This retrospective observational study aims to develop treatment-specific machine learning models for predicting tooth-level periodontal treatment outcomes among teeth treated with non-surgical periodontal treatment, conventional flap surgery, or regenerative periodontal surgery. The study uses a multidimensional dataset including baseline clinical periodontal parameters, radiographic findings, documented treatment modalities, and patient-level demographic and clinical characteristics.

The analytical unit of the study is the tooth. Only periodontally involved teeth with complete baseline and follow-up clinical records, radiographic assessment, clearly documented treatment modality, and measurable periodontal outcomes are included in the predictive analyses. Full-mouth periodontal information is used for patient-level disease characterization, including periodontal staging and grading according to the 2017 AAP/EFP classification.

Because treatment allocation was not randomized, the models are intended to support treatment-specific outcome prediction and clinical interpretability rather than to establish causal superiority between treatment modalities.

Periodontitis is a chronic, multifactorial inflammatory disease characterized by progressive destruction of the supporting periodontal tissues. Although contemporary periodontal classification systems provide a structured framework for diagnosis, staging, and grading, prediction of treatment response remains challenging because outcomes may vary according to patient-level characteristics, local tooth-level conditions, defect morphology, baseline periodontal status, and treatment modality.

Periodontal treatment may include non-surgical periodontal therapy, conventional flap surgery, or regenerative periodontal surgery, depending on clinical indication and local periodontal findings. In routine clinical practice, treatment decisions are individualized and based on clinical examination, radiographic assessment, defect characteristics, and clinician judgment. However, the ability to predict treatment response before or during treatment planning remains limited.

This retrospective observational study uses archived clinical and radiographic records to develop treatment-specific machine learning models for predicting periodontal treatment outcomes at the tooth level. The study focuses on periodontally involved teeth with documented treatment modality and measurable follow-up outcomes. Baseline clinical periodontal parameters, radiographic findings, treatment modality, and relevant patient-level characteristics are used to support outcome prediction and model interpretability.

The purpose of the study is not to establish causal superiority between treatment modalities, but to evaluate whether machine learning models can provide clinically interpretable, treatment-specific predictions of periodontal treatment response. Explainable artificial intelligence methods are used to identify variables contributing to model predictions and to support future development of personalized periodontal treatment planning.

Inclusion Criteria:

1. Patients with a confirmed diagnosis of periodontitis according to the 2017 AAP/EFP classification, supported by complete clinical and radiographic records.
2. Availability of baseline clinical periodontal examination and radiographic records before periodontal treatment.
3. Completion of active periodontal therapy, including non-surgical periodontal treatment and/or surgical periodontal treatment when clinically indicated.
4. Availability of at least one post-treatment follow-up visit after completion of active periodontal therapy.
5. Presence of at least one periodontally involved tooth meeting tooth-level eligibility criteria.
6. Availability of detailed tooth-level documentation, including baseline periodontal measurements, radiographic assessment, documented treatment modality, and corresponding post-treatment outcome records.
7. Teeth were eligible for tooth-level analysis if they received one of the predefined periodontal treatment modalities: non-surgical periodontal treatment, conventional flap surgery, or regenerative periodontal surgery.
8. Patients with a previous history of cancer were eligible if chemotherapy or radiotherapy had been completed and medical clearance for periodontal treatment had been obtained.

Exclusion Criteria:

1. Incomplete demographic, clinical, radiographic, treatment, or follow-up records.
2. Unclear or undocumented periodontal treatment modality.
3. Systemic conditions contraindicating periodontal treatment or substantially affecting periodontal treatment outcomes.
4. Pregnancy or breastfeeding at the time of periodontal treatment.
5. Ongoing chemotherapy or radiotherapy.
6. Current or previous bisphosphonate therapy affecting periodontal or surgical treatment eligibility.
7. Presence of an immunocompromised condition.
8. Acute systemic illness or active infection at the time of periodontal evaluation or treatment.
9. Teeth with missing baseline or follow-up periodontal measurements, missing radiographic assessment, unclear treatment allocation, or insufficient documentation for outcome assessment were excluded from the tooth-level analysis.

The study population consists of patients diagnosed with periodontitis who attended the Department of Periodontology, Faculty of Dentistry, Akdeniz University, between 2021 and 2025. Patients were identified retrospectively from archived clinical and radiographic records. Eligible patients had completed active periodontal therapy and had complete baseline and follow-up records. The analytical unit was the tooth; only periodontally involved teeth with documented treatment modality and measurable treatment outcomes were included in the predictive analyses.