The potential for autonomic neuromodulation to reduce perioperative complications and pain: a systematic review and meta-analysis

Abstract

Background

Autonomic dysfunction promotes organ injury after major surgery through numerous pathological mechanisms. Vagal withdrawal is a key feature of autonomic dysfunction, and it may increase the severity of pain. We systematically evaluated studies that examined whether vagal neuromodulation can reduce perioperative complications and pain.

Methods

Two independent reviewers searched PubMed, EMBASE, and the Cochrane Register of Controlled Clinical Trials for studies of vagal neuromodulation in humans. Risk of bias was assessed; I² index quantified heterogeneity. Primary outcomes were organ dysfunction (assessed by measures of cognition, cardiovascular function, and inflammation) and pain. Secondary outcomes were autonomic measures. Standardised mean difference (SMD) using the inverse variance random-effects model with 95% confidence interval (CI) summarised effect sizes for continuous outcomes.

Results

From 1258 records, 166 full-text articles were retrieved, of which 31 studies involving patients (n=721) or volunteers (n=679) met the inclusion criteria. Six studies involved interventional cardiology or surgical patients. Indirect stimulation modalities (auricular [n=23] or cervical transcutaneous [n=5]) were most common. Vagal neuromodulation reduced pain (n=10 studies; SMD=2.29 [95% CI, 1.08–3.50]; P=0.0002; I²=97%) and inflammation (n=6 studies; SMD=1.31 [0.45–2.18]; P=0.003; I²=91%), and improved cognition (n=11 studies; SMD=1.74 [0.96–2.52]; P<0.0001; I²=94%) and cardiovascular function (n=6 studies; SMD=3.28 [1.96–4.59]; P<0.00001; I²=96%). Five of six studies demonstrated autonomic changes after vagal neuromodulation by measuring heart rate variability, muscle sympathetic nerve activity, or both.

Conclusions

Indirect vagal neuromodulation improves physiological measures associated with limiting organ dysfunction, although studies are of low quality, are susceptible to bias and lack specific focus on perioperative patients.

Editor's key points.

• •Autonomic dysfunction contributes to organ injury in the perioperative period. Peripheral neuromodulation may reverse autonomic dysfunction.
• •This systematic review suggests that peripheral neuromodulation, which may increase vagus nerve activity, is effective in the experimental setting, clinical setting, or both.
• •Peripheral autonomic modulation may offer perioperative benefit by reducing pain and organ injury.

Poor exercise capacity in patients undergoing major noncardiac surgery is associated with a higher rate of complications.¹ Cardiac vagal activity is a direct determinant of exercise capacity.² Patients at higher risk of complications after noncardiac surgery are characterised by reduced vagal tone^3,4 (the major determinant of resting heart rate)⁵ and impaired arterial baroreflex control.^6,7 Major noncardiac surgery, ensuing critical illness, or both lead to further reductions in vagal activity^4,8 and, therefore, autonomic imbalance.

Independent of suboptimal exercise capacity, laboratory and clinical translational data also show that preventing the loss of, or restoring, vagal activity limits systemic inflammation,⁹ myocardial injury,^10,11 atrial and ventricular cardiac arrythmias,^12,13 gastrointestinal complications,^14,15 and lung injury.^16,17 Moreover, either direct vagal nerve stimulation,¹⁸ or stimulation of vagal afferents,¹⁹ reduces pain in experimental models. As perioperative morbidity typically involves several organs,²⁰ the restoration (or prevention of the loss of) vagal activity represents a viable therapeutic target to improve clinical outcomes after major surgery. Device-based vagal neuromodulation has gained traction as a therapeutic option in heart failure, where autonomic dysfunction is a key pathological feature.²¹ However, direct electrical stimulation of the vagus nerve is costly, invasive, imprecise, and impractical to meet the likely number of high-risk surgical patients who may benefit from redressing autonomic imbalance. Alternative, cheaper noninvasive methods of indirect (transcutaneous cervical or auricular) vagal neuromodulation have also been developed.²²

We hypothesise that autonomic vagal neuromodulation may improve pain and preserve function in organs that frequently sustain injury after noncardiac surgery. To devise a perioperative therapeutic strategy, it is essential to understand whether similar neuromodulation approaches, settings, or both may be beneficial to preserve function of various organs subjected to experimental or clinical stressors. Therefore, we systematically reviewed the literature in order to assess whether autonomic neuromodulation may play a role in attenuating end-organ dysfunction relevant to perioperative medicine and pain.

Methods

Protocol and registration

We registered this systematic review prospectively with PROSPERO (CRD42020216516). The review was performed in adherence to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist.²³ Ethical approval was not required.

Inclusion criteria

Original research articles were considered if they met the following inclusion criteria: adults ≥18 yr; RCTs, or clinical trials in which each individual acted as their own control and the timing of the intervention was not randomised; neuromodulation intervention targeting the vagus nerve; primary endpoint of organ-specific function including (but not limited to) biomarkers of inflammation, cardiovascular, gastrointestinal, and/or cognitive function or pain (as defined by experimental pain thresholds, pain symptoms, or both).

Exclusion criteria

We excluded studies where: ex vivo readouts, non-human subjects, or both were used; interventional techniques were used that did not specifically target vagus nerve activity (e.g. acupuncture); stimulation parameters for both intervention and control groups were insufficiently described; all outcome measurements were recorded more than 7 days after the intervention (to ascertain whether any effects occurred within the acute postoperative time frame).

Information sources and search

We searched PubMed, EMBASE, and the Cochrane Central Register of Controlled Clinical Trials (from the inception of each database until October 31, 2020) for relevant publications reporting data on vagal neuromodulation with the assessment of end-organ injury relevant to perioperative medicine and pain. Only RCTs (or clinical trials in which each individual acted as their own control and the timing of the intervention was not randomised) involving human participants were selected. Results were combined by the Boolean operator ‘AND’ or ‘OR’ with search terms. The PubMed search strategy is provided below as an example:

• 1.‘vagus nerve stimulation’ AND ‘inflammation’
• 2.‘vagus nerve stimulation’ AND ‘pain’
• 3.‘vagus nerve stimulation’ AND (‘myocardial ischemia’ OR ‘myocardial infarction’ OR ‘atrial fibrillation’)
• 4.‘vagus nerve stimulation’ AND ‘gastrointestinal function’
• 5.‘vagus nerve stimulation’ AND ‘cognition’

The searches were conducted independently by two authors (ABUP and VW), and reviewed by another co-author (GLA) for consistency. Differences between the reviewers were addressed through re-examination of the original sources until consensus was reached. No search filters or language restrictions were applied. We extracted records to EndNote (Thomson Reuters, Philadelphia, PA, USA) to sort and remove duplicates.

Study selection

Studies were selected for inclusion by two authors (ABUP, VW) acting independently. After merging the search results and removing duplicates, we screened the titles and abstracts. Full articles that met the inclusion criteria were retrieved. References of selected articles and published systematic reviews were also searched to identify any further relevant articles meeting the inclusion criteria. The authors of relevant papers were contacted for missing information where possible. When there was uncertainty regarding eligibility, a third reviewer was consulted (GLA).

Data collection process and data items

Data extracted for comparison from the included studies were tabulated by two independent reviewers (ABUP, VW) detailing: primary author, year of publication, study design, number of participating patients or volunteers, type of the intervention and control, stimulation parameters, outcomes, and time of intervention and outcome measurement. Means with standard deviation were extracted for continuous outcomes and numbers of events were extracted for dichotomous outcomes.

Primary outcome

The primary outcome consisted of study-specific readouts of organ function relevant to the perioperative period (cognitive function, inflammation, cardiovascular, lung, and gastrointestinal) and pain.

Secondary outcomes

The secondary outcomes were estimates of vagal activity and adverse effects of autonomic neuromodulation technique used.

Risk of bias

Risk of bias was assessed using the Cochrane Risk of Bias Tool for RCTs.²⁴ Risk of bias was assessed under the following six domains: selection, performance, detection, attrition, reporting, and other. Two review authors (ABUP, VW) independently assessed the risk of bias. When a consensus could not be reached through discussion, a third reviewer was consulted (GLA).

Statistical analysis

The meta-analysis was conducted using Review Manager software (RevMan; Computer programme, Version 5.3; The Nordic Cochrane Centre, Copenhagen, Denmark; The Cochrane Collaboration, 2014). Data entry was carried out by two investigators (ABUP and VW) acting independently. For primary outcomes, an inverse variance/random effects model, with the standardised mean difference (SMD) was used for analyses as clinical and methodological heterogeneity was expected. SMD reflects the size of the intervention effect of each study and allows for group comparisons independently of specific outcome measures. A value of zero indicates no effect.²⁵ We defined the pre-specified threshold for statistical significance as P<0.05. Between-study heterogeneity was assessed using the I² statistic test using P<0.1 as the pre-defined threshold for statistical significance. We defined significant heterogeneity by I² ≥50%. Results are presented as SMD with 95% confidence interval (CI), associated P-values, and forest plots. Potential publication bias was assessed with visual assessment of funnel plots for each meta-analysis outcome.

Results

Study selection

We initially identified 1258 publications; after removal of duplicates, 951 publications were screened with 166 publications eligible for full-text review, leaving 31 publications for analysis (Fig. 1). The majority (28/31) of studies used either transcutaneous auricular (n=23) or transcutaneous cervical (n=5) vagal nerve stimulation (VNS) (Fig. 2a). One study involved transvenous VNS at spinal level C5–C7, where the internal jugular vein runs adjacent to the vagus nerve.²⁶ None of the included studies appear to have been supported by the VNS device manufacturers. Six studies were conducted in patients undergoing surgical or interventional cardiology procedures. For VNS stimulation settings, pulse width differed between study types (Fig. 2b) but pulse frequency was similar across studies (Fig. 2c).

Fig 1.

PRISMA flow diagram of literature search results. Thirty-one RCTs were included for meta-analysis. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Fig 2.

Study design and quality. (a) Modes of stimulation, as indicated by red shading. Transcutaneous auricular vagus nerve stimulation is highlighted within the ear (which the majority of studies utilise). (b) Pulse width settings. Data are presented as mean (standard deviation [sd]) and analysed using one-way analysis of variance (anova). P-values refer to post-hoc Tukey Kramer tests conducted to determine between factor differences. Pulse width (μs) settings were different between cardiovascular and cognition studies, cognition and inflammation studies, and inflammation and pain studies. (c) Pulse frequency settings. Pulse frequency (Hz) settings were similar between all cardiovascular, cognition, inflammation and pain studies (P>0.05). (d) Cumulative risks of bias (by subtype): assessment of each risk of bias item shown as percentages for all included studies. (e) Cumulative risks of bias (by study): assessment of each risk of bias item for all included studies. Green – low risk; yellow – insufficient data; red – high risk.

Risk of bias within and across studies

The risk of selection bias was low to moderate, with details on random sequence generation and allocation concealment lacking in numerous trials (Fig. 2d and e). The risk of performance and detection bias was moderate to high, as several trials did not report if and how study participants and investigators were blinded to the intervention and outcome measurements. The risk of reporting and other types of bias was low for most studies. The majority of studies (26/31; 84%) used sham controls.

Primary outcome

Pain

Ten studies explored the impact of vagal neuromodulation (Table 1) on pain intensity,27, 28, 29, 30, 31 subjective pain scores,³² and/or evoked potentials and pain thresholds.^31,33, 34, 35, 36 Sites of stimulation were either the tragus or cymba conchae for studies using transauricular VNS (Fig. 2a). The majority (seven of 10) reported that VNS increased experimental pain thresholds (healthy volunteers) or improved pain symptoms (patients) (SMD=2.29 [95% CI, 1.08–3.50]; P=0.0002; I²=97%; Fig. 3a). Two studies found that transauricular VNS at the cymba conchae reduced pain induced by heat and pressure.^28,36 A high level of publication bias was likely, as indicated by asymmetrical funnel plot shape for the primary outcome measures (Fig. 3b).

Table 1.

Clinical trials investigating the effects of vagal neuromodulation on pain and cognition. – indicates value unspecified; n refers to intervention/control. AF, atrial fibrillation; CABG, coronary artery bypass grafting; LVEDV, left ventricular end diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end systolic volume; PCI, percutaneous coronary intervention; PMID, PubMed Identification Number; STEMI, ST elevation myocardial infarction; taVNS, transcutaneous auricular vagus nerve stimulation; tcVNS, transcutaneous cervical vagus nerve stimulation; VNS, vagus nerve stimulation.

Authors, ref. no., year (PMID)	Study design	Subjects (n)	Age (yr)	Sex (% female)	Intervention	Start of intervention	Stimulation parameters					Control	Outcome measured	Outcome time point
							On/off period (s)	Duration (min)	Pulse width (μs)	Pulse frequency (Hz)	Intensity (mA)
Pain
Lerman and colleagues27 2019 (30759089)	RCT	Volunteer (15/15)	18–54	36.7	tcVNS	9.5 min before pain challenge	120	–	1000	25	24 V	Sham	Numerical Pain Rating Scale (0–10); time to peak response from each noxious thermal stimulus (s)	16.8 min
Tassorelli and colleagues32 2018 (29907608)	RCT	Migraine (122/126)	18–75	25	tcVNS	20 min before pain assessment	120	–	200	25	60	Sham	Pain-free without using rescue medication (% of subjects)	30, 60, 120 min
Janner and colleagues28 2018 (29337730)	RCT, crossover	Volunteer (49/group)	22–28	51	taVNS	20 min before pain challenge	–	–	200	100	–	Sham and no (separate control arms)	Numeric pain scale (0–100)	48 h
De Icco and colleagues33 2018 (29154689)	RCT, crossover	Volunteer (10/group)	24–29	50	tcVNS	30 and 5 min before assessment	120	–	200	25	0–60	Sham	Nociceptive withdrawal reflex threshold (mA)	30 min
Frøkjaer and colleagues34 2016 (26728182)	RCT, crossover	Volunteer (18/group)	40–57	44.4	taVNS	5 min before assessment	–	20	250	30	0.1–10	Sham	Thresholds to bone pain (kPa)	0, 144 h
Laqua and colleagues35 2014 (25135040)	Non-RCT, crossover	Volunteer (22/group)	22–28	–	taVNS	15 min before assessment	–	30	200	2–100	–	No	Pain threshold (mA)	35 min
Busch and colleagues36 2013 (22621941)	RCT, crossover	Volunteer (48/group)	–	–	taVNS	48 h before assessment	–	60	250	25	0.25–10	No	Submodalities of pain thresholds (heat, pressure) (rating, kPa)	60 min
Napadow and colleagues29 2012 (22568773)	RCT, crossover	Chronic pelvic pain (14/group)	21–64	100	taVNS	15 min before assessment	–	–	450	2	–	No	Visual Analog Scale (0–100)	0, 15 min
Kutlu and colleagues30 2020 (32190684)	RCT	Fibromyalgia (30/30)	18–50	100	taVNS	Unclear	–	30	500	10	–	No	Visual Analog Scale (0–100)	168 h
Alt and colleagues31 2020 (32592516)	RCT, crossover	Volunteer (23/group)	18–45	60.9	tcVNS	60, 30, 15, and 5 min before assessment	120	10	200	25	0–24 V	Sham	Nociceptive flexor and pain thresholds (mA); Numerical Pain Rating Scale (0–10); Somatosensory evoked potentials (μV)	5, 15, 30, 60 min
Cognition
Bretherton and colleagues37 2019 (31358702)	Non-RCT, crossover	Volunteer (14/group)	68–71	35.7	taVNS	N/A	–	15	200	30	2–4	Sham	Mood disturbance (Profile of Mood States (POMS) score, no units)	168 h
Burger and colleagues,38 2019 (30710565)	RCT	Volunteer (48/49)	18–25	80.4	taVNS	20 min before cognitive tests	30	15	250	25	0.5	Sham	Worry (time spent worrying during 5 min period, %)	15 min
Fischer and colleagues,39 2018 (29693214)	RCT, crossover	Volunteer (21/group)	19–22	85.7	taVNS	28 min before cognitive tests	–	–	200–300	25	0.4–3.3	Sham	Conflict-triggered adjustment of cognitive control (N2 and P3 event-related potential amplitude, μV)	–
Colzato and colleagues,40 2018 (29326067)	RCT	Volunteer (40/40)	17–33	62.5	taVNS	15 min before cognitive tests	30	–	200–300	25	0.5	Sham	Creativity performance (Alternate Uses Task, Idea Selection Task, Remote Associates Test and Creative Problem Solving Task scores, no units)	25 min
Colzato and colleagues41 II, 2018 (29128975)	RCT	Volunteer (40/40)	17–33	62.5	taVNS	15 min before cognitive tests	30	–	200–300	25	0.5	Sham	Creativity performance (Alternate Uses Task, Idea Selection Task, Remote Associates Test and Creative Problem Solving Task scores, no units)	25 min
Sellaro and colleagues,42 2015 (26226074)	RCT	Volunteer (20/20)	18–30	87.5	taVNS	15 min before cognitive tests	30	–	200–300	25	0.5	Sham	Post-error slowing (Response times, ms; Accuracy, %)	–
Jacobs and colleagues,43 2015 (25805212)	RCT, crossover	Volunteer ‘older’ (30/group)	58–63	50	taVNS	90 min before cognitive tests	–	17	200	8	5	Sham	Associative memory (Response and reaction times, ms)	90 min
Wittbrodt and colleagues,44 2020 (32659483)	RCT	Volunteer without PTSD (11/8)	18–65	47.4	tcVNS	5 min before neuroimaging	120	–	40	25	0–60	Sham	Emotional stress/fear memory (Brain activity by: size, number of voxels; site, Brodmann area/Talairach coordinates)	5 h
Borges and colleagues,45 2020 (32523510)	RCT, crossover	Volunteer (32/group)	19–27	28.1	taVNS	4 min before cognitive tests	30	13	200–300	25	2.2	Sham	Core executive functions (Reaction times, ms; Error, %)	24, 48 h
Warren and colleagues,46 2020 (32630571)	Non-RCT, crossover	Volunteer (42/group)	18–25	81.0	taVNS	At least 15 min before cognitive tests	30	–	200–300	25	0.5	Sham	Attention-triggered processes (P3 event-related potential amplitude, μV)	–
Steenbergen and colleagues,47 2015 (25869158)	RCT, crossover	Volunteer (30/group)	18–27	86.7	taVNS	20 min before cognitive tests	30	55	200–300	25	0.5	Sham	Action cascading (Reaction times, ms; Error rates, %)	25 min

Fig 3.

Effects of VNS on organ function and pain. Pain: (a) Forest plot analysis showed VNS reduced pain compared with the control group (SMD=2.29 [95% CI, 1.08–3.50]; P=0.0002; I²=97%). (b) Funnel plot analysis showed asymmetrical shape, suggestive of a high level of publication bias. Cognition: (c) Forest plot analysis showed VNS improved cognitive function compared with the control group (SMD=1.74 [95% CI, 0.96–2.52]; P<0.0001; I²=94%). (d) Funnel plot analysis showed asymmetrical shape, suggestive of a high level of publication bias. Inflammation: (e) Forest plot analysis showed VNS reduced inflammatory markers compared with the control group (SMD=1.31 [95% CI, 0.45–2.18]; P=0.003; I²=91%). (f) Funnel plot analysis showed symmetrical shape, suggestive of a low level of publication bias. Cardiovascular: (g) Forest plot analysis showed VNS improved cardiovascular function compared with the control group (SMD=3.28 [95% CI, 1.96–4.59]; P<0.00001; I²=96%). (h) Funnel plot analysis showed asymmetrical shape, suggestive of a high level of publication bias. CI, confidence interval; sd, standard deviation; VNS, vagus nerve stimulation; SMD, standardised mean difference. The difference in mean values is attributable to heterogeneity of units.

Cognition

Eleven studies reported that cognitive function was improved by mostly transauricular VNS (sites of stimulation were either the tragus or cymba conchae)37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47 (Table 1) compared with sham stimulation (SMD=1.74 [0.96–2.52]; P<0.0001; I²=94%; Fig. 3c). Studies in volunteers were not age-matched and did not use comparable models. The majority of studies used similar stimulation parameters for transauricular VNS. A high level of publication bias was likely, as indicated by asymmetrical funnel plot shape for the primary outcome measures (Fig. 3d).

Systemic inflammation

Five of six studies using similar stimulation parameters reported a reduction in circulating cytokine levels after variable periods of VNS (Table 2)48, 49, 50, 51, 52 compared with sham stimulation (SMD=1.31 [0.45–2.18]; P=0.003; I²=91%; Fig. 3e). In a randomised double-blind sham-controlled trial of experimental human endotoxaemia, transvenous VNS failed to reduce symptoms, cytokine levels, or measures of the innate immune response.²⁶ A low level of publication bias was likely, as indicated by symmetrical funnel plot shape for the primary outcome measures (Fig. 3f).

Table 2.

Clinical trials investigating the effects of vagal neuromodulation on inflammation. – indicates value unspecified; n refers to intervention/control. AF, atrial fibrillation; CABG, coronary artery bypass grafting; LVEDV, left ventricular end diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end systolic volume; PCI, percutaneous coronary intervention; PMID, PubMed Identification Number; STEMI, ST elevation myocardial infarction; taVNS, transcutaneous auricular vagus nerve stimulation; tcVNS, transcutaneous cervical vagus nerve stimulation; VNS, vagus nerve stimulation.

Authors, ref. no., year (PMID)	Study design	Subjects (n)	Age (yr)	Sex (% female)	Intervention	Start of intervention	Stimulation parameters					Control	Outcome measured	Outcome time point
							On/off period (s)	Duration (min)	Pulse width (μs)	Pulse frequency (Hz)	Intensity (mA)
Stakenborg and colleagues,48 2019 (30472681)	RCT	Surgical (10/10)	60–68	70	Abdominal VNS	16 and 2 h preoperatively	–	20	1000	20	2.5	Sham or no (combined control arm)	IL-6 and IL-8 (pg ml−1)	24, 48 h
Stavrakis and colleagues,49 2017 (29759717)	RCT	Surgical (26/28)	43–72	29.6	Transvenous VNS	On arrival in ICU (time point unspecified)	–	4320	1000	20	–	Sham	TNF-α, IL-6 and IL-10 (pg ml−1), CRP (mg L−1)	24, 72 h
Kox and colleagues,26 2015 (26049730)	RCT	Volunteer (10/10)	23–27	0	Transvenous VNS	10 min before intravenous administration of E. coli LPS	–	30	1000	20	2–10 V	Sham	TNF-α, IL-6, IL-8 and IL-10 (pg ml−1)	48 h
Stavrakis and colleagues,50 2015 (25744003)	RCT	Paroxysmal AF needing ablation (20/20)	53–69	35	taVNS	Pre-induction of general anaesthesia (time point unspecified)	–	60	1000	20	–	Sham	TNF-α, IL-6 and IL-10 (pg ml−1), CRP (mg L−1)	20 min
Salama and colleagues,51 2020 (32358638)	RCT	Surgical (50/50)	54–72	50	taVNS	24 h preoperatively	–	40	200	–	0.23	Sham	TNF-α, IL-6, IL-1β and IL-10 (pg ml−1), CRP (mg L−1)	24, 96 h
Yu and colleagues,52 2017 (28797427)	RCT	STEMI needing PCI (47/48)	18–80	25.3	taVNS	35 min before reperfusion in catheterisation laboratory	5	155	1000	20	–	Sham	IL-6 and TNF-α (ng L−1), IL-1β (pg L−1), high-mobility group-box 1 protein (ng ml−1)	72 h

Cardiovascular

Six studies reported that autonomic neuromodulation by VNS reduces cardiovascular morbidity, mortality, or both (Table 3), compared with the control group (SMD=3.28 [1.96–4.59]; P<0.00001; I²=96%; Fig. 3g). Three studies reported that VNS improved echocardiographic indices, biomarkers for severity of heart failure, or both.52, 53, 54 Two studies found that VNS suppressed atrial fibrillation in individuals undergoing cardiac surgery and those with paroxysmal atrial fibrillation.^49,50 Auricular VNS also reduced levels of biomarkers for myocardial ischaemia, improved left ventricular contractility, and reduced reperfusion-related arrhythmias in two studies conducted in patients undergoing percutaneous coronary intervention for acute myocardial infarction⁵² and bypass graft surgery for coronary artery disease.⁵³ A high level of publication bias was likely, as indicated by an asymmetrical funnel plot shape for the primary outcome measures (Fig. 3h).

Table 3.

Clinical trials investigating the effects of vagal neuromodulation on cardiovascular measures. – indicates value unspecified; n refers to intervention/control. AF, atrial fibrillation; CABG, coronary artery bypass grafting; LVEDV, left ventricular end diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end systolic volume; PCI, percutaneous coronary intervention; PMID, PubMed Identification Number; STEMI, ST elevation myocardial infarction; taVNS, transcutaneous auricular vagus nerve stimulation; tcVNS, transcutaneous cervical vagus nerve stimulation; VNS, vagus nerve stimulation.

Authors, ref. no., year (PMID)	Study design	Subjects (n)	Age (yr)	Sex (% female)	Intervention	Start of intervention	Stimulation parameters					Control	Outcome measured	Outcome time point
							On/off period (s)	Duration (min)	Pulse width (μs)	Pulse frequency (Hz)	Intensity (mA)
Stavrakis and colleagues,49 2017 (29759717)	RCT	Surgical (26/28)	43–72	29.6	Transvenous VNS	On arrival in ICU (time point unspecified)	–	4320	1000	20	–	Sham	Postoperative AF occurrence (% of patients)	24, 72 h
Yu and colleagues,52 2017 (28797427)	RCT	STEMI needing PCI (47/48)	18–80	25.3	taVNS	35 min before reperfusion in catheterisation laboratory	5	155	1000	20	–	Sham	Reperfusion-related arrhythmia (number of events); creatine kinase-MB (ng m−1), myoglobin (μg L−1), N-terminal pro-B-type natriuretic peptide (ng L−1); echocardiographic parameters including LVEF (%), LVEDV and LVESV (ml)	72 h
Stavrakis and colleagues,50 2015 (25744003)	RCT	Paroxysmal AF needing ablation (20/20)	53–69	35	taVNS	Pre-induction of general anaesthesia (time point unspecified)	–	60	1000	20	–	Sham	AF duration (min)	20 min
Zamotrinsky and colleagues,53 2001 (11474540)	Non-RCT	Coronary artery disease needing CABG (8/10)	48–58	0	taVNS	Not clear	–	15	–	–	–	None	Cardiac noradrenergic nerve density (volume % in atrial tissue); LVEF (%)	–
Tran and colleagues,54 2019 (30560316)	RCT, crossover	Heart failure (24/group)	57–80	46	taVNS	40 min before assessment	10–20	60	200	20	–	Sham	LV global longitudinal strain (% reduction on echocardiography)	60 min
Clancy,92 2014 (25164906)	RCT	Volunteer (34/14)	20–62	50	taVNS	N/A	–	15	200	30	10–50	Sham	Muscle sympathetic nerve activity (μV)	–

Other organ systems

Three of four (75%) studies reported that VNS improved gastrointestinal motility^34,48,55 compared with sham stimulation (Supplementary Table S1). A single study found that transauricular VNS reduced the incidence of postoperative pneumonia and pro-inflammatory cytokine levels.⁵¹

Secondary outcome: estimates of autonomic modulation

Only six of 31 studies determined whether VNS altered autonomic function, of which five reported higher vagal activity, reductions of sympathetic activity, or both, as reflected by heart rate variability analysis (Supplementary Table S2), muscle sympathetic nerve activity, and inhibition of noradrenaline release.⁵³

Adverse effects of vagal neuromodulation techniques

One study reported more burning/stinging sensation with VNS than with sham stimulation.³⁸ Common adverse effects were otherwise not reported although there were no safety concerns noted with any intervention. Only seven of 31 studies explicitly reported on safety, tolerability, and adverse effects.

Discussion

Our prospectively registered systematic review suggests that autonomic neuromodulation aimed at increasing or preserving vagal activity in non-surgical settings has the potential to reduce organ dysfunction relevant to perioperative medicine and pain. The majority of studies used transauricular VNS. Few studies met the minimum reporting standards for transcutaneous VNS-based research, as recommended by an international consensus-based review,²² which include detailing the precise site for auricular stimulation. We noted that VNS applied 24 h before surgical or interventional cardiology procedures reduced inflammation, organ injury, or both, although only one study was conducted in noncardiac surgical patients. Although only ∼20% of studies directly examined the autonomic effects of stimulation, these findings suggest that transauricular VNS offers therapeutic potential for common complications that occur during the perioperative period.

We specifically looked at proof-of-concept studies in patients and human volunteers that inform whether the intervention may be useful in the perioperative setting. Pain, cognitive dysfunction (in part caused by injury to brain tissue), and dysfunction of other organs cluster together after major surgery.⁵⁶ Many of the volunteer studies involved individuals who were administered painful stimuli^{27,28,31,35,36} or found to have underlying subclinical neuropsychiatric states (e.g. anxiety, depression, confusion, post-traumatic stress, and memory disorders)^37,38,43,44 before VNS was administered. Although these subjects have been selected because they are deemed to be healthy (and often young) without any morbidity, they still experience pain and cognitive challenges before surgery. Indeed, preoperative anxiety and pain are clinically relevant modifiers of surgical stress and inflammation.⁵⁷ Proof-of-concept data from the volunteer studies are useful for the assessment of the potential for vagal neuromodulation to reduce organ dysfunction relevant to perioperative medicine and pain. By analogy, a novel analgesic would not be introduced into the perioperative setting without similar proof-of-concept studies in healthy volunteers.

Similarly, interventions aimed at preserving cognition would require proof-of-concept studies in human volunteers before moving into the perioperative arena. Several routine perioperative interventions can affect cognition in healthy subjects. For example, an experimental study conducted in young healthy volunteers undergoing brachial plexus blockade assessed cognitive performance. Change in cognition before and after brachial plexus blockade were assessed using a left/right hand task, involving motor imagery processes and perception illusions pertaining to hand posture. Participants performed less quickly and accurately on the task during regional anaesthesia. These findings suggested that brachial plexus anaesthesia disrupts cognition relevant to the perioperative period.⁵⁸ General anaesthetic agents⁵⁹ and opioids also reduce memory, cognitive performance, or both, as assessed by performing executive function tasks.⁶⁰ Therefore, demonstration of efficacy of vagal neuromodulation in healthy younger subjects is relevant to perioperative care, even though they are not likely to be at risk of serious cognitive impairment. However, our review of healthy volunteer studies in this context should be treated with caution, as these studies measured surrogate measures of cognitive dysfunction.

Non-electrical, non-pharmacological techniques such as sham feeding with chewing gum also preserve vagal activity, augment vagal activity, or both, and reduce gastrointestinal dysfunction,^61,62 although many operations preclude this approach. Nevertheless, these physiological interventional trials support the paradigm that vagal tone may be modulated to improve gastrointestinal motility disorders and reduce pain. Although very few studies have attempted to directly modulate autonomic function in the perioperative setting, acupuncture has demonstrated a potential perioperative role.63, 64, 65 Acupuncture reduces the consumption of anaesthetics and analgesics.⁶³ Perioperative acupuncture reduced perioperative opioid consumption, in concert with a 30–50% reduction in plasma cortisol and epinephrine within 24 h of surgery.⁶⁶ Transcutaneous electrical nerve stimulation also reduces postoperative incisional site pain,^67,68 targeting the ilioinguinal, iliohypogastric, and/or genitofemoral nerves. In the context of our findings, auricular acupuncture also appears to be effective for the treatment of preoperative anxiety. Aside from the need for advanced and specific skills, the exact mechanism of action, neurophysiologic target, and efficacy of acupuncture remain to be established. The neurophysiology underlying acupuncture is highly relevant to VNS. Acupuncture reduces splenic and serum inflammatory responses in experimentally induced acute inflammation.⁶⁹ Vagotomy and splenic neurectomy reverses these anti-inflammatory effects, suggesting that acupuncture stimulation may confer its therapeutic benefits via vagal neuromodulation of inflammatory responses observed in the spleen, and potentially other organs.⁶⁹ A meta-analysis of 17 RCTs has demonstrated that 75% of localised auricular acupuncture points, targeted during pain therapy, are found in regions predominantly innervated by the auricular branch of the vagus nerve (ABVN).⁷⁰ These clinical findings are consistent with the paradigm that stimulation of the ABVN may be a key analgesic mechanism of auricular acupuncture.⁷⁰

Patients who develop complications after noncardiac surgery rarely experience single organ dysfunction.²⁰ Even apparently relatively minor complications reduce life expectancy, a finding that has been repeatedly reported across different healthcare systems for the past three decades.71, 72, 73 Mechanistically, the clustering of complications is plausibly driven by the failure of inter-organ crosstalk, which is required for effective resolution of inflammation and minimising organ injury. The clustering of pain with other postoperative complications is common and may directly promote or prolong other complications, or both.⁷⁴ The majority of VNS studies assessing pain and pain threshold outcomes involved the administration of painful stimuli, bearing clear relevance to the surgical period. However, an increase in a pain threshold (although indicative of an analgesic effect) does not necessarily translate to organ protection, and it may increase the risk of organ dysfunction through numerous mechanisms.

Translational and clinical studies have demonstrated that maintained efferent vagal activity, augmented efferent vagal activity, or both, reduces the pathophysiological effects of renal,⁷⁵ neurological,⁷⁶ and ventilator-induced lung injury¹⁷ and provides cardioprotection.^10,11 Aside from direct organ innervation, experimental data also show that augmenting vagal activity limits systemic inflammation through innate and adaptive immune mechanisms. VNS reduces haemorrhage in experimental surgical tissue injury by increasing coagulation factor activity.⁷⁷

A substantial number of noncardiac surgical patients at higher risk of complications demonstrate vagal autonomic dysfunction before surgery, which is strongly linked to reduced exercise capacity.² Mechanistically, inhibition of vagal neurones in the dorsal vagal motor nucleus reduce exercise capacity by 80% in rats.⁵ Conversely, activating the same neurones optogenetically enhances exercise capacity to the same degree as treadmill training.⁵ Thus, vagal dysfunction is likely to be a key feature of high-risk, deconditioned surgical patients. These observations are supported by the high prevalence of preoperative baroreflex dysfunction in surgical patients, which is associated with impaired cardiac function, and an inability to respond to goal-directed therapy.⁷ After surgery, prolonged bed rest impairs baroreflex sensitivity,^78,79 which is linked to increased risk of postoperative cardiac and infectious complications,⁸⁰ and mortality.⁸¹

Postoperative delirium and cognitive dysfunction (POCD) occur commonly after noncardiac surgery and are associated with prolonged hospitalisation and higher mortality.⁸² Neuroinflammation contributes to POCD.⁸³ Transcutaneous cervical VNS reduced inflammation generated by microglia in a murine model of Alzheimer's disease.⁸⁴ A similar effect on hippocampal inflammation has been demonstrated in a model of exploratory laparotomy in aged mice.⁸⁵ The positive impact of VNS in numerous cognitive studies in humans suggests that this approach may also have perioperative utility in reducing POCD.

The drawbacks of the volunteer studies that assess pain and cognition are that they do not involve patients with organ dysfunction. However, as considered before, these studies do provide proof-of-concept data. Pain and cognition are domains associated with worse postoperative outcomes; about 12% of individuals with no evidence of preoperative cognitive dysfunction will develop symptoms of POCD after anaesthesia and noncardiac surgery,⁸⁶ and postoperative pain is poorly controlled in many apparently previously well patients.⁸⁷ Our review is also limited by studies that are not age-matched, do not use comparable models, and demonstrate significant heterogeneity. The studies also vary widely in terms of the time points at which VNS was applied, with variable outcomes assessed. There is a lack of data on either optimal or dose response, or both.⁸⁸ One systematic investigation of three combinations of different pulse width and frequency demonstrated that a pulse width of 500 μs and frequency of 10 Hz produced the greatest reduction in heart rate.⁸⁹ Although no eligible non-English full-text articles were found, the exclusion of case reports and observational cohort and case-control studies are also potential limitations. Although funnel plot analyses provided evidence of publication bias, inclusion of less than 10 studies in individual funnel plots renders real asymmetry difficult to distinguish from chance.⁹⁰

Lack of blinding is a further potential limitation, although sham stimulation served as an effective control for three reasons. First, in most studies, investigators measuring outcomes were masked to treatment allocations. Second, many studies were designed so that the stimulation settings were not observable or detectable to participants; investigators increased the current intensity until it was detectable, before reducing the intensity to just below this tactile threshold (for active stimulation) or off (for sham stimulation). Placebo interventions are effective, particularly within the context of pain.⁹¹ Therefore, for the included studies in which the stimulation settings were neither observable nor detectable to subjects, any difference detected between sham and stimulation intervention groups is likely attributable to VNS. However, a minority of studies were designed such that subjects could distinguish between sham and active stimulation. Third, ∼50% of the trials were crossover studies, in which each subject acted as their own control for both sham and stimulation interventions.

The cumulative results report an improvement of ∼2–3 SMD in treatment effect, supporting the use of VNS in perioperative patients. However, the majority of studies are small (<30 subjects), and many studies report inadequate or poor evidence of allocation concealment, blinding of participants, investigators, and outcome assessments, or both. This increases susceptibility to high levels of bias, a point emphasised by the asymmetrical funnel plots. Therefore, the poor quality of studies necessitates larger studies with better design and control, before it can be concluded that VNS should universally be used perioperatively.

In summary, current studies are of low quality, are susceptible to publication bias, and utilise surrogate measures with an implied link to actual perioperative organ dysfunction. However, several proof-of-concept studies in humans suggest that autonomic neuromodulation strategies aimed at maintaining or augmenting vagal activity may be utilised to reduce organ dysfunction and pain during the perioperative period, but definitive studies are lacking.

Authors' contributions

Development of study protocol: ABUP

Systematic search: ABUP, VW

Data extraction and analysis: ABUP, VW

Drafting of the manuscript: ABUP

Supervision of review from development of protocol through to write up stage: GLA, AVG

Declarations of interest

GLA is an editor of the British Journal of Anaesthesia and is a consultant for GlaxoSmithKline, unrelated to this work. AVG is a founder and Chief Scientific Officer for Afferent Medical Solutions. The other authors declare no competing financial interests.

Funding

British Heart Foundation (RG/19/5/34463 to GLA and AVG) and a National Institute for Health Research (NIHR) Advanced Fellowship (NIHR 300097 to GLA).

Handling editor: Jonathan Hardman

Appendix A. Supplementary data

The following is the Supplementary data to this article: