Case study database
40+ operational audits, one framework.
150–200 word structured micro-audits across four domains. New audits are added as the research pipeline produces them.
Weekend surge wait-time forecasting
A predictable weekly peak, dissolved into an unremarkable number by seven-day averaging.
Batch 01 · ER triagePriority queue collapse during flu season
Mid-acuity patients starve while median wait stays flat and nothing crosses a threshold.
Batch 01 · ER triageLeft-without-being-seen as a lagging indicator
Counted only after the patient has gone, when the shift that caused it is already over.
Batch 01 · ER triageAmbulance diversion decided too late
Declared once saturation has already happened, with no warning to the hospitals absorbing it.
Batch 01 · ER triageTriage acuity drift during understaffed shifts
The same presentation scores lower when the queue is long, and nothing compares scores against load.
Batch 01 · ER triageWaiting room deterioration between reassessments
A stable patient deteriorates unobserved because nothing measures time since last reassessment.
Batch 01 · ER triageBoarding time absorbed into ED length-of-stay averages
Two unrelated processes averaged into one number, so worsening boarding hides inside improving treatment time.
Batch 01 · ER triageShift-change arrival pileups
Assessment capacity approaches zero twice a day at times known months in advance.
Batch 01 · ER triagePediatric surge masked by aggregate volume
Pediatric arrivals spike, adult volume offsets it, and no aggregate number moves.
Batch 01 · ER triageFast-track capacity mismatched to acuity mix
A fixed-capacity area routed by a static rule that cannot see load on either side.
Batch 02 · ICU turnoverStep-down unit discharge delays
The constraint sits one level downstream, but the cost shows up as an ICU capacity problem.
Batch 02 · ICU turnoverVital-sign degradation as a transfer trigger
The trajectory was in the chart for hours, but only the final value crossed a threshold.
Batch 02 · ICU turnoverDischarge orders written but not executed
The bed is counted as occupied for hours after the decision to release it was made.
Batch 02 · ICU turnoverWeekend discharge throughput collapse
A weekly deficit that nets to zero in the number used to judge it.
Batch 02 · ICU turnoverICU readmission as a turnover signal
A premature discharge and its readmission are recorded as two separate successes.
Batch 02 · ICU turnoverPost-op bed reservations held against unclear need
A binary hold that cannot distinguish a certain need from an unlikely one.
Batch 02 · ICU turnoverTransfer refusals invisible in capacity counts
Beds reported as available are declined in practice, and the refusal leaves no trace.
Batch 02 · ICU turnoverNight-shift transfer bottlenecks
The longest part of the delay sits outside the window anyone measures.
Batch 02 · ICU turnoverIsolation-room capacity tracked separately from bed counts
The binding constraint is managed as an infection-control concern, never as a capacity number.
Batch 02 · ICU turnoverCleaning and turnover time unmeasured between patients
Beds counted as free while still being cleaned, because the intermediate state has no representation.
Batch 03 · EHR alertsAlert fatigue among triage nurses
Alert volume is monitored, precision is not, so a firing rule and a useful rule look identical.
Batch 03 · EHR alertsBatch-update latency in bed-tracking logs
Every figure is presented as current, and latency is invisible in the interface.
Batch 03 · EHR alertsDrug-interaction alerts overridden by default
The safety net is present in the record and absent in practice, and the override data is unusable.
Batch 03 · EHR alertsSepsis alert thresholds tuned once and never revisited
Performance is measured at procurement and then never again, so degradation produces no signal.
Batch 03 · EHR alertsModel drift after an EHR version upgrade
A renamed field, and the model keeps scoring against inputs whose meaning has changed.
Batch 03 · EHR alertsAlerts routed to the wrong role
The delay is not clinical judgement, it is routing by location rather than by required action.
Batch 03 · EHR alertsDuplicate alerts from parallel systems
Three systems alert on one event because nothing owns the concept of an event.
Batch 03 · EHR alertsDocumentation burden displacing bedside time
The data needed to attribute time to specific forms exists, and is only analysed for compliance.
Batch 03 · EHR alertsCopy-forward notes corrupting model inputs
Inherited findings are indistinguishable from reasserted ones, so models score a patient who no longer exists.
Batch 03 · EHR alertsSilent failure of an interface feed
An open idle connection looks identical to a working one, because volume is never compared against expectation.
Batch 04 · Regional allocationInter-hospital transfer coordination gaps
Every call is placed blind because capacity data exists inside each hospital and is shared with none.
Batch 04 · Regional allocationCross-facility bed tracking during surge
Update frequency is set by reporting convenience, not by how fast capacity actually changes.
Batch 04 · Regional allocationRegional capacity dashboards updated by manual report
A human in the hot path, least reliable exactly when demand makes it most important.
Batch 04 · Regional allocationRural transfer distance absent from capacity models
Matching optimises on bed availability alone, so the model minimises the wrong quantity.
Batch 04 · Regional allocationSpecialty coverage gaps on nights and weekends
Capability is stored as a flag with no time dimension, though the rota gap is known in advance.
Batch 04 · Regional allocationSurge plans triggered by occupancy alone
Escalation triggers on a level, so the measures activate after they could have helped.
Batch 04 · Regional allocationAmbulance offload delays propagating regionally
A queue inside one department becomes a region-wide vehicle shortage that nobody measures as one.
Batch 04 · Regional allocationPediatric and behavioral health beds counted as general capacity
A bed is a bed in the schema, so reported headroom exceeds usable headroom.
Batch 04 · Regional allocationStaffing agency lead times excluded from surge models
Beds opened on paper that cannot be staffed for days, because staff are modelled as elastic.
Batch 04 · Regional allocationPost-disaster demand persisting past the surge window
Structures stand down on a schedule set by the incident, not by the demand that outlives it.