STATISTICAL METHODS IN CLINICAL TRIALS
AND GENETIC EPIDEMIOLOGY

Clinical trials have become one of the most widely accepted tools in the search for more effective cancer therapies. We are developing methodologies to improve the design of clinical trials and to provide more efficient analytical tools in phase I/II clinical trials. As a consequence, we expect that more patients will be treated with therapeutic doses of a promising new agent and fewer patients will be overdosed and suffer from the agent's toxic effects.

The goal of genetic epidemiology in cancer research is to reach an understanding of the interactions between genetic and environmental components in cancer etiology. The intricate nature of these interactions, in addition to the complex structure of pedigree data, can only be studied with complex statistical methods. We are developing our methodology in the areas of segregation analysis, gene mapping, and risk prediction. Segregation analysis reveals the genetic mechanisms responsible for familial clustering of individuals with a specific cancer. Gene mapping assigns genes to a particular location in a chromosome. We are using these methodologies to improve the prediction of the risk that an individual will develop a particular cancer.

The three main goals in phase I clinical trials are to establish a maximum tolerated dose (safe dose) for a new drug, to determine the toxicity of the drug, and to look for evidence of anti-tumor activity of the new agent. We developed EWOC (Escalation With Overdose Control), a new approach for dose escalation in phase I clinical trials. EWOC is a method for selecting doses of a cytotoxic agent while controlling the probability of exceeding the maximum tolerated dose (MTD). Cytotoxic agents, as the name implies, can induce unacceptably toxic reactions at dose levels required for optimal anti-tumor activity. Identifying the dose that provides an optimal balance between the therapeutic and harmful aspects of a cytotoxic agent is the purpose of cancer phase I trials. More specifically, since the phase I trial represents the first application of a new agent or drug combination to humans, the emphasis is placed on controlling the toxicity rather than the efficacy of the new agent. Consequently, the goal of a cancer phase I trial is to determine the highest dose (called MTD or Target Dose) associated with an acceptable level of toxicity. EWOC is designed to approach the Target Dose as safely and as quickly as possible.

As a result of ongoing collaboration with clinical investigators, EWOC has been used to design nine phase I studies approved by the Research Review Committee and the Institute Review Board of the Fox Chase Cancer Center. The first completed study designed with EWOC was "Phase I study of 5FU, leucovorin and topotecan in patients with advanced cancer" (IRB95073; L.M. Weiner, P.I.). In this trial, patients received all three drugs daily for five days. The level of leucovorin was fixed and the escalation of 5-fluorouracil (5FU) and topotecan was conducted in two stages. In the first stage topotecan remained constant while 5FU dose was escalated to a MTD. In the second stage, 5FU was fixed at the MTD and topotecan dose escalated to MTD.

During this trial, patients were treated one at a time and the probability of dose limiting toxicity was set to 1/3. The probability of exceeding the MTD was initially set to 0.25, reflecting the high level of uncertainty about the MTD at the start of the trial. Subsequent to patient 14, this probability was gradually increased to 0.5 by increments of 0.05 reflecting the decline in uncertainty as the trial progressed. Figure 1 shows the marginal posterior distribution of the MTD after the trial was completed. The mode of this distribution is 1.44 mg/m² and is the most likely value of the MTD based on the information available at this time. If the trial were to continue, EWOC would recommend that the next patient be given 1.46 mg/m². This value is the median of the posterior distribution and is approximately the same as the mode. The 95% Bayesian confidence interval for the target dose of topotecan is [1.14 mg/m², 1.75 mg/m²].

ESCALATION WITH OVERDOSE CONTROL -- APPLICATION SOFTWARE. ROGATKO, BABB

Two computer programs have been written to implement EWOC and are available in both Unix and Windows 95/NT platforms. The Unix version, written in FORTRAN 77, was developed and tested on a DEC 3000/300 (alpha) running on OSF 1 (v3.2). The executable can be obtained by email (a_rogatko@fccc.edu or babb@canape.fccc.edu). The Windows 95/NT version, written in Visual FORTRAN 90, is a user-friendly, dialog-based, stand-alone application. The self-extracting file can be downloaded from the web site http://www.fccc.edu/users/rogatko/. The dialog window from the Windows version is illustrated in Figure 2. The current implementations of EWOC allow the specification of the following input parameters: the probability of dose limiting toxicity, the probability of exceeding the target dose, the initial and maximum doses, and minimum dose increment. They generate as output the EWOC recommended dose, Bayesian confidence interval, a plot of the marginal posterior distribution of the MTD (e.g., Figure 1), and a tree of recommended doses for up to the next four patients.

STATISTICAL ASSESSMENT OF MULTI-VALUED DIAGNOSTIC TESTS. ROGATKO, in collaboration with REBBECK,^a VIANA^b

The goal of diagnostic tests is to confirm or exclude the presence of a certain disease. Medical diagnosis and test interpretation involves considerable uncertainty, and probability serves as a natural language to describe the association between diagnostic test results and presence of disease. Diagnostic tests can be applied either as confirmatory tools in a clinical setting or as a mass screening procedure. However, they should be used only when they alter the management of a given patient. We developed a language to characterize diagnostic tests with multivalued outcomes. The proposed language allows one to describe a diagnostic test, and to compare the utility of several diagnostic tests. We showed that a complete description of a diagnostic test includes not only its sensitivity and specificity, but also the disease prevalence. Thus, the diagnostic usefulness of the test depends on the likelihood that the test result may alter the management of a patient and on the prevalence of the disease (the marginal probability that the patient actually has the disease).

We applied our methods to assess the utility of genotypes at cytochrome P450 1A1 (CYP1A1) and glutathioneStransferase mu (GST) in susceptibility to develop lung cancer. The results of these analyses indicate that CYP1A1 is a useful measure of lung cancer susceptibility only when the lung cancer prevalence in the population under study is between 28% and 53%. Thus, it is unlikely that CYP1A1 will be a good marker of lung cancer susceptibility by itself. When both CYP1A1 and GST were considered simultaneously, the joint genotypic information was a good measure of lung cancer susceptibility when the prevalence of disease in the population under study was between 21% and 60%. These results indicate that the twolocus (CYP1A1 plus GST) genotype provides a better measure of lung cancer susceptibility than the singlelocus (CYP1A1) genotype. CYP1A1 and GST may only be useful predictors of lung cancer susceptibility in highprevalence populations, and not in populations with cancer prevalence less than 20%. These methods should be widely applicable to the objective assessment of diagnostic or predictive test utility for a wide range of diseases. We plan to apply these methods to assess the predictive utility of risk factors in a case-control study.

Table 1 shows a preliminary application of our method to assess the predictive utility of the combined use of number of biopsies and number of relatives with breast cancer. Data is from the Breast Cancer Detection Demonstration Project. A preliminary analysis indicates that the combined use of number of biopsies and number of relatives with breast cancer is a useful measure of breast cancer susceptibility only when the breast cancer prevalence in the population under study is between 19.6% and 55.1%.

TABLE 1. Number of biopsies, number of relatives with breast cancer and frequencies of cases with breast cancer (BC), controls, sensitivities, specificities, and likelihood ratios from BCDDP

PUBLICATIONS

BABB, J., ROGATKO, A., ZACKS, S. Bayesian sequential and fixed sample testing of multihypotheses. In Asymptotic Methods in Probability and Statistics, edited by B. Szyszkowicz. Elsevier Science, BV, Amsterdam, pp. 801-809, 1998.

DEXTER, D.W., REDDY, R.K., GELES, K.G., BANSAL, S., MYINT, M.A., ROGATKO, A., LEIGHTON, J.C., GOLDSTEIN, L.J. Quantitative reverse transcriptase-polymerase chain reaction measured expression of MDR1 and MRP in primary breast carcinoma. Clin. Cancer Res. 4:1533-1542, 1998.

ROGATKO, A., BABB, J. Escalation with overdose control. User's guide. Version 1.beta. The URL address is http://www.fccc.edu/users/rogatko/user_guide.html, 1998.

ROGATKO, A., BABB, J., JORDAN, H., ZACKS, S. Constructing meiotic maps with known error probability. Genet. Epidemiol. 16:274-289, 1999.

ROGATKO, A., BABB, J., REBBECK, T. Genetic counseling. In Encyclopedia of Biostatistics, edited by P. Armitage, T. Colton. John Wiley and Sons, Chichester, UK, Vol. 2, pp. 1672-1674, 1998.

VIANA, M., ROGATKO, A., REBBECK, T. Statistical assessment of multi-valued diagnostic tests. Can. J. Stat. (in press).

BABB, J., ROGATKO, A., ZACKS, S. Cancer phase I clinical trials: efficient dose escalation with overdose control. Stat. Med. 17:1103-1120, 1998.

KHATER, C., LAUB, P., GALLO, J.M., ROGATKO, A., O'DWYER, P.J. Phase II clinical trials in oncology. In Anticancer Drug Development Guide: Preclinical Screening, Clinical Trials, and Approval (Cancer Drug Discovery and Development), edited by B. Teicher. Humana Press, Totowa, NJ, Chapter 13, pp. 249-270, 1997.

ZACKS, S., ROGATKO, A., BABB, J. Optimal Bayesian-feasible dose escalation for cancer phase I trials. Statistics & Probability Letters 38:215-220, 1998.

^§   Fox Chase researcher

^a   T. Rebbeck: Present address--University of Pennsylvania, Philadelphia, PA 19104

^b   M. Viana: University of Illinois, Chicago, IL 60612

Illustrations or unpublished data in these reports should not be used without permission of the author.

Fox Chase Cancer Center

Scientific Report 1998