Use of the self-organising map network (SOMNet) as a decision support system for regional mental health planning

Abstract

Background: Decision-making in mental health systems should be supported by the evidence-informed knowledge transfer of data. Since mental health systems are inherently complex, involving interactions between its structures, processes and outcomes, decision support systems (DSS) need to be developed using advanced computational methods and visual tools to allow full system analysis, whilst incorporating domain experts in the analysis process. In this study, we use a DSS model developed for interactive data mining and domain expert collaboration in the analysis of complex mental health systems to improve system knowledge and evidence-informed policy planning.

Methods: We combine an interactive visual data mining approach, the self-organising map network (SOMNet), with an operational expert knowledge approach, expert-based collaborative analysis (EbCA), to develop a DSS model. The SOMNet was applied to the analysis of healthcare patterns and indicators of three different regional mental health systems in Spain, comprising 106 small catchment areas and providing healthcare for over 9 million inhabitants. Based on the EbCA, the domain experts in the development team guided and evaluated the analytical processes and results. Another group of 13 domain experts in mental health systems planning and research evaluated the model based on the analytical information of the SOMNet approach for processing information and discovering knowledge in a real-world context. Through the evaluation, the domain experts assessed the feasibility and technology readiness level (TRL) of the DSS model.

Results: The SOMNet, combined with the EbCA, effectively processed evidence-based information when analysing system outliers, explaining global and local patterns, and refining key performance indicators with their analytical interpretations. The evaluation results showed that the DSS model was feasible by the domain experts and reached level 7 of the TRL (system prototype demonstration in operational environment).

Conclusions: This study supports the benefits of combining health systems engineering (SOMNet) and expert knowledge (EbCA) to analyse the complexity of health systems research. The use of the SOMNet approach contributes to the demonstration of DSS for mental health planning in practice.

Keywords: Decision support systems; Evidence-informed policy planning; Expert knowledge; Expert-based collaborative analysis; Health systems engineering; Interactive visual data mining; Key performance indicator; Mental health system; Self-organising map network.

Fig. 1

Three different phases of data processing (pre-, mid- and post-processing) for decision support systems are compared between knowledge discovery in databases (KDD) and ‘Expert-based Collaborative Analysis’ (EbCA). a The process of KDD [13]. b The operational process in EbCA [11]

Fig. 2

A self-organising map (SOM) example using the resource utilisation (USE) dataset. a The USE dataset table with six indicators (column) for 106 mental health areas (row). b The USE-SOM created by learning the USE dataset. c The visualisation examples of the USE-SOM. (c1) The indicator value planes underneath the USE-SOM. A mental health area, B5 is located on a neuron of the USE-SOM where its six indicator value locations are the same through the planes. (c2) The mental health areas are interpolated into the USE-SOM. The area labels are coloured in blue for Biscay, red for Gipuzkoa and black for the Catalonia areas. (c3) The property shape plane of the USE-SOM. A shape in a neuron represents the six indicator values of the neuron. The star glyph shape is created by marking the indicator values on the six evenly angled branches from the centre. (c4) A weight distribution is visualised on the USE-SOM. If a neuron has high weight its colour is red, while a low weight is indicated in blue

Fig. 3

The SOMNet analysis procedure applied to this study based on the EbCA process. The partial EbCA is shown (the full version is in Fig. 1b). The SOMNet process is indicated in black in the grey shaded area of the mid-processing phase of the EbCA. The analytical processes of the SOMNet are iterative until the analytical goals are achieved for knowledge discovery

Fig. 4

The visual identification of the system outliers in the initial USE-SOM. a The initial USE-SOM with the small mental health areas labelled (black, blue and red for Catalonia, Biscay and Gipuzkoa areas, respectively). The identified outlier areas are circled in orange. b The six indicator value planes of the USE-SOM showing the extreme values of the circled areas. The darker grey colour indicates the higher indicator value

Fig. 5

The visual identification of the global and local patterns of the mental health systems in Spain. The input SOMs are in a AVA-SOM, b PLA-SOM and c WOF-SOM, and the output SOM is in d USE-SOM. Their data property shape planes are visualised in (a'), (b'), (c') and (d'), respectively. The corresponding indicator values are pointed on the evenly angled star branches and connected to yield the shape. The legend of the property shape for each SOM is given with the order of indicators shown clockwise

Fig. 6

Visual analysis of the input and output indicator patterns by the SOMNet. a The output USE-SOM with areas. b The value planes of the output indicators, U1, U2 and U3. c The output pattern comparison using the input-driven analysis. d The input WOF-SOM with areas. e The value plane for the input indicator, W1. f The input pattern comparison using the output-driven analysis. c' The output USE-SOM pattern for the newly given W1 by the SOMNet analysis

Fig. 7

The feasibility results of the decision support systems model using the SOMNet approach. The average scores of the four feasibility evaluation dimensions are compared between the SOMNet and the other operation and visualisation approaches previously used in mental health studies. Two other dimensions (novelty and potentiality) are used separately to assess the SOMNet approach