Business as not usual

I ventured from the science pages on the Guardian website on Friday by commenting in a piece about the change in energy prices here. I was asked to look at a time series data set of wholesale prices and price to the customer. I did a very basic analysis by separating the time series into the trend, seasonality and noise using a mulitplicative model. I did this with the gas and electricity data and it showed some pretty obvious behaviour. In the main everything has an upward trend, to be expected with some minor deviations which I'm sure an economist can point at with specific causes. The seasonality effects are smaller for the customer than the wholesale but there is a clear bias towards customer increases rather than decreasing when comparing between plots. A simple regression on the trend indicates diverging costs but I'm sure there are lots of details that could be taken in to account. Basically, in a non-academic non-rigorous kind of way, it reasonable to comment that the curves for customer get a worse deal than those for wholesale. Saying that though, I'm sure there are things not captured by this data that need to considered to really make hard statements about overall cost. For example, the consumer plots are smoother which I suppose is for a practical reason. There'd be a good proper research project in all of this...

Inverse (sq)care law

The inverse care law says that the quality of the healthcare is inversely related to the population need. It is hypothesised that this could result from the increase in patient choice. Maybe not.

IGT Care Call Award

I was briefly involved in this project in the latter stages, producing descriptive statistics and plots for the output of the study. The idea was to support patients who are at high risk of becoming diabetic to avoid this happening. The support is in the form of telephone calls and shared goal setting with things like more exercise and healthy eating proposed.

Although there was no control group, as an open study there were some obvious temporal patterns which we can `point at' (in a Greenfield sense) the effect of the intervention.

Discrete Event Simulation Approximation

The DES algorithm we've implemented for the CHD model is, in the terminology of a paper by Gillespie, a direct stochastic simulation model (SSM). This basically means that each individual is simulated in continuous time until the specified end point e.g. death. The downside with this is that, especially for small transition times and large number of states, the runtime is prohibitively sloooooow. One way of addressing this is to approxaimate the process. The approximation should improve the computational performance and we should still be able to obtain the output of interest, though perhaps in a more course form.

Simple alternative version include a Poisson count process over a specified time window or a "tau jump" model which uses a Binomial approximation to generate transition numbers. A refined version can be used to ensure there are no more jumps from a state than allowed. I've implementented a version of this latter algorithm for the CHD model but used a multivariate distribution rather than a simpel Binomial, to account for the competing risks.

This approach thus relies on specifying the discrete transition probabilities for the distributions. At present these values are produced via a model fitting step where data is combined in the form of both rates and discrete probabilities. To simplify the implementation further, I fitted a logistic regression through the age categories for given from and to states, e.g. from Healthy to CHD Death for 0-25, 26-35, 36-45, 46-55, 56-65, 66-75, 76-85, 85+. This reduces down the necessary storage.

IMPACT II paper draft

Current paper draft of an informatics/statistics IMPACT II paper

IMPACT II Clinical Side Poster