Comparative accuracy of responses of chatbots and machine learning platforms for queries related to traumatic dental injuries- a systematic review

Review question. What is the comparative accuracy of responses of chatbots and machine learning platforms for queries related to various traumatic dental injuries? Searches. A search strategy based upon the Population, Intervention, Comparator, and Outcome has been formulated. It will be electronically performed in PubMed, EMBASE, Web of Science, Lilacs, and Scopus on 19 June 2025. The search shall include MeSH terms and keywords and it shall be modified as per the database. The search results will be saved in EndNote Classic. Grey literature search will be done by using Google Scholar and Open Grey. Hand-searching of the references of the included articles will also be performed. Population (P)- Queries related to different domains and subdomains of traumatic dental injuries Intervention (I)- Any chatbot or machine learning platform Comparator (C)- Comparable chatbot or machine learning platform Outcome (O)- Comparative accuracy of the chatbots or machine learning models Field 1: Traumatic-Dental-Injuries OR TDI OR Dental-Trauma OR Tooth-avulsion OR Tooth-Fracture OR Tooth-Trauma OR Tooth-Injuries OR Tooth-Luxation OR Dental-injuries OR Crown-Fracture OR Crown-Root-fracture OR Root-Fracture OR Dento-alveolar-fracture OR Lateral-Luxation OR Intrusive-Luxation OR Extrusive-Luxation OR Fractured-Tooth Field 2: ChatBot OR ChatGPT OR Artifical-Intelligence OR Machine-learning OR Deep-learning OR Alexa OR Copilot OR Gemini OR Siri OR Bixby OR AI Population. Queries related to different domains and subdomains of traumatic dental injuries Intervention. Any chatbot or machine learning platform Comparator. Comparable chatbot or machine learning platform Types of study to be included. All descriptive, observational, cohort studies, cross sectional and case control studies having elements described in the PICO of research questions were included in the review. Case reports, case series, short communications, systematic and narrative reviews will be excluded from the review. Context. Traumatic dental injuries Main outcome. Comparative accuracy of the chatbots or machine learning models The accuracy will be measured in form of frequency or proportion of the correct responses. The comparisons will be made in terms of mean difference. Additional outcomes. The effect of the various factors such as type of question, type of injury, geographical region, and language will be assessed. The influence of the factors will be measured in terms of odds ratio or risk ratio. Data extraction (selection and coding). Two authors NT, AK will assess the titles and abstracts to determine eligibility for full text evaluation. NT, AK will evaluate the full text on the basis of inclusion and exclusion criteria. MR and AK will perform the data extraction independently using a specially designed pre piloted (on 5 studies) data extraction form. In event of any disagreement, the Senior Reviewer CS will be consulted for agreement. The main data fields to be extracted will be- 1. Author (first author) 2. Year 3. Journal 4. Aim of the study described in the Population (P), Intervention (I), Comparator (C), Outcome (O). 5. Study setting 6. The type and details of chat bots and machine learning platforms 7. Type and details of questions used 8. Information about the development of questions 9. Number of times the question cycles were repeated 10. Method of assessment accuracy of the responses 11. The accuracy of responses and their characteristics 12. Overall accuracy 13. Conclusion Risk of bias (quality) assessment. The risk of bias will be assessed by using Jonna Briggs Critical Appraisal Checklist for analytical cross sectional studies. This will be performed by AK and MR and in the event of any disagreement, GT will be consulted for consensus. Strategy for data synthesis. The extracted data will be discussed by the research group and strategy shall be formulated for its analysis and clear presentation. In the event of less heterogeneity, the meta-analysis will be performed by using a random effects model with effect size of either the percentages or risk/odds ratio. The heterogeneity will be defined by I2 values and the weight of the studies and 95% Confidence Intervals will be calculated. Publication bias shall be assessed by using the Funnel plot. Analysis of subgroups or subsets. If sufficient data is available, the subgroup analysis will be done for type of question, type of injury, geographical region, and language.