 Hello, everyone. Our next speaker is John McIntosh, an NHS data analyst based in Inverness, Scotland, a frequent vlogger who has the courage to do a live lightning talk. So take it away, John. Thank you. I'm going to talk about quality improvement data with R. So I'm based in Inverness and I work for a health board which covers a very large part of Scotland. And I have been supporting a national safety program for the last few years. And that covers several programs such as adult primary care, mental health, operating in about 30 hospitals, 100 GP practices and about 80 clinical ward areas. We use run charts to assess our process, our measures and how well we're progressing. They are simple and easy to construct. So they can be used at the front line. However, the information also needs to be collated and fed through the organisation. And that's where I came in. So right there, there's an example of a care bundle. It's got five steps. All of these have to be operating at about 95% reliability. And we use run charts to tell us if that's the case. So a run chart is a time series. The main thing is the median line here. So we take a baseline measurement of the start and we extend that into the future. We continue to plot our information and look for patterns in the data. I was struggling with about 400 charts. And one day I sat and mapped out the future. And I realised that spreadsheets were not going to cut it. So here we go. We have several thousand potential charts basically. And the analysis is quite complicated sometimes. So there are several rules that we apply. Shift, which is a run of six or more points above the line, or below the line. Trend, which is five or more increasing points. An astronomical data point or too few or too many runs. There are some critiques of these rules, however. The trend rule has been described as virtually useless. The astronomical data point is hard to programme, quite subjective. Too many runs will never usually be a sign of an improvement. It's usually two processes operating at once, like day shift and night shift. However, too few runs is fine. And that is the basis for the QI charts to package. So that leaves me with the shift rule. There are some challenges, both analytical and sort of from a BI reporting perspective. So I built something in R. This does everything I wanted to do. Calculates my medians, finds the runs and keeps on going. It uses data table and ggplot too, so it's pretty stable. And essentially it turns this into this, which is just what I wanted. Now, there's points in the median just to show how tricky this can be. If you're looking at the data, it's maybe not a part of what's going on. The run chart finds the run below the median, but just spread over about 15 points over 9. Doesn't require much to get a plot, just feed in your three variables, the direction you're looking for, your date, your grouping variable, and your y value. And you get the plot and the data table showing the baseline and then your sustained runs. If you don't like the colours, you can change them. Some people say you shouldn't do this, but I know that this data is genuine. I have colleagues actively bringing it out of the change, it motivates front-line teams and it helps pull out the signal from the noise. I'm also working on an SPC charter, so basically it's process control charts that are a bit more robust and useful for longer-term monitoring, especially for outcome and balancing measures. It will find a run and then rebase control limits. You can build p-charts for percentages, u-charts for rates, and c-charts for counts. You can change the trip parameters here to give solid lines rather than shades. Small multiples, very easy. There are some differences. I've learned a bit more about programming with data table. So SPC charter takes bare variable names as opposed to quoted ones in run charter, and there's also an additional grouping variable. Future plans. Adapt run charter to accept NSE. Adapt SPC charter to accept larger denominators, so basically create prime p and prime u-charts. Improve the testing. There's none yet for SPC charter, and then maybe CRAN at some point. That's it. That's into my talk. You can find my packages on GitHub, and you can find me on the internet, on these places. Well, thank you. We have time for a few questions. I have one. Do you use these charts alone, or is it a collaborative method, you know, work with teammates? How does a day go when you're using the system? I use these for analysis, so static reporting. Historically, they can be used within meetings, but it's mainly for reporting for organization. So I will use these to create PDFs, images, as well as to supplement our BI tools. Thank you. Well, thank you very much. Okay, thank you.