Symbolic dynamics to enhance diagnostic ability of portable oximetry from the phone oximeter in the detection of paediatric sleep apnoea

Daniel Álvarez; Andrea Crespo; Fernado Vaquerizo-Villar; Gonzalo Cesar Gutierrez-Tobal; Ana Cerezo-Hernández; Verónica Barroso-García; J Mark Ansermino; Guy A Dumont; Roberto Hornero; Felix Del Campo; Ainara Garde

doi:10.1088/1361-6579/aae2a8

Symbolic dynamics to enhance diagnostic ability of portable oximetry from the phone oximeter in the detection of paediatric sleep apnoea

Physiol Meas. 2018 Sep 19. doi: 10.1088/1361-6579/aae2a8. Online ahead of print.

Affiliations

¹ Pneumology Service, Rio Hortega University Hospital, Valladolid, Valladolid, SPAIN.
² Biomedical Engineering Group, ETSI Telecomunicación, Universidad de Valladolid, Valladolid, Castilla y León, SPAIN.
³ Biomedical Engineering Group ETS Ingenieros de Telecommunicacion, Universidad de Valladolid, Camino del Cementerio sn, 47011 Valladoid, Valladolid, SPAIN.
⁴ The University of British Columbia, Vancouver, CANADA.
⁵ University of British Columbia, Vancouver, British Columbia, CANADA.
⁶ Universiteit Twente, Enschede, 7500 AE, NETHERLANDS.

PMID: 30230476
DOI: 10.1088/1361-6579/aae2a8

Abstract

Objective: This study is aimed at assessing symbolic dynamics as a reliable technique to characterise complex fluctuations of portable oximetry in the context of automated detection of childhood obstructive sleep apnoea-hypopnoea syndrome (OSAHS).

Approach: Nocturnal oximetry signals from 142 children with suspected OSAHS were acquired using the Phone Oximeter: a portable device that integrates a pulse oximeter with a smartphone. An apnoea-hypopnoea index (AHI) ≥5 events/h from simultaneous in-lab polysomnography was used to confirm moderate-to-severe childhood OSAHS. Symbolic dynamics was used to parameterise non-linear changes in the overnight oximetry profile. Conventional indices, anthropometric measures, and time-domain linear statistics were also considered. Forward stepwise logistic regression was used to obtain an optimum feature subset. Logistic regression (LR) was used to identify children with moderate-to-severe OSAHS.

Main results: The histogram of 3-symbol words from symbolic dynamics showed significant differences (p <0.01) between children with AHI <5 events/h and moderate-to-severe patients (AHI ≥5 events/h). Words representing increasing oximetry values after apnoeic events (re-saturations) showed relevant diagnostic information. Regarding the performance of individual characterization approaches, the LR model composed of features from symbolic dynamics alone reached a maximum performance of 78.4% accuracy (65.2% sensitivity; 86.8% specificity) and 0.83 area under the ROC curve (AUC). The classification performance improved combining all features. The optimum model from feature selection achieved 83.3% accuracy (73.5% sensitivity; 89.5% specificity) and 0.89 AUC, significantly (p-value <0.01) outperforming the other models.

Significance: Symbolic dynamics provides complementary information to conventional oximetry analysis enabling reliable detection of moderate-to-severe paediatric OSAHS from portable oximetry.

Keywords: Paediatric obstructive sleep apnoea-hypopnoea syndrome; nocturnal oximetry; pattern recognition; portable; signal processing; symbolic dynamics.