Healthcare Data Science and Clinical Analytics Training Course

Participants can identify healthcare data sources, coding systems and interoperability structures relevant to an analytic question.

• HL7 FHIR resources represent encounters, observations, medications, procedures, and conditions
• OMOP CDM tables organize person, visit, measurement, drug, and observation data
• ICD-10-CM, SNOMED CT, LOINC, and RxNorm support semantic consistency
• Claims data distinguish billed services from clinical findings and outcomes
• Data lineage records source systems, transformations, timestamps, and accountable owners
• Source-to-target mappings expose grain, keys, code sets, and missing fields

Participants can assess data structures and specify reproducible workflows for healthcare analytics datasets.

• Relational schemas identify entities, primary keys, foreign keys, and table grain
• ETL pipelines stage, transform, validate, and publish analytic datasets
• Star schemas organize fact tables and dimensions for reporting
• Reproducible data workflows use configuration files, logs, and immutable extracts
• Role-based access controls restrict identifiable data by approved responsibilities
• Data dictionaries define fields, units, permissible values, and derivation rules

Participants can write SQL queries that create and validate patient-level clinical cohorts.

• SQL SELECT, JOIN, GROUP BY, and window functions build patient cohorts
• Encounter-level tables are joined without duplicating patients or events
• Common table expressions structure readable multi-step clinical queries
• Date logic calculates index dates, washout periods, and follow-up windows
• Code-set tables make diagnosis, procedure, and medication definitions auditable
• Query validation reconciles cohort counts against source-system extracts

Participants can use Python notebooks to prepare, analyse and document healthcare datasets.

• Python notebooks combine pandas, NumPy, matplotlib, and scikit-learn workflows
• Dataframes import CSV extracts and database query results reproducibly
• Functions package repeated cleaning, feature creation, and validation steps
• Version-controlled notebooks separate parameters, code, outputs, and narrative
• Visualizations reveal distributions, missingness, outliers, and clinical subgroup differences
• Reusable scripts generate dated outputs with documented input assumptions

Participants can profile clinical data quality and document issues that could affect analytic conclusions.

• Completeness, conformance, plausibility, and timeliness tests profile clinical datasets
• Missingness mechanisms distinguish MCAR, MAR, and MNAR assumptions
• Range checks identify impossible vital signs, dates, doses, and ages
• Duplicate resolution applies patient, encounter, and event-level matching rules
• Unit harmonization converts laboratory values before longitudinal comparison
• Data quality scorecards assign thresholds, exceptions, owners, and remediation actions

Participants can define auditable computable phenotypes with inclusion, exclusion and temporal rules.

• Computable phenotypes combine code sets, temporal logic, and exclusion criteria
• Index events establish cohort entry, baseline periods, and outcome windows
• Inclusion criteria control clinical relevance, observability, and data availability
• Rule-based algorithms identify diabetes, sepsis, readmission, or medication exposure
• Gold-standard chart review estimates phenotype sensitivity and positive predictive value
• Attrition diagrams document every cohort filter and remaining patient count

Participants can describe clinical populations, outcomes and variation using appropriate exploratory methods.

• Descriptive statistics summarize demographics, utilization, laboratory results, and outcomes
• Stratified analyses expose variation by facility, specialty, deprivation, and ethnicity
• Run charts distinguish common-cause variation from special-cause signals
• Control charts apply appropriate limits to rates, counts, and continuous measures
• Clinical dashboards pair denominators, numerators, definitions, and refresh dates
• Funnel plots compare providers while accounting for differing patient volumes

Participants can design clinical visualisations that communicate evidence, uncertainty and required action.

• Chart selection matches clinical questions to distributions, trends, comparisons, and relationships
• Small multiples reveal subgroup patterns without misleading aggregated summaries
• Color scales distinguish risk categories while meeting accessibility requirements
• Annotation explains denominators, exclusions, confidence intervals, and clinical thresholds
• Dashboard wireframes prioritize actions, escalation routes, and decision timing
• User acceptance criteria test whether displays answer stated clinical questions

Participants can select and interpret statistical tests and regression models for clinical data.

• Confidence intervals communicate estimation uncertainty for clinical and operational measures
• Chi-square, t-tests, and nonparametric tests match variable distributions
• Regression models adjust associations for age, sex, comorbidity, and baseline risk
• Effect sizes report clinical importance alongside statistical significance
• Multiple-testing controls reduce false discoveries across many candidate variables
• Analysis plans pre-specify hypotheses, populations, endpoints, and subgroup tests

Participants can build baseline and machine learning models for defined clinical prediction tasks.

• Feature engineering derives utilization, laboratory trend, medication, and comorbidity variables
• Train-validation-test splits prevent information leakage across patient records
• Logistic regression provides interpretable baselines for binary clinical outcomes
• Random forests and gradient boosting model nonlinear risk patterns
• Class imbalance handling uses weighting, resampling, and threshold selection
• Pipeline design preserves identical preprocessing during training and scoring

Participants can evaluate whether a clinical prediction model is accurate, calibrated, useful and equitable.

• Discrimination metrics include AUROC, precision-recall, sensitivity, specificity, and F1
• Calibration plots compare predicted risk with observed event rates
• Decision curve analysis assesses clinical value across intervention thresholds
• Confusion matrices quantify false alerts, missed cases, and workload
• Subgroup evaluation tests performance by sex, age, ethnicity, and site
• Model cards document intended use, limitations, metrics, and monitoring requirements

Participants can analyse time-to-event outcomes while accounting for censoring and competing risks.

• Kaplan-Meier curves estimate time until death, readmission, or treatment change
• Censoring rules distinguish incomplete follow-up from absence of events
• Cox proportional hazards models estimate adjusted time-to-event associations
• Competing-risk methods address mutually exclusive outcomes such as death
• Landmark analysis prevents immortal-time bias in treatment comparisons
• Survival outputs translate hazard ratios into clinically interpretable evidence

Participants can assess and apply natural language processing methods to unstructured clinical notes.

• Clinical notes require de-identification, section parsing, and abbreviation handling
• Named entity recognition extracts symptoms, medications, findings, and adverse events
• Negation detection separates pneumonia from no evidence of pneumonia
• TF-IDF and embeddings represent note content for downstream models
• Annotation guidelines define labels, adjudication rules, and inter-rater agreement
• Error analysis categorizes extraction failures by note type and context

Participants can assess treatment and intervention effects using observational healthcare data methods.

• Target trial emulation defines eligibility, treatment strategies, assignment, and follow-up
• Directed acyclic graphs identify confounders, mediators, colliders, and adjustment sets
• Propensity scores support matching, weighting, and covariate balance assessment
• Standardized mean differences verify balance after propensity-score adjustment
• Difference-in-differences estimates intervention effects using pre-post comparison groups
• Sensitivity analyses quantify susceptibility to unmeasured confounding assumptions

Participants can analyse patient flow, demand and service changes using healthcare operational data.

• Process mining reconstructs patient pathways from timestamped event logs
• Queueing measures quantify arrival rates, waiting times, capacity, and utilization
• Length-of-stay models separate patient acuity from discharge process delays
• Readmission analyses distinguish planned returns from potentially preventable events
• Demand forecasting estimates admissions, beds, staffing, and theatre requirements
• Interrupted time series evaluates service changes using level and trend shifts

Participants can apply privacy, governance and fairness controls to clinical analytics work.

• Minimum necessary access limits identifiable data to approved purposes
• HIPAA Privacy Rule concepts guide permitted uses and disclosures
• GDPR principles address lawful basis, minimization, retention, and data subject rights
• De-identification methods suppress, generalize, pseudonymize, or perturb patient attributes
• Data protection impact assessments record risks, controls, and residual exposure
• Algorithmic fairness reviews examine disparate impact and benefit allocation

Participants can specify the technical controls required to deploy a clinical analytics product safely.

• Model registries preserve approved versions, training data, and validation evidence
• Containerized services package dependencies for repeatable scoring deployments
• Batch scoring schedules align predictions with clinical decision timing
• API contracts define inputs, outputs, authentication, and error handling
• Release gates require clinical sign-off, security review, and rollback criteria
• Audit logs record prediction requests, versions, users, and downstream actions

Participants can plan adoption of analytics outputs within a defined clinical workflow.

• Workflow co-design maps alerts, responsibilities, escalation, and documentation steps
• Implementation science uses CFIR to assess adoption barriers and facilitators
• Training materials explain model purpose, limits, overrides, and feedback routes
• Pilot protocols define sites, eligibility, success measures, and stop rules
• Benefit realization plans link analytics outputs to measurable service improvements
• Executive briefings state evidence, risk, cost, and operational implications

Participants can create monitoring processes for clinical analytics products after release.

• Data drift metrics detect changes in distributions, coding, and completeness
• Performance monitoring recalculates discrimination, calibration, and alert burden regularly
• Outcome feedback loops capture intervention delivery and patient response
• Incident management classifies failures, assigns owners, and records corrective actions
• Periodic revalidation tests model suitability after clinical pathway changes
• Retirement criteria trigger withdrawal when benefits, safety, or relevance decline

Participants can plan an accountable clinical analytics product from requirements through ongoing review.

• Requirements traceability links decisions, data elements, methods, and acceptance criteria
• RACI matrices assign clinical, technical, privacy, and operational accountabilities
• Risk registers track data, bias, security, adoption, and safety threats
• Milestone plans sequence discovery, validation, deployment, and monitoring activities
• Peer review checklists inspect code, assumptions, outputs, and documentation
• Presentation structures communicate findings, uncertainty, recommendations, and decision requests

Participants can build, review and present a complete clinical analytics case from decision framing through implementation planning.

• Problem framing specifies decision owner, patient population, and measurable outcome
• Source assessment documents provenance, coding standards, quality limitations, and access controls
• Cohort specification applies index dates, eligibility rules, and attrition reporting
• Analytic pipeline produces reproducible tables, visualizations, and model outputs
• Validation evidence reports accuracy, calibration, subgroup performance, and clinical utility
• Implementation brief states workflow, governance, monitoring, and recommended next actions