SepsiGuard · Clinical evidence pack
Hard sepsis calls, one dependable verdict each.
A curated compendium for intensivists and ED physicians: challenging sepsis-recognition and bundle-timing cases run through SepsiGuard — each shown as the clinician’s pain-point and the system’s genuine, brain-concealed output (verdict + directives; the scoring internals are withheld by design). The cases were chosen where trust is earned: septic shock, a negative qSOFA, non-clearing lactate, bundle ordering, and the sepsis mimics. SepsiGuard recommends the action; the antimicrobial dose is deferred to DoseOptimize.
How to read each verdict✅ verdict firm deterministic Sepsis-3 classification
⚑ directive the bundle action (abx <1 h · MAP ≥ 65 · cultures-first · 30 mL/kg)
⚠ caution / monitor lactate not clearing, fluid-overload, de-escalation
⚠ red flag / mimic shock, ordering violation, or a firewalled sepsis-mimic suspicion
SG-01Septic shock · expanded fluid space
SG-02qSOFA negative, organ dysfunction present
SG-03Lactate not clearing
SG-04Antibiotic-timing urgency
SG-05MAP target — why 65, not higher
SG-06Vasopressor-dependent shock · steroids
SG-07Bundle ordering violation
SG-08Mimic · TTP (PLASMIC) — firewalled
SG-09Mimic · HLH (HScore) — firewalled
SG-10Mimic · tropical suspicion (unvalidated)
SG-11Fluid-overload caution · de-escalation
SG-PNDeterminism — same case, same verdict
Try it yourself — self-submission kitFill the template with de-identified values (alter any field to explore the behaviour) and submit it at your invited intake link. Every submission is reviewed & signed by a registered clinician before an advisory is returned — there is no instant machine output, and the released advisory shows verdict + directives only.
SG-01Septic shock · expanded fluid space
58 y M · 80 kgPneumonia + septic shockLactate 5.2 · MAP 58On vasopressors
The pain-point“Shock with a near-normal creatinine — what actually has to happen this first hour?”
SepsiGuardSEPTIC SHOCK30 mL/kg crystalloid · norepinephrine to MAP ≥ 65 · broad-spectrum antibiotics < 1 h · cultures before antibiotics
SSC 2021 Hour-1 bundle
✅ verdict firm⚑ antibiotics < 1 h⚑ MAP ≥ 65dose → DoseOptimize
Why it earns trustIt names the whole Hour-1 bundle deterministically — fluids, pressor target, cultures-first, antibiotics within the hour — and hands the antibiotic dose to DoseOptimize rather than guessing it. The verdict is reproducible for identical inputs.
SG-02qSOFA negative, organ dysfunction present
71 y F · 62 kgPyelonephritisqSOFA 1 · SOFA rise ≥ 2Clinician worried
The pain-point“qSOFA is only 1 — is it safe to stand down?”
SepsiGuardSEPSIS — do not stand downsuspected infection + organ dysfunction (SOFA rise ≥ 2) defines sepsis regardless of qSOFA
Sepsis-3 (Singer 2016)
⚠ qSOFA insensitive early✅ Sepsis-3 met⚑ start SSC bundle
Why it earns trustqSOFA is a prompt to escalate, not a rule-out. The tool holds the line on the Sepsis-3 definition and starts the bundle — the case where reassurance would have been the error.
SG-03Lactate not clearing
64 y M · 78 kgAbdominal sepsisLactate 5.2 → 4.9 / 6 hFluids given
The pain-point“Lactate has barely moved at 6 h — do I just give more fluid?”
SepsiGuardLactate NOT clearingreassess fluid responsiveness, MAP target and source control — not reflexive fluid
SSC lactate-guided resuscitation
⚠ ongoing hypoperfusion⚑ reassess source controlfluid caution
Why it earns trustFailure to clear lactate is flagged as ongoing hypoperfusion that points to source control and dynamic reassessment — the tool explicitly resists the reflexive extra fluid bolus.
SG-04Antibiotic-timing urgency
52 y M · 74 kgSeptic shockCultures pendingHypotensive
The pain-point“Cultures aren’t back — should I wait before giving antibiotics?”
SepsiGuardAntibiotics within 1 h — nowobtain cultures first only if it does not substantially delay antibiotics
Kumar 2006
⚑ antibiotics < 1 h⚠ each hour delay ~7.6% mortalitydose → DoseOptimize
Why it earns trustIn shock, the strongest modifiable determinant of survival is time to effective antibiotics — the tool refuses to trade that for a complete culture set.
SG-05MAP target — why 65, not higher
69 y M · 82 kgSeptic shockChronic hypertensionOn norepinephrine
The pain-point“He’s a chronic hypertensive — shouldn’t I target a higher MAP?”
SepsiGuardTarget MAP ≥ 65 mmHgno mortality benefit at 80–85 vs 65–70; more atrial fibrillation at the higher target
SEPSISPAM (Asfar 2014)
⚑ MAP ≥ 65✅ trial-anchored
Why it earns trustIt defends the default target against a plausible-sounding higher one, citing the trial that tested exactly this — and notes the chronic-hypertension nuance without abandoning the evidence.
SG-06Vasopressor-dependent shock · steroids
60 y F · 70 kgSeptic shockAdequately fluid-resuscitatedOngoing pressor need
The pain-point“Still on pressors after adequate fluids — is it time for steroids?”
SepsiGuardHydrocortisone indicatedseptic shock with an ongoing vasopressor requirement — shortens time to shock reversal
ADRENAL / APROCCHSS (2018)
⚑ steroids indicated✅ trial-anchored
Why it earns trustThe steroid prompt fires on the right trigger — persistent vasopressor requirement after fluids — rather than on shock alone.
SG-07Bundle ordering violation
57 y M · 76 kgSepsisAntibiotics charted before culturesNo delay concern
The pain-point“Antibiotics went in before cultures were drawn — does the tool even notice?”
SepsiGuardOrdering violation flaggeddraw blood cultures before antibiotics unless doing so would substantially delay them
SSC 2021 bundle sequencing
⚠ ordering violation⚑ cultures-first
Why it earns trustThe bundle engine checks sequence, not just presence — it surfaces cultures-after-antibiotics as a violation so the next case is drawn correctly.
SG-08Mimic · TTP (PLASMIC) — firewalled
44 y F · 65 kgFever + thrombocytopeniaHemolysis, low plateletsIs this TTP?
The pain-point“Low platelets, hemolysis, shock — is this TTP rather than sepsis?”
SepsiGuardSEPSIS (verdict unchanged) · PLASMIC suspicion raisedthe mimic layer prompts the TTP differential — it does not change the Sepsis-3 verdict
Firewalled parallel layer
⚠ PLASMIC — differential raised✅ verdict unchanged (firewall)
Why it earns trustDeterminism wraps probability: the PLASMIC screen raises the differential for haematology input, while an invariant guarantees it can never move the diagnosis. You get the prompt without the diagnosis drifting.
SG-09Mimic · HLH (HScore) — firewalled
39 y M · 72 kgPersistent feverVery high ferritinCytopenias
The pain-point“Sky-high ferritin, cytopenias, fever that won’t settle — could this be HLH?”
SepsiGuardSEPSIS (verdict unchanged) · HScore suspicion raisedprompts the HLH differential; the verdict and directives are unchanged
Firewalled parallel layer
⚠ HScore — differential raised✅ firewalled from the verdict
Why it earns trustThe HScore prompt nudges you toward the HLH work-up in the patient who isn’t behaving like plain sepsis — without the tool silently reclassifying the case.
SG-10Mimic · tropical suspicion (unvalidated)
33 y M · 68 kgEndemic areaFever + thrombocytopeniaDengue / scrub typhus?
The pain-point“Endemic region, fever, low platelets — is this a tropical infection, not bacterial sepsis?”
SepsiGuardSEPSIS (verdict unchanged) · tropical suspicion flaggedexpert heuristic — unvalidated — a differential prompt only
Firewalled · explicitly unvalidated
⚠ tropical suspicion (unvalidated)✅ never moves the verdict
Why it earns trustThe tool is honest about its weakest signal: the tropical heuristic is labelled unvalidated everywhere it appears and is firewalled from the diagnosis — a prompt to think, not a claim to trust.
SG-11Fluid-overload caution · de-escalation
66 y M · 84 kgSepsis, resuscitated3 L in · new hypoxia48–72 h mark
The pain-point“Three litres in and now he’s hypoxic — keep resuscitating, or change course?”
SepsiGuardFluid-overload caution · reassesscheck dynamic responsiveness before more fluid; review source control; de-escalate antibiotics at 48–72 h
Safety + trajectory layer
⚠ fluid-overload watch⚑ de-escalate 48–72 hdose → DoseOptimize
Why it earns trustPast the resuscitation window the tool switches register — caution against more fluid, prompt de-escalation and stewardship — and still leaves the antibiotic dose to DoseOptimize.
SG-PNDeterminism — same case, same verdict
58 y M · 80 kgSeptic shock (case 01, re-entered)Independently submittedIdentical inputs
The pain-point“Would a different clinician entering the same patient get the same verdict?”
SepsiGuardSEPTIC SHOCK — identicalsame inputs, independently entered · same verdict and directives
Reproducible · operator-independent
✅ reproducible✅ mimic layer never touched the verdict
Why it earns trustAn independently re-entered, identical picture returns the identical verdict — reproducible and auditable, and the firewall means the probabilistic layer never moved it. Not a matter of who typed it in.