Lesson 1Basics of Sepsis Body Response and Clinical Signs (SIRS, qSOFA, SOFA, Sepsis-3)This lesson reviews how sepsis affects the body and the patient's response, linking these to signs seen at the bedside like low blood pressure, fast heart rate, and organ problems. It compares SIRS, qSOFA, SOFA, and Sepsis-3 signs and how they are used in sorting patients at emergency triage.
Body's fight against infection and organ failureChanges in blood flow and tiny vessel breakdownsSIRS, qSOFA, SOFA: parts and limitsSepsis-3 meaning and signs of severe septic shockWeaknesses of these signs in emergency sortingLesson 2Safety, Failure Ways, and Fixes: Wrong Positives/Negatives, Model Changes, Data Problems, Harmful InputsThis lesson points out safety dangers like wrong positives, wrong negatives, model changes over time, and bad data quality. It looks at harmful or surprise inputs, strong monitoring, safety barriers, human checks, and steps for safe model updates.
Wrong positives, wrong negatives, and harm typesChecking data quality and spotting odd thingsModel changes, adjusting again, and retrainingDealing with harmful or surprise inputsHuman checks, overrides, and rulesLesson 3Evaluation Measures and Checking Methods for Sepsis Prediction: AUROC, AUPRC, Matching, Lead Time, Decision Curve AnalysisThis lesson defines main performance measures for predicting sepsis, including AUROC, AUPRC, matching to reality, and lead time. It explains checking inside and outside, time-based checking, and decision curve analysis to assess usefulness in clinics.
AUROC, AUPRC, and uneven class problemsMatching curves and risk groupingLead time and time-specific performanceInside, outside, and time-based checkingDecision curve analysis and overall gainLesson 4Feature Building and Time Modeling: Time-Series Cleaning, Sliding Windows, Trend PullingThis lesson explains how to clean and line up emergency time-series data for modeling. It covers re-sampling, dealing with uneven times, sliding windows, trend and change features, and coding treatments and clinic context over time.
Time lining, re-sampling, and filling gapsSliding windows and prediction timesTrend, change, and rate featuresCoding treatments and care steps upDealing with uneven and thin time-seriesLesson 5Machine Learning Models for Sudden Risk Prediction: Logistic Regression, Boosted Trees, RNNs, Time Convolutional Networks, Transformer Time-SeriesThis lesson compares ways to model sudden sepsis risk prediction, from logistic regression to boosted trees and deep sequence models. It highlights strengths, weaknesses, clear understanding, and fit for emergency time limits.
Logistic regression and control choicesBoosted trees and feature weightsRecurrent networks for sequencesTime convolutional networks for time-seriesTransformers for clinic time-series dataLesson 6Data Types for Real-Time Sepsis Spotting: Vitals, Labs, Nursing Notes, Medicines, WaveformsThis lesson describes key real-time data flows in emergencies, including vital signs, lab tests, medicines, nursing records, and body wave signals. It discusses sampling speeds, trust levels, and how each type shows growing sepsis.
Vital signs and ongoing monitor feedsLab sets, cultures, and wait timesMedicine orders, fluids, and pressure drugsNursing notes, start text, and record sheetsWaveforms from monitors and bed devicesLesson 7Fitting into Emergency Workflows and Health Record Systems: Event Flows, FHIR, HL7, SMART on FHIR Apps, CDS HooksThis lesson describes 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 clinic work.
Event-based setups and data flowsHL7 and FHIR parts for sepsis signsSMART on FHIR apps for bed decision helpCDS Hooks for aware suggestionsWorkflow mapping and ease testingLesson 8Clinical Alert Design and Human Factors: Limits, Alarm Weariness Fix, Step-Up Workflows, Who Gets AlertsThis lesson covers rules for alert design for emergency doctors, including limit choices, stepped alerts, and sending to right roles. It addresses alarm weariness, alert timing, step-up paths, and showing reasons and context.
Picking limits and alert levelsAlarm weariness and stop strategiesWho gets alerts and on which waysStep-up workflows and handoff helpExplaining alerts and giving contextLesson 9Rules and Evidence Needs for Diagnostic AI: FDA/CMS Thoughts, Clinic Check Study Design, Forward Tests, Report Standards (TRIPOD, CONSORT-AI)This lesson outlines rule and evidence hopes for diagnostic AI in sepsis, including FDA paths, CMS thoughts, and clinic checks. It reviews forward tests and report standards like TRIPOD and CONSORT-AI.
FDA paths for diagnostic help toolsCMS, payment, and quality plansDesigning strong clinic check studiesForward tests and step rolloutsTRIPOD and CONSORT-AI report guidesLesson 10Deployment Speed and Delay Thoughts: Near-Real-Time Flow vs Batch Scoring, Dealing with Missing and Late DataThis lesson discusses setup ways for sepsis models, comparing near-real-time flow with batch scoring. It addresses delay limits, handling missing or late data, filling back, and watching data line health in emergencies.
Near-real-time flow vs batch scoringDelay limits and service promisesFilling for missing and late inputsBackfilling, replay, and late dataWatching lines and system strength
