Travels with R and Python

the power of data science in healthcare

Chris Beeley

Aug 2, 2023

What is data science?

• “A data scientist knows more about computer science than the average statistician, and more about statistics than the average computer scientist”

(Josh Wills, a former head of data engineering at Slack)

Drew Conway’s famous Venn diagram

Source

What are the skills of data science?

• Analysis
  • ML
  • Stats
  • Data viz
• Software engineering
  • Programming
  • SQL/ data
  • DevOps
  • RAP

What are the skills of data science?

• Domain knowledge
  • Communication
  • Problem formulation
  • Dashboards and reports

Stats and data viz

• ML leans a bit more towards atheoretical prediction
• Stats leans a bit more towards inference (but they both do both)
• Data scientists may use different visualisations
  • Interactive web based tools
  • Dashboard based visualisers e.g. {stminsights}

Software engineering

• Programming
  • No/ low code data science?
• SQL/ data
  • Tend to use reproducible automated processes
• DevOps
  • Plan, code, build, test, release, deploy, operate, monitor
• RAP
  • I will come back to this

Domain knowledge

• Do stuff that matters
  • The best minds of my generation are thinking about how to make people click ads. That sucks. Jeffrey Hammerbacher
• Convince other people that it matters
• This is the hardest part of data science

RAP

• Data science isn’t RAP
• RAP isn’t data science
• They are firm friends

Reproducibility

• Reproducibility in science
• The $6B spreadsheet error
• George Osbourne’s austerity was based on a spreadsheet error
• For us, reproducibility also means we can do the same analysis 50 times in one minute
  • Which is why I started down the road of data science

What is RAP

• a process in which code is used to minimise manual, undocumented steps, and a clear, properly documented process is produced in code which can reliably give the same result from the same dataset
• RAP should be:

the core working practice that must be supported by all platforms and teams; make this a core focus of NHS analyst training

Levels of RAP- Baseline

• Data produced by code in an open-source language (e.g., Python, R, SQL)
• Code is version controlled
• Repository includes a README.md file that clearly details steps a user must follow to reproduce the code
• Code has been peer reviewed
• Code is published in the open and linked to & from accompanying publication (if relevant)

Levels of RAP- Silver

• Code is well-documented…
• Code is well-organised following standard directory format
• Reusable functions and/or classes are used where appropriate
• Pipeline includes a testing framework
• Repository includes dependency information (e.g. requirements.txt, PipFile, environment.yml)
• Data is handled and output in a Tidy data format

Levels of RAP- Gold

• Code is fully packaged
• Repository automatically runs tests etc. via CI/CD or a different integration/deployment tool e.g. GitHub Actions
• Process runs based on event-based triggers (e.g., new data in database) or on a schedule
• Changes to the RAP are clearly signposted. E.g. a changelog in the package, releases etc. (See gov.uk info on Semantic Versioning)

Data science in healthcare

• Forecasting
  • Stats versus ML
• Text mining
  • R versus Python
• Demand modelling
  • DevOps as a way of life

Get involved!

• NHS-R community
  • Webinars, training, conference, Slack
• NHS Pycom
  • ditto…
• MLCSU GitHub?
• Build links with the other CSUs

Contact

• strategy.unit@nhs.net
• The-Strategy-Unit

• chris.beeley1@nhs.net
• chrisbeeley