Marvin Zelen's 8 Predictions for the Future of Biostatistical Sciences

Movers & Shapers| Education| Marvin Zelen |

May 14, 2015

Written by Esha Senchaudhuri

Ten years ago, in May 2005, world-renowned biostatistician Marvin Zelen was asked to deliver a keynote address before the Eastern Mediterranean Region of the Biometric Society. His address, entitled ‘Biostatisticians, Biostatistical Science and the Future,’ [1] offered a reflection on the direction which the field at large would have to take, given the anticipated leaps in software and the rise of big data.

This included eight predictions on how short and longer term advances in technology would shape the field.

As we prepare to celebrate the annual Marvin Zelen Leadership Award in Statistical Science, this year awarded to Professor Nan Laird (Harvard University) [2] [3] we can take a moment to re-examine the predictions and insights of Cytel's beloved friend and Board Member.

Where will the field of biostatistical sciences be ten years from now? Thirty years from now? How about one hundred years from now? It is perhaps a guessing game full of imprecise probabilities and chances for Bayesian updating.

A related question is how we should approach the idea of a future for the biostatistical sciences. Professor Zelen chose to approach the question from the perspective of a teacher and educator. How could the field best prepare future biostatistical scientists for the terrain in which they would necessarily find themselves?

By the time Professor Zelen gave his keynote address to the Eastern Mediterranean Region of the Biometric Society, his groundbreaking work in reforming clinical trial methodology, and the application of the statistical sciences to pressing issues in public health, had translated into unprecedented strides in moving the profession forward. His record of achievement, combined with his tireless encouragement and training of younger generations of scientists, perhaps made Professor Zelen uniquely positioned to discuss his chosen topic.

Still, he described his task, of looking into the crystal ball so to speak, as a task which ‘should only be undertaken by those possessing a blend of courage and foolhardiness.’

Professor Zelen began his talk with a brief reflection on the history of the biostatistical sciences – how one hundred years ago the science did not exist, and how in another one hundred years it will have likely evolved into several other disciplines. (Given the recent rise of departments and programs in bioinformatics, computational biology and biophysics, we are possibly bearing witness to the start of this process.)

Professor Zelen also invoked history to defend his use of the phrase ‘statistical scientist.’ The term ‘statistician,’ he pointed out, actually refers to someone who is a practitioner of political arithmetic (i.e. census taking.) The term ‘statistical science’ has historically distinguished the applications of statistical analysis to the natural sciences from their political and social scientific uses.

However, Professor Zelen quickly changed the direction of his talk from history to the future, by launching a long string of proposals regarding how the discipline ought to fashion itself. These proposals are striking, both for their breadth and for their foresight. For example, he begins by proposing that ‘our profession assemble courses on the internet which would be freely available.’

Encouraging his colleagues to tap into online communication devices, make lecture slides available to the general public and assign a particular ‘guide’ whom students outside the university might contact if they had any questions, Professor Zelen was an early advocate of MOOCs (i.e. Massive Open Online Courseware) in the biostatistical sciences.

Another of Professor Zelen's predictions included the rise of automatic data analysis systems, in which a computer program has the ability to select the correct technique of analysis after being fed a question. One such program that has come into being is Cytel's Autopilot for StatXact PROCs.

The predictions, taken together, may be summarized as follows:

There will be a growing need for MOOCs in biostatistical sciences to keep up with the demands of industry and the globalizing nature of the field.
Given the advent of big data, focus will shift from ‘traditional mathematical statistics’ to data analysis for large datasets, such as for purposes of bioinformatics.
There will be greater need for rigorous training in computing and statistical software. (Professor Zelen points out that he had first made this claim in 1983, at a time when several prominent professors disagreed with him.) [4]
We will see the rise of automatic data analysis systems (e.g. a computer program which can be given a question, and which has the ability to choose the correct statistical technique and then engage in analysis.) For an example of this, see Cytel's Autopilot for StatXact PROCs.
Computer programs will be able to inform scientists of methodological inadequacies in their statistical evaluations (e.g. problematic assumptions, etc.)
"Clinical Researchers" will be able to conduct risk-benefit analysis using information on individual genome profiles and the chemical composition of drugs.
There will have to be closer collaboration between industry and academia.
Biostatistical scientists will have to engage with policy makers and take leadership in devising sound public policy.