Background

Central venous catheter (CVC)-related venous thromboembolism (VTE) is a common complication in the pediatric cardiac institute care unit (PCICU) that can lead to pulmonary embolism, post-thrombotic syndrome, and cerebrovascular events. Patients in the PCICU are in a hypercoagulable state following surgery and the endothelial damage and stasis of blood flow that accompanies the insertion of CVCs, completing Virchow’s triad for thrombosis formation. Indeed, pediatric patients undergoing surgery for congenital heart anomalies show a VTE incidence rate of 2–18% [1, 2]. Clinically, suspicion of thrombosis is aroused by clinical observations of edema, pain, tenderness, and phlebitis, and thrombosis is subsequently confirmed with ultrasound. However, almost 40% of thrombotic events are asymptomatic, and frequent ultrasound surveillance is extremely costly. Once detected, VTE can then be treated with anti-coagulation therapy. Thus, while detection and treatment are relatively straightforward, initial suspicion relies solely on clinical observation.

Our research aims to develop an alert system that can automatically suspect thrombosis from high-frequency central venous pressure (CVP) data, notifying the clinical team of events requiring ultrasound confirmation. As an important step toward the VTE predictive system, we developed a method to successfully remove baseline drift and impulsive noise from the CVP waveform data, reducing false positives due to motion and fluid/medication administration artifacts.

Methods

Patients in the PCICU at Dell Children’s Medical Center from December 2018 to January 2021 with ultrasound-confirmed VTE were included in this single-center retrospective trial. The type of CVC (peripherally inserted central catheter versus central venous line), the anatomical location of the CVC, the size of the CVC, the date of CVC insertion, and the age of the patient at the time of insertion were all recorded. The CVP waveforms were sampled at 125 Hz and recorded with the Sickbay Platform (Medical Informatics Corp., Houston, TX, USA). These signals are nonlinear and non-stationary and contain both impulsive noise and baseline drift artifacts.

Iterative filtering (IF), a signal decomposition technique that can handle nonlinear, nonstationary signals [3], was used to decompose CVP waveforms into a series of intrinsic mode functions (IMFs). Baseline drifts were removed by early termination of the decomposition and removal of the remainder. Instantaneous frequency analysis of the resulting IMFs (i.e., the IMFogram [4]) was used to generate a time-frequency (TF) representation of the signal. Outlier detection on the IMFogram was used to remove impulsive noise. Manual curation of impulsive noise in the 4 hours after CVC insertion served as labels to assess algorithm performance.

Results

Manual curation of impulsive noise labeled 6% ± 3.5% of data as noise. The TF representation provides useful features for removing this impulsive noise within CVP data. Notably, this approach based on IF techniques can rapidly identify and remove noise due to the numerical performance of IF techniques compared to other TF analysis techniques, enabling eventual real-time translation. Furthermore, resampling IMFs after noise removal helps preserve signal decomposition despite
nonlinearities.

Implications

By removing the noise from high-frequency CVP data, we now have clean CVP data that can be used to develop CVC-VTE
detection algorithms for use in congenital heart surgery patients. Future research will evaluate adaptive procedures that
personalize noise removal and fault detection for individual patients.

References

[1] L. D. Murphy, B. D. Benneyworth, E. A. S. Moser, K. M. Hege, K. M. Valentine, and C. W. Mastropietro, “Analysis ofpatient characteristics and risk factors for thrombosis after surgery for congenital heart disease,” Pediatric Critical CareMedicine, vol. 19, no. 12, pp. 1146–1152, Dec. 2018.

[2] E. H. Steen, J. J. Lasa, T. C. Nguyen, S. G. Keswani, P. A. Checchia, and M. M. Anders, “Central venous catheter-related deep vein thrombosis in the pediatric cardiac intensive care unit,” Journal of Surgical Research, vol. 241, pp. 149–159, Sep. 2019.

[3] A. Cicone and H. Zhou, “Numerical analysis for iterative filtering with new efficient implementations based on FFT,” Numerische Mathematik, vol. 147, no. 1, pp. 1–28, Jan. 2021.

[4] A. Cicone, W. S. Li, and H. Zhou, “New theoretical insights in the decomposition and time-frequency representation of nonstationary signals: The IMFogram algorithm,” Applied and Computational Harmonic Analysis, vol. 71, p. 101634, 2024.