How the Atlas is made

Sources

• Burden estimates — Institute for Health Metrics and Evaluation (IHME), Global Burden of Disease Study, via the GHDx results tool: deaths, DALYs, prevalence, and incidence; global, both sexes, age-standardized rates and counts, 1990 onward, with 95% uncertainty intervals. GBD is the standard reference for comparable burden across every disease. The data layer is source-agnostic; a WHO Global Health Estimates adapter exists as a fallback backbone.
• Drug → indication links — ChEMBL (European Bioinformatics Institute), DRUG_INDICATION and MOLECULE_DICTIONARY, approved indications only, licensed CC BY-SA 3.0. Approval-history detail from Drugs@FDA / openFDA (public domain).
• Trials — ClinicalTrials.gov, through Clin2’s own registry catalog (refreshed nightly). Trial counts here are registered interventional studies matched to each disease by its MeSH descriptors.
• Vocabulary — the GBD reporting cause hierarchy and the National Library of Medicine’s MeSH descriptors (courtesy of NLM; no endorsement implied) as the crosswalk spine between burden data, drugs, and trials.

The math

All trend math runs on the age-standardized rate per 100,000, log-transformed — so a 2% fall means the same thing for a common and a rare disease, and population growth or aging never masquerades as progress. Raw counts are shown but never differentiated.

• Trend line: LOESS (span 0.35, quadratic) over the log series.
• Velocity: the slope of a 5-year window on that smoothed curve, reported as annual % change. Negative = burden falling.
• Acceleration: the year-over-year change in velocity, in percentage points. Negative when a decline is speeding up.
• Uncertainty: 200 bootstrap resamples through GBD’s own uncertainty intervals; bands show the 10th–90th percentile. The first and last two years use one-sided windows and render dashed.
• Turning points: a year is flagged when velocity shifts by more than 1.5 standard deviations of all velocity changes across every disease and year, and is a local extremum. We keep the three largest per disease.
• Progress Index: the DALY-weighted average of velocity across all diseases — one number for “is humanity gaining ground this year?”

Second derivatives are noisy, especially for low-burden diseases. We state the direction anyway and attach a confidence read: clearly (the uncertainty band excludes zero), appears to be (the sign has held three-plus years), or may be — low confidence. Being directionally clear and occasionally wrong is a deliberate trade against saying nothing.

The treatability ladder

Each disease carries a rung describing what medicine can do at its best — never what is available everywhere. Access is unequal; the ladder is about capability, and the caveat travels with every badge.

Where AI is used — and where it isn’t

Every burden number, velocity, acceleration, and turning-point detection is deterministic math over the sources above — no AI involved. AI (open-weight models via OpenRouter, with logging-free providers) is used for five jobs: matching GBD causes to MeSH descriptors, scoring milestone significance, proposing non-drug milestones, explaining turning points, scoring the treatability ladder, and writing the plain-language narratives and FAQ. The chat guide answers only from the same data through fixed read-only lookups.

• Every AI output stores its model, prompt version, and timestamp.
• Anything unreviewed carries an explicit AI-generated label; AI-proposed milestones don’t appear at all until a human approves them.
• Turning-point explanations below 50% confidence render as “Unattributed” — many bends are data-model changes, coding shifts, or events like conflict, not medicine.
• Ladder scores require checkable citations, and a sample is human-reviewed each cycle.

What the Atlas is not

It describes populations, never people. It offers no diagnosis, no prognosis, no treatment guidance, and the chat declines personal medical questions by design. For finding a clinical trial that might fit a real person, the front door is Clin2’s trial search.

Citations & licenses

• Global Burden of Disease Study. Institute for Health Metrics and Evaluation (IHME), University of Washington. Used under the IHME free-of-charge non-commercial user agreement; commercial licensing in progress.
• ChEMBL. EMBL-EBI. CC BY-SA 3.0. Milestone indication data are derivative and shared under the same terms.
• Drugs@FDA / openFDA. U.S. Food & Drug Administration. Public domain.
• ClinicalTrials.gov. U.S. National Library of Medicine.
• MeSH. Courtesy of the U.S. National Library of Medicine.