See Article, page 955

This issue of Anesthesia & Analgesia includes a pro–con debate paper1 with persuasive arguments favoring universal continuous postoperative surveillance systems, based on extensive longitudinal experience from the proponent’s group. It also posits why we might adopt a more focused approach by identifying the higher-risk patient populations in the postanesthesia care unit (PACU). The 2 groups offer complementary methods to enhance monitoring efficiencies, and both of their perspectives could benefit from adopting key human factors engineering strategies to further enhance their effectiveness. This safety systems commentary views this current pro–con debate through the lenses of evidence-based medicine as well as failure modes and effects. Enhanced postoperative monitoring—universal or focused—promotes patient safety by reducing or eliminating the failure to recognize state change, which if unchecked, increases the likelihood of adverse outcomes. In addition to this failure mode, the failure to relay information or to escalate care in a timely manner and the failure of timely, appropriate clinical care responses all lead to failure to rescue (FTR). A schematic of the rescue loop structure with failure modes relevant to this commentary is depicted in the Figure.

ALERTS: PRECISION AND ABSOLUTE EFFECT

Standardized monitoring data acquisition and interpretation facilitate the determination of the need to relay to a rapid response team (RRT) or other escalation steps. While automated paging alert notifications will undoubtedly increase efficiencies in recognizing a physiologically relevant state change, the fidelity of this system directly influences the accuracy of notifications. In at least one previous study, 25% of alert notifications were missed, presumably from system errors.2 As described by the proponents of surveillance models, more advanced systems have shown significantly better performance3 but remain reliant on system or local network performance. Nonetheless, both systems are significantly superior to a system that entirely relies on humans for the timely recognition of risk state change.

A highly effective afferent limb system aims to accurately identify deteriorating patients. The desired outcome here is appropriate escalation of care, including RRT activation and intensive care unit (ICU) admission. The majority of alerts are for conditions that can be addressed in under 1 minute, which suggests that these pertain to causes such as pulse oximeter sensor dislodgement, transient airway obstruction, a displaced oxygen mask or nasal cannula, or need for additional oxygen therapy, and they are unlikely to include more intense interventions like the institution of positive airway pressure therapy.2,4

Figure.:

Schematic representation of the rescue arc showing modes contributing to failure to rescue. EWS indicates early warning score; Rx, treatment.

Two metrics can be used to evaluate monitoring interventions in the afferent limb and to better understand human factors involved in FTR. The positive predictive value (PPV) of the alerts reflects the afferent limb’s precision. Adapted from the number needed to screen,5 the number needed-to-monitor (NNM) is calculated as 100 divided by absolute risk change. The NNM can be used to estimate the number of patients who need to be monitored to identify 1 patient who requires a higher (more intense) level of care. The term NNM is used here to differentiate from evaluations of the efferent or rescue limb, where the more traditional number-needed-to-treat (NNT) has been used in this commentary. These are calculated similarly but provide additional semantic clarity when describing the different limbs of the rescue arc (afferent →monitoring versus rescue →treatment).

After the implementation of a surveillance system, a previous report observed 4 daily alerts per patient with an average length of stay of 3.68 days across 2841 patients (9092 patient days) or 36,368 total alerts.3 As 23 patients needed ICU admission, the precision of each alert (PPV) in identifying high-risk states needing escalation of care was 0.03% (23/36,368) or 1 in 4000 alerts. Additionally, over 99% of alerts were resolved without escalating to RRT.4 A similar study of postoperative orthopedic patients on patient-controlled analgesia reported a total of 7 alerts per patient, of which 4 were related to desaturation.2 The PPV of these alerts (710 generated) for the unplanned need for continuous positive airway therapy (1 patient) was 0.14%.

The NNM to identify need for ICU care with surveillance systems can be estimated from published data, where the absolute risk difference for ICU admission dropped from 50 of 2841 patients (1.63%) in the preimplementation cohort to 23 of 3118 patients (0.82%) in the postimplementation cohort, after adjusting for study durations,3 with a resulting NNM of approximately 125.

HUMAN RESPONSE TO ALERT NOTIFICATION

Variability of human responsiveness to alert notifications limits reliable, appropriate, and timely decision-making and rescue intervention. Escalation pathways typically involve the breaching of trigger thresholds, but they are over-reliant on human behavior, with resultant variability in risk assessment and individual likelihood to escalate. Delayed or dropped relays are associated with an increased risk of mortality, which would be defined as FTR. Although continuous monitoring increases the likelihood of timely escalation, it does not guarantee it. Failure to escalate is independently associated with FTR, and in 1 study, it occurred more often among monitored patients.6 Several fixed risk factors, such as emergent admission, not-for-resuscitation orders, advanced age, specific disease conditions like cancer diagnoses, and prolonged hospitalization, were associated with failure to escalate. Taken together with the NNM estimates from earlier sections, variability in human response to alerts has the potential to nullify the described benefits of continuous monitoring systems.

RESCUE EFFECTIVENESS: NNT AND FTR

Two relevant measures of clinical rescue effectiveness are NNT and FTR. A previous publication on a surveillance system reported postoperative mortality rates of 0.13% (4 of 3118) in the preimplementation phase vs 0.07% (2 of 2841) in the postimplementation phase, for an absolute risk difference of 0.06%.3 The NNT thus equates to 1666 (100 ÷ 0.06) to prevent 1 death, when combined with a rigorous process to optimize RRT activation, or 1 to 2 deaths prevented per year in the study population. The FTR rates for postoperative surveillance system were 8% (4 deaths among 50 estimated RRT escalations) in the preimplementation phase vs 8.7% (2 of 23) in the postimplementation phase of the surveillance system. In summary, during the preimplementation phase, mortality was significantly higher, but FTR was marginally lower when compared to the postimplementation phase.

FTR AND COMPLICATION RATES HAVE AN INVERSE RELATIONSHIP

Across Michigan hospitals, the incidence of major complications was systematically higher in hospitals in the low FTR tertile compared to those in the high FTR tertile. Major complications included postoperative infection, respiratory failure, pulmonary embolism, acute renal failure, myocardial infarction, or need for blood transfusion.7 Other studies have confirmed this finding.8 In the Michigan study, postoperative mortality rates across the lowest to highest FTR tertiles were 3.6% vs 4.7%. Low FTR hospitals were more likely to have board-certified intensivists, closed-model ICUs, hospitalists, advanced practice providers, residents, overnight coverage, and increased use of RRT. Based on this study, one can postulate that hospitals that identify complications (higher complication rates) and manage them effectively have excellent rescue rates (low FTR). If all these resources were considered as one bundled intervention, the NNT (100 ÷ ARR) for this rescue bundle to prevent postoperative mortality would be 100 ÷ 1.1 = 91.

It is reasonable to surmise that one of the likely and perhaps, desirable effects of a highly effective afferent limb is to increase the rates of documented postoperative complications, RRT activation, and rescue interventions.9 This paradox introduces vexing challenges for hospitals that are ranked both for postoperative complication rates with publicly reported metrics such as the Center for Medicare & Medicaid (CMS) Patient Safety Indicator – 90,10 its component complications, and FTR.

NEW FRONTIERS FOR MONITORING: RELEVANCE FOR POSTOPERATIVE SLEEP-DISORDERED BREATHING

Globally, the evidence for improved outcomes from enhanced monitoring in hospitals still remains sparse. In a meta-analysis of continuous multi-parameter monitoring, a 39% decrease in risk of mortality was reported (risk ratio [RR], 0.61; 95% confidence interval [95% CI], 0.39–0.95] when compared to patients with regular (routine) intermittent monitoring. However, none of the studies included in this meta-analysis independently reported improvements in mortality. The same study reported no impact on RRT activation, ICU transfer, or hospital length of stay.11 These findings were reproduced in another meta-analysis, which also showed no difference in identification of deterioration, RRT activation, in-hospital cardiac arrest, mortality, ICU utilization, or length of hospital stay.12

It is thus clear that the clinical rescue arc interventions do not all have the same NNT for different drivers of patient deterioration. When the various elements and failure modes of the rescue arc are taken together, the significance of the 2-step rule becomes salient. First, the alert or trigger state should induce a change in treatment, and second, the treatment should change the outcome. A large study of the Nationwide Inpatient Sample reported that patients with a documented diagnosis of sleep-disordered breathing (SDB) had significantly higher rates of postoperative intubation but shorter ICU and hospital stay after intubation.13 General postoperative locations where >65% to 75% of trigger events have a single driver such as airway obstruction,2,4 lend themselves to higher continuous monitoring value, and human responses to alert states are also expected to be more predictable, with little need for variation. In contrast, the same surveillance systems have been less effective in medical wards, probably due to the significantly lower contribution of respiratory causes, which typically drops to 20%, resulting in a significant increase in alarm fatigue.11 Taken together, this may strengthen the hypothesis that SDB as a driver of postoperative complications is more attuned and responsive to monitoring and early intervention than nonairway or nonrespiratory causes.

OPPORTUNITIES FOR INNOVATION

With absolute risk change values for continuous monitoring systems of 0.03% to 0.14% for identification of patients needing escalation of care, and PPV values of 1 in 4000 alerts, it is not surprising that variability in human responsiveness remains a major failure mode for FTR. Based on these data, the case can be made for focused approaches that maximize human responsiveness. There is also an urgent need for such systems to redefine alarm states. One strategy to improve substantially the precision of intelligent systems would to incorporate information compiled over epochs of time and to move away from individual alerts for episodic signal change. Interestingly, the currently highlighted progroup introduced several interventions, including an 80% Spo2 threshold, which reduced the number of alarms by 88%, and the addition of a 15-second delay on top of default delays, which further reduced alerts by 71%.4 With these changes, the average alert load decreased from 83 to 4.2 per patient per 24 hours. Safety as measured by postoperative complication rates were significantly improved, while FTR was largely unchanged.

Alternate approaches to improving the NNM can be considered by extrapolating the value described by the currently highlighted con-group.14 The trigger signature in the con-group’s study was high sleep apnea clinical risk score and evidence of recurrent desaturation events in the PACU. In this study, 21 of 143 (14.6%) patients who had recurrent desaturation events required ICU care—as opposed to 0.7% of patients without recurrent desaturation. With an absolute risk difference of 13.9%, the NNM for this approach would be 7.1, a vast improvement on pure threshold alerts. Extrapolating this, futuristic system alerts for recurrent desaturations (and not every episodic desaturation event), would greatly improve the precision of the alert system. In a similar study of PACU desaturation in a broader study population, including those who underwent sedation and regional anesthesia for procedures, we previously reported high-risk monitoring features, including the need for supplemental oxygen, prolonged desaturation periods, and low nadir Spo2 (<89%).15 A futuristic alert system that triggers based on these high-risk features would require monitoring 45 patients (NNM of 45 = 100 ÷ 2.2) to identify those who need higher level respiratory support: again, a significantly superior precision than threshold alerts alone. Using automated systems to consolidate temporal recurrent episodic changes may help identify high-risk patients for appropriate therapy in a future state.

CONCLUSIONS

From the presented data relating to precision, specifically, the NNM and NNT of monitoring systems, it is clear that there is need for the development of intelligent systems to risk-stratify and focus resources across diverse, complex patient populations. Surveillance systems need to evolve from unidimensional, threshold-based alert models to multifaceted learning systems that integrate temporal data from multiple inputs, including monitor data, baseline risk states, medication administration, and dynamic responses, to name a few. This space is ripe with opportunity for innovation and research. The best is yet to come!

DISCLOSURES

Name: Satya Krishna Ramachandran, MD, MBA.

Contribution: This author helped to conceptualize the article, performed the data synthesis and analysis, authored the manuscript, edited and modified drafts, and read and approved the final manuscript. He is the guarantor of the contents of the article who takes responsibility for all parts of this manuscript from conceptualization to publication.

Conflicts of Interest: S. K. Ramachandran reports receiving honoraria from Fresenius Kabi and grant funding from Fresenius Kabi and CRICO.

This manuscript was handled by: Thomas R. Vetter, MD, MPH, MFA.

REFERENCES 1. Blike GT, McGrath SP, Ochs Kinny MA, Gali B. Pro-con debate: universal vs selective continuous monitoring of postoperative patients. Anesth Analg. 2023;138:955–956. 2. Voepel-Lewis T, Parker ML, Burke CN, et al. Pulse oximetry desaturation alarms on a general postoperative adult unit: a prospective observational study of nurse response time. Int J Nurs Stud. 2013;50:1351–1358. 3. Taenzer AH, Pyke JB, McGrath SP, Blike GT. Impact of pulse oximetry surveillance on rescue events and intensive care unit transfers: a before-and-after concurrence study. Anesthesiology. 2010;112:282–287. 4. McGrath SP, Taenzer AH, Karon N, Blike G. Surveillance monitoring management for general care units: strategy, design, and implementation. Jt Comm J Qual Patient Saf. 2016;42:293–302. 5. Rembold CM. Number needed to screen: development of a statistic for disease screening. BMJ. 1998;317:307–312. 6. Tirkkonen J, Ylä-Mattila J, Olkkola KT, Huhtala H, Tenhunen J, Hoppu S. Factors associated with delayed activation of medical emergency team and excess mortality: an Utstein-style analysis. Resuscitation. 2013;84:173–178. 7. Ward ST, Dimick JB, Zhang W, Campbell DA, Ghaferi AA. Association between hospital staffing models and failure to rescue. Ann Surg. 2019;270:91–94. 8. Burke JR, Downey C, Almoudaris AM. Failure to rescue deteriorating patients: a systematic review of root causes and improvement strategies. J Patient Saf. 2022;18:e140–e155. 9. Bellomo R, Ackerman M, Bailey M, et al.; Vital Signs to Identify, Target, and Assess Level of Care Study (VITAL Care Study) Investigators. A controlled trial of electronic automated advisory vital signs monitoring in general hospital wards. Crit Care Med. 2012;40:2349–2361. 10. CMS Patient Safety Indicators PSI 90 (NQF #0531). National quality strategy domain: patient safety. Accessed November 24, 2023. https://www.cms.gov/priorities/innovation/files/fact-sheet/bpciadvanced-fs-psi90.pdf. 11. Sun L, Joshi M, Khan SN, Ashrafian H, Darzi A. Clinical impact of multi-parameter continuous non-invasive monitoring in hospital wards: a systematic review and meta-analysis. J R Soc Med. 2020;113:217–224. 12. Cardona-Morrell M, Prgomet M, Turner RM, Nicholson M, Hillman K. Effectiveness of continuous or intermittent vital signs monitoring in preventing adverse events on general wards: a systematic review and meta-analysis. Int J Clin Pract. 2016;70:806–824. 13. Mokhlesi B, Hovda MD, Vekhter B, Arora VM, Chung F, Meltzer DO. Sleep-disordered breathing and postoperative outcomes after elective surgery: analysis of the nationwide inpatient sample. Chest. 2013;144:903–914. 14. Gali B, Whalen FX, Schroeder DR, Gay PC, Plevak DJ. Identification of patients at risk for postoperative respiratory complications using a preoperative obstructive sleep apnea screening tool and postanesthesia care assessment. Anesthesiology. 2009;110:869–877. 15. Ramachandran SK, Thompson A, Pandit JJ, Devine S, Shanks AM. Retrospective observational evaluation of postoperative oxygen saturation levels and associated postoperative respiratory complications and hospital resource utilization. PLoS One. 2017;12:e0175408.