Rationale: In patients with obstructive sleep apnea (OSA) treated with continuous positive airway pressure (CPAP), the apnea hypopnea index (AHI) measured off CPAP may be decreased relative to baseline AHI preceding CPAP treatment. Semi-invasive "endo-phenotyping" sleep studies attribute this fall in AHI primarily to improved ventilatory control stability. Phenotyping Using Polysomnography (PUP) attempts to reproduce these studies using routine polysomnography (PSG). Objectives: To determine whether changes in AHI following CPAP associate primarily with changes in PUP-estimated ventilatory control stability (loop gain, LG1) or with changes in other PUP-estimated pathophysiologic mechanisms. Methods: PUP analyses were performed on existing PSGs in research participants who underwent baseline PSG, 4.4±2.2 months CPAP therapy, and CPAP withdrawal with repeat PSG on night 2 of withdrawal. Pre-CPAP PUP-estimated LG1, arousal threshold (ArTH), and upper airway collapsibility (Vpassive) and compensation (Vcomp) were compared to corresponding values during CPAP withdrawal. Mixed effects models were constructed to determine which PUP estimate best explained changes in AHI. Results: PSG data were available for 35 participants (age 47±10.8 years; 12 female; BMI 38.5±8.6 kg/m2, AHI3A 58.8±33.1 events/hr, 9 mild/moderate OSA, 26 severe OSA). Following CPAP, AHI decreased, but the change was not statistically significant. However, a significant decrease was observed in those with severe OSA (pre-CPAP 68.2 [32.6-86.3] versus CPAP withdrawal 49.0 [36.1-74.4] events/hr). Across all participants, changes in PUP estimates did not exceed test-retest agreement limits. For those with severe OSA, decrease in LG1 (0.86 [0.61-1.13] pre-CPAP versus 0.71 [0.61-0.99] on CPAP withdrawal) and increase in Vpassive (64.8 [5.4-88.4] %Veupnea pre-CPAP versus 76.4 [20.7-92.7] %Veupnea on CPAP withdrawal) exceeded test-retest agreement limits. Increased Vpassive, decreased LG1, and decreased ArTH were predictors of decreased AHI in mixed effects models. Vpassive had the greatest estimated effect on AHI. After accounting for Vpassive, additional estimates did not improve model performance. However, Vpassive and LG1 were correlated, and post hoc analyses suggest these estimates may be influenced by both upper airway collapsibility and ventilatory control. Conclusions: According to PUP physiologic estimates, decreases in AHI following several months of CPAP therapy are primarily attributable to improved upper airway collapsibility.