zelen

Marvin Zelen

Lemuel Shattuck Research Professor of Statistical Science
Member of the Faculty of Arts and Sciences (Emeritus)

Department of Biostatistics

Department of Biostatistics

655 Huntington Avenue
Building II Room 443
Boston, Massachusetts 02115
Phone: 617.432.4914

Other Affiliations

Faculty of Arts and Sciences

Research

Marvin Zelen’s current research consists of:

  • Creation of stochastic models for the early detection of disease:

Early detection of chronic diseases has the potential of raising cure rates and lengthening survival. Zelen and colleagues (mainly Dr. Sandra J. Lee) have developed the main theoretical basis for the early detection of chronic diseases principally motivated by problems arising in cancer—especially breast cancer. These theoretical models can help answer questions which empirical studies, such as clinical trials, cannot answer due to feasibility considerations. Among this class of problems are: optimal scheduling of screening examinations (age to begin special exams and screening intervals), estimation of probability of over diagnosis, prediction of mortality benefit without very long term follow-up for clinical trials, assessing how benefit is related to both improvements in therapy and dissemination of screening programs.  Some current research findings relating to public health have shown that (1) biennial mammogram screening may have almost the same benefit as annual screening for breast cancer with half the cost. (These findings are under consideration by the U.S. Task Force on Prevention  who recommends policies on these matters for the U.S.); (2) there is modest mortality benefit for screening women in the 40-49 age group with respect to breast cancer; (3) over diagnosis of breast cancer from screening is not an issue; (e.g. the probability is about 10% that a woman, age 70 who is diagnosed early by screening, dying from other causes before the breast cancer would have become clinical).

  • Clinical trials, randomization and inference

All of the statistical methods, ordinarily used for analyzing multi-center trials, require a random sample of patients. However almost no clinical trials have a random sample of patients.   Patients entering randomized trials can simply be described as a “collection” of patients who have consented to enter a clinical trial. Consequentially, without a random sample of patients (or hospitals), one cannot make a statistical inference about beneficial treatments for the population with disease. New techniques have been developed which only rely on the randomization process as the basis for making an inference with regard to treatments. These new methods do not require a random sample of patients or of hospitals. However the inference using these new methods applies only to the group of patients who have entered the trial; i.e. best treatment for the patients who have entered the trial. This work is joint with Dr. Lu Zheng.