Analytics, visualisation and machine learning of general practitioner prescribing using open health data

Student thesis: Doctoral Thesis

Abstract

This thesis examined open General Practitioner (GP) prescribing data from 2015 onward to investigate the nature of GP practices in Northern Ireland (NI). Contribution to knowledge is embodied in the linking of multiple open data sources to create a novel data set, the use of data analytics techniques and machine learning to develop a method for the categorisation of GP practices based on location and prescribing behaviours, the comparison of these categories to discover differences in prescribing behaviours and possible contributing factors. One unexpected factor, COVID-19, and the national lockdown changed prescribing behaviours, and this was also examined. Finally, people’s attitudes to the concept of citizen science via a GP prescribing dashboard was surveyed as a possible next step in making open data more accessible for anyone to analyse. It was found that whilst registered patients in NI had risen in line with population, the number of GP practices had fallen by 3.6% and comparing levels to that of other UK nations, NI had the highest prescribing in 6 of the 20 British National Formulary (BNF) chapters. The new method of categorisation employing machine learning clustering techniques found that two types of GP practice exist in NI, Metropolitan and Non-Metropolitan. Whilst they had similar prescribing patterns, prescribing levels were higher in half of the BNF chapters for Metropolitan practices with the largest variation being in the prescribing of Antidepressants and Analgesics. Possible factors contributing to the variations observed found a possible link to deprivation as a larger proportion of Metropolitan practices were in areas of high deprivation. The effects on prescribing due to the national lockdown showed a pattern of peak, trough, recovery. Antibiotic prescribing however did not recover to pre lockdown levels. Attitudes to citizen science were positive with 15.1% of participants contributing comments on resulting graphical output.
Date of AwardSept 2022
Original languageEnglish
SponsorsDepartment for the Economy
SupervisorMaurice Mulvenna (Supervisor) & Raymond Bond (Supervisor)

Keywords

  • Geolocation
  • K-means clustering
  • General practice
  • Prescriptions

Cite this

'