Just a Spoonful of Technology Makes the Protocol Go Down

Important therapeutic questions in critical care, including questions about fluid management, discontinuation of mechanical ventilation, management of acute lung injury, and nutritional support, do not involve testing a single drug at a fixed dose against a placebo (1–4). We are interested in evaluating hypotheses such as “routine screening for extubation readiness reduces duration of mechanical ventilation” or “minimizing alveolar distention in acute lung injury reduces mortality.” We do this by transforming these mechanistic principles into testable, reproducible protocols. Protocols do more than make vague recommendations—they provide precise and reliable instructions that are modified over time in response to data measured from the patient.

Unfortunately, the very aspects of protocols that allow them to test complex treatment strategies make them difficult to evaluate in randomized trials. First, investigators must make multiple, often arbitrary, choices in constructing the protocol. Evidence is rarely available to inform each of the decisions necessary to build even the simplest protocol. When bedside clinicians try to implement the results of these trials or when consensus groups try to make recommendations based on them, they are often faced with questions about which aspects of the protocol evaluated in the study can be modified without sacrificing its efficacy. Second, for feasibility reasons, trials can only test a limited number of different protocols. A winning strategy can only be said to be better than its comparator. Therefore, the control arms of trials of protocols are now scrutinized as carefully as the intervention protocol (5). Third, it is rarely possible to blind clinicians in the study to treatment assignment. Therefore, bias from differential use of co-interventions or assessment of outcomes can never be completely excluded. Given these design limitations, it is not surprising that clinical trials of protocols generate a great deal of debate.

This issue of the AJRCCM (pp. 894–900) presents the results of a multicenter, multinational, randomized trial of computer-driven protocolized weaning that reduced median duration of mechanical ventilation by 4.5 d (6). The idea of substituting a computer for a physician in the feedback loop between patient and ventilator is not novel; however, clinical trials demonstrating superiority of this approach are limited (7, 8). This study by Lellouche and colleagues represents the culmination of a research program directed at developing a closed-loop knowledge based system for weaning patients (9). The device incorporates clinical data including end-tidal CO₂, body weight, type of humidification, and a history of the patient's respiratory rate response to varying pressure support levels to gradually lower ventilatory support while keeping patients in a comfort zone.

Control arm aficionados will note that these investigators chose a “usual care” control arm in which individual sites relied on their own written guidelines, principles, and protocols (the authors use these terms almost interchangeably) to discontinue mechanical ventilation. Unfortunately, it is difficult, based on the information provided, to tell exactly how the care differed between the arms. In fact, the computer algorithm and the guidelines at the participating hospitals appear quite similar. Patients in both arms of the trial received nearly identical modes of mechanical ventilation on most days, were supposed to receive at least daily spontaneous breathing trials, and received similar (and quite limited) use of sedation after initiation of the weaning phase. However, we do not know how many successful spontaneous breathing trials lead to extubation or how long it took to act on this information in either arm. We don't really know how carefully the sites followed their own ventilator guidelines or how rapidly pressure support levels were reduced. The challenging part of this study and, in fact, the challenging part of any study with a control arm in which usual care varies considerably among sites and patients is to figure out exactly what care was superseded. We know more about the type of care the patients in the control arm were supposed to have received than the care they actually received.

Should we be surprised that a protocolized approach to discontinuing mechanical ventilation performed by a non-physician (in this case a computer) that incorporates limited sedation, gradual reduction in pressure support that maintains patient comfort, and a prompt to extubate after a successful spontaneous breathing trial reduces duration of mechanical ventilation? Since this is essentially what current guidelines recommend, I suspect few readers will find these results surprising (3). It would appear that the computer was simply more effective at implementing current guidelines than whatever strategies were being used to implement similar strategies at the participating hospitals. Again, unfortunately, we are provided with limited information on whether these hospitals used non-physicians empowered to independently implement their weaning guidelines as recommended by the current literature. I suspect they did not. In their pilot study, the authors suggest that one of the advantages of this device in their hospitals was that it did not require the “presence of a physician at the bedside 24 h a day to apply a protocol” (10). Of course, this ignores the fact that much of the literature on protocolized weaning does not rely on the continuous presence of a physician at the bedside (11). Written guidelines for physicians, as implemented in the study hospitals in the absence of protocolized orders, feedback, or other implementation tools, are known to have little effect on practice (12).

Every year in industrialized nations, hundreds of thousands of people require mechanical ventilation for acute respiratory failure. We have evidence to improve outcomes in these areas with protocols for acute lung injury and weaning, but need strategies to make sure every patient has access to this care. Some physicians may find computer-based protocols more acceptable than sharing responsibility with a respiratory therapist or a nurse (13). Computer-based protocols may be more effective, less expensive, or more reliable than implementation by non-physician clinicians. Hopefully, future studies of this technology will explore these questions with appropriate comparisons and measurements. In the meantime, it is refreshing to see the debate move from whether we should have non-physician–driven protocolized mechanical ventilation to how we might disseminate and implement these protocols.