Deep representation learning of patient data from Electronic Health Records (EHR): A systematic review

Si, Yuqi; Du, Jingcheng; Zhao, Li; Jiang, Xiaoqian; Miller, Timothy A.; Wang, Fei; Zheng, W. Jim; Roberts, Kirk

doi:10.1016/j.jbi.2020.103671

Cited by 109 publications

(66 citation statements)

References 105 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…PPMI and PDBP collect longitudinal records in a board spectrum of clinical assessments and biomarkers of patients, which enable us to model the complex symptom progression trajectories of PD to identify subtypes. Sample sizes of the cohorts are relatively large, while from the data-driven perspective, it's always more data the better especially in model training, as abundant data can help avoid model underfitting and overfitting and hence lead to the model capturing true underlying patterns from data 8,9,34 . Meanwhile, data quality, especially missing values may affect the robustness of the identified subtypes.…”

Section: Discussionmentioning

confidence: 99%

Identification of Parkinson PACE subtypes and repurposing treatments through integrative analyses of multimodal clinical progression, neuroimaging, genetic, and transcriptomic data

Hou

Brendel

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

The Parkinson's disease (PD) is a heterogeneous neurodegenerative disease, of which the etiological and pathological mechanisms remain unclear to date. PD has been associated with diverse movement dysfunctions and non-motor symptoms (i.e., symptom heterogeneity) and progression patterns of these symptoms differ from patient to patient (i.e., progression heterogeneity). To address these, the present investigation aims at comprehensively considering full progression course of early PDs to identify subtypes, each of which can reflect unique PD progression pattern. We retrospectively analyzed the Parkinson's Progression Markers Initiative (PPMI) and the Parkinson Disease Biomarkers Program (PDBP) as the development and validation cohorts, respectively. An unsupervised deep learning model was built to model progression trajectories in diverse clinical manifestations and cerebrospinal fluid (CSF) biomarkers to produce a representation vector for each patient, encoding his/her symptom progression profile. Then by performing clustering analysis on the patients' representation vectors, we identified three subtypes with distinct PD progression patterns in the PPMI cohort: Subtype I, mild baseline severity and mild symptom progression; mild baseline severity and moderate progression; and Subtype III, rapid symptom progression. Replication in the PDBP validation cohort demonstrated reproducibility of the subtypes. After that, we explored demographic factors, CSF biomarkers, neuroimaging biomarkers in brain regional atrophy, and genetic factors of the subtypes. Last, to enhance usability of the subtypes, predictive model of subtypes that relies on data at baseline and 1-year follow-up was trained. In conclusion, the identified subtypes revealed significant symptom progression patterns of PDs. Patients with similar baseline severities can even suffer from different progression pattern, leading to distinct prognosis. Demographic factors, biomarkers, and genetic components of the subtypes suggested distinct biological mechanisms and pathways potentially leading to those progression patterns. Our findings may benefit pathophysiological study, clinical practice, and clinical trials to advance PD.

show abstract

Section: Discussionmentioning

confidence: 99%

Identification of Parkinson PACE subtypes and repurposing treatments through integrative analyses of multimodal clinical progression, neuroimaging, genetic, and transcriptomic data

Hou

Brendel

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Benefitting from the advances of deep learning, a great number of studies have been conducted in the representation learning of objects, widely known as object2vector, where object means the original form of human-readable knowledge. Examples of studies in medicine include word2vector [22,78], senten-ce2vector [79], document2vector [80], image2vector, speech2vector [81], EHR2vector [82], and patient2vector [83].…”

Section: Managing Multimodal Data Cognitive Computing-basedmentioning

confidence: 99%

Cognitive Computing-Based CDSS in Medical Practice

Chen

Huang

et al. 2021

Health Data Sci

View full text Add to dashboard Cite

Importance. The last decade has witnessed the advances of cognitive computing technologies that learn at scale and reason with purpose in medicine studies. From the diagnosis of diseases till the generation of treatment plans, cognitive computing encompasses both data-driven and knowledge-driven machine intelligence to assist health care roles in clinical decision-making. This review provides a comprehensive perspective from both research and industrial efforts on cognitive computing-based CDSS over the last decade. Highlights. (1) A holistic review of both research papers and industrial practice about cognitive computing-based CDSS is conducted to identify the necessity and the characteristics as well as the general framework of constructing the system. (2) Several of the typical applications of cognitive computing-based CDSS as well as the existing systems in real medical practice are introduced in detail under the general framework. (3) The limitations of the current cognitive computing-based CDSS is discussed that sheds light on the future work in this direction. Conclusion. Different from medical content providers, cognitive computing-based CDSS provides probabilistic clinical decision support by automatically learning and inferencing from medical big data. The characteristics of managing multimodal data and computerizing medical knowledge distinguish cognitive computing-based CDSS from other categories. Given the current status of primary health care like high diagnostic error rate and shortage of medical resources, it is time to introduce cognitive computing-based CDSS to the medical community which is supposed to be more open-minded and embrace the convenience and low cost but high efficiency brought by cognitive computing-based CDSS.

show abstract

“…For example, a parallel can be drawn between the two types of data by treating a medical visit as a sentence and the medical codes contained within as words. Because of this parallel, NLP techniques have proved to be successful in the clinical field as well [2]. In 2018, Zhang et al used the Word2Vec approach in Patient2Vec [3] to learn a personalized representation for each patient to use it in future hospitalization prediction.…”

Section: Natural Language Processing and Patient Representation Learningmentioning

confidence: 99%

“…Most research [2] based on patient representation learning using EHR uses Recurrent Neural Networks (RNN) with Gated Reccurent Units (GRU) or bi-directional Long Short Term Memory (bi-LSTM) networks. Given that NLP techniques have shown good results in the particular task of patient representation learning, the plan of this research is to explore several NLP models in order to apply them to the available data.…”

Section: Learning Patient Representationsmentioning

confidence: 99%

Patient representation learning from EHR data

Grandpré

2021

Proceedings of the Canadian Conference on Artificial Intelligence

View full text Add to dashboard Cite

Deep representation learning of patient data from Electronic Health Records (EHR): A systematic review

Cited by 109 publications

References 105 publications

Identification of Parkinson PACE subtypes and repurposing treatments through integrative analyses of multimodal clinical progression, neuroimaging, genetic, and transcriptomic data

Identification of Parkinson PACE subtypes and repurposing treatments through integrative analyses of multimodal clinical progression, neuroimaging, genetic, and transcriptomic data

Cognitive Computing-Based CDSS in Medical Practice

Patient representation learning from EHR data

Contact Info

Product

Resources

About