Residual subjective daytime sleepiness under CPAP treatment in initially somnolent apnea patients: A pilot study using data mining methods

Background and purposeDespite correct treatment with positive airway pressure (PAP), obstructive sleep apnea (OSA) patients sometimes remain subjectively somnolent. The reliability of the Epworth Sleepiness Scale (ESS) has been established for healthy subjects and patients under stable conditions; the ESS may eventually vary among treated OSA patients, biasing the results of a cross-sectional analysis of persisting sleepiness. The objective of this study was to depict the evolution of subjective vigilance under treatment using an index of ESS variability (ΔESS).MethodsIn 80 OSA patients (apnea–hypopnea index [AHI] = 54 ± 26/h), initially somnolent (ESS = 15 ± 3) and treated with auto-titrating PAP (APAP) (oxyhaemoglobin desaturation index 3% [ODIapap] = 3.4 ± 2.2/h; daily APAP use = 5.3 ± 1.5 h) during 434 ± 73 days, ESS scores were regularly collected four times every 109 ± 36 days. DESS was calculated and data mining methods (Segmentation and Decision Tree) were used to determine homogeneous groups according to the evolution of ESS scores.ResultsWhen assessed cross-sectionally, 14–25% of the subjects were recognized as somnolent, depending on the moment when ESS was administered. Using data mining methods, three groups were clearly identifiable: two without residual somnolence – group 1, n = 38 (47%), with high ΔESS = −2.9 ± 0.8, baseline ESS = 16.3 ± 3.3, AHI = 58.5 ± 26.1/h, mean ESSapap = 5.1 ± 2.4 and group 2, n = 31 (39%), with low ΔESS = −1.1 ± 0.5, baseline ESS = 13.2 ± 1.4, AHI = 53 ± 27.3/h, mean ESSapap = 8.8 ± 1.9; and one with persisting sleepiness; group 3, n = 11 (14%), with low ΔESS = −0.3 ± 0.8, baseline ESS = 16.3 ± 3, AHI = 38.7 ± 10.8/h, mean ESSapap = 14.1 ± 1.9. Compliance to PAP was high and comparable in the three groups. Age and body mass index (BMI) did not differ.ConclusionData mining methods helped to identify 14% of subjects with persisting sleepiness. Validation needs to be done on a larger population in order to determine predictive rules.