An R package designed for the analysis of CPRD prescription electronic healthcare record (EHR) data.
rdrugtrajectory provides an API for analysing raw CPRD EHRs. Functions available include:
- loading & saving CPRD raw text files as R dataframes.
- creating mock CPRD data sets.
- works over the medical.txt and product.txt data files (part of the CPRD data dictionary - not supplied with RDrugTrajectory).
- retrieving patient IDs, gender, age and index of multiple scores.
- retrieve all prescriptions with a matching disease date.
- retrieve the first prescription event for a patient cohort.
- drug prescription incidence rate.
- a timeline generator using drug prescription data to use with Cox regression and survival curves (for packages such as REDA).
- drug prescription demographic data analysis.
- mean cumulative function plots of drug prescriptions.
- drug prescription incidence rates.
- many more functions.
Coming soon in version 0.2.2
- bug fixes (notably checking the quality of the data passed into each function).
- combining clinical and therapeutic consid (consultation IDs) to increase the reliability of the FirstDrugObject (FDO) results.
- the ability to subset a FirstDrugObject.
- Event (medical/therapeutic) covariate table building - a binary matrix to indicate whether an event for a patient during a time period was recorded.
- interrupted time series analysis for CPRD data.
- segmented regression analysis for CPRD data.
- HTML time series (uses the R packages timevis and htmltools) plots of patient medical history (example contains fabricated data).
CPRD electronic healthcare records are delivered as flat text files. Researchers have two choices, pay a lot of money to use existing tools or write tools of their own. Having faced this problem myself I started developing my own tools three years ago.
This R package came about as the result of supervising a very talented post-graduate Clinical Neuroscience student (please see Credits). The student's project, titled "A Longitudinal Cohort Study of Migraine Preventative Medication Usage in UK CPRD" required immediate access to primary care clinical and prescription records of patients suffering with headache disorders. The CPRD data can be very overwhelming, especially to those with little experience in electronic healthcare records or the manipulation of large data sets. To mitigate this concern I decided to develop a drug-prescription themed R API. Whilst I was writing the R package my student was putting it to use, performing the analysis necessary to answer the research aims whilst reporting bugs and suggesting additional features. My goal is to build an R package that can interrogate CPRD records whilst requiring minimal R language experience from the user (although some experience goes a long way).
The data attached to the R package, images and examples presented here, and the images and examples in the accompanying publication (expected soon), is fabricated and does not in any part represent real patient data. Our accompanying researech articles have ISAC (Independent Scientific Advisory Committee - UK Gov) approval.
Under pre-alpha stage until initial release - expected, early December 2020. Learn more about software life-cycles here. Whilst in active development, I cannot guarantee the results and all responsibility rests with the user.
Example results using fabricated electronic healthcare records.
Plotting the number of patients by their first drug prescription (Figure 1) is achieved by simply accessing the frequency data.frame in the FirstDrugObject list-type. The first element holds a named-list (by the drug ID) of patient id vectors, the second entry reflects the first entry in structure but holds the event dates, and the third element is a data.frame of drug by number of first prescriptions on record. With all of this information, one is able to analysis each drug type by social-demographic factors (functions provided).
Figure 1. First drug prescription (matched to a disease event e.g., headache) frequency.
rdrugtrajectory can produce the data structure necessary to perform a mean cumulative function (MCF) over several groups. Having first filtered the cohort for only medication of interest, and MCF plot reveals prescription burden by social-demographic factors, for example by social deprivation scores (Figure 2), or whole cohort populations.
Figure 2. Cumulative drug prescriptions stratified by patient IMD score.
The first codedraft of rdrugtrajectory is able to plot the change in a patients medication starting from their first prescription (Figure 3). Later releases will factor in time.
Figure 3. Prescription drug changes. From first drug prescription matched with disease. The example does not take into account time.
Whilst in alpha development phase, please download a release and install locally. The installation happens in two steps:
(1) Install all dependencies first. For example, install.packages("foreach"). The dependencies are:
plyr, dplyr, foreach, doParallel, data.table, parallel, splus2R, rlist, reda, ggplot2, ggalluvial, stats, utils, useful
(2) Download the latest release of rdrugtrajectory and install using:
install.packages("path/to/tar/file", source = TRUE, repos=NULL)
library(rdrugtrajectory)
Ideally you should also have a copy of the medical.txt and product.txt files that can be found inside the Windows installation of the CPRD data dictionary. These files are used to name products (e.g., drugs) and medical terms. In their absence the R package will still work, however, the medcode and prodcode entries will remain coded. Please read the instructions on how to structure your data.
The forth coming publication will contain a number of examples, for now, please check out the reference manual.
The conventional R code style with original Java/C++ code blocks (showing my age). I do not use the R dot-notation for naming variables and I have avoided the %>% infix notation where possible.
Please head over to the wiki page for scenario based tutorials.
Please see the available reference manual and forth coming publication.
There are basic unit tests one can run. Instruction will be added here in due course.
Please raise an issue for bug fixes or suggested features/improvements. Please note: the code is only maintained by myself (Anthony Nash). Unfortunately, issues won't be resolved overnight! I have students to supervise, fellowships and papers to write and my own research to be getting on with. Please be patient.
Dr Anthony Nash PhD, University of Oxford, Nuffield Department of Clinical Neurosciences - Design, development, testing, research, supervisor.
Tingyee Chang MSc, University of Oxford, Nuffield Department of Clinical Neurosciences - Testing and research.
Benjamin Wan, King's College London, Pharmacology and Therapeutics - Testing.
Dr Zameel Cader DPhil, University of Oxford, Nuffield Department of Clinical Neurosciences - Group lead.
We are grateful to the Oxford Science Innovation, NIHR Oxford Biomedical Research Centre and NIHR Oxford Health Biomedical Research Centre (Informatics and Digital Health theme, grant BRC-1215-20005) for funding. The views expressed are those of the authors and not necessarily those of the UK National Health Service, the NIHR, or the UK Department of Health.
MIT © Anthony Nash 2020