Open AccessPB2225 ARTIFICIAL INTELLIGENCE AIDING IN DIAGNOSIS OF JAK2 V617F NEGATIVE PATIENTS WITH WHO DEFINED ESSENTIAL THROMBOCYTHEMIA
Author(s) -
Belcic T.,
Cernelc P.,
Sever M.
Publication year - 2019
Publication title -
hemasphere
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.677
H-Index - 11
ISSN - 2572-9241
DOI - 10.1097/01.hs9.0000567380.33673.ae
Subject(s) - essential thrombocythemia , medicine , algorithm , hematology , referral , bone marrow , gastroenterology , platelet , computer science , family medicine
Background: Smart Blood Analytics (SBA) is an artificial intelligence program using machine learning algorithm in order to provide a list of most probable disease diagnosis in numerical order based on patient's blood test results. It was successfully tested in diagnosing hematological diseases and is especially targeted for medical doctors without detailed knowledge in the field of hematology. Aims: Our aim was to determine the applicability of SBA algorithm in patients with suspected essential thrombocythemia (ET). Methods: We analyzed 237 patients that presented at UMC Ljubljana between 04/2011 and 09/2016 with the referral diagnosis of ET. Based on the presence of CALR or MPL mutation and bone marrow examination we defined patients with true ET according to WHO 2008 or modified WHO criteria. All 237 patients were subsequently analyzed by SBA algorithm in order to determine the most probable hematologic diagnosis based on patients’ blood test results obtained at the time of referral. The diagnosis of interest according to ICD10 classification was D47 describing other neoplasms of uncertain behavior of lymphoid, hematopoietic and related tissue, including ET. We defined sensitivity and specificity of ET diagnosis by using SBA machine learning algorithm. Results: 10.5% (25/237) of patients were diagnosed with ET according to WHO 2008 or modified WHO criteria. 22 patients were CALR positive, 1 patient was MPL positive and 3 patients were triple negative. 89.4% (212/237) of patients were diagnosed with either secondary thrombocythemia or other neoplastic disease, such as myelodysplastic syndrome. The sensitivity of SBA algorithm to detect patients with ET was 100% and the specificity was 50.4%. Summary/Conclusion: The number of patients with true ET referred to our center was low. An improved referral and diagnostic approach for patients with thrombocythemia should be used. The SBA machine learning algorithm could be useful, especially in circumstances where hematologic consultation is not available. Further development of SBA machine learning algorithm is warranted.