- About
- Our solutions
- Resource Center
January 31 2022
|
Reading time 12 minutes
A great number of tasks that physician is encountered with in their practice require quick and accurate response. Diagnosis and treatment must take into account:
• examination data
• patient’s individual characteristics
• results of laboratory and instrumental methods of analysis.
Routine operations, such as medical paperwork, monitoring of the patients’ condition, and control over the observation of prescriptions, create an additional workload for specialists.
The emergence of new rules of medical treatment, including clinical recommendations, standards and protocols, complicate the situation. Clinicians do not always have an opportunity to update their knowledge due to the increasing volume of specialized information and its sources (1).
In this context, it is important to ensure the timeliness and safety of clinical interventions: to detect a disease in time and start the correct treatment. Partial automation of diagnosis and treatment and specialist information support are aimed at reducing the burden on the physician.
Artificial intelligence implemented in a Clinical Decision Support System (CDSS) can assist physicians in solving clinical problems
CDSS is a computer program that provides a physician with patient’s clinical data and assist them in making decisions. Immediate information support allows for the developing the right medical tactics. Specialist’s clinical thinking combines with artificial intelligence to raise the quality of diagnosis and treatment (2).
Artificial intelligence in the CDSS is applied by means of crisp logic and fuzzy logic, machine learning methods and natural language processing (3). Computer vision and deep learning help to recognize medical images and interpret them (4).
As new data arrives, AI is able to learn on its own and improve the accuracy of decision-making (3).
• Reference and information support
The service provides a specialist with the latest clinical recommendations and protocols. Information about medicines helps to refine certain aspects of therapy (5).
• Assistance with medical records
The system sorts and accounts electronic medical records. The patient’s disease is coded according to the International Classification of Diseases Version 10 (ICD-10). When making diagnosis, a physician chooses relevant code (6).
• Identification of the degree and severity of the disease
The algorithm assesses patient’s condition using machine learning classifications. The physician receives a conclusion about the risk group as an output (7).
• Warning signals generation
The CDSS is able to reveal hidden patterns that a physician may not notice. The algorithms warn about possible complications, for example, elevated glucose level in diabetes mellitus (9), infection in the postoperative period (9), decompensation of the disease (10).
• Assistance in diagnosis
AI is able to carry out a “consultation” based on the input: the user’s symptoms and complaints. The system provides a diagnosis that serves as a guide for a physician (11). Another direction is a diagnostic visualization. The CDSS recognizes medical images and highlights suspicious areas (2).
• Treatment optimization
The CDSS helps to find the correct treatment based on the electronic medical records data (10), to prescribe the optimal dosage of medicines (12), predict the length of hospital stay and monitor the therapy (13).
• The attainment of economic efficiency
The introduction of AI helps to direct resources of a medical organization properly. This is achieved due to the efficient use of diagnostic equipment (14) and electronic documentation (10).
Diagnosis with the use of the CDSS is based on an organic combination of the professional skills of physicians and technical capabilities of AI. AI’s conclusion has a probabilistic nature and serves as a guide for the clinician. The system automates routine processes in medicine.
Google supported the development of an algorithm for automatic detection of diabetic retinopathy, retinal damage in diabetes mellitus. Training data included 128,175 images of retina (15).
Sets for clinical tests consisted of 9,963 and 1,748 images respectively. Ophthalmologists evaluated the quality of the image and the presence of visual pathology on it. Their conclusions were compared with those of the neural network. According to the researchers, the sensitivity of the sets was 97.5% and 96.1% (15).
The Russian service Care Mentor AI uses AI to enhance the accuracy of the radiologist’s conclusions. The expert system analyses the results of radiology examinations to detect signs of thoracic organs diseases (16).
Care Mentor AI Service makes it possible to (16):
• detect a damage in lung tissue, presence of a foreign body or a cavity with liquid level
• identify areas of pathological changes
• calculate the size of suspicious formations
• formulate a final conclusion.
In Russia the SberMedAI team is implementing a multi-component medical decision-making support system:
• TOP-3: analyzes data from the initial appointment and provides 3 most probable diagnosis under the ICD-10
• CT Stroke: searches for areas of acute cerebral circulation disorder on tomograms
• CT Lungs: analyzes x-rays of the thoracic organs to look for the signs of viral pneumonia, including caused by COVID-19 or cancer. The service highlights affected areas and minimal nodular densities in the lung tissue.
• Mammography: identifies suspicious masses in the mammary gland
• Short ECG: processes the ECG results with online display
• AI Resp mobile app: studies acoustic characteristics in respiratory system diseases
• AI Skin mobile app: identifies skin rashes by photo.
In order to choose treatment, clinicians evaluate the degree of the disease assigning a particular group or category to it. They use generally accepted international classifications as a guideline.
Clinical Decision Support System accelerates this task. AI uses input data about the patient’s health to do it, including gender, age, previous diseases, symptoms and laboratory measures (10).
The researchers of the Cancer Institute of Bari Named After John Paul II in Italy have published preliminary results on the prediction of breast cancer recurrence. The used machine learning-based model. The results of examination and treatment of 256 research center patients were used as the input. The probability of cancer recurrence was assessed in 5 and 10 year terms after the diagnosis (17).
The Italian researchers highlighted the efficiency of the AI-based auxiliary tool: prediction accuracy was 77.5% and 80.39%, and sensitivity was 92.31% and 95.83% for 5 and 10 year terms respectively (17).
The CDSS information support can help a physician to make reasonable decisions when choosing therapy (5):
• to see recommended dosages
• to check possible contraindications and unwanted medicine interactions
• to look for cheaper generics
• to make sure that the prescribed treatment is correct and safe.
German scientists from the University Hospital of Mannheim compared the decisions made by 77 physicians and 89 medical students using various sources of information. The participants were asked to prescribe antibiotic therapy for upper urinary tract infections. The sources of information included Internet, guidebooks and CDSS . The system provided the user with recommendations taking into account patient’s individual features (18).
The scientists noted that only 27.1% of participants made the correct diagnosis, while only 19.4% managed to prescribe treatment in accordance with the standards. This shows the necessity of informational support of medical specialists (18).
The participants whose decisions were based on AI showed the best results: 57.1% and 40.5% among those who made the correct diagnosis and prescribed the right treatment respectively (18).
In Russia there are services to improve the quality of pharmacotherapy. The Electronic Clinical Pharmacologist system provides a physician with up-to-date information on medicines – State Drug Register and clinical recommendations (19).
The CDSS facilitates the evaluation of qualitative and quantitative measures of patients’ health in dynamics. The use of AI allows a physician to concentrate on each stage of the disease. Control of the individual features is aimed at timely correction of therapeutic and diagnostic measures.
A publication in The Lancet by British scientists describes a model that monitors patients’ condition. The Oracle algorithm makes predictions about the results of the psychotherapy. To do it, the Oracle:
• uses data from patients’ questionnaires;
• evaluates depression and anxiety symptoms during each psychotherapy session
• updates prediction of the model after each new session (20).
High rates of prediction accuracy allowed the authors to consider the Oracle applications in medicine. The algorithm can help a physician to identify patients at risk already at the first sessions (20).
The American researchers from the Duke University in North Carolina have developed machine learning models to evaluate postoperative risks of complications and mortality. The algorithm was processing a large set of clinical and surgical patient data. The analysis covered 37 million clinical cases and involved 194 indications (20).
According to the authors, the created model managed to identify patients with high risks after surgery with sensitivity and specificity of 76% (21).
Both domestic and foreign research reviews show the active development of intellectual systems. The effectiveness of the AI use is under study. However, even now technological solutions are yielding the first positive results in diagnosis and treatment (2, 22-24).
At this stage we can identify the following aspects of the CDSS use (25):
• High computer literacy level.Knowing the basics of bioinformatics will help physicians take full advantage of medical technology.
• Transparent decisions. Big data processing methods may not be clear to a user. A clinician needs to know the general principles of how the technology works: how the AI came to this or that conclusion and which performance characteristics and limits the system has.
• Ethics. Algorithmic bias may make a decision based not on the patient’s features, but on mathematically calculated use. A physician follows the non-maleficence principle and cares about each human life. A specialist take into account the patient’s life situation and preferences in the choice of treatment.
References
1. S.M. Stivaros, A Gledson, G Nenadic et al. Decision Support Systems for Clinical Radiological Practice — Towards the Next Generation // Br J Radiol. 2010.
2. Reed T. Sutton, David Pincock, Daniel C. Baumgart et al. An Overview of Clinical Decision Support Systems: Benefits, Risks, and Strategies for Success // npj Digital Medicine. 2020; 3(17). DOI: https://doi.org/10.1038/s41746-020-0221-y
3. Expert.ai. team What is Machine Learning? A Definition. // Expert.ai. 2020. URL: https://www.expert.ai/blog/machine-learning-definition/ (accessed: 24/12/2021).
4. What is Computer Vision? [Online] // IBM. URL: https://www.ibm.com/topics/computer-vision/ (accessed: 24/12/2021).
5. Muslimah Ithnin, Mohd Dzulkhairi Mohd Rani, Zuraidah Abd Latif et al. Mobile App Design, Development, and Publication for Adverse Drug Reaction Assessments of Causality, Severity, and Preventability // JMIR Mhealth Uhealth. 2017;5(5):e78. DOI: https://dx.doi.org/10.2196/mhealth.6261
6. Christopher M Bell, Arash Jalali, Edward Mensah. A Decision Support Tool for Using an ICD-10 Anatomographer to Address Admission Coding Inaccuracies: A Commentary // Online J Public Health Inform. 2013. DOI: https://doi.org/10.5210/ojphi.v5i2.4813
7. Guangyao Wu, Pei Yang, Yuanliang Xie et al. Development of a Clinical Decision Support System for Severity Risk Prediction and Triage of COVID-19 Patients at Hospital Admission: an International Multicentre Study // Eur Respir J. 2020; 56(2): 2001104. DOI: https://doi.org/10.1183/13993003.01104-2020
8. Ashenafi Zebene Woldaregay, Eirik Årsand, Ståle Walderhaug et al. Data-Driven Modeling and Prediction of Blood Glucose Dynamics: Machine Learning Applications in Type 1 Diabetes // Artif Intell Med. 2019. DOI: https://doi.org/10.1016/j.artmed.2019.07.007
9. Thara Tunthanathip, Sakchai Sae-heng, Thakul Oearsakul et al. Machine Learning Applications for the Prediction of Surgical Site Infection in Neurological Operations // Neurosurg Focus. 2019. DOI: https://doi.org/10.3171/2019.5.FOCUS19241
10. Chris Giordano, Meghan Brennan, Basma Mohamed et al. Accessing Artificial Intelligence for Clinical Decision-Making // Front. Digit. Health. 2021. DOI: https://doi.org/10.3389/fdgth.2021.645232
11. Michael M. Segal, Alanna K. Rahm, Nathan C. Hulse et al. Experience with Integrating Diagnostic Decision Support Software with Electronic Health Records: Benefits versus Risks of Information Sharing // EGEMS (Wash DC). 2017; 5(1): 23. DOI: https://doi.org/10.5334/egems.244
12. Athanasios Tsoukalas, Timothy Albertson, Ilias Tagkopoulos. From Data to Optimal Decision Making: A Data-Driven, Probabilistic Machine Learning Approach to Decision Support for Patients With Sepsis // JMIR Med Inform 2015;3(1):e11. DOI: https://doi.org/10.2196/medinform.3445
13. Konstantinos C. Siontis, Xiaoxi Yao, James P. Pirruccello et al. How Will Machine Learning Inform the Clinical Care of Atrial Fibrillation? // Circulation Research. 2020;127:155–169. DOI: https://doi.org/10.1161/CIRCRESAHA.120.316401
14. C. Craig Blackmore, Robert S. Mecklenburg, Gary S. Kaplan. Effectiveness of Clinical Decision Support in Controlling Inappropriate Imaging // J Am Coll Radiol. 2011; 8(1): 19-25. DOI: https://doi.org/10.1016 / j.jacr.2010.07.009
15. Varun Gulshan, Lily Peng, Marc Coram et al Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs // JAMA. 2016; 316(22): 2402-2410. DOI: https://doi.org/10.1001/jama.2016.17216
16. Care Mentor AI: [Online]. URL: https://carementor.ru/ (accessed: 26/12/2021).
17. Raffaella Massafra, Agnese Latorre, Annarita Fanizzi et al. A Clinical Decision Support System for Predicting Invasive Breast Cancer Recurrence: Preliminary Results // Front Oncol. 2021 Mar 11;11:576007. DOI: https://doi.org/10.3389/fonc.2021.576007
18. M. Neugebauer, M. Ebert, R. Vogelmann. A Clinical Decision Support System Improves Antibiotic Therapy for Upper Urinary Tract Infection in a Randomized Single-Blinded Study // BMC Health Services Research. 2020; 20,185. DOI: https://doi.org/10.1186/s12913-020-5045-6.
19. Electronic Clinical Pharmacologist (Электронный клинический фармаколог, ЭКФ): [Online]. URL: https://www.ecp.umkb.com/ (accessed: 26/12/2021).
20. КClaire Bone, Melanie Simmonds-Buckley, Richard Thwaites et al. Dynamic Prediction of Psychological Treatment Outcomes: Development and Validation of a Prediction Model Using Routinely Collected Symptom Data // The Lancet. 2021. DOI: https://doi.org/10.1016/S2589-7500(21)00018-2
21. Kristin M. Corey, Sehj Kashyap, Elizabeth Lorenzi et al. Development and Validation of Machine Learning Models to Identify High-Risk Surgical Patients Using Automatically Curated Electronic Health Record Data (Pythia): A Retrospective, Single-Site Study // PLoS Med 15(11): e1002701. DOI: https://doi.org/10.1371/journal.pmed.1002701
22. Taku Harada, Taiju Miyagami, Kotaro Kunitomo et al. Clinical Decision Support Systems for Diagnosis in Primary Care: A Scoping Review // Int J Environ Res Public Health. 2021; 18 (16): 8435. DOI: https://doi.org/10.3390/ijerph18168435
23. Kupeeva I. A., Raznatovskiy C. I. Ravodin R. A. et al. Efficiency Assessment of Clinical Decision Support System // Vestnik of St Petersburg University. Applied Mathematics. Computer Science. Control Processes. 2016; 2: 62-68. DOI: https://doi.org/10.21638/11701/spbu10.2016.206
24. Rebrova O. Yu. Efficacy of Clinical Decision Support Systems: Methods and Estimates // Clinical and Experimental Thyroidology. 2019; 15(4): 148-155. DOI: https://doi.org/10.14341/ket12377
25. Tamra Lysaght, Hannah Yeefen Lim, Vicki Xafis, Kee Yuan Ngiam AI-Assisted Decision-Making in Healthcare // Asian Bioethics Review. 2019; 11: 299-314. DOI: https://doi.org/10.1007 / s41649-019-00096-0
