A modular approach for estimating nociception during general anesthesia

Background: Nociception of an anaesthetized patient is difficult to assess, and no numeric methods exist for assessing it during surgery. Since the adequacy of analgesia depends on the interaction of several aspects, a multi-variate approach may be needed. We developed a response index of nociception (RN) to correlate with the clinical assessment of the level of probable nociception at the time of incision during propofol-remifentanil anaesthesia. For the clinical assessment a new index (Stimulus, Analgesia, and clinical Signs Score, SASS) was defined. The SASS provides a combination score for the estimated effect site concentration of delivered analgetic, the stimulus intensity (e.g., size of incision), and the presence or absence of any clinical signs of inadequate analgesia (movements, coughing etc.). Methods: The study was approved by the local ethical committee and written informed consent was obtained from each patient. Electrocardiographic (ECG), photoplethysmographic (PPG) and electroencephalographic (EEG) spectral entropy signals were recorded in 55 females during propofol-remifentanil anaesthesia. Depth of anaesthesia was targeted to maintain the response entropy (RE) value between 35 and 60, and remifentanil plasma concentration levels of 1, 3 or 5 ng/ml were obtained randomly before the moment of the first incision. Two independent clinicians estimated the SASS at the time of incision on the basis of the annotations made by a trained nurse, who was instructed to carefully note any clinical sign showing possibly inadequate analgesia. The SASS was from 0 to 12 corresponding to no nociception vs. extreme nociception, respectively. Physiological signals were collected using a Datex-Ohmeda S/5 Anesthesia Monitor and were stored on a PC. Features representing signal shape descriptors and covering time- and frequency domain information were extracted off-line from these signals at the time of the incision. Absolute as well as relative features were calculated. An algorithm using combinations of signal features was developed to map the monitored signal features to the three different components constituting the SASS (i.e., level of analgesic drug, level of stimulation, and presence of clinical signs). The outputs of these three modules were then combined to provide the RN (i.e., estimation of SASS on the basis of monitored signals). The performance of the RN was assessed by the root mean square error (RMSE) and Pearson's correlation coefficient for the module estimating remifentanil level, and prediction probability (pk) for the components reflecting clinical observations and stimulus intensity. Results: Different features of HRV, response entropy (RE) and PPG variability were combined for different modules: HRV and RE for estimation of remifentanil level and presence of clinical signs, and HRV and PPG variability for estimation of stimulus intensity. For remifentanil the correlation coefficient, r = 0.75 (p = 8e-11) and RMSE 1.1 ng/ml; for clinical observation pk = 0.62 (SE=0.04); and for stimulus intensity pk=0.57 (SE=0.04) were reached. The pk of the eventual SASS estimation is 0.77 (SE=0.03). Discussion: A modular algorithm was developed to estimate the nociception at the time of incision. This modular approach provides a more robust, easy maintainable and tunable, and intuitive implementation as compared to the one in which the final index would be estimated directly in one step.