By Oleksandr Gamaniuk, Hipa.ai

52 open datasets: US clinical trials recruiting now

At Hipa.ai we help people find US clinical trials that are enrolling now. To do that we keep a clean, filtered copy of ClinicalTrials.gov: only the studies that are actually open, with the US sites that are still taking patients. Today we are publishing that data as open datasets on Hugging Face: one nationwide dataset and 51 regional ones, for every US state plus the District of Columbia.

What's in them

The nationwide dataset has 27,165 trials and 157,706 trial × site rows across 3,330 US cities. The snapshot was taken on November 18, 2026 from an AACT import dated November 13, 2026.

A trial is included when its overall status is RECRUITING (21,921 trials), NOT_YET_RECRUITING (3,248) or ENROLLING_BY_INVITATION (1,996), and it has at least one open site in the United States. About four in five are interventional studies (21,958); the rest (5,207) are observational.

Each dataset has two CSV files and two configs:

  • trials: one row per trial
  • sites: one row per trial × open US site, which joins to trials on nct_id

Why one dataset per state

Most questions people ask about trials are local. A cancer center in Ohio, a health reporter in Arizona or a student building a demo for Colorado does not want to download 56 MB and filter it. Each state dataset holds every active trial with at least one open site in that state, and only that state's site rows. Sizes run from Wyoming (50 trials, 82 site rows) to California (6,379 trials, 16,189 site rows).

One detail to keep in mind: in state files, us_site_count and states still describe a trial's nationwide US footprint, so you can tell a single-site local study apart from a 300-site national one.

Schema

The main columns of trials:

Column What it holds
nct_id ClinicalTrials.gov ID (primary key)
brief_title, official_title Study titles
overall_status RECRUITING, NOT_YET_RECRUITING, ENROLLING_BY_INVITATION
phase, study_type e.g. PHASE2; INTERVENTIONAL / OBSERVATIONAL
conditions, interventions Pipe-separated (`\
{% raw %}sponsor, sponsor_class Lead sponsor and its class (INDUSTRY, NIH, FED, OTHER…)
enrollment, start_date, primary_completion_date Size and timing
min_age, max_age, sex, healthy_volunteers Basic eligibility
us_site_count, states Nationwide open US footprint
brief_summary Registry summary, trimmed to ~1,000 characters
hipa_url, clinicaltrials_gov_url Links to the full record

And sites: nct_id, facility_name, city, state, zip, country, site_status, latitude, longitude, hipa_url.

How we built them

  1. Source. ClinicalTrials.gov, read through AACT, the Clinical Trials Transformation Initiative's relational copy of the registry.
  2. Active trials only. We keep the three enrolling statuses listed above.
  3. Open US sites only. A site counts as open unless its own status is COMPLETED, WITHDRAWN, SUSPENDED, TERMINATED or ACTIVE_NOT_RECRUITING. Sites with no status on the registry count as open. Trials with no open US site are dropped.
  4. No personal contact data. Names, email addresses and phone numbers of study contacts, investigators and site coordinators are excluded. Emails and phone numbers that appear inside free text (titles, summaries, sponsor names) are replaced with [email removed] and [phone removed]. To reach a study team, follow hipa_url or clinicaltrials_gov_url.
  5. Normalization. Whitespace is normalized, list fields use |, ZIP codes stay text so leading zeros survive, and dates are ISO YYYY-MM-DD.

We regenerate the datasets periodically; the snapshot date is in each README.

What the data shows

All numbers below come from the nationwide trials.csv and sites.csv of the November 18 snapshot.

1. California, Texas and New York lead, but every state has trials

Top 15 states by active clinical trials

California has 6,379 active trials, Texas 5,400, New York 4,906, Florida 4,234 and Pennsylvania 3,481. Massachusetts (3,466) and Maryland (2,754) rank far above their population share, which reflects their concentration of academic medical centers and federal research. A trial with sites in several states counts once in each, so these numbers add up to more than 27,165. On Hipa.ai the same per-state view is a page per state, for example hipa.ai/paid-clinical-trials-ca.

2. Cancer dominates the condition list, followed by obesity

Top 15 conditions in US trials recruiting now

We split conditions on |, merged case and possessive variants ("Alzheimer's Disease" and "Alzheimer Disease"), and counted each trial once per condition. Breast cancer leads with 463 trials, then obesity (331), prostate cancer (328), the generic label "Cancer" (312), depression (276) and stroke (255). Seven of the top 15 are cancers. Conditions are free text as registered, so "Lung Cancer" (163) and "Non-Small Cell Lung Cancer" (134) are still separate entries; mapping them to MeSH terms is a good first cleanup task. The label "Healthy" (239 trials) is left out of the chart because it marks healthy-volunteer studies, not a disease.

3. Most active trials have no drug phase

Phase mix of US trials recruiting now

41% of trials (11,107) are interventional studies with phase "NA": devices, behavioral programs, procedures and diagnostics. Another 19% (5,207) are observational. Among drug studies, Phase 2 is the largest group (3,619), followed by Phase 1 (2,324) and Phase 3 (1,544).

4. Industry runs one trial in five, but almost half of all sites

Who sponsors US trials recruiting now

By lead sponsor class, 72.4% of trials (19,678) are led by academic centers, hospitals and nonprofits (class OTHER), 21.2% (5,754) by industry, 3.4% (914) by NIH and 2.1% (565) by other federal agencies. Footprint tells a different story: industry trials have a median of 5 open US sites against 1 for academic ones, so they account for 46% of all site rows. Industry also leads 1,046 of the 1,544 Phase 3 trials (68%). Overall, two thirds of active trials (67%) recruit at a single US site, while 455 run at 50 sites or more.

5. About one trial in four accepts healthy volunteers

6,252 trials (23.0%) state that they accept healthy volunteers; 20,598 do not and 315 don't say. Among interventional trials the share is 21.3%. Eligibility by sex: 24,285 trials take all sexes, 2,092 are female-only and 788 male-only.

6. A handful of cities host thousands of trials

Counting distinct trials with an open site in each city: New York has 3,424, Houston 3,081, Boston 2,887, Chicago 2,160, Los Angeles 2,001 and Philadelphia 1,977. Below those, the long tail is wide: 3,330 cities have at least one open site.

Load it

With the datasets library (configs trials and sites):

from datasets import load_dataset trials = load_dataset("hipa-ai/us-clinical-trials-nationwide", "trials", split="train") sites = load_dataset("hipa-ai/us-clinical-trials-nationwide", "sites", split="train") print(trials[0]["nct_id"], trials[0]["brief_title"]) # One state only ca = load_dataset("hipa-ai/clinical-trials-california", "trials", split="train") 

With pandas, through the hf:// path (needs huggingface_hub installed):

import pandas as pd base = "hf://datasets/hipa-ai/us-clinical-trials-nationwide/" trials = pd.read_csv(base + "trials.csv", dtype={"nct_id": str}) sites = pd.read_csv(base + "sites.csv", dtype={"zip": str}) # Top conditions, one count per trial cond = (trials.assign(c=trials["conditions"].str.split("|")) .explode("c")["c"].str.strip().str.lower()) print(cond.value_counts().head(10)) # Recruiting Phase 3 trials with a site in Houston houston = sites.loc[sites["city"].eq("Houston"), "nct_id"].unique() p3 = trials[trials["nct_id"].isin(houston) & trials["phase"].eq("PHASE3")] print(len(p3)) 

Ideas for using them

  • Trial matching. Combine conditions, age range, sex and site coordinates to match a patient profile to nearby studies. This is the problem Hipa.ai works on, and the Space shows a simple version.
  • Eligibility NLP. Titles and summaries are a ready corpus for condition normalization (to MeSH or SNOMED), intervention extraction or plain-language rewriting.
  • Geo analysis. latitude and longitude on each site row let you map trial deserts, compute driving distance to the nearest open site, or compare rural and urban access.
  • Recruitment research. Compare sponsor classes, site counts, enrollment targets and start dates across states or conditions.
  • Retrieval and QA benchmarks. Questions like "Which Phase 2 obesity trials recruit in Ohio?" have exact answers in the data.

License and attribution

The data comes from ClinicalTrials.gov, a service of the U.S. National Library of Medicine (NLM) at the National Institutes of Health, obtained through AACT. Reuse is subject to the ClinicalTrials.gov Terms and Conditions. NLM asks re-users to credit the source and give the retrieval date; the snapshot and source dates are in every README. The listing of a study does not mean it has been evaluated by the U.S. Federal Government.

The datasets were compiled, filtered and formatted by Hipa.ai. If you use them, please cite ClinicalTrials.gov and link to hipa.ai.

These datasets are for research and information. They are not medical advice, and only a study team can confirm whether someone qualifies for a trial.

Feedback

If you find an error, want another field, or build something with the data, open a discussion on the dataset page. To search the same trials by condition and city, with a plain-language summary and every US site for each study, visit hipa.ai.