Evaluation of a Novel Intelligent Interruption Management System in High-Fidelity Simulated Cardiac OR to Improve Surgical Safety

INTRODUCTION

Surgical flow disruptions during complex cardiac surgery procedures may cause preventable errors and patient harm. Prior work has shown that timing interruptions during moments of lower cognitive workload (CogL) can minimize the cost of interruption and risk of inducing primary task errors [1,2]. Heart rate variability (HRV) can serve a reliable proxy for cognitive load (CogL) in surgical contexts [3]. We evaluated a novel intelligent interruption management system (IIMS) which estimates surgeons’ CogL from real-time HRV measurements. We hypothesize that IIMS is feasible in a realistic simulation environment and, if implemented in clinical practice, may decrease preventable errors and improve safety.

MATERIALS AND METHODS

Nine surgical trainees (PGY 2–9) at an academic medical center were recruited to construct a microvascular coronary anastomosis between a silicon based mammary artery model and a left anterior descending coronary artery (see Figure 1). Surgeons were equipped with a Polar H10 heart rate sensor [4] to measure HRV, which was used as a proxy for CogL. Data for one participant was excluded due to equipment malfunctioning.

Figure 1.

Realistic Beating Heart Model (CABG HEARTS#1259; The Chamberlain group, Great Barrington, MA, USA).

Surgeons’ baseline HRV was established immediately before the surgical task during a 5-minute relaxing video depicting nature imagery and sounds. Custom Python software was developed to estimate surgeons’ CogL in near real-time by wirelessly streaming inter-beat interval (IBI; msec) data from the Polar H10 sensor via Bluetooth. HRV was computed as the LF/HF ratio over a 30-second rolling window and CogL terciles were estimated as follows (see Figure 2): “High” CogL was defined as HRV >2 times their baseline, “Medium” CogL as HRV >1.5 but <2 times their baseline, and “Low” CogL as HRV <1.5 times their baseline. Thresholds were defined according to prior work on real-time, HRV-informed cognitive-alert generation [5].

Figure 2.

Simplified Python Code of CogL Estimation Algorithm.

A peak detection algorithm was applied retrospectively to identify peaks (i.e., local maxima) and valleys (i.e., local minima) in surgeons’ CogL during the task. Peaks were defined as an increase in surgeons’ current HRV by more than on standard deviation from their baseline HRV within 5 seconds. Valleys were defined as a decrease in surgeons’ current HRV by more than one standard deviation of their baseline HRV within 5 seconds.

A display positioned behind the surgeon in the simulated OR was dynamically updated to reflect the surgeon’s current CogL (see Figure 3). A confederate observed the display and delivered six scripted interruptions, spaced at least one minute apart, during each simulation session. Surgeons completed a brief survey following the simulation to assess the perceived utility of the IIMS. The study was designed according to best medical simulation scenario practices and performed in a state-of-the-art simulation center. This project was approved by the local IRB and all participants completed an informed consent process.

Figure 3.

Intelligent Interruption Management System.

RESULTS

Results are reported as median (interquartile range [IQR]). The duration of the anastomotic task was 14 minutes (IQR: 13–15 minutes). Surgeons’ HRV increased by 227.3% (p<.05) after starting the task compared to their baseline HRV (see Figure 4), confirming that the task was appropriately challenging.

Figure 4.

Difference in LF/HF ratio between task (Median: 5.0, IQR: 3.6–8.6) and baseline (Median: 2.2, IQR: 1.2–8.3).

Surgeons spent 41% (IQR: 10–72%) of the task duration in high CogL state, 50% (IQR: 21–79%) of the task duration in low CogL, and 9% (IQR: 5–14%) in medium CogL.

Table 1 summarizes the frequency, duration, and temporal spacing of CogL peaks and valleys captured by the IIMS. Peaks occurred more frequently than valleys (26 vs. 14) but were shorter in duration (6.8 sec vs. 38.7 sec). 100% of surgeons reported that they would want interruptions in the OR to be timed based on CogL and that they would like other team members to be aware of their CogL.

Table 1.

Frequency, duration, and temporal spacing of CogL peaks and valleys. Results are reported as median (interquartile range).

	CogL Peaks	CogL Valleys
Frequency	26 (13–56)	14 (9–20)
Duration (in seconds)	6.8 (5.8–7.1)	38.7 (27.3–47.3)
Time Apart (in seconds)	25.5 (16.0–57.8)	56.8 (36.7–96.5)

CONCLUSION AND DISCUSSION

Our novel IIMS captured dynamic changes in surgeons’ CogL during a complex simulated surgical task, allowing proper timing of information exchanges. Surgeons exhibited sustained periods of low CogL with intermittent instances of high CogL occurring across the task duration. However, surgeons spent comparable amounts of time in low and high CogL states overall. These findings highlight the importance of intelligently identifying fluctuations in surgeons’ CogL to avoid interrupting the surgeon during instances of high CogL. Confirmation of these findings in a larger sample size is warranted. This cognitive engineering approach could lead to mitigation of preventable errors and patient harm.