November 2019: Artifical Intelligence and Coagulation Testing

by Donna Castellone, MS, MT (ASCP) SH • October 31, 2019



What is artificial intelligence (AI) and how is it used in healthcare, and in particular in the coagulation laboratory? More than 2500 years ago the abacus was invented and contributed as a precursor to the computer and AI. Its principle incorporated the process of repeating calculations faster than the human brain. AI uses this concept with the speed of computers, as well as using algorithms and decision trees. These tools are used to solve complex medical problems. Big data and advances in technology are driving opportunities for the application of AI and machine learning (ML) in health care and clinical decision-making at an unprecedented pace.(1) AI mimics human cognitive functions. It can be used in the detection and diagnosis, treatment and outcome prediction and prognosis evaluation. This process has been used in cancer, neurology and cardiology.(2)

In the clinical laboratory, as early as 1984 Rutgers University developed “EXPERT” a consultation system tool that enabled sequential laboratory testing and interpretation. Systems are not capable of learning themselves, but based on the data that they have programmed, they make decisions on accumulated knowledge(3). Precision medicine and personalized treatments are venues for clinical decision making.(3)


Clinical diagnostic algorithms allow computers to relate diverse clinical literature with patient data to generate outcome reports. AI looks at images, graphs, clinic notes and literature and converts this unstructured data into a structured format. Aligning software with other knowledge banks helps the system become intelligent.(1)

 

STROKE:

AI has been used in stroke detection, treatment and outcomes. This disorder affects more than 500 million people worldwide and costs about $689 billion in medical expenses. Up to 85% of strokes are caused by a thrombus in the vessel.(2)

Early stroke detection allows for timely treatment. Studies have applied ML along with neuroimaging data to assist with stroke diagnosis. A detection process was implemented that included a human activity recognition stage and stroke onset detection stage. When the movement of a patient is different from the normal pattern, there would be an alert for stroke and treatment would begin. Accuracy for stroke identification was at 87.6%.(4) ML methods were also used to analyze CT scans from stroke patients. Since it is difficult to distinguish carotid plaque from free floating thrombus on CT imagining, ML algorithms were used to classify by quantitative shape analysis with an accuracy of 65.2% and 75.4%.(4)

For treatment of stroke, support vector machine (SVM) methods were used to predict whether patients with tPA treatment would develop symptomatic intracranial hemorrhage. This incorporated using practice guidelines, meta-analysis and clinical trials as well as 56 different variables and 3 decisions for analysis.(5)

Management of stroke is a complicated process that requires several clinical decision points. Research has focused on a single or limited clinical questions while ignoring the continuous nature of stroke management. Using large amounts of available data available on this chronic disease AI can look at more complicated scenarios by looking at real life clinical questions, resulting in better decision making in managing stroke patients.(6)

 

DISSEMINATED INTRAVASCULAR COAGULATION (DIC)

DIC is a complex condition resulting from an underlying event such as sepsis, severe trauma or advanced cancer. This occurs due to intravascular activation of coagulation including the simultaneous activation of both thrombin and plasmin. There is no one specific test to diagnosis DIC. Scoring systems have been developed to aid in the diagnostic criteria for DIC. Results from the PT, platelet count, fibrinogen and fibrin markers (D-dimer or FDP) are also evaluated. Several scoring systems lack the sensitivity required, while others have limitations based on underlying criteria.

Patients (n=837) from Seoul, Korea suspected of DIC were evaluated in a study. After exclusion criteria, there were 656 suspected cases requiring an evaluation of DIC. A full set of laboratory results and clinical diagnosis process were obtained. Tests included: CBC, PT, aPTT, D-dimer, Fibrinogen and or Protein C. Each case was reviewed by two experts. If there was a discrepancy, a third expert was consulted.(7) Using the dataset, several ML algorithms were tested. The model enrolled 32 clinical parameters. The study uncovered that there were overlooked clinical parameters in the scoring system used to aid in diagnosis.

Results revealed data classified 228 cases as DIC and 428 cases as non-DIC. Several ML algorithms were tested and an artificial neural network was established. When compared to several established scoring systems (ISTH, JMHW, JAAM) the ANN model had higher area under the curve for both the DIC and non-DIC.(7)

It was found that additional variables such as anticoagulant use, bleeding, thrombosis, hematological malignancy, hepatic failures, and tissue damage can further enhance and facilitate the diagnosis of DIC. Performance can be further improved by adding more data and applying advanced algorithms and parameters.(7)

A limitation that occurred in this study is the classification of DIC by medical experts, which could have been incorrect. Developing a ML model that looks at consecutive changes of variable laboratory parameters could eliminate variations in expert opinions. However, ML cannot go beyond what is contained in the data, so including additional tests such as thromboelastography and tests to identify, histone DNA complexes, and circulating histones may improve performance. It may be possible to implement this approach into the EMR as a clinical decision support system in the future.(7)

 

CONCLUSIONS:

There are 2 hurdles with AI technologies: The first is regulatory requirements and the lack of standards to evaluate the safety and efficacy of AI systems. The FDA has made an attempt to provide guidance for assessing these systems which can be classified as general wellness products with low risk to patients. It also clarifies that rules for the adaptive design in clinical trials. The second hurdle is data exchange. In order for these AI systems to work, continuation of data supply is crucial for further development and improvement of the system.(2)

AI can contribute to the evolution of laboratory technology in reducing human error and providing objective guidance on optimal patient therapy. It can also contribute to reducing health care costs by reducing human labor. There are limitations such as upfront costs to implement AI, the reliability of results and the limitations of the current electronic health record. Other considerations include FDA approval, data security and questions around reimbursement. AI tools will help to advance patient care, management of disease prevention and treatment. It can have a significant role in laboratory workflow and testing.(8)

 


REFERENCES:

  1. Khan, A., Is Laboratory Medicine Ready for Artificial Intelligence? Clinical Laboratory News, AUG.1.2017
  2. Jiang F, Jiang Y, Zhi H, et al Artificial intelligence in healthcare: past, present and future Stroke and Vascular Neurology 2017;2\
  3. Menon, PK., Effect of Artificial Intelligence in the Clinical Laboratory https://www.medlabme.com/magazine/en/Issue/2018-issue-21
  4. Bentley P , Ganesalingam J ,Carlton Jones AL, et al Prediction of stroke thrombolysis outcome using CT brain machine learning. Neuroimage Clin 2014;4:635–40.
  5. Love A , Arnold CW , El-Saden S , et al Unifying acute stroke treatment guidelines for a bayesian belief network. Stud Health Technol Inform 2013;192:1012
  6. Ye H , Shen H , Dong Y , et al . Using Evidence-Based medicine through Advanced Data Analytics to work toward a National Standard for Hospital-based acute ischemic Stroke treatment. Mainland China, 2017
  7. Yoon JG, Heo J, Kim M, et al. Machine learning-based diagnosis for disseminated intravascular coagulation (DIC): Development, external validation, and comparison to scoring systems. PLoS One. 2018;13(5):e0195861. Published 2018 May 2.
  8. Kim, A., Monique, J, Artificial Intelligence and In Vitro Diagnostics: Advancing Patient Care, March 2019
    https://www.iqvia.com/blogs/2019/03/artificial-intelligence-and-in-vitro-diagnostics-advancing-patient-care