Bronchiectasis Phenotype Identification Model

The Bronchiectasis Phenotype Identification Model (BPIM) is developed within the Assiut University prospective bronchiectasis translational research platform as a supervised baseline phenotype-translation framework for adults with non-cystic fibrosis bronchiectasis (NCFB).

The Bronchiectasis Phenotype Identification Model (BPIM) follows a two-step analytical architecture. First, latent class trajectory analysis (LCTA) identifies trajectory-derived bronchiectasis phenotype classes using prospectively collected longitudinal disease-signature data. Second, the Bronchiectasis Phenotype Identification Model (BPIM) translates the accepted latent class trajectory analysis (LCTA)-derived phenotype structure into a supervised baseline classifier using the locked baseline disease-signature architecture.

The Bronchiectasis Phenotype Identification Model (BPIM) is methodologically separated from the Bronchiectasis Assessment of Severity and Exacerbations (BASE) framework while remaining scientifically linked to it. The Bronchiectasis Assessment of Severity and Exacerbations Severity model (BASE-S) classifies current bronchiectasis severity at baseline. The Bronchiectasis Assessment of Severity and Exacerbations Prognostic model (BASE-P) predicts 12-month bronchiectasis exacerbation risk. The Bronchiectasis Phenotype Identification Model (BPIM) predicts the accepted latent class trajectory analysis (LCTA)-derived phenotype class.

The study uses a prospective observational development-validation design. The development cohort will be used to execute the prespecified latent class trajectory analysis (LCTA) hierarchy, identify the accepted trajectory-derived phenotype structure, assign phenotype labels according to the locked convention, and train the supervised Bronchiectasis Phenotype Identification Model (BPIM) classifier. The validation cohort will be used only to evaluate the locked Bronchiectasis Phenotype Identification Model (BPIM) classifier without refitting, recalibration, predictor substitution, phenotype relabeling, threshold retuning, or post hoc classifier rescue.

The latent class trajectory analysis (LCTA) component will use a prespecified top-down variable-combination hierarchy based on longitudinal functional, oxygenation, and inflammatory disease-signature domains. Three-domain latent class trajectory analysis (LCTA) options will be attempted first. If no acceptable three-domain solution is identified, two-domain options will be attempted. If all three-domain and two-domain options fail, one-domain options will be attempted as the final fallback level. Class-number selection, acceptability criteria, failure criteria, and phenotype-labeling rules are prespecified in the locked protocol and statistical analysis plan.

Following acceptance of the latent class trajectory analysis (LCTA) solution, phenotype classes will be ordered according to increasing composite inflammatory and functional disease burden. Depending on the accepted class number, phenotype labels may include Stable phenotype, Progressive/Frequent Exacerbator phenotype, Frequent Exacerbator/Inflammatory phenotype, Advanced Multidomain phenotype, and End-stage/Terminal-risk phenotype according to the locked labeling convention.

The Bronchiectasis Phenotype Identification Model (BPIM) classifier will be trained as a supervised baseline classifier. Binary logistic regression will be used if the accepted latent class trajectory analysis (LCTA) solution contains two classes. Multinomial logistic regression will be used if the accepted latent class trajectory analysis (LCTA) solution contains three or four classes. Classification performance will be evaluated using confusion matrix, overall accuracy, class-specific sensitivity, class-specific specificity, positive predictive value, negative predictive value, macro-average F1 score where applicable, and agreement between predicted Bronchiectasis Phenotype Identification Model (BPIM) class and accepted latent class trajectory analysis (LCTA)-derived class. Where predicted class probabilities are generated, probability calibration will be assessed using calibration plots, observed-versus-predicted class probability summaries, and calibration metrics where appropriate.

The Bronchiectasis Phenotype Identification Model (BPIM) study is observational and non-interventional. No treatment, medication, intervention, exposure, or management strategy is assigned by this protocol. All clinical care follows routine institutional practice and treating physician judgment. The study is not designed to estimate causal treatment effects.

The locked methodological disclosure, protocol, deterministic statistical analysis plan, latent class trajectory analysis (LCTA) variable ledger, phenotype-labeling convention, supervised classifier structure, and validation governance are archived in the version-specific Zenodo record: https://doi.org/10.5281/zenodo.20157926.

The related Bronchiectasis Assessment of Severity and Exacerbations (BASE) structural lock is archived separately at: https://doi.org/10.5281/zenodo.20143505.

• Accepted Latent Class Trajectory Analysis-Derived Phenotype Class [Time frame: Baseline to 12 months follow-up.]
• Agreement Between Bronchiectasis Phenotype Identification Model-Predicted Class and Accepted Latent Class Trajectory Analysis-Derived Class [Time frame: Baseline to 12 months follow-up.]
• Class-Specific Sensitivity of the Bronchiectasis Phenotype Identification Model [Time frame: Baseline to 12 months follow-up.]
• Class-Specific Specificity of the Bronchiectasis Phenotype Identification Model [Time frame: Baseline to 12 months follow-up.]
• Positive Predictive Value of the Bronchiectasis Phenotype Identification Model for Phenotype Classification [Time frame: Baseline to 12 months follow-up.]
• Negative Predictive Value of the Bronchiectasis Phenotype Identification Model for Phenotype Classification [Time frame: Baseline to 12 months follow-up.]
• Macro-Average F1 Score of the Bronchiectasis Phenotype Identification Model [Time frame: Baseline to 12 months follow-up.]
• Probability Calibration of the Bronchiectasis Phenotype Identification Model [Time frame: Baseline to 12 months follow-up.]
• Confusion Matrix of Bronchiectasis Phenotype Identification Model Classification [Time frame: Baseline to 12 months follow-up.]
• Relationship Between Bronchiectasis Phenotype Identification Model Phenotype Classes and Bronchiectasis Assessment of Severity and Exacerbations Severity Categories [Time frame: Baseline]
• Relationship Between Bronchiectasis Phenotype Identification Model Phenotype Classes and Bronchiectasis Assessment of Severity and Exacerbations Prognostic Risk Categories [Time frame: Baseline to 12 months follow-up.]
• Bronchiectasis Exacerbation Occurrence by Bronchiectasis Phenotype Identification Model Phenotype Class [Time frame: Baseline to 12 months follow-up.]