Introduction Sleep Evaluation in Humans Has Been Usually Per- Formed with Polysomnography (psg), a Technique Considered the Gold Standard for Sleep

SLEEP EVALUATION IN HUMANS HAS BEEN USUALLY PERFORMED WITH POLYSOMNOGRAPHY (PSG), A TECHNIQUE CONSIDERED THE GOLD STANDARD FOR SLEEP STUDIES. In fact, PSG has been used in clinical trials in spite of its limitations for longitudinal and more naturalistic studies. Actigraphs, instruments to measure wrist motor activity and also called activity monitors, provide an indirect assessment of sleep through the use of algorithms. These are automatic scoring methods specially developed to distinguish sleep from wakefulness. The use of actigraphy is considerably increasing in clinical studies due to its low cost, and possibilities opened for long-term studies without interfering with the volunteer’s or patient’s routine. Studies employing a variety of algorithms have shown that wrist actigraphy correlates well with PSG recordings, and the agreement rates have been based on an epoch-by-epoch analysis. The algorithm developed by Cole et al1 distinguishes sleep from wakefulness in samples of healthy volunteers and patients with sleep and psychiatric disorders, yielding an 88% agreement rate with PSG. The Sadeh et al2 algorithm also gives high agreement rates with PSG (± 90%) in adults as well as in adolescents. Jean Louis et al3 developed another algorithm, reporting a slightly higher agreement rate (97%) with PSG in healthy volunteers, associated with significant correlation coefficients for some sleep parameters, including a sample of patients with insomnia.4 Despite the fact that agreement rates (epoch-by-epoch) and correlation coefficients (sleep parameters) have been used to compare PSG and actigraphy in many studies,4-11 the use of these techniques as a measure of concordance has not been considered fully appropriate.12,13 First, it has not been taken into account that the testing periods, which are usually conducted during the night in healthy volunteers, are almost exclusively composed by sleep. Consequently, the probability of a good concordance is high. In fact, Sadeh et al6 have shown that the concordance could be as high as 92% by just scoring all epochs as sleep.12 Much more information would be achieved if the principles of sensitivity, specificity, and accuracy were applied.14 Sensitivity would reflect the ability of the actigraphy to detect sleep when the PSG has also scored sleep; specificity, the ability of the actigraphy to detect wakefulness when the PSG did the same; and accuracy, the ability of the actigraphy to detect both sleep and wakefulness compared to PSG. In addition, the use of correlation coefficient has been criticized because it does not measure the concordance between variables but the strength of their relationship.13 In other words, the correlation coefficient measures the degree of association between two quantities. Since it does not inform how closely they agree, a high correlation does not mean a good concordance.13 More specifically, a perfect agreement occurs if the points lie along the line of equality in a plot of one variable against the other, but a perfect correlation can also be achieved if the points lie along any straight line, providing the slope is not zero or infinite.15 In addition, correlation coefficient depends on the range of the true quantity in the sample: if this is wide, the correlation will be greater than if it is narrow.13 Bland and Altman have suggested a procedure to estimate concordance in a sample based on the plot of the difference against the mean of two measures from each subject.13,15 Briefly, the mean difference between actigraphy and PSG measurements for all subjects would reflect the systematic difference between them, and the standard deviation of the mean difference indicates the random fluctuations around this mean. The purpose of this study was to evaluate the concordance between PSG and two actigraphic algorithms (Cole et al, 1992; Sadeh et al, 1994). The actigraphic sensitivity, specificity, and accuracy were calculated from the algorithm, and the concordance among the sleep parameters was estimated with the Bland and Altman technique.