Munim Technologies

Research/Medicine

Who in NIH's dbGaP can be studied, and under which consent

A census of public dbGaP metadata finds that 55.5% of European-ancestry enrolments sit in studies with a general-research-use consent group, against 43.4% of African American and as little as 31.7% of Latin American enrolments.

Oct 14, 2026 · 6 min read

NIH's database of Genotypes and Phenotypes (dbGaP) holds thousands of genetic studies, but a researcher can only use the participants whose consent covers the project. We crossed dbGaP's computed ancestry counts with each study's consent groups across all 3,612 public studies. 55.5% of European-ancestry enrolments sit in studies with a general research use (GRU) consent group, against 43.4% of African American enrolments and 31.7% and 43.7% under dbGaP's two Latin American ancestry codes. A researcher whose project fits only general research use can reach a smaller share of these groups than of European-ancestry participants. Most of the gap closes under the next-broadest consent, health, medical and biomedical use.

3,612

public dbGaP studies harvested on 2026-10-14

742

studies that report dbGaP's computed ancestry

70.8%

of ancestry-counted enrolments are European ancestry

85.4%

of studies with consent groups allow for-profit use in at least one group

The question

Diversity in genomic research is usually measured by who was enrolled. But enrolment is not access. Each dbGaP study splits its participants into consent groups, such as general research use (GRU), health, medical and biomedical research (HMB), or disease-specific use (DS), often with a modifier that rules out for-profit use (NPU). Whether a participant's data can serve a given project depends on those groups. dbGaP shows computed ancestry one study at a time. The closest earlier work mapped dbGaP consent groups to a standard ontology from a 2021 snapshot but did not cross them with ancestry. We found no corpus-wide table of ancestry by consent class.

What we did

We harvested every public study from dbGaP's open FHIR API, along with public data dictionaries, using only public metadata and no controlled-access data. Of 3,612 studies, 3,611 report consent groups and 742 report dbGaP's computed (GRAF-pop) ancestry counts. Each enrolment was counted under the most open consent group of its study. For each ancestry group we then asked what share of its enrolments sit in studies with a GRU group, with at least an HMB group, and with a group that allows for-profit use.

Across all studies, general research use is the most common most-open consent: 2,069 studies have a GRU group. But 776 studies go no further than HMB, 642 no further than disease-specific use and 124 offer only other restricted uses.

Studies by their most open consent group

3,611 public studies that report consent groups. No study reported an NRES (no restrictions) group. Source: census/2026-10/census.md

What we found

European-ancestry enrolments make up 70.8% of all enrolments with computed ancestry. Their GRU share is 55.5%. African American (AFA) enrolments, the second-largest group, reach 43.4%. The two Latin American codes are lower still: 43.7% for LAC and 31.7% for LEN. Not every non-European group is behind. East Asian (56.9%), African (AFO, 63.3%) and other Asian (OAS, 65.6%) enrolments are at or above the European share.

Share of each ancestry group's enrolments in studies with a general-research-use consent group

dbGaP computed (GRAF-pop) ancestry codes. Study-level upper bounds: each study is counted under its most open consent group. Source: census/2026-10/census.md

The picture changes one step down. Under HMB or more open consent, African American enrolments reach 76.5% and LAC 75.7%, about the same as European (75.1%). So the gap is mostly about projects that fit only general research use, such as methods work or studies outside health. For LEN and East Asian enrolments, though, about a third sit in studies whose most open group is disease-specific or narrower, so even an HMB project reaches only 67.0% and 67.2%. For-profit use is possible for 70% to 77% of most groups' enrolments, but only 58.5% of other Asian enrolments.

Computed ancestry by how openly the study can be used

AncestryEnrolmentsShare of allGRUHMB or more openFor-profit allowed
EUR1,717,43570.8%55.5%75.1%74.3%
AFA334,59213.8%43.4%76.5%74.5%
LEN111,4524.6%31.7%67.0%76.7%
EAS90,6033.7%56.9%67.2%70.4%
LAC52,1482.1%43.7%75.7%75.5%
AFO44,1061.8%63.3%78.5%84.7%
OTR28,3361.2%48.5%71.9%76.5%
OAS27,5841.1%65.6%78.1%58.5%
SAS20,8580.9%50.7%74.8%71.6%
Enrolments, not people: a person in several studies is counted once per study. Source: census/2026-10/census.md.

What it means

Counting who is enrolled overstates how diverse the usable data are for some kinds of research. The gap is real but bounded: it mostly affects general-research-use projects, and it is concentrated in African American and Latin American enrolments. For people designing consent forms for new cohorts, the table suggests where narrower consent options have quietly reduced what later researchers can do with participants' data. For researchers planning a data access request, it is a reason to check consent groups as carefully as sample sizes.

The census is a by-product of an open search tool built from the same public metadata. It ranks dbGaP studies for a plain-English feasibility question and serves as a baseline for NLM's SPARK dbGaP Task 2 challenge. We do not report its development scores here: they are measured on labels we built ourselves from the same metadata the ranker reads, and they do not predict the official challenge score.

More research