The Trouble with Counting Breaths

AA patient whose breathing stops three hundred times in a night and a patient whose breathing stops thirty times are not, by any ordinary reckoning, suffering the same illness; yet for four decades sleep medicine has described both with a single number. The apnoea-hypopnoea index, or AHI, counts the breathing interruptions captured during an overnight study and divides that total by the hours slept. Thresholds followed: five events an hour marks the beginning of disease, fifteen makes it moderate, thirty severe. The convenience of the arrangement is hard to overstate, since a diagnosis, a payment decision and a research cohort can all be settled by one integer. What the integer discards is almost everything about the events themselves, including how long each lasted, how far the blood oxygen fell, and whether the sleeper woke at all. Two research programmes have spent the past decade arguing about what ought to replace it, and they disagree, at bottom, about what a replacement is for.

BKaisa Fontaine, a respiratory physiologist at the Lindqvist Sleep Laboratory in Uppsala, treats the index less as a convention than as a measurement error. Between 2016 and 2022 her group re-scored the overnight recordings of 1,140 patients, ignoring the count altogether and integrating instead the area beneath each dip in blood oxygen, a quantity her papers name the hypoxic burden. Patients in the highest quarter of hypoxic burden proved 2.4 times as likely to have suffered a cardiovascular event within eight years as those in the lowest quarter, while the AHI, applied to the identical recordings, separated the two groups not at all. Fontaine is careful about what this establishes. The re-scoring predicts an outcome; it does not show that lowering the burden would alter one, and she notes that her cohort was drawn entirely from patients already referred to a specialist clinic, a group whose complaints were troubling enough to reach one. The claim, she insists, concerns a better ruler and not a better treatment.

CHer argument reaches beyond oxygen. Fontaine's laboratory holds that an obstructed airway is produced by four physiological traits acting in combination: a collapsible upper airway, a sluggish muscular response to that collapse, an unstable ventilatory control loop, and an arousal threshold so low that the sleeper wakes before the airway can be reopened. Two patients recording an AHI of twenty-two may owe it to quite different combinations, which is why, on her account, one appliance succeeds in the first and fails in the second. The difficulty is practical. Measuring the four traits has until recently demanded an oesophageal catheter and a full night of laboratory instrumentation, at a cost her own unit puts at eleven times that of a recorder worn at home. Her group has since published algorithms that estimate the traits from ordinary breathing signals, but reports that the estimates agree with catheter measurements on roughly seventy per cent of nights, and she declines to call that sufficient.

DTomás Reiner, a clinical epidemiologist at the Brackwell Institute for Health Evidence, regards the hunt for a finer ruler as premature. His objection begins with repetition. When 213 patients in his 2020 study wore a home recorder on four consecutive nights, an individual's AHI wandered so widely that 41 per cent of them crossed at least one diagnostic threshold during that week, and a fifth crossed two. A quantity as unstable as that, he argues, cannot be refined; it can only be averaged over more nights. His larger claim rests elsewhere. Pooling nine randomised trials in which patients with moderate disease were assigned either to airway pressure or to conservative care, he found no reduction whatever in cardiovascular events, although daytime sleepiness and reported quality of life both improved substantially. Reiner concedes two weaknesses. Adherence in those trials averaged under four hours a night, and trials recruiting the sleepiest patients had been judged unethical to run at all.

ENone of this has stayed inside the journals. Insurance schedules across much of Europe release funding for a pressure device only above an AHI of fifteen, and the licensing authorities of several countries suspend a commercial driving licence at the very same figure, so a number both researchers distrust decides who is treated and who may work. A 2021 review by a national sleep society acknowledged the criticisms in detail and then retained the threshold anyway, on the ground that no alternative had yet been validated in a population sample rather than a clinic. Fontaine replies, awkwardly for her own case, that the objection is a fair one and that her burden index has never been tested outside referred patients. Reiner's complaint runs the other way: regulators quote his null trials as proof that milder disease needs no attention, when his own figures show sleepiness improving. Both dislike one habit above all, namely the treatment of a threshold invented for administrative convenience as though it marked a boundary in biology.

FThe dispute is not about whether the index is crude, which neither of them denies, but about what a replacement is supposed to do. Fontaine's programme asks which physical injury a night of interrupted breathing inflicts, and answers with a quantity that tracks later illness; Reiner's asks which patients are made better by being treated, and answers, so far, the ones who are sleepy. A ruler that predicts harm and a ruler that predicts benefit need not be the same instrument, and no published study has yet set one against the other in the same patients. Until one does, the field will go on producing two literatures that look as though they contradict each other while in fact answering different questions, and the clinics will go on sorting patients by the integer that both sides have spent a decade calling inadequate.