Python for Public Health Research Training Course

Participants can inspect, select and reshape health datasets using pandas data frames.

• pandas Series and DataFrames for participant, facility, and event records
• Inspect dimensions, column types, unique values, and descriptive record counts
• Select rows and columns using Boolean conditions, loc, and iloc
• Create derived variables from dates, categories, numeric values, and text
• Sort, rename, recode, and reorder variables without altering source files
• Reshape wide and long data for repeated measurements and indicator reporting

Participants can apply documented checks to identify and correct common health data defects.

• Missingness profiles by variable, facility, reporting period, and population group
• Range checks for ages, dates, laboratory values, doses, and vital signs
• Detect duplicate persons, visits, notifications, and specimen identification numbers
• Standardise text fields, categorical codes, dates, and administrative location names
• Cross-field validation between diagnosis, treatment, outcome, and follow-up dates
• Cleaning logs recording transformations, exclusions, rationale, and unresolved anomalies

Participants can join and aggregate related health datasets while preserving analytical traceability.

• Merge person, visit, laboratory, and facility files using documented keys
• Assess match rates, unmatched records, duplicate keys, and join consequences
• Concatenate monthly extracts while checking schema consistency and reporting completeness
• Group records by age, sex, geography, service, and reporting period
• Aggregate counts, means, medians, proportions, and denominators for indicators
• Create analysis-ready datasets with explicit inclusion and exclusion criteria

Participants can calculate and interpret core descriptive and epidemiological measures in Python.

• Person, place, and time summaries for descriptive epidemiological investigation
• Calculate counts, proportions, rates, risks, prevalence, and incidence measures
• Construct denominators from census, catchment, enrolment, and person-time data
• Direct age standardisation using reference populations and age-specific rates
• Compute case-fatality ratio, mortality rate, and standardised mortality ratio
• Stratify measures by demographic factors, exposure groups, geography, and time

Participants can summarise public health data and quantify uncertainty using appropriate statistical methods.

• Distributional summaries using mean, median, standard deviation, quantiles, and IQR
• Visual checks for skewness, outliers, heaping, and implausible values
• Confidence intervals for proportions, means, rates, risk ratios, and rate ratios
• Bootstrap confidence intervals for statistics without straightforward parametric assumptions
• Compare groups using t-tests, chi-square tests, Fisher's exact test, and Mann-Whitney tests
• Distinguish statistical significance, practical importance, precision, and public health relevance

Participants can design clear charts that communicate patterns, uncertainty and comparisons responsibly.

• Matplotlib and seaborn charts for distributions, groups, trends, and relationships
• Epidemic curves showing onset dates, reporting delays, and outbreak progression
• Rate charts using appropriate denominators, scales, labels, and uncertainty intervals
• Small multiples comparing regions, facilities, age groups, and reporting periods
• Colour choices supporting accessible interpretation and avoiding misleading categorical emphasis
• Publication-ready figures with titles, captions, source notes, and consistent formatting

Participants can translate public health research questions into defensible analytical specifications.

• PICO framing for intervention, exposure, outcome, comparator, and target population
• Cross-sectional, cohort, case-control, ecological, and interrupted time-series study designs
• Selection bias, information bias, recall bias, and misclassification mechanisms
• Directed acyclic graphs for confounder identification and adjustment decisions
• Define index dates, follow-up windows, censoring rules, and outcome ascertainment
• Analysis plans linking research questions, variables, estimands, and model choices

Participants can fit and interpret regression models for common public health outcome types.

• Linear regression assumptions, residual diagnostics, transformations, and interpretation limits
• Logistic regression for binary outcomes, odds ratios, predicted probabilities, and confidence intervals
• Poisson regression for counts with population offsets and incidence rate ratios
• Negative binomial models for overdispersed surveillance and service-use counts
• Confounding assessment through crude, adjusted, and stratified effect estimates
• Interaction terms for effect modification by age, sex, exposure, or geography

Participants can evaluate model adequacy and report regression results transparently.

• Check linearity, residual patterns, influential observations, and multicollinearity diagnostics
• Variance inflation factors and correlation matrices for predictor screening
• Assess logistic model calibration, discrimination, classification thresholds, and residuals
• Compare nested models using likelihood ratio tests and information criteria
• Generate adjusted predictions, marginal effects, and interpretable effect plots
• Regression tables reporting estimates, intervals, reference groups, and sample exclusions

Participants can analyse time-to-event outcomes with censoring and follow-up information.

• Define time origin, event date, censoring date, and observation periods
• Kaplan-Meier curves for retention, treatment outcomes, mortality, and recovery
• Log-rank tests comparing survival distributions across exposure and treatment groups
• Cox proportional hazards models with hazard ratios and confidence intervals
• Test proportional hazards assumptions using Schoenfeld residuals and visual diagnostics
• Competing risks considerations for mortality, discharge, transfer, and treatment outcomes

Participants can prepare and interpret routine surveillance time series for public health action.

• Aggregate weekly and monthly counts with complete reporting-period calendars
• Seasonality, trend, anomalies, and reporting artefacts in routine surveillance data
• Moving averages and seasonal decomposition for exploratory disease pattern analysis
• Calculate baselines and thresholds for simple aberration detection methods
• Interrupted time-series models for policy, campaign, and service-change evaluation
• Interpret surveillance signals alongside case definitions, testing volume, and reporting delays

Participants can prepare geographic health data and create defensible area-level maps.

• GeoPandas objects, coordinate reference systems, geometries, and attribute joins
• Join case counts and denominators to administrative boundary datasets
• Choropleth maps using rates, classification choices, and suppression rules
• Point maps for facilities, cases, environmental exposures, and service locations
• Spatial aggregation, geocoding limitations, boundary changes, and modifiable areal unit concerns
• Map interpretation avoiding causal claims from area-level association patterns

Participants can create an auditable Python research workflow suitable for review and reuse.

• Research project conventions for scripts, notebooks, parameters, outputs, and archived inputs
• Git repositories, commits, branches, diffs, and meaningful commit messages
• Requirements files recording Python versions and package dependencies
• Functions reducing repeated code in data preparation and indicator calculations
• Automated checks for expected columns, valid values, and output totals
• README files documenting data access, workflow order, assumptions, and limitations

Participants can manage analytical work with appropriate privacy, governance and reporting controls.

• Data minimisation, pseudonymisation, access controls, retention, and secure storage
• Disclosure control for small cells, rare events, sensitive groups, and maps
• Ethics approvals, data-sharing agreements, consent limitations, and secondary data use
• STROBE reporting items for observational study methods, results, and limitations
• Provenance records linking analytical outputs to authorised source data
• Communicate uncertainty, bias, missingness, and generalisability without overstating conclusions

Participants can build, review and present a complete reproducible Python analysis addressing a public health research question.

• Frame a decision-focused research question using available public health data
• Build an auditable pipeline from source data through final analytical outputs
• Apply cleaning rules, descriptive measures, visualisations, and appropriate statistical models
• Document assumptions, missingness decisions, validation findings, and analytical limitations
• Review code, tables, figures, and interpretations against STROBE reporting items
• Present findings, recommendations, uncertainty, and next analytical steps to stakeholders