# PartSA to Markov After Calibration

After you have calibrated model parameters to match observed data using the "simple" PartSA to Markov conversion, do you continue using the new numbers from the calibration in the "simple" conversion model or do you need to adjust the initial PartSA parameters and use the "complex" conversion? I wasn't sure when the "complex" model was used, especially since I need to account for background mortality.

The "complex" conversion initially was able to more accurately model the survival curves but there aren't instructions on how to calibrate a "complex" conversion. This is my first time with PartSA to Markov conversion and I just want to make sure I'm not doing something wrong.

• Official comment

Neil

I think there are two parts to this question. When you have a Partitioned Survival Analysis model, Calibration may help you to better identify the parameters required so your model matches observed data. The Partitioned Survival Analysis (PartSA) model can then be considered as complete.

Separate to this is converting your PartSA model to a Markov model. As you can see, you either use Simple or Complex conversion. Complex transition probabilities attempt to replicate state membership from the original PartSA model. But its just an algorithm, and so those transition probabilities may still need adjusting.

In regards to how to calibrate a Markov model, this is just the same process you used on the PartSA model. You can calibrate any Markov model, including one you have converted from a PartSA model. Look in Chapter 44 in the Help files and there is an explicit example of calibrating a Markov model.

If you have more detailed questions, email Support: support@treeage.com

• Note that PartSA to Markov conversion generates a separate independent Markov model that is not linked back to the PartSA model. Any adjustments to either model are also independent.

Also, note that Markov models involve many different individual event probabilities (e.g., progression, death, etc.), while PartSA models use distributions (or tables) to describe state membership directly without any individual event probabilities. Therefore, the two techniques require different inputs and there is no perfect conversion between the two.