Posts

My colleague Lia Ali pointed me to this – at this year’s Royal College of Psychiatrist’s International Congress, there’s a debate titled Clinic appointment with Skynet? This house believes that the RCPsych should embrace Artificial Intelligence and Big Data in guiding clinical decision making and service development.

1 Purpose 2 Posterior Predictive Distributions 2.1 Conditional Probability 2.2 Law of Total Probability 2.3 Conditional Independence 2.4 Bayes Theorem 2.5 Bayes with Extra Conditioning 2.6 Deriving the PPD 3 Sampling and Integration 4 Logits, Probabilities and Odds 4.

Here’s the problem set-up: we have a group of diseases/disorders/conditions (A,B,C and D) and associated known treatments (RxA, RxB, RxC and RxD). We want to build graphs that represent associations between the conditions and treatments for a group of patients.

In clinical decision making for serious but rare events, there has been discussion about how to use predictive models as tools in decision making. One example is in decision making for assessment and treatment in people at risk of suicide.

In the previous post, loss functions where considered in the context of estimating measures of central tendency for distributions. In this post, I want to look at the computation of loss functions in situations that might arise in a clinical predictive model.

For a while, I’ve been thinking about the deployment of predictive algorithms in clinical decision support. Specifically, about the difference between what we understand about a model’s performance from the publication describing it and how this might be less informative when deployed.

The motivation for this post is a series of discussions I’ve had with colleagues over the past few weeks involving the potential for electronic health/medical records (EHRs or EMRs) for improving clinical care and their utility for research.

In this series of blogposts, we look at some models of clinical state. The motivation is to document exploratory work with a colleague (Nick Meyer, who runs the SleepSight study) as we try and apply some theoretical ideas – for example (Nelson et al.

The motivation for looking at this topic was primarily my naivety and getting counter-intuitive results when I was looking at serial measures of response to treatment. In particular, I wanted to use outcome measures based on published conventions in the literature for threshold response (the problems with this approach are a topic for another day).