Dataset Information

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.

SUBMITTER: Tang S

PROVIDER: S-EPMC7727385 | biostudies-literature | 2020 Dec

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

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

Tang Shengpu S Davarmanesh Parmida P Song Yanmeng Y Koutra Danai D Sjoding Michael W MW Wiens Jenna J

Journal of the American Medical Informatics Association : JAMIA 20201201 12

<h4>Objective</h4>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 fro ...[more]

PMID: 33040151

Similar Datasets

Project description:BackgroundAs the most stable and experimentally accessible epigenetic mark, DNA methylation is of great interest to the research community. The landscape of DNA methylation across tissues, through development and in disease pathogenesis is not yet well characterized. Thus there is a need for rapid and cost effective methods for assessing genome-wide levels of DNA methylation. The Illumina Infinium HumanMethylation450 (450K) BeadChip is a very useful addition to the available methods for DNA methylation analysis but its complex design, incorporating two different assay methods, requires careful consideration. Accordingly, several normalization schemes have been published. We have taken advantage of known DNA methylation patterns associated with genomic imprinting and X-chromosome inactivation (XCI), in addition to the performance of SNP genotyping assays present on the array, to derive three independent metrics which we use to test alternative schemes of correction and normalization. These metrics also have potential utility as quality scores for datasets.ResultsThe standard index of DNA methylation at any specific CpG site is ? = M/(M + U + 100) where M and U are methylated and unmethylated signal intensities, respectively. Betas (?s) calculated from raw signal intensities (the default GenomeStudio behavior) perform well, but using 11 methylomic datasets we demonstrate that quantile normalization methods produce marked improvement, even in highly consistent data, by all three metrics. The commonly used procedure of normalizing betas is inferior to the separate normalization of M and U, and it is also advantageous to normalize Type I and Type II assays separately. More elaborate manipulation of quantiles proves to be counterproductive.ConclusionsCareful selection of preprocessing steps can minimize variance and thus improve statistical power, especially for the detection of the small absolute DNA methylation changes likely associated with complex disease phenotypes. For the convenience of the research community we have created a user-friendly R software package called wateRmelon, downloadable from bioConductor, compatible with the existing methylumi, minfi and IMA packages, that allows others to utilize the same normalization methods and data quality tests on 450K data.

Project description:Background: As the most stable and experimentally accessible epigenetic mark, DNA methylation is of great interest to the research community. The landscape of DNA methylation across tissues, through development and in disease pathogenesis is not yet well characterised. Thus there is a need for rapid and cost effective methods for assessing genome-wide levels of DNA methylation. The Illumina Infinium HumanMethylation450 (450K) BeadChip is a very useful addition to the available methods but its complex design, incorporating two different assay methods, requires careful consideration. Accordingly, several normalization schemes have been published. We have taken advantage of known DNA methylation patterns associated with genomic imprinting and X-chromosome inactivation (XCI), in addition to the performance of SNP genotyping assays present on the array, to derive three independent metrics which we use to test alternative schemes of correction and normalization. These metrics also have potential utility as quality scores for datasets. Results: The standard index of DNA methylation at any specific CpG site is ? = M/(M + U + 100) where M and U are methylated and unmethylated signal intensities. Betas calculated from raw signal intensities (the default GenomeStudio behaviour) perform well, but using 11 methylomic datasets we demonstrate that quantile normalization methods produce marked improvement, even in highly consistent data, by all three metrics. The commonly used procedure of normalizing betas is inferior to the separate normalization of M and U, and it is also advantageous to normalize Type I and Type II assays separately. More elaborate manipulation of quantiles proves to be counterproductive. Conclusions: Careful selection of preprocessing steps can minimise variance and thus improve statistical power, especially for the detection of the small absolute DNA methylation changes likely associated with complex disease phenotypes. For the convenience of the research community we have created an R software package called wateRmelon, compatible with the existing methylumi, minfi and IMA packages, that allows others to utilise the same normalization methods and data quality tests on 450K data. Bisulfite converted DNA from the 11 cohorts (N=695, including 36 technical replicates) were hybridised to the Illumina Infinium 450k Human Methylation Beadchip v1.2

Project description:Background: As the most stable and experimentally accessible epigenetic mark, DNA methylation is of great interest to the research community. The landscape of DNA methylation across tissues, through development and in disease pathogenesis is not yet well characterised. Thus there is a need for rapid and cost effective methods for assessing genome-wide levels of DNA methylation. The Illumina Infinium HumanMethylation450 (450K) BeadChip is a very useful addition to the available methods but its complex design, incorporating two different assay methods, requires careful consideration. Accordingly, several normalization schemes have been published. We have taken advantage of known DNA methylation patterns associated with genomic imprinting and X-chromosome inactivation (XCI), in addition to the performance of SNP genotyping assays present on the array, to derive three independent metrics which we use to test alternative schemes of correction and normalization. These metrics also have potential utility as quality scores for datasets. Results: The standard index of DNA methylation at any specific CpG site is β = M/(M + U + 100) where M and U are methylated and unmethylated signal intensities. Betas calculated from raw signal intensities (the default GenomeStudio behaviour) perform well, but using 11 methylomic datasets we demonstrate that quantile normalization methods produce marked improvement, even in highly consistent data, by all three metrics. The commonly used procedure of normalizing betas is inferior to the separate normalization of M and U, and it is also advantageous to normalize Type I and Type II assays separately. More elaborate manipulation of quantiles proves to be counterproductive. Conclusions: Careful selection of preprocessing steps can minimise variance and thus improve statistical power, especially for the detection of the small absolute DNA methylation changes likely associated with complex disease phenotypes. For the convenience of the research community we have created an R software package called wateRmelon, compatible with the existing methylumi, minfi and IMA packages, that allows others to utilise the same normalization methods and data quality tests on 450K data.

Dataset Information

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

Publications

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

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets