Emergency-rescue stress matrix (2026-11)

SIMULATION — fictional patients — educational use only. Not a medical device. Not medical advice. Every vital sign below is illustrative model output of the simulator (engine/physiology.js, engine/pd.js); nothing here is a clinical efficacy claim or a validation.

Generated by node tools/emergency-stress-report.mjs (driver: tests/emergency-stress.mjs; a representative subset runs in npm test, tests/emergency-stress-2026-11.test.mjs). Do not edit by hand — regenerate.

Question: do the rescue interventions the engine offers work in its scripted emergencies across extreme fictional patients, and do the safety rules hold in every run, including deliberately abusive dosing?

Matrix size

Safety assertions (every run, including the aggressive arm)

Refusals by drug and reason (numbers masked as #):

Accepted with a warning or after a confirmation (engine policy, SPEC §5.2/§6: the regimen range is a warning; drug.ceiling values and label repeat intervals refuse; simulator teaching ceilings refuse until confirmed):

Aggressive arm extremes (432 runs): max MAP 192 mmHg, max HR 200 /min; seconds at a physiology clamp (MAP/HR 25 or 200, SpO2 40): MAP 0, HR 8262 (in 12 runs: sepsis_emergency), SpO2 0; all arms: 17691 clamp-seconds (control map anaphylaxis_hypotension: 7830; aggressive hr hemorrhage: 4347; aggressive hr anaphylaxis_hypotension: 3915; control map hemorrhage: 1578; rescue_im map anaphylaxis_hypotension: 21). Per run, accepted within 600 s: phenylephrine bolus ≤ 200 mcg, ephedrine ≤ 50 mg, epinephrine IV ≤ 0.55 mg, epinephrine IM ≤ 1.5 mg, glycopyrrolate ≤ 1 mg, crystalloid ordered up to 60.5 L.

Efficacy (directional, illustrative)

Recovery = the emergency's primary vital sign(s) back inside the limits for ≥ 60 s, timed from the onset, counted only after the first deranged sample (a run that never deranges counts as recovered at 0 s). Primary criteria: hemorrhage MAP ≥ 65; vagal bradycardia HR ≥ 50; bronchospasm SpO2 ≥ 92 and internal PaCO2 < 60 mmHg; laryngospasm SpO2 ≥ 92; anaphylaxis MAP ≥ 65 and SpO2 ≥ 92; light anaesthesia MAP ≤ 120, HR ≤ 120 and processed-EEG index ≤ 60. All thresholds are reused simulator values (alarm thresholds, GA band, model hypercapnia threshold). The composite (MAP ≥ 65, HR ≥ 50, SpO2 ≥ 92 together) is in the per-patient table. Observation window 1500 s (laryngospasm 900 s after extubation).

"better" = recovers more often, or as often and faster (median), or spends less time with the primary sign deranged, or has a smaller median deficit (shortfall beyond the primary limit summed over the window, e.g. mmHg·s below MAP 65). The aggressive arm is a cap probe, not a treatment; its "efficacy" mostly reflects overshoot.

Per patient: control vs rescue

Emergencies combined with faults and holds (rescue arm)

Findings (illustrative model behaviour; reported, not tuned)

1. Model limit — pre-emergency hypotension at the default plan (documented only; no dose or physiology change, owner decision). With the ga_opioid template (label-default fentanyl, propofol induction, 1.0 × age-adjusted MAC sevoflurane, or propofol TIVA when sevoflurane is locked), MAP was already < 65 just before the onset in 7/12 patients (seed 1, early onset): age18_asa1 63, age90 61, wt40_bmi15 63, asa4_hfref_cad 38, beta_blocker_brady 63, hypovolaemia 48, sepsis_emergency 41 mmHg (processed-EEG index 31–40). Consistent with the documented deep reading of the illustrative depth model at label-default maintenance (TIMING_CONFIG DEPTH_LOW_ADVISORY note). Many hemorrhage/anaphylaxis "recoveries" therefore start from a pre-existing derangement; arms are still compared on the same pre-onset state.
2. Hemorrhage: fluids + accepted label vasopressor advisories do not restore MAP ≥ 65 in every patient class. Not recovered within 1500 s in: asa4_hfref_cad 3/6 (worst MAP 36); sepsis_emergency 3/6 (worst MAP 39). The bleeding never stops in the scenario (holdSec null; surgical control is not modelled), crystalloid offsets at most 80 % of the loss (FLUID_MODEL), the driver gives at most 4 × 500 mL, and the advisory stops at the phenylephrine 200 mcg / ephedrine 50 mg cumulative label maxima ("label maximum reached — escalate; not simulated"). Control: 1/72 runs recovered.
3. Vagal bradycardia: the scenario resolves by itself; glycopyrrolate 0.1 mg (label rescue dose [S8]) helps only a little. Control median time to HR ≥ 50: 119 s; rescue 107 s. Median seconds with HR < 50: control 102, rescue 88; worst nadir HR control 25, rescue 25 /min. Cause: the event lasts 10 s ramp + 90 s + 30 s resolve, and glycopyrrolate uses the generic illustrative PK (no label onset statement; ke0 0.3 /min, effect-site half-time ≈ 2.3 min), so its effect arrives as the event ends. The largest benefit is in the beta-blocker patient (control 239 s vs rescue 131 s). "Stop the stimulus" cannot be modelled (gap).
4. Bronchospasm does not desaturate a ventilated patient at FiO2 0.5. Worst SpO2 across all arms 98 %; the scenario shows as hypercapnia (control worst internal PaCO2 67 mmHg, rescue 58 mmHg). The rescue (FiO2 1.0, deepen once, IV epinephrine 50 mcg when EtCO2 > 45 or SpO2 < 90) keeps PaCO2 below 60 mmHg; deepening alone does little because the hypnotic effect is already near maximal (bronchospasm is countered by 0.4 × hypnosis). The IV epinephrine dose is the perioperative-anaphylaxis guideline dose [S44, S45] — no bronchospasm-specific IV regimen is in the library.
5. Laryngospasm desaturates only the obese patients (FiO2 left at 0.5 through extubation): control worst SpO2 wt200_bmi65_osa 43, obese_osa 47; every other patient stays ≥ 92 % during the 120-s spasm. Rescue (FiO2 1.0, jaw thrust, positive-pressure mask ventilation, which breaks the spasm after 45 s in the scenario) prevents the desaturation. With FiO2 1.0 before extubation (checked while building the matrix) no patient desaturated.
6. Light anaesthesia / strong stimulus is fully blunted at the default maintenance. Control: no patient exceeded MAP 120, HR 120 or index 60 (worst MAP 103, worst HR 118, worst index 49), so efficacy cannot be tested there ("never deranged"). In the "lightened" variant (hypnotic at the bottom of its label range 300 s before the stimulus, every arm) the index exceeded 60 in 21/36 control runs (worst index 83), while MAP stayed ≤ 109 and HR ≤ 115; of those deranged runs the control recovered in 1 (the anaesthetic stays light), the rescue (fentanyl maintenance boluses [S13], deepen once within the regimen range) in 21/21 (median 485 s). The model's stimulus response (GAINS.STIM_MAP_GAIN 0.4 × stimulus blunted by analgesia and hypnosis) cannot take MAP above 120 from these baselines once label-default analgesia is present.
7. Anaphylaxis: IV epinephrine 50 mcg [S44, S45] + fluids recovers fastest; IM 0.3 mg [S40] is slower; the control recovers only when the scripted event ends (control median 1242 s, rescue IV 65 s, rescue IM 214 s). Slowest rescue: sepsis_emergency 495 s, asa4_hfref_cad 468 s, hypovolaemia 236 s. The phenylephrine/ephedrine advisory is suppressed during anaphylaxis by design.
8. Aggressive arm after the owner decisions (repeat refusal, teaching ceilings). Before (commit 21637da) every too-close repeat was accepted with a warning: up to 60.5 mg IM epinephrine, 6.05 mg IV epinephrine and 12.1 mg glycopyrrolate per run (104,616 warned-and-accepted doses in 432 runs). Now, in 432 runs: repeats refused — phenylephrine 4752, epinephrine IM 50976, glycopyrrolate 41400; teaching-ceiling refusals (confirmation required) — epinephrine IV 47952, crystalloid 50334; per run at most epinephrine IM 1.5 mg (one 0.5 mg injection per 5 min), IV 0.55 mg (0.5 mg ceiling + the one deliberately confirmed dose), glycopyrrolate 1 mg. Per-injection caps, the phenylephrine 200 mcg / ephedrine 50 mg cumulative caps and all rate ceilings held as before. Ephedrine ("as needed") and IV epinephrine ("titrated") still have no label interval; ephedrine is limited by its 50 mg label total, IV epinephrine by the teaching ceiling. Midazolam's repeat rule stays warn-only (owner decision; its label wording — "wait an additional 2 or more minutes to fully evaluate the sedative effect" — is titration guidance rather than a stated dosing interval).

Fixes and owner decisions

• Label repeat intervals of rescue boluses were not checked at all (silent). A phenylephrine, glycopyrrolate (rescue) or IM/SC epinephrine bolus repeated sooner than the label interval was accepted with no warning and no record (only the advisory knew the phenylephrine interval). Fix: engine/timing.js TIMING_CONFIG PHENYLEPHRINE_REPEAT_MIN_SEC 60 s [S18], GLYCOPYRROLATE_RESCUE_REPEAT_MIN_SEC 120 s [S8] (not in emergence, where glycopyrrolate goes with neostigmine), EPINEPHRINE_IM_REPEAT_MIN_SEC 300 s [S40] (lower bounds of the label intervals), read by repeatBolusHold / labelRepeatHold (first a warning for manual repeats in commit 21637da, now a refusal — next item); the bolus route is now recorded in the dose log so IM and IV epinephrine are gated separately. Ephedrine ("as needed") and IV epinephrine ("titrated to effect") have no stated interval and are not gated.
• Owner decision: too-close clinician-entered repeats are refused. sim.administer refuses (BLOCKED, logged with the label interval and the seconds remaining) a manual or accepted-advisory repeat of phenylephrine IV < 60 s, glycopyrrolate IV < 120 s outside emergence or IM/SC epinephrine < 300 s after the previous one (engine/timing.js labelRepeatHold); planned repeats keep waiting. Midazolam and the rocuronium top-up stay warn-only.
• Owner decision: simulator teaching ceilings (engine/teaching-ceilings.js, provenance simulator convention, no sourceIds; searched 2026-11-06 — no label or guideline states a cumulative maximum for intraoperative glycopyrrolate or perioperative IV epinephrine boluses; crystalloid has no label dose): IV epinephrine boluses 0.5 mg total (10 × the 50-mcg guideline titration step; = the label IM per-injection maximum), glycopyrrolate 1 mg total (10 × the 0.1 mg rescue dose; = the top of the emergence regimen range), crystalloid 30 mL/kg TBW total (the Surviving Sepsis 2021 initial volume already shown by the fluid model). A dose that would exceed one is refused until the clinician presses "Confirm dose above the simulator teaching ceiling" (confirm_teaching_ceiling:<key>, logged; covers one dose within 60 s). Guided/automatic modes never confirm (a planned step above the ceiling is refused and skipped). The Run screen shows the ceiling and running total next to the dose form and the fluid buttons, with the confirm button after a refusal; the debrief has a "Simulator teaching ceilings" table.

Gaps the engine cannot model (not invented here)

• Stopping the trigger (anaphylaxis: no causative drug is modelled; vagal bradycardia: the surgical stimulus cannot be stopped — both events run on their scripted time course).
• Surgical control of bleeding and blood products (hemorrhage never stops; fluids temporise only).
• Anticholinergic other than glycopyrrolate (no atropine regimen in the library); glycopyrrolate onset uses generic illustrative PK.
• Inhaled bronchodilators, airway pressure / peak pressure, and a bronchospasm-specific epinephrine regimen; only the anaphylaxis guideline IV dose is available.
• A laryngospasm-specific propofol or succinylcholine dose: the scenario resolves on any propofol/succinylcholine dose, but the library has only induction/sedation regimens, so the rescue arm uses airway manoeuvres and oxygen only.
• A sourced antihypertensive (MAP > 120 advisory is text only) and a vasopressor infusion in the advisory (phenylephrine infusion exists in the library but is a manual choice).
• A sourced cumulative maximum for IV epinephrine boluses, glycopyrrolate and crystalloid (simulator teaching ceilings stand in for them); fluid overload.

Patients, plans and strategies

Fictional patients, each built through derivePatient (validation applies). Plans: ga_opioid technique template via buildTemplatePlan (sevoflurane → propofol TIVA fallback for the MH-susceptible patient); hold combinations use ga_opioid_free (dexmedetomidine, ketamine, lidocaine infusions). Every plan passes the plan-safety check.

Arms (every dose from recommend(); label/guideline sources in brackets):

• control — nothing; the rescue advisor is attached but never accepted (it must never dose by itself).
• rescue — hemorrhage: lactated Ringer's 500 mL when MAP < 65 (one at a time, ≤ 4) + every phenylephrine/ephedrine advisory accepted [S18, S19]; vagal bradycardia: glycopyrrolate 0.1 mg IV when HR < 50, ≥ 120 s apart [S8] + advisories; bronchospasm: after 30 s FiO2 1.0 and deepen once (sevoflurane +0.5 % or propofol +50 mcg/kg/min, within the regimen range), epinephrine 50 mcg IV when EtCO2 > 45 or SpO2 < 90, ≥ 120 s apart [S44, S45]; laryngospasm: 5 s after extubation FiO2 1.0, jaw thrust, positive-pressure mask ventilation until resolved and SpO2 ≥ 94 for 30 s; light anaesthesia: when MAP > 120, HR > 120 or index > 60, fentanyl maintenance bolus (≤ 4, ≥ 180 s apart) [S13] and deepen once; anaphylaxis: FiO2 1.0, fluids as hemorrhage, epinephrine 50 mcg IV when MAP < 65 ≥ 120 s apart, after 3 boluses an epinephrine infusion from 0.05 mcg/kg/min IBW, +0.05 every 10 min while MAP < 65 [S40, S41, S44, S45].
• rescue_im (anaphylaxis) — as rescue but epinephrine 0.3 mg IM when MAP < 65, ≥ 300 s apart (label [S40]); no IV epinephrine.
• aggressive — every 5 s for 600 s: the regimen-maximum bolus of phenylephrine, ephedrine, epinephrine IV and IM, glycopyrrolate, a 500 mL fluid bolus and a refused 1000 mL bolus; every 30 s twice a hard cap (phenylephrine cumulative, epinephrine per IV and IM injection, ephedrine cumulative) and twice the ephedrine regimen maximum; at the onset every infusion at twice and at exactly its rate ceiling (epinephrine, phenylephrine, propofol, dexmedetomidine) and the sevoflurane dial at twice its ceiling; at 60 s a direct state write of an epinephrine rate (must be re-validated).

Driver conventions (simulated clinician; not clinical recommendations):

Plausible bounds: the physiology clamps MAP and HR to 25–200 and SpO2 to 40–100 (engine/physiology.js); the asserted bounds are a little wider than these clamps (so a clamp failure would be caught) and everything outside them is non-physiological. Seconds spent at a clamp are reported above as a sign of model saturation, not a pass/fail criterion. Timeline: premed 300 s, induction 360 s (FiO2 1.0), airway secured, maintenance FiO2 0.5, incision after 300 s, onset 300 s (early) or 1200 s (late) after incision; laryngospasm is armed at that time and starts at extubation (at readiness, or readiness + 180 s for "late"), after 1500 s of surgery.