Prediction of continuous positive airway pressure in obstructive sleep apnea

Continuous positive airway pressure (CPAP) prediction formulas can potentially simplify the treatment of obstructive sleep apnea (OSA). However, they can be difficult to derive and validate. We tested a statistical method to derive and validate a CPAP prediction formula using the same sample population. Seventy-six OSA patients underwent polysomnography and CPAP titration. Anthropometric measures, sleep parameters, and the Epworth sleepiness scale (ESS) were evaluated as predictors. All subsets regression was used to determine the optimum number of variables in the model. The Bayes information criterion was used to find the best-fit model. The model was then evaluated by a tenfold cross-validation procedure. Subjects were obese (BMI 31.3 ± 5.4) and had significant daytime somnolence (ESS 11.9 ± 5). Mean respiratory disturbance index (RDI) was 53.5 ± 31.3. The ESS was not predictive of titrated CPAP. The best-fit model included three variables (CPAP_pred = 30.8 + RDI × 0.03 − nadir saturation × 0.05 − mean saturation × 0.2). This model explained 67% of the variance. Our data and the literature suggest that a combination of two to three factors is predictive of titrated CPAP: RDI, oxyhemoglobin saturation, and obesity. Except for RDI, the specific factors vary in each population. A CPAP prediction formula that explains a high proportion of the titrated CPAP variance can be easily derived from parameters measured during the diagnostic work-up of OSA patients using a unique statistical model that allows derivation and validation of the formula in the same test population.