Instructions

The Bayescores platform is comprised of two tools:

1) Tracer is used mainly for trials of metastatic disease where cure rates are less common and the main goal is life prolongation with a smaller proportion of long-term survivors.

2) Curalyze is intended to be used mainly for trials of localized tumors treated with curative intent.

Bayescores can be used by the trialists or any other stakeholders with access to individual patient data for the trial of interest. However, such data may not always be accessible. In this case, users can reconstruct individual patient data from published Kaplan-Meier curves using free online tools like IPDfromKM tool:
https://biostatistics.mdanderson.org/shinyapps/IPDfromKM

Alternatively, manual digitization may also be performed as described for example:
https://www.pharmasug.org/download/sde/rtp2023/PharmaSUG-NCSDE_2023-05.pdf

Another frequently used free tool is WebPlotDigitizer which can be used for point-by-point digitization:
https://apps.automeris.io/wpd

For a more in-depth approach, consider reconstructing the dataset by solving the Kaplan-Meier equations in reverse, often done using Guyot's method. R libraries like survHE or ipdfromkm can be helpful for this.

The individual patient data can then be used in Bayescores to comprehensively assess treatment efficacy.

Tracer

Tracer uses Bayesian Accelerated Failure Time (AFT) models such as Weibull and lognormal, chosen to best fit the data. The option is provided to check model fit. The approach produces posterior probability distributions of Time Ratios (TRs) and cure fractions to represent treatment effects and their uncertainty. Its use of mixture models incorporating cure fractions is well-suited for treatments that yield long-term efficacy, such as immunotherapies.

Curalyze

Curalyze uses Bayesian AFT models to compare treatments used with curative intent. The scoring weights can be adapted to account for quality of life and adverse event burden.

Score Assignment:

Time ratio and cure fraction cutoffs can be adapted based on tumor type and clinical setting.

Cumulative distribution functions (CDFs) are used to transform continuous metrics into scores.

The score weights can be based on contextual factors like social cost or toxicity.

Uncertainty is emphasized by presenting scores as probability distributions with quartiles and highest density intervals.

For more detailed technical recommendations and best practices, refer to the document.