Methodological Quality of Prognostic Models for Periodontitis: A Systematic Review and Critical Appraisal
Maryia Karaban, Debora R. Dias, Giuseppe Troiano, Sahar Baniameri, Ishan Ahuja, Divya Joseph, Matteo Serroni, Andrea RavidàABSTRACT
Aim
To critically appraise the methodological quality and historical evolution of prognostic models for periodontitis.
Materials and Methods
A systematic search was conducted in three databases. Eligible studies were classified as conceptual tools, category‐based tools and data‐driven models. Category‐based and data‐driven models were evaluated with PROBAST. Risk of bias (ROB) was compared across four domains: Participants, Predictors, Outcome and Analysis.
Results
Twenty‐seven studies were included. Prognostic methods evolved from qualitative risk categories and expert‐based classifications towards quantitative, individualised probability estimates generated through multivariable statistical and machine‐learning models, alongside progressive improvements in outcome definition, prediction time and risk expression formats. Conceptual tools were not evaluable under PROBAST due to the absence of empirical data. Category‐based tools showed limitations in predictor handling and validation rather than bias per se. Data‐driven models demonstrated progressive refinement: earlier studies lacked consistent validation, whereas recent models more frequently report internal validation, calibration metrics and external validation. However, all studies except one were classified as high ROB. Domain‐level assessment showed strengths in the Participants domain but persistent limitations in the Analysis domain. Methodological improvements after TRIPOD were modest.
Conclusion
Prognostic models reflect distinct methodological profiles across design type. Conceptual and category‐based frameworks remain clinically relevant historically, while data‐driven models represent the contemporary standard but retain important methodological gaps.
Trial Registration
PROSPERO number: CRD42024525505