Applying artificial neural network models to clinical decision making.

Abstract

Because psychological assessment typically lacks biological gold standards, it traditionally has relied on clinicians' expert knowledge. A more empirically based approach frequently has applied linear models to data to derive meaningful constructs and appropriate measures. Statistical inferences are then used to assess the generality of the findings. This article introduces artificial neural networks (ANNs), flexible nonlinear modeling techniques that test a model's generality by applying its estimates against "future" data. ANNs have potential for overcoming some shortcomings of linear models. The basics of ANNs and their applications to psychological assessment are reviewed. Two examples of clinical decision making are described in which an ANN is compared with linear models, and the complexity of the network performance is examined. Issues salient to psychological assessment are addressed.In the area of psychological assessment, a computer-based algorithm approach has relied heavily on the expert knowledge of clinicians. In a more empirically based approach-used when insufficient knowledge is available, existing knowledge is debated, or a special population is investigated-linear models such as factor analysis or regression models are frequently applied to empirical data to derive meaningful constructs and appropriate measures. In this approach, statistical inferences are needed to assess the generality of constructs and rules because the gold standard of expert knowledge may not be applied. mortality. The details of ANN methodology in these contexts include ANN parameter specification, software, cross-validation, and the receiver operating characteristic (ROC) evaluation. The performance of the ANN is compared with the performance of the linear statistical analyses frequently used in the area of psychological assessment: linear and quadratic discriminant analyses and logistic regression analyses. We examine the performance of ANN models with varying numbers of hidden neurons to show when complex network architecture is needed to achieve a better level of prediction than that achieved in linear models. We then discuss issues for research practitioners to consider in order to make optimal use of ANNs. Future directions of ANN applications and current salient issues in psychological assessment are discussed. Artificial Neural Networks for Psychological AssessmentBackground

The use of artificial neural networks and multiple linear regression in modelling work–health relationships: Translating theory into analytical practice

“…Multiple regression analysis can then be used to supplement this information by discerning the relationship between specific work characteristics and health outcomes (e.g., the partial derivatives approach for risk estimation in a multivariate curvilinear context, Karanika-Murray et al, under review). A range of other approaches, such as principal components, discriminant, or cluster analysis, can also be used to examine neural network structure (Price et al, 2000). As Sargent (2001) A range of methods for the study of nonlinear systems have been developed (Barton, 1994;Eidelson, 1997;Somers, 2001).…”

“…social relationships, behavior) (see Lowe et al, 2003). Ways to 'open the black box' include, for example, examination of weights and confidence intervals, use of graphical methods, estimation criteria, and diagnostics (e.g., Price et al, 2000;Sale, 1994). These are not readily applicable yet, but "methodology and statistics are computer-dependent enterprises [...] as computer technology advances, the ability of researchers to advance the field increases dramatically" (Shadish, 2002, p. 12).…”

“…As a general rule, low R can Modeling Work-Health Relationships 24 indicate incomplete underlying theory or high measurement error in the data. Any type of analysis is as good as the data against which the conceptual models are examined (e.g., Price et al, 2000). Inclusion of additional predictors in future studies could further improve the models.…”

“…The neural network literature provides limited guidance on required sample size (Price et al, 2000) and optimal numbers of hidden layers and of hidden nodes (Anderson & McNeill, 1992). Estimations are based on the basis of considerations such as data noisiness (errors or imperfections) and model complexity.…”

“…Thus, complexity and nonlinearity are embedded in the network's architecture rather than the exact relationships between variables, thus allowing for a supple conceptualization of complex phenomena. On the negative side, many have described neural networks as a 'black box' approach (Paruelo &Tomasel, 1997;Price et al, 2000): it is not possible to discern the relationships between a specific predictor and the outcome in a neural network, in the same way that statistical formulae allow. However, as mentioned, this is often neither necessary in occupational health risk assessment nor feasible in organizational Modeling Work-Health Relationships 10 settings.…”