Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

HEALTH CARE

Individualized sepsis treatment using reinforcement learning

Nature Medicine volume 24, pages 1641–1642 (2018)Cite this article

Reinforcement learning is applied to two large databases of electronic health records for patients admitted to an intensive care unit to identify individualized treatment strategies for correcting hypotension in sepsis.

Sepsis is a severe, life-threatening infection and requires instant correct treatment to avoid death in the clinic. For example, sepsis resulting in an extended period of low blood pressure can put a patient at risk of organ failure because of the body’s inability to pump adequate blood to the distal organs. Early fluid therapy may enhance or maintain tissue perfusion by increasing venous return and cardiac output. However, fluid administration may also cause edema within vital organs, leading to organ dysfunction and impairment of oxygen delivery. Conversely, vasopressors may be administered to reverse hypotension and maintain perfusion while limiting the administration of fluid. Both strategies have some evidence to support their use but lack robust data to confirm the benefit of one strategy over the other, creating clinical and scientific challenges1. Furthermore, they both come with potential toxicities, including volume overload for fluid administrations and neurological damage for vasopressor administration. These are both strategies that the clinician must decide between upon first contact with the patient2. If hypotension can be reversed without vasopressors, the patient may not need to be admitted to the intensive care unit (ICU) and incur unnecessary health care costs. How best to treat sepsis has thus been the topic of much debate3,4,5. Komorowski et al.6 now explore the application of reinforcement learning to determine dosing decisions on the basis of more granular patient characteristics.

This is a preview of subscription content, access via your institution

Relevant articles

Open Access articles citing this article.

Access options

Rent or buy this article

Get just this article for as long as you need it

$39.95

Learn more

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Komorowski et al. use RL to develop and validate a best-practice algorithm for vasopressor versus fluid dosing.

References

  1. Self, W. H. et al. Ann. Emerg. Med. 72, 457–466 (2018).

    Article  Google Scholar 

  2. Jaehne, A. K. & Rivers, E. P. Crit. Care Med. 44, 2263–2269 (2016).

    Article  Google Scholar 

  3. Malbrain, M. L. N. G. et al. Ann. Intensive Care 8, 66 (2018).

    Article  Google Scholar 

  4. Bai, X. et al. Crit. Care 18, 532 (2014).

    Article  Google Scholar 

  5. Marik, P. & Bellomo, R. Br. J. Anaesth. 116, 339–349 (2016).

    Article  CAS  Google Scholar 

  6. Komorowski, M., Celi, L. A., Badawi, O. & Gordon, A. C. Nat. Med. https://doi.org/10.1038/s41591-018-0213-5 (2018).

    Article  CAS  Google Scholar 

  7. Sutton, R. S. & Barto, A. G. Introduction to Reinforcement Learning 135 (MIT Press, Cambridge, MA, USA, 1998).

    Book  Google Scholar 

  8. Xu, Y., Xu, Y. & Saria, S. A Bayesian Nonparametric Approach for Estimating Individualized Treatment-Response Curves. in Machine Learning for Healthcare Conference 282–300 (2016).

  9. Prasad, N. et al. A reinforcement learning approach to weaning of mechanical ventilation in intensive care units. Preprint at https://arxiv.org/abs/1704.06300 (2017).

  10. Nemati, S., Ghassemi, M. M. & Clifford, G. D., 2016, August. Optimal Medication Dosing from Suboptimal Clinical Examples: A Deep Reinforcement Learning Approach. in 2016 IEEE 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) 2978–2981 (2016).

  11. Schulam, P. & Saria, S. Discretizing logged interaction data biases learning for decision-making. Preprint at https://arxiv.org/abs/1810.03025 (2018).

Download references

Author information

Authors and Affiliations

  1. Department of Computer Science and Applied Math and Statistics, Whiting School of Engineering, Johns Hopkins University, Baltimore, MD, USA

    Suchi Saria

  2. Department of Health Policy and Management, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD, USA

    Suchi Saria

  3. Armstrong Institute for Patient Safety and Quality, Johns Hopkins School of Medicine, Johns Hopkins University, Baltimore, MD, USA

    Suchi Saria

Authors
  1. Suchi Saria
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Suchi Saria.

Ethics declarations

Competing interests

S.S. has received research support from the American Heart Association (Dallas) and an honorarium from AbbVie (Chicago) and has an ownership interest in Patient Ping (Boston) and Bayesian Health (Baltimore).

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Saria, S. Individualized sepsis treatment using reinforcement learning. Nat Med 24, 1641–1642 (2018). https://doi.org/10.1038/s41591-018-0253-x

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41591-018-0253-x

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing