| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
Albuquerque, NM
Correspondence to: Lee K. Brown MD, FCCP, Professor of Medicine and Pediatrics and Vice Chair, Department of Internal Medicine, University of New Mexico School of Medicine, UNMH Sleep Disorders Center, 1101 Medical Arts Ave NE, Building 2, Albuquerque, NM 87102; e-mail: lkbrown{at}alum.mit.edu
"On two occasions I have been asked [by members of Parliament], ‘Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?’ I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question."
Charles Babbage
Medical technology has become increasingly complex and difficult for nonengineers to understand, while at the same time the exigencies of the marketplace have made medical equipment manufacturers reluctant to divulge detailed descriptions of the technology used in their devices. In some cases, this is of minimal or no importance; for instance, the mechanical details of how a bronchoscope tip is deflected are not critical when the clinician can directly observe the results. However, medical devices that collect, filter, and analyze data, providing only the processed version to the clinician, or those that interpose their own judgment in diagnosing a condition or treating a patient must be subject to much more investigation before we can fully trust their functionality. In truth, we can rely on most automated technology only under certain conditions, and these limitations should be fully vetted before the devices are made available to clinicians.
We only need to recall the disastrous case of a computer-controlled radiotherapy device some years ago to realize the potential hazards involved in automated technology that is incompletely tested.12 In that case, a sequence of control commands not anticipated by the designer prevented the deployment of a target that should have been interposed between a high-energy radiation beam and the patient in order to generate a therapeutic-level secondary emission. With the target missing, patients were subjected to radiation levels far above therapeutic, unfortunately with dire consequences. Closer to home for pulmonary and sleep physicians, it has been well-documented that the brand and model of pulse oximeter influences whether a given respiratory event qualifies as a hypopnea under Medicare guidelines.3 Variations in response time and other computational details inherent in each device alter the lowest value of saturation that will be displayed subsequent to a respiratory event; this can be of critical importance when the event definition requires a 
4% desaturation.
Autotitrating continuous positive airway pressure (CPAP) devices are classic examples of the "black box," defined in this usage as "any small... box containing a secret, mysterious, or complex mechanical or electronic device."4 In essence, a black box is only known in terms of its output for any particular input; how that output is determined remains hidden. Under US Food and Drug Administration regulations, marketing approval for an autotitrating CPAP device seems to require only that clinical studies demonstrate equivalent ability, in a defined patient group, to suppress sleep-disordered breathing events in comparison with a previously approved apparatus. In most cases, detailed descriptions of the current algorithms that determine the response of the machine to changes in upper airway mechanics and airflow are not explicitly disseminated, and when requested are said to be proprietary. Those descriptions that have been published5678 are associated with early iterations of each apparatus (now > 10 years old) and cannot with certainty be applied to the commercially available versions.
In the absence of detailed information on the algorithms used by these devices to control therapy, they are fair game for studies that evaluate their operation under simulated or real-world conditions. Thus, the two reports on performance testing of autotitrating CPAP machines that appear in this issue of CHEST (see pages 343 and 350) are a welcome addition to the literature. Rigau and colleagues9 modified their previously described lung simulator, which reproduced real flow and snoring signals, to also vary airway resistance consistent with the flow signal. All events could include simulated mask leaks, and both open-loop and closed-loop testing (the simulated breathing characteristic changing in appropriate ways with variations in CPAP pressure) could be programmed. Three of the tested generators measured airway resistance by forced oscillation, while seven machines relied only on snoring and airflow.
During open-loop testing, the measured performance differed between machines in important ways: With respect to obstructive apneas, they found variations in the rate at which pressure was increased, the amount of the change per interval, and the maximal pressure achieved. Simulated central apneas caused the devices without forced oscillation technology to increase pressure as if they were sensing obstructive apneas. Obstructive hypopneas without snoring resulted in some machines not changing pressure at all; for obstructive hypopneas with snoring, one apparatus still did not increase pressure, while some generators increased pressure more rapidly than without snoring. When tested with flow limitation alone, some machines did not increase applied airway pressure, including one with forced oscillation and several without forced oscillation. Simulating an isolated mouth leak identified four devices that increased pressure as if detecting an obstructive event, and only two machines appropriately responded to a replicated "mask leak" on top of obstructive apneas. Finally, in the closed-loop test only three generators (one with forced oscillation) proved to be capable of inducing normal breathing in the simulated patient.
Lofaso and coworkers10 also used a lung simulator to test five different autotitrating CPAP generators. Unlike the apparatus used by Rigau et al,9 this simulator was not designed to replicate changes in airway resistance, mask leaks, or mouth leaks, and only open-loop behavior was tested. They used a variety of synthesized as well as patient-derived flow waveforms that were meant to simulate common degrees of inspiratory flow limitation. None of the devices detected all of the flow-limited waveforms, and many exhibited different sensitivity to the flattening of inspiratory flow depending on the level of CPAP being delivered. Some machines displayed seemingly aberrant behavior, cycling between extremes of pressure. The authors concluded that manufacturers should disseminate the technical details and algorithms employed by their devices so that physicians could select the appropriate machine according to the needs of each individual patient.
Since this is not currently the case, what is the clinician to do, and more importantly, how should we respond to this situation as a community of sleep medicine physicians? The clinician can certainly react by carefully following the symptomatic response of patients started on autotitrating CPAP, assuming that a complaint is present in the first place. Drawbacks to this approach include a noticeable placebo effect that has been demonstrated in randomized controlled trials of CPAP,11 as well as the frequent lack of a strong correlation between the severity of sleep apnea and any given symptom.12 At the other end of the spectrum would be the empiricist’s approach of testing the patient in the sleep laboratory during the application of the proposed autotitrating device. This is a technique that I have indeed applied on occasion in patients with complex combinations of sleep-disordered breathing events, and it at least provides the comfort of knowing whether the patient is responding adequately even if one does not exactly know why. An intermediate approach might be to depend on the quantification of respiratory events by the device itself, which is a form of circular reasoning that does not seem very appealing. Rigau and colleagues9 suggested that a consensus be reached among manufacturers to standardize the signals measured and the algorithms used, a strategy that I am afraid would stifle further research and development and is probably unworkable, given the competitive nature of the industry.9 I would rather propose a fifth strategy, consisting of the independent, standardized, verifiable testing of each apparatus using techniques such as those reported by Rigau et al9 and Lofaso et al.10 Such an approach would be equivalent to the testing of appliances or automobiles by some not-for-profit consumer organizations, and could be performed by one or another of our professional societies or even by collaboration among them. With this information, the clinician could make an informed choice about using a specific autotitrating CPAP generator in each individual clinical situation. Can there be any doubt that our patients would benefit from opening up the black box to the light of day?
Footnotes
Dr. Brown has received grant support from the ResMed Foundation and ResMed, Inc.
References
This article has been cited by other articles:
|
|
 |
|
  L. K. Brown Filling in the Gaps: The Role of Noninvasive Adaptive Servoventilation for Heart Failure-Related Central Sleep Apnea Chest, July 1, 2008; 134(1): 4 - 7. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
