# 39. Partitioned Survival Analysis

Partitioned Survival Analysis (PartSA) Models mirror real disease progression by mapping the state of the model cohort directly from observed survival data. PartSA models are frequently used to track disease progression in the area of Oncology.

PartSA models include health states like Markov models. However, transitions among states are not driven by transition probabilities. Instead, state membership is estimated based on Survival Functions fitted to the original survival data.

Traditional Markov models provide more flexibility with patient pathways by allowing for both health states and events. However, it can be hard to create patient pathways and transition probabilities that result in progression that matches to the known survival data. PartSA models avoid that problem by using the survival data more directly.

The cohort flow within a Partitioned Survival model is defined by survival curves over time. Each survival curve describes membership in its underlying health state.

For example, 100% of the cohort may start in a Progression-Free Survival state (PFS) at time 0, but after 1 year, only 90% of the cohort is still progression free. Over time, the percentage in PFS would continue to decrease. The membership in the PFS state is then calculated as the area below the PFS Survival Curve.

If the model has additional non-dead health states, these are represented by multiple Survival Curves. Each health state can accumulate value – typically cost and effectiveness – as is appropriate based on the cost of treatment and the state's utility value.