Smoothing seasonality - your feedback needed
We received feedback from users that some of our measures are difficult to understand, because they exhibit a high level of seasonal variation. Antibiotics show the most seasonality amongst our measures (e.g. openprescribing.net/measure/ktt9_antibiotics). As you can see below, it takes a lot of staring at this graph to work out whether the deciles (blue lines) are going up or down overall, and whether the red line (the data from a single CCG) is changing in relation to them.
When doing the analysis for our soon to be published antibiotics trends paper (at this link once it’s online: doi.org/10.1093/jac/dky377), we experimented with smoothing these lines to make the trends over time clearer. The method we used is called “locally weighted scatterplot smoothing”, which draws a trend line through each seasonal line to smooth it out. The result is the lower graph below.
It’s immediately clear in the second graph that there is a consistent downward trend in the deciles, while the prescribing of the CCG in question stays more or less steady, meaning that it moves from around the 20th centile to the 70th centile. While it does seem much clearer, removing the seasonality could be viewed as ignoring an important part of the picture of antibiotic prescribing.
We have lots of ideas for new features on OpenPrescribing.net, but would like your help in prioritising them. Is this a feature you’d like to see on OpenPrescribing? Let us know at email@example.com.
We’ve had some really good feedback, and one suggestion we especially like is to try to show both smoothed and unsmoothed on one plot. Below I’ve added the actual prescribing for the CCG on top of the smoothed prescribing trend. This would need a clear explanation, but it shows a more complete picture of the prescribing.