Loading...
Loading...
Contributors:
Dataset URI:
10.5281/zenodo.23651698Data used in every analysis of a systematic review and meta-analysis of 105 randomised trials of interactive automated voice and text systems (interactive voice response, scripted messaging with replies, conversational agents, and chatbots built on large language models) for medication adherence and self-management in adults with chronic disease (PROSPERO CRD420261495582): effects by trial and outcome domain with arm-level data, trial-level coding, study attrition, use of the system, deaths, the within-trial contrast of dialogue against the same messages, and sensitivity variants; risk-of-bias judgements (RoB 2); trustworthiness judgements (INSPECT-SR); decisions of each screening pass; excluded and non-retrieved reports; the counts of the flow diagram; the full list of departures from the protocol; the language model used at each step; the blind coding of who closes the loop, with the location and quote of each code; the arm-level data and data-handling notes; the coding of autonomy used in the moderator tests; the statistical analysis plan, in Portuguese with an English translation; the codebook; the instructions given to the language models; an R script that reproduces the main results of the article from the dataset alone; the full extraction dataset and the exact inputs of every model; the outputs of the final statistical round, with the leave-one-out analysis of every outcome domain, the trial sequential analysis, and study attrition by who closes the loop; the tables, figures, and methods removed from Supplement 1 of the article; the scripts of the review as run; and the working folders of the search rounds of Oct 23–24, 2026.Aggregate data from published reports only; no individual participant data. The contents and the instructions for reproduction are in README.md. Data and documentation are licensed under CC BY 4.0; the code is licensed under the MIT licence (LICENSE-code.txt).Version 10 (October 25, 2026): the set of trials is unchanged (105 trials in 116 reports). The main analyses include all eligible trials, as registered (departure No 118); the exclusion of the seven trials with serious trustworthiness concerns (INSPECT-SR, reported only in this deposit) is a sensitivity analysis, and every analysis was run again in one round on Oct 25, 2026. Objectively measured adherence SMD 0·22 (0·09 to 0·35) (26 trials); self-management 0·21 (0·09 to 0·33). New: the full text of all 122 departures from the protocol (Nos 118 to 122 added), the outputs of the round of Oct 25 (results/synthesis_20261025/: leave-one-out of every outcome domain, trial sequential analysis, study attrition by who closes the loop), the material removed from Supplement 1 (docs/supplement1_moved/: tables, forest plots in standardised mean difference of HbA1c and systolic blood pressure, methods of the trustworthiness assessment), the albatross plots of every outcome domain and the use of the system (docs/appendix_full/), the scripts of Figures 3 and 4 (code/project_scripts/figures/), and the current Supplement 2 (docs/language_model_use.md). The R script reproduces 43 of the 43 published values checked. The main estimates are given in README.md.Note on the use of full texts. The language models were run under consumer terms of service with the setting that allows inputs to be used for model training switched off, so inputs were not used to train the models. Full-text reports, obtained through institutional access, were processed only to support the review, were not redistributed, and are not part of this deposit.
Keyword(s):
blind coding; chatbot; chronic disease; conversational agent; interactive voice response.
MeSH Term(s):
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
Supplementary Data:
No Supplementary Data Available
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)