PB1118 - Construction of a new model for estimating causes of APTT prolongation using AI for clot waveform analysis (CWA)

Tuesday, June 27, 2023

18:30 – 19:30 ET

Room: Exhibition

Presenting Author(s)

Daiki Shimomura, PhD

Vice-Head Medical technologist
Department of Laboratory Medicine, Tenri Hospital
Tenri, Nara, Japan

Co-Author(s)

TM

Tomoko Matsumoto

Associate Professor
Department of Clinical Laboratory , Tenri Health Care University
Tenri, Nara, Japan
Masato Matsuda, PhD

Assistant professor
Fukushima Medical University
Fukushima, Fukushima, Japan
YT

Yuka Tabuchi

Employee
Sysmex Corporation
Kobe, Hyogo, Japan
AK

Aya Kouno

Medical Technologist
Department of Laboratory Medicine, Tenri Hospital
Tenri, Nara, Japan
YH

Yoshiki Hoshiyama

Head Medical technologist
Medical Laboratory Division, Niigata University Medical and Dental Hospital
Niigata, Niigata, Japan
TS

Takeshi Suzuki

Employee
Sysmex Corporation
Kobe, Hyogo, Japan
KN

Keisuke Nishi

Employee
Sysmex Corporation
Kobe, Hyogo, Japan
NA

Nobuo Arai

Employee
Sysmex Corporation
Kobe, Hyogo, Japan
HK

Hiroshi Kurono

Employee
Sysmex Corporation
Kobe, Hyogo, Japan
MM

Masato Moriyama

Associate Professor
Department of Medical Oncology, Niigata University Graduate School of Medical and Dental Sciences.
Niigata, Niigata, Japan
MK

Mikio Kamioka

Director
Department of Laboratory Medicine, Tenri Hospital
Tenri, Nara, Japan

Background: Activated partial thromboplastin time (APTT) is a screening test of the blood coagulation mechanism in routine clinical laboratory tests, and the causes of prolongation vary. Recently, a method for discriminating the cause of APTT prolongation using parameters obtained from CWA has been reported. However, many of the CWA parameters were not automatically calculated, and low versatility was a problem.

Aims: We attempted to construct a model to estimate the causes of APTT prolongation by applying machine learning, an artificial intelligence technology, to CWA parameters, and compared the sensitivity due to differences between reagent lots.

Methods: The subjects were 651 samples with prolonged APTT. The causes of APTT prolongation were confirmed based on medications administered and clinical information, and included samples of patients administered with heparin (n=99), direct oral anticoagulants (DOACs) (n=249), warfarin (n=105), and samples of lupus anticoagulant (LA) positive (n=76), and reduced factor VIII, factor IX, or factor VIII inhibitor positive (FVIII/FVIIIi/FIX) (n=122). CWA was measured using automated coagulation analyzers CN-6000 and Revoheme™ APTT SLA (Sysmex, Japan). Based on previous reports, we constructed a model for estimating the cause of APTT prolongation using 22 CWA parametersobtained from clot waveform, first- and second-derivative curves. In addition, the differential performance of the cause of prolongation was cross-validated.

Results: At a specificity of >95%, sensitivity was 82.8% for heparin, 84.7% for DOACs, 98.1% for warfarin, 78.9% for LA-positive, and 94.3% for FVIII/FVIIIi/FIX. In addition, the sensitivities of two different lots of APTT reagents were similar, and the difference between lots was therefore considered to be small.

Conclusion(s): We constructed a machine learning model to estimate the cause of APTT prolongation. The model is expected to be able to estimate the cause of prolongation by using the waveform information obtained from the APTT measurement. This model could contribute as a highly versatile screening test in hospital laboratories.