Scenario catalogue

The module ships with 65 scenarios, one clinical question each. Most replicate a real study: 29 were estimated from open patient-level data, 7 from teaching datasets of the author’s own studies, and the others from the published tables of the study. Use the catalogue below to find one for a domain, a study design or a teaching point. In jamovi, all of them are in the scenario picker of Simulated study and Compare designs, grouped by domain.

How the scenarios were made

From open data. Many classic studies are distributed as datasets with R packages (survival, MASS, datasets, HSAUR3, TH.data, pROC, medicaldata, aplore3, NHANES). For each one the module’s own Derive scenario engine estimated:

  • the distribution of every baseline variable (normal, log-normal, gamma, counts, categories);
  • their correlations, on a latent scale (a Gaussian copula);
  • the conditional effects on the outcome, from the model the study would use (logistic, Cox, linear, Poisson or negative binomial).

When a study had few events for its number of variables, the effects were shrunk towards no effect by one common factor, so the confounding between them is kept.

From teaching datasets. Several come from datasets of the author’s own studies used for years in teaching (cervical cancer and HPV, microsatellite instability, colorectal cancer and diet, relapse of stage II colon cancer, dyspepsia, cardiac surgery, cirrhosis), estimated the same way; the scenario keeps their parameters, not their data.

From a published paper. When no data are open, the parameters come from the paper’s tables: the baseline characteristics (Table 1), the frequency of the outcome, and the adjusted estimates of the main analysis. A few papers were chosen for critical reading: their scenarios reproduce what the paper reports, errors included, and the teaching note points the errors out.

Finishing by hand. Data cannot tell causes from correlates, so each scenario was then given:

  • a causal structure (which variable affects which: confounders, mediators);
  • reference categories and units;
  • the designs that make sense for the question;
  • labels and column names in English, Spanish and Catalan;
  • a teaching note.

Variables that matter in the literature but had no effect in the data are kept as null variables, useful for teaching multiple testing and confounding.

Checking. For every scenario and design, a large simulated study was analysed as a student would analyse it. Its estimates were compared with the values the data were generated with and with the source’s main result. Examples:

  • the alcohol odds ratios of the Ille-et-Vilaine study;
  • the attenuated hazard ratio of local histology in the Wilms tumour study;
  • the vaccine efficacy of the Salk trial;
  • the AUC of high-sensitivity troponin.

What a scenario is, and is not. A scenario stores parameters, never the original data. The data it produces resemble the source study in their distributions, associations and main results, so analyses give the same conclusions, but no row corresponds to a real person. The effect sizes are teaching values that reproduce the source; they are not meta-analytic estimates, and they should not be quoted as evidence. Every scenario lists its sources, and the original investigators deserve the credit for the data.

Catalogue

Filter by any column, or type in the search box (it searches names, populations, outcomes and teaching notes). Click a column header to sort, and click a row for its population, outcomes, teaching note and references.

Scenario Domain Designs Outcome Source Variables

Coverage by design

Number of scenarios in the current selection that can be simulated with each design.

Choosing a scenario

  • For a design, filter by design. Most scenarios fit several designs, and Compare designs simulates them from the same population: the classic contrasts are a case-control vs a cohort study for a rare cancer, an ecological vs an individual analysis (water hardness, dietary fat), or a parallel vs a crossover trial (hypnotics, asthma).
  • For a statistical idea, use the features and the search box:
    • confounding: free light chains, lung cancer and smoking, NHANES;
    • mediation: myopia, birth weight;
    • effect modification: RECOVERY, ISIS-2, breast cancer risk;
    • competing risks: MGUS, PBC, colon cancer;
    • misclassification: Wilms tumour;
    • overdispersion: epilepsy;
    • dropout: Beat the Blues;
    • overfitting with many markers: glaucoma;
    • rare outcomes and sample size: polio vaccine.
  • For dentistry and podiatry courses, filter the domains Odontology (periodontal treatment in pregnancy, fluoride varnish, periodontitis) and Podiatry (diabetic foot ulcer, onychomycosis, plantar fasciitis, ankle- and toe-brachial index).
  • For your own course, export a scenario from Library as Excel or YAML, change it, and load it back; or derive a new one from your own data with Derive scenario.