Clinical Trial Results Table QA

Check normalized clinical-trial exports before a join, analysis or buyer delivery. Supply two CSV strings; receive a deterministic audit with missing fields, invalid NCT identifiers, duplicate keys, orphan outcomes and malformed result dates. No AI key, web scraping or source credentials required.

Input contract

• studiesCsv: trial_id,condition,phase; one row per study.
• outcomesCsv: trial_id,outcome_id,outcome_type,primary_endpoint,result_status,results_first_post_date; one row per normalized outcome.
• Identifiers must match NCT followed by eight digits. Outcomes join exactly to studies on trial_id. Outcome identity is the compound (trial_id,outcome_id).
• Outcome type: PRIMARY, SECONDARY, or OTHER_PRE_SPECIFIED. Despite its name, primary_endpoint is the endpoint text column for every outcome type in this export contract.
• Date: a valid full YYYY-MM-DD date. Partial dates need explicit upstream handling, never guessed here.
• Phase and result status are checked for presence, not assigned clinical meaning. Use this profile only for posted-results exports; it intentionally fails incomplete result rows.
• UTF-8, comma delimiter, standard quoted CSV, optional BOM. Whitespace around cell values is ignored for checks. Additional columns are allowed and not interpreted. Headers are exact and case-sensitive.

This is a normalized two-table contract, not a native ClinicalTrials.gov JSON/CSV parser. Map source columns explicitly before running. The Actor does not fetch data, repair it, reinterpret outcomes or verify trial existence.

Output

The default dataset contains one audit; the default key-value store has REPORT. report.schema.json defines the JSON result. PASS means every implemented structural check passed. FAIL is a completed audit with data defects; inspect issueCounts and issues. Invalid configuration, malformed CSV or exceeded limits fail the Actor run.

Findings contain table name, one-based data-record number, field name and issue code, never cell values. Coverage is the non-empty fraction for each required column. Hashes bind the exact submitted CSV bytes. At most 1,000 detailed findings are returned; aggregate counts remain complete. issuesTruncated signals truncation.

Example and local run

The bundled example is synthetic and contains no patient or production submission data.

python3 validator.py < examples/input.json
python3 -m unittest discover -s tests -v
# Python 3.12, dedicated virtual environment recommended:
pip install -r requirements.txt
# Apify CLI, after putting examples/input.json in local INPUT storage:
apify run

See examples/output.json for the matching successful audit. A malformed date or unknown parent study gives a failing audit, not an invented replacement.

Limits, privacy and resource use

Maximum combined CSV size: 5 MiB. Maximum 10,000 rows per table, 40 columns, approximately 128 KiB per CSV field. The local CLI also caps JSON transport input at 6 MiB. Recommended cloud configuration: 256 MiB memory, 60-second timeout, limited permissions, no standby mode. The image runs as a non-root user. There are zero external source requests, retries or source rate-limit consumption; only Apify storage/event operations occur on-platform. No user-supplied URL is fetched. Logs contain only counts, status and elapsed time.

Input and output follow the customer's Apify storage/access/retention settings. Do not provide individual patient data or secrets. Source values are not echoed in audit output, but the customer's INPUT record still contains the submitted CSV. Whole-file hashes are integrity metadata, not anonymization.

Pricing proposal

Proposed price: $0.25 per completed audit, up to the stated limits. A FAIL report is chargeable because finding defects is the service. Input errors are not charged by the custom audit-completed event. The event hook is implemented; no paid listing or price has been activated. Configure this custom event as one-time per run and remove default per-dataset-item charging before publication. Disable usage pass-through. Platform start fees, if imposed, must be disclosed in the final listing.

A private cloud billing smoke test must verify the configured event and spending cap before public publication. No automatic run resurrection is proposed. See monetization-proposal.json for the budget and approval boundary.

Rights and limitations

Original validator code draws on DataForge's prior clinical export QA patterns; it copies no buyer dataset or competitor code. The customer must be entitled to process their input. This service neither licenses nor redistributes a source dataset and does not certify clinical accuracy, source authenticity, recency, statistical validity, legal reuse or regulatory compliance. It is not affiliated with NIH/ClinicalTrials.gov.

For reproducible issues provide the run ID and a minimal synthetic example through the Actor Issues tab. Do not post confidential CSVs or tokens.