An extension of that positive control law based on the total system mass will be developed in order to lead to an individualized drug administration, as an alternative to the well known TCI. Moreover, since in most practical cases, the physical state of the system cannot be determined by direct observation, a state observer will be considered. It will also be analysed the performance of the previous control law for the target control of the total mass, when the unknown state of the system is replaced by its estimate provided by an observer.

The main idea is to explore online learning algorithms for change detection and appropriate adaptation strategies, taking into account intra and inter patient variability. For example, while monitoring Bispectral Index BIS or IOC signals, the learning algorithm FCWM (Fixed Cumulative Windows Model) that monitors the distance between distributions in two time windows can be feasibly adapted to detect changes in the signal. Also cumulative sum procedures may be used to address this problem, namely the Page-Hinckley statistic and the detection system DSKC composed by Kalman filters associated with Cumulative Sum of Recursive Residual (CUSUM).

The purpose of this task is to have consistently built databases with the NMB and DoA records from the EPE/HGSA, HPH and also from Hospital Clinic, Barcelona. The already available database of the team from Hospital Clinic, Barcelona, has about 200 clinical cases collected in endoscopic procedures with sedation-analgesia using TCI systems. This represents a valuable resource, where each clinical record contains a variety of items, such as text with information about the patient clinical history, diagnosis and other clinical comments, and data from propofol infusion, BIS and IOC. Following the same structure, data collected by the closed-loop software Hippocrates at EPE/HGSA and HPH will be added to the initial database. Moreover, DoA records along with drug rates will also constitute a new entries of this database. This will be an ongoing task, running after the 36 months period. The number of cases fully documented can always be increased with time, depending on the possibilities of the EPE/HGSA and HPH clinicians. It turns out that a large number of real reported cases will be available for algorithm development on modeling, identification and control purposes.

The aim of this task consists in the development of algorithms for parameter and in PK/PD models. The backbone of of PK/PD models are compartmental models, together with models that relate the drug concentration in the effect compartment with the variable measuring the effect. As such, PK/PD models have a Wiener type structure, with a linear part corresponding to the PK model and the relation between the plasma drug concentration and the effect concentration, and a nonlinear static part that relates the effect concentration with the effect variable. The estimation methods to use in this task have to comply with a number of constraints: 1) The Wiener structure of the model; 2) Strong parametric uncertainty; 3) Identifiability issues; 4) A relatively reduced number of data points; 5) The need for a fast convergence to an acceptable estimate in order to avoid initial transients that would yield unacceptable control actions. When considering neuromuscular blockade for general anaesthesia, another important constraint is the type of drug administration profile applied in the beginning. This consists of a bolus of drug that forces the neuromuscular blockade level to fall quickly to a near saturated value. Parameter identifiability issues will be the subject of an a priori study of PK/PD models. Parameters that are not identifiable will be fixed at typical values. In order to avoid adaptation transients and given the small numbers of data points, Bayesian estimation methods will be used. These methods rely on a prior parameter density that characterises the patient population and is estimated from previously available data.

The purpose of this task is to develop a software platform suitable for:
Data acquisition, i.e. to collect data about infusion rates and physiological signals related with patient's anaesthesia state for the purpose of anaesthesia model improvement, model identification and validation.
Personalized drug administration, i.e, to use the software platform to implement control algorithms whose descriptions are on previous items, and execute them in a laptop computer, in real time. That is, to automatically adjust the infusion rates of anaesthetic drugs using the information of patient's anaesthesia state. It turns out that GALENO platform must interact with syringe pumps and with DATEX, and must implement software communication functions using the proprietary data communication protocols of the syringe pumps and DATEX. The requirements of the software platform are, to be able to collect data at a 20s period for the neuromuscular blockade monitor and at a 5s period for the DoA signal, and be able to execute the identification strategies, the novel population model proposed and a variety of control algorithms developed within this framework. Nevertheless the requirements imposed by the sampling of the data collection may need a huge processing time that is not compatible with the available computing resources. Hence, a supervisory system should be implemented in order to drive the system throughout its global complexity.

Propofol
We will implement, in our laboratory, an HPLC method with fluorescence detection based on the previously described method to quantify whole blood and plasma propofol concentrations (7). The method will then be validated.

Remifentanil HPLC/UV
We will implement, in our laboratory, an HPLC method with UV detection based on the previously described method to quantify whole blood remifentanil concentrations (7). The method will then be validated.

For the validation of analytical techniques we determine linearity, detection and quantification limit for selected drugs, as well as the precision and accuracy for each of the methods tested by Topic Q2B, Step 4 Note for Guidance on Validation of Analytical Procedures: Methodology (CPMP/ICH/281/95 - adopted December 95).

The experimental results of plasma concentration of propofol and remifentanil and variables collected from the patients enrolled in the study and will be used to establish the population pharmacokinetic-pharmacodynamic (PK-PD) model of drugs.

The clinical validation is to be done in three different settings:
- Operating room: where validation can be performed with respect to neuromuscular blocking (NMB) agents and with respect to deep hypnosis without electromyographic (EMG) contamination because of the effect of NMB agents.
- Sedation area (endoscopy): where validation can be performed with respect to superficial hypnosis (and/or analgesia) but with the contamination of EMG because this patients must be kept in spontaneous ventilation.
- Intensive care unit: where validation can be performed specially with respect to monitoring of hypnosis to control sedation of the patients, and artificial sources of noise can be added or removed to test the system.

Elaboration of a final report on the basis of the written documentation issued during the different tasks of the project. The developed algorithms and methods will be gradually integrated in the system platform and a continuous evaluation under different environments will be performed. The results of this evaluation will report to the necessary methodology improvements and will provide a crucial research guide to the design of the different integrated methodologies.