On 26 of September 2019, the RSS local group in Glasgow hosted an event 'Transforming Health and Social Care Publications in Scotland' by two members of the Information Services Division (ISD) of National Services Scotland, part of NHS Scotland. The event was attended by 31 people from all universities in Glasgow as well as some members of industry.
The main focus of the talks was the Transforming Publishing Programme (TPP), which is modernising the presentation of data analysis at ISD. The TPP vision focuses on: engaging people and thus, encouraging them to come back for more; accessible content; connected content to make clear how the ISD data fits together; comparable data across different areas; insightful content to aid the understanding of trends in the data and objective content which can be trusted.
In the first part of the event, data visualiser Jack Hannah presented on the use of RShiny applications. The RShiny apps were inspired by the fact that ISD regularly produces pdfs of around 70 pages which are difficult to parse if one is interested in a specific piece of information. The apps allow for comparability, granularity of the data (as much as possible) and that the data are free from bias (political or otherwise). R Shiny apps and D3 are introduced as part of the shift away from using SPSS and Excel to R. Furthermore, GitHub is used for version control and peer review. As much as possible, data are released as open data which results in fewer emails to the analysts and helps them focus on further analysis.
However, not all data could be displayed using R Shiny apps because they are hosted on a server in the USA rather than locally. An example was given on Mental Health Inpatient Activity. The initial app took six months to develop and has been added to over time.
Not all pages are created using RShiny. The landing page provides a data summary (which, for instance, can be used by non-specialists such as journalists), data trends, data explorer, data files, glossary of terminology and questions. The app shows the major trends (eg: geographical, diagnostic groupings and regions) in the data. Data can be downloaded from the app; however plots and summaries are interactive and can be displayed at a granular level. People who use screen readers would struggle too as screen readers do not work properly with the app as the screen readers tend to read the code underlying the charts. Further limitations are a result of two scrollers on the same page which are hard to differentiate.
It should be noted that the apps are created in a manner that allows them to be use by those who are colour-blind and can be navigated without using the mouse. The next developments are mostly related to moving the servers to ISD in order to be able to use the apps for confidential data and spending more time on user research (currently limited to ISD and collaborators from academia).
The second talk was given by data wrangler David Caldwell who presented on Reproducible Analytics Pipelines (RAP). The team developed a seven-level classification of RAP with teams encouraged to reach at least level 4:
- Ad hoc R code
- R project
- R project under version control (VC)
- a) R project under VC and peer reviewed (wrangling)
b) Replicable report in R markdown (publication)
- Near RAP (as above plus data quality assurance and package management)
- Full RAP (as above plus unit testing and documentation)
- R package.
As proof of concept, David attempted to reach level seven using the quarterly Hospital Standardised Mortality publication which was successful. Published in August 2019, it consists of 40-page PDF, a two-page summary and seven Microsoft Excel tables. After publication, the code was made available on GitHub, which helps with transparency and makes it easier to provide and acquire feedback. The final product is available in the R package HSMR, which includes 17 functions.
The processes of producing the report has been completely automated, which has eliminated prone errors that can occur when producing the report manually. The results were compared to the SPSS output to ensure the methodology had been correctly translated into R. The main benefit in creating the package is that a report which took seven days to create now takes 50 minutes. This has allowed the team to pursue other projects. Currently, David hopes that the RAP process will expand to more teams using the buddy systems. Another publication produced using RAP is the Scottish Bowel Screening Programme report.
The talks were very informative and resulted in a lively discussion on the benefits of using Shiny apps. The two speakers were very open and honest about shortfalls and received a number of suggestions on different approaches to address some of the current issues.
The most important take away was the free availability of data from the R Shiny apps. Pictures from the talks are available on the Twitter page of the local RSS group @RSSGlasgow1.