Table 2.

Identifying High versus Low Subgroups by the Polysomnographic Method

			All Data				Cross-Validation
Trait†	High/Low Definition (Gold Standard Median)‡	PSG Cutoff§	Accuracy (% ± SEM)	Sensitivity, Specificity (%)	Gold Standard Values within PSG Subgroups*		Accuracy (% ± SEM)	Sensitivity, Specificity (%)	Gold Standard Values within PSG Subgroups*
					High	Low			High	Low
Diaphragm EMG
Collapsibility, Vpassive	66.4	86.6	83 ± 7||	93, 73	75 ± 5	41 ± 9¶	72 ± 8¶	71, 73	73 ± 7	49 ± 8**
Collapsibility, Vactive	85.9	81.7	86 ± 6||	100, 73	92 ± 2	45 ± 10||	86 ± 6||	100, 73	92 ± 2	45 ± 10||
Compensation, Vactive − Vpassive	10.9	8.0	69 ± 9**	40, 100	36 ± 9	7 ± 4¶	69 ± 9**	40, 100	36 ± 9	7 ± 4¶
CPAP drops
Collapsibility, Vpassive	7.7	95.5	71 ± 9**	64, 79	29 ± 10	−10 ± 13**	68 ± 9**	64, 71	27 ± 9	−11 ± 14**
Collapsibility, Vactive	61.0	96.5	93 ± 5||	93, 93	108 ± 10	25 ± 12||	93 ± 5||	93, 93	108 ± 10	25 ± 12||
Compensation, Vactive − Vpassive	60.2	0.9	70 ± 9**	85, 57	81 ± 9	25 ± 14¶	52 ± 10	46, 57	75 ± 10	49 ± 14

Definition of abbreviations: CPAP = continuous positive airway pressure; PSG = polysomnographic method; Vactive = ventilation at maximum ventilatory drive without arousal; Vpassive = ventilation at normal ventilatory drive.

^*Gold standard values (mean ± SEM) within high and low polysomnographic subgroups were compared by Student’s t tests. Classification analyses were repeated using leave-one-out cross-validation (right side of table) for a more conservative estimate of accuracy (the selected PSG cutoff was redefined for each subject based on all the other subjects). Note that CPAP drop compensation accuracy was reduced to chance levels with the cross-validation testing (optimal threshold fluctuated between 0.9 and 5.4; both of these thresholds separately yielded 70% accuracy [P < 0.05] using all data).

^†Traits are presented in units of ventilation as a percentage of eupneic levels.

^‡Patients were defined as having a high versus low value of a given trait, based on the median gold standard value (shown).

^§The optimal polysomnographic cutoff used to identify high versus low subgroups was determined by maximizing sensitivity plus specificity (23); this cutoff was used to define the polysomnographic subgroups (left side of table).

^||Accuracy is the sum of correct predictions divided by total number of patients (%): P < 0.05 for accuracy (better than chance, normal approximation method) or high versus low subgroups (Student’s t test).

^¶Accuracy is the sum of correct predictions divided by total number of patients (%): P < 0.01 for accuracy (better than chance, normal approximation method) or high versus low subgroups (Student’s t test).

^**Accuracy is the sum of correct predictions divided by total number of patients (%): P < 0.001 for accuracy (better than chance, normal approximation method) or high versus low subgroups (Student’s t test).