Selection process
The outcome of the study was screened and selected using an open-source
machine learning (ML)-aided pipeline applying active learning: ASReview,
Active learning for Systematic Reviews (Van de Schoot et al., 2021).
ASReview is a tool that increases the efficiency of screening titles and
abstracts by determining prioritization with active learning. The
ASReview tool is extensively tested and validated and has shown to
achieve better performance in SR’s than manually evaluation titles and
abstracts (Van de Schoot et al., 2021). The tool was initially trained
for the current study with 10 relevant and 10 irrelevant publications
selected by two independent researchers (ARU & KvH). After feeding the
tool with the training publications, the tool returned the set of hits
ordered according to relevance priority. These results were checked by
the same two independent researchers. In case of several irrelevant
results among the top priority hits, the tool was further trained by
manually screening at least 1% of the total number of publications in
the whole set. Publications selected for further full-text review (n=99)
were those prioritized by ASReview. For each assigned publication,
authors checked each criterion and assessed the inclusion of only those
publications that met all criteria. Each publication was reviewed by a
second independent author following concordant and stratified criteria.
The full list of studies included for full-text review as well as the
inclusion and exclusion criteria can be consulted in Supplementary
Material; Table S2. For the selected publications (n=47), authors
annotated some additional publication details (e.g., country of the
study, participants included, disease area, factors affecting adherence
considered, study design, type of experimental design, etc.). The total
number of records after each screening round was documented using the
PRISMA flow diagram template (see Figure 1
below).
–Figure 1 should come around here –
Subsequently, the data related to the effect of interventions to
increase adherence were extracted from each study. These effects were
grouped according to the following dimensions: Socioeconomic, Treatment,
Condition, Personal, and Healthcare-related factors, which were based on
WHO’s 5 dimensions of treatment adherence (Sabaté, 2003). For each
adherence factor, both the inclusion and exclusion in each of the
reviewed studies were reported, as well as evidence of a significant
association of that specific factor with treatment adherence. Variables
related to the characteristics of the study, study sample, and study
intervention were also extracted.