Hospital nursing commonly involves non‑standard hours, extended shifts and rotating day–night schedules that disrupt circadian rhythms and can affect sleep, fatigue and health. The authors used linked electronic rostering and sickness records to investigate how recent configurations of working time are associated with registered nurses’ sickness absence on adult acute inpatient wards. The primary focus was on exposures measured in rolling 7‑day windows prior to each worked shift or sickness episode.
This retrospective longitudinal analysis used anonymised e‑roster and sickness absence records from all adult acute inpatient wards in two large NHS hospital Trusts in England. Records covered April 2015 to September 2020. The analytic dataset comprised 1,367,497 worked shifts and 19,876 sickness episodes from 7,515 registered nurses across 95 wards. Administrative pay bands identified registered nurses (band 5 or above). Demographic details (for example, age or tenure) were not available due to data governance restrictions.
Shift pattern variables captured configurations within 7‑day exposure windows. Main exposures included:
Models also adjusted for total working hours in exposure windows, hours worked as bank, number of prior sickness episodes and part‑time status (below median 0.75 FTE). Records of sickness episodes preceded by zero working hours in the prior month were excluded.
Random‑intercept logistic mixed models estimated associations between shift variables and onset of sickness absence, clustering repeated observations within nurses. Nurse‑level clustering was significant (adjusted ICC=0.710); ward‑level clustering was not included. Univariable and full multivariable models were run, with quadratic and cubic terms added where exploratory analyses suggested nonlinearity. Variance inflation factors indicated low multicollinearity among predictors (VIF <5, excluding expectedly elevated polynomial terms).
Two sensitivity analyses were conducted: re‑creating exposures using 28‑day lookback windows and restricting to pre‑March 2020 records to assess potential pandemic influences. Model fit comparisons used AIC and BIC.
The dataset included more shifts from full‑time nurses (60% of shifts). Median sickness episode duration was 4 days (IQR 2–8). Median shift length and the proportion of long shifts were similar between full‑time and part‑time staff, but full‑time nurses worked more night shifts, quick returns and rotations. Over the study period there were modest increases in the mean number of long shifts, night shift proportion and quick returns.
In the full multivariable model, several shift features in the previous 7 days were associated with higher odds of sickness absence:
Nonlinear analyses for proportions of long shifts and night shifts indicated increased odds of sickness at higher proportions. Working about 70% of shifts as long (≥12 hours) was associated with the highest increased odds compared with working no long shifts (OR 1.68, 95% CI 1.53–1.84). Working 100% night shifts slightly increased odds relative to no night shifts (OR 1.07, 95% CI 1.01–1.13).
One finding showed working long consecutive spells (≥6 shifts) was associated with decreased odds (OR 0.09, 95% CI 0.01–0.66), but this configuration was rare (only 1,937 cases) and estimates are likely unstable.
Using a 28‑day exposure window returned the same directions of effect for most variables but with attenuated effect sizes; AIC and BIC were higher for 28‑day models, indicating the 7‑day exposure was a more parsimonious fit. Restricting data to pre‑March 2020 records produced similar directions and magnitudes of effects, suggesting primary findings were not driven by pandemic‑period changes.
Strengths of the study include linkage of comprehensive electronic rostering data with sickness records at shift level, a large sample across multiple wards and modelling that captured nonlinearity and within‑nurse clustering. Limitations include absence of demographic covariates and inconsistent overtime records due to governance constraints, which required inclusion of proxy covariates such as previous sickness and total worked hours. Source data cannot be shared because of data sharing agreements.
The analysis indicates that organisational scheduling characteristics — particularly short recovery periods (quick returns), frequent shift rotations, consecutive intense working spells and high proportions of long or night shifts — are associated with higher odds of nurses’ sickness absence, most notably over weekly (7‑day) exposure windows. These findings can inform shift‑planning policies and proactive management strategies aimed at reducing adverse working‑time configurations to support staff well‑being and potentially mitigate associated organisational costs of increased sickness absence.