condLSTM-Q: A novel deep learning model for predicting COVID-19 mortality in fine geographical scale

Abstract

Background: Modern machine learning-based models have not been harnessed to their total capacity for disease trend predictions prior to the COVID-19 pandemic. This work is the first use of the conditional RNN model in predicting disease trends that we know of during development that complemented classical epidemiological approaches.

Methods: We developed the long short-term memory networks with quantile output (condLSTM-Q) model for making quantile predictions on COVID-19 death tolls.

Results: We verified that the condLSTM-Q was accurately predicting fine-scale, county-level daily deaths with a two-week window. The model’s performance was robust and comparable to, if not slightly better than well-known, publicly available models. This provides unique opportunities for investigating trends within the states and interactions between counties along state borders. In addition, by analyzing the importance of the categorical data, one could learn which features are risk factors that affect the death trend and provide handles for officials to ameliorate the risks.

Conclusion: The condLSTM-Q model performed robustly, provided fine-scale, county-level predictions of daily deaths with a two-week window. Given the scalability and generalizability of neural network models, this model could incorporate additional data sources with ease and could be further developed to generate other valuable predictions such as new cases or hospitalizations intuitively.

Author summary

Predictive models benefit governments and healthcare systems to combat the COVID-19 pandemic. Here we present the conditional long short-term memory networks with quantile output (condLSTM-Q), a well-performing model for quantile predictions on COVID-19 death tolls at the county level with a two-week forecast window. This fine geographical scale is a rare but valuable feature in publicly available predictive models, significantly benefit state-level officials to coordinate resources within the state. The quantile predictions from condLSTM-Q inform people about the distribution of the predicted death tolls, allowing better evaluation of the possible trajectories of the pandemic. Given the scalability and generalizability of neural network models, this RNN-based model could incorporate additional data sources with ease and could be further developed to generate other helpful predictions such as new cases or hospitalizations intuitively.

Graphical abstract

Keywords

COVID-19 / machine learning / deep learning / epidemiology / time series forecast

References

[1] IHME COVID-19 health service utilization forecasting team, Murray C. J.. ( 2020) Forecasting COVID-19 impact on hospital bed-days, ICU-days, ventilator-days and deaths by US state in the next 4 months. medRxiv, doi: https://doi.org/10.1101/2020.03.27.20043752

[2] Abu-MostafaY.. ( 2020) Caltech CS156 COVID-19 Model. https://cs156.caltech.edu/. Accessed: Accessed: February 1, 2021

[3] Khan,N., Shahid,S., Juneng,L., Ahmed,K., Ismail,T. ( 2019). Prediction of heat waves in Pakistan using quantile regression forests. Atmos. Res., 221 : 1– 11

[4] BrennanA., CrossP. C.. ( 2015) Managing more than the mean: using quantile regression to identify factors related to large elk groups. J. Appl. Ecol., 52, 1656– 1664

[5] IHME. ( 2020) Institute for health metrics and evaluation COVID-19 estimate. http://www.healthdata.org/covid/data-downloads. Accessed: February 1, 2021

[6] HochreiterS.. ( 1997) Long short-term memory. Neural Comput., 9, 1735– 1780

[7] Pal,R., Sekh,A. A., Kar,S. Prasad,D. ( 2020). Neural network based country wise risk prediction of COVID-19. Appl. Sci. (Basel), 10 : 6448

[8] MelinP., MonicaJ. C., SanchezD.. ( 2020) Multiple ensemble neural network models with fuzzy response aggregation for predicting COVID-19 time series: The case of Mexico. Healthcare (Basel), 8, 181

[9] Rémy. ( 2020) Conditional RNNs made easy with Tensorflow and Keras. https://github.com/philipperemy/cond_rnn. Accessed: February 1, 2021

[10] Karpathy,A. ( 2015). Deep visual-semantic alignments for generating image descriptions. In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 3128– 3137

[11] Breiman,L. ( 2001). Random forests. Mach. Learn., 45 : 5– 32

[12] FisherA., RudinC.. ( 2019) All models are wrong, but many are useful: Learning a variable’s importance by studying an entire class of prediction models simultaneously. J. Mach. Learn Res., 20, 177

[13] Analytics ( 2020) DELPHI Model. https://www.covidanalytics.io/projections/. Accessed: February 1, 2021

[14] LosAlamos National Laboratory. ( 2020) LANL Model. https://covid-19.bsvgateway.org/. Accessed: February 1, 2021

[15] TanW.. ( 2020) The cardiovascular burden of coronavirus disease 2019 (COVID-19) with a focus on congenital heart disease. Int. J. Cardiol., 309, 70– 77

[16] ChenR., LiangW., JiangM., GuanW., ZhanC., WangT., TangC., SangL., LiuJ., NiZ.,, ( 2020) Risk factors of fatal outcome in hospitalized subjects with coronavirus disease 2019 from a nationwide analysis in China. Chest, 158, 97– 105

[17] ChenY., GongX., WangL.. ( 2020) Effects of hypertension, diabetes and coronary heart disease on COVID-19 diseases severity: a systematic review and meta-analysis. medRxiv, doi: 10.1101/2020.03.25.20043133

[18] ChenC., ChenC., YanJ. T., ZhouN., ZhaoJ. P. WangD.. ( 2020) Analysis of myocardial injury in patients with COVID-19 and association between concomitant cardiovascular diseases and severity of COVID-19. Zhonghua Xin Xue Guan Bing Za Zhi (in Chinese), 48, 567– 571

[19] ZhangY., TianH., ZhangY.. ( 2020) Is the epidemic spread related to GDP? Visualizing the distribution of COVID-19 in Chinese mainland. arXiv, 2004.04387

[20] U.S.Bureau of Economic Analysis (BEA). ( 2020) Local area gross domestic product, 2018. https://www.bea.gov/system/files/2019-12/lagdp1219.pdf. Accessed: February 1, 2021

[21] NationalInstitutes of Health. ( 2019) COVID-19 treatment guidelines panel. Coronavirus disease 2019 (COVID-19) treatment guidelines. National Institutes of Health. https://www.covid19treatmentguidelines.nih.gov/. Accessed: February 1, 2021

[22] U.S. Food and Drug Administration Pfizer-BioNTech COVID-19 Vaccine. https://www.fda.gov/emergencypreparedness-and-response/coronavirus-disease-2019-covid-19/pfizer-biontech-covid-19-vaccine. Accessed: February 1, 2021

[23] Centersfor Disease Control. COVID-19 vaccinations in the United States. https://covid.cdc.gov/covid-data-tracker/vaccinations. Accessed: February 1, 2021

[24] TheCovidvis team. ( 2020) Covidvis. https://covidvis.berkeley.edu/. Accessed: February 1, 2021

[25] Centersfor Disease Control. ( 2020) Pneumonia and influenza mortality surveillance from the national center for health statistics mortality surveillance system. https://gis.cdc.gov/grasp/fluview/mortality.html. Accessed: February 1, 2021

[26] ZimmermannH. TietzC.. ( 2012) Forecasting with recurrent neural networks: 12 tricks. In Neural Networks: Tricks of the Trade. pp. 687– 707. Springer

[27] KingmaD. P.. ( 2017) Adam: A Method for Stochastic Optimization. arXiv, 1412.6980

ACKNOWLEDGEMENTS

The authors thank Prof Yaser Abu-Mostafa, and the Teaching Assistants of CS156 in Caltech for organizing the COVID19 prediction initiative and for providing the data pipeline for parsing data sources. We thank Isaac Yen-Hao Chu, M.D. for reading the manuscript. Yu-Li Ni was supported by Taipei Veterans General Hospital Yang-Ming University Excellent Physician Scientists Cultivation Program (No.103-Y-A-003).

COMPLIANCE WITH ETHICS GUIDELINES

The authors HyeongChan Jo, Juhyun Kim, Tzu-Chen Huang, and Yu-Li Ni declare that they have no conflict of interest or financial conflicts to disclose. All procedures performed in studies involving animals were in accordance with the ethical standards of the institution or practice at which the studies were conducted, and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

OPEN ACCESS

This article is licensed by the CC By under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

RIGHTS & PERMISSIONS

2022 The Author (s). Published by Higher Education Press.