Completed Project (2014-2016)

Research Team: Nicoletta Nicolaou (Marie Curie Fellow), Lorena Goncalves de A. Freitas, Timothy Constandinou
Collaborators: Geoffrey Lockwood (Hammersmith Hospital), Marcela P. Vizcaychipi (Chelsea and Westminster Hospital), Julius Georgiou (University of Cyprus)
Funding: EU FP7 Programme/Marie Curie Actions (FP7/2007-2013) under REA grant agreement no623767.

Suddenly, I was aware something had gone very wrong. I could hear what was going on around me, and I realised with horror that I had woken up in the middle of the operation, but couldn’t move a muscle. I heard the banal chatter of the surgeons, and I was aware of many people in the room bustling about, doing their everyday clinical jobs and minding their own business, with absolutely no idea of the cataclysmic event that was unfolding from my point of view. While they fiddled, I lay there, frantically trying to decide whether I was about to die, and what options were open to me”

Excerpt from 5th National Audit Project (NAP5) of the Royal College of Anaesthetists and the Association of Anaesthetists of Great Britain and Ireland, 'Accidental Awareness during General Anaesthesia in the United Kingdom and Ireland', Sept. 2014


Modern anesthesia is a chemical cocktail of agents, each one targeting a different integral part of anesthesia (unconsciousness, amnesia, analgesia and immobility). Given that there are no strict dosages for these combinations of agents, the anesthetist follows a set of approximate dosage guidelines and adjusts these based on patient characteristics and past experience. However, administering an incorrect dosage could have serious complications. Overadministration leads to unnecessarily deep anesthesia, with dangerous and potentially lethal consequences (e.g. cardiovascular depression). Under-administration leads to light anesthesia, with increased risk of accidental regaining of awareness during surgery.

The patient testimony above highlights the severity of accidental regaining of awareness during surgery. Even though intra-operative awareness (IOA) does not happen often, the consequences for patients who do experience it during their surgery are traumatizing. Such an experience can lead to long-term psychological problems, including post-traumatic stress disorder and fear of having another surgery. In addition, the majority of surgically-related insurance compensation claims to patients are related to accidental awareness during surgery.

The Association of Anesthetists of Great Britain and Ireland has recently published the findings of a comprehensive study (NAP5) of accidental awareness during general anesthesia in the UK and Ireland [1]. The study reports that approximately 300 patients in the UK and Ireland have experienced awareness during their surgery in one year, resulting to an estimated incidence of awareness at 1 patient in 15,000. Other similar studies in the USA report a higher incidence at approximately 0.1% (e.g. [2]).

Putting this into perspective, considering that an estimated 234.2 million patients undergo surgery with general anesthesia every year worldwide [3], a conservative estimate of 15,613 patients will experience intraoperative awareness. These estimates are likely to be even higher, as some patients who experience awareness may not remember this due to the amnesic effects of the administered agents and, hence, a number of IOA cases remain unreported.

To address the problem of intra-operative awareness, the state-of-the-art has led to the development of commercial devices that monitor the patient level of hypnosis. Their operation is based on information extracted from the patient’s electrical brain activity (electroencephalogram – EEG), as some of the effects of anesthetics cause identifiable changes in EEG activity [4].

The most widely used devices are the BIS® (Aspect Medical Systems, USA) and E-Entropy (GE, Germany). Despite their existence, they failed to become part of routine surgery (reported use in 2.8% of surgeries [1]). This is due to a number of issues related to robustness, intersubject variability and the large delay between the actual time of awareness and the time of change in the monitoring index [5].

An additional factor is cost of initial purchase, as well as long-term operational cost. The biggest cost is associated with the purchase of consumable sensors, whose prices vary more widely. The NHS cost for the BIS® Vista and the BIS® Vista Bilateral monitoring systems comes to £4,350 and £5,025 respectively, while the BIS® Quatro sensors cost £362.50 per box of 25 and the BIS® bilateral sensors cost £210 per box of 10 [6].

As part of a past project at University of Cyprus, we collected EEG activity from 47 patients undergoing routine surgery at Nicosia General Hospital. This activity was analysed with different methods (e.g. permutation and approximate entropy [7], granger causality [8], synchrony methods [9] [10]). Our general findings verified a decrease of EEG complexity during anesthesia and we identified specific changes in brain connectivity and synchrony during anesthetic induced unconsciousness (an example can be seen in figure 1). Even though the majority of features investigated showed high performance in wakefulness / anesthesia discrimination, our interest focused more on granger causality-based features, as these provided high discriminatory ability (98%) while also capturing general mechanisms of anesthetic-induced changes in brain connectivity. More specifically, we identified a stable and reversible increase in granger causality in the frontoposterior direction during anesthesia, independently of the particular anesthetic protocol [8].

Based on our previous findings, “AnaeWARE” (August 2014 – July 2016) challenges the state-of-the-art and addresses the question whether EEG alone is sufficient for reliable monitoring of anesthetic state. A promising direction for a new generation of monitoring devices is the extension from a single- to a multi-modal approach, utilizing additional information that is already available as part of routine monitoring during surgery, e.g. heart rate. This will not impose additional costs to already limited national healthcare funds and could offer a solution for more reliable monitoring.

The AnaeWARE project set out to investigate and characterise the effect of anaesthetics on biological signals monitored during surgery, either routinely (e.g. cardiovascular signals) or as an additional modality (brain activity). The motivation behind this approach is that anaesthetics cause measurable and significant changes to the relationships between these signals, with the main goals of improving patient monitoring and understanding the general mechanisms of anaesthetic action. The project focused on the following two objectives:


Objective 1: To explore the anaesthetic-induced changes in (and between) various biological signals during surgery (multi-modal signals)

The fulfilment of this objective focused on obtaining and analysing relevant data. After obtaining relevant regulatory and ethical approval data collection was initially conducted at a single hospital site, with a second site added at a later stage. Data from 25 patients were collected from both sites, and together with similar data from an additional 40 patients (obtained from a collaborator, Dr. Vizcaychipi) and brain activity data from 20 healthy participants (available online from Chennu et al.: were analysed for the purposes of the project. Preliminary analysis of this multi-modal dataset indicates that anaesthetics induce: (a) significant changes in nonlinear relationships (phase-amplitude coupling) at specific frequency ranges between brain activity and electrocardiogram, electromyogram, heart rate and bispectral index; and (c) significant changes in nonlinear relationships (phase-amplitude coupling) at specific frequency ranges between different brain areas (preliminary results published in Abstract form and presented at the 10th FENS Forum of Neuroscience); and (c) changes in time-frequency measures in cardiovascular activity that are significantly different during wakefulness and anaesthesia. The findings suggest that a better description of the patient state of hypnosis can be obtained from a combined analysis of these signals, which supports the benefits of a multi-modal approach to patient monitoring during surgery. Analysis also inspired the development of a nonlinear, multivariate and nonparametric measure of causality, which can be applied to study the anaesthetic-induced causal interactions in the recorded multi-modal activity. The measure was published in a scientific article (frontiers in Neuroinformatics) and will be utilised in future work emerging from the current project.

Objective 2: To link the observed signals and signal changes with underlying models of anaesthetic action

For the purposes of this objective a model of thalamocortical activity was initially implemented in software (Python) and hardware (as part of a student project supervised by the Fellow). However, the model was inaccurate and there were numerous discrepancies between its theoretical description in a published article and the actual model parameters. As a result, the particular model was abandoned. Nonetheless, important skills were learnt during its implementation, such as programming in Python using the Brian module, and theoretical knowledge on neuronal network interactions. Another model of anaesthetic effects on the brain activity was then investigated. The particular model also describes burst-suppression, which is an endpoint of deep anaesthesia. After personal discussions with one of the model’s co-authors (Prof. Bojak), it was concluded that extending the model to include multi-modal interactions, even though useful and important, this was highly non-trivial and, thus, not feasible within the scope of the project. Instead, the data-driven modelling approach described above (phase-amplitude coupling) was followed. Discussions, however, led towards interesting future investigations in linking actual data with modelled data: validating the burst-suppression model with data recorded during the project, which has not been attempted previously. An additional component of the project was to investigate whether it was feasible to implement some of the analysis methods using off-the-shelf hardware solutions for future prototyping of a monitoring device. This task was undertaken as a final year student project supervised by the Fellow, which included the implementation of causality on a Field Programmable Gate Array, and was successfully completed.


The main project results indicate that anaesthetics cause measurable changes to the communication mechanism between different signals of the body, i.e. they affect the human body as a whole and, as such, a reliable and robust index for monitoring patients during surgery should combine information from the entire body (multi-modal). The project also generated new research directions and ideas, both as a continuation (multi-modal analysis) and as a new direction (e.g. creating models of multi-modal signal interactions for prediction of awareness, hardware implementation of an online multi-modal analysis prototype). The project findings support the development of a new generation of monitoring devices that are more reliable, but also more cost effective (utilisation of information that is already monitored routinely during surgery). Target groups for whom the research could be relevant are anaesthesiologists, surgical patients and biomedical engineers. Through the project the Fellow was able to share ideas and thoughts within a highly multi-disciplinary group of people, had numerous opportunities for integration and knowledge transfer, and was the recipient of unique training courses offered by the host organisation. All these contributed positively towards reinforcing the Fellow’s potential for reaching professional maturity.

The project results were disseminated to the scientific community (journal and conference publications, and organisation/chairing of a special session at IEEE BioCAS 2015) and the public (e.g. demonstration at the Science Museum, public talk at Pint of Science, talk targeting 15-17 year olds at Westminster School, and articles for the Host Organisation’s Annual Research Reports). The project has also led to the development of independent collaboration between the Fellow and Dr. Vizcaychipi (the lead PI of the data recordings at the second site, Chelsea & Westminster Hospital). Common interests in developing improved systems for perioperative patient monitoring has led to fruitful collaboration and a joint application for funding is planned in the near future.




  • M. Ardissino, N. Nicolaou, and M. Vizcaychipi,  "Non-invasive real-time autonomic function characterization during surgery via continuous Poincaré quantification of heart rate variability," Journal of clinical monitoring and computing,  Vol. 33, pp.627-635, 2018. doi: 


  • N. Nicolaou and T. G. Constandinou, “Phase-amplitude coupling during propofol-induced sedation: an exploratory approach,” in FENS Forum of Neuroscience, 2016
  • N. Nicolaou and T. G. Constandinou, “A nonlinear causality estimator based on non-parametric multi- plicative regression,” Frontiers in Neuroinformatics, vol. 10, no. 19, 2016. doi:


  • T. Lauteslager, N. Nicolaou, T. S. Lande, and T. G. Constandinou, “Functional neuroimaging using UWB impulse radar: a feasibility study,” in IEEE Biomedical Circuits and Systems (BioCAS) Confer- ence, pp. 406–409, 2015. doi: 


  • E. Demarchou, N. Nicolaou, T. G. Constandinou, and J. Georgiou, “Anesthetic-induced changes in EEG activity: a graph theoretical approach,” in Proc. IEEE Biomedical Circuits and Systems (BioCAS) Con- ference, 2014. doi: