DistinCT: Deciphering CT Indications in Lung Cancer Survivors

Overview

DistinCT is an R package designed for abstraction of CT imaging indications in long-term lung cancer survivors. The package provides tools for reading and pre-processing electronic health records (EHR), including CT scan reports, extracting structured EHR and NLP features, and applying a fitted logistic regression model. The logistic regression predicts whether CT scans are performed for surveillance or other reasons.

Package Installation

You can install the development version of DistinCT from GitHub with:

# install.packages("devtools")
devtools::install_github("Aparajita-K/DistinCT")

Usage

The DistinCT package is straightforward to use. Below is a basic workflow demonstrating how to load sample data, extract features, and predict CT scan indications.

Example usage:

Read and Validate Input Data

The input data should be a data frame or in CSV or Excel format with the following columns:

• patient_id: Unique identifier for the patient (character or numeric).
• diagnosis_date: Date of initial diagnosis in "YYYY-MM-DD" format.
• ct_date: Date of the CT scan in "YYYY-MM-DD" format.
• provider_type: Type of provider ordering the CT scan (character).
• report: Text of the CT scan report (character).
• symptom_diagnosis, lungdisease_diagnosis, Xray_count: Additional structured EHR fields.

The read_data function reads and validates data from a CSV or Excel file, or an existing data frame.

library(DistinCT)

# Load sample data
path_to_data <- system.file("data", "SyntheticData.xlsx", package = "DistinCT")
data <- read_data(path_to_data)

Feature Extraction

The feature_extract function processes the input data to extract structured EHR and NLP features. It prepares the data for subsequent analysis and prediction.

# Extract features
feature_data <- feature_extract(data)

Predict CT Scan Indications

The predict_indication function applies a pre-trained logistic regression model to predict the indication for CT scans in long-term lung cancer survivors. It classifies each scan as either 'Surveillance' or 'Other Reasons' based on structured EHR and NLP features.

Parameters:

• feature_data: A data frame containing features extracted by the feature_extract function, including EHR Features (e.g., 'priorCT_6mon', 'provider_med', 'provider_onc', 'symptom_binary', 'LungDis_binary', 'Xray_count') and NLP Features (e.g., 'DefinitiveTreat', 'Surveillance', 'Recurrence', 'FollowUp', 'Metastasis', 'Symptom', 'LC_Treat_Drug').
• binarize_thr: A number in [0, 1] to specify the binarization threshold for classifying 'Surveillance' vs 'Other Reasons'. If set to NA (default), the threshold from the saved model is used.
• writeFile: A logical value indicating whether to write the output to an Excel file. Defaults to TRUE.
• outputFilePath: The file path for the Excel output file. Defaults to "Predictions.xlsx" in the current working directory.

Output:

The function returns a data frame identical to feature_data but with two additional columns: Prediction_Probability and CT_indication. Prediction_Probability contains the real-valued probability of a CT scan being for 'Surveillance', and CT_indication contains the prediction 'Surveillance' or 'Other Reasons'. If writeFile is TRUE, the output is also saved as an Excel file.

Example Usage:

# Predict CT scan indications and write to an Excel file
predictions <- predict_indication(feature_data, 
                                  writeFile = TRUE, 
                                  outputFilePath = "Model_Predictions.xlsx")

# Predict CT scan indications with a custom binarization threshold
predictions_custom_thr <- predict_indication(feature_data, binarize_thr = 0.5)

Contact Information

For any questions or feedback regarding the DistinCT package, feel free to reach out to the authors:

• Aparajita Khan - aparjita@stanford.edu
• Summer Han - summerh@stanford.edu