From a technical perspective, it is relatively easy to program various alerts into electronic anesthesia record-keeping systems. The difficulty is that even “obvious” alerts may actually prove harmful if they distract clinicians without providing useful guidance. Thus, we believe that potential alerts should be formally tested, rather than implemented simply because they seem logical. Our recent report describes the first randomized evaluation of an intraoperative alert.1
In this initial trial, we evaluated an essentially trivial condition, namely a systolic pressure of 80 mm Hg, which no anesthesiologist considers acceptable. Essentially, we were looking for clinicians who were “asleep at the switch” or otherwise distracted. Apparently, neither was an issue at Hillcrest Hospital since the Cleveland Clinic attending physicians and nurse anesthetists took a median of only 1 minute to correct blood pressure with or without alerts.
An obvious conclusion from our results is that decision alerts may be more helpful in more complicated situations. For example, although triple low conditions2,3 can be detected by clinicians, doing so continually would require considerable mental effort that might distract from other tasks. Thus, we are conducting another randomized trial, this one providing alerts for triple low conditions, which we expect to be a better test of decision-support alerts (NCT00998894).
Kappen et al note our conclusion that “decision support may only be useful in more complicated decision situations” and add that “this conclusion presumes that intraoperative hypotension is a simple problem.” We said nothing of the kind! Nowhere in our report1 do we assert or imply that hypotension is simple, solved, or unimportant. All we say is that a simple alert for severe hypotension is unhelpful. In fact, we were the first to report that even mild hypotension is strongly associated with myocardial injury and death,4,5 findings that have since been confirmed.6 We and others are actively working to identify hypotensive thresholds associated with various harms and whether the relationships are causal.
Kappen et al assert that “randomization of patients may have caused study group contamination because of a learning effect” and then suggest that either a cluster-randomized trial or before-and-after design may have been a better choice. A cluster-randomized trial (eg, randomizing entire hospitals to either use or not use the alert) would remove some of the within-provider learning effect because all providers at an institution would either have or not have the alerts for all of their patients. However, cluster design trials are costly and both logistically and statistically complex. Realistically, few investigators would attempt a cluster randomized trial without first doing a single-center study to estimate treatment effect. There is nothing in our results to suggest that a cluster trial would be worthwhile for the alert we tested. The before-after design that Kappen et al recommend is far weaker than our randomized approach, most obviously because it is impossible to make reliable causal inferences when treatment effects are confounded by time.
In contrast to the suggestion of Kappen et al., it seems unlikely that providers would learn instantly from the initial alerts and then treat all future patients—with and without the alert—with the same increased urgency. If there had been a noticeable treatment effect early in the trial that dissipated as providers “learned,” we would have evidence of a systematic “learning effect” over time. For both randomized groups, there was a gradual and small decrease in the mean time to achieve the outcome over time, but the trends were almost identical for the 2 groups. As evidence of this, the estimated treatment effect (ie, effect of alert on time for mean arterial pressure to return to ≥80 mm Hg) did not decrease significantly during the course of our study period (P = 0.59).
In the context of our trial, learning would apply only if the participating attending anesthesiologists and nurse anesthetists needed to be taught that a systolic pressure ≤80 mm Hg was unacceptable. It seems highly unlikely that any were not already aware of this fact—a supposition fully supported by aggressive and effective treatment of hypotension in all patients throughout the trial.
We are grateful to Kappen et al for identifying an error in our article. We specified that our pilot cohort accrued from July to December 2012; that is, in the middle of the actual trial. Of course the sample size was estimated from data obtained before the trial started, specifically between July and December 2011. More importantly, the pilot incidence was based on 3 minutes of hypotension with either direct or oscillometric measurements, whereas alerts in the actual trial were triggered after only a single oscillometric measurement, thus explaining the higher incidence in the trial.
In summary, decision alerts should be formally evaluated just like other diagnostic systems and therapeutic interventions. We present an initial example of a rigorous evaluation that is nonetheless technically simple and inexpensive. The results were clear: an alert for severe hypotension was ineffective because clinicians recognized hypotension on their own and effectively intervened. There is nothing in this result that implies that intraoperative hemodynamic management is simple or solved; it is not and remains an active area of investigation. But it is clear from our results that decision support is more likely to prove beneficial in more complicated clinical situations.
Daniel I. SesslerDepartment of Outcomes ResearchAnesthesiology InstituteCleveland ClinicCleveland, OhioDS@OR.org
Andrea KurzDepartments of Outcomes Researchand General AnesthesiologyAnesthesiology InstituteCleveland ClinicCleveland, Ohio
Edward J. MaschaDepartments of Quantitative Health Sciences and OutcomesLerner Research Institute andAnesthesiology InstituteCleveland ClinicCleveland, Ohio
1. Panjasawatwong K, Sessler DI, Stapelfeldt WH, Mayers DB, Mascha EJ, Yang D, Kurz A. A randomized trial of a supplemental alarm for critically low systolic blood pressure. Anesth Analg 2015;121:15007.
2. Sessler DI, Sigl JC, Kelley SD, Chamoun NG, Manberg PJ, Saager L, Kurz A, Greenwald S. Hospital stay and mortality are increased in patients having a “triple low” of low blood pressure, low bispectral index, and low minimum alveolar concentration of volatile anesthesia. Anesthesiology 2012;116:1195203.
3. Willingham MD, Karren E, Shanks AM, O’Connor MF, Jacobsohn E, Kheterpal S, Avidan MS. Concurrence of intraoperative hypotension, low minimum alveolar concentration, and low bispectral index is associated with postoperative death. Anesthesiology 2015;123:77585.
4. Walsh M, Devereaux PJ, Garg AX, Kurz A, Turan A, Rodseth RN, Cywinski J, Thabane L, Sessler DI. Relationship between intraoperative mean arterial pressure and clinical outcomes after noncardiac surgery: toward an empirical definition of hypotension. Anesthesiology 2013;119:50715.
5. Mascha EJ, Yang D, Weiss S, Sessler DI. Intraoperative mean arterial pressure variability and 30-day mortality in patients having noncardiac surgery. Anesthesiology 2015;123:7991.
6. Monk TG, Bronsert MR, Henderson WG, Mangione MP, Sum-Ping ST, Bentt DR, Nguyen JD, Richman JS, Meguid RA, Hammermeister KE. Association between intraoperative hypotension and hypertension and 30-day postoperative mortality in noncardiac surgery. Anesthesiology 2015;123:30719.