NIH Real-World Evidence Innovation Challenge 2027: Up to $6 Million for Reproducible Clinical Analytics

NIH Real-World Evidence Innovation Challenge 2027: Up to $6 Million for Reproducible Clinical Analytics

The National Institutes of Health, through the National Center for Advancing Translational Sciences (NCATS), has opened a prize challenge for analytic methods that can produce dependable results across different health systems. Its central question is practical: when a model, causal method, phenotyping algorithm, or other workflow is moved from one clinical data environment to another, does it still work in a reproducible and interpretable way?

The NIH Real-World Evidence (RWE) Innovation Challenge has two phases. Phase 1 asks eligible participants to describe and document a fully specified method that could be tested in the National Clinical Cohort Collaborative (N3C) secure environment. Selected teams then move to Phase 2, where NIH evaluates locked analytic workflows across multiple independent sites and pooled datasets. The challenge is aimed at validation and transportability. It is not simply a competition to report the best score on one development dataset.

Registration is due by February 2, 2027, at 5:00 p.m. Eastern Time. The Phase 1 package is due separately by March 26, 2027, at 5:00 p.m. Eastern Time. NIH lists a total prize purse of up to $6 million, including as many as twelve Phase 1 awards of $50,000 and possible Phase 2 awards of up to $1 million each. These are maximum planned amounts, not guaranteed awards; prize counts and amounts may change, and NIH may decline to award prizes if submissions do not meet the requirements.

Key details

What the challenge is designed to reward

Clinical research teams increasingly have access to harmonized health record data from multiple institutions. Yet a common data model does not guarantee that an analytic method behaves the same way everywhere. Coding practices, patient populations, clinical workflows, measurement timing, missing data, and treatment patterns can differ from one site to another. A model can appear successful in its development setting and still be poorly calibrated or unstable elsewhere.

This competition asks participants to make those limitations testable. NIH describes reproducibility, transportability, and stability as central measures of reliable real-world evidence. Methods may include predictive models, causal inference, phenotyping algorithms, survival or time-to-event models, risk scores, or other clinical and translational analytic workflows. Methods may draw on conventional statistics, machine learning, artificial intelligence, propensity scores, inverse probability weighting, target trial emulation, instrumental variables, or a combination of approaches.

The challenge prioritizes a meaningful clinical question and a credible evaluation plan over novelty by itself. Participants need to explain what question the method answers, how the inputs and outputs are defined, what assumptions it relies on, and how it could be run consistently in a secure multi-site setting. A polished concept without a reproducible implementation plan is unlikely to be responsive.

NIH names cancer, kidney disease, and pediatrics as high-priority areas, while stating that the competition is open to other clinically meaningful use cases that satisfy the rules. Each entry must address one use case or priority area. A participant may submit more than one entry, but each needs its own complete package.

Who can participate and who can receive prize money

Participants may register as an individual, a group of individuals, or an entity. The eligibility rules distinguish participation from eligibility to receive cash. Individual participants, including a team lead, must be U.S. citizens or permanent residents to receive a monetary prize. Non-U.S. citizens and non-permanent residents may contribute as team members when the team otherwise qualifies, but they cannot receive any portion of a cash prize. For a private entity, the entity must be incorporated in the United States and maintain its primary place of business there.

Every participant must be at least 18 years old at the time of submission, register under the challenge rules, and comply with the announcement. Federal entities and federal employees acting within their employment are excluded. HHS employees acting in a personal capacity are also excluded. Employees of other federal agencies should consult their agency ethics officials about whether prize acceptance is permitted. Challenge judges, certain people involved in designing or delivering the competition, and specified immediate family members are also barred.

Teams should confirm their intended structure before registration. A group must identify one Team Lead who handles registration, submissions, and sponsor communications. An entity must designate a point of contact; the cash prize, if won, is paid to the entity rather than that contact. The registration form also asks for a scientific lead, if different from the point of contact. The person listed as the lead matters if the group seeks NIH Early Stage Investigator recognition.

Federal award recipients may participate. However, they may only use funds from a grant, cooperative agreement, or other transaction award to develop or support a submission when that use is permitted by the award’s purpose, terms, and conditions. If a participant intends to use federal award funds, the registration must be made as an entity on behalf of the awardee organization. Teams should resolve this with their institutional grants office before committing funds or submitting.

How the two phases work

Phase 1 is a methods and readiness review. Participants provide a precise analytic approach that can be evaluated in a multi-site environment. NIH says Phase 1 does not require teams to have demonstrated performance inside N3C already. It does require enough evidence and implementation detail for reviewers to judge methodological quality, internal validation, reproducibility, and feasibility for secure execution. Up to twelve participants may be selected to advance.

Phase 2 is the validation and benchmarking phase. Only Phase 1 selectees are invited. Finalists are expected to deploy a locked, containerized workflow inside the N3C secure enclave. Evaluation occurs across individual health systems and progressively pooled datasets, including predefined permutations of which sites are included. Predictive methods may be assessed for discrimination, calibration, subgroup stability, site-level consistency, and performance degradation as data are pooled. Causal methods may be assessed for consistency and robustness of estimated effects across sites and datasets.

Training and evaluation data partitions are separated. A method is trained or optimized in a predefined training partition, then evaluated in separate site-level and pooled partitions. Teams cannot iteratively tune the workflow against evaluation data or retrain it after seeing those results. The challenge therefore rewards a defensible approach to generalization and a workflow that can be locked before evaluation.

NIH expects Phase 2 entrants to provide all components needed for execution, such as code, trained artifacts or training scripts, dependency manifests, configuration, feature generation, final cohort and outcome specifications, training and inference scripts, input and output schemas, and documentation. External APIs, external data access, and site-specific manual changes are incompatible with the described evaluation conditions.

Application and registration timeline

The process has two early deadlines, and registration is a separate prerequisite rather than a substitute for the submission. The official announcement gives this sequence:

NIH expects registration forms and submissions to be obtained and sent by email, not through the NIH challenge webpage. The announcement gives separate addresses for registration and submission. Check the live official page for the current forms and exact submission instructions before sending. Late or incomplete submissions may be treated as non-responsive. The assistance webinar is optional; joining it is not listed as an eligibility condition.

What to prepare for Phase 1

The Phase 1 package is a PDF and has a maximum 12-page narrative, excluding the cover page and references. The announcement specifies U.S. letter page size, margins of at least one inch, font size of at least 11 points in Arial, Georgia, Helvetica, or Palatino Linotype, and line spacing of at least 1.0. The submission must be in English. Participants may not use HHS, NIH, or NCATS logos or official seals, and may not suggest federal endorsement.

The cover page identifies the challenge and entry, participant type, contacts, team members and affiliations, use case, analytic type, and agreement to the rules. The executive summary must explain the question, method, clinical or translational relevance, and why multi-site validation is appropriate. It must end with a two- or three-sentence plain-language summary about the public-health relevance.

The core narrative should define the clinical question, cohort, inclusion and exclusion criteria, outcome, model or analytic framework, assumptions, intended use, and expected performance across health systems. Predictive submissions should define the prediction target, time window, candidate predictors, model structure, and task-relevant metrics. Causal submissions should make the exposure or intervention, comparator, outcome, estimand, assumptions, and confounding adjustment plan explicit.

Teams must define inputs and explain how features and covariates are derived from real-world data. They should show why those data elements are likely to exist across health systems, describe alignment to a common data model such as OMOP where applicable, and anticipate missingness, measurement timing, coding variation, and workflow differences. This is a good place to expose limitations instead of assuming harmonization removes them.

The package also includes internal validation evidence and a reproducibility package. NIH lists cross-validation, temporal validation, independent datasets where available, simulation, and sensitivity analyses as possible validation approaches. Predictive work should select metrics suited to the task, potentially including AUROC, AUPRC, calibration slope or intercept, Brier score, sensitivity, specificity, or positive predictive value. Causal work should provide evidence such as covariate balance, positivity, sensitivity analyses, uncertainty estimates, and robustness checks. In either case, the narrative should explain bias risks and steps taken to address overfitting, leakage, or confounding.

Reproducibility material should include runnable code or pseudocode, software and package dependencies, preprocessing and feature-construction steps, training and inference procedures, random seeds where relevant, and instructions sufficient for an independent evaluator. The challenge encourages code that can be containerized. A fully operational container is not required for Phase 1 unless final instructions add that requirement. The disclosure appendix should explain technical choices, model selection and tuning, dependencies, and any proprietary components or licensing restrictions. NIH says participants retain other intellectual property rights, but must provide enough information for independent review and secure validation.

How to make a competitive, credible entry

Start by choosing one question narrow enough to specify completely and important enough to matter clinically or translationally. The intended use should be understandable: who would use the output, what decision or evidence question it informs, and what the method is not designed to do. A broad claim such as “improve patient outcomes with AI” does not tell reviewers what cohort, outcome, horizon, or validation should be assessed.

Design the data representation with portability in mind. For every feature, state its source, definition, timing, and transformation. Mark which data are essential and which are optional. Explain how the method will behave if an input is missing, measured on a different schedule, or encoded differently. If a variable depends on a local workflow or proprietary field, either justify its availability or redesign the method around broadly present data.

Build an internal validation plan that answers the actual question. Explain the split strategy, prevent information leakage between training and validation, select metrics before reviewing results, and report uncertainty. Avoid presenting only a headline accuracy number. For prediction, calibration and subgroup stability may matter alongside discrimination. For causal analyses, explain the estimand, assumptions, confounding adjustment, overlap or positivity, and sensitivity to plausible unmeasured bias. A modest but interpretable result with a sound design is more useful evidence than an inflated score with unclear validation.

Treat reproducibility as part of the scientific argument. Make the workflow understandable to someone who did not build it. Document the environment, dependencies, parameters, preprocessing, and execution order. Identify any network calls, external data, manual steps, or licensed components early. The challenge’s secure enclave setting makes these operational details central to feasibility.

Finally, address how failure will be detected. State which sites or subgroups might expose a weakness, what kind of degradation would matter, and how results should be interpreted if the method is unstable. The official review criteria allocate substantial weight to methodological rigor, reproducibility, validation evidence, multi-site feasibility, and generalizable impact. The proposal should make a direct case in each area rather than relying on a novelty claim.

Common mistakes to avoid

• Confusing registration and submission deadlines. A team that registers by February 2 still has to send the full package by March 26.
• Writing a model-development pitch instead of a validation-ready method. NIH says proposals focused only on model development, platform building, clinical deployment, or policy without a testable analytic method are non-responsive.
• Relying on site-specific inputs. Proprietary or local variables, inaccessible external services, or manual site-by-site modifications undermine the stated multi-site purpose.
• Leaving assumptions implicit. Define cohorts, outcomes, estimands, exclusions, time windows, and data transformations so an independent reviewer can understand the method.
• Reporting performance without credible validation. Explain the split design, leakage protections, uncertainty, limitations, and task-specific metrics.
• Treating the total purse as guaranteed funding. Prize amounts are conditional. NIH may adjust awards or make no award if submissions do not meet its standards.
• Ignoring team and payment eligibility. Verify citizenship or entity status, age, federal ethics rules, and the role of each team member before registration.
• Submitting an incomplete or incorrectly formatted PDF. Check page limits, fonts, margins, English language, logos, required cover-page wording, and separate email submission instructions.

Frequently asked questions

Does Phase 1 require access to N3C?

The announcement says Phase 1 does not require demonstrated multi-site performance inside N3C. It does require a method that is specified and feasible for later structured validation. Phase 2 finalists will conduct evaluation in the N3C secure enclave.

Can an international researcher join?

Yes, an international participant may be part of a team that otherwise meets the rules. That participant is not eligible to receive any share of a cash prize. Individual prize recipients and team leads must meet the U.S. citizenship or permanent-residency rule; an entity must satisfy the U.S. incorporation and primary-place-of-business conditions.

Is the challenge limited to cancer, kidney disease, and pediatrics?

No. NIH identifies these as high-priority areas but says clinically meaningful real-world evidence questions in other areas may be considered if they meet the challenge requirements. Entries should still clearly identify one priority area or use case.

Must a Phase 1 submission include a Docker container?

The Phase 1 materials should explain containerization and secure-execution readiness. NIH encourages container-ready code but says a fully operational container is not required in Phase 1 unless the final instructions say otherwise. Phase 2 finalists should expect to provide a locked, containerized workflow.

Does the prize require giving NIH ownership of the underlying intellectual property?

The announcement says participants retain other intellectual-property rights and are not required to transfer ownership. However, prize recipients grant NIH a nonexclusive worldwide license for certain public uses of the participant or entity name, title, and executive summary, including the plain-language summary. Teams should review the official terms for the complete license and other participation conditions.

Where do participants submit?

The announcement specifies email for both registration and Phase 1 submissions, using different addresses. The NIH challenge page provides the current forms and submission contacts. Do not assume that registering on a web portal or emailing the wrong address completes either step.

Official source and next steps

Read the official NIH challenge announcement before deciding to participate. It contains the operative rules, exact email addresses, registration form, submission instructions, prize conditions, and updates. NIH also describes an optional February 16, 2027 technical assistance webinar for registrants; questions for that session are encouraged by February 9.

Before registering, settle the participant structure and cash-prize eligibility. Then select a single clinically meaningful question, check the method against the Phase 1 criteria, and draft the package from the required components. Build in time for scientific, statistical, data-engineering, and institutional review. Confirm the two deadlines, the contact addresses, and any changes on the official page before sending materials.