To diagnose disease using a continuous metric implies a model whereby there is a bimodal distribution of the metric where disease can be more or less separated from normal by a threshold value. As pointed out previously, data in large populations show that the AHI is rarely 0, even in normal individuals. The association of sufficiently elevated AHI with negative outcomes confirms that some intensity of the sleep-disordered breathing event rate is abnormal. However, technology has led us to refine what events are being counted; thus, little attention has been paid to changing this threshold value. It seems intuitive that the threshold of disease should change as we refine what is counted, but the values of 5 and 15 events/h have been stubbornly present as cut points above which we say sleep apnea syndrome is present. This has led to many of the objections to the value of the AHI (for diagnosis). If the AHI comprises only complete or severe (eg, desaturating) individual obstructions, perhaps a low value can provide a sufficiently specific indicator of disease. In contrast, if all subtle events that produce at least a minimal physiologic consequence in the EEG or sympathetic nervous system are included, a somewhat higher value may increase sensitivity without too much loss of specificity. However, testing and refining these thresholds in clinical samples has been complicated by the extreme nonspecificity of the outcome measures used to define the “gold standard” of significant sleep apnea. Defining an epidemiologically healthy population is critical to deciding on a threshold value for disease, and this varies with the outcome being evaluated (eg, sleepiness, hypertension). Several studies have suggested that the threshold AHI number of events for disease is not low. For example, the Sleep Heart Health Study showed a median AHI, by an inclusive definition, of nearly 30/h in a large nonclinic population, and in the São Paulo cohort, it was reported that even in those with an AHI < 5/h and no daytime symptoms, evidence of upper-airway collapsibility during sleep (flow limitation) was very prominent (the 95 percentile cutoff of flow limitation was 30% of the time asleep). To reconcile these observations, I would argue that the AHI is useful at its extreme values but less so in the midrange. Despite evidence of widespread occurrence of obstructive events in the sleep of otherwise normal subjects, little doubt exists that in individuals with a very-low AHI (by any definition) compared with those with an elevated number (generally > 30/h), excessive sleepiness, hypertension, and other cardiovascular consequences are more frequent in the latter. Furthermore, many studies show improvement in clinical symptoms if the AHI is reduced by treatment in symptomatic subjects, again attesting to the ability of an elevated AHI to define disease. However, rather than trying to find an exact cutoff between normal and diseased, many of the same studies can be reinterpreted as showing that an intermediate AHI (ie, between 5 and 30/h, with some consideration of which AHI) confers no diagnosis but rather a rising probability of disease defined by clinical outcomes. This reasoning is different from but may underlie the common practice of diagnosing sleep apnea based on the AHI alone if > 15/h but requiring symptomatic consequences if between 5 and 15/h. The situation is similar to the use of an elevated BP taken in the clinic to define hypertension. The cutoff for hypertension was initially defined using measurements of BP norms in large, defined, nonclinic populations. These values were validated when it was shown that hypertension (ie, an elevated BP measurement) had consequences for health. We now use intermediate BPs in a different way in individuals with other risk factors (eg, diabetes, age) than in those who appear healthy. In a similar way, a high AHI is clearly a marker of disease. Thus, I conclude that AHI is a useful metric in defining the presence of OSA if severely elevated, and useful to define the risk of OSA if moderately increased.