Case:

A 68 year old woman presents to the emergency department, brought in by paramedics, complaining of generalized weakness. She lives alone at her home and has not been able to get up to prepare her meals for the last 24 hours. She is alert and oriented, but appears weak. Her vital signs are normal. She has no fever.

She gets blood tests, urine tests and X-rays done in the emergency, which are all normal except for her chest X-ray which shows a patchy infiltrate in the R lung field. She is diagnosed with pneumonia. She is started on antibiotics and admitted to the medical ward.

Later that evening she becomes altered mentally and her blood pressure drops when the nurse comes to check her evening vital signs. She is diagnosed with septic shock and is transferred to the ICU for aggressive intravenous fluids and close monitoring.

 

Diagnosis:

Sepsis is a major problem with a high mortality rate in hospitalized patients. In the US, there are more than 750,000 admissions to hospital for the condition. Their stay in the hospital is much longer and is more expensive when compared to the average hospital patient. Serum lactate levels are elevated in sepsis.

Sepsis can have more favorable outcomes when diagnosed early and treated aggressively with intravenous fluids and intravenous antibiotics. The sooner the treatment is started the better the outcomes.

There are no reliable biomarkers or tests to predict sepsis, before its onset. If there was one such test that could predict the onset of sepsis, treatment could be initiated early, which will have a positive effect on the mortality, morbidity and the duration of hospitalization.

 

Challenge:

Sepsis presents a prediction challenge, where machine learning techniques can be applied to find a solution. As mentioned, the advantage of prediction and/or early diagnosis is that the outcomes can be improved when treated early.

Most prediction methods have relied on vital signs, laboratory data and clinical observation data. The problem with the latter two is they require a physician’s suspicion of sepsis to trigger the action. Also, there is lag time between ordering a lab test, the test being done, and the results being reported. This time can be precious when we are counting each hour the problem has not been recognized.

 

Way Forward:

An ideal situation is where minimal data is required and that this data is routinely collected. Machine learning algorithms have been developed that can use routine vital signs data to predict sepsis several hours before its onset.

This has been done using retrospective data with good results. It needs to be validated and tested in a prospective study. This will help alert the physician to the possibility of sepsis developing in a patient. He/she can then order confirmatory tests and/or start treatment early to prevent worsening of the condition and turning the clinical situation around faster with better outcomes.

In summary, sepsis is a serious clinical condition that needs early detection and aggressive treatment to achieve favorable outcomes. The challenges are that it is insidious in onset and hard to detect early with traditional methods. The advantages are that it responds well to treatment when diagnosed and treated early.

Machine learning algorithms that can predict sepsis early are poised to make a significant change in the time to diagnosis and outcomes for this group of patients. Although it is early, we can expect significant improvement once clinical studies prove the utility of this methodology.

 

References:

  1. Mao Q, Jay M, Hoffman JL, Calvert J, Barton C, Shimabukuro D, Shieh L, Chettipally U, Fletcher G, Kerem Y, Zhou Y, Das R. Multicentre validation of a sepsis prediction algorithm using only vital sign data in the emergency department, general ward and ICU. BMJ Open. 2018 Jan 26;8(1):e017833. doi: 10.1136/bmjopen-2017-017833.
  2. Calvert J, Hoffman J, Barton C, Shimabukuro D, Ries M, Chettipally U, Kerem Y, Jay M, Mataraso S, Das R. Cost and mortality impact of an algorithm-driven sepsis prediction system. J Med Econ. 2017 Jun;20(6):646-651. doi: 10.1080/13696998.2017.1307203. Epub 2017 Apr 3.
  3. Desautels T, Calvert J, Hoffman J, Jay M, Kerem Y, Shieh L, Shimabukuro D, Chettipally U, Feldman MD, Barton C, Wales DJ, Das R. Prediction of Sepsis in the Intensive Care Unit With Minimal Electronic Health Record Data: A Machine Learning Approach.  JMIR Med Inform. 2016 Sep 30;4(3):e28.

 

Author Bio:

Uli K. Chettipally, MD., MPH. is a physician, researcher and an innovator.  He is passionate about delivering artificial intelligence-enabled solutions to the physicians in order to improve patient outcomes. He is the Chief Technology Officer of CREST Network, a collaborative research network at Kaiser Permanente, covering 21 hospitals.  He designed, developed and implemented a region-wide clinical decision support platform to deliver real-time predictive analytics to the physicians at the point of care – for which he received the “Pioneer” award for Innovation.

His other roles are President, Society of Physician Entrepreneurs, San Francisco Bay Area chapter; Member, Board of Directors, San Mateo County Medical Association; Assistant Clinical Professor of Medicine, University of California San Francisco.    Contact: LinkedIn or website: www.InnovatorMD.com