Evaluating the effectiveness of Bayesian feature selection

A practical Bayesian approach for inference in neural network models has been available for ten years, and yet it is not used frequently in medical applications. In this chapter we show how both regularisation and feature selection can bring significant benefits in diagnostic tasks through two case studies: heart arrhythmia classification based on ECG data and the prognosis of lupus. In the first of these, the number of variables was reduced by two thirds without significantly affecting performance, while in the second, only the Bayesian models had an acceptable accuracy. In both tasks, neural networks outperformed other pattern recognition approaches.

Development and trialling of a behavioural intervention for patients with atrial fibrillation initiating oral anticoagulation:the 'treat' study

Purpose: Atrial fibrillation (AF) is the most common heart arrhythmia and is associated with an increased risk of stroke. Stroke risk is commonly treated with oral anticoagulation (OAC) with a narrow therapeutic range (INR 2.0 to 3.0); which is poorly controlled in practice. Barriers to adherence include poor knowledge, and inaccurate perceptions surrounding illness and medications. Trial registration: ISRCTN93952605. Systematic review: Seven trials of educational, self-monitoring and decision aid interventions were included in a systematic review. Pooled analysis suggested education OR, 95% CI 7.89 (5.54-10.24) and self monitoring OR (95% CI) 5.47(2.55-8.39) significantly improve TTR; whereas decision aids are no more effective in reducing decision conflict than usual care, OR (95% CI) -0.10 (-0.17 to -0.02). Intervention development: The intervention was theoretically-driven (utilising the common sense and beliefs about medication models) and developed with expert patient feedback. Described using behavioural change techniques, the one-off group session included an educational booklet, ‘expert-patient’ focussed DVD, and worksheet. Methods: Ninety seven warfarin-naïve AF patients were randomised to receive the intervention (n=43), or usual care (n=54). The primary endpoint was time within therapeutic range (TTR), secondary endpoints included knowledge, quality of life (AF-QoL-18), beliefs about medication (BMQ), illness perceptions (IPQ-B), and anxiety and depression (HADS). Results: Intervention group had significantly higher TTR than usual care (78.5% vs. 66.7%; p=0.01). Knowledge changed significantly across time (F (3, 47) = 6.4; p<0.01), but not between groups (F (1, 47) = 3.3; p = 0.07). At six months knowledge predicted TTR (r=0.245; p=0.04). Illness concern negatively correlated with TTR (r= - 0.199; p=0.05). General Harm scores at one month predicted TTR (F (1, 72) = 4.08; p=0.048). There were significant differences in emotional representations (F (3, 49) = 3.3 (3, 49); p= 0.03), anxiety (F (3, 46) = 25.2; p<0.01) and depression (F (3, 46) = 37.7; p<0.01) across time. Conclusion: A theory-driven educational intervention can improve TTR in AF patients and potentially reduce the risk of adverse clinical outcomes. Improving education provision for AF patients is essential to ensure efficacious treatment.