Notes from ERS Congress 2026

In early September I was in Barcelona for the European Respiratory Society (ERS) International Congress 2026, held at Fira Gran Via from September 5 to 9. Our late-breaking abstract had been accepted as a poster, and Dayoon Lee and I attended together and presented it on Tuesday, September 8.

Our poster

Reading screening spirometry as a trajectory predicts incident airflow obstruction earlier: a landmark deep-learning analysis of 221,903 adults (Kyung Hyun Lee, Dayoon Lee, Jin Young Lee, Byung Jae Lee).

In health screening, spirometry is repeated over many years, yet each result is usually read on its own. A person can decline steadily while every individual test still falls inside the normal range, and so still looks reassuring. The question we asked was whether the trajectory of repeated spirometry carries information beyond the latest value, and whether a nonlinear model is needed to capture it.

We drew on 557,503 health-screening spirometry tests from 221,903 adults with normal airflow at their first test (FEV1/FVC ≥ 75%). At a fixed landmark three years after the first test, we predicted incident airflow obstruction (FEV1/FVC < 0.70) over the next five years, using only the tests taken before the landmark. Three readings of the same history were compared: the latest value alone, a linear trajectory summary (slope, variability, test count, span), and a sequence model that reads every measurement in order.

What we found:

• Direction matters before obstruction. Among people starting in the same normal band, those on a falling trajectory had a 5-year risk of 5.17%, against 1.88% for a stable trajectory and 0.74% for a rising one — an age- and sex-adjusted relative risk of 6.4 (95% CI 4.2–9.6) for falling versus rising. At the moment of prediction, both groups were still above 0.70; what differed was the direction from which they arrived.
• More history improves prediction. Both trajectory models improved on the latest value, and the full sequence model was best (AUPRC 0.462 → 0.534).
• The advantage grows further ahead. When outcomes occurring within one or two years of the prediction were excluded, the sequence model’s C-index advantage over the latest value widened from +0.018 to +0.026. The gain held under a stricter two-in-a-row outcome definition and in a cohort restricted to a baseline FEV1/FVC ≥ 80%, while the linear summary was not consistently better than the latest value.

Taken together, the results suggest that repeated screening spirometry is better read as a longitudinal signal than as a series of isolated tests, and that change over time may flag risk years before a threshold-based label appears.

The study has clear limits: it uses retrospective single-centre screening data, with no smoking, diagnosis, or bronchodilator-response labels; the outcome is spirometric airflow obstruction rather than clinical COPD; and validation is internal. The next steps are to analyse clinically performed pulmonary function test records now being acquired from hospital departments, and to carry out a planned multicenter external validation of transportability, calibration, and subgroup performance.

At the poster

Thanks to Prof. Jin-Young Lee and Prof. Byung-Jae Lee for their guidance on this work, to Dayoon for attending and presenting it with me, and to everyone who stopped by the poster.