Based on this paper

Michael Roimi, Rom Gutman, Jonathan Somer, Asaf Ben Arie, Ido Calman, Yaron Bar-Lavie, Arnona Ziv, Danny Eytan, Malka Gorfine & Uri Shalit

Special credit goes to Malka Gorfine, Asaf Ben Arie and Jonathan Somer for implementation of the model and code used for analysis, as well as setting up this Github repository.

This package presents a multi-state survival analysis model which can be used to predict covid-19 patients' illness trajectories from time of hospitalization up to time of recovery or death.

Transitions between states (such as: "severe" to "death") are modeled using survival models with competing risks, and the patient's journey is estimated via Monte-Carlo path sampling over these transitions, while updating time-dependent patient covariates.

The following examples introduce the model and provide examples for typical usage. We suggest following the examples in order.

Each notebook has a corresponding pre-rendered html file with the same name as the *.Rmd file. Reviewing the html files is the fastest way to review the examples and decide if they are useful for your setting.

A similar Python package by Hagai Rossman, Ayya Keshet and Malka Gorfine is also available at: https://hrossman.github.io/pymsm/

This notebook provides a simple working example, while introducing the basic components of the model and dataset used to fit the model.

By following the steps described in this notebook you can fit the model to your own data and make predictions.

If you do not have access to Covid-19 patient data, you can perform predictions using our model which was fit to national Israeli Covid-19 patient data up to early May. The model will be updated as more recent data becomes available.

This notebook shows how to estimate the following values, for a single patient:

• probability of death
• probability of future critical state
• quantiles of predicted time in hospital
• quantiles of predicted time in critical state
• Cumulative Distribution Function (CDF) for time at hospital

This notebook provides an aggregate view of the future for any set of patients. These could be all patients in a certain hospital, ward or even an entire country. Specifically, we provide:

• The expected number of deaths over time
• The expected number of hospitalized patients over time
• The expected number of hospitalized critical patients over time

The notebook also shows how to include expected future hospitalizations in the estimates.