SMARTp

SMARTp analyzes non-surgical treatment data for chronic periodontitis using a Sequential Multiple Assignment Randomized Trial (SMART) design to develop and evaluate dynamic treatment regimes (DTRs) while accounting for non-Gaussian outcomes, spatial clustering, and nonrandom missingness.

Key Features:

• Dynamic Treatment Regimes (DTRs): Implements and compares DTRs for adaptive, response-driven non-surgical periodontal treatments within a SMART design.
• Statistical Analysis Plan: Provides methods to analyze periodontal data with non-Gaussian distributions and spatial dependencies in clustered tooth-level measurements.
• Cluster-Level Sample Size Calculation: Calculates cluster-level sample sizes for periodontal studies accounting for skewness and tail behavior of clinical attachment level (CAL) outcomes via a skew-t distribution model.
• Handling Nonrandom Missingness: Employs a shared parameter model to address nonrandom missing data and support robust inference.
• Tooth-Level Covariance Structure: Models tooth-level spatial relationships using a conditionally autoregressive framework.
• Inverse Probability Weighting and Method of Moments: Uses inverse probability weighting and the method of moments to derive expected values and variances/covariances of sample means across DTRs.
• Simulation Studies: Conducts simulation studies to evaluate finite-sample performance of proposed sample size formulas under varied outcome-generating scenarios.

Scientific Applications:

• Personalized Treatment Strategies: Enables development and evaluation of tailored treatment plans for chronic periodontitis using DTRs.
• Robust Statistical Analysis: Facilitates reliable inference for periodontal treatment effects in the presence of non-Gaussian outcomes, spatial clustering, and nonrandom missingness.
• Sample Size Optimization: Supports determination of cluster-level sample sizes that account for skewed CAL distributions and tooth-level spatial correlation.

Methodology:

Analysis of SMART design data and DTRs; skew-t distribution modeling for CAL outcomes in cluster-level sample size calculations; shared parameter models for nonrandom missingness; conditionally autoregressive modeling of tooth-level covariance; inverse probability weighting and method of moments to obtain expected values and variances/covariances of sample means across DTRs; and simulation studies to assess finite-sample performance.