Two decades of evidence. One unmeasured variable.

The study: 197,961 admissions and 176,696 eight-hour nursing shifts across 43 units at one large academic medical center. Staffing was measured shift by shift against each unit’s own target — not as an annual average. The strongest design in this literature for our purposes, because the exposure is the shift itself.

What it found: Each below-target shift raised a patient’s hazard of death by 2% (HR 1.02). The average patient was exposed to three such shifts — roughly 6% higher mortality than patients on always-staffed units. High-turnover shifts carried a 4% increase (HR 1.04).

The limit: one hospital, observational. Exposure was measured precisely; what the unit did about the shortfall was not measured at all. Read the summary ↗

The study: More than 223,000 operations, with shift-level nursing coverage linked to surgical outcomes.

What it found: Each one-nurse reduction per shift was associated with 7% more complications and 13% higher mortality. Understaffing was not the exception — at 45% of shifts, it was close to the norm.

The limit: observational and associational, drawn from national data. It counts coverage; it cannot see response. Read the finding ↗

The study: Nearly 550,000 medical-surgical patients, analyzed in support of a proposed safe-staffing policy for Pennsylvania hospitals.

What it found: Each additional patient per nurse was associated with 8% higher odds of death within 30 days — and 33% higher odds of nurse burnout. The clinical and workforce effects move together.

The limit: a policy model built on observational data; effects are modeled, not trialed. Read the finding ↗

The study: 8,584 U.S. hospital units, linking unit-level nurse turnover to patient falls.

What it found: Workforce instability itself tracks harm. In a hospital with 1,000 inpatients per day, a 10-point rise in turnover was associated with roughly 36 additional falls a year — the cost of continuity lost, measured in injuries.

The limit: unit-level analysis; falls are one harm among several, and turnover is a proxy for everything a departing nurse takes with her. Read the study ↗

The study: 168 Pennsylvania hospitals, 10,184 staff nurses, 232,342 surgical patients. The study that defined the field.

What it found: Workload was associated not just with mortality, but with failure to rescue — death after a complication the system should have caught. Rescue is a response function: surveillance, early detection, timely intervention. The study measured the workload around it, never the rescue itself.

The limit: cross-sectional. Associations, not randomized effects. Read the abstract ↗

The study: 729 inpatients on 20 units in 2 hospitals, asked directly what care never arrived.

What it found: Missed care concentrates in the basics — mouth care, ambulation, getting out of bed, bathing, information. And patients who reported adverse events during that stay — medication errors, new infections, pressure injuries, IV problems — reported significantly more missed care. Care was vanishing, and until someone asked the patient, it left no record anywhere.

The limit: patient self-report, two hospitals. But it is the only study here that measures omission — the break that never becomes an event. Read the study ↗

The study: The industry’s annual benchmark: 450 hospitals, 844,205 healthcare workers.

What it found: RN turnover costs the average hospital $3.9–$5.78 million a year. Each 1% change in turnover is worth about $289,000. An experienced-RN vacancy takes ~83 days to fill; national RN turnover runs 16.4%. Instability is not only a clinical exposure — it is one of the largest controllable lines in the budget.

The limit: an industry survey, self-reported by participating hospitals. Directionally consistent with the clinical literature above. Read the summary ↗