Clinical prediction rules in obstructive sleep apnoea syndrome.

I have read with interest the paper in which DEEGAN and MCNICHOLAS [1] try to find prediction rules for obstructive sleep apnoea syndrome (OSAS). Yet, there are several questions I would like to pose: 1) The need for clear-cut guidelines for assessment of OSAS patients has been recently stressed [2, 3]. In this sense, my question is: why is an apnoea-hypopnoea index (AHI) of 15 events·hr-1 used to define OSAS, instead of the "classical" 5–10 events·hr-1 [4–6] or the "new" 5 events·hr-1 [2, 7]? It is well known that modifications in the threshold of positivity produce changes in the prevalence of a certain disease and in the performance of diagnostic tests (as can be seen in table 6 [1]). Moreover, I wonder if considering a threshold of 5 or 10 events·hr-1, could have had another consequence: modifying the parameters included in the prediction rules, and their relative weight, so that other prediction equations could have been obtained. I would be pleased if the authors could answer this question. 2) There is another index derived from oximetric recordings that can be of practical interest, the percentage of time during which arterial oxygen saturation (Sa,O2) is <90% (CT90). Perhaps the addition of CT90 to the prediction model could add accuracy. 3) The authors do not state how many of their patients had chronic obstructive pulmonary disease (COPD). Since it has been shown that oximetry is an inaccurate screening procedure for OSAS in COPD [5, 8], one may wonder if the diagnostic performance of the oximetric predictive model is higher for non-COPD patients than for COPD patients. Have the authors analysed this potential difference in behaviour? 4) The multiple linear regression model proposed by the authors yields a determination coefficient of 0.189, or a correlation coefficient (r) of 0.435, which differs significantly from 0. Is this enough? A determination coefficient of 0.189 means that only 18.9% of the total variation in AHI is explained by the three explicative variables (alcohol consumption, BMI and age), and that 81.1% of the total variation remains unexplained. Considering these data and the conditions for applying a regression equation (all the variables involved follow a Gaussian distribution, that r differs from 0, and that r is high enough, i.e. at least 0.5, and preferably at least 0.6 [9]), I cannot accept that AHI can be estimated from these clinical features, and my conclusion would be that further studies are needed. 5) The authors propose that a clinical probability score (CPS) of <0.26 can be considered to exclude OSAS, and a CPS≥0.54 is very helpful in confirming OSAS. In the same way, an oximetric probability score (OPS) of <0.2 excludes OSAS, and OPS ≥0.86 confirms OSAS. So, from these data, it can reasonably be concluded that CPS <0.26 + OPS <0.20 = non-OSAS, and that CPS ≥0.54 + OPS ≥0.86 = OSAS. However, what is the situation for intermediate values of both scores, or those instances in which one of these scores clearly supports or rejects the diagnosis of OSAS and the other yields an intermediate values? It seems sensible to suggest that intermediate values can correspond to hypopnoeic patients (with no apnoeas), upper airways resistance syndrome, or mild OSAS. Nevertheless, there may be instances of apparently doubtful cases that are not included in any of these diagnoses. For instance, can a 31-year-old female, who does not snore every night, has observed apnoeas, does not doze while driving, does not drink and is not obese (CPS=0.361) but who has 129 desaturations ≥4% (desaturation index of 19 hr-1) and a minimum Sa,O2 of 74% (OPS=0.976) be considered as having an a priori nonpredictable overall prepolysomnography study? There can be no doubt that this patient has OSAS. Furthermore, how should we consider a 40 yr old, everyday-snoring, driving-dozing, apnoeic, nondrinking, nonobese male (CPS=0.551) who has 40 desaturations and whose lower Sa,O2 is 88% (OPS=0.468)? Could a joint score (weighted or non-weighted) be useful in these "borderline" situations? Or, should the clinical or the oximetric score be more important in case of disagreement or indetermination? I agree that more studies are needed to assess clinically valid prediction rules, but first we must reach a consensus on basic issues concerning diagnosis and management [2, 3].