Deident: An R package for data anonymization

The delivery of quality health care is a constant act of balancing demand against capacity, with emerging, data intensive, artificial intelligence (AI) and machine learning (ML) approaches poised to bridge gaps in the large, resource-limited sector (Harwich & Laycock, 2018; Nelson et al., 2019; Wilson, 2019; Yu et al., 2018). The scale of data required in such projects magnifies the importance of existing ethical and legal frameworks for research with human participants (Sales & Folkman, 2000; UKRI, 2022), notably around the risks posed by the processing of personally identifiable data (PID) and pseudo-PID (variables which if used together can identify
an individual) (ICO, 2023).
One approach to dealing with PID concerns is to apply transformations to the data, e.g. encryption of names, or aggregation of ages, which can limit the risk of identification at the cost of nuance (Tachepun & Thammaboosadee, 2020). Hence, we demonstrate an extendable
package of tools for the implementation and application of deidentification techniques to panel data sets.