**2. Decision support systems for anesthesia in the operating room**

easily understood, the rules process should be intuitive and open for collaboration, all deci‐

DSSs in medicine could play a role in every field: a modern DSS is conceived to predict rehabil‐ itation protocol for patients with knee osteoarthritis [7]. Another example of a modern DSS is a system that uses anthropometric information and questionnaire data to predict obstructive sleep apnea [8]. The use of DSSs has been proposed to treat major depression [9]; a DSS has been validated recently to diagnose the common flu [10]; a DSS has been developed to support the treatment of epilepsy [11]. Another DSS has been presented in the field of gynecology [12].

At present, it is not clear if an improvement of medical performance can always be transfer‐ red into an improvement of patient outcomes [13, 14] [15], and although better adherence to guidelines is proven, this cannot always be translated into abandoning habits of wrong-do‐ ing [16]. Furthermore, there are some considerable barriers to the widespread diffusion of

These systems are usually produced with limited private funds; mass production is limited by economic pressures. Lack of standardization often represents a "political" problem. There are always emotional barriers for physicians and other health care providers to 'rely'

these systems, like costs, cultural issues and lack of standards [2] [17] [18].

on the help of devices in order to make proper decision.

sions should be reproducible and the user interface easy to use (Figure 1) [6].

**Figure 1.** Graphical user interface [6].

18 Decision Support Systems

Anesthesiologists in the operating room have to provide direct patient care. Anesthesiolo‐ gists are considered the "pilots of human biosphere" [21], and terms like "takeoff" and "landing" for the process of inducing anesthesia and reversing it, are very common; since these are the two dominant and critical moments of anesthesia, often, maintenance of anes‐ thesia receives less attention [22]. To assure safe and good patient care during the surgical procedure, an anesthesiologist interacts with several devices: he becomes "the mediator be‐ tween patient and machine while the machine is mediating between patient and anesthesiol‐ ogist; all are hybrids in action and each is unable to act independently" [22]. It is impossible to consider the anesthetic work without machines just as it is impossible to imagine a pilot without his joysticks, buttons and computers.

Decision support systems for anesthesia in the milieu of the operating room are software shaped to assist the anesthesiologist in his difficult work during the surgical procedure. Let's divide DSSs for anesthesia in the operating room into three classes: DSSs designed for perioperative use, DSSs for one single intraoperative problem (*simple DSSs*) and DSSs for multiple problems (*complex DSSs*).

#### **2.1. Organizational DSSs and implementation in AIMS in the perioperative context**

In his everyday activity, the anesthesiologist deals not only with patient-related issues, but al‐ so with many kinds of organizational problems, like strictly hierarchical command structures or deficits in providing important drugs or devices that can cause serious accidents. Reason [23] has proposed a scheme of the development of an organizational accident (Figure 2).

It is not possible to consider the anesthesiologist's responsibility only during the surgical in‐ tervention; as a pilot has to control his systems before the flight, anesthesiologists must con‐ tinuously assess the patient status, from pre-operative assessment till post-operative care. As a 'commander-in-chief', he has to make the final check of everything 'anesthetic' in the operating room, despite the presence of nurses or respiratory technicians. One type of DSS can deal with organizational problems in order to prevent accidents.

The first example of how DSSs may improve safety in the operating environment is a DSS whhi generates dynamically configured checklists for intraoperative problems [24]. It is interesting that the database built with 600 entries of two anesthesia textbooks and organ‐ ized in problems and corresponding abnormalities, considers also technical hitches, like e.g. inefficiency in anesthesia machines or incorrect position of an endotracheal tube. For each abnormality detected by monitors and confirmed by the practitioner, the software formulates a list of questions, starting with a recognized "high-impact abnormality" (ev‐ ery abnormality uniquely associated with a problem); questions about the "high-impact abnormality" are presented to users as closed-type questions, i.e. they can be answered as "yes" or "no", to facilitate a quick response.

sic anthropometrical features (Figure 3) to predict easy (Cormack I and II) or difficult intu‐ bation (Cormack III and IV). The system showed an average classification accuracy of 90%.

Decision Support Systems in Medicine - Anesthesia, Critical Care and Intensive Care Medicine

http://dx.doi.org/10.5772/51756

21

**Figure 3.** The 13 variables for Cormack classification and their encoding schemes. BMI, for body mass index; TMD, ty‐ ro-mental distance; EAJ, atlanto-axial joint; IIG, interincisor gap; MMT, modified Mallampati test. Binary values (0, 1) were used for variables with only two attributes. Values as 0, 0.5 and 1 were used for variables with three attributes.

Anesthesia information management systems (AIMS) can reduce the anesthesiologist's workload. Implementation of DSSs in AIMS represents a natural evolution of information technology: DSSs can use data stored in AIMS to give diagnostic or therapeutic messages.

Values as 0, 0.33, 0.67, 1, were used for variables with four attributes [30].

This development increases the usefulness of both systems [31, 32].

**Figure 4.** Percentage of patients involved in prophylaxis. \*Statistically significant difference [33].

A recent example of how a DSS combined with an AIMS can improve performance and outcomes is shown in a study about automated reminders for prophylaxis of postopera‐

**Figure 2.** The development of an organizational accident [23].

Preoperative tests are crucial for the stratification of the anesthetic risk, for the choice of the anesthesia technique but also to define the anesthesiologist's behavior. A Canadian group [25] found that the mean cost of investigations was reduced from \$124 to \$73 if data for pa‐ tients were assessed by staff anesthesiologists. Another study [26] demonstrated that, fol‐ lowing definite preoperative diagnostic guidelines, possible savings per 1000 patients would be €26287 and €1076 if duplicated tests were avoided.

A DSS for this purpose, the System for Pre-Operative Test Selection (SPOTS), has been de‐ veloped to assist physicians in selecting the right preoperative, individualized and clini‐ cally relevant tests [27]. The software uses a database comprising of patient data, clinical history, a list of surgical procedures, standard guidelines for preoperative investigations, type and cost of investigations, and investigation results: the DSS then suggests the tests and performs a cost comparison.

Airway management represents one of the most important challenges for the anesthesiolo‐ gist. The main causes of anesthesia-related mortality are respiratory and cardiocirculatory events [28, 29]. One of the most important aims of preoperative assessment is predicting a difficult intubation; it means to timely prepare airway devices to facilitate a possibly diffi‐ cult procedure. Currently, the gold standard for the evaluation of the difficulty of intubation is the Cormack and Lahane classification, but it's feasible only through direct laryngoscopy. A DSS for estimating the Cormack classification was presented in 2009 [30]; it was based on data of 264 medical records from patients suffering from a variety of diseases. It used 13 ba‐ sic anthropometrical features (Figure 3) to predict easy (Cormack I and II) or difficult intu‐ bation (Cormack III and IV). The system showed an average classification accuracy of 90%.

ized in problems and corresponding abnormalities, considers also technical hitches, like e.g. inefficiency in anesthesia machines or incorrect position of an endotracheal tube. For each abnormality detected by monitors and confirmed by the practitioner, the software formulates a list of questions, starting with a recognized "high-impact abnormality" (ev‐ ery abnormality uniquely associated with a problem); questions about the "high-impact abnormality" are presented to users as closed-type questions, i.e. they can be answered as

Preoperative tests are crucial for the stratification of the anesthetic risk, for the choice of the anesthesia technique but also to define the anesthesiologist's behavior. A Canadian group [25] found that the mean cost of investigations was reduced from \$124 to \$73 if data for pa‐ tients were assessed by staff anesthesiologists. Another study [26] demonstrated that, fol‐ lowing definite preoperative diagnostic guidelines, possible savings per 1000 patients would

A DSS for this purpose, the System for Pre-Operative Test Selection (SPOTS), has been de‐ veloped to assist physicians in selecting the right preoperative, individualized and clini‐ cally relevant tests [27]. The software uses a database comprising of patient data, clinical history, a list of surgical procedures, standard guidelines for preoperative investigations, type and cost of investigations, and investigation results: the DSS then suggests the tests

Airway management represents one of the most important challenges for the anesthesiolo‐ gist. The main causes of anesthesia-related mortality are respiratory and cardiocirculatory events [28, 29]. One of the most important aims of preoperative assessment is predicting a difficult intubation; it means to timely prepare airway devices to facilitate a possibly diffi‐ cult procedure. Currently, the gold standard for the evaluation of the difficulty of intubation is the Cormack and Lahane classification, but it's feasible only through direct laryngoscopy. A DSS for estimating the Cormack classification was presented in 2009 [30]; it was based on data of 264 medical records from patients suffering from a variety of diseases. It used 13 ba‐

"yes" or "no", to facilitate a quick response.

20 Decision Support Systems

**Figure 2.** The development of an organizational accident [23].

be €26287 and €1076 if duplicated tests were avoided.

and performs a cost comparison.

**Figure 3.** The 13 variables for Cormack classification and their encoding schemes. BMI, for body mass index; TMD, ty‐ ro-mental distance; EAJ, atlanto-axial joint; IIG, interincisor gap; MMT, modified Mallampati test. Binary values (0, 1) were used for variables with only two attributes. Values as 0, 0.5 and 1 were used for variables with three attributes. Values as 0, 0.33, 0.67, 1, were used for variables with four attributes [30].

Anesthesia information management systems (AIMS) can reduce the anesthesiologist's workload. Implementation of DSSs in AIMS represents a natural evolution of information technology: DSSs can use data stored in AIMS to give diagnostic or therapeutic messages. This development increases the usefulness of both systems [31, 32].

**Figure 4.** Percentage of patients involved in prophylaxis. \*Statistically significant difference [33].

A recent example of how a DSS combined with an AIMS can improve performance and outcomes is shown in a study about automated reminders for prophylaxis of postopera‐ tive nausea and vomiting (PONV) [33]. A database was implemented with PONV prophy‐ laxis guidelines. The comparison of two groups (one with only AIMS and the other with AIMS and also DSS), found that automated reminders were more effective for adherence to PONV prophylaxis (Figure 4).

study, authors found that 70% of all surgical patients received their antibiotics within 60 min

Decision Support Systems in Medicine - Anesthesia, Critical Care and Intensive Care Medicine

http://dx.doi.org/10.5772/51756

23

of incision (Figure 6); after one year, the adherence increased to about 92%.

**Figure 6.** Administration of antibiotic: gradually increasing to about 92% [40].

**Figure 7.** Anesthesia information management system screen overlaid by SAM screen [41].

It also showed a reduction of inappropriate administration of PONV prophylaxis medica‐ tion to low-risk patients: automated reminders not only are effective in promoting correct actions, but may also prevent unnecessary prescription of medication, hence reducing drug costs. Although the DSS significantly improved adherence to the PONV guidelines, guidelines adherence decreased to the level before use of the DSS after its withdrawal from clinical routine (Figure 5).

**Figure 5.** Guidelines adherence for high risk patients by week [16].

Surgical wound infections are relatively common, as they are considered the second most common complications occurring in a hospitalized patient [34, 35], and the second most common nosocomial infections, occurring in 2%–5% of surgeries and in up to 20% of ab‐ dominal surgeries [36]. They have a significant economical impact, because patients affected spend more time in the hospital and are more in danger to be admitted to an intensive care unit, to be readmitted to the hospital after discharge, or to die [37]. Antimicrobial prophylax‐ is is most effective when administered before surgical incision, with an optimal time to be within 30 minutes before incision or within 2 hours if vancomycin is administered [38, 39].

In order to facilitate timely administration, DSSs were implemented in AIMS to obtain better adherence with those guidelines. One of these is an automated computer-based documenta‐ tion that generates automatic reminders to the anesthesia team and the surgeon[40]. In this

study, authors found that 70% of all surgical patients received their antibiotics within 60 min of incision (Figure 6); after one year, the adherence increased to about 92%.

**Figure 6.** Administration of antibiotic: gradually increasing to about 92% [40].

tive nausea and vomiting (PONV) [33]. A database was implemented with PONV prophy‐ laxis guidelines. The comparison of two groups (one with only AIMS and the other with AIMS and also DSS), found that automated reminders were more effective for adherence

It also showed a reduction of inappropriate administration of PONV prophylaxis medica‐ tion to low-risk patients: automated reminders not only are effective in promoting correct actions, but may also prevent unnecessary prescription of medication, hence reducing drug costs. Although the DSS significantly improved adherence to the PONV guidelines, guidelines adherence decreased to the level before use of the DSS after its withdrawal

Surgical wound infections are relatively common, as they are considered the second most common complications occurring in a hospitalized patient [34, 35], and the second most common nosocomial infections, occurring in 2%–5% of surgeries and in up to 20% of ab‐ dominal surgeries [36]. They have a significant economical impact, because patients affected spend more time in the hospital and are more in danger to be admitted to an intensive care unit, to be readmitted to the hospital after discharge, or to die [37]. Antimicrobial prophylax‐ is is most effective when administered before surgical incision, with an optimal time to be within 30 minutes before incision or within 2 hours if vancomycin is administered [38, 39]. In order to facilitate timely administration, DSSs were implemented in AIMS to obtain better adherence with those guidelines. One of these is an automated computer-based documenta‐ tion that generates automatic reminders to the anesthesia team and the surgeon[40]. In this

to PONV prophylaxis (Figure 4).

22 Decision Support Systems

from clinical routine (Figure 5).

**Figure 5.** Guidelines adherence for high risk patients by week [16].

**Figure 7.** Anesthesia information management system screen overlaid by SAM screen [41].

Another DSS for antibiotic prophylaxis, the so-called Smart Anesthesia Messenger (SAM) [41], analyzes AIMS documentation data in real-time. Conceived as the final stage of inter‐ vention, after implementation in an AIMS, SAM transmits reminder messages to the AIMS screen to improve compliance of antibiotic administration before surgical incision (Figure 7). The addition of real-time reminders and feedback via SAM achieved near 100% compliance.

A follow-up study investigated the impact of the same DSS on the re-dosing of antibiotic therapy [42] in comparison with the use of only AIMS. Re-dosing could be important to maintain the necessary serum concentration of drug, to reduce the risk of postoperative wound infections in procedures that exceed of two half-lives of an antibiotic drug [43, 44]. In this study, a reminder message of re-dosing was effectuated every 3 hours (the short‐ est re-dose interval in guidelines of University of Washington Medical Center). The SAM detected the eventual administration of the prophylactic antibiotic drug, if necessary, it triggers an internal timer specific to that antibiotic and generates reminder icons 15 min prior to the time of re-dosing; these messages are repeated every 6 minutes until the dose is administered and documented (Figure 8). The employment of real-time decision sup‐

Decision Support Systems in Medicine - Anesthesia, Critical Care and Intensive Care Medicine

http://dx.doi.org/10.5772/51756

25

A further example of advantageous use of DSS integrated in AIMS is an electronic re‐ minder to switch on the ventilator alarms after separation from cardiopulmonary bypass (CPB) [45]. In cardiac surgery, during the CPB period, monitor alarms are often disabled; the alarms are frequently not reactivated. The software detects the separation from CPB by return of aortic and pulmonary blood flow, the resumption of mechanical ventilation and the reappearance of end-tidal CO2. If alarms have not been reactivated after the sepa‐ ration from CPB, an electronic reminder appears on the AIMS screen (Figure 9). The

A simple DSS combines a small amount of data to deal with one particular problem; it is like an electronic textbook about a specific issue, with the capability of giving the important in‐ formation at the right time. Usually, problems for which these DSSs are created are very common or insidious. A simple DSS could represent the first step for the progressive devel‐

An example of a simple DSS is a system that detects 'light' anesthesia using as input the changes of mean arterial pressure (MAP) [46]. Krol and Reich considered a 12% change in median MAP in comparison with the median value of MAP over the previous 10 min period

Another DSS involved in the detection of light anesthesia is an algorithm that relates differ‐ ent MAC values of volatile anesthetics to different intravenous sedative or hypnotics agents

The introduction of fuzzy logic for setting up DSSs is founded on the ability of fuzzy-logic in dealing with the incompleteness and vagueness that often characterize medical data and knowledge [3]; in 1997, a fuzzy-logic based DSS to control the supply of oxygen in a patient

A more recent Fuzzy-Logic Monitoring System (FLMS) has been developed [49]; this is a DSS conceived to detect critical events during anesthesia; it is able to detect only hypovo‐ lemia, using as inputs heart rate (HR), blood pressure (BP) and pulse volume (PV). Hypo‐ volemia is classified as mild, moderate or severe. The FLMS was evaluated in 15 patients

port improved the success rate to 83.9%.

alarm reactivation increased from 22% to 83%.

opment of a more complex DSS.

administered at the same time [47].

**2.2. Simple DSSs for a single intraoperative problem**

a parameter to trigger warnings for recognition of light anesthesia.

during low-flow/closed-loop anesthesia was presented (Figure 10) [48].

**Figure 8.** Messages for antibiotic re-dose. (A) Message reminding anesthesia team about need for re-dose. (B) Mes‐ sage about documenting re-dose [42].

**Figure 9.** Main window of anesthesia information management system with the electronic reminder [45].

A follow-up study investigated the impact of the same DSS on the re-dosing of antibiotic therapy [42] in comparison with the use of only AIMS. Re-dosing could be important to maintain the necessary serum concentration of drug, to reduce the risk of postoperative wound infections in procedures that exceed of two half-lives of an antibiotic drug [43, 44]. In this study, a reminder message of re-dosing was effectuated every 3 hours (the short‐ est re-dose interval in guidelines of University of Washington Medical Center). The SAM detected the eventual administration of the prophylactic antibiotic drug, if necessary, it triggers an internal timer specific to that antibiotic and generates reminder icons 15 min prior to the time of re-dosing; these messages are repeated every 6 minutes until the dose is administered and documented (Figure 8). The employment of real-time decision sup‐ port improved the success rate to 83.9%.

A further example of advantageous use of DSS integrated in AIMS is an electronic re‐ minder to switch on the ventilator alarms after separation from cardiopulmonary bypass (CPB) [45]. In cardiac surgery, during the CPB period, monitor alarms are often disabled; the alarms are frequently not reactivated. The software detects the separation from CPB by return of aortic and pulmonary blood flow, the resumption of mechanical ventilation and the reappearance of end-tidal CO2. If alarms have not been reactivated after the sepa‐ ration from CPB, an electronic reminder appears on the AIMS screen (Figure 9). The alarm reactivation increased from 22% to 83%.

#### **2.2. Simple DSSs for a single intraoperative problem**

Another DSS for antibiotic prophylaxis, the so-called Smart Anesthesia Messenger (SAM) [41], analyzes AIMS documentation data in real-time. Conceived as the final stage of inter‐ vention, after implementation in an AIMS, SAM transmits reminder messages to the AIMS screen to improve compliance of antibiotic administration before surgical incision (Figure 7). The addition of real-time reminders and feedback via SAM achieved near 100% compliance.

**Figure 8.** Messages for antibiotic re-dose. (A) Message reminding anesthesia team about need for re-dose. (B) Mes‐

**Figure 9.** Main window of anesthesia information management system with the electronic reminder [45].

sage about documenting re-dose [42].

24 Decision Support Systems

A simple DSS combines a small amount of data to deal with one particular problem; it is like an electronic textbook about a specific issue, with the capability of giving the important in‐ formation at the right time. Usually, problems for which these DSSs are created are very common or insidious. A simple DSS could represent the first step for the progressive devel‐ opment of a more complex DSS.

An example of a simple DSS is a system that detects 'light' anesthesia using as input the changes of mean arterial pressure (MAP) [46]. Krol and Reich considered a 12% change in median MAP in comparison with the median value of MAP over the previous 10 min period a parameter to trigger warnings for recognition of light anesthesia.

Another DSS involved in the detection of light anesthesia is an algorithm that relates differ‐ ent MAC values of volatile anesthetics to different intravenous sedative or hypnotics agents administered at the same time [47].

The introduction of fuzzy logic for setting up DSSs is founded on the ability of fuzzy-logic in dealing with the incompleteness and vagueness that often characterize medical data and knowledge [3]; in 1997, a fuzzy-logic based DSS to control the supply of oxygen in a patient during low-flow/closed-loop anesthesia was presented (Figure 10) [48].

A more recent Fuzzy-Logic Monitoring System (FLMS) has been developed [49]; this is a DSS conceived to detect critical events during anesthesia; it is able to detect only hypovo‐ lemia, using as inputs heart rate (HR), blood pressure (BP) and pulse volume (PV). Hypo‐ volemia is classified as mild, moderate or severe. The FLMS was evaluated in 15 patients using off-line data and was found to be in good agreement with the anesthetist's diagno‐ sis. An upgrading of this system, FLMS-2 [50], tested in 20 off-line patients, has demon‐ strated a sensitivity of 94%, specificity of 90% and predictability of 72%. The user interface of FLMS-2 is shown in Figure 11.

**2.3. Complex DSSs for intraoperative use**

sidered as "intelligent textbooks".

A complex DSS is software dealing with multiple problems. According with the complexity of the issue, it usually requires the collection of a certain number of information to combine with mathematical algorithm. It does not respond only to one problem, but can recognize different questions, sometimes inherent in a same category. These systems have to be con‐

Decision Support Systems in Medicine - Anesthesia, Critical Care and Intensive Care Medicine

http://dx.doi.org/10.5772/51756

27

One of the first complex DSSs for critical events in anesthesia was SENTINEL [51]. Based on fuzzy logic templates, this system used signals to establish a diagnosis despite missing information: it calculated the impact of lack of one or more signals for a certain condition via the estimation of the *completeness factor* [52]; the combination of some signals was judged as more important than others. The likelihood of a given diagnosis is measured considering two parameters of evidence: the *belief* (total of data supporting the evidence of a diagnosis) and the *plausibility* (the amount of data that do not contradict the diagno‐ sis). At the beginning, this system was designed to detect only one problem, malignant hyperpyrexia (MH, between 1:5000 and 1:100000 episodes [53]). Lowe and Harrison [54] set up rules based on characteristic patterns of changes in heart rate, end-tidal carbon di‐ oxide and temperature found in the literature and tested their software in a human simu‐ lator (Human Patient Simulator, version 1.3, University of Florida). During open surgery, the algorithm detected MH 10 minutes before the anesthetist; during laparoscopic sur‐ gery, in a condition with some similarities to MH (high end tidal CO2, cardiovascular changes), the diagnosis was only transient. Afterwards, SENTINEL was implemented with other rules to deal with other six conditions (Table 1). The interface of the system is depicted in Figure 12. SENTINEL was only tried in off-line tests, and its diagnostic alarms were compared with the annotations of anesthetists, showing a sensitivity of 95% and a

specificity of 90% (during the period between induction and recovery phases).

**Table 1.** Diagnoses and their descriptions for the fuzzy trend templates [52].

**Figure 10.** Scheme of fuzzy logic control system. Volume of the reservoir bag (BAGVOL) and his rate of change (DEL‐ TAVOL) are the inputs data for the first module (FZ module 1) to calculate the supply of oxygen (OXSUP); this value is sent as output data together with generated alarms (AL 1 and AL 2) to the second module (FZ module 2), that corre‐ lates them with oxygen concentration values in inspired (INSO) and expired air (EXPO) to generate simple diagnostic messages including obstructions (OBS), overfilling (OFILL), leakage (LEAK), and entrapment ((ENTR) in the system and metabolism (METAB), cardiovascular (CVS) and other (OTHER) problems with patient [48].

**Figure 11.** Graphic user interface [50].

#### **2.3. Complex DSSs for intraoperative use**

using off-line data and was found to be in good agreement with the anesthetist's diagno‐ sis. An upgrading of this system, FLMS-2 [50], tested in 20 off-line patients, has demon‐ strated a sensitivity of 94%, specificity of 90% and predictability of 72%. The user

**Figure 10.** Scheme of fuzzy logic control system. Volume of the reservoir bag (BAGVOL) and his rate of change (DEL‐ TAVOL) are the inputs data for the first module (FZ module 1) to calculate the supply of oxygen (OXSUP); this value is sent as output data together with generated alarms (AL 1 and AL 2) to the second module (FZ module 2), that corre‐ lates them with oxygen concentration values in inspired (INSO) and expired air (EXPO) to generate simple diagnostic messages including obstructions (OBS), overfilling (OFILL), leakage (LEAK), and entrapment ((ENTR) in the system and

metabolism (METAB), cardiovascular (CVS) and other (OTHER) problems with patient [48].

**Figure 11.** Graphic user interface [50].

interface of FLMS-2 is shown in Figure 11.

26 Decision Support Systems

A complex DSS is software dealing with multiple problems. According with the complexity of the issue, it usually requires the collection of a certain number of information to combine with mathematical algorithm. It does not respond only to one problem, but can recognize different questions, sometimes inherent in a same category. These systems have to be con‐ sidered as "intelligent textbooks".

One of the first complex DSSs for critical events in anesthesia was SENTINEL [51]. Based on fuzzy logic templates, this system used signals to establish a diagnosis despite missing information: it calculated the impact of lack of one or more signals for a certain condition via the estimation of the *completeness factor* [52]; the combination of some signals was judged as more important than others. The likelihood of a given diagnosis is measured considering two parameters of evidence: the *belief* (total of data supporting the evidence of a diagnosis) and the *plausibility* (the amount of data that do not contradict the diagno‐ sis). At the beginning, this system was designed to detect only one problem, malignant hyperpyrexia (MH, between 1:5000 and 1:100000 episodes [53]). Lowe and Harrison [54] set up rules based on characteristic patterns of changes in heart rate, end-tidal carbon di‐ oxide and temperature found in the literature and tested their software in a human simu‐ lator (Human Patient Simulator, version 1.3, University of Florida). During open surgery, the algorithm detected MH 10 minutes before the anesthetist; during laparoscopic sur‐ gery, in a condition with some similarities to MH (high end tidal CO2, cardiovascular changes), the diagnosis was only transient. Afterwards, SENTINEL was implemented with other rules to deal with other six conditions (Table 1). The interface of the system is depicted in Figure 12. SENTINEL was only tried in off-line tests, and its diagnostic alarms were compared with the annotations of anesthetists, showing a sensitivity of 95% and a specificity of 90% (during the period between induction and recovery phases).


**Table 1.** Diagnoses and their descriptions for the fuzzy trend templates [52].

With the implementation of a Multi-Modal Alarms System (MMAS) [58], RT-SAAM was able to diagnose also sympathetic activity, relative hypovolemia and inadequate anesthesia; diagnostic messages and alerts were sent every 10 seconds to MMAS. Every outcome alarm was connected to a specific sound that was directly transmitted to the anesthetist through a bluetooth headset. The MMAS display had two different modalities of presentation, de‐

Decision Support Systems in Medicine - Anesthesia, Critical Care and Intensive Care Medicine

http://dx.doi.org/10.5772/51756

29

In 2008, Perkin and Leaning presented Navigator, a DSS involved in the therapeutic con‐ trol of the circulation and the oxygen delivery optimization and management [60]. They developed a mathematical model to create an algorithm for the control of circulation based on the values of the effective circulating volume (Pms), systemic vascular resistance (SVR) and heart performance (Eh). Using mathematical techniques, the values were de‐ rived from measured circulatory variables, the mean arterial pressure (MAP), the right at‐ rial pressure (RAP) and the cardiac output (CO): corrected with a factor, *c*, that correlates

Through the combination of these values, Navigator supports the decision process with con‐ tinuous therapeutic informations about the hemodynamic status and the oxygen delivery in‐ dex related to the cardiac output, based on the entered hemoglobin and the arterial oxygen saturation (Spo2). The system display (Figure 15) is organized as such: on the right side, there is the current status of the patient, with his current values acquired from the monitors, target val‐ ues and other data; on the left side, there is the patient's position (the red dot in the yellow ar‐ row) on an orthogonal graph, in which: *x*-axis is the resistance axis (SVR values); *y*-axis is the volumetric axis (Pms values); MAP and CO are shown as lines corresponding to their upper

pending on the presence or not of the symptoms (Figure 14).

**Figure 14.** Alert modality of presentation [59].

with height, weight and age of the subject.

**Figure 12.** Prototype of SENTINEL user interface. Weak diagnosis of absolute hypovolemia (AHV) [55].

In the wake of SENTINEL, another DSS for critical events in anesthesia was presented in 2007, called Real Time-Smart Alarms for Anesthesia Monitoring (RT-SAAM) [56]. Initially, it was proposed to recognize and suggest treatment options of hypovolemia and decreas‐ ing cardiac output. Based on the evidence that hypovolemia can be detected by monitor‐ ing systolic pressure variations (SPV) in patients artificially ventilated [57], the DSS filtered the blood pressure (BP), pulse volume (PV), end-tidal carbon-dioxide (ETCO2) waveforms and calculated the SPV and the absolute PV values, providing diagnostic in‐ formation on the monitor in real-time (Figure 13). Tested in 18 patients in retrospective tests and in 8 patients during real-time tests, a moderate level of agreement between the DSS and the anesthesiologist was determined.

**Figure 13.** RT-SAAM screen with windows diagnoses. AHW (acute hypovolemia), hypovolemia; fall in cardiac out‐ put (FCO) [56].

With the implementation of a Multi-Modal Alarms System (MMAS) [58], RT-SAAM was able to diagnose also sympathetic activity, relative hypovolemia and inadequate anesthesia; diagnostic messages and alerts were sent every 10 seconds to MMAS. Every outcome alarm was connected to a specific sound that was directly transmitted to the anesthetist through a bluetooth headset. The MMAS display had two different modalities of presentation, de‐ pending on the presence or not of the symptoms (Figure 14).

**Figure 14.** Alert modality of presentation [59].

**Figure 12.** Prototype of SENTINEL user interface. Weak diagnosis of absolute hypovolemia (AHV) [55].

DSS and the anesthesiologist was determined.

put (FCO) [56].

28 Decision Support Systems

In the wake of SENTINEL, another DSS for critical events in anesthesia was presented in 2007, called Real Time-Smart Alarms for Anesthesia Monitoring (RT-SAAM) [56]. Initially, it was proposed to recognize and suggest treatment options of hypovolemia and decreas‐ ing cardiac output. Based on the evidence that hypovolemia can be detected by monitor‐ ing systolic pressure variations (SPV) in patients artificially ventilated [57], the DSS filtered the blood pressure (BP), pulse volume (PV), end-tidal carbon-dioxide (ETCO2) waveforms and calculated the SPV and the absolute PV values, providing diagnostic in‐ formation on the monitor in real-time (Figure 13). Tested in 18 patients in retrospective tests and in 8 patients during real-time tests, a moderate level of agreement between the

**Figure 13.** RT-SAAM screen with windows diagnoses. AHW (acute hypovolemia), hypovolemia; fall in cardiac out‐

In 2008, Perkin and Leaning presented Navigator, a DSS involved in the therapeutic con‐ trol of the circulation and the oxygen delivery optimization and management [60]. They developed a mathematical model to create an algorithm for the control of circulation based on the values of the effective circulating volume (Pms), systemic vascular resistance (SVR) and heart performance (Eh). Using mathematical techniques, the values were de‐ rived from measured circulatory variables, the mean arterial pressure (MAP), the right at‐ rial pressure (RAP) and the cardiac output (CO): corrected with a factor, *c*, that correlates with height, weight and age of the subject.

Through the combination of these values, Navigator supports the decision process with con‐ tinuous therapeutic informations about the hemodynamic status and the oxygen delivery in‐ dex related to the cardiac output, based on the entered hemoglobin and the arterial oxygen saturation (Spo2). The system display (Figure 15) is organized as such: on the right side, there is the current status of the patient, with his current values acquired from the monitors, target val‐ ues and other data; on the left side, there is the patient's position (the red dot in the yellow ar‐ row) on an orthogonal graph, in which: *x*-axis is the resistance axis (SVR values); *y*-axis is the volumetric axis (Pms values); MAP and CO are shown as lines corresponding to their upper and lower target ranges; the equivalent delivery oxygen indices are shown on the CO lines; the heart performance (Eh values) is displayed like a vertical axis parallel to the Pms axis.

Pellegrino et al [62] assessed Navigator in postoperative cardiac surgical patients. Fifty-seven patients received DSS-guided care and were compared with 48 patients who received conven‐ tional care. The performance of the system, considered as "average standardized distance" (ASD) between actual and target values of MAP and CO, was statistically not inferior to the control, and there were no significantly differences in the hospital length of stay (Table 2).

Decision Support Systems in Medicine - Anesthesia, Critical Care and Intensive Care Medicine

http://dx.doi.org/10.5772/51756

31

Sondergaard et al. [61] tested the Navigator's hemodynamic control and oxygen delivery during elective major abdominal surgery. They compared two groups of patients, one treat‐ ed using DSS and the other one treated by expert anesthetists. They found a high concord‐

Another complex DSS conceived to assist the anesthetist during surgery is Diagnesia [63]. It uses the input from the anesthesia panel to estimate the likelihood or unlikelihood of a diag‐ nosis; it then gives the five most probable diagnoses in descending order (from the most to the least likely) with respective information that support or are against the evidence (Figure 16). Tested in 12 realistic situations from simulated anesthesia monitoring displays, its diag‐ noses were compared with those of a group of anesthesiologists, and in 11 test cases (92%), the most probable diagnosis was the same; however, the system couldn't distinguish be‐ tween two or more specific problems from the same category and couldn't deal with diag‐

ance between the advices of the system and the intervention of the anesthetists.

nosis in which the indicators were only observable but not measurable.

**Figure 16.** Graphical user interface [64].

**Figure 15.** Navigator display [61].


**Table 2.** ASD=average standardized distance, MAP=mean arterial pressure, CO=cardiac output, AF=atrial fibrillation, SOFA=Sequential Organ Failure Assessment [62].

Pellegrino et al [62] assessed Navigator in postoperative cardiac surgical patients. Fifty-seven patients received DSS-guided care and were compared with 48 patients who received conven‐ tional care. The performance of the system, considered as "average standardized distance" (ASD) between actual and target values of MAP and CO, was statistically not inferior to the control, and there were no significantly differences in the hospital length of stay (Table 2).

Sondergaard et al. [61] tested the Navigator's hemodynamic control and oxygen delivery during elective major abdominal surgery. They compared two groups of patients, one treat‐ ed using DSS and the other one treated by expert anesthetists. They found a high concord‐ ance between the advices of the system and the intervention of the anesthetists.

Another complex DSS conceived to assist the anesthetist during surgery is Diagnesia [63]. It uses the input from the anesthesia panel to estimate the likelihood or unlikelihood of a diag‐ nosis; it then gives the five most probable diagnoses in descending order (from the most to the least likely) with respective information that support or are against the evidence (Figure 16). Tested in 12 realistic situations from simulated anesthesia monitoring displays, its diag‐ noses were compared with those of a group of anesthesiologists, and in 11 test cases (92%), the most probable diagnosis was the same; however, the system couldn't distinguish be‐ tween two or more specific problems from the same category and couldn't deal with diag‐ nosis in which the indicators were only observable but not measurable.

**Figure 16.** Graphical user interface [64].

and lower target ranges; the equivalent delivery oxygen indices are shown on the CO lines; the

**Table 2.** ASD=average standardized distance, MAP=mean arterial pressure, CO=cardiac output, AF=atrial fibrillation,

heart performance (Eh values) is displayed like a vertical axis parallel to the Pms axis.

**Figure 15.** Navigator display [61].

30 Decision Support Systems

SOFA=Sequential Organ Failure Assessment [62].

Lastly, a hybrid system for conscious sedation (HSS) with DSS, was presented [65]. This system integrates closed loop sedation with a DSS, offering pop-up menus as smart alarms with several treatment advices for hemodynamic or respiratory adverse events, which need to be confirmed by the anesthetic team by clicking respective touch buttons on a touch screen (Figure 17).

the whole health care team [69-72]. These considerations lead to a request for decision sup‐ port. There are several difficulties for designing systems for this environment and testing them. In a study of 2003, the diagnostic performance of two DSSs planned for an emergency department [73] were compared with expert decision making. Only in one third of the cases,

Decision Support Systems in Medicine - Anesthesia, Critical Care and Intensive Care Medicine

http://dx.doi.org/10.5772/51756

33

We discuss further DSSs for in-hospital emergency medicine and DSSs for pre-hospital and

Chest pain and abdominal pain represent principal causes of admission to the emergency room [74]. An example of a DSS for in-hospital emergencies is an artificial neural network designed to diagnose acute myocardial infarction [75]. It was tested in 118 patients, with a sensitivity of 92% and specificity of 96%; without the input of the electrocardiogram, the sensitivity decreased to 86%, while the specificity was 92%. In a later study, the same DSS was tested in 331 patients with anterior chest pain in emergency department [76]. Whereas sensitivity and specificity of physicians in diagnosing myocardial infarction was 77.7% and 84.7%, the DSS values were higher at 97.2% and 96.2%, respectively. This suggests that the

A simple DSS that could be useful in emergency medicine is an artificial neural network designed to detect microembolic Doppler signals [77]; it resulted in a specificity of 56.7% and sensitivity of 73.4%, increasing to 75.9% in patients with mechanical prosthetic car‐ diac valves. However, this study did not test this system in emergency situations. Results about a favorable use of DSS to diagnose pulmonary embolism in emergency medicine in fact are not uniform. For example, the study of Roy et al. tried to evaluate the importance of the introduction of a computer-handled DSS in diagnosing pulmonary embolism com‐ pared with the use of paper guidelines [78]. The software yields a list of tests, specifying which of them is appropriate or inappropriate considering the pre-test probability, en‐ tered before by the physician according to the revised Geneva score; the DSS recom‐ mends as first choice the least invasive investigation among the appropriate tests. The system was compared with paper guidelines in two groups of patients, one with DSS and one without it. In the intervention group (the DSS group), there was an increase of appro‐ priate diagnostic testing by 30.2% while in the other group it only increased by 10.9%. The DSS was used in 80% of cases during real time intervention, suggesting a good per‐ formance in the emergency environment and good acceptance by physicians. A study by Drescher evaluated the impact of the integration of a computerized DSS in a computer‐ ized physician order entry on the frequency of positive CT angiography results for pul‐ monary embolism and the staff's acceptability of such a DSS[79]. The DSS was designed based on a modified Wells score to give diagnostic options to physicians when ordering a CT angiography or D-dimer testing (Figure 18). Although the study showed a superior performance of the DSS rather than physicians in ordering diagnostic investigations, ad‐ herence and acceptability of the DSS were quite low. Probably, the principle reasons were

the time needed to enter data and mistrust in the effectiveness of the software.

the experts' diagnosis was within the top five diagnosis generated by the software.

**3.1. Decision support systems for in-hospital emergency medicine**

out-of-hospital emergency medicine.

system performed better than physicians.

Tested on two groups of 50 patients, the detection of critical events was significantly im‐ proved by the DSS, as shown in Table 3.


**Table 3** Comparison of detecting critical events by the time [65].

**Figure 17.** Pop-up menu for respiratory critical event [65].
