COPD Stage Progression and Regression: a Multistate Transition Analysis of The COPDGene Cohort

To the Editor:

Chronic obstructive pulmonary disease (COPD) is typically diagnosed only after patients develop severe airflow obstruction.¹ However, even in its early stages, the disease causes reduced respiratory function and increased mortality.² A better understanding of the disease course, including the timing of disease progression and possible regression, may improve early detection, allowing time to address risk factors, slow progression, and improve prognosis.²

To better characterize the COPD disease course, we used data from the COPDGene study, a US observational longitudinal cohort of adults aged 45–80 years old who smoked ≥10 pack-years, with and without COPD diagnosis.³ Data were collected at baseline (2008–11) with two 5-year follow-up intervals (2012–16, 2018–21). We included only individuals who had ever smoked and had post-bronchodilator spirometry values (FEV₁ and FVC) for at least two time points. Transition probabilities were estimated from 8,487 observation pairs (5,746 individuals) across the three study phases. We categorized each individual at each time point into a severity stage using the GOLD classification (FEV₁/FVC ratio; predicted FEV₁): GOLD 0 (normal: ≥0.7 and ≥80%, respectively); Preserved Ratio Impaired Spirometry (PRISm; ≥0.7 and <80%); GOLD 1 (<0.7 and ≥80%); GOLD 2 (<0.7 and 50–80%); and GOLD 3–4 (<0.7 and <50%); or deceased.

We developed a Markov multistate transition model to estimate progression, regression, and mortality propensities at each COPD stage. Biologically, disease progression was assumed to occur sequentially, through (potentially unobserved) intermediate stages, without skipping stages. We assumed no direct transitions between PRISm and GOLD 1, consistent with the low numbers (<25) of observed transitions between these stages, but individuals in each of PRISm and GOLD 1 could regress to GOLD 0 or progress to GOLD 2.

We used the estimated transition rates to compute expected transition probabilities for individuals in each stage after 1 and 5 years.⁴ Additionally, we calculated transition hazard ratios for pairs of transitions, estimating the relative propensity of progression vs. regression in the early stages of the disease. Finally, we calculated the mean sojourn time spent in each disease severity stage, i.e., expected time in that stage before progressing, regressing, or dying. Analysis was done in R(v4.4) using the msm package.

At baseline, the prevalence of the disease severity stages among the participants (age range 38–90, mean 59.5± 8.6 years) were 47.3% GOLD 0, 8.8% GOLD 1, 19.3% GOLD 2, 12.2% GOLD 3–4, and 12.3% PRISm. GOLD 0 had the longest mean sojourn time (16.1 years, 95% CI 15.0–17.2; Figure 1), with 94.1% and 76.0% of participants remaining in that stage after 1 and 5 years, respectively. Individuals in GOLD 0 had similar rates of progression to GOLD 1 vs PRISM (hazard ratio [HR] 1.1, 95% CI 0.93–1.30), resulting in similar transition probabilities (2.5% and 2.3%, respectively) after 5 years (Figure 2).

Figure 1. Estimated mean sojourn time of years spent in a single COPD severity stage before transitioning.

Of the 19,266 observations (10,652 individuals), 3.7% were excluded due to missing or never smoking status, 4.5% had missing spirometry values, and 22.9% had just one observation across the three phases. Transition probabilities were estimated from 8,487 observation pairs from our final analytical sample of 13,269 observations (5,746 individuals).

Figure 2. One-year (A) and five-year (B) transition probabilities between PRISm, GOLD severity stages and death, estimated from all-phases sample.

Individuals may make multiple transitions between observations, so an individual may appear to skip intermediate disease stages, e.g., the probability of transitioning from GOLD 1 to GOLD 3 is non-zero because it accounts for the probability of transitioning from GOLD 1 to GOLD 2 and from GOLD 2 to GOLD 3 between observations.

GOLD 1 and PRISm were the least persistent disease stages, with <50% of participants remaining in those stages after 5 years. Their mean sojourn times were 6.0 years (95% CI, 5.5–6.6) and 6.8 years (95% CI, 6.2–7.4), respectively, before transitioning with an approximately equal likelihood of about 20% for each direction after 5 years (HR for progression vs. regression from GOLD 1: 1.15 [95% CI 0.94–1.42] and from PRISm: 1.09 [95% CI 0.9–1.30]).

GOLD 2 had a mean sojourn time of 8.0 years (95% CI, 7.5–8.6), with 57.2% remaining in GOLD 2 after 5 years. Individuals in GOLD 2 were about equally likely to progress to GOLD 3–4 as to regress to PRISm or GOLD 1 (about 15% each after 5 years); the rate of regression may have been lower to GOLD 1 vs. PRISM (HR 0.78, 95% CI 0.59–1.04), but the difference was not statistically significant. Individuals in GOLD 3–4 had a mean sojourn time of 10.5 years (95% CI 9.6–11.5), with 63.2% remaining in GOLD 3–4, a low probability of regression (8.4%) and the highest probability of dying (27.3%) after 5 years.

In summary, in a cohort of US ever-smoking adults, we found that mild disease severity stages (GOLD 1, PRISm) were the most dynamic COPD stages, with similar likelihoods of progression or regression, while GOLD 3–4 had a low probability of regression and a high risk of mortality. Consistent with findings from other studies,² this result underscores the importance of recognizing early spirometric abnormalities as potential windows for interventions to modify COPD long-term outcomes.^5,6

Our results are consistent with a prior report by Wan et al.,⁷ also using COPGene, which found that around 50% (51.9%−53.9%) of individuals transition in and out of PRISm within 5 years compared to ≥86% that remain with COPD. Here, we reported that after 5 years, only 50.3% of participants stayed at PRISm, and only 45.8% remained at GOLD 1. Other studies in other populations have also found that the early stages of COPD are the most dynamic.⁸

We showed that individuals with COPD can regress to a less severe stage, especially from GOLD 1 or PRISm. Other studies have reported this regression phenomenon in this and other populations.^7,8 It is possible that early lung injury may resolve, particularly among the PRISm group. However, it may also represent a certain degree of bronchodilator reversibility or spirometry variability, particularly for mild and moderate obstructions.⁹ We also showed that disease regression was less likely at severe stages (GOLD 3–4), consistent with indications that patients with severe disease have a higher number of exacerbations, hospitalizations, and comorbidities that accelerate the FEV₁ decline.¹⁰

Our study is one of the few to report transitions among airflow limitation categories across the entire spectrum of COPD severity and PRISm. Stratifying COPD by disease severity allowed us to develop insights into the variability of progression within disease stages and estimate time spent in each stage. However, we only considered spirometry-defined COPD, without accounting for obstructive vs restrictive patterns,¹¹ treatment,¹² exacerbations, biomarkers, CT imaging, or environmental determinants.

This study reinforces that early detection is likely key to reducing COPD burden. Disease progression may be delayed by early diagnosis, timely treatment, and risk-factor interventions. However, the dynamic nature of early COPD stages and limitations of spirometry pose a challenge for early detection, highlighting potential limitations in current GOLD criteria and the need for improved diagnostic tools and guidelines. Nevertheless, enhancing early diagnostic strategies and timely intervention could mitigate disease progression and improve patient outcomes, reducing morbidity and mortality.

Data Sharing:

The code used to replicate this analysis is publicly available in a Zenodo repository: https://doi.org/10.5281/zenodo.17144121