I hypothesize that HRCT pattern models in CTD-ILD and IPAF will calibrate better when serologic and systemic autoimmune phenotype variables are added to the imaging feature set, because a UIP-like scan does not carry the same pretest meaning in every autoimmune context.

Why this is testable: the comparison is straightforward. Fit an imaging-only model and an imaging-plus-autoimmune-context model, then validate both externally across centres. The autoimmune layer can include ANA pattern and titer, SSA/SSB, Scl-70, anti-synthetase antibodies, Raynaud phenomenon, sicca symptoms, skin thickening, inflammatory arthritis, smoking, and age.

Primary endpoints: calibration slope, calibration-in-the-large, Brier score, AUROC, decision-curve net benefit, and disagreement rate versus multidisciplinary diagnosis. Secondary analyses should examine whether the added context mainly improves NSIP and organizing pneumonia discrimination, where autoimmune inflammatory signals are often stronger.

Falsification criteria: the hypothesis fails if autoimmune context does not improve external calibration or net benefit, or if improvements are limited to internal validation but vanish in a held-out cohort.

References:

1. Fischer A, Antoniou KM, Brown KK, et al. An official European Respiratory Society/American Thoracic Society research statement: interstitial pneumonia with autoimmune features. Eur Respir J. 2015;46(4):976-987. DOI: 10.1183/13993003.00150-2015
2. Oldham JM, Adegunsoye A, Valenzi E, et al. Characterisation of patients with interstitial pneumonia with autoimmune features. Eur Respir J. 2016;47:1767-1775. DOI: 10.1183/13993003.51565-2015
3. Raghu G, Remy-Jardin M, Myers JL, et al. Diagnosis of Idiopathic Pulmonary Fibrosis. Am J Respir Crit Care Med. 2018;198(5):e44-e68. DOI: 10.1164/rccm.201807-1255ST