Close

NIH Data Management & Sharing Plan

Tips for Writing an NIH Data Management & Sharing Plan for Computational Modeling 

In launching an updated Data Management and Sharing Policy in January 2023, the National Institute of Health aimed to increase the reproducibility of scientific research by requiring investigators to proactively plan how they will organize and share their data. The policy requires researchers to submit a detailed Data Management and Sharing Plan (DMSP) along with their funding proposal, and then to comply with the approved plan over the course of the grant performance period. The policy applies broadly, even to projects that may not generate data in a traditional sense (see the official NIH list of applicable activity codes and exceptions (PDF). This category includes many modeling projects such as those produced in the MIDAS Network.


The MIDAS Network Coordination Center supports the NIH’s goal of making research more transparent, reproducible, and reusable. Yet computational modelers may feel like they are trying to fit a square peg into a round hole when describing their research products for a Data Management and Sharing Plan. Many infectious disease models produce predicted trajectories of outbreaks, where results produced by a model depend on the assumptions, arguments, and pre-existing data fed into it. Although the NIH policy does not clearly identify any of those components as scientific data, broad sharing of modeling results is consistent with the spirit of openness implicit in NIH data sharing initiatives.


With that ambiguity in mind, here are some recommendations for writing the plan to satisfy both the spirit and the letter of the new policy. (All references to specific element numbers refer to the NIH’s list of Elements to Include in a Data Management and Sharing Plan).

Regarding what data must be described in the plan, the NIH defines scientific data as “data commonly accepted in the scientific community as of sufficient quality to validate and replicate research findings, regardless of whether the data are used to support scholarly publications” (NIH, “Research Covered Under the Data Management & Sharing Policy.”) This does not need to be data generated by the individual researcher, as investigators are asked to summarize the types of data “to be generated and/or used” in Element 1, “Data type.” Therefore, researchers should look carefully at any data output by a model or used for creating, training, or testing, and ask themselves the following questions: 

  • Are these data necessary for validating and replicating your research finding, e.g. that the proposed model is a useful tool for predicting or analyzing a phenomenon of interest? For example, if you use case counts in a given area to predict hospital admission numbers, both the initial case counts and the output hospital admission numbers could be used to validate your model, if only as a sample dataset. Similarly, the results of your modeling efforts might be necessary for validating and replicating results. Include these data in your data type discussion, even if you do not generate all of them yourself. 
  • Are the data already publicly available in the same form in which you will use them? If so, you do not need to preserve and share the data again, but be sure to state where you will obtain the data and who is responsible for maintaining them. If you plan on expanding or significantly transforming the data to a point at which they constitute a new dataset, you should preserve and share the new data. 
  • Some modeling-related projects like U24 data coordination centers or educational programs require a Data Management and Sharing Plan but do not generate or use data even in the ways described above. If your project falls into this category, are there any outcomes-related data that you will collect, like performance metrics, student surveys, or usage statistics? These should be described in Element 1. 

If your modeling project uses data produced and shared by another source, it’s important to identify those datasets in Element 1 as discussed above. However, many of the sections in Element 4, “Data preservation, access, and associated timelines,” and Element 5, “Access, Distribution, or Reuse Considerations,” will not apply, as you do not control access to the original data. Write something to the effect of “Data preservation and access is controlled by the original data creators” in the appropriate sections so that the busy program officer who is looking over your plan does not lose track of where your data will come from. 

Even when there’s no data to share in a modeling project, the code used to construct or run the model can be distributed. This isn’t strictly scientific data under the NIH’s definition; however, it’s important to share for the reproducibility of computational modeling research. Discuss your plans to share your code both in Element 2, which is all about computational tools and code, and in Element 4, where you would otherwise discuss sharing your data. 

 

As for where to share: putting your code in GitHub is a start, but GitHub isn’t built to maintain public access to your code in perpetuity. (What if Microsoft decides to shut it down or start charging for access?) You can archive a snapshot of your GitHub repository in Zenodo, a public research repository maintained by CERN that meets NIH’s checklist of desirable characteristics of a data repository. This will make your code far more discoverable as a research object than it would be in GitHub alone, and you’ll receive a DOI that will make it formally citable in publications. 

Element 3, “Standards,” asks researchers to identify what standards will be applied to the scientific data and metadata. The NIH gives relatively little guidance on what standards might be used, but a review of program officer comments on submitted plans from one MIDAS-affiliated institution shows that program officers want to see details in this element. To address this element, consider: 

  • Will the data that you use or generate be organized according to any particular data formatting standard, or use any sort of standardized terminology such as SNOWMED-CT?
  • Will the data that you use or generate include any Common Data Elements (CDEs), which make data harmonization easier? 
  • Will you share your data or code in a repository that has standardized metadata? This information can usually be found in a repository’s documentation or policy pages. 


If none of this applies and you are using data that you will not preserve or share because it is managed by a third party, state this clearly in your response for Element 3. 

DMPTool is a free resource created by data librarians and research administrators to offer customized templates for a variety of data management and sharing plans, including that required by the NIH.

 

To use a template, log into DMPTool with your institutional email address or sign up for a new account, and then create a new plan with the NIH as your chosen funder. You’ll see a fill-in-the-blank interface with guidance from the NIH and from DMPTool curators, with sample answers to illustrate potential strategies for addressing each element. When you’re done, export your plan as a PDF to attach to your proposal before submission.