Table 1.
Review protocol, compliant with PRISMA-P (Preferred Reporting Items for Systematic review Protocols 1.
| Methods | Item No. | Checklist Item |
|---|---|---|
| Eligibility criteria | 8 | Only original articles and reviews in English covering dental topics with objective AI utilization were eligible. On 19 April 2022, PubMed, Scopus, Web of Science—Core Collection, and Google Scholar were queried. The focus for the 1st objective was on publications in the years from 1 January 2011 until 31 December 2021. The focus for the 2nd objective was on publications in the years from January 2021 until 19 April 2022. Only published publications were considered. Only publications truly implementing AI (including subfields) in the context of a dental topic or dental specialization were eligible. Respected subfields of AI were: (#1) Machine Learning, (#2) Deep learning, (#3) Neural Networks, (#4) Cognitive Computing, (#5) Natural Language Processing, (#6) Computer Vision. Study design analysis:
|
| Information sources | 9 | Only the following four electronic databases were queried on 19 April 2022:
|
| Search strategy | 10 | PubMed, web search: https://pubmed.ncbi.nlm.nih.gov/advanced/ software: Harzing’s Publish or Perish for batch export, with API key Query: “artificial intelligence”AND(dental OR dentistry OR tooth OR teeth OR dentofacial OR maxillofacial OR orofacial OR orthodontic OR endodontic OR periodontal OR prosthodontic) Type: Articles and reviews in English Relevant publication dates for the 1st objective: 1 January 2011–31 December 2021. Relevant publication dates for the 2nd objective: 2021–19 April 2022, Scopus, web search: https://www.scopus.com/search/form.uri?display=advanced software: Harzing’s Publish or Perish for batch export, with API key Query: “artificial intelligence”AND(dental OR dentistry OR tooth OR teeth OR dentofacial OR maxillofacial OR orofacial OR orthodontic OR endodontic OR periodontal OR prosthodontic) Limited to subject area: Dentistry Type: Articles and reviews in English Relevant publication dates for the 1st objective: 1 January 2011–31 December 2021. Relevant publication dates for the 2nd objective: 2021–19 April 2022 Web of Science—Core Collection web search: https://www.webofscience.com/wos/woscc/advanced-search software: Publish or Perish for batch export, IP based access with login/password Query: “artificial intelligence”AND(dental OR dentistry OR tooth OR teeth OR dentofacial OR maxillofacial OR orofacial OR orthodontic OR endodontic OR periodontal OR prosthodontic) Limited to subject area: Dentistry Oral Surgery Medicine Type: Articles or review articles in English Relevant publication dates for the 1st objective: 1 January 2011–31 December 2021. Relevant publication dates for the 2nd objective 2021–19 April 2022 Google Scholar, web advanced search: https://scholar.google.com/#d=gs_asd&t=1650796298842 software: Harzing’s Publish or Perish for batch export Query: “artificial intelligence” AND (dental OR dentistry OR tooth OR teeth OR dentofacial OR maxillofacial OR orofacial OR orthodontics OR endodontics OR periodontics OR prosthodontics) -motor -rotors -gears -”amelogenesis Imperfecta” Defined Custom year range Type: All, excluding patents and citations Relevant publication dates for the 1st objective: 1 January 2011–31 December 2021. Relevant publication dates for the 2nd objective: 2021–19 April 2022 |
| Study records: | ||
| Data management | 11a | Search and batch export with software: Harzing’s “Publish or Perish” (https://harzing.com/resources/publish-or-perish (accessed on 23 March 2022)). Version for Microsoft Windows: 8.2.3944 (from 23 March 2022), (Windows 11 pro; Microsoft Corp., Redmond, WA, USA) and processed in Microsoft Excel (Excel 365; Microsoft Corp., Redmond, WA, USA). For export and data manipulation, Google Sheets (Alphabet Inc., Mountain View, CA, USA) were also used. This is an online spreadsheet program included as part of the free, web-based Google Docs Editors suite offered by Google. |
| Selection process | 11b | Three independent reviewers conducted each phase of the review. |
| Data collection process | 11c | As the first objective was solely quantitative, the need to extract data from reports was valid only for the second qualitative objective. Publication title and abstract analysis defined the focus of each paper; this was performed separately by three reviewers. Upon the preliminary evaluation of registered papers, 22 dental topics/specializations were defined. Only in cases where an eligible publication was not clearly classifiable from the title and abstract, was a full text analyzed by three independent reviewers. After the reviewers reached a consensus, either the paper’s focus was recorded, or (more frequently) the paper was removed as irrelevant. Only one qualitative marker was assigned to each selected publication for the 2nd objective. In cases of discrepancy between independent reviewers, a consensus had to be reached. The dominant dental topic regarding AI utilization was evaluated, unless the article was equally balanced between two specializations. There were 5 areas that covered all undividable interdisciplinary intersections, which were mostly found between dental radiology, orthodontics, and surgery. For these cases, five border groups had to be defined. |
| Data items | 12 | All 22 qualitative variables included five interdisciplinary groups of papers (# 4,5,6,8,9). |
| Outcomes and prioritization | 13 | The primary outcome of the 1st objective was the list of selected publications for each year. The primary outcome of the 2nd objective was the list of qualitatively evaluated publications distributed into 22 focus groups/topics. Papers focused on more than two qualitative groups were assigned to “general scope” group. Papers addressing dental education in at least one of the defined topics/fields were assigned separately to the “dental education” group. |
| Risk of bias in individual studies | 14 | Each eligible study was evaluated independently by three reviewers. To minimize risk of bias of individual studies, only the studies truly dealing with AI implementation were included. As a number of discrepancies occurred between the three independent reviewers, further tools used to assess the risk of bias in the systematic review were considered. Each initial disagreement was noted. |
| Data synthesis | 15a | The 1st objective provided a list of selected papers. These were grouped according to their publication year and their topics summarized for each year. Percentual increments per year were calculated. The 2nd objective provided a list of selected papers to be distributed into defined focus groups. |
| 15b | For simplicity, no combining of data from other studies was planned. | |
| 15c | No other additional analyses were proposed (such as sensitivity or subgroup analyses, or meta-regression). | |
| 15d | If quantitative synthesis was not appropriate, a systematic narrative synthesis was provided with information presented in the text and tables to summarize and explain the impact of included studies. | |
| Meta-bias(es) | 16 | The weight of each study included was equivalent. Overall, the selective reporting of study results (and the failure to publish small or nonsignificant results) leads to the overestimation of intervention effects in systematic reviews, a phenomenon called meta-bias. |
| Confidence in cumulative evidence | 17 | For the simplicity of this review design, only a “summary of findings” tables are included with a summary of the amount of evidence, in accordance with the GRADE framework (GRADE Working Group, 2004; Guyatt et al. [72]) which combines considerations of risk of bias, directness, heterogeneity, precision, and publication bias [72]. |