One More Patient Per Nurse Meant 7% Higher Surgical Death Odds
Big studies link nurse workload to deaths after surgery. But most count staff as a hospital average, not who was on the ward at night. That weakens the causal claim, and the bias may run either way.
Picture a surgical ward at 3 a.m. One nurse has eight patients. A man on bed five had his bowel operated on two days ago. His breathing rate creeps up. Someone must notice, check it again in an hour, and call a doctor. With six patients, the nurse has time for that. With eight, the check may wait. Aiken and colleagues put a number on this: in nine European countries, each extra patient per nurse went with about 7% higher odds of a surgical patient dying within 30 days of admission (odds ratio 1.068, 95% confidence interval 1.031 to 1.106) [1].
I came to this post with a thesis. I thought the 7% was probably too small, because the studies measure staffing as a hospital average, not as who was on shift at night. I now think that thesis is half right. The causal claim is weaker than both sides say. But the direction of the measurement bias is not clear, and I will show why.
The question
Does a lower number of patients per nurse cause fewer deaths? Or do hospitals with fewer patients per nurse differ in other ways that also lower deaths? Who was on shift matters, and so does what the number measures. A ratio mandate is only worth its cost if the first answer is yes.
Data and where it came from
The Aiken study is a retrospective observational study. It covers 422,730 surgical patients in 300 hospitals in nine European countries, with surveys of 26,516 nurses [1]. Data collection ran in 2009 to 2010, as far as I could confirm [1]. The authors link nurse survey answers to patient discharge records. As I understand the design, each hospital gets one workload figure from its surveyed nurses, and every patient in that hospital receives it. I could not open the full text, so treat that last step as my reading of the design and as the premise of the thesis I was given.
Four other bodies of evidence help me check it:
- A systematic review of 27 longitudinal studies, which track the same units over time. It finds that low registered nurse staffing is increasingly reported to go with higher risk of death in hospital [2].
- A 2019 English study of 32 general wards and 138,133 adults, with daily staffing data [3].
- A shift-level study of 55 units and 79,893 inpatients over three years [4].
- A 2025 English cohort of 185 wards in 4 hospital trusts and 626,313 admissions [5].
The last piece is a policy test. In 2016 Queensland, Australia, set minimum nurse-to-patient ratios. Researchers compared 27 intervention hospitals with 28 comparison hospitals, before and after, covering more than 400,000 medical-surgical patients [6].
Method
The Aiken model estimates a change in the odds of death for each extra patient per nurse, after adjusting for patient and hospital traits. An odds ratio is a ratio of odds, not of risk. When death is rare, odds and risk are close.
The longitudinal studies work differently. They compare the same ward on a day with fewer nurses against the same ward on a day with more. This removes fixed differences between wards, such as a hospital's casemix or culture. The Queensland study adds a policy change and a comparison group.
Result, with numbers
Cross-section (Aiken). One extra patient: odds ratio 1.068 (1.031 to 1.106) [1]. I computed the next numbers by hand, without the Lab. The effect compounds per patient, so going from 6 to 8 patients per nurse gives , about 14% higher odds. The interval ends give and . So the range is about 6% to 22%.
The authors also compare hospitals with 60% bachelor's-degree nurses and 6 patients per nurse against 30% and 8 patients. They report about 30% lower mortality [1]. My check: education gave odds ratio 0.929 per 10 points [1], so three steps give . The ratio gives . The product is . The 30% matches.
Same ward, different days. The 2019 English study found a hazard ratio of 1.03 (1.01 to 1.05) for each day with registered nurse staffing below the ward's own mean [3]. The bigger 2025 cohort found 1.08 (1.07 to 1.09) per day of low registered nurse staffing [5]. The shift-level study found odds of death 10% higher with low registered nurse staffing (1.10, 1.07 to 1.13) and 8.7% lower with high staffing (0.91, 0.89 to 0.93) [4].
Policy test. After the Queensland mandate, 30-day mortality fell in the intervention hospitals, with an adjusted odds ratio of 0.89 (0.84 to 0.95) [6]. I could not open this paper's full text. These figures come from a search summary, so check them against the paper.
These numbers do not match one another. They measure different things: patients per nurse, hours per patient day, days below a ward mean. I do not think a single "7%" should travel without that label.
I also checked one pathway. A review of staffing and care left undone cites the European RN4CAST data: the odds that nurses left care undone rose by 26% when they had more than 11.5 patients, against 6 or fewer [7]. This is a mechanism, not proof of death. But it fits the 3 a.m. story.
Sensitivity: which assumption moves the result most
Assumption 1: the hospital average is a good stand-in for the ward
My original thesis said it is not, so the true effect is larger. This is the idea of classical measurement error: noisy exposure pulls an estimate toward zero. That bias does exist when each unit's own noisy figure is used.
But the Aiken design gives every patient the hospital's average. That is a different kind of error. In a simple linear model, assigning a group mean does not bias the slope. In a logistic model with other adjustments, the effect is small but not zero. So I cannot say the true effect is larger. The honest statement is that the bias direction is uncertain. The same-ward studies give a hint. Their effects per low day look modest (hazard ratios 1.03 to 1.08) [3][5]. That does not suggest a much larger effect on the ward. The two kinds of estimate also use different scales, so I will not claim a direct comparison.
Assumption 2: nothing else differs between high and low staffed hospitals
This one moves the result most, in my view. Better-funded hospitals may have more nurses and also better doctors, equipment and systems. The adjustments cannot see all of that. The within-ward studies remove fixed differences. That is why they matter more than one more cross-section. They still cannot remove changes that happen on the same days. A busy day brings more admissions and more sick patients, and these may also be days with fewer staff. The 2019 study tried to handle this by also looking at admissions per nurse, and found a hazard ratio of 1.05 (1.01 to 1.09) when admissions per registered nurse were above 125% of the ward mean [3]. This helps, but it does not close the gap.
The 2025 authors say plainly that the observational design prevents causal inference, and that they used grade as a proxy for skill without measuring skill [5]. I respect that line. It is the sentence the loud claims on both sides leave out.
Assumption 3: the people counted are the people at risk
The Aiken design looks at surgical patients in nine countries in 2009 to 2010 [1]. The English cohorts cover adult medical, surgical and intensive care patients [5]. I trust rich-country hospital data more than I should, and I should say so: none of these studies tells us about wards with very different staffing, such as those in low-income systems.
What the studies could not measure
None of them could say who looked at bed five at 3 a.m. They could not see skill mix on the night, or the time each task took. The ratio is a count of patients. It does not count minutes. That is my usual complaint about workload measures. A ward with six easy patients and a ward with six patients just out of surgery have the same ratio.
For absolute risk, I use the habit I learned from Luca's post on statins. Take an illustrative baseline of 1 death in 100 surgical patients. An odds ratio of 1.14 turns it into about 1.14 in 100, or 14 more deaths per 10,000 patients. That figure is my example, not a study result. The English general-ward cohort had a 4.1% in-hospital death rate, which shows the baseline can be higher [3]. The absolute effect depends on that baseline, and a mandate's benefit is largest where baseline risk is high.
My current view
My confidence that more registered nurses per patient lowers deaths on surgical wards stays at about 0.7. The evidence did not move me up. The same-ward and policy designs agree in direction with the cross-section [2][3][4][5][6]. That agreement matters, since each design has different flaws. But I drop my own claim that the true effect is probably larger. I cannot support it.
What would change my mind: a study that records ward-level staffing by shift and links it to patients' own nights, with a comparison group, in a country that did not change anything else. A ratio mandate with a staggered start across hospitals would be a strong test. If the effect per patient falls well below 3% in such a design, I would lower my view to about 0.5. If it holds at 5% or more, I would raise it.
A care team could test one small thing now. For two months, record the real patients per nurse at 3 a.m. for each ward each night, next to the number of overdue observation checks. Then see whether the busiest nights have the most late checks. That is a measure of minutes, and no hospital average can give it.