Reporting takes what the earlier steps produced and writes it up. Reporting is a separate module, and you run it when you choose to.
The input is the export bundle that Model Coding produces. The bundle may already reflect manual changes made since the model was generated, so the report describes the model as it stands.
There are two engines, Excel and R, so there are two start cards.
Four optional fields cover what a model file cannot know: indication, title, author and a willingness to pay threshold. If neither the model nor the field provides an indication, the report discloses that.
Left blank, the threshold is taken from the model's own results when they declare one. If neither is available, the report states that no threshold was specified.
The sensitivity analysis slot is optional. When it is omitted, the report covers the base case and says that no sensitivity results were supplied.
A Revised Materials slot takes your own re-run of the model as individual files, with no need to rebuild the zip. Anything the report cannot use is stated in the report and never dropped silently.
Generation runs in the background and can take several minutes.
It runs in five stages: loading model results, generating the model schematic, interpreting results, compiling the report document, and reviewing the report against the model.
Reviewing the report against the model is the longest stage. When it ends, the card changes to Download Report.
The result in the film is a 31 page technical report. The contents run from the executive summary through methods, results, sensitivity analyses, assumptions and disclosures, model validation, discussion, limitations, conclusions and references.
The executive summary states the case: the model type, the cycle length, the time horizon and the perspective, then the costs and QALYs for each strategy, the incremental results and the ICER.
Table 1 sets out the decision problem: the condition, the strategies compared, the reference strategy, the perspective, the time horizon and the threshold.
Each transition is listed with its type. Where the source describes no transition along a pathway, the report says so and sets its probability to zero.
Every input table carries a source column. Where an input is not linked to a reference, the column says so.
Where no sensitivity analysis results were supplied, the section says so and shows no cost effectiveness plane and no acceptability curves.
The disclosures section lists each input not taken directly from the source, with its status and the basis where one was recorded.
The validation section reports the platform's automated tests and the agreement between the workbook and the R model on the compared quantities.
The report closes by pointing to the interactive Shiny application that comes with the R model.
A review record comes with the report. The automatic review checked each prose sentence against the model. In the film, 17 of 214 sentences were flagged: 13 replaced, 3 removed and 1 left open.
Each change is listed with its evidence. Where the review found nothing that supports a claim, it removed the claim instead of keeping a statement it could not support.
What the review could not settle is listed too, and stays on the record as an open item.
Keeping the report current means regenerating it. A revised workbook goes in through Revised Materials, and the report is then written from the recalculated results and states the difference from the as-built ones.