DNP 679 Module 7 Final Paper Critique: Statistics in an Observational Study of Nurse Staffing and Surgical Mortality Example

Seven Percent per Patient: A Statistical Critique of a Landmark Study Linking Hospital Nurse Staffing to Surgical Mortality

Student Name

Edson College of Nursing and Health Innovation, Arizona State University

DNP 679: Biostatistics: Principles of Statistical Inference

Instructor Name

Month Day, Year

Seven Percent per Patient: A Statistical Critique of a Landmark Study Linking Hospital Nurse Staffing to Surgical Mortality

Introduction

Few nursing studies have shaped policy debates as much as Aiken et al. (2002), which tied heavier nurse workloads to higher odds that surgical patients would die within a month of admission. The finding has been cited in arguments for mandated nurse-to-patient ratios and for hospital staffing investment. Because the study is so influential, its statistical foundation deserves close scrutiny. This critique evaluates the research question, design, data sources, measurement, analytic approach and interpretation, applying the principles of inference covered in the course, with particular attention to logistic regression.

Research Question and Design

The study asked whether hospital nurse staffing, measured as the number of patients per nurse, was linked to surgical patients' risk-adjusted deaths within 30 days and to failure to rescue, and to nurse burnout and job dissatisfaction. The design was cross-sectional and observational, linking three data sources from 168 nonfederal adult general hospitals in Pennsylvania: a survey of 10,184 staff nurses, discharge abstracts for 232,342 general, orthopedic and vascular surgery patients and hospital administrative data.

An observational design was the only feasible choice: hospitals cannot be randomly assigned to staffing levels for research. The design's strength is scale and realism, since it covers nearly all eligible hospitals in a state and uses routine data. Its weakness is that it can establish association, not causation (Polit & Beck, 2021). Well-staffed hospitals are likely to differ in other respects too, such as resources, physician quality, technology and culture, that also affect survival. The cross-sectional timing adds a further limit: staffing was measured from a single survey, so the analysis cannot show that staffing levels preceded the outcomes, although reverse causation, in which deaths lead to lower staffing, is implausible for most hospitals.

Measurement

Mortality and failure to rescue were derived from discharge abstracts linked to vital statistics records. Mortality is an objective and important outcome, though it is relatively rare in surgical populations, which is one reason a very large sample was needed. Failure to rescue, death after a complication, focuses on the outcome most sensitive to nursing surveillance, which strengthens the conceptual link between staffing and outcome.

The staffing measure is the critical variable. It was derived from nurses' survey reports of how many patients they cared for on their last shift, averaged to the hospital level. This approach has strengths: it reflects actual workload rather than budgeted positions, and averaging across many nurses reduces individual error. It also has weaknesses. Nurses who are dissatisfied may report heavier loads; a single shift may be unusual; and averaging across all units assumes that staffing on medical or other units reflects conditions for surgical patients. Any random measurement error in staffing would tend to bias the association toward zero, but systematic error linked to burnout could bias the nurse outcome models in either direction, since burnout is measured from the same survey.

Statistical Analysis

Because the outcomes are binary (died or survived; burned out or not), logistic regression is the appropriate model. It estimates the odds of the outcome as a function of predictors and expresses each predictor's association as an odds ratio, holding the other variables constant. For the patient outcomes, the models adjusted for an extensive set of patient characteristics, including demographics, the type of surgery and comorbid conditions, and for hospital features including bed count, teaching role and high-technology services. Adjustment of this kind is the main defense against confounding in an observational study, and the breadth of the risk adjustment is a strength.

Two analytic issues deserve attention. First, patients are clustered within hospitals, and staffing is a hospital-level variable. Patients in the same hospital share many conditions, so their outcomes are not independent. If the models ignored clustering, standard errors would be too small and confidence intervals too narrow, overstating precision. A careful reader should check how the methods section accounts for this clustering, because the 168 hospitals, not the 232,342 patients, are the effective sample for the staffing estimate. Second, residual confounding is always possible. Adjustment can address only the variables that were measured; unmeasured hospital qualities, such as surgeon skill or organizational culture, could explain part of the association. Later work by the same team adding nurse education and work environment to similar models suggests that staffing is one of several related hospital features, which is consistent with this concern.

Results and Their Interpretation

The central estimate: for every extra patient in a nurse's average load, the odds that a surgical patient died within 30 days of admission rose by about 7% (OR 1.07, 95% CI [1.03, 1.12]), with an identical 7% rise for failure to rescue (OR 1.07, 95% CI [1.02, 1.11]). For nurses, each additional patient was associated with 23% higher odds of burnout (odds ratio 1.23, 95% CI [1.13, 1.34]) and 15% higher odds of job dissatisfaction (odds ratio 1.15, 95% CI [1.07, 1.25]).

Reporting confidence intervals alongside the odds ratios is a strength: none of the intervals include 1, and the lower bounds indicate that even the smallest plausible associations are positive. Interpretation needs care, however. An odds ratio is not a relative risk. When an outcome is rare, as surgical mortality is, the odds ratio approximates the relative risk, so the 7% figure can reasonably be read as about a 7% higher risk per additional patient. For common outcomes such as burnout, which affected a large share of nurses, the odds ratio overstates the relative risk, and the 23% figure should not be read as a 23% higher probability of burnout.

The authors also extrapolated the per-patient estimate across a range of staffing levels, reporting that patients in hospitals with eight patients per nurse would have about 31% higher odds of death than those in hospitals with four. Because the model treats the association as constant for each additional patient, this extrapolation assumes a linear relationship on the log-odds scale across the full range. That assumption is reasonable within the range of observed staffing but should not be extended beyond it.

Conclusions and Implications

The authors concluded that improving nurse staffing could reduce preventable deaths and improve nurse retention. The statistical evidence supports a robust and consistent association between staffing and outcomes, adjusted for many patient and hospital factors, with precise intervals drawn from a large population. It does not, on its own, establish that changing staffing in a given hospital would produce the predicted reduction in deaths, because of the observational design, the possibility of residual confounding and the reliance on self-reported workload. The evidence is strongest when read alongside later studies with different designs. A longitudinal study that tracked individual patients' exposure to understaffed shifts in one hospital system found that each additional understaffed shift was associated with higher mortality (Needleman et al., 2011). Because that design compares exposure within the same hospitals over time, it is less vulnerable to the between-hospital confounding that limits the 2002 analysis, and its agreement with the earlier finding strengthens the case that staffing itself matters.

For a DNP-prepared nurse leader, the study offers a model of how linked administrative and survey data can inform policy, and a caution about translating odds ratios into promises. When presenting staffing evidence to administrators, it is accurate to say that lower patient loads are consistently associated with lower surgical mortality; it is an overstatement to say that adding one nurse will save a specific number of lives.

References

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), 1987-1993. https://doi.org/10.1001/jama.288.16.1987

Needleman, J., Buerhaus, P., Pankratz, V. S., Leibson, C. L., Stevens, S. R., & Harris, M. (2011). Nurse staffing and inpatient hospital mortality. New England Journal of Medicine, 364(11), 1037-1045. https://doi.org/10.1056/NEJMsa1001025

Polit, D. F., & Beck, C. T. (2021). Nursing research: Generating and assessing evidence for nursing practice (11th ed.). Wolters Kluwer.