Background:
Intensive care physicians make hundreds of decisions per shift under sustained cognitive and emotional demands. Decision fatigue, the tendency towards less effortful choices as cumulative decision load increases, may affect clinician wellbeing and decision processes. However, the associations of decision volume and temporal workload structure with within-shift fatigue and mood in intensive care unit (ICU) physicians remain unclear. This study explored whether decision volume and temporal clustering show distinct associations with fatigue, sleepiness, mood, and post-shift decision responses.

Methods:
Five physicians (two residents, three consultants) at a surgical ICU of a German university hospital were shadowed during 38 twelve-hour shifts (456 hours) between October and December 2025. Clinical decisions were classified by severity using a study-specific, non-validated Decision Severity Score. Because recruitment was feasibility-based and involved only five physicians, all analyses were considered exploratory and hypothesis-generating. Psychometric assessments of decision fatigue, sleepiness, and mood (Decision Fatigue Scale, Karolinska Sleepiness Scale, Profile of Mood States – Short Form) were administered at morning, midday, and shift end. Following each shift, physicians completed a standardised scenario task requiring three binary treatment decisions. Temporal workload structure was quantified using the burstiness metric; shifts were classified into four types by median split of decision density and burstiness (quiet-smooth, quiet-clustered, busy-smooth, busy-clustered). Linear and binomial mixed-effects models with physician as random intercept were used throughout.

Results:
Across 17,953 recorded decisions (median 474 per shift; range 366–589), a higher mean decision rate was associated with greater end-of-shift decision fatigue: an increase of 0.5 decisions per 5-minute interval corresponded to a 2.1-point higher decision-fatigue score (95 % CI 0.6 to 3.6). Higher burstiness was associated with less mood disturbance, with a 0.10 increase in burstiness corresponding to a 6.3-point lower mood score (95% CI −10.2 to −2.3). Averaged across all shifts, changes between morning and evening were small and not statistically significant for any of the three scores. Within-shift trajectories differed by shift type, however, with mood disturbance increasing over the shift in quiet-smooth shifts only, a difference already apparent at midday. Quiet-smooth shifts, the least objectively demanding, showed the least favourable psychometric values. Response certainty in the post-shift scenario task was lower after shifts with more task interruptions, with each ten additional interruptions corresponding to a 0.29-point lower certainty score (p = 0.012), whilst a descriptive trend towards more conservative scenario responses was observed in busy-clustered shifts (odds ratio 2.25, p = 0.555).

Conclusions:
Decision volume and temporal workload structure showed different associations with psychometric outcomes: higher decision rates were associated with greater end-of-shift decision fatigue, whereas greater temporal clustering was associated with lower mood disturbance, already apparent from midday onwards. That more quiet, uniformly paced shifts showed the least favourable psychometric values suggests that reducing decision volume alone may not improve physician wellbeing; this observation requires replication. Mitigation strategies may therefore need to extend beyond reducing decision volume to protecting concentrated, meaningful clinical activity and minimising hindrance stressors such as interruptions. Future multicentre studies should determine whether these associations are reproducible and whether temporal workload structure is related to objective decision quality or patient-safety outcomes.