A command-line tool for classifying patients from a tumor registry as having a Single Primary Cancer (SPC) or Multiple Primary Cancers (MPC), based on ICD-O-3 morphology and ICD-10 topology codes.

Install dependencies:

pip install -r requirements.txt

A tab-separated file (.tsv or .txt) with one row per tumor record. The following columns are required:

Optional columns (used if present, ignored if absent):

See toy_dataset.txt for a minimal working example.

All classification logic is driven by mpc_config.yaml, which must be present in the working directory when the script is run. The config defines:

cd mpc_classifer

python mpc_classifier.py <input_file> <output_label_name> [output_dir]

Example:

python mpc_classifier.py my_cohort.tsv cohort_v1 results/

Four files are written to the output directory, all date-stamped:

See the outputs directory for minimal outputs.

Code for running a mock example of second primary cancer risk prediction (breast-ovary), contained in the mpc_predict_models/ directory. It applies a pre-fitted penalized Fine–Gray model (saved coefficients) to toy data to demonstrate a simple application of the models. For the full methodology, refer to the Materials and Methods and Supplementary Methods of the manuscript. Models for the remaining cancer pairs are available from the authors upon reasonable request.

Requires R and the tidyverse package.

Verify R is installed:

​bash R --version ​

Install tidyverse if missing (run the top of the .Rmd block)

The model uses the following features, derived from each patient's first primary cancer record and linked germline/registry data. All predictors are measured at or before the model landmark time (within one year of first cancer diagnosis).