rdoe_persona_userprofile
rdoe_persona_select_up
Article

The powerful possibilities of artificial intelligence for sepsis diagnosis

A patient’s age, medical history, vital signs, and lab test results are all long-recognized data points that individually provide relevant information when it comes to clinical decision-making. However, bring them together and they provide powerful insights for determining the correct diagnosis, rapid prognosis, and the best course of treatment. In today’s world, the collection of these data points amounts to about a trillion gigabytes of healthcare information produced annually.1 But what do we do with data so vast that it is becoming humanly impossible to arrive at data insights easily, or quickly? 

Fortunately, Artificial Intelligence (AI) is stepping in to help, with its ability to support fast data mining, analysis, and machine learning. One of the latest innovative uses of AI technologies in healthcare is medical algorithms that can act as clinical decision support (CDS) tools for physicians. When used at the right time and in the right way, these AI-based tools can alleviate resource pressures on administrative and laboratory work while also helping to enhance the efficiency and accuracy of clinical decisions in diagnostics.2

In honor of Sepsis Awareness Month, let's take a dive into the benefits that artificial intelligence-based CDS tools can bring in the context of sepsis, a life-threatening organ dysfunction caused by a systemic inflammatory response. We look at the resounding impact AI can have on sepsis detection, prognosis, and survival.

Article highlights:

  • Sepsis is a life-threatening condition that claims more lives annually than all forms of cancer combined, especially due to late diagnosis.
  • Artificial intelligence (AI)-based medical algorithms integrated into clinical decision support tools have the potential to improve sepsis diagnosis and management.
  • The potential of AI in the prompt detection of sepsis is enormous, but it must undergo rigorous clinical validation, regulatory approval, and seamless integration into the clinical workflow.

animated speaking bubble

Join our community and stay up to date with the latest laboratory innovations and insights.

Subscribe now

The need to change the way we diagnose sepsis

Globally, sepsis is responsible for approximately 11 million deaths annually—nearly 1 in 5 deaths—surpassing the combined mortality of all cancers.4,5 Time is critical in sepsis management; each hour of delayed treatment increases the risk of death by 7.6%, and current guidelines recommend administering effective antimicrobial therapy within one hour of recognition.6,7

Early identification of sepsis relies on host-specific parameters, such as recognizing temperature abnormalities, infection signs, mental decline, and extreme illness (TIME™), along with other clinical scores such as the Sequential Organ Failure Assessment (SOFA) score.3,8 Unfortunately, mainly due to the slow turnaround time of blood cultures, which typically takes 3 to 5 days, accurate and timely sepsis detection still remains a challenge.9

In order to help physicians save lives and reduce the substantial burden of sepsis on hospital resources, such as prolonged stays and high readmission costs, there is an urgent need for more rapid diagnostic and risk stratification tools.9-11 This situation raises an important question: Can AI support healthcare providers and systems in making earlier, more confident clinical decisions?

Current unmet medical needs and opportunities in sepsis diagnosis

Despite being classed as a medical emergency, emergency room doctors and intensive care units are still met with major challenges when it comes to diagnosing sepsis.9 First, septic patients often present with vague and heterogeneous symptoms. Second, sepsis is a syndrome of various physiological, pathological, and biochemical abnormalities that lacks an accepted, unanimous definition within the classical diagnostic approach. 

Utilizing the vast amount of healthcare data available to train artificial intelligence (AI) models, opens the door for medical algorithms to support more accurate diagnosis of sepsis. By finding global patterns in the diversity and complex ambiguity of sepsis presentation, AI can help us understand the likelihood that a patient will become septic and/or the potential progressions of the disease.9,12

Moreover, the ability of AI to quickly amalgamate data from a patient’s own labs, vitals, and demographics in real time could improve the number of patients correctly identified as septic. AI-based CDS tools for sepsis diagnosis have the potential to provide an accurate, quick holistic, patient-specific interpretation for better clinical decision-making and patient care.9,12 This includes:

  1. Rapid diagnosis: Biomarkers and AI-assisted tools might help physicians diagnose sepsis earlier in the emergency room/intensive care unit, and in postoperative patients, along with rapid organism and resistance gene identification.
  2. Triage of patients: Biomarkers and AI-assisted tools might help physicians identify patients at high risk of sepsis or developing sepsis within 24 hours of assessment, and also indicate when it is safe to discharge a patient from the intensive care unit after sepsis.
  3. Outcome prediction/severity assessment: AI-assisted tools might ensure correct patient management, as escalation and de-escalation alert and confirmation systems must be in place for patients at high or low risk of developing sepsis.
  4. Therapy guidance: Ultimately, better patient care is achievable when biomarkers and AI-assisted tools are implemented to monitor therapy response and guide personalized medicine.
From artificial intelligence opportunity to reality for sepsis

While the promise of artificial intelligence (AI) for sepsis diagnosis and management is tremendous, as with all pharmaceutical and diagnostic devices, AI devices must be subjected to a robust evidence-based clinical validation process to ensure safety and efficacy prior to use on patients.13-16 This requires rigorously testing AI algorithms against sufficiently large and diverse datasets, comparing their performance to other standard tools/methods, and ensuring that the technology functions reliably across different patient populations within its intended use. Clinical validation is non-negotiable and provides a transparent and evidence-based foundation for clinical decision-making that increases confidence in the tool's accuracy, predictions, and recommendations. This confidence is also conveyed by the AI tools’ regulatory classification, granted by a regulatory agency such as the United States Food and Drug Administration (US FDA).17

Beyond clinical validation, data privacy must also be ensured wherever patient health data is processed. As with all medical institutions handling such data, AI-based tools must follow local data privacy laws and regulations to ensure that patient data is protected and secure.18

AI and the importance of real-world implementation

Currently, there are still too many potential septic patients underdiagnosed and undertreated, or overdiagnosed and overtreated. While some (AI-)tools are available to assist clinicians in sepsis detection (mainly the so-called “early warning systems”), the significant potential of AI remains relatively untapped in terms of diagnostic and prognostic performance in real clinical settings.19,20

One of the reasons for this is that the essential role of very large, diverse real-world datasets in training and validating accurate AI models, both retrospectively and prospectively, has not yet been fully exploited. Without accurate AI diagnostics tools, AI can actually create more harm than good, by missing truly septic patients or worsening the overuse of antibiotics in non-septic patients.21-23 Accurate AI diagnostic tools must also ensure that healthcare professionals do not encounter alert fatigue through too many automated “false alarms”. 24,25

Real-world validation and collaborative intelligence

Healthcare professionals are responsible for making the key decisions along a sepsis patient continuum of care, from emergency/intensive care unit admission and intervention, to discharge or continued patient monitoring. 

When this expertise, or human-based intelligence, is augmented with artificial intelligence, there is potential for even greater intelligence. This is conveyed by the concept of “collaborative intelligence”, defined as the use of advanced analytics and computing power with an understanding that we are responsible both for the data it is offered and the fair interpretation of its outputs, with the intention of together becoming more intelligent.26

The intention behind utilizing AI-based CDS tools for sepsis diagnosis is to leverage this collaborative intelligence so that healthcare professionals have the best possible chance of identifying rising-risk individuals and directing them to timely care in the most appropriate location.9,12 By employing collaborative intelligence, CDS tools can help physicians improve quality and safety measures such as response times to changing patterns.26

The future of sepsis diagnosis

Medical algorithms and AI-based CDS tools offer healthcare professional insights that come from models built on tens of thousands of cases, effectively creating a tool that brings the experience of hundreds of physicians to just one or a few physicians making a life-or-death decision. 

2024 looks to be an exciting year for sepsis diagnosis, with new AI-based CDS tools getting either authorization or clearance from regulatory agencies. With the development of these game-changing technologies, hopefully, we can begin changing the way we diagnose sepsis and improve patient care. 

Want to be the first to receive the latest insights from industry leaders? Sign up for our newsletter.

  1. Roche. (2019). Article available from https://www.roche.com/stories/machine-learning-in-personalised-healthcare [Accessed September 2024]
  2. Navify.com. (2022). Article available from https://navify.roche.com/wp-content/uploads/2022/11/navify-Event-Editorial-Healthcare-Transformer-AI-MC-10689-Nov11-2022-v6.pdf [Accessed September 2024]
  3. Singer M et al. (2016). JAMA. 315(8), 801-10. Paper available from https://jamanetwork.com/journals/jama/fullarticle/2492881 [Accessed September 2024]
  4. Rudd KE et al. (2020). Lancet 395(10219), 200-211. Paper available from  https://www.thelancet.com/article/S0140-6736(19)32989-7/fulltext [Accessed September 2024]
  5. Bray F et al. (2024). CA Cancer J Clin. 74(3), 229-263. Paper available from https://acsjournals.onlinelibrary.wiley.com/doi/10.3322/caac.21834 [Accessed September 2024]
  6. Kumar A et al. (2006). Crit Care Med. 34(6), 1589-96. Paper available from https://journals.lww.com/ccmjournal/abstract/2006/06000/duration_of_hypotension_before_initiation_of.1.aspx   [Accessed September 2024]
  7. Evans L et al. (2021). Intensive Care Med. 47(11), 1181-1247. Paper available from https://link.springer.com/article/10.1007/s00134-021-06506-y [Accessed September 2024]
  8. Sepsis Alliance. Website available from https://www.sepsis.org/about/its-about-time/  [Accessed September 2024]
  9. Duncan CF et al. (2021). Curr Infect Dis Rep. 23(12), 22. Paper available from  https://link.springer.com/article/10.1007/s11908-021-00765-y [Accessed September 2024] 
  10. Paoli CJ et al. (2018). Crit Care Med. 46(12),1889-1897. Paper available from https://journals.lww.com/ccmjournal/fulltext/2018/12000/epidemiology_and_costs_of_sepsis_in_the_united.1.aspx [Accessed September 2024]
  11. Gadre SK et al. (2019) Chest. 155(3), 483-490. Paper available from https://journal.chestnet.org/article/S0012-3692(18)32895-2/abstract [Accessed September 2024]
  12. O'Reilly D et al. (2023). J Intensive Med. 4(1), 34-45. Paper available from https://www.sciencedirect.com/science/article/pii/S2667100X23000816?via%3Dihub [Accessed September 2024]
  13. Lenharo M. (2024). Nature. 632(8026), 722-724. Paper available from https://www.nature.com/articles/d41586-024-02675-0 [Accessed September 2024]
  14. Chouffani El Fassi S et al. (2024). Nat Med. Paper available from  https://www.nature.com/articles/s41591-024-03203-3 [Accessed September 2024]
  15. O'Driscoll F et al. (2024). JMIR Form Res. 8, e56241. Paper available from  https://formative.jmir.org/2024/1/e56241  [Accessed September 2024]
  16. Park SH et al. (2021). Korean J Radiol. 22(3), 442-453. Paper available from  https://www.kjronline.org/DOIx.php?id=10.3348/kjr.2021.0048 [Accessed September 2024]
  17. FDA. (2024).  Article available from https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device  [Accessed September 2024]
  18. Roche. (2023). PDF available from https://assets.roche.com/f/176343/x/401c28049f/roche-ai-ethics-principles.pdf   [Accessed September 2024]
  19. Qayyum SN et al. (2024). Ann Med Surg (Lond). 86(5), 2411-2412. Paper available from https://journals.lww.com/annals-of-medicine-and-surgery/fulltext/2024/05000/ai_integration_in_sepsis_care__a_step_towards.15.aspx [Accessed September 2024]
  20. Kamran et al. (2024). NEJM AI 1(6). Paper available from https://ai.nejm.org/doi/full/10.1056/AIp2400247  [Accessed September 2024]
  21. Infectious Diseases Society of America (IDSA). (2015). Report available from https://www.idsociety.org/globalassets/idsa/policy--advocacy/current_topics_and_issues/diagnostics/statements/better-tests-better-care-for-policymakers.pdf  [Accessed September 2024]
  22. WHO. (2023). Fact sheet available from https://www.who.int/news-room/fact-sheets/detail/antimicrobial-resistance  [Accessed September 2024]
  23. Rhee C et al. (2020). JAMA Netw Open. 3(4), e202899. Paper available from https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2764581  [Accessed September 2024]
  24. The Lancet Respiratory Medicine. (2018). Lancet Respir Med. 6(3), 161. Paper available from https://www.thelancet.com/journals/lanres/article/PIIS2213-2600(18)30072-9/fulltext  [Accessed September 2024]
  25. Hwang MI et al. (2020). West J Emerg Med. 21(5), 1201-1210. Paper available from https://escholarship.org/uc/item/4wv4c7fk  [Accessed September 2024]
  26. Bhatt AB, Bae J. (2023). NPJ Digit Med. 6(1), 177. Paper available from https://www.nature.com/articles/s41746-023-00920-w  [Accessed September 2024]