ORI runs deterministic pipelines against openFDA, CMS, PubMed, and clinical claims data. Every readout is sourced to the underlying record, every synthesis step is labelled, and every number on this page is the current pipeline output.
Every pipeline is code you can audit. Every readout is deterministic where it can be, AI-synthesised where it cannot, and labelled transparently at every step.
Extract all FDA Adverse Event reports for a drug class. Score each drug's female-toxicity rate against an epidemiology-calibrated expected-usage baseline. Flag drugs where toxicity skew is structural, not demographic.
Medicare and Medicaid claims at ICD-10 pathway resolution. Surfaces sex-stratified misrouting where obstetric origins are erased into primary-care coding, and where diagnostic delay compounds onto state budget lines.
Maps each autoimmune condition to its X-chromosome inactivation escape mechanism, TLR7/TASL biallelic expression pattern, and downstream pharmacological implications. Reframes 80 percent female predominance as cellular architecture, not estrogen.
Prices the unpaid family caregiver subsidy absorbed out of hospital discharge margins, by jurisdiction and diagnosis category. Turns the Sandwich Generation from a sociology term into a line item.
The FDA's expected-usage baseline for each drug class is calibrated to epidemiology (a decongestant given equally to both sexes has a 50 percent female baseline; a thyroid medication at 85 percent). The pipeline flags every drug where the toxicity signal in FAERS exceeds that baseline by more than 20 percentage points. Six examples follow. The full list is available on request.
Pyrimidine synthesis inhibitor. Rheumatoid arthritis affects women at a 3-to-1 ratio; leflunomide is prescribed accordingly. But the tox signal exceeds the prescribing ratio, consistent with a pharmacokinetic amplification rather than a usage artifact.
Pseudoephedrine is sold over-the-counter to both sexes in equal proportion. There is no usage-based reason for a 40-point female skew in toxicity. Aggregate pharmacovigilance averages smooth this signal; a sex-stratified view surfaces it.
Fioricet-class headache combination. Prescribed to both sexes for tension and post-traumatic headache, but overwhelmingly dispensed to women because migraine is sex-skewed 3-to-1. The toxicity signal exceeds the prescribing skew, consistent with the broader finding that analgesics metabolize more slowly in female livers (see the Zucker 76-of-86 pharmacokinetic signal).
Prescribed for opioid and alcohol dependence, which skew male in the general population. Yet women file 85 percent of toxicity reports. A prevalence-adjusted filter would absorb this into the average; the sex-stratified view makes it visible.
A potent topical steroid for eczema, psoriasis, and dermatitis. Conditions are not sex-skewed at the population level, yet 90 percent of toxicity reports are female. Worth an inquiry into thinner-skin-volume systemic absorption, an area with almost no published pharmacokinetic trial data stratified by sex.
Injectable corticosteroid used in orthopaedic and rheumatologic indications. Orthopaedic injections skew male; rheumatologic injections skew female. The 35-point female tox gap suggests the injection-volume reactions concentrate specifically in the rheumatologic population, a pattern invisible to aggregate reporting.
Clinical Synthesis Agent queries NCBI PubMed, openFDA FAERS, FDA Labels, and the ORI SQLite claims database. Every retrieved record carries a PMID or DOI. No AI in this layer.
Routes structured-JSON generation to Claude Sonnet 4.6 (default) or Gemini 2.0 Flash (fallback), driven by an env flag. Every LLM-synthesised field is labelled synthesis_type: LIVE with a timestamp and model string.
Converts the clinical and fiscal output into a jurisdiction-specific intervention memo: CalAIM for California, NHM for Andhra Pradesh, TARDOC for Switzerland. Output labelled AI-drafted and routed for human review.
No enterprise contract. No sales cycle. A scoped brief, a deterministic pipeline run, and a written findings memo, with the JSON artefact attached.