Early detection of HVAD pump thrombosis based on technical analysis and power consumption measurements

Background Continuous‐flow left ventricular assist devices (LVADs) have been extensively used in a strategy of bridge to orthotopic heart transplant and destination therapy. The usage of LVAD, however, is not free from limitations such as device‐related adverse events, including pump thrombosis (PT)...

Full description

Saved in:
Bibliographic Details
Published in:Artificial organs 2022-06, Vol.46 (6), p.1142-1148
Main Authors: Krysiński, Mateusz, Gawlikowski, Maciej, Biełka, Agnieszka, Krysińska, Małgorzata, Małyszek‐Tumidajewicz, Justyna, Copik, Izabela, Pacholewicz, Jerzy, Zembala, Marian, Zembala, Michał
Format: Article
Language:English
Subjects:
Algorithms
early detection
early event detection
Heart Failure - diagnosis
Heart Failure - surgery
Heart Transplantation
Heart-Assist Devices - adverse effects
Humans
HVAD pump
left ventricular assisting device
Parameters
Patients
Power consumption
pump thrombosis
Retrospective Studies
Signal processing
Thromboembolism
Thromboembolism - etiology
Thrombosis
Thrombosis - diagnosis
Thrombosis - etiology
Thrombosis - therapy
Ventricle
Ventricular assist devices
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Background Continuous‐flow left ventricular assist devices (LVADs) have been extensively used in a strategy of bridge to orthotopic heart transplant and destination therapy. The usage of LVAD, however, is not free from limitations such as device‐related adverse events, including pump thrombosis (PT). We aimed to develop an algorithm of early PT detection based on the maintenance parameters monitored by the implanted device. Methods We analyzed log files of 101 patients implanted with HeartWare pump (HVAD) with 18 PT events among them. For signal processing, we used the open‐high‐low‐close format transformation and typical price (TP) technical analysis indicator. Model parameters were tuned with 5‐fold cross‐validation, and the final performance was measured on a separate group of patients. Results Our algorithm achieved 100% sensitivity and 100% specificity of indications. In the final evaluation, alarms preceded the clinical acknowledgement of events by 2 days and 20 h on average. In the worst‐case scenario, an alarm was raised 1 day and 8 h prior to the event. Conclusions The proposed algorithm could be installed to work directly with the device controller and provide clinicians with automatic readings analysis, raising an alarm when there is a high probability of thromboembolism. Early event detection could enable better thrombosis management and improve prognosis in patients implanted with HVAD. Patient's characteristics are gathered over 2 days. For every consecutive datapoint, an alarm is triggered if the specified probability threshold is reached. Algorithm achieved 100% sensitivity and 100% specificity in cross‐validation and on the test group. It could be installed to work directly with the device controller and provide clinicians with automatic readings analysis.
ISSN:0160-564X
1525-1594
DOI:10.1111/aor.14163