Department of Biomathematics

Section of Biostatistics

Staff and Activities

Staff

Peter F. Thall, Ph.D. (1975) Florida State University
Associate Professor of Biomathematics
Areas: Clinical trial design; survival analysis; statistical modeling; nonlinear regression; longitudinal and repeated measures data; computer-intensive methods in statistics; applied Bayesian methods

Su-Chun Cheng, Sc.D. (1995) Harvard School of Public Health
Assistant Professor of Biomathematics
Areas: statistical methods for survival data; design and analysis of clinical trials

Kenneth R. Hess, Ph.D. (1992) The University of Texas School of Public Health
Assistant Professor of Biomathematics
Areas: Survival data analysis; tree-based statistical models; statistical consulting; statistical education for medical researchers; prognostic factor studies

J. Jack Lee, Ph.D. (1989) University of California at Los Angeles
Assistant Professor of Biomathematics
Areas: Biostatistics; survival data analysis; design and analysis of clinical trials; longitudinal data analysis; statistical computing and graphics

J. Lynn Palmer, Ph.D. (1988) The University of Texas School of Public Health
Assistant Professor of Biomathematics
Areas: Biostatistics; design and analysis of clinical trials; longitudinal data analysis; Bayesian methods of design and analysis

Yu Shen, Ph.D. (1994) The University of Washington School of Public Health
Assistant Professor of Biomathematics
Areas: Design and analysis of clinical trials; survival analysis; estimation of natural history of diseases

Terry L. Smith, M.S. (1986) The University of Texas School of Public Health
Assistant Professor of Biomathematics
Areas: Design and analysis of clinical trials; evaluation of prognostic factors and their use in treatment comparisons

Mary Ann Gregurich, Ph.D. (1993) The University of Texas School of Public Health
Research Associate
Areas: Biostatistics; statistical consulting; survey analysis; statistical computer program packages; Bayesian methodology

Hai-An Hsu, Ph.D. (1995) The University of Texas School of Public Health
Programmer Analyst II
Areas: Clinical trials and survival analysis

Chris Schacherer, Ph.D. (1996) Rice University
Programmer Analyst II
Areas: Database management; user interface design; psychometric analysis

Qiong Dong, M.S. (1993) Iowa State University
Programmer Analyst II
Areas: Clinical trials and survival data analysis

Hsi-Guang Sung, M.S. (1996) University of Washington
Programmer Analyst II
Areas: Statistical computing

Chi-Hong Tseng, M.S. (1994) Iowa State University
Programmer Analyst II
Areas: Data analysis and statistical genetics

Jesus Rodriguez, B.S. (1993) The University of Houston
Computer Programmer II
Areas: Database development and programming; data analysis; statistical programming

Activities

To collaborate with staff members of other Departments on the design, analysis, interpretation, and reporting of clinical, observational, and laboratory studies in cancer research.

To perform research in biostatistics on methodological problems arising in consulting.

To provide biostatistical consultation and resources, including educational lectures to clinical research scientists engaged in the planning, conduct, analysis, and interpretation of clinical research studies.

Investigator-Initiated Research

1. Skin Cancer Control. Dr. T. E. Moon.
   Skin cancers are the most common neoplasms in the United States, commonly classified as melanoma, basal cell carcinoma (BCC) and squamous cell carcinoma (SCC). Over 700,000 new skin cancers are projected to be diagnosed in 1995, equalling approximately 58% of the number of all other types of cancers combined. Skin cancer annual incidence has increased for at least the past two to three decades, with melanoma incidence showing the second highest increase of all types of cancer. Although the survival rate for melanoma has increased, skin cancer represents an increasingly important health and economic problem. Health implications resulting from the increasing incidence of skin cancer in the U.S. are further intensified by the continued population migration to sunnier, more southern regions and by the reduced stratospheric ozone. Thus, there is increasing need to better understand the etiology and to control skin cancer. Surveillance of skin cancer incidence for defined populations is essential to accurately calculate the magnitude of the cancer burden and also provides an important reference to evaluate cancer control interventions. Improved statistics and adaptation of biostatistical methods are critical to the successful use of surveillance data. A publication is in preparation on skin cancer surveillance.
   The retinoid skin cancer prevention clinical trials have been completed. These trials tested the hypothesis that vitamin A and its synthetic derivatives (retinoids) reduce the risk of skin cancers (SCC or BCC) in high risk individuals. Over 3,500 individuals participated in these two complementary vitamin A clinical trials. By use of more contemporary biostatistical methods (CART and other regression models), the value of vitamin A supplementation will be determined and the specific subgroup of individuals that benefitted from retinoids identified. These results will not only benefit the control of skin cancer but also common cancer types (squamous cell type) that arise in other parts of the body. One publication has been submitted and three other publications are in preparation.
   
2. Evaluation Of Cancer Biomarkers. Dr. T.E. Moon, Dr. R. Lotan, Dr. T.C. Shu, and Dr. H.N. Ananthaswamy.
   The evaluation of biologic (bio) markers of carcinogenesis is continuing to determine how to identify individuals at high risk of cancer, biologic explanations of their high risk, and how to cost effectively select interventions to control cancer. The success of molecular biology research has resulted in the description of an increasing number of putative biomarkers. Combined with established clinical, medical and environmental risk factors, the clinical value of such biomarkers requires adapting and developing biostatistical methods. Also, the influence of measurement error and heterogeneity of study subjects is included in the evaluation.
   A population based case control study of all three major types of skin cancer has been completed and provides data on multiple risk factors and putative biomarkers, including dietary intake, blood micro and macro nutrients, tissue fatty acid profiles plus exposure and susceptibility measures. Data analyses indicate a substantial variation in skin cancer risks by ultraviolet light exposure and skin pigmentation. Unfortunately, these two risk factors are of modest value in predicting who will develop skin cancer. Various statistical techniques are being evaluated to determine cost effective ways of integrating biomarkers into the design of new clinical studies. A publication is in preparation on the evaluation of biomarkers and other risk factors for the three different types of skin cancer.
   
3. Egyptian Cancer Research and Education Programs. Dr. T. Moon, Dr. A. Soliman, Dr. M. Shebl, and Dr. K. El-Beltagy.
   Our activities in Egypt focus on epidemiologic research for identifying the risk factors and biomarkers of skin and liver cancer. Other projects include developing a teaching curriculum for cancer control and prevention and improving faculty skills in epidemiology and biostatistics through a series of teaching workshops, coupled with training Egyptian junior faculty at The University of Texas M.D. Anderson Cancer Center.
   
4. Time Dependence of Prognostic Effects. Dr. K. Hess.
   In a typical analysis of time-to-event data, covariates are assumed to have a constant effect over the entire follow-up period. However, some covariates violate this so-called proportional hazards assumption by having effects which vary over time. For example, some covariates are good predictors of short-term survival but poor predictors of long-term survival. Methods were reviewed and some new methods developed for identifying the presence and nature of the time dependence and for modelling the time dependence of the covariate effects. Alternative methods for demonstrating the presence of time-dependent covariate effects were also pursued.
   A powerful and straightforward method is to partition the follow-up time into mutually exclusive intervals and then to compute survival curves and relative risks within each interval. Plotting the relative risks against the interval midpoints can be helpful in assessing time-dependence. Instead of viewing the relative risk as a step function of time, we can model it as a continuous function of time. A convenient method is to use an extended Cox regression model which incorporates a time-by-covariate interaction term modelled as a product between the covariate and a function of time. Smoothed plots of the Grambsch-Therneau-Schoenfeld residuals are indispensable in deciding the correct function of time to use. It is possible to fit simple transformations of time, polynomial functions of time, and even piecewise polynomials of spline functions of time (Recent Publications, #34 and #35).
   
5. Metastatic Patterns. Dr. J. Abbruzzese and Dr. K. Hess.
   The goals of this study are to describe and analyze the patterns of metastases for a cohort of patients with single primary tumors and to determine the feasibility of predicting the location of the primary based on the pattern of metastases and certain clinical and demographic factors. Patients admitted to UTMDACC during 1992 with one or more metastases and a single primary were identified through the Patient Studies' EPIMAST database. Data on age, race, sex, histology and metastatic pattern were acquired for 2,903 adult patients with one of the 13 primaries and one of the four major histologic categories of interest. Preliminary analyses using linear discriminant analysis and classification tree analysis were performed on (1) women with colon, lung, breast and ovary primaries and (2) men with buccal, colon, lung and prostate primaries. The primary location was correctly predicted in 85% to 90% of patients. Ultimately, the results will be compared to data from the Unknown Primary Clinic to assess our ability to predict the location of primaries in patients in whom the primary could not be found clinically.
   
6. A Better Confidence Interval for Kappa on Measuring Agreement Between Two Raters with Binary Outcomes (Recent Publications, #43). Dr. J. Lee and Ms. Z. N. Tu.
   Although the kappa statistic is widely used in measuring interrater agreement, it is known that the standard confidence interval estimation behaves poorly in small samples and for nonzero kappas. Efforts have been made to improve the estimation through transformation and Edgeworth expansion. However, the results remain unsatisfactory when kappa is far from 0, even with the sample size as large as 100. In this research, we reparameterize the kappa statistic to reveal its relationship with the marginal probability of agreement. The reparameterization not only gives a more meaningful interpretation of kappa but also clearly demonstrates that the range of kappa depends on the marginal probabilities. Various 2-D and 3-D plots are shown to illustrate the relationship among these parameters. The immediate application is to construct a new confidence interval based on the profile variance and reparameterization. Extensive simulation studies show that the new confidence interval performs very well in almost all parameter settings even when other methods fail.
   
7. Likelihood-Weighted Confidence Intervals for the Difference of Two Binomial Probabilities. Dr. J. Lee, Dr. B.W. Brown, and Mr. D.M. Serachitopol.
   Two new methods for computing a confidence interval for the difference of two binomial proportions are proposed. Both methods are based on a weighted likelihood approach which maps the tail probabilities in the two-dimensional space into a function of the one-dimensional difference so that the weighted tail probability is /2 at each end. The probability computation is based on the exact distribution rather than the large sample approximation. This procedure may be regarded as a natural extension of the Clopper and Pearson (Biometrika, 26:404-413, 1934) interval to the two-sample case. An extensive computation study was carried out to evaluate the coverage probability and width of the likelihood weighted intervals, and compared them to several other methods reported in the literature. The likelihood weighted intervals compare very favorably with the standard asymptotic interval and with intervals proposed by Hauck and Anderson (The American Statistician, 40:318-322, 1986); Cox and Snell (Analysis of Binary Data, 2nd ed., Chapman and Hall: New York, 1989); Santer and Snell (Journal of the American Statistician, 75:386-394, 1980); Santer and Yamagami (Communication in Statistics, 18:1325-1341, 1993); and Peskun (Journal of the American Statistical Association, 88:656-661, 1994). In particular, the likelihood weighted intervals provide the best balance between accurate coverage probability and short interval width.
   
8. A Versatile One-Dimensional Distribution Plot: The BLiP Plot. Dr. J. Lee and Ms. Z. N. Tu.
   A versatile graphical tool, the BLiP plot, was developed for displaying one-dimensional data. The basic building blocks are boxes, lines, and points. Similar to many standard one-dimensional distribution plots, the BLiP plot is capable of displaying individual data value in points or lines and grouped information in lines or boxes. In addition, the BLiP plot includes many useful new features such as variable-width plots. The central idea of the BLiP plot is to provide basic graphical elements in a friendly and flexible environment so that users can construct anything from a simple, standard plots to a complex, customized plot to best present their data. The program is available in the S functional form and may be utilized on many computer platforms.
   
9. Statistical Analysis of the Reliability and Classification Error of the Mutagen Sensitivity Assay. Dr. J. Lee, Dr. Z. Trizna, Dr. T. C. Hsu, and Dr. W. K. Hong.
   Statistical analysis was applied to evaluate the reliability and classification errors of the in vitro bleomycin sensitivity assay(BSA). It has been demonstrated that BSA can be used for determining an individual's genetic susceptibility to cancer. The standard practice is to read 50 metaphases in each sample and compute the mean chromatid breaks per cell (b/c). Since scoring breaks is a time-consuming task, various fixed number and variable number designs were evaluated to establish efficient and reliable reading strategies. A collection of 160 cases with 100 consecutive readings for each case was obtained. The b/c values were between 0.14 and 1.30 (mean: 0.61). Standard errors(SE) based on scoring 50 and 100 metaphases were 0.15 and 0.11, respectively. After evaluating the first 50 metaphases, the gain in reducing SE is less than 1% with each additional reading. Comparing to 100 readings, the misclassification error for scoring 50 metaphases is less than 4% but the maximum bias can be as high as 0.35. Variable number designs reduce the bias of the fixed number design by reading additional metaphases until the estimated SE is within a prespecified bound. The maximum bias is reduced to 0.13 when SE<0.15. Reading 50 metaphases is sufficiently accurate for bleomycin-resistant(b/c<1.0) subjects, while the variable number design should be applied to increase the reliability for bleomycin-sensitive (b/c 1.0) subjects. Efficient and reliable scoring methods can be applied in large scale cancer epidemiology studies for individual risk assessment.
   
10. Nonparametric Analysis of Longitudinal Data. Dr. J. G. Staniswalis and Dr. J. Lee.
    Methods were developed for analyzing data involving repeated measures of the outcome variable in multiple groups over irregular time intervals. It is demonstrated that the functional form of the response curve can be recovered through nonparametric and semiparametric modeling. The group effect, time effect, and group x time interaction can be estimated and tested. Practical considerations such as bandwidth selection in nonparametric smoothing, model robustness to outliers, estimation accuracy in terms of sample size and signal-noise ratio were also studied. The strength and limitation of this approach will be compared to other methods reported in the literature, such as the Laird-Ware model and GEE methods.
    
11. Repeated Confidence Intervals and Bayesian Implementation of Equivalence Trials. Dr. J. Lee and Mr. J.A. Dubin.
    "Proving the null hypothesis" approach in equivalence trials (Blackwelder, Controlled Clinical Trials, 3:345-353, 1982) is not suitable for interim analysis in the group sequential setting because trials cannot be stopped early when the experimental treatment is much worse than the standard. More appropriate methods were proposed by constructing the repeated confidence intervals (Durrleman and Simon, Biometrics, 46:329-336, 1990; Jennison and Turnbull, Biometrics, 49:31-43, 1993). An alternative is applying the Bayesian method for interim monitoring of equivalence trials by adding the "handicap" proportion (Grossman, et al., Statistics in Medicine, 13:1815-1826, 1994). The performance of various repeated confidence methods and the Bayesian methods were compared in simulation studies. It was found that while all methods preserve the overall significance level, the Bayesian approach is more efficient. As the number of tests increases, the averaged sample size decreases for all methods. However, the power decreases for repeated confidence intervals but increases for the Bayesian method. The Bayesian method is also robust among several choices of non-informative priors. Confidence interval approaches and the Bayesian method offer more specific conclusions (equivalent, non-equivalent, or inconclusive) compared to the standard hypothesis testing method. They can also be easily adapted to the group sequential designs to reduce the averaged sample size.
    
12. Bayesian Optimal Design in Population Models of Hematologic Data. Dr. J.L. Palmer, Dr. M. Korbling, Dr. R.E. Champlin, and Dr. P. Müller.
    A population model was designed to investigate an optimum apheresis schedule for breast cancer patients. Blood stem cells are collected by apheresis prior to the patient's undergoing high-dose chemoradiotherapy and these cells are returned after treatment to enable reconstitution of white blood cells. Maximizing the number of blood stem cells collected in as few aphereses as possible is desirable. We use a longitudinal data model with random effects to describe profiles of individual patients. A hierarchical prior model introduces common mean profiles for patients undergoing different treatments. The optimal apheresis schedule for a new patient is found by minimizing an expected loss over the posterior predictive distribution of the patient's predicted CD34 profile. Estimation of the model and solution of the optimal design problem are implemented by a simulation approach, which allows us to accommodate arbitrary shapes for the profiles and realistic loss functions which include relative penalties for the number of scheduled stem cell collections and for collecting fewer than a specified target quantity of total collected stem cells.
    
13. Bayesian Estimates of Autocorrelation in a Regression Model. Dr. J.L. Palmer, Dr. L.D. Broemeling, and Dr. L.I. Pettit.
    Two Bayesian estimates of autocorrelation were derived assuming the errors of a multiple linear regression model follow a first order autoregressive process. The first estimate is based on the conditional posterior distribution of the autocorrelation parameter, , given the regression coefficients. A test of this estimate can be made by conditioning on the usual least-squares estimate of the regression parameters and constructing a region of highest posterior density (HPD). The resulting t-test is convenient and easily implemented using standard software packages. The second estimate is based on marginal posterior distributions of regression and autocorrelation parameters determined by a Gibbs sampling approach. For both estimates, posterior inferences for estimation and testing have been made, with emphasis on testing for no autocorrelation in the errors. Simulation results show effects of sample size, values of the autocorrelation parameter, and other sampling properties of the estimates of autocorrelation (Recent Publications, #60).
    Difficulty may be experienced when using the Gibbs sampler to estimate the autocorrelation coefficient in a linear regression model when moderate-to-high positive autocorrelated errors are present. Although no problems in estimation were found, when the true autocorrelation was near -1 estimates became unstable and sometimes failed to converge when the true autocorrelation was greater than 0.5. In order to determine the cause of this problem, we simulated data with a positive autocorrelation of 0.9 and, at each iteration of the Gibbs sampling process, recorded and later evaluated estimates of , ß, and ß₁. Results suggest that these estimation difficulties are resolved when a vague but proper prior is used rather than a non-informative, improper prior (Recent Publications, #61).
    
14. Strategies for Planning and Conducting Phase I Clinical Trials of Chemotherapeutic Agents. Mrs. T. Smith, Dr. J.J. Lee, Dr. M. Raber, and Mr. D. Serachitopol.
    The objective of most Phase I trials of chemotherapeutic agents is to establish a maximum tolerated dose of the agent which can then be tested for anti-tumor efficacy in subsequent trials. The usual approach is to administer increasing dose levels to small cohorts of patients until some acceptable level of toxicity is exceeded. Recent proposals in the statistical literature for more quantitative methods of dose determination include a "continual reassessment ethod" in which toxicity outcomes associated with all previous patient doses are used to update a dose-response model and decide the dose for an incoming patient. In a survey of Phase I trials conducted recently at MDACC, it was learned that all were conducted by conventional methods. Trials fell into distinct categories, classified, for example, by whether or not a trial represented the first administration of an agent in humans. Information from this survey will be used to propose practical alternatives to the continual reassessment and other methods, and computer programs already developed to implement these methods will be appropriately modified. Discussions are ongoing with clinicians regarding suitable trials in which some of these methods can be implemented.
    
15. Uses of Prognostic Factors in the Planning and Analysis of Clinical Studies. Mrs. T. Smith.
    Since heterogeneity among patients with respect to prognostic factors often leads to more variability in outcomes than is dependent upon the particular therapy administered, it is of great importance to account properly for the prognostic features of the patients. Knowledge of prognostic factors is essential for both the design and analysis of clinical trials, and also in evaluating clinical importance of the many new biological parameters which can now be measured in laboratory investigations. Application of standard methods such as proportional hazards regression analysis in the case of survival-type endpoints often leads to impressive relative risk estimates and highly significant P-values for certain factors. Additional analyses which yield estimates of the amount of variation in an outcome variable explained by those factors may provide the clinical investigator a more realistic understanding of the disease being studied. Various aspects of these problems are under investigation in clinical studies, e.g. in leukemia, prostate cancer, breast cancer, lymphoma, head and neck cancer.
    
16. Evaluation of the Adequacy of Follow-up Time Available in Investigations Involving a Survival-Type Endpoint. Mrs. T. Smith and Dr. M. Schemper.
    Much clinical research is based on the comparison of survival- type endpoints to draw conclusions regarding treatments, prognostic factors, or disease course. In designing randomized trials, a minimum amount of follow-up duration must be specified and adhered to in order to achieve the required power for the hypothesis being tested. Studies of prognostic factors are usually not prospectively designed, and may be based on a cohort of patients drawn from various sources and with no specified follow-up requirements. In such studies, there may be some unmeasured influence determining which patients are followed more completely. Various methods are used to describe the amount of follow-up when studies comparing survival are reported in the clinical literature, and some reports do not address the issue of follow-up at all. The value of the various methods for determining duration of follow-up when not all cases have been followed to completion (i.e. censored) is being investigated, with the intent to make recommendations for more meaningful summaries of available follow-up.
    
17. A Bayesian Strategy for Screening Cancer Treatments Prior to Phase II Clinical Evaluation (Recent Publications, #79). Dr. P. Thall and Dr. E. Estey.
    We address the problem of selecting a treatment for phase II evaluation when several candidate treatments emerging from phase I testing are available. A pre-phase II Bayesian selection design which randomizes patients among treatments is proposed. The patient group in the trials has a prognosis more favorable than that of phase I patients but less favorable than the target group of the subsequent phase II trial. The patient response rate distribution in each treatment arm is continually updated during the trial for comparison with early termination cutoffs, and the best final treatment must satisfy a minimal posterior efficacy criterion. The primary aim is to replace the usual informal treatment selection process with a fair comparison formally based on a combination of prior opinion and clinical data.
    
18. New Designs for the Selection of Treatments to be Tested in Randomized Clinical Trials (Recent Publications, #73). Dr. R. Simon, Dr. P. Thall, and Dr. S. Ellenberg.
    The most important aspect of phase III randomized clinical trials is the selection of the experimental treatments to be tested. Often this decision is based on uncontrolled phase II trials. Substantial statistical attention has been focused on the design of phase III trials. Much less statistical effort has been devoted to the design and analysis of phase II trials for screening effective experimental treatments to determine whether they are sufficiently effective, relative to standard treatments, to warrant the conduct of a large randomized phase III trial. This problem is particularly acute in the development of drug combinations where many regimens are possible. We review several designs for such screening trials which we have developed.
    
19. Practical Bayesian Guidelines for Phase IIB Clinical Trials (Recent Publications, #82). Dr. P. Thall and Dr. R. Simon.
    A phase IIB clinical trial typically is a single arm study aimed at deciding whether a new treatment E is sufficiently promising, relative to a standard therapy S, to include in a large-scale randomized trial. Thus, phase IIB trials are inherently comparative even though a standard therapy arm usually is not included. Uncertainty regarding the response of S is rarely made explicit, either in planning the trial or interpreting its results. We propose practical Bayesian guidelines for deciding whether E is promising relative to S in settings where patient response is binary and the data are monitored continuously. The design requires specification of an informative prior for , a targeted improvement for E and bounds on the allowed sample size. No explicit specification of a loss function is required. Sampling continues until E is shown to be either promising or not promising relative to S with high posterior probability, or the maximum sample size is reached. The design provides decision boundaries, a probability distribution for the sample size at termination, and early stopping probabilities under fixed response probabilities with E.
    
20. An Optimal Three-Stage Design for Phase II Clinical Trials (Recent Publications, #23. Ms. L. Ensign, Dr. E.A. Gehan, Mr. D.S. Kamen, and Dr. P.F. Thall.
    A phase II clinical trial in cancer therapeutics is usually a single-arm study to determine whether an experimental treatment (E) is sufficiently promising to warrant further testing, usually in a much larger randomized comparative trial. Patient response is generally defined as a binary variable denoting treatment success/failure, with the parameter p = Pr (success) the criterion for evaluating treatment efficacy. The design for two-stage phase II cancer clinical trials proposed by Gehan (1961) provided the minimum number of patients required so that if all patients were non-responders, further study of the therapy could be discontinued with a given type II rejection error.
    Simon (1989) considered two-stage procedures that permitted rejection of a therapy at either of two stages, but acceptance only at the final stage. The rationale was that stopping a study early was undesirable when a therapy appeared to be effective, but desirable when the treatment seemed ineffective. Simon's design was optimal in that it minimized null expected sample size at given levels of significance and power. One drawback to Simon's design is that it does not allow early termination if there is a long run of failures at the start.
    We propose a three-stage design that permits early stopping when a moderately long sequence of initial failures occurs. In Stage 1, n₁ patients are treated and the trial terminates if all n₁ are treatment failures. If one or more successes are observed in Stage 1, then Stage 2 and (possibly) Stage 3 are implemented, the latter two stages having the same decision rule structure as in Simon's Stages 1 and 2. The optimal three-stage design is parameterized to minimize the null expected sample size.
    
21. A Bayesian Approach to Establishing Sample Size and Monitoring Criteria for Phase II Clinical Trials (Recent Publications, #81). Dr. P. Thall and Dr. R. Simon.
    Thall and Simon propose a Bayesian approach to phase II clinical trials with binary outcomes and continuous monitoring. The efficacy of an experimental treatment E is evaluated relative to that of a standard treatment S based on data from an uncontrolled trial of E, an informative prior for and a non-informative prior for . The trial continues until Eis shown with high posterior probability to be either promising or not promising, or until a predetermined maximum sample size is reached. Operating characteristics are evaluated under fixed values of the success probability of E.
    We propose two extensions of this decision structure, describe sample size and monitoring criteria, and provide numerical guidelines for implementation. The first extension gives criteria for early termination of trials unlikely to yield conclusive results, based on the marginal (predictive) distribution of the observed success rate. The second extension allows early termination only if E is found to be not promising compared to S. Operating characteristics of each of these designs are evaluated numerically over a range of design parameterizations. We also examine the effects of intermittent monitoring on the design's properties. An application of this approach to a leukemia bio-chemotherapy trial is described.
    
22. Bayesian Sequential Monitoring Designs for Single-Arm Clinical Trials with Multiple Outcomes (Recent Publications, #84). Dr. P. Thall, Dr. R. Simon, and Dr. E. Estey.
    We present a Bayesian approach for monitoring multiple outcomes in single-arm clinical trials. Each patient's response may include both adverse events and efficacy outcomes, possibly occurring at different study times. We use a Dirichlet-multinomial model to accommodate general discrete multivariate response. We present Bayesian decision criteria and monitoring boundaries for early termination of studies with unacceptably high rates of adverse outcomes or low rates of desirable outcomes. Each stopping rule is constructed either to maintain equivalence or to achieve a specified level of improvement of a particular event rate for the experimental treatment, compared with standard therapy. We avoid specification of costs and a loss function. We evaluate the joint behavior of the multiple decision rules using frequentist criteria. Applications include trials where response is the cross-product of multiple simultaneous binary outcomes, and hierarchical structures that reflect successive stages of treatment response, disease progression and survival. We illustrate the approach with a variety of single-arm cancer trials, including bio-chemotherapy acute leukemia trials, bone marrow transplantation trials, and an anti-infection trial, with up to seven elementary patient outcomes in each, and up to four monitoring boundaries running simultaneously. We provide general guidelines for eliciting and parameterizing Dirichlet priors and for specifying design parameters.
    
23. Recent Developments in the Design of Phase II Clinical Trials (Recent Publications, #83). Dr. P. Thall and Dr. R. Simon.
    This article reviews statistical methods for the design and conduct of phase II clinical trials. Following a description of the basic phase I-II-III paradigm, the single-stage phase II trial design is described in detail. The limitations of this design are discussed, and more general approaches to treatment development are described. These approaches include designs for selecting treatments prior to phase II, randomized phase II trials, conduct of a sequence of phase II trials, and an integrated phase II/III approach. Practical considerations in the design and conduct of phase II trials are discussed, followed by descriptions of specific designs which incorporate historical data, Bayesian designs, designs with early stopping rules, and designs which account for multiple patient outcomes.
    
24. Analysis of CD7 Expression in Acute Myelogenous Leukemia: Martingale Residual Plots Combined with "Optimal" Cutpoint Analysis Reveals Absence of Prognostic Significance (Submitted Publications, #10). Dr. S.M. Kornblau, Dr. P.F. Thall, Dr. Y.O. Huh, Dr. E.H. Estey, and Dr. M. Andreeff.
    Conflicting results regarding the prognostic significance of CD7 expression in acute myelogenous leukemia (AML) have appeared in the literature. Differences in the method for determining CD7 positivity, the antibody used, the therapy administered, and the CD7 level used as a cutoff for reducing it to a binary variable all have been postulated to account for the discordant findings. In this study, CD7 level was determined by flow cytometry in 331 patients with newly diagnosed AML. Martingale residual plots from fitted Cox regression models of survival on CD7 and an exhaustive cutpoint search were used to obtain "optimal" cutpoints for determining whether a patient was CD7 positive or negative. After accounting for cytogenetic category and other important prognostic variables (age, performance status, albumin), CD7 was shown to have no additional prognostic value, either in its raw numerical form or as a binary variable.
    
25. Estimating Genomic Category Probabilities from Fluorescent in situ Hybridization Counts With Misclassification. (Recent Publications, #80 and Recent Publications, Section of Mathematical Biology, #36). Dr. P.F. Thall, Mr. D. Jacoby, and Dr. S.O. Zimmerman.
    Fluorescent in situ hybridizating (FISH) is used in many medical settings to identify the specific genetic alteration or chromosomal abnormality characterizing a disease. Using FISH techniques, a sample of a patient's cells may be classified into genomic categories, one or more of which is associated with the disease. Estimating the proportion of cells of a particular type based on count data is complicated by classification error inherent in FISH methodology. We address the problem of estimating the proportions of cells of each category in the patient when background counts on cells of known type are available. We do this by modelling the misclassification probabilities and then estimating the parameters of the model and the patient's cell-type proportions jointly using maximum likelihood. Our objective is to obtain more reliable estimates by accounting formally for misclassification error. The method is applied to data arising in chronic myelogenous leukemia (CML), where FISH is used to identify the translocation between chromosomes 9 and 22 which characterizes CML.
    
26. Large Sample Properties of Some Survival Estimators in Heterogeneous Samples. (Recent Publications, #68). Dr. Y. Shen and Dr. T. R. Fleming.
    In studies having right-censored failure time data, the Kaplan-Meier estimator is often used to estimate the corresponding survival distribution. However, the Kaplan-Meier estimator can be biased for a heterogeneous sample, in particular, when the censoring mechanism is dependent on the survival mechanism. We investigated a mean survival estimator that incorporates important covariates associated with survival. The estimator is constructed based on Cox's regression model and a weighting according to any specified distribution of the baseline covariates. A mean hazard function estimator is also derived corresponding to the multiplicative intensity model which allows for recurrent outcome events.
    
27. Weighted mean survival test statistics: a class of distance tests for censored survival data (Recent Publications, #69, #70). Dr. Y. Shen and Dr. T. R. Fleming.
    A class of test statistics is introduced which is sensitive against the alternative of stochastic ordering in the two-sample censored data problem. The test statistics for evaluating a cumulative weighted difference in survival distributions are developed while taking into account the imbalances in baseline covariates between two groups. This procedure can be used to test the null hypothesis of no treatment effect, especially when baseline hazards cross and prognostic covariates need to be adjusted. The statistics are semiparametric, not rank based, and can be written as integrated weighted differences in estimated survival functions, where these survival estimates are adjusted for covariate imbalances.
    
28. Estimating Asymptomatic Duration in Cancer: The AIDS Connection . (Recent Publications, #25). Dr. R. Etzioni and Dr. Y. Shen.
    Many chronic diseases, including cancer and AIDS, do not manifest themselves clinically until some time after their inception. In studies of diseases' natural history, the duration of the asymptomatic period is of interest -- in AIDS, to predict the epidemic's course, and in cancer, to develop efficient screening strategies. By adapting AIDS methodology to cancer, the article identifies a non-parametric method for estimating the duration of the asymptomatic period in cancer. The method is designed to be applied to data from a cohort of individuals, screened periodically. The adapted EM algorithm which, at convergence, provably yields a maximum or saddle point of the likelihood. We investigate the performance of the algorithm by means of a simulation study, and explore the effect of adding a smoothing step to the estimation procedure.
    
29. Parametric Likelihoods for Multiple Non-Fatal Competing Risks and Death. (Submitted Publications, #24). Dr. Y. Shen and Dr. P.F. Thall.
    Clinical trials of fatal diseases often focus on one or more non-fatal events, in addition to survival, both to characterize morbidity and to improve survival estimates. The statistical complications are that the time to each non-fatal event and subsequent residual survival may be either positively or negatively associated, the times to death with or without an antecedent event often have very different distributions, and death may censor some of the non-fatal event times. Consequently, the overall survival time distribution is a mixture of the distributions corresponding to the possible antecedent non-fatal events. In this paper, we consider a general parametric model for multiple non-fatal competing risks and death. The model accounts for positive or negative association between the time of each non-fatal event and subsequent survival while accommodating covariates and the usual administrative censoring. Each event time distribution is specified marginally, and the time of each non-fatal event and subsequent residual survival are combined under a bivariate generalized Morgenstern distribution. The approach is illustrated by application to two data sets from clinical trials in colon cancer and acute leukemia.
    
30. Inference for Long Term Survival by Incorporating Historical Control Information. (Submitted Publications, #23). Dr. Y. Shen and Dr. T.R. Fleming.
    In this article, we present an approach to using historical control data to augment information from a randomized controlled clinical trial, when the clinical trial does not provide long term follow-up for the control regimen. Based on an adjustment procedure to the historical control data, estimation of the long term survival function for the clinical trial control group and inference for the long term treatment effect are obtained. Advantages of this approach are its simple interpretation, validity in important settings, and ability to be readily implemented using existing software. Data from the first and second National Wilms' Tumor studies are used to illustrate the method.
    
31. Age Specific PSA: A Re-assessment. (Submitted Publications, #6) Dr. R. Etzioni, Dr. Y. Shen, Dr. J.C. Petteway and Dr. M.K. Brawer
    Prostate specific antigen (PSA) has emerged as the most useful of all serum tumor markers in prostate oncology. In this article, we have attempted to compare the expected survival advantage when screening for prostate cancer when a positive screen is defined as PSA>4.0 ng/ml, with that expected when a positive screen is defined as PSA greater than an age-specific bound. By extensive statistical simulations, we found that using a bound of 4.0 ng/ml for all ages is more efficient in identifying men with cancer in a screening cohort and will ultimately provide increased population longevity.
    
32. Analysis of Failure Time Data with Linear Transformation Models. (Recent Publications, #14 & #15) Dr. S.C. Cheng, Dr. L.J. Wei and Dr. Z. Ying
    In many clinical trials the goals are to evaluate the effects of patient covariates on survival or disease progression time, and to predict survival probabilities for future patients with given characteristics. Statistical inference is complicated by the presence of right censoring, due to the facts that some patients are lost to follow-up, while others are still alive or have not suffered disease progression or death at the end of the study. For this common data structure, namely censored time-to-event outcomes with covariates, the proportional hazards model introduced by Cox (1972) is often used. The Cox model may not fit the data well, however, when the proportional hazards assumption is not met.
    In this research we consider a large class of semi-parametric transformation models, under which an unknown transformation of the survival time is linear related to the covariates with various specified error distributions. This class of regression models includes the proportional hazards and proportional odds models. Inference procedures, which are derived from a class of generalized estimating equations, are proposed to examine the covariate effects and to predict survival probabilities for future patients. These methods provide a more general and practical alternative to the Cox regression model.
    
33. Distribution-Free Confidence Intervals for Pr(X < Y) Based on Censored Survival Data. (Submitted Publications, #5) Dr. S.C. Cheng and Dr. P.F. Thall
    In randomized clinical trials the primary goal is to compare the efficacy of two treatments in terms of patient survival or disease progression time. If X and Y denote the respective progression/survival times of a random patient under the two treatments, treatment comparison may be carried out by constructing a confidence interval for Pr(X < Y). This quantifies the degree of superiority of one treatment over the other and conveys more information than a p-value. Cox's proportional hazards model (1972) provides a semiparametric basis for estimating Pr(X < Y) based censored survival data. Cheng, Wei and Ying (1995) can also be used to derive semiparametric confidence interval for Pr(X < Y) under an extension of the Cox model.
    In this project we propose a robust nonparametric confidence interval for Pr(X < Y) based on two independent samples of censored survival data. Our procedure generalizes the nonparametric confidence interval proposed by Mee (1990), which only accommodates the uncensored case. Mee's procedure is a modification and improvement of the method of Halperin, Gilbert and Lachin (1987), who in turn rely on the approach used by Ghosh (1979) to construct a confidence interval for a binomial parameter in the one sample case. We study the nonparametric confidence interval by simulation under a range of distributions having proportional hazards, proportional odds, and neither of these properties, and under both moderate and heavy censoring.
    Our simulation results indicate that, in terms of closeness to nominal coverage, the nonparametric interval is superior to the Cox model-based interval under departures from proportional hazards, and is superior to the semiparametric procedure of Cheng, Wei and Ying (1995) for small to moderate sample sizes under heavy censoring.
    
34. Prediction of Cumulative Incidence Function under the Proportional Hazards Model. (Submitted Publications, #4) Dr. S.C. Cheng, Dr. J.P. Fine and Dr. L.J. Wei
    In the presence of dependent competing risks in survival analysis, the Cox model can be utilized to examine the covariate effects on the cause-specific hazard function for the failure type of interest. For example, competing risks can be different causes of death, or local recurrence versus recurrence involving a distant metastasis. In this situation, the cumulative incidence function provides an intuitively appealing summary curve for marginal probabilities of the interested event.
    In this project we show how to construct confidence intervals and bands for such a function under the Cox model for future patients with given covariates. Our proposals are illustrated with data from two breast cancer studies conducted at M.D. Anderson Cancer Center.
    
35. Survival Analysis with Linear Transformation Models When Covariate Variables are Measured with Error. Dr. S.C. Cheng, Dr. N. Wang and Dr. L.J. Wei
    In many biomedical studies, patient covariates are often measured with error, or may not be directly observable. For analysis of survival data, a naive and quite common approach to this problem is to use the Cox model with the observed value of the covariates. Unfortunately, this usually produces biased parameter estimates and unreliable inferences. Existing statistical methods for censored survival data accommodating covariates subject to measurement errors have focused on the estimation of the parameters in the Cox model. In this project we will explore the effect of covariate measurement error on the estimation of regression parameters as well as the prediction of survival probabilities in the context of a much more general class of models.
    
36. Two-Stage Selection and Testing Designs for Clinical Trials Monitoring Both Safety and Efficacy Outcomes. Dr. P.F. Thall and Dr. S.C. Cheng
    This project addresses the general problem of evaluating several new "experimental" treatments and comparing them to an established "standard" treatment. In particular, it is highly desirable to determine which, if any, of the experimental treatments are promising compared to the standard as quickly but reliably as possible, due to limited resources. Thall, Simon, and Ellenberg (1988, 1989) developed optimal two-stage phase II/III designs which first select the best experimental treatment and, if it is promising, subsequently compare it to the standard in a larger second stage. These designs are limited to trials with a single, binary "success" outcome, hence do not accommodate safety monitoring. We propose to develop optimal two-stage selection and testing designs which deal with bivariate patient responses, including both adverse and efficacy outcomes. These outcomes are similar to those treated by the Bayesian continuous-monitoring design of Thall, Simon, and Estey (1996). Additionally, decision rules will be based on a two-dimensional representation of clinically meaningful trade-offs between safety and efficacy.
    
37. A Bayesian Analysis for Estimating the Common Mean of Independent Normal Populations using the Gibbs Sampler. (Recent Publications, #29) Dr. Mary Ann Gregurich and Dr. Lyle D. Broemeling.
    Combining information from several independent normal populations to estimate a common parameter has applications in meta-analysis and is an important statistical problem. For this application, a Bayesian technique via the Gibbs sampler is adopted. Given several normal independent populations with a common mean and different variances, it is possible to perform a complete Bayesian analysis that determines the posterior distribution of the important parameter, the common mean, by using the Gibbs sampler. The methodology is illustrated using two examples. Characteristics such as the mean and the 95% Credible Region are presented. In example 2, a hypothesis test is performed.
    
38. Bayesian Estimation of Multinomial Logit Model Regression Coefficients. Dr. Mary Ann Gregurich.
    A Bayesian approach to estimation of the regression coefficients of a multinomial logit model with ordinal scale response categories is presented. A Monte Carlo method is used to construct the posterior distribution of the link function. The link function is treated as an arbitrary scalar function. The the Gauss-Markov theorem is used to determine a function of the link which produces a random vector of coefficients. The posterior distribution of the random vector of coefficients is used to estimate the regression coefficients. The method described is referred to as a Bayesian generalized least square (BGLS) analysis. Two cases involving multinomial logit models have been described. Case I involves a cumulative logit model and Case II involves a proportional-odds model. All inferences about the coefficients for both cases are described in terms of the posterior distribution of the regression coefficients. The results from the BGLS method are compared to Maximum likelihood estimates of the regression coefficients. The BGLS method avoids the nonlinear problems encountered when estimating the regression coefficients of a generalized linear model. The method is not complex or computationallly intensive. The BGLS method offers several advantages over Bayesian approaches.
    
39. The Internet Survey. Dr. Mary Ann Gregurich and Dr. Marilyn Greer.
    The purpose of this study was to evaluate employee satisfaction with the Internet, the quality of in-house services, and use of the Internet as a tool int was developed and tested with a focus group of employees. The survey was distributed to all faculty and staff using institutional mail. The response rate was 35.4% (615 respondents); respondent, reliability, and validity analysis was completed.
    The majority of respondents were faculty members (61.2%) between 36 and 50 years of age (61.4%). E-mail was the most widely used application (94.1%), followed by the World Wide Web (77.5%). Internet users (87.8%) responded that the Internet was a valuable tool in the workplace and 75% responded that it was easy to use. Only 29% of respondents were satisfied with in-house Internet assistance. Comments suggested that Internet users were dissatisfied with in-house training availability and home access to the Internet.
    Employee use of the Internet will improve the institution's effectiveness as a research center and cancer hospital. It is recommended that the study be repeated every two years to assess employee satisfaction and use of the Internet.

Scientific Consultations and Collaborations

1. Surveillance Committee Review for Protocols:
   • Dr. Peter F. Thall
   • Dr. Su-Chun Cheng
   • Dr. Kenneth R. Hess
   • Dr. J. Jack Lee
   • Dr. J. Lynn Palmer
   • Dr. Yu Shen
   • Mrs. Terry L. Smith
   • Dr. Mary Ann Gregurich
2. Breast cancer prognosis in young women (Recent Publications, #31). Dr. V. Guinee, Ms. S. Taylor, and Dr. K. Hess.
   This study was undertaken to determine possible prognostic factors in women less than 30 years old with breast cancer. The clinical presentation and course of treatment was documented for 407 women aged 20-29 years with histologically-confirmed local or regional invasive breast carcinoma who registered from 1978 through 1988 at one of nine centers participating in the International Cancer Patient Data Exchange System. Analysis was performed using Cox proportional hazards regression analysis. For patients whose breast cancer was diagnosed during pregnancy, the relative risk of dying from beast cancer (compared to those who had not been pregnant) was 3.2 (p = 0.0005, 95% confidence interval: 1.8 to 5.8). The relative risk, adjusted for number of metastatic axillary nodes and tumor diameter, was 2.6 (p = 0.036, 95% confidence interval: 1.1 to 5.8). For each 1-year increment in the interval between a prior pregnancy and breast cancer diagnosis, the risk of dying decreased by 15 percent, a relative risk of 0.85 (p = 0.011, 95% confidence interval: 0.74 to 0.97). Insufficient data were available to assess the effect of subsequent pregnancy on survival. However, the magnitude of the effect of a concurrent or recent pregnancy is such that it should be considered as a contributing factor in the acknowledged poor prognosis of young women with breast cancer.
   
3. Risk factors for neurosurgical complications. Dr. R. Sawaya, Dr. M. Hammoud, and Dr. K. Hess.
   Data were collected on 400 craniotomies to determine predictors of postoperative complications. Complications were scored as major or minor and systemic or neurological. Data on twelve potential predictors were analyzed with multiple logistic regression and classification tree analysis. A total of 53 (13%) cases developed a major complication: 34 neurologic deficit, 12 other neurologic, and 11 systemic. Only three patients had multiple types of major complications. Logistic regression identified age and grade (lesion in nonfunctional part of brain, near functional part, in functional part) as significant independent predictors of the occurrence of major complications. Classification tree analysis also demonstrated that among the cases with tumors in the nonfunctional parts of the brain, pathology (glioma or metastasis) was predictive. Of the 79 patients with metastatic lesions in nonfunctional areas, only one (1%) developed a major complication. Analysis of factors related to length of stay was performed using linear regression and regression tree analyses (with a log (log) transformation of hospital stay). In addition to age, grade, and pathology, lower performance scores were found to be associated with prolonged stays.
   
4. Fecal occult blood testing in a community setting. Dr. B. Levin, Ms. C. Johnson, and Dr. K. Hess.
   A mass screening (80,000 kits distributed) with fecal occult blood tests was performed in cooperation with a local television station (Channel 13) and local grocery store chain pharmacies (Randall's). The goal was to compare three tests manufactured by Smith-Kline Diagnostics. A Scantron questionnaire was included with the test kits. A total of 10,633 kits were returned with 1,748 testing positive on one or more tests. These patients were urged to see their doctors for further testing. The test positivities ranged from 5% to 15%, while the positive predictive values in the 1029 patients with clinical follow-up ranged from 6% to 13%. Non-compliance with dietary and medication restrictions and the presence of gastrointestinal symptoms increased the positivity rates.
   
5. Unknown primary tumors (Recent Publications, #1). Dr. J. Abbruzzese and Dr. K. Hess.
   A primary objective of this study was to identify new and verify existing prognostic factors in patients with unknown primary carcinoma. Among 927 consecutive patients referred to the Unknown Primary Clinic between 1987 and 1992, a standardized clinical evaluation identified 657 patients with unknown primary carcinoma. Data on age, gender, race, histology and pattern of involved organ sites were correlated to survival time using Cox proportional hazards regression analysis. Increasing numbers of involved sites, adenocarcinoma histology, male gender, and liver involvement were significant independent predictors of shorter survival times. Neuroendocrine carcinoma histology and lymph node involvement were significant independent predictors of longer survival. Based on the prognostic index computed from the Cox regression analysis, we profiled the 139 lowest-risk patients and the 100 highest-risk patients. We also analyzed the two-way interactions between the prognostic factors to determine whether the effects of some factors depended on the levels of the other factors. Among the patients with isolated lymph node involvement, it was demonstrated that patients with supraclavicular nodal involvement had shorter survival and among patients with isolated liver involvement. Patients with neuroendocrine metastases had longer survival. In the future, prognostic stratification will be performed using classification tree analysis. This information will allow the selection of patients for specific therapeutic interventions.
   
6. Chemoprevention aerodigestive PO1 project (CAP). Dr. W. Hong, Dr. S. Lippman, Dr. J. S. Lee, Dr. R. Lotan, Dr. W. Hittelman, Dr. M. Tainsky, and Dr. J. Lee, and Dr. T. Moon.
   The CAP study is a five-year project which includes two Phase III clinical trials and four basic science research to study the efficacy and mechanism of retinoid chemoprevention in the aerodigestive tract. Dr. Lee is the statistician of "Chemoprevention Trial in Oral Premalignancy" and is responsible for the statistical issues involved in the design, implementation, and analysis of this trial. He also provides statistical consultation to the basic science studies.
   
7. Phase III double-blind randomized trial of 13-cis retinoic acid to prevent second primary tumors (SPT) in stage I non-small cell lung cancer. Dr. S. Lippman, Dr. S. Benner, Dr. R. Winn, Dr. J. J. Lee, Mr. Joel Dubin, and Ms. N. Tu.
   As of February, 1995, there were 736 patients registered in this multi-center clinical trial and among them, 593 were randomized. It is expected that it will take another two years to reach the accrual goal of enrolling 1,260 patients. With an additional three to four years of follow-up, the study will have 80% power in detecting a 50% reduction in the SPT incidence rate. The study is moving along well. The accrual rate and compliance rate are on target. QA and QC procedures are in place to ensure the quality of the data. Statistical reports are provided every six months.
   
8. Suppression of retinoic acid receptor b in oral premalignant lesions and its upregulation by isotretinoin. Dr. R. Lotan, Dr. S. Lippman, and Dr. J. Lee.
   Data from 52 oral leukoplakia patients indicated that only 40% of the lesions expressed RAR-ß at baseline while 100% expression was found in normal controls. It was found that, after 3-month of high-dose isotretinoin treatment, 90% of the lesions expressed RAR-ß (p=.00001, McNemar's test). The upregulation of RAR-ß is correlated with clinical response (p=.039, Fisher's exact test).
   The result suggests that RAR-ß expression is an excellent biomarker for retinoid chemoprevention trials in oral carcinogenesis.
   
9. Relationship of P-glycoprotein and CEA expression in human colon carcinoma to local invasion, DNA ploidy, and disease relapse. Dr. F. Sinicrope and Dr. J. Lee.
   Among a total of 52 Stage B2 colon cancer patients, P-glycoproteins were detected in 44 and 42 patients by JSB-1 and HYB-241 and CEA was found in 50 patients. In a multivariate analysis, a high level of P-gp expression, DNA aneuploidy, microinvasion, and single carcinoma cell invasion were found to be significant predictors of disease relapse. The results indicate that diffuse P-gp expression is present in the majority of Stage B2 colon cancer and can be of prognostic value in predicting disease recurrence.
   
10. Prognostic value of size of lymph node metastases in patients with cutaneous melanoma. Dr. A. Buzaid, Dr. S. Legha, Dr. R. Benjamin, Dr. J. Lee, and Ms. N. Tu.
    In a retrospective study of 442 melanoma patients, it was found that the number of positive nodes, age, and tumor thickness were significant prognostic factors for survival. On the other hand, the size of the nodal mass by either physical examination or pathology were not of prognostic value and should be eliminated from the current staging system.
    
11. p53 and retinoid chemoprevention of oral carcinogenesis. Dr. S. Lippman, Dr. D. Shin, and Dr. J. Lee.
    The patterns of p53 expression, association with retinoid treatment and modulation were studied in 40 patients with oral premalignant lesions. It was found that the accumulation of p53 had a direct association with histology grade and an inverse correlation was found between the baseline p53 expression and treatment response.
    
12. Trisomy 7: A potential cytogenetic marker of human prostate cancer rogression. Dr. M.G. Bandyk, Dr. L. Zhao, Dr. P. Troncoso, Dr. L.L. Pisters, Dr. A.C. von Eschenbach, Dr. L.W.K. Chung, Dr. J.C. Liang, and Dr. J.L. Palmer.
    The technique of fluorescence in situ hybridization (FISH) was employed to demonstrate that chromosome 7 trisomy is associated with the progression of human prostate cancer to advanced stages and to metastatic sites such as the bone marrow. Thirty-five human specimens including 14 primary prostate tumors, 16 metastatic lesions, and 5 normal prostate tissues were examined for the presence of chromosome 7 aneuploidy. In addition two prostatic tumor cell lines of different tumorigenic potential were also evaluated for the copy number of chromosome 7. Results showed that the androgen-unresponsive and tumorigenic cell line PC-3 exhibited a significantly higher ratio of chromosome 7 to total chromosome number than the androgen-responsive, non-tumorigenic cell line. In the human prostate specimens, significantly elevated frequencies of trisomy 7 cells were observed in the advanced stage tumors as compared to the early stage tumors and normal prostatic tissue. Furthermore, metastases were found to have a significantly higher frequency of trisomy 7 cells than primary tumors. Trisomy 7 was apparently not due to triploidy of the cells because when chromosome 9 was used as an internal control, all primary tumors and metastases were predominantly disomic for this chromosome. Also, in two patients with paired primary and metastatic tumors, the percentage of trisomy 7 cells increased from 4-7% in the primary tumors to 42-45% in the metastatic prostate tumor cells in the bone marrow. Therefore, our data suggest that trisomy 7 may be a common feature associated with the local and metastatic progression of human prostate cancer and may serve as a novel marker for human prostate cancer progression.
    
13. p53 protein accumulation in the progression of human prostate carcinoma. Dr. N.M. Navone, Dr. P. Troncoso, Dr. L.L. Pisters, Dr. A.C. von Eschenbach, Dr. C.J. Conti, and Dr. J.L. Palmer.
    Nuclear accumulation of the p53 protein has been shown to be strongly associated with p53 mutations. Previous studies on accumulation of the p53 protein in prostate carcinoma have been confined to primary tumors. We studied the accumulation of p53 protein in specimens obtained from primary and metastatic sites of prostate carcinoma. This study was designed to understand the role of p53 mutation in the progression of prostate carcinoma and to determine the relationship between accumulation of this protein with stage, histologic grade and androgen responsiveness of the tumor. We studied the accumulation of the p53 protein by means of immunohisto-chemical methods using the polyclonal antibody to human p53. The material studied consisted of formalin fixed, paraffin-embedded tissue obtained from primary tumors and metastases of 92 cases of prostate carcinoma. All tumors with p53 accumulation were metastatic, poorly differentiated and androgen independent. Results showed that p53 accumulation had a strong association with stage, grade, and androgen sensitivity. Logistic regression analysis demonstrated that androgen sensitivity predicted p53 outcome better than stage or grade alone.
    
14. Significance of tumor size and radiation dose to local control in stage I-III diffuse large cell lymphoma treated with chop-bleo and radiation. Dr. L.M. Fuller, Dr. P. McLaughlin, Dr. A. Rodriguez, and Dr. J.L. Palmer.
    The purpose of this study was to evaluate the possible effect of adjunctive involved field radiotherapy on long-term local control for patients with Ann Arbor Stage I-III diffuse large cell lymphoma who achieved a complete remission on a combined modality program which included cyclophosphamide, doxorubicin, vincristine, prednisone, and Bleomycin (CHOP-Bleo). One hundred and ninety patients were included in this study. Analyses were made to determine response to treatment according to stage, extent of maximum local disease and irradiation dose. Relapse patterns were also studied. A complete remission was achieved in 162 patients. Among patients who achieved a complete response, local control was better for those who received tumor doses of 40 Gy (97%) than for those who received < 40 Gy (83%). Among those with extensive local disease, the corresponding control rates were 88% and 71%, respectively.
    
15. Analysis of lymphangiogram and laparotomy findings in relation to staging in adult patients with Hodgkin's disease. Dr. L. Fuller, Dr. N. Mirza, and Dr. J.L. Palmer.
    The objective of this study is to determine prognostic factors which correctly predict a positive finding of Hodgkin's disease by lymphangiogram; and for those patients with a negative lymphangiogram, which factors would predict a correct positive finding of the disease by laparotomy. Laparotomies are being eliminated as sequential staging procedures, but a large data base is available for past procedures performed at M.D. Anderson.
    
16. Second-line treatment for breast cancer patients. Dr. R. Winn, Dr. M. Weitzner, Dr. C. Meyers, Dr. J. Palmer and others.
    The objectives of this study are to determine what variables predict whether or not a breast cancer patient will receive second-line chemotherapy and which predict whether or not a breast cancer patient who receives second-line chemotherapy will respond. The study will also attempt to determine outcomes (in terms of survival, quality of life) for breast cancer patients who do and who do not receive second-line chemotherapy (and who do and do not respond). Since this is not a randomized study, efforts were made to insure that patients compared were equivalent in terms of major clinical variables.
    
17. Evaluation of interferon therapy for treatment of chronic myelogenous leukemia (CML). Dr. H. Kantarjian and Mrs. T. Smith.
    Approximately one-third of patients with early stage CML achieve a cytogenetic response (significant reduction of Philadelphia chromosome), sometimes after prolonged therapy with interferon. We were able to show that risk of death is reduced in these patients, compared to those with no or lesser response. In addition to the Philadelphia chromosome, patients with CML often develop other chromosomal abnormalities during the long course of their disease. Recursive partitioning methods are being used in an exploratory phase of investigation into the impact on survival of the various types of abnormalities. This analysis is based on a large series of patients treated with interferon.
    
18. Staging of chronic lymphocytic leukemia (CLL). Dr. M. Keating and Mrs. T. Smith.
    Classical staging systems, such as those of Rai and Binet, were developed prior to the advent of effective chemotherapy agents such as fludarabine. Since 1984, several hundred CLL patients have been treated at MDACC on fludarabine regimens. Factors associated with treatment response and survival have been investigated in this population; inclusion of additional factors resulted in prediction models that categorized patients more accurately than traditional staging systems when tested in an independent group of CLL patients. A large patient population with early stage CLL not requiring fludarabine therapy is also being investigated to determine if inclusion of information on such characteristics as B2-microglobulin can improve on previous staging systems.
    
19. Evaluation of breast cancer treatment programs. Dr. G. Hortobagyi, Dr. A. Buzdar, Dr. F. Holmes, and Mrs. T. Smith.
    Many large clinical trials are underway evaluating various strategies for the treatment of both early stage and advanced forms of breast cancer, and statistical methods are applied for the design and analysis of these trials. An investigation of the role of radiotherapy, and whether it may be associated with the development of second cancers, is underway. Other studies involve the role of hormone therapy, intensified therapy with bone marrow transplantation, and the timing of various components of therapy.
    
20. Predicting the need for blood transfusion following surgery for head and neck cancer. Dr. R. Weber and Mrs. T. Smith.
    A model incorporating several pre-surgery factors has been developed to predict which patients undergoing surgery for head and neck cancer are likely to require blood transfusion. This would allow more efficient decisions regarding directed blood donations. Data are being collected on recent patients so that the model may be subjected to testing before it is recommended for prospective application.
    
21. Adenocarcinoma as an independent risk factor for recurrence in patients with stage IB cervical carcinoma. Dr. P. Eifel and Mrs. T. Smith.
    The objective of this investigation is to determine the influence of histologic type on outcome in patients with stage IB carcinoma of the cervix treated with radiotherapy. A draft manuscript has been written and is under review.
    
22. Use of G-CSF before, during, and after fludarabine+ ara-C induction therapy of newly-diagnosed AML or MDS: Comparison of fludarabine + ara-c without G-CSF (Recent Publications, #24). Dr. E. Estey, Dr. P. Thall, Dr. M. Andreeff, Dr. M. Beran, Dr. H. Kantarjian, Dr. S. O'Brien, Dr. S. Escudier, Dr. L.E. Robertson, Dr. C. Koller, Dr. S. Kornblau, Dr. S. Pierce, Dr. E.J Freireich, Dr. A. Deisseroth, and Dr. M. Keating.
    Two lines of evidence suggest that the use of granulocyte colony-stimulating factor (GCSF) might be beneficial in the management of untreated acute myelogenous leukemia (AML). Japanese investigators have found that administration of GCSF following completion of chemotherapy shortens the duration of neutropenia by about one week and also reduces the infection rate among previously treated AML patients. In vitro data suggest that GCSF administered before, during, and shortly after ara-C may sensitize leukemia cells to the drug. In this paper, the clinical effects of GCSF administered in this manner with fludarabine + ara-C (FLAG), compared to fludarabine + ara-C alone (FA), are investigated. Logistic regression analyses of 112 FLAG patients and 85 FA patients, from consecutive studies, showed accelerated neutrophil recovery with FLAG, but no significant difference in complete remission rate between FLAG and FA.
    
23. Phase II study of accelerated fractionation radiation therapy with carboplatin followed by PCV chemotherapy for the treatment of glioblastoma multiforme (Submitted Publications, #14). Dr. V.A. Levin, Dr. P.F. Thall, Dr. M.H. Maor, Dr. W.K.A. Yung, Dr. J. Bruner, Dr. R. Sawaya, Dr. A. Kyritsis, Dr. S. Woo, Dr. L. Rodriguez, and Dr. M.J. Gleason.
    In the past 20 years, numerous drug-radiation combinations have been tried in an attempt to improve the survival of patients with glioblastoma multi-forme (GBM). This paper describes a phase II one-arm study to evaluate the long-term efficacy and safety of accelerated fractionated radiotherapy combined with intravenous carboplatin for patients with previously untreated glioblastoma multiforme tumors.
    Eighty-three patients received 1.9-2.0 Gy radiation three times a day with 2-h infusions of 33 mg/m² carboplatin for two 5-day separated by 2 weeks; following radiotherapy, patients received procarbazine, lomustine (CCNU), and vincristine (PCV) for 1 year or until progression.
    Seventy-four of the 83 patients (89%) received one or more courses of PCV; their median survival was 55 weeks. Total resection was performed in 20%, subtotal resection in 69%, and biopsy in 11%; reoperation (total or subtotal resection) was performed in 37% of patients. Survival was worst for those 61 years old (median 35 weeks). Covariates individually predictive of improved survival were younger age, smaller radiation volume, total or subtotal resection vs. biopsy, and higher Karnofsky performance status. A multivariate analysis showed that age and extent of initial surgery together were predictive of a better survival with no other variables providing additional significance. Only 8.4% of patients had therapy-associated neurotoxicity ("radiation necrosis").
    When comparable selection criteria were applied, the survival in this study is similar to the results currently attainable with other chemoradiation approaches. The relative safety of accelerated fractionated radiotherapy, as used in this study with carboplatin, enables concomitant full dose administration of chemotherapy or radio-sensitizing agents in GBM patients.
    
24. Prognostic significance of preoperative MRI scans in gliobastoma multiforme (Submitted Publications, #7). Dr. M.A. Hammoud, Dr. R. Sawaya, Dr. W. Shi, Dr. P.F. Thall, and Dr. N.E. Leeds.
    Tumor necrosis, enhancement, and associated edema in patients with gliobastoma multiforme (GBM) represent biological variables that can be quantitated on preoperative MRI scans. We reviewed 48 high selected patients, all of whom had supratentorial lesions, underwent gross total tumor resection, and received adjuvant radio- and chemotherapies. None had prior surgery. The median age was 50 years. The median Karnofsky performance score was 80. Cox regression analysis revealed that the strongest prognostic variable is the amount of tumor necrosis on preoperative scan, with median survivals of 42, 24, 15, and 12 months for tumor necrosis grades of 0, I, II, and III, respectively. The intensity of enhancement of the tumor nodule is also prognostic, with median survivals of 35, 18, and 13.5 months for enhancement grades of 0, I, and II, respectively. The extent of peritumoral edema has a quadratic effect, with grades I, II, and III surviving for 24, 12, and 20 months respectively. In conclusion, GBM patients with little or no necrosis and with less tumor nodule enhancement on preoperative MRI survive longer than patients with greater amounts of necrosis and with less tumor nodule enhancement on preoperative MRI survive longer than patients with greater amounts of necrosis and greater degrees of tumor enhancement. In addition, a moderate degree of peritumoral edema is associated with worse prognosis.
    
25. Evaluation of Short-Chain Fatty Acid Enemas: Treatment of Radiation Proctitis (Recent Publications, #3). Dr. R. Al-Sabbagh, Dr. F.A. Sinicrope, Dr. J.H. Sellin, Dr. Y. Shen, and Dr. L. Roubein.
    Radiation proctitis is a common complication of abdominal and pelvic radiotherapy; unfortunately, there is no established effective therapy for radiation proctitis. Short-chain fatty acids (SCFA) have been effectively used to treat a variety of colitides. We sought to determine whether SCFA enemas have a role in the treatment of radiation proctitis. Results: Four weeks of treatment with SCFA enemas resulted in clinical improvement in all patients. There were modest, but not significant, changes in endoscopic and pathological parameters. Therefore, SCFA is a promising therapeutic option in radiation proctitis.
    
26. Elevated Soluble Fas Levels in Nonhemotopoietic Human Malignancy (Submitted Publications, #16). Dr. G.P. Midis, Dr. Y. Shen and Dr. L.B. Schaub
    Fas is a widely expressed membrane-anchored protein that induces apoptosis. Soluble Fas (sFas), generated by alternative mRNA splicing, can antagonize cell-surface Fas function. We have investigated sFas in 104 cancer patients with nonhematopoietic malignancies using a Fas-specific ELISA and immunoprecipitation. Our studies demonstrate an elevated 40-42 kDa sFas species in both patient serum and tumor explants. These observations provide the first evidence that sFas is increased in patients with solid tumors in a manner reflective of disease stage and tumor burden and argue that sFas can be synthesized and released both systemically and locally within the tumor microenvironment.
    
27. Increased Apoptosis Accompanies Neoplastic Development in the Human Colorectum (Recent Publications, #75). Dr. F.A. Sinicrope, Dr. G. Roddey, Dr. T.J. McDonnell, Dr. Y. Shen, Dr. K.R. Cleary and Dr. C. Stephens.
    A disturbance in the balance between cell proliferation and cell loss, or apoptosis, may underlie neoplastic development. Therefore, we determined spontaneous apoptotic and proliferative rates in normal, hyperplastic, adenomatous, and malignant colorectal epithelia. In normal mucosa, luminal epithelial cells demonstrated higher rates of apoptosis compared to cells in the proliferative zone. Neoplastic transformation was associated with a significant increase in rates of apoptosis and proliferation. However, apoptosis, but not proliferation, decreased at the adenoma to carcinoma transition coincident with expression of mutant p53. In carcinomas, both mutant p53 and bcl-2 protein levels were associated with attenuated apoptotic rates. In conclusion, apoptosis is an important regulator of growth in normal and neoplastic epithelia.
    
28. Loss of p21 Protein Expression Accompanies Neoplastic Progressionin Human Colorectalum (Submitted Publications, #19). Dr. G. Roddey, Dr. S. Ruan, Dr. C. Stephens, Dr. M.L. Frazier, Dr. G. Glober, Dr. Y. Shen, Dr. W. Zhang, and Dr. F.A. Sinicrope.
    p21, a cyclin-dependent kinase inhibitor, can be transcriptionally activated by p53 and mediates cell cycle arrest. We analyzed p21 expression during colorectal tumorigenesis and its relationship to p53, bcl-2, and to rates of proliferation and apoptosis. No relationship between p21 and proliferative or apoptotic labeling indices, or with bcl-2 expression, was found in neoplasms. In conclusion, a reduction in the frequency of p21 expression and altered topography accompany colorectal neoplastic development and progression. These findings suggest that p21 is an early target of alteration and that loss of negative cell cycle regulation may be a critical event in colorectal tumorigenesis. The lack of correlation between p21 and p53 expression suggests that p21 may be regulated by p53-independent pathways.

Other Consultations and Collaborations

1. Critical determinants of aggressive skin cancer program project (application). Dr. M. Kripke, Dr. T. Moon, and Dr. J. Lee.
2. Chemoprevention trial in oral premalignancy. Dr. S. Lippman, Dr. R. Winn, Dr. J. Lee, Dr. G. Clayman, and Dr. T. Moon.
3. Genetic alterations in PUVA-induced skin cancers. Dr. H. Ananthaswamy and Dr. T. Moon.
4. Medical informatics patient studies unit. Dr. M. Morris and Dr. T. Moon.
5. Institutional data monitoring. Dr. M. Keating and Dr. T. Moon.
6. Molecular investigation of etiological factors in hepatocellular carcinoma in the U.S. Dr. Y. Patt, A. Hogue, and Dr. T. Moon.
7. Therabite mobilizer for maxillary resection. Dr. R. Jacob and Dr. T. Moon.
8. Smoking cessation among school children. Dr. A.V. Prokhorov and Dr. T. Moon.
9. Retrograde cerebral perfusion in pigs. Dr. H. Safi and Dr. K. Hess.
10. Cerebral spinal fluid drainage and distal aortic perfusion in surgery for thoracoabdominal aortic aneurysms. Dr. H. Safi and Dr. K. Hess.
11. Bcl-2 in medulloblastoma. Dr. J. Bruner and Dr. K. Hess.
12. p53 correlations in breast cancer. Dr. M. Cervantes and Dr. K. Hess.
13. Radiosurgery for brain metastases. Dr. R. Sawaya, Mr. R. Bindal, and Dr. K. Hess.
14. MIB-1 antigen and BudR in meningioma. Dr. L. Langford, Ms. C. Cooksley, and Dr. K. Hess.
15. Prognostic factors in melanoma of the foot. Dr. J. Reilly, Mr. G. Giacco, and Dr. K. Hess.
16. Pelvic exenteration for rectal carcinoma. Dr. J. Skibber, Mr. G. Giacco, and Dr. K. Hess.
17. Improved contrast-enhanced MRI of the breast. Dr. K. Wright and Dr. K. Hess.
18. Fibroblast growth factor receptor alterations in human gliomas. Dr. R. Morrison and Dr. K. Hess.
19. Latent primary malignancies. Dr. J. Abbruzzese and Dr. K. Hess.
20. Reoperation for metastatic brain tumor. Dr. R. Sawaya, Mr. R. Bindal, and Dr. K. Hess.
21. Melanoma metastatic to GI tract. Dr. H. Gutman and Dr. K. Hess.
22. Contrast-enhanced MRI of the liver. Dr. M. Fenstermacher and Dr. K. Hess.
23. Detecting axillary lymph nodes in breast cancer. Dr. S. Wallace and Dr. K. Hess.
24. Heart rate variability. Dr. T. Vybiral and Dr. K. Hess.
25. Sequential annual immunization with influenza vaccine. Dr. W. Keitel and Dr. K. Hess.
26. High doses of purified influenza vaccine. Dr. W. Keitel and Dr. K. Hess.
27. Aggressive meningeal tumors. Dr. R. Sawaya, Mr. G. Younis, and Dr. K. Hess.
28. BUdR labeling in glioblastoma multiforme. Dr. R. Sawaya, Dr. A. Ritter, and Dr. K. Hess.
29. Sarcoma brain metastases. Dr. R. Sawaya, Mr. R. Bindal, and Dr. K. Hess.
30. Phase II and III Trials for comparing the combined effect of cis-retinoic acid, alpha-interferon and radiation versus radiation alone in metastatic cervical cancer. Dr. S. Lippman, Dr. I. Krakoff, and Dr. J.J. Lee.
31. Correlation of PSA, retinol, and 13-cRA level in oral leukoplakia patients. Dr. S. Lippman, Dr. H. Fritsche, and Dr. J. Lee.
32. Expression of p53 oncoprotein in non-small cell lung cancer: A favorable prognostic factor? Dr. J.S. Lee, Dr. J. J. Lee, and Ms. N. Tu.
33. Phase I trial of chemoradiation therapy with taxol and brachytherapy for patients with NSCLC. Dr. J. S. Lee and Dr. J. J. Lee.
34. Bronchial squamous metaplasia chemoprevention trial. Dr. W.K. Hong, Dr. J.S. Lee, and Dr. J. Lee.
35. Lung cancer chemoprevention with synthetic retinoid 4-HPR. Dr. W. K. Hong, Dr. E. Gehan, and Dr. J. Lee.
36. Primary and secondary prevention of lung cancer. Dr. W. K. Hong, Dr. J. J. Lee, Dr. M. Spitz, and Dr. C. Amos.
37. Lung cancer chemoprevention research program (U19). Dr. W.K. Hong, Dr. J. Kurie, Dr. W. Hittelman, Dr. R. Lotan, and Dr. J. Lee.
38. The prognostic effect of p53 expression in head and neck cancer patients. Dr. D. Shin, Dr. W. Hittelman, Dr. J. Lee, and Ms. N. Tu.
39. Combined therapy of high-dose 13cRA, alpha interferon and CDDP in patients with stage III/IV head and neck SCC. Dr. D. Shin and Dr. J. Lee.
40. High proliferation cell nuclear antigen expression predicts poor survival in patients with head and neck squamous cell carcinoma. Dr. D. Shin and Dr. J. Lee.
41. A phase II study of topotecan in patients with advanced non-small cell lung cancer previously untreated with chemotherapy. Dr. R. Perez-Soler and Dr. J. J. Lee.
42. Micronuclei analysis on the high-dose induction, low-dose maintenance, oral leukoplakia trial. Dr. S. Benner and Dr. J. J. Lee.
43. Phase I trials of new anti-cancer agents. Dr. M. Raber, Dr. J. Lee, and Dr. E. Gehan.
44. DNA repair in biochemotherapy of melanoma. Dr. F. Ali-Osman, Dr. A. Buzaid, and Dr. J. Lee.
45. Macrophages and cytokines in biochemotherapy of melanoma. Dr. E. Grimm, Dr. A. Buzaid, and Dr. J. Lee.
46. Protective effects of in vivo 13-cis-retinoic acid treatment against mutagen-induced genetic damage. Dr. Z. Trizna and Dr. J. Lee.
47. A randomized study of vitamin E versus placebo in the prevention of treatment induced mucositis. Dr. C. Verschraegen and Dr. J. Lee.
48. Melphalan-taxol dose finding study. Dr. D. Gershenson, Dr. P. Thall, and Dr. J. J. Lee.
49. Is follow-up of lung cancer patients following resection medically indicated and cost effective? Dr. G. Walsh and Dr. J. Lee.
50. Lung cancer risk assessment among ex-smokers. Dr. M. Spitz and Dr. J. Lee.
51. Phase II study of adriamycin, ifosfamide and platinum on sarcomas. Dr. S. Patel and Dr. J. Lee.
52. Master agreement application -- Phase I Chemoprevention Agents. Dr. B. Levin and Dr. J. Lee.
53. Subcutaneous heparin in reducing the pulmonary embolism and venous thrombosis in patients receiving thoracic surgery. Dr. J. B. Putnam, Jr. and Dr. J. Lee.
54. A randomized comparison of intravenous versus thoracic epidural sufentanil for analgesia in adult patients. Dr. J. C. Nesbitt and Dr. J. Lee.
55. The combination therapy of 13-cis-retinoic acid with alpha-interferon in treating aggressive squamous carcinoma of the skin. Dr. G. Clayman, Dr. S. Lippman, and Dr. J. Lee.
56. Quantitative DNA analysis by PCR. Dr. J. Roth, Dr. R. A. White and Dr. J. Lee.
57. Calcium carbonate therapy of diarrhea in intestinal bypass patients. Dr. G. Steinbach and Dr. J. Lee.
58. Studies on the role of H. Pylori in the immunopathogenesis of MALT lymphomas. Dr. G. Steinbach and Dr. J. Lee.
59. Using p65 staining as a screening tool for diagnosing colon cancer. Dr. G. Steinbach, Dr. J.J. Lee, and Ms. N. Tu.
60. Expression of the bcl-2, p53, and p-glycoproteins during colorectal tumorigenesis. Dr. F. Sinicrope and Dr. J. J. Lee.
61. The design and analysis of human colorectal epithelial cell proliferation studies using the labeling index as an intermediate marker. Dr. M. Wargovich, Dr. G. Steinbach, Dr. R. Winn, and Dr. J. Lee.
62. Reliability analysis of computerized scoring system of colon epithelial cell proliferation index. Dr. M. Wargovich, Dr. J. J. Lee, and Mr. J. Dubin.
63. Protocol DM 89-08: Utility of a panel of intermediate markers in predicting colonic adenomatous polyp recurrence. Dr. R. Winn, Dr. M. Wargovich, Dr. J. J. Lee, and Ms. N. Tu.
64. Time to hematopoietic recovery after high-dose cyclophosphamide, etoposide and cisplatin. Dr. R. Champlin, Dr. D. Seong, and Dr. J. Palmer.
65. High dose tandem chemotherapy with or without autologous bone marrow transplantation. Dr. D. Seong and Dr. J. Palmer.
66. Blood stem cell transplantation in patients with high risk hematologic malignancies. Dr. M. KÖrbling and Dr. J. Palmer.
67. Pancreatic patient quality-of-life study. Dr. R. Winn, Dr. R. Pazdur, and Dr. J. Palmer.
68. Phase I clinical trials of anticancer agents. Dr. J. Abbruzzese, Dr. M. Raber, and Dr. Palmer.
69. Survival differences in three groups of colon cancer patients. Dr. R. Pazdur and Dr. J. Palmer.
70. A phase I-II study of induction chemotherapy followed by radiotherapy and concomitant boost chemotherapy in patients with locally advanced nasopharyngeal carcinoma. Dr. B. Glisson and Dr. J. Palmer.
71. Phase II study of topotecan and ara-C in MDS and CML patients. Dr. M. Beran, Dr. H. Kantarjian, and Dr. J. Palmer.
72. Genotoxic effects of multiple cycles of eight chemotherapeutic agents on mouse germinal stem cells. Dr. P. Sen, Dr. M.L. Meistrich, Dr. M. Kattan, Dr. J.C. Liang, and Dr. J.L. Palmer.
73. Androgen independent growth in prostate cancer. Dr. C.J. Conti, Dr. L. Chung, Dr. P. Troncoso, Dr. A.C. von Eschenbach, Dr. N. Navone and Dr. J.L. Palmer.
74. Clinical trials of biological response modifiers. Dr. J.L. Murray, Dr. R. Freedman, Dr. H.M. Vriesendorp, Dr. S.M. Quadri, Dr. D.A. Podoloff, Dr. D.J. Macey, Dr. L.P. Kasi, Dr. H.Z. Zhang, Dr. R. Katz, Dr. M. Raber, and Dr. J.L. Palmer.
75. Quality of life measurement in newly diagnosed patients with intermediate grade non-Hodgkin's lymphomas. Dr. M.A. Rodriguez, W. Swanson, and Dr. J.L. Palmer.
76. Phase I-II studies for patients with small cell lung cancer. Dr. B. Glisson, Dr. R. Komaki, and Dr. J.L. Palmer.
77. A comparison of the sensitivity of blood and bone marrow for the presence of the Ph+ chromosome in chronic myelogenous leukemia by fluorescent in situ hypridization. (Submitted Publications, #22). Dr. D. Seong, Dr. P.F. Thall, Dr. H. Kantarjian, Dr. J. Swantkowski, Dr. J. Xu, Dr. Y. Shen, Dr. S. Zimmerman, Dr. A. Glassman, Dr. L. Ramagli, Dr. E. Freireich, Dr. R. Champlin and Dr. M. Siciliano.
78. Multi-institutional phase I trial of Cox-2 inhibitor, piroxicam-DFMO, piroxicam versus sulindac control for patients with colorectal segment or duodenal polyps. Dr. Steinbach, Dr. Shen, Dr. Zimmerman, et al.
79. Cooperative family registry for epidemiologic studies of colon cancer. Dr. P. Lynch, Dr. C. Amos, Dr. Y. Shen, et al.
80. Study of mechanisms of Fos induced osteosarcoma and cell transformation. Dr. P. Chiao, Dr. D. Cromeens, Dr.Y. Shen, et al.
81. Estimate the misclassification probabilities and test for the technician effect associated with the FISH method. Dr. A. Glassman, Dr. Y. Shen, Dr. P.F. Thall.
82. A Phase II clinical trial for preoperative chemoradiation therapy in patients with gastric adenocarcinoma. Dr. P. Mansfield, Dr. Y. Shen, et al.
83. Risk factors for time to death in advanced colon carcinoma patients. Dr. G. Roddey, Dr. Y. Shen, and Dr. F.A. Sinicrope.

CCSG Shared Resource:

Biostatistical Resource Group

Background

The University of Texas M.D. Anderson Cancer Center has been committed to clinical research for over 50 years. Currently, there are over 400 clinical studies in progress which have a range of biostatistical needs, including study design, assessment of feasibility, quality assurance of database management, statistical analysis, and reporting of results. Studies underway encompass pilot, phase I, II, and III clinical trials as well as observational studies of therapeutics, translational and cancer control objectives.

Studies vary in their requirements for biostatistical collaboration. Some require only simple statistical analysis. Most require extensive interactions in the planning and design phase, the conduct and database management phase, and the analysis and reporting phase. To maximize the cost effectiveness of contemporary clinical research, more sophisticated biostatistical methods must be used. Frequently, adaptation and development of biostatistical methods are required to ensure maximal use of data from cancer research. Biostatistical education and training of physicians and basic scientists is essential to high quality clinical and translational research.

An institution-wide Biostatistical Resource Group was established for all clinical investigators for training in biostatistics and for collaboration in clinical research. The Biostatistical Resource Group was organized in 1986 and has been funded as a Shared Resource of the Cancer Center Support ("Core") Grant, CA16672, to The University of Texas M.D. Anderson Cancer Center (Dr. F.F. Becker, Principal Investigator).

Functions

Major functions include:

1. Provision of biostatistical consultation and resources in statistical computing and database management to institutional scientists engaged in the planning, conduct, analysis, interpretation, and reporting of clinical trials and observational studies. Collection of the data accruing on studies is the responsibility of the scientist. Advice is provided on computerized database management and the establishment and quality assurance of appropriate data management procedures.
2. Continue the education and training in biostatistical methods and clinical research. An entire course and a series of lectures is being prepared in the areas of design, conduct, analysis, interpretation, and reporting of clinical research.
3. Continue research on biostatistical methods for clinical research.

The program of the Biostatistical Resource Group thus complements the ongoing activities of the Section of Biostatistics and provides a visible mechanism for enhancing the quantitative research environment at The University of Texas M.D. Anderson Cancer Center.

Publications of the Section of Biostatistics

Table of Contents

Section of Computer Science