Lesson 1Foundations of sepsis pathophysiology and clinical criteria (SIRS, qSOFA, SOFA, Sepsis-3)This lesson reviews sepsis biology and the body's response to infection, linking these to signs at the bedside like low blood pressure, fast heart rate, and organ failure. It compares SIRS, qSOFA, SOFA, and Sepsis-3 criteria and their role in sorting patients in emergency departments.
Body's response to infection and organ failureBlood flow changes and small vessel failureSIRS, 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 identifies safety risks like false alarms, missed cases, model changes over time, and bad data quality. It explores handling tricky or surprise inputs, strong monitoring, safety barriers, human checks, and safe ways to update models.
False alarms, missed cases, and harm typesData quality checks and oddity spottingModel shift, recalibration, and retrainingHandling tricky or surprise inputsHuman checks, overrides, and oversightLesson 3Evaluation metrics and validation strategies for sepsis prediction: AUROC, AUPRC, calibration, lead time, decision curve analysisThis lesson defines main performance measures for sepsis prediction, including AUROC, AUPRC, calibration, and early warning time. It explains internal and external checks, time-based validation, and decision curve analysis for real-world usefulness.
AUROC, AUPRC, and uneven class balanceCalibration graphs and risk groupingEarly warning time and specific horizon performanceInternal, external, and time-based validationDecision curve analysis and overall gainLesson 4Feature engineering and temporal modeling: time-series preprocessing, sliding windows, trend extractionThis lesson explains cleaning and lining up emergency department time-series data for modelling. It covers resampling, dealing with uneven gaps, sliding windows, trend and variation features, and coding treatments and clinical background over time.
Time lining up, resampling, and filling gapsSliding windows and prediction periodsTrend, variation, and change featuresCoding treatments and care steps upDealing with uneven and thin time-seriesLesson 5Machine learning models for acute risk prediction: logistic regression, gradient boosted trees, RNNs, temporal convolutional networks, transformer-based time-seriesThis lesson compares modelling ways for sudden sepsis risk prediction, from logistic regression to boosted trees and deep sequence models. It highlights strengths, weaknesses, clarity, and fit for emergency time limits.
Logistic regression and control choicesBoosted trees and feature valueRecurrent networks for sequencesTime convolutional networks for time-seriesTransformers for clinical time-series dataLesson 6Data modalities for real-time sepsis detection: vitals, labs, nursing notes, medication, waveformsThis lesson describes key real-time data flows in the emergency department, including vital signs, lab tests, medications, nursing records, and body signal waves. It discusses sampling speeds, trustworthiness, and how each type shows developing sepsis.
Vital signs and ongoing monitoring feedsLab panels, cultures, and result timesMedication orders, fluids, and pressure drugsNursing notes, intake text, and record sheetsWaves from monitors and bedside toolsLesson 7Integration with ED workflows and EHR systems: event streams, FHIR, HL7, SMART on FHIR apps, CDS HooksThis lesson describes how AI sepsis models fit into emergency workflows and electronic health records. It reviews event flows, HL7, FHIR resources, SMART on FHIR apps, and CDS Hooks, stressing ease of use, reliability, and little upset to clinical work.
Event-based setups and data flowsHL7 and FHIR resources for sepsis signsSMART on FHIR apps for bedside helpCDS Hooks for aware suggestionsWorkflow mapping and ease testingLesson 8Clinical alert design and human factors: thresholds, alarm fatigue mitigation, escalating workflows, who receives alertsThis lesson covers alert design principles for emergency clinicians, including threshold picking, tiered alerts, and sending to right roles. It addresses alert tiredness, timing, step-up paths, and showing reasons and background.
Picking thresholds and alert levelsAlert tiredness and cut-back waysWho gets alerts and on which pathsStep-up workflows and handover 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 regulatory and proof needs for diagnostic AI in sepsis, including FDA paths, CMS thoughts, and clinical checks. It reviews forward pilots and reporting rules like TRIPOD and CONSORT-AI.
FDA paths for diagnostic help toolsCMS, payment, and quality plansDesigning strong clinical check studiesForward pilots and step rolloutsTRIPOD and CONSORT-AI reporting guidesLesson 10Deployment frequency and latency considerations: near-real-time streaming vs batch scoring, handling missing and delayed dataThis lesson discusses setup ways for sepsis models, comparing near-real-time flows with batch scoring. It addresses time limits, handling missing or late data, filling back, and watching data line health in the emergency department.
Near-real-time flows vs batch scoringTime limits and service promisesFilling in for missing and late inputsBackfilling, replay, and late dataWatching lines and system strength
