What does the research actually show about nurse staffing and patient mortality?
Three landmark studies found the same association between nurse workload and death within 30 days of admission, on the odds scale, in designs that cannot establish cause.
Reviewed against primary sources on July 19, 2026 by the Soon operations research team
The evidence in one line
Across three landmark cross-sectional studies, hospitals whose nurses carried heavier patient loads also showed higher odds of surgical patients dying within 30 days of admission. The largest of them, Aiken et al. (2014), measured the average number of patients per nurse across 300 European hospitals and found each additional patient in that average associated with a little under 7% higher odds of death, OR 1.068 in the fully adjusted model. None of these designs can establish cause, so what the literature supports is a consistent association rather than a demonstrated causal relationship.
What the three studies measured
The best known estimate comes from Aiken et al. (2014), the RN4CAST study published in The Lancet. It covered 422,730 surgical patients aged 50+ treated in 300 hospitals across 9 European countries, linked to a parallel survey of the nurses working in those same hospitals. The exposure was the hospital-level mean number of patients per nurse. The outcome was death within 30 days of admission. In the fully adjusted model, each additional patient in that average workload was associated with 7% higher odds of a surgical inpatient dying within 30 days of admission, OR 1.068 (95% CI 1.031-1.106), p=0.0002.
A dozen years earlier, Aiken et al. (2002) had reported a nearly identical number in Pennsylvania hospitals, built on the same kind of exposure, a hospital-level mean patient-to-nurse ratio rather than any unit or shift measurement. Each additional patient per nurse was associated with 7% higher adjusted odds of 30-day mortality, OR 1.07 (95% CI 1.03-1.12, p<.001), and with 7% higher odds of failure-to-rescue, OR 1.07 (95% CI 1.02-1.11, p<.001). Failure-to-rescue counts deaths among patients who developed a complication, which is a useful second lens because it isolates what happened after something went wrong rather than how sick the admitted population was.
Aiken et al. (2010) then estimated the same relationship separately in three states. Per additional patient per nurse, 30-day mortality odds ratios were 1.13 (1.07-1.20) in California, 1.10 (1.01-1.22) in New Jersey and 1.06 (1.00-1.12) in Pennsylvania. Failure-to-rescue followed the same pattern at 1.15 (1.09-1.21) in California, 1.10 (1.01-1.21) in New Jersey and 1.06 (1.00-1.12) in Pennsylvania. Two of the Pennsylvania intervals touch 1.00 at the lower bound, which is worth saying out loud rather than rounding away.
The scale matters more than the number
Every figure above is an odds ratio, and odds ratios are widely misread. OR 1.068 means the odds of dying within 30 days of admission were about 7% higher. It does not mean 7% more patients died, and it does not mean the mortality rate was 7% higher. Odds ratios and risk ratios are different quantities built from different arithmetic, and the three studies estimated and reported odds. The odds scale is the only scale these figures should be quoted on.
The conversion that goes wrong most often turns a coefficient into a body count. OR 1.068 is routinely restated as 7% more deaths. Nothing in Aiken et al. (2002), Aiken et al. (2010) or Aiken et al. (2014) supports that restatement. What the studies measured is how the odds of an outcome varied with a staffing measure across hospitals, in a single snapshot, with statistical adjustment for patient and hospital characteristics. Anyone quoting the number has to carry the scale along with it, because the scale is what the number means.
The detail that separates a real reading from a slogan
Aiken et al. (2014) reported more than one specification, and the difference between them is the most instructive part of the paper. The partly adjusted model, which accounted only for the country a hospital sat in, produced OR 1.005 with p=0.816. That is a null result. The 7% figure comes from the fully adjusted model, after patient and hospital characteristics were taken into account.
This is not a flaw, and it is not evidence of fishing. Sicker patients are concentrated in larger, more specialized hospitals, which also tend to staff differently, so an unadjusted or lightly adjusted comparison mixes the staffing signal with case mix and hospital type. Adjustment is how the analysis separates them. But it does mean the association is conditional on the model. Anyone citing the 7% figure is citing a fully adjusted estimate and should say so, because a reader who checks the paper will find a null coefficient sitting a few lines above it.
The same model carried a second coefficient worth keeping attached to the first. Every 10 percentage point increase in the proportion of nurses holding a bachelor degree was associated with 7% lower odds of death, OR 0.929 (95% CI 0.886-0.973), p=0.002. Workload and education were estimated simultaneously, so the honest summary is that both were independently associated with the outcome. Quoting one without the other narrows a two-part finding into a single talking point.
Why the consistency is credible without being causal
All three studies are cross-sectional. Staffing and mortality were observed in the same window rather than staffing being altered and outcomes then followed, which means each analysis compares hospitals with each other at one moment instead of comparing a hospital with its own later self. The authors are direct about what that costs them. Aiken et al. (2014) caution against reading their coefficients as causal, and the authors of Aiken et al. (2010) state plainly that they cannot establish causality. Those sentences are in the papers, and carrying them along is part of citing the work accurately.
Replication is still worth something. The same directional association appeared in Pennsylvania in 2002, in three US states in 2010, and across 9 European countries in 2014, under different health systems, payment models, nursing labor markets and decades. Coincidence and local quirks are poor explanations for a pattern that survives that much variation in context. What replication cannot do is upgrade a design. Consistent snapshots are still snapshots.
The alternative explanations are ordinary ones. Hospitals that staff nursing units well may also invest in monitoring equipment, frontline supervision, overnight physician coverage, escalation protocols and clinical information systems, none of which these models fully measured, and any of those could be carrying part of the association attributed to workload. A reverse pathway is open too, since institutions under financial strain cut nursing hours and are likely to be cutting or deferring other things that bear on surgical outcomes at the same time. The conclusion that survives all of this is that nurse workload is a robust marker of something that matters for surgical patient outcomes, and that no one has yet shown, in these designs, that changing the workload changes the outcome.
What this means for your schedule
- State the finding on the odds scale, as higher odds of dying within 30 days of admission, never as a mortality rate or a count of extra deaths.
- Quote the fully adjusted Aiken et al. (2014) estimate, OR 1.068, and be ready to explain that the partly adjusted model was null at OR 1.005, p=0.816, because anyone who opens the paper will find both.
- Keep the claim inside its population, surgical inpatients largely aged 50+, and refuse to extend it to medical, pediatric or obstetric units.
- Report workload the way the studies measured it, as a hospital-level mean of patients per nurse. Tracking it by unit or by shift is a deliberate refinement beyond the exposure these odds ratios are attached to, and should be described that way rather than as replicating the research.
- Track nurse education mix alongside workload, since Aiken et al. (2014) estimated the bachelor-degree coefficient, OR 0.929, in the same model as the staffing coefficient.
The business case
Three separate studies, one across nine European countries and two in the United States, more than a decade apart, found the same direction and roughly the same magnitude: heavier nurse workloads went with higher odds of surgical patients dying within 30 days of admission. That degree of replication is rare in health services research, and it is enough to justify treating nurse workload as a measured, reported operating quantity rather than a residual of the budget.
What it does not justify is a promise. Every one of these estimates is correlational, a limit the authors themselves are explicit about, so the defensible executive position is that workload is a risk indicator worth monitoring, not a lever with a proven mortality return.
If workload is going to be reported, report it on the terms the evidence uses: a hospital-level mean of patients per nurse on surgical inpatient services, with the education mix of the nursing staff beside it, since both coefficients came out of a single model. A workload figure standing alone represents half of what was estimated.
Frequently asked questions
- What did the RN4CAST study actually measure?
- Aiken et al. (2014) estimated two coefficients simultaneously in one fully adjusted model: each additional patient in the average nurse workload at OR 1.068 (95% CI 1.031-1.106, p=0.0002) for death within 30 days of admission, and each 10 percentage point rise in the share of nurses holding a bachelor degree at OR 0.929 (95% CI 0.886-0.973, p=0.002). The exposure is a hospital average rather than any individual nurse assignment.
- Has the same association been found outside Europe?
- Yes. Aiken et al. (2010) estimated the association separately in three states and found the same direction in each: OR 1.13 in California, 1.10 in New Jersey, and 1.06 in Pennsylvania for death within 30 days of admission. The intervals overlap heavily, so the states are better read as agreeing than as ranking against one another.
- What kind of study would settle whether staffing changes outcomes?
- One that watches staffing change over time and follows outcomes afterwards, with a comparison group that did not change, so the shift in staffing can be separated from everything else moving in the same period. The closest thing this literature has is California's legislated staffing floor. McHugh et al. (2012) documented that the mandate was followed by a measurable drop in patients per licensed nurse across California hospitals, which establishes that a legislated floor moved staffing. The authors state they were unable to determine the causal effect, and no verified causal estimate of a resulting change in mortality exists in either direction. The claim that the law proved itself and the claim that it proved nothing both go past the evidence.
- Does the finding apply to every hospital patient?
- No. All three studies drew their samples from surgical inpatients, and Aiken et al. (2014) restricted its population to patients aged 50 and over. Surgical populations suit this kind of research because the reason for admission is well defined and risk adjustment is tractable, which is exactly why the results do not transfer automatically. A medical admission, an intensive care stay, a delivery and a pediatric admission each involve different nursing work, different deterioration patterns and different rescue opportunities. Extending an odds ratio estimated on surgical inpatients aged 50+ to any of those settings is an assumption a reader is entitled to challenge, not a finding any of these papers reported.
Sources
Every figure on this page is drawn from a cited primary source and checked against the original publication.
Aiken, L. H., Sloane, D. M., Bruyneel, L., Van den Heede, K., Griffiths, P., et al. (2014). Nurse staffing and education and hospital mortality in nine European countries: a retrospective observational study. The Lancet, 383(9931), 1824โ1830. https://pmc.ncbi.nlm.nih.gov/articles/PMC4035380/
Retrospective cross-sectional observational study (422,730 surgical patients, 300 hospitals, 9 countries)
Aiken, L. H., Clarke, S. P., Sloane, D. M., Sochalski, J., & Silber, J. H. (2002). Hospital Nurse Staffing and Patient Mortality, Nurse Burnout, and Job Dissatisfaction. JAMA, 288(16), 1987โ1993. https://doi.org/10.1001/jama.288.16.1987
Cross-sectional analysis of survey and discharge data (Pennsylvania hospitals)
Aiken, L. H., Sloane, D. M., Cimiotti, J. P., Clarke, S. P., Flynn, L., Seago, J. A., Spetz, J., & Smith, H. L. (2010). Implications of the California Nurse Staffing Mandate for Other States. Health Services Research, 45(4), 904โ921. https://doi.org/10.1111/j.1475-6773.2010.01114.x
Descriptive cross-state comparison (California, New Jersey, Pennsylvania)
None of the studies cited here evaluated Soon.They examine scheduling practices, shift patterns, and working hours as studied by independent researchers, so their findings describe what those practices are associated with, not what any particular software produces.
This article summarizes published research for scheduling and operations decisions. It is not medical advice. Individual health questions belong with a qualified clinician.
Keep reading
Does nurse staffing affect patient outcomes?
A four-article pillar on nurse workload and patient outcomes: what the landmark studies measured, what the California mandate research settled, and where the causal line actually sits.
Read →Do mandated nurse-to-patient ratios save lives?
A legislated ratio floor was followed by a real drop in how many patients each nurse carried, but whether it changed patient survival has not been established, and the two numbers most often quoted as proof are model projections.
Read →What does understaffing do to nurses?
The workload association with burnout and job dissatisfaction is the strongest signal in the nurse staffing literature, and it stands independently of the contested mortality question.
Read →