 weeks thereafter. A complete response to therapy was defined as a complete disappearance of all clinically detectable malignant disease for at least 4 weeks. A 50% or greater decrease in tumor size for at least 4 weeks, without an increase in size of any known malignant disease or appearance of new lesions, constituted a partial response. Stable disease was defined as no significant change in measurable disease for at least 4 weeks, and progressive disease was defined as the appearance of new lesions or a 25% or greater increase in the area of tumor size. Statistical Design. The primary purpose of the study was to determine the dose of IFN-y that produces the greatest objective response rate. The essential features of this study design involve randomization into a broad range of dose levels, followed by nonlinear statistical modeling to establish the optimal biological and therapeutic dose, with a two-stage stopping rule to permit early termination of a nonresponsive agent. In the first stage of accrual, 84 patients would be allocated among each of seven dose groups ranging from 0.01 mg/m 2 to the MTD, using permuted blocks of size 7. The dose-response curve would be estimated from the data using methods described later in this section. If the estimated dose-response curve was below 20% for all doses up to the MTD, the study would be terminated. Otherwise, the study would proceed to accrue an additional 84 patients, who would be allocated randomly among four doses equally spaced in the range of doses with at least a 20% estimated response rate. The data from both stages of the study would be used to estimate the final dose-response curve and to determine the dose producing the peak response. Research. on January 6, 2018. © 1996 American Association for Cancer clincancerres.aacrjournals.org Downloaded from Clinical Cancer Research 31 Table 2 Signs and symptoms associated with acute, flu-like syndrome associated with IFN Mild Moderate Severe Life-Threatening Grade 1 2 3 4 Lack of appetite after Miss one meal Miss two meals Continue to miss two or more treatment meals on the day of therapy despite repeated IFN treatments (2 wk) In bed for >6 h on the day of therapy despite repeated IFN treatments (2 wk) Acute fatigue and In bed for <6 h In bed for >6 h malaise after after treatment after treatment treatment Chills Rigors that do Rigors requiring not require parental parental medications medications for relief for relief Myalgias/arthralgias Subjectively Subjectively mild moderate Rigors unrelieved by parental medications despite repeated IFN treatments (2 wk) Subjectively severe or with significant disability and unrelieved with medications In the first stage of the study, it was assumed that little or nothing was known about the dose-response relationship. To have the best chance of finding a range in which the therapy was effective, the decision was made to use equally spaced doses between zero and the MTD. The best metric for the dose level was unknown but was assumed to be either log or linear. The choice of dose levels, 0.01, 0.03, 0.10, 0.30, 0.50, 0.70, and 0.90 mg/m 2, is a compromise between equal spacing on the log dose scale, 0.01, 0.03, 0.10, 0.33, and 1.0 mg/m 2, and equal spacing on the linear dose scale, 0.10, 0.30, 0.50, 0.70, and 0.90 mg/m 2. The use of seven compared with five dose levels provided a hedge against improper scaling of the dose axis. If the first stage of the study was successful in detecting a therapeutic range, this range would determine the dose levels for the second stage of the study. Specifically, let A be the minimum dose having at least a 20% response rate and B be the maximum dose, and let C = (B A) /3 . The four dose levels in the second stage would be spaced equally across the therapeutic range at A, A + C, A + 2C, and B. At the completion of follow-up at each stage of the study, the dose-response curve would be estimated from the data. Usually in a dose-response or dose-ranging study, the doseresponse function is often assumed to be nondecreasing over the therapeutic range and is modeled using a logistic model, which relates the logit or log odds of the response probability to a linear function of the dose, i.e., logit [p(d) = logLo(d)/(1-p(d)] = a + [3d. However, previous studies have suggested that the effect of IFN-~/ may peak at a dose below the MTD. It was proposed that the dose-response curve be estimated using a four-parameter logistic model: p(d) = pl(d) pz(d), where logit Pi(d) = ai + f3id (11). Here, ai and [3 i are parameters, and d is the dose level of IFN-~/expressed as a percentage of the MTD. If the data suggested that this model was inappropriate (e.g., a bimodal response curve was observed), then other techniques, which do not assume a parametric model, would be used (12). The final estimate of the dose-response curve would incorporate response data from both stages of the study. Simulation studies were used to assess the properties of our statistical design for five different hypothetical underlying doseresponse curves, each representing a potential true situation against which we would want our design to perform well (Fig. 1). These dose-response curves are described by the four-parameter logistic model, with parameters selected to fix the response levels at doses 5, 25, 50, and 75% of the MTD to prese|ected values. The first curve is unimodal, with probability of response of 0.05, 0.15, 0.35, and 0.15 at these dose levels. The second curve is flat, with constant response probability of 0.15. The third dose-response curve is gently rising, with probability of response of 0.05, 0.15, 0.30, and 0.35 at dose levels 5, 25, 50, and 75%, respectively, of the MTD. The fourth doseresponse curve rises more steeply and then reaches a plateau, with response probabilities of 0.05, 0.30, 0.35, and 0.35 at these dose levels. The fifth curve reaches a peak response probability of 0.35 at 5% of the MTD quickly and then falls to zero at dose levels 25, 50, and 75% of the MTD. Monte Carlo simulations were used to assess the performance of the design described above and to compare it with that of other candidate designs. In the simulation, it was assumed that roughly 10% of the patients would be discovered to be ineligible due to pathology review after enrollment, leaving 75 eligible for evaluation of responses after the first stage and 150 after the second stage. Each replicate of the simulation corresponds to conducting the clinical trial once from start to finish, using random-number generators to allocate patients to dose levels and to create response data with a probability of response given by one of the four parameter logistic models described above. In all, the conduct of the trial was replicated 100 times to ascertain the probability of events such as early stopping for each of the proposed designs. The results of the simulation for each of the proposed designs were summarized by averaging the mean squared prediction error, the best response ratio, and the probability of early stopping across the (n = 100) replicate trials according to the formula given below. The mean squared prediction error for the ith replicate was obtained as follows: the squared differences in response rate between the true dose-response curve, p(d), and estimated dose-response curves, pi(d), were averaged over 100 dose levels, d, ranging from 1 to 100% of the MTD. The best Research. on January 6, 2018. © 1996 American Association for Cancer clincancerres.aacrjournals.org Downloaded from 32 IFN-~/Trial in Metastatic Melanoma