by Hannah Johns

When I tell people that I work in statistics, the usual response is an eyes-glazed stammering about how complicated it sounds. Even among researchers, a lot of what people like myself do seems to come across like a whole lot of numerical wizardry, which can only be understood by selling your soul to the gods of MS Excel.

Despite what any researcher wants, the data that we get from clinical trials is complicated, and often messy. Even if we never had missing data (we will always have missing data), the factors which impact on things like patient outcomes are complicated and interact with one another. This has especially been the case with trying to understand the data from the AVERT clinical trial.

AVERT demonstrated that early and intense physiotherapy interferes with recovery, an unexpected finding which will change clinical guidelines for the better. But this finding is only part of the story, and there’s more to be learned from the data. Unfortunately, the data we collected in AVERT is complicated. Instead of receiving a conceptually simple intervention (e.g. either receiving an experimental drug or a placebo), stroke rehabilitation is characterised across multiple factors:

  • rehabilitation intensity can vary across a number of days,
  • therapy can be spread out across a different number of sessions, and
  • a patient might get better or worse during rehabilitation, which might change how their therapy is delivered.

All of this complexity means that on the statistics side of things, we have to resort to the sort of techniques which give statisticians their reputation for being number wizards. This can make results harder to interpret, which makes it harder to improve treatment guidelines and patient outcomes. So how can we get around this?

For the AVERT data at least, the answer to this is data visualisation. Lots and lots of data visualisation. One of the things that has been incredibly useful in further analysis of the AVERT dataset has been the development of a tool we’ve been calling the AVERT Atlas, an app for visualising patient data.

Each patient is represented by a tile that shows all information about their therapy stay:

  • colour shows a patient’s health state, and
  • bar height shows therapy intensity across days.

Each tile is also annotated with additional information about patient health and therapy received. These tiles can be arranged next to each other, allowing researchers to sift through the data to find patterns. On top of that, if a researcher wants to examine a specific subgroup of patients (e.g. older or younger, smokers or non-smokers), they can easily filter out patients to just match those relevant to their questions. Researchers can identify patterns among patients by simply looking at the data.

As we continue to learn from the data gathered in AVERT, data visualisation helps to make sure that new discoveries are understood in terms of the patient, without getting lost in the details. This means that clinical researchers can focus on what is important: Improving outcomes for stroke survivors.