We harness the power of R and Python to create impactful open-source Data Science software. National Health Service (NHS) trusts receive thousands of patient feedback texts. The text contains invaluable information for managers and staff who are devoted to improving patient experience. The sheer volume of text records prohibits efficient processing by humans alone. We use Text Mining to revolutionize the analysis of patient feedback in two ways: to surface key information from the text, e.g. positivity/negativity in comments, burning issues for patients etc, and to classify texts by theme.

We use Python’s Scikit-learn, with its suite of fast Machine Learning algorithms and pre-processing tools for text classification. We do detailed Sentiment Analysis with tidytext (R) and TextBlob (Python). We summarize our complex findings in an interactive Shiny (R) dashboard that makes them accessible to non-technical staff. The dashboard uses the Golem framework to create a stable, maintainable, production-grade Shiny application that can be shipped as an R package and used with any NHS trust feedback dataset across the country. We demonstrate how we got the best of the R and Python worlds to make real change in the NHS.

Submission by:

• Andreas D Soteriades, Data Scientist, Nottinghamshire Healthcare NHS Foundation Trust.
• Chris Beeley, Senior Data Scientist, Nottinghamshire Healthcare NHS Foundation Trust.

Quite often I hear conversations about how R isn’t suited for production and how Python is the only way to go.

While Python has its strengths, if you make a model in R, you most definitely can go to production. In this talk, I share with you my experience of developing and deploying MakuSafe’s ML models - all powered by R. The EDA, model training & scoring, and final implementation are all powered by various R packages.

The “secret sauce” is packaging the essentials into an R package and deploying within a Docker container. This creates a reproducible and predictable environment that can run and yield the same results regardless of where it is deployed. And with this approach, we can update models and deploy within minutes!

What kind of activities might inspire a sixth-form student to consider data science as an exciting career path - and challenge their preconceptions (and misconceptions) of what “data science” is?

I will deliver a 10-minute overview of a new data science course for sixth-form students (ages 16-18) developed for their extended curriculum. Combining best DS practice from industry with the instructional expertise of teaching staff, the first instalment of the course engaged the students in data wrangling, R programming and modelling, with activities including prediction of the UK singles chart (with current data!), analysing and visualising the sentiment across the Harry Potter books (with a side helping of regular expressions), and using simulation in R to verify published results on the cost of completing a football sticker album.

The talk includes feedback from students and staff, along with recommendations for those looking to develop similar courses for others - whether sixth-formers or adult non-experts.

Since March 2020 we’ve been collecting, processing and publishing data and analysis on the COVID-19 pandemic in London.

At the start of the pandemic sub-national data was very patchy and hard for non-technical audiences to access. We wanted to provide information and tools so policy makers and the public could better understand the effect of the pandemic within London.

The process for delivering this has developed from trying to gather as much information from news stories and Wikipedia to an automated daily process sourcing data from PHE, NHS and ONS. We provide daily updates of detailed analysis, high level summaries for the main GLA website and social media and ad-hoc analysis for the mayoral, communication and health teams. The information covered includes data on cases, deaths, use of NHS services, vaccinations and changes in behaviour.

The whole process is carried out in R, from collecting the data to generating RMarkdowns for final outputs and sending out alert emails when data is updated.