PSM Developer builds a partitioned survival model from a published cost-effectiveness paper or from your own specification. The film starts from a paper, a cost-effectiveness analysis of pembrolizumab with chemotherapy in biliary tract cancer, and follows it to a working model.
You choose what you are starting from, a published paper or your own specification.
In the film the paper goes in as a PDF with its seven supplements. The paper's tables carry the survival curve parameters, the costs and the utilities, and the Word supplements are read as well. One of them holds the adverse-event costs.
Where the paper reports more than one analysis, the module asks which one to model. In the film the paper reports China and the United States, and the choice is the United States.
The module reads the paper and the supplements.
Before anything is built, you review what was read from the paper: the settings, the treatments and the arms.
Then you review the survival curves and the health states. Every change is described in plain words, and Finalize the structure locks it in once you are satisfied.
The module works out what the paper leaves open. Everything that needs an answer is listed under Input needed.
You answer in plain words. In the film the paper does not say how its survival parameters are defined, and the answer is that the paper used R with the hesim package, so the R flexsurv definitions apply.
The answer comes back as a proposed change. Nothing changes until you press Apply, and then the model rebuilds with the flexsurv convention.
A table question works the same way. In the film the paper lists two subsequent therapies without saying when they start, and one answer, that they begin when patients progress in both arms, sets both rows.
Drug-level details are asked once. In the film one answer on administration, a single visit shared by the drug lines, covers every line.
When every item is answered, Input needed shows zero and Confirmed shows seventeen in the film. Confirm Model builds the deterministic model.
Where the paper is silent, the model uses a placeholder and labels it. In the film this is the proportion female. Every choice the paper left open is listed in the specification.
The results open on one page: costs, QALYs and life years for each arm, with the time in each health state drawn as curves below.
Costs are split by category, so you can see where the differences between the arms come from.
The survival curves tab shows overall survival and progression-free survival for visual inspection.
The model checks itself. In the film twenty-four of twenty-six checks pass. The two Excel checks error, because this model has no workbook.
You take the model away: the R model, the model specification, the session record, the machine-readable files and the testing report. The technical report is available on request.
In the specification every value shows its status and its source, with the quote from the paper and the page it came from. The life table follows the market and is never asked, and the placeholder is labelled as one.
The R model also runs on its own, outside the platform, and gives the same results.
The model also comes as a Shiny app, with the introduction, the results and the survival curves for both arms.