[Prediction models for respiratory event types in OSA patients based on hypoxic parameters].

ABSTRACT

Objective: To analyze the hypoxic parameters in patients with obstructive sleep apnea (OSA), to explore the difference and association between different types of respiratory events and to construct predictive models for respiratory event types. Methods: Fifty patients [including 41 males and 9 females with age 18-74(45.72±13.39) years ] with OSA diagnosed by polysomnography (PSG) were selected for retrospective analysis, and all respiratory events with pulse oximetry (SpO2) desaturation in the recorded overnight data were divided into hypopnea group (Hyp, 3 316), obstructive apnea group (OA, 5 552), central apnea group (CA, 1 088) and mixed apnea group (MA, 1 369) according to the type of events, and all event records were exported separately from the PSG software as comma-separated variable (.csv) files, which were imported and analyzed using the in-house built Matlab software. A total of 13 hypoxic parameter differences were compared among the four groups, including minimum oxygen saturation of events (e-minSpO2), the depth of desaturation (ΔSpO2), the duration of desaturation and resaturation (DSpO2), the duration of desaturation (d.DSpO2), duration of resaturation (r.DSpO2), duration of SpO2<90% (T90), duration of SpO2<90% during desaturation (d.T90), duration of SpO2<90% during resaturation (r.T90), area under the curve of SpO2<90% (ST90), area under the curve of SpO2<90% during desaturation (d.ST90), area under the curve of SpO2<90% during resaturation (r.ST90), oxygen desaturation rate (ODR) and oxygen resaturation rate (ORR). Hyp model (H), OA model (O), CA model (C) and MA model (M) were constructed respectively; group differences for the different hypoxia parameters were assessed using single factor analysis and Kruskal-Wallis H test. For different categories of respiratory events, binary logistic regression was used to identify the variables included in the regression model. Receiver operating characteristic (ROC) curves were generated to assess and compare the sensitivity, specificity, positive predictive value and negative predictive value of the four models, thereby gauging the predictive precision of each model. Results: ΔSpO2, ODR, ORR, T90, d.T90, r.T90, ST90, d.ST90 and r.ST90 for each type of respiratory events showed MA>OA>CA>Hyp, and e-minSpO2 showed MALogistic regression showed that e-minSpO2, ΔSpO2, d.DSpO2, r.DSpO2, ODR, ORR, d.T90, r.T90, d.ST90 and r.ST90 were independent predictors for Hyp, ΔSpO2, d.DSpO2, r.DSpO2, ORR, d.T90, r.T90, d.ST90 and r.ST90 were independent predictors for OA, ΔSpO2, d.DSpO2, r. DSpO2, ODR, ORR, r.T90, d.ST90 and r.ST90 were independent predictors for CA, while e-minSpO2, ΔSpO2, d.DSpO2, r.T90, d.ST90 and r.ST90 were independent predictors for MA. The area under the ROC curve (AUC) of the H, O, C, and M models were 0.875, 0.751, 0.755, and 0.749, respectively, and all models had high specificity (0.865, 0.722, 1.000, and 0.993, respectively) and negative predictive values (0.871, 0.692, 0.904, and 0.881, respectively).

Conclusions:

Four predictive models for respiratory event types can be constructed based on hypoxic parameters, providing a feasible novel tool for applying nocturnal SpO2 to automatically identify respiratory event types.