Distilling Knowledge from Publicly Available Online EMR Data to Emerging Epidemic for Prognosis

Abstract

Due to the characteristics of COVID-19, the epidemic develops rapidly and overwhelms health service systems worldwide. Many patients suffer from life-threatening systemic problems and need to be carefully monitored in ICUs. An intelligent prognosis can help physicians take an early intervention, prevent adverse outcomes, and optimize the medical resource allocation, which is urgently needed, especially in this ongoing global pandemic crisis. However, in the early stage of the epidemic outbreak, the data available for analysis is limited due to the lack of effective diagnostic mechanisms, the rarity of the cases, and privacy concerns. In this paper, we propose a distilled transfer learning framework, DistCare, which leverages the existing publicly available online Electronic Medical Records to enhance the prognosis for inpatients with emerging infectious diseases. It learns to embed the COVID-19related medical features based on massive existing EMR data. The transferred parameters are further trained to imitate the teacher model's representation based on distillation, which embeds the health status more comprehensively on the source dataset. We conduct Length-of-Stay prediction experiments for patients in ICUs on real-world COVID-19 datasets. The experiment results indicate that our proposed model consistently outperforms competitive baseline methods. In order to further verify the scalability of DistCare to deal with different clinical tasks on different EMR datasets, we conduct an additional mortality prediction experiment on End-Stage Renal Disease datasets. The extensive experiments demonstrate that DistCare can benefit the prognosis for emerging pandemics and other diseases with limited EMR.

MedML: Fusing medical knowledge and machine learning models for early pediatric COVID-19 hospitalization and severity prediction

“…For general COVID-19 related clinical tasks such as diagnosis (Feng et al, 2020;Zoabi et al, 2021), length of stay (Dan et al, 2020;Ma et al, 2021), or severity risk (Jamshidi et al, 2022;Wynants et al, 2020;Yan et al, 2020) The integration of domain knowledge into machine learning models is another ongoing research topic. Most existing works extract domain knowledge from medical literature or clinical concept hierarchies such as ICD codes.…”

“…Finally, we evaluate the models on the testing set and report the prediction performance. The evaluation metrics are AUROC (area under the receiver operating characteristic curve), AUPRC (area under the precision-recall curve), and the maximum of the minimum between precision and sensitivity under the same threshold, known as Min(Re,Pr), following existing works (Ma et al, 2020(Ma et al, , 2021. An example of Min(Re,Pr) is shown in Figure 3, which evaluates whether the model can achieve the balance between precision and recall.…”

“…Most are designed with limited features and are more likely to perform poorly on large heterogeneous datasets. For example,Ma et al (2021) conducted length-of-stay prediction for COVID-19 patients from the HM Hospitals in Spain using a total of 66 lab test features for each patient,ArticleandYan et al (2020) conducted mortality prediction for COVID-19 patients from the Tongji Hospital in China using 74 lab test features. Most of these methods also have limited explainability due to their use of ''black box'' deep neural networks limiting their clinical utility.…”

A Comprehensive Benchmark for COVID-19 Predictive Modeling Using Electronic Health Records in Intensive Care: Choosing the Best Model for COVID-19 Prognosis

“…In the past two years, many machine learning and deep learning models have been proposed to conduct COVID-19 clinical prediction tasks, including diagnosis prediction 13,14 , length-of-stay prediction 15,16 , severity and mortality prediction [3][4][5][6][7][8][9][10][11][12] , etc. Yan et al 3 conducted mortality prediction for COVID-19 patients from the Tongji Hospital in China.…”

“…Gao et al 12 used deep learning and tree-based models to predict COVID-19 severity and hospitalization risks. Ma et al 15 conducted length-of-stay prediction for hospitalized COVID-19 patients from the HM Hospitals in Spain. Though these works have achieved good prediction performance on their own data, these different models are applied to different datasets, and most of them are not publicly available or have strict access restrictions.…”