A question of trust: can we build an evidence base to gain trust in systematic review automation technologies?
Open accessThe DOI matches its Crossref record; 38 of the 38 references it cites were found in a registry or an index; 5 of the 21 sources its claims cite were checked for support.
Abstract
BACKGROUND: Although many aspects of systematic reviews use computational tools, systematic reviewers have been reluctant to adopt machine learning tools. DISCUSSION: We discuss that the potential reason for the slow adoption of machine learning tools into systematic reviews is multifactorial. We focus on the current absence of trust in automation and set-up challenges as major barriers to adoption. It is important that reviews produced using automation tools are considered non-inferior or superior to current practice. However, this standard will likely not be sufficient to lead to widespread adoption. As with many technologies, it is important that reviewers see "others" in the review community using automation tools. Adoption will also be slow if the automation tools are not compatible with workflows and tasks currently used to produce reviews. Many automation tools being developed for systematic reviews mimic classification problems. Therefore, the evidence that these automation tools are non-inferior or superior can be presented using methods similar to diagnostic test evaluations, i.e., precision and recall compared to a human reviewer. However, the assessment of automation tools does present unique challenges for investigators and systematic reviewers, including the need to clarify which metrics are of interest to the systematic review community and the unique documentation challenges for reproducible software experiments. CONCLUSION: We discuss adoption barriers with the goal of providing tool developers with guidance as to how to design and report such evaluations and for end users to assess their validity. Further, we discuss approaches to formatting and announcing publicly available datasets suitable for assessment of automation technologies and tools. Making these resources available will increase trust that tools are non-inferior or superior to current practice. Finally, we identify that, even with evidence that automation tools are non-inferior or superior to current practice, substantial set-up challenges remain for main stream integration of automation into the systematic review process.
Overview
CoChat Summary
TL;DR
This commentary argues that slow adoption of machine learning/automation in systematic reviews is multifactorial, with lack of trust and set-up/workflow compatibility as major barriers. It proposes that automation tools need evidence of non-inferiority or superiority, better reporting, shared datasets, and credible early adopters, while noting integration challenges remain.
Key findings
- Adoption of higher-level automation remains rare: most machine-assisted screening is Level 2, where the tool reorders citations by relevance probability but the human reviewer still makes the final decision on all citations.
- At the time of writing, 159 software tools were indexed on the systematic review toolbox website, but it is unclear how many can seamlessly fit each unique review team workflow.
- Of 33 studies reporting approaches to automated screening, no study provided all criteria for reproducible reporting, according to the cited review by Olorisade et al.
- The authors propose that automation assessments can be framed like diagnostic test evaluations, using metrics such as precision, recall, F1, sensitivity, specificity, AUC, and time-saved measures, and that shared validated datasets and early adopters are needed to build trust.
Methods
- Discussion/commentary paper, not an empirical study; it applies diffusion of innovations theory to systematic review automation adoption, focusing on trust and set-up challenges.
- It draws on examples such as the Cochrane Schizophrenia Group workflow described by Knight et al. and an electric car analogy, then proposes study designs, outcome metrics, reporting standards, and dataset-sharing formats.
- No sample, data collection, or statistical analysis is reported.
Limitations
- The authors state that some barriers require cultural change and that how to effect that change is beyond the scope of the manuscript; they focus on trust and set-up challenges.
- The authors frame the barriers as hypotheses (“We hypothesize”) and propose approaches rather than empirically validating them.
- Evident from the text: the paper presents no primary data or formal systematic review, so its recommendations are conceptual and depend on future studies and community adoption.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
2 Valid · 4 Not valid · 10 Not checked · 42 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“For example, despite numerous studies that are over a decade old documenting the use of machine learning approaches to screening citation records, this technology remains rarely used in peer-reviewed systematic reviews [15–20].”
§ A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? > Background
What the sources say
- Not checked yetcites [15] Aphinyanaphongs Y, Tsamardinos I, Statnikov A, Hardin D, Aliferis CF. Text categorization models for high-quality article retrieval in internal medicine. J Am Med Inform Assoc. 2005;12:207–16. source
- Partly supportedcites [16] O'Mara-Eves A, Thomas J, McNaught J, Miwa M, Ananiadou S. Erratum to: using text mining for study identification in systematic reviews: a systematic review of current approaches. Syst Rev. 2015;4:5. sourceExperimental — accuracy not yet measured
The Results (Chronological developments; page not provided) identify 44 studies from 2006–2014 evaluating text mining for screening. The Results (Deployed systems; page not provided) say only six systems were available in a packaged form, supporting limited practical availability, but the paper does not establish how rarely the technology was used in peer-reviewed systematic reviews. Nor does its 2015 account establish that numerous studies were already over a decade old. (LLM, conf=0.85)
- Partly supportedcites [17] O'Mara-Eves A, Thomas J, McNaught J, Miwa M, Ananiadou S. Using text mining for study identification in systematic reviews: a systematic review of current approaches. Syst Rev. 2015;4:5. https://doi.org/10.1186/2046-4053-4-5 . sourceExperimental — accuracy not yet measured
The Results (Chronological developments; page unavailable) identify 44 studies of text mining for screening published from 2006 to 2014. The Discussion (Summary of key findings; page unavailable) reports few deployed systems, supporting limited practical uptake, but it does not establish how often machine learning was used in peer-reviewed systematic reviews. The paper also does not characterize the studies as over a decade old. (LLM, conf=0.82)
- Not checked yetcites [18] Bekhuis T, Demner-Fushman D. Towards automating the initial screening phase of a systematic review. Stud Health Technol Inform. 2010;160(Pt 1):146–50.
- Partly supportedcites [19] Wallace BC, Trikalinos TA, Lau J, Brodley C, Schmid CH. Semi-automated screening of biomedical citations for systematic reviews. BMC Bioinformatics. 2010;11:55. https://doi.org/10.1186/1471-2105-11-55 . sourceExperimental — accuracy not yet measured
The Abstract (Results) documents this 2010 paper’s machine-learning approach to screening citations, so it supports the claim that such studies existed over a decade ago. The Background (On Systematic Reviews) describes manual screening, but the paper does not establish that machine-learning screening remains rarely used in peer-reviewed systematic reviews. (LLM, conf=0.85)
- Not checked yetcites [20] Hemens BJ, Iorio A. Computer-aided systematic review screening comes of age. Ann Intern Med. 2017;167:210. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Further, with increasing movement toward living reviews, such efforts will require technological advances in order to be sustainable [11, 12].”
§ A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? > Background
What the sources say
- Partly supportedcites [11] Vandvik PO, Brignardello-Petersen R, Guyatt GH. Living cumulative network meta-analysis to reduce waste in research: a paradigmatic shift for systematic reviews? BMC Med. 2016;14:59. sourceExperimental — accuracy not yet measured
In “Background > Living cumulative NMA” (page unavailable), the paper says improved tools and health information technology could make living network meta-analyses easier to produce and update. It does not say technological advances are required for sustainability, and its discussion is narrower than living reviews generally. (LLM, conf=0.84)
- Supportedcites [12] Elliott JH, Turner T, Clavisi O, Thomas J, Higgins JP, Mavergames C, Gruen RL. Living systematic reviews: an emerging opportunity to narrow the evidence-practice gap. PLoS Med. 2014;11:e1001603. source
The Production of Living Systematic Reviews section calls improved production efficiency a key enabler of living reviews and describes technological developments that could make their widespread production feasible. The Bridge from Evidence to Practice section says new technologies can resolve the trade-off between rigorous and up-to-date reviews. Together, these support the attributed need for technological advances to sustain living reviews. (LLM, conf=0.90)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
In the paper
“Currently, few systematic review teams use automation technology to take over all, or some, of the cognitive tasks or to take on a higher level of decision-making or interpretation (Levels 3 and 4) [14].”
§ A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? > Background
What the sources say
- Not checked yetcites [14] Knight I, Wilson M, Brailsford D, Milic-Frayling N. “Enslaved to the trapped data”: a cognitive work analysis of medical systematic reviews. In: In Proceedings of 2019 ACM SIGIR Conference on Human Information Interaction and Retrieval. Glasgow: CHIIR2019; 2019. 10–14 March 2019: 10 pages.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, the evidence that these automation tools are non-inferior or superior can be presented using methods similar to diagnostic test evaluations, i.e., precision and recall compared to a human reviewer.”
§ A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? > Abstract > Discussion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“Although automation tools capable of Level 3 and Level 4 tasks are rapidly becoming available for systematic reviewers, surveys suggest that adoption of these automated technologies by the systematic review community is slow [7].”
§ A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? > Background
What the sources say
- Not checked yetcites [7] Al-Zubidy A, Carver JC, Hale DP, Hassler EE. Vision for SLR tooling infrastructure: prioritizing value-added requirements. Inf Softw Technol. 2017;91:72–81. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
11 more claims
Create a free account to see all 58 claims and to verify the 16 references still open.
- Probably not a substantial claim. 36
- It may state two separate claims as one. 20
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“They are particularly protracted processes (Borah et al., 2017; O’Connor et al., 2019)”
From SYMBALS: A Systematic Review Methodology Blending Active Learning and Snowballing (2021) · cites this paper as O’Connor et al., 2019
The cited source says what the claim attributes to it.
In the citing paper
“The lack of trust in systematic review automation technologies (O’Connor et al., 2019) is not fully solved by SYMBALS.”
From SYMBALS: A Systematic Review Methodology Blending Active Learning and Snowballing (2021) · cites this paper as O’Connor et al., 2019
The cited source says what the claim attributes to it.
In the citing paper
“Despite numerous tools for automation, adoption of these tools has lagged [8, 22].”
From Machine learning for screening prioritization in systematic reviews: comparative performance of Abstrackr and EPPI-Reviewer (2020) · cites this paper as [22]
The cited source says what the claim attributes to it.
In the citing paper
“Reasons for slower uptake may include pragmatic challenges from incorporating tools into current workflow, and lack of trust by the systematic review community [22].”
From Machine learning for screening prioritization in systematic reviews: comparative performance of Abstrackr and EPPI-Reviewer (2020) · cites this paper as [22]
In the citing paper
“While AI may not be ready to fully replace human screeners in the task of study selection, studies suggest that optimizing, accelerating, and reducing screening burden through the use of AI-informed screening methods represents a viable option.”
From An evaluation of DistillerSR’s machine learning-based prioritization tool for title/abstract screening – impact on reviewer-relevant outcomes (2020) · cites this paper as [25]
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? > References
“16. O'Mara-Eves A Thomas J McNaught J Miwa M Ananiadou S Erratum to: using text mining for study identification in systematic reviews: a systematic review of current approaches Syst Rev. 2015 4 5 10.1186/2046-4053-4-5 25588314 PMC4320539”
§ A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? > References
“17. O'Mara-Eves A, Thomas J, McNaught J, Miwa M, Ananiadou S. Using text mining for study identification in systematic reviews: a systematic review of current approaches. Syst Rev. 2015;4:5. 10.1186/2046-4053-4-5. 10.1186/2046-4053-4-5 PMC4320539 25588314”
Citations
Cites
Perceived usefulness, perceived ease of use, and user acceptance of information technology
Davis, F. D. (1989) · MIS Quarterly · doi:10.2307/249008 · Cited by 68,483
Rogers, E. M., Singhal, A., & Quinlan, M. M. (2019) · An Integrated Approach to Communication Theory and Research · doi:10.4324/9780203710753-35 · Cited by 37,322
Valente, T. W. (2003) · Genetics in Medicine · doi:10.1097/01.gim.0000061743.67794.c4 · Cited by 13,723
STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies
Bossuyt, P. M., Reitsma, J. B., Bruns, D. E., Gatsonis, C. A., Glasziou, P. P., Irwig, L., Lijmer, J. G., Moher, D., Rennie, D., de Vet, H. C. W., et al. (2015) · BMJ · doi:10.1136/bmj.h5527 · Cited by 3,660
Self-citationBrereton, P., Kitchenham, B. A., Budgen, D., Turner, M., & Khalil, M. (2007) · Journal of Systems and Software · doi:10.1016/j.jss.2006.07.009 · Cited by 2,394
Ten Simple Rules for Reproducible Computational Research
Sandve, G. K., Nekrutenko, A., Taylor, J., & Hovig, E. (2013) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1003285 · Cited by 880
Borah, R., Brown, A. W., Capers, P. L., & Kaiser, K. A. (2017) · BMJ Open · doi:10.1136/bmjopen-2016-012545 · Cited by 737
O’Mara-Eves, A., Thomas, J., McNaught, J., Miwa, M., & Ananiadou, S. (2015) · Systematic Reviews · doi:10.1186/2046-4053-4-5 · Cited by 681 · cited 1×
“relaxed variants in natural language processing (NLP) tasks: these and other summary measures of relevance have been described elsewhere [17].”
BackgroundSelf-citationThe Use of Research Evidence in Public Health Decision Making Processes: Systematic Review
Orton, L., Lloyd-Williams, F., Taylor-Robinson, D., O'Flaherty, M., & Capewell, S. (2011) · PLoS ONE · doi:10.1371/journal.pone.0021704 · Cited by 599
Living Systematic Reviews: An Emerging Opportunity to Narrow the Evidence-Practice Gap
Elliott, J. H., Turner, T., Clavisi, O., Thomas, J., Higgins, J. P. T., Mavergames, C., & Gruen, R. L. (2014) · PLoS Medicine · doi:10.1371/journal.pmed.1001603 · Cited by 596
Self-citation
Research works printed but not linked: 1
- Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 [updated March 2011]. (Higgins J, Green S eds.): The Cochrane Collaboration; 2011.
Cited by
Automation of systematic literature reviews: A systematic literature review
van Dinter, R., Tekinerdogan, B., & Catal, C. (2021) · Information and Software Technology · doi:10.1016/j.infsof.2021.106589 · Cited by 432
Artificial intelligence for literature reviews: opportunities and challenges
Bolaños, F., Salatino, A., Osborne, F., & Motta, E. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-024-10902-3 · Cited by 246
“…compatibility with other software systems [148, 149], iii) distrust , as there is uncertainty about the reliability and the mechanisms of these tools [150, 151], and iv) financial obstacles , predominantly arising from licensing expenses, along with feature restrictions in trial versions [152].”
Error rates of human reviewers during abstract screening in systematic reviews
Wang, Z., Nayfeh, T., Tetzlaff, J., O’Blenis, P., & Murad, M. H. (2020) · PLOS ONE · doi:10.1371/journal.pone.0227742 · Cited by 217
“[19, 24] It is intuitive to assume that these tools should achieve a zero error rate in order to be implemented to generate evidence used for decision-making (i.”
BackgroundMethodChai, K. E. K., Lines, R. L. J., Gucciardi, D. F., & Ng, L. (2021) · Systematic Reviews · doi:10.1186/s13643-021-01635-3 · Cited by 190
“For such a tool to be considered seriously by researchers, clinicians and policy-makers, they require evidence to support the effectiveness of the semi-automation tool and the potential time savings associated with its adoption [28].”
BackgroundScoping Reviews, Systematic Reviews, and Meta-Analysis: Applications in Veterinary Medicine
Sargeant, J. M., & O'Connor, A. M. C. (2020) · Frontiers in Veterinary Science · doi:10.3389/fvets.2020.00011 · Cited by 198
Self-citation
Metadata
- Authors
- O’Connor, A. M., Tsafnat, G., Thomas, J., Glasziou, P., Gilbert, S. B., & Hutton, B.
- Affiliations
- Macquarie University (AU)
- University College London (GB)
- Bond University (AU)
- Iowa State University (US)
- Ottawa Hospital Research Institute (CA)
- Venue
- Systematic Reviews
- Published
- 2019-06-18
- Volume
- 8
- Issue
- 1
- Pages
- 143-143
- ISSN
- 2046-4053 (ISSN-L)
- PMC
- PMC6582554
- OpenAlex
- W2953029151
- Cited by
- 167
- References
- 38
- Citations by year
- 2026: 19 · 2025: 23 · 2024: 29 · 2023: 18 · 2022: 14 · 2021: 28 · 2020: 31 · 2019: 5
- FWCI
- 3.95
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0/ (tdm)
- creativecommons.org/licenses/by/4.0/ (vor)
- Locations
- Systematic Reviews (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Bond University Research Portal (Bond University) (submittedVersion, cc-by)
- Iowa State University Digital Repository (Iowa State University) (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- UCL Discovery (University College London) (submittedVersion)
- uO Research (University of Ottawa) (submittedVersion, public-domain)
- Funders
- U.S. Environmental Protection Agency
- University College London
- Medical Research Council (MR/J005037/1)
- National Institute of Environmental Health Sciences
- Topics
- Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
- Physical Sciences › Computer Science › Information Systems › Software Engineering Techniques and Practices
- Social Sciences › General Social Sciences › Computational and Text Analysis Methods
- Keywords
- systematic review automation, machine learning tools, technology adoption, trust in automation, precision and recall, diagnostic test evaluation, benchmark dataset
- MeSH terms
- Machine Learning, Systematic Reviews as Topic, Humans, Reproducibility of Results, Trust, Evidence-Based Practice
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-04, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-04 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? https://app.cochat.ai/papers/10.1186/s13643-019-1062-0
- The paper
- O’Connor, A. M., Tsafnat, G., Thomas, J., Glasziou, P., Gilbert, S. B., & Hutton, B. (2019). A question of trust: can we build an evidence base to gain trust in systematic review automation technologies? Systematic Reviews. https://doi.org/10.1186/s13643-019-1062-0
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.