Electronic Surveillance of Ventilator Settings and Airway Pressures Can Increase the Use of Lung-Protective Ventilation
By David J. Pierson, MD, Editor, Professor, Pulmonary and Critical Care Medicine, Harborview Medical Center, University of Washington, Seattle, is Editor for Critical Care Alert.
Snyopsis: This study from a high-volume ICU setting with a fully integrated electronic medical record system evaluated the impact of automated computer-generated notification to tell respiratory therapists and intensivists when ventilated patients with acute lung injury were receiving potentially injurious tidal volumes and airway pressures. Compared to pre-implementation data, use of the notification algorithm was associated with significant reductions in tidal volumes and airway pressures, as well as the number of hours patients were exposed to these risks for ventilator-induced lung injury.
Source: Herasevich V, et al. Limiting ventilator-induced lung injury through individual electronic medical record surveillance. Crit Care Med 2011;39:34-39.
In this study from the Mayo Clinic in Rochester, MN, Herasevich et al tested an electronic algorithm that incorporated patient characteristics and ventilator data and notified clinicians immediately when potentially injurious ventilator settings were being used. Using data from the institution's electronic medical record (EMR), the "sniffer algorithm" for ventilator-induced lung injury (VILI) risk required satisfaction of the following criteria for patients at least 16 years old who underwent invasive mechanical ventilation for > 24 hours in each of three ICUs:
- Presence of acute lung injury (ALI):
- PaO2/FIO2 < 300 mm Hg, and
- The words "bilateral" and "infiltrates," or the word "edema," on the radiologist's report of the portable chest radiograph
- Use of potentially injurious ventilator settings:
- Plateau pressure > 30 cm H2O or peak inspiratory pressure (PIP) > 35 cm H2O, and
- Set tidal volume > 8 mL/kg predicted body weight (PBW)
At the authors' institution, the EMR acquires all of the above data automatically, within 1 hour of measurement or radiographic exposure. When all the listed criteria were met, a text page was sent to the respiratory therapist assigned to that patient, and also to the critical care fellow on call. The therapist was expected to go to the patient's bedside to assess the validity of the alarm. Subsequent actions, such as discussion of possible ventilator changes among the therapist, fellow, and primary managing physician, were at the discretion of the therapist and fellow receiving the page. Text notification occurred only when the potentially injurious ventilator data thresholds were met for four consecutive 15-minute periods (total 1 hour), and was done only once in any 24-hour period for a maximum of 3 days on any given patient.
The intention was to reduce patient exposure to potentially injurious ventilator settings, as measured by the number of hours that patients continued to have the threshold ventilator-related measurements, as well as to assess clinician responses to the "VILI alert" notification. The electronic surveillance system was already in place prior to the study, so that the authors could compare data on patients ventilated during the preceding 8 months to those collected during the 12 months after the VILI alert system was implemented. In addition, respiratory therapists working in the study ICUs completed a brief on-line satisfaction survey about the system, and incremental costs of implementing the notification system were determined.
Of 9888 potentially eligible patient admissions during the historical control and prospective study periods, 1159 patients were ventilated > 24 hours and 490 cases of ALI were identified using the screening criteria. The study population comprised these 490 patients (42% of the eligible ventilated population), 186 of them from prior to VILI alert implementation and 304 afterwards. Patients in the two groups were well matched by demographics, primary diagnoses, and several measures of illness severity. One hundred eleven text alerts were sent during the study period, of which 65 (in 80 patients) were deemed by the investigators to represent valid VILI risk (positive predictive value of alert, 59%). Of these 65 alerts, all of which included tidal volumes in excess of 8 mL/kg PBW, 12 were from plateau pressures > 30 cm H2O and 53 were from PIPs > 35 cm H2O.
The number of alerts generated per day trended downward during the study period, from a mean of 22 in the first month to six in the final month. Mean patient exposure to potentially injurious mechanical ventilation decreased significantly during the study period, from 40.6 ± 74.6 hours to 26.9 ± 77.3 hours (P < 0.05). There were no changes in ICU or hospital mortality, or in ICU or hospital length of stay. The 27 respiratory therapists who completed the satisfaction survey and had received VILI alerts were asked whether the system was useful; 48% agreed, 33% were neutral, and 19% disagreed. The authors estimated that running the program took about 10 minutes per day on the part of the trained JAVA programmer who implemented it, with a total cost to the institution of $10,400.
This study demonstrates the feasibility and potential effectiveness of fully automated EMR surveillance of mechanically ventilated patients at risk for VILI. As the authors point out, with impending implementation of the Health Information Technology for Economic and Clinical Health Act, as well as current trends toward practice standardization and increased information technology-facilitated clinical decision support, it very likely offers a preview of the future of ICU care.
That future is not yet here, however, and this study was not perfect. Use of design features such as randomization or at least concurrent rather than historical controls would have been better. Using PIP rather than plateau pressure (which accounted for nearly half of the alerts sent) is problematic, as the former may reflect airway phenomena and other factors unrelated to the risk for VILI, and is not used in most assessments of this or in tailoring lung-protective ventilation (LPV). The system would not distinguish between ALI and cardiogenic pulmonary edema, although, as the authors point out, the use of LPV in the latter condition would not be expected to be harmful.
Comment should also be made about the potential generalizability of the study's findings. It was carried out in a large tertiary referral center, with a fully implemented EMR, that manages large numbers of patients with ALI. The system was thus both easier (and cheaper) to implement and more likely to have an important impact on patient outcomes than might be the case in an institution with fewer ventilated patients or one without the technological means for automated detection of the input variables. On the other hand, the fact that LPV was already widely implemented in the authors' institution at the time of the study suggests that the demonstrated benefits could be even greater at other centers where this is not the case.
The ALI diagnostic criteria used in this study were based on the American-European consensus criteria,1 which have stood the test of time and are most widely accepted both conceptually and practically for identifying patients with ALI and the acute respiratory distress syndrome (ARDS). However, only two of the four criteria were used, and only a minimalist version of the radiographic component could be extracted into the EMR. Bilateral pulmonary infiltrates compatible with pulmonary edema remain the least precise and most contentious component of the ALI diagnosis. Rubenfeld and associates showed that even an assemblage of recognized ARDS investigators could not agree on which chest X-rays were consistent with a diagnosis of ARDS in the absence of clinical context,2 so it is hardly surprising that the VILI alerts generated in the present study included numerous false positives.
A page-based system to promote better adherence to LPV as employed in this study is imperfect at best. It is conceptually attractive to imagine closed-loop mechanical ventilation in which LPV could be initiated and maintained automatically using input data similar to those used in this study. This would assure that this life-saving ventilatory strategy was properly and promptly applied to all patients who could benefit from it. However, we are a long way from such a system because of the complexities involved and not least in the determination of which patients are appropriate.
At present it is hard to imagine a non-human system that could identify and account for all the factors that might influence whether and how LPV should be implemented. Such factors could include hemodynamic instability, acute brain injury, non-pulmonary causes for high airway pressures, other explanations for the oxygenation or radiographic findings, patient preferences and code status affecting the aggressiveness of management, and therapy for coexisting conditions, among many others. In the absence of true closed-loop mechanical ventilation, inclusion in the protocol of the bedside assessment and judgment of an experienced clinician once potential VILI risks have been identified as incorporated into this study remains a crucial component of safe and effective mechanical ventilation.
- Bernard GR, et al. The American-European Consensus Conference on ARDS: Definitions, mechanisms, relevant outcomes, and clinical trial coordination. Am J Respir Crit Care Med 1994;149:818-824.
- Rubenfeld GD, et al. Interobserver variability in applying a radiographic definition for ARDS. Chest 1999;116:1347-1353.