Data and code to accompany the paper "Screening for in vitro systematic reviews: a comparison of screening methods and training of a machine learning classifier" published in Clinical Science.
All code is written in R using R Markdown documents. R version number and package version numbers are included in each script. Details of each script are below. Run each script in order.
- 1-1_data cleaning.Rmd: Prepare screening method comparison data for analysis; remove excluded data
- 1-2_calculate_performance.Rmd: Calaculate the performace (sensitivity and specificity) of screening methods at various thresholds
- 1-3_analyse_performance.Rmd: Plot the performance in a ROC curve and determine the optimal threshold for regex screening methods
- 1-4_search_term_retrieval_comparison.Rmd: Additional analysis comparing retrieval of studies using the planned vs actual search terms
Data are stored in the following folders:
- data-raw: raw data to be processed
- data: data which have undergone some processing
- data-analysis: final clean datasets
- data-ml_input: input data required to train ML
- data-ml_output: output data from ML
Machine learning (ML) functions are in the functions folder. Please not that the information required to configure the ML API are not included as we do not have permission to share this.
Plot outputs (in PDF file format) are in the figures folder.
- regex_histogram.pdf: histograms showing number of regex matches against (a) tiab and (b) full text
- screening_roc.pdf: figure from screening comparison part of project
- ml_roc.pdf: figure from machine learning part of project