Lesson 1Foundations of sepsis pathophysiology and clinical criteria (SIRS, qSOFA, SOFA, Sepsis-3)This lesson looks at how sepsis affects the body and the body's response, then connects these to signs doctors see at the bedside like low blood pressure, fast heart rate, and failing organs. It compares SIRS, qSOFA, SOFA, and Sepsis-3 rules and how they help sort patients in emergency triage.
Body's response to infection and organ problemsChanges in blood flow and small vessel failuresSIRS, qSOFA, SOFA: parts and limitsSepsis-3 meaning and septic shock rulesLimits of bedside scores in emergenciesLesson 2Safety, failure modes, and mitigation: false positives/negatives, model drift, data quality issues, adversarial inputsThis lesson points out safety dangers like wrong positive or negative alerts, models changing over time, and bad data quality. It looks at handling tricky or surprise inputs, strong monitoring, safety barriers, human checks, and ways to update models safely.
Wrong positives, negatives, and harm waysChecking data quality and spotting odd thingsModel changes, adjusting, and retrainingDealing with tricky or surprise inputsHuman checks, overrides, and rulesLesson 3Evaluation metrics and validation strategies for sepsis prediction: AUROC, AUPRC, calibration, lead time, decision curve analysisThis lesson explains main ways to measure how well sepsis prediction works, including AUROC, AUPRC, how well it matches reality, and warning time. It covers checking inside and outside, time-based checks, and decision curve analysis to see real clinical value.
AUROC, AUPRC, and uneven class problemsMatching curves and risk groupingWarning time and time-specific resultsInside, outside, and time checksDecision curve analysis and net gainLesson 4Feature engineering and temporal modeling: time-series preprocessing, sliding windows, trend extractionThis lesson teaches how to clean and line up emergency department time data for models. It includes adjusting samples, dealing with uneven times, sliding windows, trend and change features, and noting treatments and clinical background over time.
Time lining, adjusting samples, filling gapsSliding windows and prediction timesTrend, change, and rate featuresNoting treatments and care steps upDealing with uneven and thin time dataLesson 5Machine learning models for acute risk prediction: logistic regression, gradient boosted trees, RNNs, temporal convolutional networks, transformer-based time-seriesThis lesson compares ways to model sudden sepsis risk, from simple logistic regression to boosted trees and deep sequence models. It points out strong points, weaknesses, clear explanations, and fit for quick emergency times.
Logistic regression and control choicesBoosted trees and feature valueLooping networks for sequencesTime convolutional networks for time dataTransformers for clinical time dataLesson 6Data modalities for real-time sepsis detection: vitals, labs, nursing notes, medication, waveformsThis lesson describes main real-time data flows in emergencies, like vital signs, lab tests, medicines, nurse notes, and body signal waves. It talks about how often sampled, reliability, and how each shows growing sepsis.
Vital signs and ongoing watch feedsLab sets, cultures, and wait timesMedicine orders, fluids, and pressure drugsNurse notes, start checks, and sheetsWaves from watchers and bed toolsLesson 7Integration with ED workflows and EHR systems: event streams, FHIR, HL7, SMART on FHIR apps, CDS HooksThis lesson shows how AI sepsis models fit into emergency workflows and health records. It reviews event flows, HL7, FHIR parts, SMART on FHIR apps, and CDS Hooks, stressing ease of use, trust, and little change to doctor work.
Event-based setups and data flowsHL7 and FHIR parts for sepsis signsSMART on FHIR apps for bed helpCDS Hooks for aware suggestionsWorkflow lining and ease testingLesson 8Clinical alert design and human factors: thresholds, alarm fatigue mitigation, escalating workflows, who receives alertsThis lesson covers ways to design alerts for emergency doctors, including picking limits, stepped alerts, and sending to right roles. It deals with alert tiredness, timing, step-up paths, and giving reasons and background.
Picking limits and alert stepsAlert tiredness and cut-back waysWho gets alerts and on what linesStep-up workflows and hand-off helpExplaining alerts and giving backgroundLesson 9Regulatory and evidence requirements for diagnostic AI: FDA/CMS considerations, clinical validation study design, prospective pilots, reporting standards (TRIPOD, CONSORT-AI)This lesson outlines rules and proof needs for diagnostic AI in sepsis, including FDA paths, CMS thoughts, and clinical checks. It reviews forward pilots and report rules like TRIPOD and CONSORT-AI.
FDA paths for diagnostic help toolsCMS, payment, and quality plansPlanning strong clinical check studiesForward pilots and step rolloutsTRIPOD and CONSORT-AI report guidesLesson 10Deployment frequency and latency considerations: near-real-time streaming vs batch scoring, handling missing and delayed dataThis lesson talks about setup ways for sepsis models, comparing almost real-time flows with group scoring. It covers time limits, dealing with missing or late data, filling back, and watching data line health in emergencies.
Almost real-time flow vs group scoringTime limits and service promisesFilling for missing and late inputsFilling back, replay, and late dataWatching lines and system strength
