Answers are grounded in this paper and cite their sources.
Background Quantitative computed tomography (CT) and radiomics may support differential diagnosis of established immune checkpoint inhibitor-related pneumonitis (CIP), but transportable accuracy and clinical utility remain uncertain. Methods Six databases were searched in June 2026, with a documented public-index update through 31 August 2026. Current etiologic attribution was distinguished from future-risk prediction. Revision-stage author assessments and final adjudication were supplemented by a complete 3,900-record disposition log. Abstracts and preprints were included as a distinct evidence stratum; the accuracy synthesis used full-length published reports. Risk of bias was reassessed using tailored Quality Assessment of Diagnostic Accuracy Studies, version 3 (QUADAS-3) at the accuracy-estimate level, with separate descriptive-study appraisal. Results The review includes 17 reports: 13 full-length publications (11 accuracy reports and two descriptive comparisons), three abstracts and one indexed preprint. One included abstract is a companion, without duplicate cohort counting. Twenty-one areas under the receiver operating-characteristic curve (AUCs) and 10 sensitivity/specificity pairs were appraised; 20 estimates had high risk and 11 had insufficient information. These estimates were not independent cohorts. Validation was predominantly internal; one report included mixed-institution rather than purely external validation. The only compatible patient-level CIP-versus-radiation-pneumonitis hold-out operating point involved 40 patients: sensitivity 0.75 (exact 95% CI 0.51–0.91) and specificity 0.60 (0.36–0.81). No meta-analysis was performed. Prospective artificial intelligence (AI)-assisted reader or management-benefit evaluation was not identified among included reports. Conclusions The evidence supports within-dataset discrimination, not established transportable diagnostic utility. Unavailable full texts, historical validation limitations and incomplete cross-source linkage constrain evidence completeness. Independent attribution, locked patient-level testing, genuinely external validation and prospective clinical-impact evaluation remain priorities.
No takes yet. Share an insight, caveat, or question.
Huang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: