Machine learning feature data from EHR, labels, and estimates for next generation sequencing-based assay
Data files
Nov 28, 2024 version files 17.83 MB
-
Estimates.csv
1.74 KB
-
Feature_data_anon.csv
17.82 MB
-
README.md
3.30 KB
Abstract
Next-generation sequencing-based tests have advanced the field of medical diagnostics, but their novelty and cost can lead to uncertainty in clinical deployment. The Heme-STAMP is one such assay that tracks mutations in genes implicated in hematolymphoid neoplasms. Rather than limiting its clinical usage or imposing rule-based criteria, we propose leveraging machine learning to guide clinical decision-making on whether this test should be ordered. We trained a machine learning model to predict the outcome of Heme-STAMP testing using 3,472 orders placed between May 2018 and September 2021 from an academic medical center and demonstrated how to integrate a custom machine learning model into a live clinical environment to obtain real-time model and physician estimates. The model predicted the results of a complex next-generation sequencing test with discriminatory power comparable to expert hematologists (AUC score: 0.77 [0.66, 0.87], 0.78 [0.68, 0.86] respectively) and with capacity to improve the calibration of human estimates.
README: Machine learning feature data from EHR, labels, and estimates for next generation sequencing-based assay
https://doi.org/10.5061/dryad.nzs7h450b
Description of the data and file structure
These datasets were utilized to train and evaluate a machine learning model that predicts the outcome of the Heme-STAMP test, a next generation sequencing assay that tracks mutations in genes implicated in hematolymphoid neoplasms. The feature_data_anon.csv was used to train/test a Random Forest model and uses features such as demographics, lab results, medications, diagnoses, etc. Numerical values were binned by their distribution. For example, "Age0" would correspond to the 1st bucket of values while "Age_3" would correspond to the 4th bucket. The estimates.csv contains the estimations generated by the ordering physician and the machine learning model on the orders that were prospectively collected.
Files and variables
File: Feature_data_anon.csv
Description: Used to train/test a Random Forest model and uses features such as demographics, lab results, medications, diagnoses, etc. Numerical values were binned by their distribution. For example, "Age_0" would correspond to the 1st bucket of values while "Age_3" would correspond to the 4th bucket.
Variables
- Anonymized Observation ID: Numerical value that can be used to identify features that belong to the same observation ID (unique for each patient and order).
- feature_type: Feature category
- feature: Subtype within each category. Details the specific medication, lab test, etc.
- value: Indicates the presence of the corresponding feature in the patient's EHR up to the date/time the Heme-STAMP order was placed for this patient. To save memory, the absence of features is not explicitly shown. In other words, if one observation ID has a specific feature but another does not, then the former will have a row with that feature and value = 1 while the latter will not have a row for that feature.
- label: Binary (0 or 1) value that indicates the presence of at least one pathogenic mutation in the Heme-STAMP test result.
File: Estimates.csv
Description: Contains the estimations generated by the ordering physician and the machine learning model on the orders that were prospectively collected.
Variables
- Anonymized Observation ID: Numerical value that can be used to identify features that belong to the same observation ID (unique for each patient and order).
- Ordering Physician Estimate: Expert hematologists selected one of the following estimation buckets: <10%, 10-30%, 30-50%, 50-70%, 70-90%, >90%
- Model Estimate: Continuous value between 0 and 1, inclusive, generated by the model.
Code/software
Any software that is capable of opening a CSV file can be used to view the data. To simply view, use Excel or some other spreadsheet software. To wrangle the data or use it for other purposes, use an IDE of your preference.
Access information
Data was derived from the following sources:
- STAnford medicine Research data Repository (STARR)
- Stanford Pathology Department
- Machine learning model and participating hematologists (described in corresponding paper)
Methods
The feature data was pulled from the STAnford medicine Research data Repository (STARR) and further processed to meet the needs of this study and privacy guidelines. Labels were obtained through the Stanford Pathology Department. Ordering physician estimates were generated by participating physicians and model estimates were generated by the machine learning model used in the study.