Six datasets worth an afternoon

Under ten items, on purpose. Each one is something you can actually open and query this week, not a data portal you will bookmark and never return to.

1. MIMIC-IV. De-identified critical care records from Beth Israel Deaconess. The catch: credentialing takes real time, budget a week before you get access, not a day.
2. CMS Synthetic Public Use Files. Fully synthetic Medicare claims data, no access request needed. The catch: it is synthetic, so patterns are realistic but individual records are not real people, do not draw clinical conclusions from them.
3. Global Health Observatory (WHO). Country-level indicators going back decades. The catch: granularity stops at the country level, useful for policy questions, useless for anything patient-level.
4. HCUP State Inpatient Databases. State-level hospital discharge data. The catch: most states require a data use agreement and a modest fee, this is not a same-day download.
5. PhysioNet Challenge datasets. Curated, competition-ready clinical datasets with defined tasks. The catch: many are narrowly scoped to one prediction problem, great for a focused project, poor for general exploration.
6. Our World in Data, health section. Clean, versioned, well-documented global health metrics. The catch: it is aggregated and secondary, always trace a surprising number back to its original source before citing it.