Using Clinical Notes with Time Series Data for ICU Management

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 6432–6437, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 6432 Using Clinical Notes with Time Series Data for ICU Management Swaraj Khadanga University of Minnesota Minneapolis, MN khada004@umn.edu Karan Aggarwal University of Minnesota Minneapolis, MN aggar081@umn.edu Shafiq Joty Nanyang Technological University Singapore srjoty@ntu.edu.sg Jaideep Srivastava University of Minnesota Minneapolis, MN srivasta@umn.edu Abstract Monitoring patients in ICU is a challenging and high-cost task. Hence, predicting the con- dition of patients during their ICU stay can help provide better acute care and plan the hospital’s resources. There has been continu- ous progress in machine learning research for ICU management, and most of this work has focused on using time series signals recorded by ICU instruments. In our work, we show that adding clinical notes as another modal- ity improves the performance of the model for three benchmark tasks: in-hospital mor- tality prediction, modeling decompensation, and length of stay forecasting that play an important role in ICU management. While the time-series data is measured at regular intervals, doctor notes are charted at irreg- ular times, making it challenging to model them together. We propose a method to model them jointly, achieving considerable improvement across benchmark tasks over baseline time-series model. Our implemen- tation can be found at https://github. com/kaggarwal/ClinicalNotesICU. 1 Introduction With the advancement of medical technology, pa- tients admitted into the intensive care unit (ICU) are monitored by different instruments on their bedside, which measure different vital signals about patient’s health. During their stay, doc- tors visit the patient intermittently for check-ups and make clinical notes about the patient’s health and physiological progress. These notes can be perceived as summarized expert knowledge about the patient’s state. All these data about instru- ment readings, procedures, lab events, and clin- ical notes are recorded for reference. Availabil- ity of ICU data and enormous progress in ma- chine learning have opened up new possibilities for health care research. Monitoring patients in ICU is a challenging and high-cost task. Hence, predicting the condition of patients during their ICU stay can help plan better resource usage for patients that need it most in a cost-effective way. Prior works (Harutyunyan et al., 2019; Ghassemi et al., 2015; Suresh et al., 2018; Song et al., 2018; Caballero Barajas and Akella, 2015) have focused exclusively on modeling the problem using the time series signals from medical instruments. Ex- pert knowledge from doctor’s notes has been ig- nored in the literature. In this work, we use clinical notes in addition to the time-series data for improved prediction on benchmark ICU management tasks (Harutyunyan et al., 2019). While the time-series data is mea- sured continuously, the doctor notes are charted at intermittent times. This creates a new challenge to model continuous time series and discrete-time note events jointly. We propose such a multi- modal deep neural network that comprises of re- current units for the time-series …

Content truncated to 500 words