Skip to content
fatiguemanagement.co.uk

Limitations of fatigue models

Fatigue models — including fatigue risk index (FRI) methods based on HSE research — translate roster and workload inputs into scores such as a Fatigue Index (FI) or Risk Index (RI).

Used well, they support planning and documentation. Used poorly, they create false confidence or unnecessary conflict. This page explains the boundaries every competent reviewer should understand before acting on model output.

What fatigue models can do

Section titled “What fatigue models can do”

Models grounded in research (e.g. HSE RR446) can:

  • Compare roster options on a consistent basis before implementation
  • Highlight patterns with elevated modelled exposure — timing, cumulative load, or job factors
  • Support dialogue between planners, safety teams, and workforce representatives
  • Document due diligence when assumptions, scores, and decisions are recorded

HSE describes its fatigue/risk index as a tool to compare shift patterns and identify when fatigue risks may be excessive — as part of wider risk assessment, not as a standalone verdict.

What fatigue models cannot do

Section titled “What fatigue models cannot do”
Limitation Plain-English explanation
Measure real-time alertness Models estimate exposure from schedule data. They do not know how a specific person slept last night.
Determine fitness for duty Fitness for duty involves individual state, task context, and policy. Models do not replace that judgement.
Predict specific accidents Risk indices express relative modelled risk — not whether an incident will occur on a given shift.
Capture all lifestyle factors Secondary jobs, caring responsibilities, health conditions, and off-duty activity are largely invisible to roster-based models.
Replace supervision and reporting Workers and supervisors observe impairment that models miss. Reporting routes remain essential.
Guarantee safety at low scores A favourable score does not prove a shift is safe — controls, workload, and individual factors still matter.
Mandate action at high scores Response depends on organisational policy and competent review — not the model alone.

Population averages vs individuals

Section titled “Population averages vs individuals”

Fatigue models are built from population-level research on sleep, circadian rhythms, and performance. They describe what tends to happen for a typical worker on a given pattern — not what will happen for every individual.

Two workers on the same roster may experience very different sleep and alertness because of:

  • Age and chronotype (natural morning/evening preference)
  • Off-duty responsibilities and environment
  • Caffeine, nutrition, and hydration
  • Stress and acute illness
  • Undiagnosed sleep disorders (organisations should signpost occupational health — not diagnose via models)

A model may show moderate exposure while a specific worker is struggling — or vice versa. Never dismiss credible fatigue reports because a score looks acceptable.

Input quality: garbage in, garbage out

Section titled “Input quality: garbage in, garbage out”

Model outputs are only as good as inputs. Common failures:

Input problem Effect on output
Missing previous shifts in the schedule Cumulative component wrong — often the largest error
Planned times used when actual hours differ Underestimates exposure after disruption or overtime
Unrealistic break assumptions Understates job/break component
Wrong commute or travel data Misstates total burden where travel is in scope
Incorrect shift classification (day/night) Misaligns circadian component

Reassess when assignments, timings, travel, or actual hours change materially from the last calculation.

Sleep outside work

Section titled “Sleep outside work”

Models focus on work-related contributors to fatigue exposure. They generally cannot know:

  • Whether the worker slept during the rest window
  • Quality of sleep environment
  • Whether rest days were genuinely free from call-ins

Organisations still have a role in designing rosters that allow adequate recovery — even if they cannot control private sleep behaviour.

Workload, environment, and task demand

Section titled “Workload, environment, and task demand”

Roster timing is not the whole story. The same shift pattern may feel very different when:

  • Workload spikes during an outage or possession
  • Tasks require sustained vigilance with few breaks
  • Environmental conditions (heat, noise, isolation) add strain
  • Monotonous machine-paced work increases mental fatigue

Some models include workload and break parameters; many assessments still under-specify these fields. Competent review should ask whether the task during the shift matches the assumptions in the score.

Travel and commuting

Section titled “Travel and commuting”

Travel before and after duty can add to overall fatigue burden even when it is not paid working time. Whether and how travel is included depends on organisational policy and methodology — not all tools treat travel the same way. See duty, travel, and commute and door-to-door time.

Planners should not ignore long commutes simply because they fall outside a duty-time limit in a contract or model.

Policy thresholds are not physics — or law

Section titled “Policy thresholds are not physics — or law”

Many organisations apply colour bands or trigger levels to FI and RI scores. These thresholds are policy decisions configured to context — they are not universal constants and should not be presented as statutory limits.

HSE’s research framework provides methods to calculate indices; acceptable scores for action are for the organisation — subject to competent review, consultation, and any contractual requirements that apply to your work. HSE does not currently specify threshold scores for fatigue.

Governance judgement still decides outcomes

Section titled “Governance judgement still decides outcomes”

A defensible process usually records:

  • What the model showed (scores, assumptions, version)
  • Who reviewed it (named competent person or role)
  • What was decided (proceed, mitigate, redesign)
  • What controls apply if proceeding (supervision, task adjustment, recovery plan)

Absence of a recorded decision when exposure is elevated is a governance gap — regardless of what software displayed.

When to soften or avoid model reliance

Section titled “When to soften or avoid model reliance”

Consider reducing reliance on model scores alone when:

  • Credible fatigue reports conflict with low scores
  • Operations are highly disrupted or ad hoc
  • Roles are novel or rotations not well represented in the research base
  • Individual medical or occupational health factors are in play
Section titled “Related pages”

References

Section titled “References”