Home » Featured, Science Policy

Statisticians and Clinicians: Collaborations Based on Mutual Respect

1 February 2012 25,411 views 9 Comments
This month’s guest columnist, Don Berry, describes his policies and philosophy for developing collaborations between statisticians and clinicians. Since starting at ASA, I’ve heard about the challenge of statisticians to be engaged early in an experiment or study. I invited Berry to do this piece after hearing about his widely respected efforts to address that challenge. ~Steve Pierson, ASA Director of Science Policy

Contributing Editor
Donald A. Berry is professor of biostatistics at The University of Texas MD Anderson Cancer Center, where, until 2011, he served as head of the division of quantitative sciences and chair of the department of biostatistics.

Collaborative research is about mutual respect. The same principles apply in any relationship, whether business, science, or marriage. You cannot respect me if I don’t respect you.

I moved to MD Anderson in 1999 to found a department of biostatistics. MD Anderson is the biggest cancer center in the United States, with an annual budget of more than $3 billion and more than 1,500 faculty members, a number that has approximately doubled in the last 10 years. More than 10,000 patients per year participate in our clinical research. Our statisticians work with our clinicians to design and run hundreds of clinical trials each year.

The department of biostatistics expanded into a division of quantitative sciences (DQS)—including a department of bioinformatics and computational biology—with more than 40 faculty members and about 45 statistical analysts. Most of our analysts have master’s degrees; some have PhDs. The analysts are organized into about 10 teams, each with a team leader. They report on a line separate from division faculty up to the director of quantitative research, who reports to the division head. However, analysts work closely with division faculty on all projects.

Faculty and analysts in DQS collaborate with faculty and other personnel in every academic department in the institution, of which there are more than 50. Every academic department is assigned a faculty member and analyst team leader in biostatistics. These are initial contacts for clinical and basic scientists in the respective departments.

Our statisticians become specialists in the diseases within which they collaborate. A statistician might be responsible for a single, large department. More typically, they are assigned several departments that focus on related diseases. They learn about the standard treatments, depending on disease and disease subtype. They learn about the biology of the disease, the role of biomarkers, etc. My pet peeve is the statistician who designs a clinical trial by asking for the null rate, clinically important difference, and accrual rate and uses standard software to produce a sample size. Where are the questions about the disease, its standard treatment, its prevalence, and its biology?

We send clear messages to our clinical collaborators that we are as interested in curing cancer as they are. We work as a team. Even though we have tools, we are not mechanics.

As a National Cancer Institute (NCI)–designated comprehensive cancer center, we get a major “core grant” from the NCI. One of the major cores is biostatistics. Another is bioinformatics. These are the only two of the 20-plus cores that do not charge for their ‘services.’ I have always insisted on this policy, and it is supported by our administrations, even though it is not popular with everyone. We are collaborators. Collaborators don’t pay each other to collaborate. However, I encourage joint grants and funding for statistical collaborators on clinical, translational, and basic science grants.

Every protocol initiated within the institution must have a statistical collaborator. And every protocol undergoes review by the biostatistics department. (In view of the large number of protocols presented at the nearly weekly meetings, the statistical reviewer who presents the protocol is limited to three minutes, with discussion being open-ended.) The department develops a consensus review, and this review is communicated to the institution’s review boards (IRBs). Our IRBs have developed great respect for these reviews, in part because they represent the whole department and not a single reviewer. Not incidentally, these review sessions are wonderful learning experiences for students and young faculty.

From the beginning, and supported by our president, my policy was that 50% of faculty time would be devoted to developing methodology and other purely statistical research. This is consistent with a traditional 50:50 division of teaching and research at major research universities and it made us competitive in hiring top faculty. In practice, the 50% statistical research is synergistic with our collaborative cancer research. For example, a statistician works with a clinician in building a design for an actual trial and then they publish the design in a statistics journal, run the trial, and publish the results in a clinical journal. The statistician is the first author on the former and second author on the latter.

A major consequence of our faculty’s statistical research is that we have established an excellent reputation in our discipline. We publish in top statistics journals, as well as in top medical journals, and we obtain research grants in both quarters. Our clinical colleagues know about our strong statistical reputation and they respect us for it. So do reviewers for clinical grants. It happens that we are best known for developing and applying Bayesian methods in cancer research. Perhaps the only relevance of the Bayesian approach as regards the subject of this article is that clinicians relate to it. It’s the way they think. It provides a bond between us. But it is not a necessary component of a healthy collaborative environment.

As much as I avoid using the word “service,” we do provide services to the institution. Very early on, I established a ‘drop-in’ statistical clinic, analogous to a medical clinic. It is staffed by statistical analysts and is ‘open’ three times per week. Anyone in the institution can bring problems to the clinic. There is no charge. Typical services are doing straightforward analyses and providing advice for incorporating statistical analyses into projects and manuscripts. Some projects turn out to be large, requiring more than an hour or so, and are referred to the statistical faculty member or team leader assigned to the particular client’s department.

MD Anderson provides a great environment for promoting collaborations. Its motto is “making cancer history,” something collaborations between our statisticians, clinicians, and basic scientists are helping to do.

My advice to statisticians who want to establish equal partnerships relates to my opening sentence: Foster mutual respect. Respect yourselves, including by establishing a statistical reputation outside of your institution. Be confident, but keep learning, especially about science and medicine. Seek to achieve scientific and medically important goals. Think outside of your box. Listen more than you talk. Don’t interrupt. If you hear rot, guide your collaborators down a path that helps them see the rot without you having to announce it.

1 Star2 Stars3 Stars4 Stars5 Stars (6 votes, average: 4.00 out of 5)


  • Rob Califf said:

    Don, nice job. For most of my career i have worked in an environment as you describe here–the DCRI. The research group was founded on a collaboration of experts in the “holy trinity”– biostatistics, informatics and clinical expertise. It has been interesting to see how different it can be in other areas of research. Your advice in this blog is excellent.

  • Bill B. said:

    Based on hearing Dr. Berry’s talks, it’s too bad he doesn’t afford other statisticians (in particular, non-Bayesians) the same respect he gives clinicians.

  • Dick L said:

    A very nice model for biostatistics collaboration.

  • Jeffrey Morris said:

    Great writeup and summary, Don! I totally agree — mutual respect is crucial to a good collaborative experience.

  • Steve Goodman said:

    There is little doubt that the MD Anderson set-up is the model for how statistical consulting-partnership should be conducted, in no small part due to Don. But Don doesn’t mention perhaps the most important locus of respect, which is that of the institution’s leadership for the contributions of Biostatistics-Bioinformatics. That you were allowed to build a team of 40 faculty members and 45 analysts at MDA is amazing, and critical to the kind of operation you describe above. Even though most Biostat faculty pay for themselves about five minutes after they join, it is the very rare institution that will provide the space or resources to hire adequate numbers. Every institution I have been familiar with bemoans the lack of statistical resources for collaborative research, yet none will add the #s of statisticians needed to address it. Even if they could come close, they would demand that those hired spend 100% time on collaborative / service activities; not a formula for getting the best people.

    So, while everything you say above about mutual respect is true, it’s a lot easier to be an effective partner when you are working on 5 projects in similar areas than 23 in disparate ones, not counting the grant prep, reviewing, SS calcs and informal consults that can fill up a week. So, institutional recognition of the proper sizing of biostat/bioinformatics depts. is a critical concomitant to the success of the professional respect model you outline above. You are lucky to have that at MDA, and I hope others take the #s you outline above to their own institutions. BTW, how many research protocols did you review and collaborate on each year, and roughly how many per faculty+analyst?

  • Diana Miglioretti said:

    Don – Nice blog! Thanks for sharing. And Steve G. has very good points about the challenges most of us face in this area in terms of time and numbers.

  • Karen Messer said:

    A nice model indeed, and helpful to have this appear in a professional journal so that we can wave it under various noses- thank you Dr. Berry. The argument that support for research is necessary in order to attract top-flight faculty is one that resonates with our Dean and other academic leaders. I wonder what are the various models in the biostats community for obtaining institutional support for research time? It would be very useful to have several successful models at hand.

  • Steve Simon said:

    I would like to second Dr. Berry’s comment about the importance of not charging for your services. Collaboration often requires the room to explore new ideas and areas. If you are worried about how much of a limited budget you are spending, you won’t be eager to wander down different paths–you’ll want the fastest answer that will get the job done.