Transforming Pulmonology: Predicting Pulmonary Fibrosis and Learning from COVID-19 Trials

Pulmonologists face an era reshaped by COVID-19, presenting both challenges and opportunities in predicting and managing pulmonary fibrosis. Advanced CT analytics can sharpen risk stratification and prompt earlier, structured follow-up in respiratory clinics.

Leveraging deep learning and radiomics can enhance prediction of fibrotic risk—for example, the risk of radiographic fibrotic progression on CT—potentially facilitating earlier intervention pathways. Early studies integrating deep learning with radiomics report promising performance in identifying radiographic fibrotic changes in post-COVID cohorts, but these tools should be considered preliminary and not yet determinative for therapy decisions. Preliminary reports describe performance using metrics such as AUROC in early cohorts, with external validation still needed before clinical adoption. However, real-world benefit depends on external validation, integration with clinical risk scores, and demonstration of utility in prospective studies.

Moreover, in the realm of post-COVID care, the early prediction of fibrosis bears significant implications. Identifying fibrosis early enables timely management, including consideration of antifibrotic therapy in selected cases, which may help slow disease progression and improve functional measures; evidence for mortality or morbidity reduction in post-COVID ILD remains limited and under active study. For context on evolving therapeutic considerations, see a concise review on antifibrotics in post-COVID ILD. These caveats underscore why imaging-derived risk tools should be tested alongside clinical variables and functional measures before influencing treatment pathways.

Addressing long-term respiratory complications in patients with post-acute sequelae of SARS-CoV-2 infection (PASC), particularly those with pulmonary involvement, poses ongoing challenges as clinical phenotypes evolve. Clinics are seeing heterogeneity—from transient post-inflammatory scarring to persistent fibrosing interstitial lung disease—necessitating careful phenotyping, longitudinal monitoring, and judicious use of therapies. Multidisciplinary discussion remains central to interpreting imaging changes in the context of symptoms, pulmonary function tests, and exposure histories.

Recent studies have illustrated a new frontier in predictive diagnostics that may redefine physician approaches, specifically through advancements seen in COVID-19 clinical trial designs. These trials have underscored the importance of rapid, collaborative, and adaptive models that are vital in crisis settings for accelerating evidence generation and assessing treatment efficacy, as discussed in the NEJM Evidence perspective on adaptive trials. At the same time, adaptive designs introduce operational and statistical complexities and require careful pre-specification and data quality safeguards to control bias and Type I error. Adaptive trials conducted during the pandemic not only expedited the evaluation process for COVID-19 therapies but also reshaped pulmonology research by highlighting where flexible, pre-specified adaptive designs can be valuable when Type I error is controlled—though they are not universally applicable. For instance, imaging-derived risk scores could be used to enrich enrollment or trigger sample-size re-estimation in antifibrotic trials for post-COVID ILD.

Such advancements are transforming clinical practices, allowing more precise interventions tailored to individual patient profiles. The same imaging advances can serve as early, quantitative endpoints in adaptive trials, tightening the feedback loop between discovery and clinical testing. These experiences illustrate how streamlined, pre-specified protocols can function in rapidly shifting clinical landscapes, without implying blanket regulatory endorsement. Lessons cataloged in the Wellcome Open Research analysis of adaptive designs in COVID-19 reinforce the value of coordination, platform infrastructures, and transparent statistical plans.

Building validation pathways for imaging biomarkers is essential. First, model development should include clear definitions of end points (e.g., radiographic fibrotic progression on serial CT) and pre-specified thresholds. Second, internal validation with cross-validation should be followed by external validation across diverse scanners, sites, and populations, with attention to domain shift and calibration. Third, analytical validity (repeatability, reproducibility) needs demonstration with phantom and test-retest studies. Finally, clinical validity should connect imaging risk scores to patient-centered outcomes such as dyspnea trajectories, oxygen requirements, and quality-of-life scores.

An example adaptive antifibrotic trial schema for PASC-related fibrosing ILD could proceed in stages. An initial enriched cohort would be screened using a validated imaging risk score combined with physiological markers (e.g., DLCO decline). A seamless phase 2/3 design might use early imaging-based surrogates (quantitative fibrosis extent on CT) as interim decision points for dropping or graduating arms, with pre-specified rules for controlling Type I error. Sample-size re-estimation could be triggered if event rates or biomarker responsiveness diverge from assumptions. Throughout, blinded independent review of imaging endpoints and centralized data quality checks would be mandatory.

Patient-centered outcomes must remain front and center. Beyond survival, endpoints should include symptom burden (e.g., cough, dyspnea), functional capacity (6-minute walk distance), health-related quality of life, and return-to-work metrics. Embedding patient-reported outcomes and home spirometry can capture day-to-day variability, while pragmatic follow-up schedules minimize burden. Importantly, shared decision-making around antifibrotic therapy should incorporate uncertainty about benefits in PASC-related disease and weigh adverse effects and monitoring requirements.

Implementation considerations will determine whether predictive tools and adaptive evidence translate to practice. Health systems will need workflows for structured CT acquisition protocols, automated quality checks, and secure pipelines for running imaging algorithms. Interdisciplinary case conferences can mediate between algorithmic predictions and nuanced clinical judgment. Prospective registries that harmonize imaging, physiology, and outcomes data can provide real-world evidence and post-deployment monitoring for model drift.

Equity and generalizability also require attention. Models trained predominantly on hospitalized or urban cohorts may underperform in community settings or among underrepresented groups. Ensuring diverse data sources, transparent reporting, and accessible deployment (including edge-computing options for low-bandwidth clinics) can mitigate disparities. Governance frameworks should clarify data use, consent, and patient communication around algorithm-informed decisions.

Finally, collaboration will be the engine of progress. Near-term priorities include: (1) externally validating imaging predictors across institutions and scanners; (2) integrating risk scores into clinical pathways that tie imaging, physiology, and follow-up cadence; and (3) designing adaptive antifibrotic trials that use these markers as early endpoints under rigorous, pre-specified statistical plans. By aligning imaging science with methodologically sound trial designs and patient-centered care, pulmonology can convert post-pandemic lessons into durable gains for those at risk of fibrosing lung disease.

Key Takeaways:

• Imaging AI and radiomics show promising but preliminary performance for identifying radiographic fibrotic risk in post-COVID cohorts; external validation is essential.
• Early identification supports timely, individualized management; antifibrotics may slow progression in selected cases, but definitive morbidity/mortality benefits in post-COVID ILD remain unproven.
• Adaptive trial methods can accelerate learning when pre-specified and well-controlled, and imaging biomarkers can serve as interim endpoints or enrichment tools.
• Implementation, equity, and patient-centered outcomes should guide deployment of predictive tools and trial designs.