Epidemiology and Biostatistics Using Stata Training Course

Participants can select study designs and sampling strategies suited to defined public health questions.

• Cross-sectional, case-control, cohort and trial designs compared systematically
• Target population, study population and sampling frame specified precisely
• Probability and non-probability sampling methods assessed for selection bias
• Simple random, stratified, cluster and multistage sampling designs contrasted
• Sample size components linked to precision, power and expected effect
• Eligibility criteria written to support valid participant selection

Participants can import, document and clean health data using traceable Stata procedures.

• Excel, CSV and delimited files imported with verified variable formats
• Codebooks produced using describe, codebook, summarize and tabulate commands
• Missing values, impossible values and duplicate records identified systematically
• Value labels and variable labels applied for interpretable outputs
• Dates converted, checked and formatted using Stata date functions
• Cleaning decisions recorded within annotated and reusable do-files

Participants can restructure and derive variables required for epidemiological analysis.

• Generate and replace commands create derived epidemiological variables safely
• Recode methods create meaningful categorical exposure and outcome measures
• Data reshaping uses reshape wide and reshape long appropriately
• Merge and append commands combine facility, patient and survey datasets
• Collapse and contract commands construct aggregated indicator datasets
• Preserve, restore and tempfile commands protect source data integrity

Participants can describe person, place and time patterns using appropriate Stata outputs.

• Frequency tables report counts, proportions and valid denominators clearly
• Central tendency and dispersion summarise continuous health measurements appropriately
• Distributional shape assessed using histograms, box plots and quantile summaries
• Person, place and time patterns structured for descriptive epidemiology
• Two-way tables compare exposure, outcome and subgroup distributions
• Stata graphs communicate trends, disparities and unusual data patterns

Participants can calculate and interpret disease frequency measures using correct numerators and denominators.

• Point and period prevalence calculated from cross-sectional population data
• Cumulative incidence estimated for defined populations at risk
• Incidence rates calculated using person-time denominators and follow-up
• Mortality, case fatality and proportionate mortality distinguished correctly
• Attack rates used to characterise outbreaks in defined groups
• Denominator definitions checked against population and service utilisation data

Participants can estimate measures of association and population impact from epidemiological data.

• Risk ratio, rate ratio and odds ratio calculated correctly
• Risk difference and attributable risk interpreted for programme decisions
• Population attributable fraction estimated using exposure prevalence assumptions
• Crude rates compared before considering demographic population structure
• Direct age standardisation applies external standard population weights
• Standardised mortality ratios calculated using indirect standardisation methods

Participants can use statistical inference to quantify uncertainty and test defined hypotheses.

• Probability distributions linked to binary, count and continuous outcomes
• Sampling distributions explain standard errors and confidence intervals
• Null and alternative hypotheses specified before statistical testing
• Type I and Type II errors related to public health decisions
• t-tests, chi-square tests and non-parametric tests selected appropriately
• P-values interpreted alongside effect estimates and confidence intervals

Participants can identify threats to validity and plan analyses that address confounding.

• Selection bias, information bias and recall bias distinguished clearly
• Confounding assessed through subject knowledge and causal assumptions
• Directed acyclic graphs identify adjustment sets and inappropriate controls
• Stratified analysis compares crude and adjusted association estimates
• Mantel-Haenszel methods estimate pooled adjusted risk and odds ratios
• Effect modification assessed using stratum-specific estimates and interaction terms

Participants can fit, diagnose and interpret linear regression models for continuous health outcomes.

• Linear regression models mean differences in continuous outcome variables
• Regression coefficients interpreted with units, direction and uncertainty
• Dummy variables represent categorical predictors with reference categories
• Residual plots assess linearity, normality and constant variance assumptions
• Influential observations examined using leverage, Cook's distance and residuals
• Adjusted predictions communicate modelled outcome differences across groups

Participants can build and interpret logistic regression models for binary health outcomes.

• Logistic regression models binary outcomes using log odds
• Odds ratios interpreted with confidence intervals and reference groups
• Univariable screening distinguished from theory-driven model specification
• Factor-variable notation specifies categorical predictors and interaction effects
• Goodness-of-fit assessed using calibration and classification measures
• Marginal predicted probabilities communicate adjusted findings accessibly

Participants can model count outcomes and incidence rates using appropriate regression methods.

• Poisson regression models event counts and incidence rate ratios
• Exposure time incorporated through offsets in rate models
• Overdispersion assessed using deviance, Pearson statistics and residual patterns
• Negative binomial regression addresses overdispersed count outcome data
• Zero-inflated models considered when structural zeros are plausible
• Rate model findings translated into expected event burden differences

Participants can analyse time-to-event outcomes using Kaplan-Meier and Cox regression methods.

• Time origin, event definition and censoring rules specified explicitly
• stset declares survival time, failure events and censoring records
• Kaplan-Meier curves estimate event-free survival across comparison groups
• Log-rank tests compare survival distributions without covariate adjustment
• Cox proportional hazards models estimate adjusted hazard ratios
• Proportional hazards assumptions assessed using Schoenfeld residual methods

Participants can analyse complex survey data while accounting for the sampling design.

• Survey weights correct unequal selection probabilities and non-response adjustments
• Primary sampling units, strata and finite population corrections identified
• svyset declares survey design variables before estimation commands
• Taylor series linearisation estimates design-based standard errors
• Weighted proportions and means reported with valid confidence intervals
• Subpopulation analysis avoids incorrect restriction of survey datasets

Participants can assess missingness, document data quality issues and select defensible responses.

• Missing completely at random, MAR and MNAR assumptions distinguished
• Missingness patterns explored by outcome, exposure and participant characteristics
• Complete-case analysis limitations assessed against likely missingness mechanisms
• Multiple imputation by chained equations introduced for appropriate datasets
• Imputation diagnostics compare observed and imputed variable distributions
• Data quality reports document completeness, consistency and timeliness findings

Participants can evaluate diagnostic test performance and interpret screening results.

• Sensitivity and specificity calculated against a defined reference standard
• Positive and negative predictive values linked to disease prevalence
• Likelihood ratios quantify diagnostic evidence from test results
• Receiver operating characteristic curves assess continuous test thresholds
• Area under the curve summarises discrimination performance
• Screening programme interpretation considers false positives and false negatives

Participants can produce documented Stata analyses and publication-quality epidemiological outputs.

• Project folder structures separate raw data, code, outputs and documentation
• Do-files use comments, macros and version control statements
• Table shells specify denominators, estimates, uncertainty and comparison groups
• Regression results formatted into interpretable tables and forest plots
• STROBE reporting items applied to observational study results
• Methods and limitations written with accurate statistical interpretation

Participants can critically appraise epidemiological analyses before using findings for decisions.

• Research questions checked against study design and analytic approach
• Internal validity assessed through bias, confounding and measurement criteria
• Precision evaluated using sample size, intervals and estimate stability
• Clinical and public health significance distinguished from statistical significance
• External validity considered against target population and programme context
• Recommendations tested against findings, uncertainty and implementation constraints

Participants can build, review and present a complete Stata-based epidemiological analysis addressing a public health question.

• Analysis question, target population and estimand defined for capstone dataset
• Stata do-file documents import, cleaning, transformation and analysis decisions
• Descriptive tables characterise participants, exposures, outcomes and missingness
• Regression model selected, diagnosed and interpreted against causal assumptions
• Results report presents estimates, uncertainty, limitations and recommendations
• Peer review challenges methods, conclusions and public health implications