Welcome to the BERD Reviews Home page

NOTE: This web is only open to BERD members.

Dear BERD Colleagues,

We respectfully request your feedback on the draft manuscript entitled “Strategies for Developing Biostatistics Resources in Academic Health Centers,” below. In particular, we welcome comments on scope, focus, tone, and potential interest. Our intended audience is senior leadership at Academic Health Centers (e.g. deans, clinical department chairs) as well as principal investigators and senior biostatistics leadership. As we plan to submit this for publication, please do not distribute or publicize the contents at this time.

This manuscript is a product of the Promotion, Tenure, and Professional Development working group within the BERD Evaluation subgroup.

Please provide your name, affiliation and comments in the box below. If you prefer to commet on the document directly, please use the .doc or .pdf versions then send to Leah Welty using the link below. To ensure that the manuscript is published in a timely manner, we request your feedback no later than Tuesday, February 21, 2012. We greatly appreciate your time and effort.

Leah Welty, on behalf of the Promotion, Tenure, and Professional Development working group

Draft: "Biostatistics Units in AHC" on CTSpedia wiki

Draft: "Biostatistics Units in AHC" in Word 2007

Draft: "Biostatistics Units in AHC" in PDF

Please Comment on Paper Below

Please add comments and then click on the "Add comment" button.

%COMMENT{type="belowthreadmode"}%

WendyMack - 20 Feb 2012 - 10:20

Very nice paper detailing major issues related to centralized biostatistics. A couple of points that have been relayed by others but I think cannot be emphasized enough, in particular as this paper is directed to senior AHC leadership --

1. Biostatistical support for "unfunded but promising collaborations", mentoring, training etc. requires direct and sustained institutional support, i.e., representing an ongoing institutional investment (but with ultimately large payoff for the institution).

2. The 10% minimum effort rule is an ideal strategy but difficult to achieve with small grants and very difficult funding as we are now experiencing. I think you should keep this ideal strategy, but suggest alternatives when it is not possible for an investigator to budget at that level.

3. I like the suggestion to include support for methods development. However it must be then emphasized that the investigators' allocation of funding for the biostatistician's efforts in methodological research must be accompanied by a sound budget justification linking the methods development to the grant's specific aims.

4. As this is directed to AHC leadership, some discussion of the "con's" of centralization is warranted, with some well-considered balancing of the strengths of centralization. I would imagine the con's would be more fiscal and logistical in nature.

5. We also support the hourly recharge concept, as it is very efficient for short-term consultations and we also find is a natural lead-in to the development of long-term collaborations that include named effort of both biostatistical faculty and staff. Note a small typo in abstract (should be "retaining biostatisticians"). Also some funny wording on page 8 that needs to be fixed ("they should also want for").

DianneFinkelstein - 16 Feb 2012 - 14:48

Looks good! Dianne

RobertParker - 16 Feb 2012 - 13:23

I have reviewed the document and have the following comments / suggestions, from the perspective of someone who is actually trying to sell such a group within an institution that does not have one.

General comments:

You seem to be arguing that biostatistics should be centralized rather than distributed across an AHC. If so, you should make it clear what the alternative is, and provide a succinct comparison of the two, preferable in a list or table approach, showing why you favor the centralized approach.

You seem to feel that somehow the centralization will allow for all sorts of "nice activities" (e.g., teaching, unfunded but promising collaborative projects, IRB support etc.) but this needs to be supported. You do not mention how this is to happen.

I think that you need to face the reality that such a group requires not only that individuals be (primarily) self-supporting [through collaborations with others] but that the institution needs to be committing resources for (a) start-up of the group and (b) support for "institutional goods", such as assistance in grant preparation, training of physicians, etc. etc.

There also seems to be an implicit assumption that all staff have 20% to pursue their own research interests, and that their collaborators should provide this type of support (e.g., if I'm supported 25% on a core grant, then only 20% work is expected, with the other 5% being time for me to pursue my own research interests). The reality is that investigators supporting a statistician for 25% are far more likely to expect 30% effort than 20% effort. If you are serious about having this as part of the value of such a group then you need to accept that this has to be an overhead paid for by some of the core support of a group -- not by individual investigators.

Some of the advice and suggestions are not feasible, e.g. that individuals should not devote less than 10% effort on any grant. This is probably reasonable advice for R01 clinical trials funded near the $500K/year limit -- where 10% effort + 20% effort for MS / analyst support might be only 6-7% of the annual budget, but totally impractical for most smaller grants (R21, R34, etc.). You need to deal with the reality that in many situation an investigator has two choices: small support or no support, and provide possible solutions for this. One aspect is making it quite clear to the investigator what the little support actually will be (setting appropriate expectations) and that the support will not provide for significant scientific input, and is largely being used to "buy a figurehead". Even 5% buys approximately 75 hours of input. There are several alternatives that spring to mind The best (from the statistician / scientific collaboration) is when multiple bits of small support on related projects can be combined to provide adequate support for the subject area collaboration. Another is to have a consultation service available to supervise the MS level staff and provide general design support, without any collaboration connection -- and that this service be part of the core biostatistics group.

Finally the conclusions appear somewhat overstated:

a) you provide no information about how to actually start such a group, so there is no information about "building" b) you provide no discussion of institutional support, so I do not see how you have material for "supporting such a group".

Overall, I would very much love to see you actually discuss how to build and maintain such a group and compare the basic alternatives:

i) bootstrapping, ie.. when the X current individuals are over commited enough to fully support the X+1

ii) continuing institutional support for Y individuals to be spread out over the X+Y individuals available

with a detailed discussion of the tradeoffs of such an approach.

Starting up: i) some institutional start up funds to set up such a group but no long-term institutional committment to keep it going vs ii) long-term commitment with (potentially) less startup funding.

Specific comments:

I feel rather strongly that if an investigator / department insists on having control over hiring (page 7/8) that they should be allowed veto power over a proposed candidate identified by the biostatistics group [applicable only when they are guaranteeing 100% support for the position] or (b) be cut lose and have no access to the core biostatistics group. Even if there is a centralized group, I would argue against (a) forcing others in the AHC to use the group, and I would equally argue (b) that no group at the AHC has the right to expect the biostatistics group to cater to them. The basic idea one hopes for is to have a partnership. If in investigator / department insists on control the biostatistics group either picks the candidate or they are insulated from the department / investigators demands. (Perhaps not the "collegial" thing to say, but the reality in many organizations -- and such a department does not want a partnership in the first place.)

Infrastructure (pages 11/12): if the costs are not recruited in grants, then how is the infrastructure developed and maintained? This is a theme throughout -- that by having a larger group good things happen, but there is never a consideration of how this will actually be supported.

XuR - 13 Feb 2012 - 15:42

I think it's a wonderful idea to have such a pub out there, and the paper has many many good points. Thank you all for the great efforts that have been put into it. I wonder if the case can be made even stronger by: 1) providing some data, how many AHC's are without a centralized biostat unit (on a minor point, you mentioned on page 5 'those groups that focus primarily on consultation..' but did not seem to come back to this later in the paper); 2) early on, maybe a bit more focus on why the senior leaders should consider a centralized biostat unit in the big picture, since this is the main selling point. The focus itself can be efficient use of the biostat resources, for example.

MimiKim - 09 Feb 2012 - 16:36

We tend to discourage the hourly charge approach at Einstein because it seems to perpetuate the negative stereotype that biostatisticians are service technicians rather than true scientific collaborators. In fact, a cardiologist once said to me after we completed an analysis for him to "just send me a bill for the work the way my auto mechanic does." Furthermore, many junior investigators don't have the resources to pay for services by the hour and have to go to their department heads for funding. If, on the other hand, clinical departments are willing to provide partial FTE support for a statistical collaborators, junior faculty would be more inclined to involve a statistician early in the research process. Long-term FTE support from clinical departments can also be used to hire more biostatistics faculty; it's been one of the main ways we've been able to expand our group. But clearly the fee-for-service model has been very successful at UCSF. Maybe we should think about a separate paper on the pros/cons of different funding mechanisms for biostatistical support!

PeterBacchetti - 09 Feb 2012 - 12:14

Regarding Mimi's comment, at UCSF our biostatistical "consulting" service (somewhat mis-named) has fostered many substantial collaborations, with big impacts on the projects and co-authored papers produced. These were done by hourly charge.

MimiKim - 09 Feb 2012 - 11:50

Excellent and much needed paper! I'm grateful there will now be a reference I can point to whenever the issue of hiring statisticians into different clinical departments comes up at our institution. In the section on "Collaboration versus Consultation" it may be worthwhile to provide specific recommendations for how productive and long-term collaborations can be fostered between biostatisticians and investigators, e.g., partial FTE support for biostatistician from grant or departmental funding as opposed to fee-for-service or hourly charge; including a biostatistician as a co-investigator in grant applications; appropriate acknowledgement for intellectual contribution on papers, i.e., co-authorship, etc...

PeterBacchetti - 08 Feb 2012 - 14:33

Page 10 near top: \x93A doctoral-level biostatistician necessarily possesses data management and programming skills.\x94 I believe that data management is increasingly done by programmers and other specialists who are not statisticians. At UCSF, we have separate services for data management, and many master\x92s-level and faculty statisticians dislike being called on to participate in data management and feel unqualified to do so. So I would add some mention that non-statisticians (not just master\x92s-level statisticians) may be better choices for data management tasks.

PeterBacchetti - 08 Feb 2012 - 14:32

I would argue that the sentence near the top of page 9 is not correct: \x93Clinical trials biostatisticians are able to design studies with the minimum sample size required to answer the clinical question\x94. No particular sample size can accurately be said to be \x93needed to answer the clinical question\x94; this is what I have called the \x93threshold myth\x94 about sample size. A concrete example is the Women\x92s Health Initiative study on dietary fat and breast cancer: Ross Prentice is a good clinical trials biostatistician, but the trial still turned out not to be conclusive by any reasonable standard (although widely misinterpreted to be conclusively negative). I would also on more general grounds recommend against seeming to endorse the idea that single RCTs usually settle a clinical question. Positive results often turn out to be wrong or exaggerated (as shown by Ioannidis), so-called \x93negative\x94 results are very often not conclusive, and meta-analysis of multiple RCTs is now widely thought to play a key role in guiding clinical practice.

MarleneEgger - 07 Feb 2012 - 17:13

This is a lovely, thoughtful, well-expressed article on the care and feeding of academic biostatisticians. Which journal do you have in mind? Perhaps an AHC VP should read it and suggest ways to motivate AHC leadership to want to retain biostatisticians. I think you have covered al the issues, but sometimes the taregt audience can suggest how to say it. For example, depending on the journal, starting with or having a box with a concrete John-and-Mary case study, real or hypothetical, might be effective with the target audience.

Are there any hard data on projects that foundered because the PhD statistician left? Funding agency statistics on projects rejected because of inadequate statistical support? When an agency does not want to fund a proposal, they 'throw the book' at it, including the statistics, but even a reviewer survey on how much biostatistics played a role would be interesting.

It would be interesting to take the concepts you have outlined and develop them into a scale of institutional biostatistics health. Then one could use them to predicting funding and project success.

Not in this paper, though.....I think it is ready, unless you want to ask the target audience about how to sell it to them.

EmiliaBagiella - 07 Feb 2012 - 15:20

Very nice paper. It touches crucial issues for both junior and senior Biostatisticians who collaborate with clinicians. In the last section "Collaboration versus consultation" i would also stress out the importance on the part of the clincians to keep their comitment toward their biostatistical collaborators. Too often happens that a biostatistician is listed as a co-investigator on a grant and he/she is then dropped as the grant is funded or the funding is reduced.

Click here to contact Leah Welty directly

This topic: ReviewsBERD > WebHome
Topic revision: 20 Feb 2012, WendyMack

---