This study aimed to evaluate the applicability of a clinical decision support system (CDSS) based on a Bayesian network, which aims to support the interdisciplinary team in risk screening for laryngotracheal aspiration in hospitalized adults.
This is a methodological study, as it allowed the development of a technological tool, associated with the collection of data and organization of information necessary for the prototyping of a CDSS to be executed on an Android mobile device, aiming at the adoption of diagnostic recommendations or strategies [31].
The conclusions of this CDSS were defined by colors, percentages, and risk levels. The definitions by percentage and risk level were defined by the specialists who developed this study. They consider the scores described in the validation of the Gugging Swallowing Screen (GUSS) [17]. The 4 risk levels were defined as high risk 55.01–100%; moderate risk 30.01% − 55%; low risk 5.01% − 30% and minimal risk 0%-5%.
The color definition used in the CDSS was stratified in the form of four colors, which were related to the severity of the risk: red (high risk), orange (moderate risk), yellow (low risk), and green (minimum risk). This is according to the criteria defined in the Manchester protocol and adapted by the specialists who developed this Bayesian network [18].
The development of the CDSS follows the approach described in Fig. 1.
Figure 1: Schematic representation of the system development phases from this study.
Identification Of Risk Indicators For Laryngotracheal Aspiration
From a systematic review, 14 scientific articles (SA) related to LA risk indicators were selected. Among these, 11 SA exclusively describe LA risk indicators [19–20–21–22–23–24–25–26–27–28–29] and 3 SA contain variables related to LA risk [11-12- 13]. Of the 14 SA analyzed, 25 variables were selected, listed, analyzed, and grouped by the specialists of this study in order to identify the most relevant.
These were classified into 5 main groups. Each of the groups contains subgroups with their corresponding classification. See Table 1.
Table 1 - Risk indicators for laryngotracheal aspiration
Qualitative Development Of The Bayesian Network (Bn)
A BN is an oriented acyclic graph, where the nodes represent random variables and the arc joining two nodes represents the probabilistic dependence between the associated variables. Each node has stored the conditional probability distribution function of the values that can be assumed by the random variable associated with the node, given the values of its parent nodes (ie, those directly linked to the node in question). The BN is a compact representation of the global joint probability distribution function of all random variables in the modeled domain [15–30]. Therefore, a BN is composed of two parts, qualitative and quantitative, described as follows:
The qualitative development was presented by a graphic model, with the variables represented by nodes and the relationship between them represented by directed arcs. The qualitative part of this BN was built from the clinical experience of the researchers. In this BN, the central node is laryngotracheal aspiration (ALT), which depends on 5 categories: current diagnosis of the patient (Dx_Ppte), history associated with swallowing (Hist_REL), other factors related to swallowing (Others_Fat), indirect aspects of swallowing (Pre_Alim ) and direct aspects of swallowing (VA). As described in Fig. 2.
Figure 2: Qualitative development of the Bayesian network
Quantitative Development Of The Bayesian Network
The quantitative relationship was developed through a set of conditional probabilities associated with each arc created in the graphic model, a priori, of the diagnostic hypotheses. The quantitative portion measures the probability with which the patient satisfies a certain criterion. The quantitative portion can be obtained through specialists in the BN knowledge domain or through a database of real cases. Due to the lack of information in the database available for this study, the specialists from the research group defined the parameters of the conditional probability tables. See Table 2.
Table 2 - System a priori variables and probabilities
The variables obtained in this study were selected by speech pathologists with clinical and academic experience. The a priori values of the variables described in Table 1 were obtained after analyzing the scientific articles that included them. A dependency relationship between these variables needed to be defined. Table 3 presents the interdependence relationship established in this study.
Table 3 - Interdependence of variables
System Evaluation
The CDSS assessment, called “SR.REAL”, was carried out following the guidelines of ISO/IEC 25010, 2011 [32]. This standard is a reference for promoting the performance of the internal and external quality of software products. It is based on three levels: features, sub-features, and metrics. Each characteristic is divided into sub-characteristics, which in turn are evaluated against a set of metrics. The evaluation of internal and external quality is divided into functionality, reliability, usability, efficiency, maintainability, and portability. Five closed-ended questions were defined by the researchers to identify the opinion of experts concerning external quality. Six open-ended questions allowed professionals to give their opinion on the possibility of implementing this tool in the hospital environment.
Expert Assessment
For the evaluation of the internal and external quality of CDSS, the participation of 9 specialists in the area of oropharyngeal dysphagia and LA was necessary. Non-probabilistic sampling by judgment was held [33]. The selection of participants to compose the CDSS internal and external quality assessment occurred as we view them are sources of information and clinical experience necessary to achieve our objective. Nurses (2), nutritionists (2), physicians (2), and speech pathologists (3) were summoned. The inclusion criteria (CI) were: CI1 professional duly registered in their professional body; CI2 proven clinical experience of at least 5 years with hospitalized adults with Oropharyngeal Dysphagia (OD) and LA; CI3: voluntary decision to participate in the study, after signing the informed consent form approved by the Research Ethics Committee of the Federal University of Health Sciences of Porto Alegre (UFCSPA).
Statistical analysis
Categorical results will be presented by frequency and percentage. The answers to the open-ended questions were grouped according to the content. The general quality and its characteristics and sub-characteristics will be presented as mean, standard deviation, and coefficient of variation. For the sub-characteristics, relevance will be classified as high (score between 70.1 and 100), medium (between 50.1 and 70), and low (between 0 and 50). For the characteristics and the general quality, the quality will be classified as excellent (score between 90.1 and 100), good (score between 75.1 and 90), satisfactory (between 60.1 and 75), regular (between 50 and 60), and unsatisfactory (between 0 and 50). Data will be analyzed using SPSS software version 25 [34].