What the NHLBI paper metrics data mean for NIH grant review

In reflecting on the profound lack of association of grant percentile rank with the citations and quantity of the resulting papers, I am struck that it reinforces a point made by YHN about grant review.

I have never been a huge fan of the Approach criterion. Or, more accurately, how it is reviewed in practice. Review of the specific research plan can bog down in many areas. A review is often derailed off into critique of the applicant’s failure to appropriately consider all the alternatives, to engage in disagreement over the prediction of what can only be resolved empirically, to endless ticky-tack kvetching over buffer concentrations, to a desire for exacting specification of each and every control….. I am skeptical. I am skeptical that identifying these things plays any real role in the resulting science. First, because much of the criticism over the specifics of the approach vanish when you consider that the PI is a highly trained scientist who will work out the real science during the conduct of same. Like we all do. For anticipated and unanticipated problems that arise. Second, because there is much of this Approach review that is rightfully the domain of the peer review of scientific manuscripts.

I am particularly unimpressed by the shared delusion that the grant revision process by which the PI “responds appropriately” to the concerns of three reviewers alters the resulting science in a specific way either. Because of the above factors and because the grant is not a contract. The PI can feel free to change her application to meet reviewer comments and then, if funded, go on to do the science exactly how she proposed in the first place. Or, more likely, do the science as dictated by everything that occurs in the field in the years after the original study section critique was offered.

The Approach criterion score is the one that is most correlated with the eventual voted priority score, as we’ve seen in data offered up by the NIH in the past.

I would argue that a lot of the Approach criticism that I don’t like is an attempt to predict the future of the papers. To predict the impact and to predict the relative productivity. Criticism of the Approach often sounds to me like “This won’t be publishable unless they do X…..” or “this won’t be interpretable, unless they do Y instead….” or “nobody will cite this crap result unless they do this instead of that“.

It is a version of the deep motivator of review behavior. An unstated (or sometimes explicit) fear that the project described in the grant will fail, if the PI does not write different things in the application. The presumption is that if the PI does (or did) write the application a little bit differently in terms of the specific experiments and conditions, that all would be well.

So this also says that when Approach is given a congratulatory review, the panel members are predicting that the resulting papers will be of high impact…and plentiful.

The NHLBI data say this is utter nonsense.

Peer review of NIH grants is not good at predicting, within the historical fundable zone of about the top 35% of applications, the productivity and citation impact of the resulting science.

What the NHLBI data cannot address is a more subtle question. The peer review process decides which specific proposals get funded. Which subtopic domains, in what quantity, with which models and approaches… and there is no good way to assess the relative wisdom of this. For example, a grant on heroin may produce the same number of papers and citations as a grant on cocaine. A given program on cocaine using mouse models may produce approximately the same bibliometric outcome as one using humans. Yet the real world functional impact may be very different.

I don’t know how we could determine the “correct” balance but I think we can introspect that peer review can predict topic domain and the research models a lot better than it can predict citations and paper count. In my experience when a grant is on cocaine, the PI tends to spend most of her effort on cocaine, not heroin. When the grant is for human fMRI imaging, it is rare the PI pulls a switcheroo and works on fruit flies. These general research domain issues are a lot more predictable outcome than the impact of the resulting papers, in my estimation.

This leads to the inevitable conclusion that grant peer review should focus on the things that it can affect and not on the things that it cannot. Significance. Aka, “The Big Picture”. Peer review should wrestle over the relative merits of the overall topic domain, the research models and the general space of the experiments. It should de-emphasize the nitpicking of the experimental plan.