Login Help / FAQ

Blog

Louise Cook

2018-09-19 17:01

Where are the results?

Will Dixon, Professor of Digital Epidemiology at the University of Manchester and study lead for Cloudy with a Chance of Pain, explains what's been happening with all that data you gave us. We also give you a sneak preview of the results data map we are developing for the public.

Analysis of those 5.1 million symptom scores

We are now on the home straight of the Cloudy with a Chance of Pain analysis. This research will address the question of whether, and how, the weather affects pain. In this blog, I hope I can explain a little of what we have been up to over the last year.

Thanks to all of you dedicated participants, we have broken records for recruitment and ongoing engagement in a smartphone health research study. This wonderful effort and huge success has provided us with a mountain of rich, complex data. However, it has generated some considerable challenges in working out how to handle the data and getting to a final result.

Why is the analysis so complicated?

The first step in the analysis is 'cleaning' the data. Not with a dustpan and brush, or a mop, but making sure we make best use of all of the data that we have. We need to put it into a format that the computer can interpret. For example, when you signed up via the app, we asked you what health conditions you had. You could select from a pre-specified list such as rheumatoid arthritis or migraine, or you could tick 'other' then write what other condition you had. In this list, we received over 350 different conditions and their various misspellings. These then need to be grouped into recognized types of health issues ready for the analysis.

A second major challenge was getting high quality weather data linked to all of your symptom data. We had set up the study to provide hourly GPS data to link to the weather. We later discovered that different smartphones provided GPS data with different frequencies, some protecting the battery life by providing GPS less frequently. Also, some weather stations provided more complete data than others, so we needed to work out how to get the best possible local weather data for every participant. Building up a complete dataset took us months before we were even able to consider the analysis.

Growing the study team

Once we had a final dataset, we then needed to work out how the weather influenced pain. This requires us to try and spot relationships between weather that is changing hour-by-hour and day-by-day, and symptoms that also have a natural fluctuation from one day to the next. We need to consider the weather today, but also the weather yesterday and the day before, and perhaps the day before that. Then is it the temperature that is important, or the pressure, or the change in pressure since yesterday? Or is it a combination of different things within the weather? Several of our participants tell us their joints are worse when it is cold and damp. We also wanted to see whether other things might be influencing this relationship, for example will the effect of weather be different if you are inside compared to outside? Or is it different in different health conditions, for example does weather affect migraine in the same way as it does rheumatoid arthritis?

We wanted the best possible minds on the challenge, and so we have set up links with international experts in statistics, epidemiology, meteorology and machine learning. Technology allows us to hold video meetings remotely, but even so, this is not always easy. Our project meetings need to happen via computer links at 4pm so our colleagues in Vancouver can join at 8am their time, whilst our colleagues in Israel need to stay late to join at 7pm local time.

Hang on… are those results at last?

In the last month, we have started to see the first results. We have taken five different statistical approaches because, as one famous statistician said, “no models are perfect, but some are useful". All of the models are giving similar results, which gives us confidence that they are giving us useful information. There's still more to do before we have the final results, but it now feels in our grasp. We will of course keep you all informed about our progress and will let you know all that we discover. Thank you for contributing and for your ongoing patience. We're nearly there!