A health economic model rarely starts from a blank page. It usually starts from the models already published on the same question, or a similar one, to see what has been done. The first task is to understand how those models differ, where they agree, and how well each one fits the question that has to be answered. In HEORACLE, that first step is part of Model Specification. When you start from several papers, Multi-Paper Model Specification compares them for you: which model fits your question best, and what in the others you may want to bring in.
You upload the published models, each with its supplementary materials where it has them, and type your research question: the patients, the treatments, the country and the perspective.
The question is the lens. Every published model is reviewed through it.
Each paper is read once, in full, and the papers are read in parallel: the model structure, the settings, the treatments and the sources.
The system shows what it understood from your question. You can correct any field or fill in one your question did not give, and what you confirm is the specific need every paper is reviewed against.
Each published model is graded against your question on population, comparators, perspective and setting, and structural fit, each scored from one to five. Recency adds one to three points.
Where your question leaves something open, it counts neither for nor against a paper.
Every score has its explanation, drawn from the paper, so you can see why a model fits or does not.
The model with the highest total is proposed as the starting point. It is a recommendation: you can accept it or start from any of the other models, because the decision is yours.
You review the chosen structure and edit anything read from the paper. The other papers stay open under Source models, with their own structures and treatments.
You can ask how another model handled the same thing. The answer quotes the paper and changes nothing until you decide.
Every change is described in your own words. The system shows exactly what it would change, and nothing changes until you apply it. A change can raise questions that need an answer before you go on.
When the structure is right, you finalise it. From then on the values are read from the paper you chose, and values from the other papers come in only where you add them yourself.
From here the run follows the same assessment as a single paper or a specification you upload yourself. Published papers have gaps in their data and can contradict each other, and the system brings back the questions that must be answered for the model to be built, each in plain words, with what the paper does give.
Each answer goes back into the model structure. Where an answer reveals another gap, the system brings it back to you rather than deciding.
Each answer is recorded with its source, and items flagged for review stay listed. When the last question is answered, the specification is complete and the downloads unlock.
The result is the model specification for your question: the structure, the treatments, every value with its source, and every assumption and decision, including the ones you made. The model is then built from it.
It downloads as the narrative, a review report, the assumptions, missing data, the interaction record, the schematic, and the files for the model build.