Trial phase and sponsor mix across sixteen conditions

Question

How many interventional studies does ClinicalTrials.gov hold for sixteen conditions, how many are tagged Phase 3, and what share carry an industry lead sponsor?

Method

Endpoint: https://clinicaltrials.gov/api/v2/studies, called with python3 urllib and the User-Agent pgx-gap-map-research/1.0 (public data note; python-urllib). Retrieval date 2026-10-30, 14:36 UTC. Cap 40 requests, 0.4 s sleep; forty were used, two failed.

Count-only queries sent pageSize=1&countTotal=true and read totalCount:

• query.cond=<condition>&pageSize=1&countTotal=true&filter.advanced=AREA[StudyType]INTERVENTIONAL
• ...&filter.advanced=AREA[StudyType]INTERVENTIONAL AND AREA[Phase]PHASE3 (and ...AREA[LeadSponsorClass]INDUSTRY)

Conditions (query.cond literals): type 2 diabetes; hypertension; Alzheimer disease; Parkinson disease; schizophrenia; major depressive disorder; tuberculosis; malaria; HIV infections; hepatitis C; breast cancer; lung cancer; pancreatic cancer; sickle cell disease; cystic fibrosis; chronic kidney disease.

A single filter.advanced accepts a boolean expression (type 2 diabetes: 9,910 interventional; 1,647 Phase 3), so the interventional restriction sits inside every query, not only the denominator.

The full matrix did not fit the cap (16 × 6 phases + 8 sponsor classes = 224 queries): Phase 3 was kept for all sixteen and industry for two; other cells are absent, not zero.

Both failures came from /api/v2/stats/field/values, which answers HTTP 400 Invalid prefix in parameter name: query.cond; logged in raw.json under failures.

A 1,000-record page per condition (pageSize=1000&sort=StudyFirstPostDate:asc) was also fetched as a shortcut; those counts sit in data.csv with a _sampled suffix and are not results.

Results

• The sixteen hold 80,880 interventional registrations, 10,984 tagged Phase 3 (13.6% pooled). Totals run from 867 (sickle cell disease) to 13,233 (breast cancer). Sickle cell disease is the only complete census (867 of 867); its counted Phase 3 records, 119, match the API total.
• Phase 3 share runs from 7.4% (pancreatic cancer) to 23.4% (hepatitis C), median 14.9%; ten of the sixteen sit between 11% and 16%.
• Phase tags overlap, so buckets cannot be summed. For chronic kidney disease, Phase 2 returns 754 and Phase 3 returns 684, while the OR query returns 1,363: 75 studies carry both tags and appear in each single-phase count.
• Industry counts exist for two conditions: breast cancer 2,863 of 13,233 (21.6%) and lung cancer 3,962 of 11,458 (34.6%). The archived page put lung cancer at 28.4%, 6.2 points lower, and its Phase 3 shares ran a median 7.9 points high.

What this does and does not show

1. Coverage is partial: five phase values (Early Phase 1, Phase 1, Phase 2, Phase 4, NA) have no condition-level count, and no sponsor class was counted except industry, for two conditions; nothing here speaks to NIH, OTHER, NETWORK, FED, AMBIG, INDIV or UNKNOWN lead sponsors.
2. Phase counts are not mutually exclusive: 75 chronic kidney disease studies carry both Phase 2 and Phase 3. A Phase 3 share is the fraction of a condition's interventional registrations whose Phase field includes PHASE3, not the fraction that are Phase 3 alone.
3. The 1,000-record pages are not a random sample. Sorted by first-post date they take the oldest registrations, overstating Phase 3 share by a median 7.9 points (range −0.5 to +22.3) and miss the lung cancer industry share. They are kept for audit, not used as estimates.
4. query.cond is the registry's match for a literal phrase; synonym or hierarchical expansion was not tested, so a count means "what ClinicalTrials.gov returns for that string", not a case definition.
5. Registrations, not trials conducted: all statuses are included, records are self-reported, and a study appears once however many countries it runs in.
6. Sponsor class is the lead sponsor, not the funder; the API also returns OTHER_GOV, absent from the requested eight-class list.

Reproduce

curl -sG "https://clinicaltrials.gov/api/v2/studies" --data-urlencode "query.cond=type 2 diabetes" \
  --data-urlencode "pageSize=1" --data-urlencode "countTotal=true" \
  --data-urlencode "filter.advanced=AREA[StudyType]INTERVENTIONAL AND AREA[Phase]PHASE3" \
  | python3 -c "import sys,json;print(json.load(sys.stdin)['totalCount'])"

The sweep, 0.4 s apart:

import csv, json, time, urllib.parse, urllib.request
UA = "pgx-gap-map-research/1.0 (public data note; python-urllib)"
CONDS = [r["condition"] for r in csv.DictReader(open("data.csv"))]
for c in CONDS:
    for adv in ("AREA[StudyType]INTERVENTIONAL",
                "AREA[StudyType]INTERVENTIONAL AND AREA[Phase]PHASE3"):
        q = urllib.parse.urlencode({"query.cond": c, "pageSize": 1, "countTotal": "true",
                                    "filter.advanced": adv}, safe="[]():,")
        req = urllib.request.Request("https://clinicaltrials.gov/api/v2/studies?" + q,
                                     headers={"User-Agent": UA})
        print(c, json.load(urllib.request.urlopen(req))["totalCount"])
        time.sleep(0.4)

raw.json stores every response verbatim, keyed by query string, plus the request log; data.csv is rebuilt from those bodies; medians and pooled shares come from its rows.

Publication status

Draft. Auto-generated from public API responses; requires editorial review before publication.