Democratizing EHR analyses with FIDDLE: a flexible data-driven preprocessing pipeline for structured clinical data

Abstract

Objective In applying machine learning (ML) to electronic health record (EHR) data, many decisions must be made before any ML is applied; such preprocessing requires substantial effort and can be labor-intensive. As the role of ML in health care grows, there is an increasing need for systematic and reproducible preprocessing techniques for EHR data. Thus, we developed FIDDLE (Flexible Data-Driven Pipeline), an open-source framework that streamlines the preprocessing of data extracted from the EHR. Materials and Methods Largely data-driven, FIDDLE systematically transforms structured EHR data into feature vectors, limiting the number of decisions a user must make while incorporating good practices from the literature. To demonstrate its utility and flexibility, we conducted a proof-of-concept experiment in which we applied FIDDLE to 2 publicly available EHR data sets collected from intensive care units: MIMIC-III and the eICU Collaborative Research Database. We trained different ML models to predict 3 clinically important outcomes: in-hospital mortality, acute respiratory failure, and shock. We evaluated models using the area under the receiver operating characteristics curve (AUROC), and compared it to several baselines. Results Across tasks, FIDDLE extracted 2,528 to 7,403 features from MIMIC-III and eICU, respectively. On all tasks, FIDDLE-based models achieved good discriminative performance, with AUROCs of 0.757–0.886, comparable to the performance of MIMIC-Extract, a preprocessing pipeline designed specifically for MIMIC-III. Furthermore, our results showed that FIDDLE is generalizable across different prediction times, ML algorithms, and data sets, while being relatively robust to different settings of user-defined arguments. Conclusions FIDDLE, an open-source preprocessing pipeline, facilitates applying ML to structured EHR data. By accelerating and standardizing labor-intensive preprocessing, FIDDLE can help stimulate progress in building clinically useful ML tools for EHR data.

FIBER: enabling flexible retrieval of electronic health records data for clinical predictive modeling

“…Noticeably, only one of the tools reviewed directly supported getting all features for a cohort as FIBER does. 21 Providing a Python interface and working on an i2b2 star schema data format, FIBER stands out in facilitating information exchange and cohort comparability between different health organizations following this schema (eg, the JSON cohort definitions can easily be shared across institutions). Generalizability of data extraction pipelines for these institutions has always been challenging, and we anticipate FIBER to alleviate this issue.…”

Phenoflow: A Microservice Architecture for Portable Workflow-based Phenotype Definitions

“…Future work will explore the impact that Phenoflow has on the portability of additional types of phenotype definitions, including probabilistic definitions, the development of which is likely to leverage data processing tools such as the Flexible Data-Driven Pipeline (FIDDLE) framework 20 . In addition, future work will investigate how the multidimension annotations of the structured definition model can be leveraged in order to introduce new search and discovery capabilities into phenotype repositories.…”

mikropml: User-Friendly R Package for Supervised Machine Learning Pipelines

“…It incorporates good practices in ML training, testing, and model evaluation (Teschendorff, 2019;Topçuoğlu et al, 2020). Furthermore, it provides data preprocessing steps based on the FIDDLE (FlexIble Data-Driven pipeLinE) framework outlined in Tang et al (Tang et al, 2020) and post-training permutation importance steps to estimate the importance of each feature in the models trained (Breiman, 2001;Fisher et al, 2018).…”

“…preprocess_data() takes continuous and categorical data, re-factors categorical data into binary features, and provides options to normalize continuous data, remove features with near-zero variance, and keep only one instance of perfectly correlated features. We set the default options based on those implemented in FIDDLE ( Tang et al, 2020 ). More details on how to use preprocess_data() can be found in the accompanying vignette .…”