CLINICAL DIGITAL TWINS IN PEDIATRIC ASTHMA: GENERATION AND DISTRIBUTIONAL VALIDATION OF A LONGITUDINAL SYNTHETIC DATABASE UNDER SAMPLE CONSTRAINTS

Abstract

Considering that pediatric clinical trials often yield small samples, incompatible with the data volume required by deep learning, and that asthma is the most prevalent chronic disease in childhood, the aim is to present and validate a reproducible methodology for generating a longitudinal synthetic database from a small-scale randomized clinical trial on diaphragmatic breathing. To this end, a deterministic Python simulator was developed, integrating three components: empirical extraction of behavioral distributions from a mobile health application; parametric calibration through truncated normal and lognormal distributions derived from published parameters; and a Markovian dynamic with engagement modeled asymmetrically between groups, capturing the distinction between therapeutic effect and placebo effect. Distributional fidelity was verified by the two-sample Kolmogorov-Smirnov test, complemented by the normalized Wasserstein distance, under predefined criteria. In this way, the procedure generated 12,491 longitudinal records of 500 synthetic patients, without violation of the biological limits and preserving the empirical adherence patterns. It is concluded that the proposed mechanistic-empirical integration constitutes an auditable and clinically coherent strategy to overcome the data constraint, enabling the database as a Clinical Digital Twin for subsequent predictive stages.