Distinct Lung Function Trajectories Identified in Idiopathic Pulmonary Fibrosis

Different lung function trajectories have been identified in individuals with idiopathic pulmonary fibrosis (IPF), which could have major implications for research and patient care in this population, according to study findings published in The Lancet Digital Health.

Recognizing that IPF is a progressive fibrotic lung disease with a variable clinical trajectory, the researchers sought to identify distinct clusters of lung function trajectory among individuals with IPF with the use of machine learning techniques. The researchers did a secondary analysis of longitudinal data on forced vital capacity (FVC) in patients with IPF enrolled in the PROFILE study (ClinicalTrials.gov Identifier: NCT01134822), a multicenter, prospective, observational cohort study.

After evaluating the performance of conventional and machine learning techniques for imputing missing data, the researchers analyzed a fully imputed dataset via unsupervised clustering using self-organizing maps. Anthropometric characteristics, genomic associations, serum biomarkers, and clinical outcomes were compared between the clusters. Additionally, the investigators performed a replication of the analysis on data derived from a cohort of patients with IPF from an independent dataset, which was obtained from the Chicago Consortium.

Overall, 71% (415 of 581) of the participants who were recruited from PROFILE met eligibility criteria for additional analysis. Results of the analysis showed that use of an unsupervised machine learning algorithm had the lowest imputation error among all the tested methods. Self-organizing maps identified 4 distinct clusters (1 through 4), which were confirmed with sensitivity analysis.

Cluster 1 comprised 34% (140 of 415) of the participants and was associated with a disease trajectory that demonstrated a linear decline in FVC over 3 years. Cluster 2, which comprised 24% (100 of 415) of the participants, was associated with a trajectory that showed an initial improvement in FVC prior to a subsequent decline. Cluster 3 comprised 27% (113 of 415) of the participants and was associated with a trajectory that demonstrated an initial decrease in FVC before subsequent stabilization. Cluster 4, which comprised 15% (62 of 415) of the participants, was associated with a trajectory that demonstrated stable lung function.

Median survival was the shortest in cluster 1 (2.87 years; range, 2.29 to 3.40 years) and cluster 3 (2.23 years; range, 1.75 to 3.84 years), followed by cluster 2 (4.74 years; range, 3.96 to 5.73 years). The median survival was the longest in cluster 4 (5.56 years; range, 5.18 to 6.62 years). Baseline forced expiratory volume in 1 second (FEV₁) to FVC ratio and concentrations of the biomarker serum surfactant protein-D (SP-D) were both significantly higher in clusters 1 and 3. Similar lung function clusters that shared some anthropometric characteristics were observed in the replication cohort.

Several limitations of the current study warrant mention. Because of the extent of missing data, only a small number of individuals with IPF were available to effectively train the imputation algorithm, which might decrease the ability of the model to effectively identify additional smaller clusters. Another limitation of the analysis is the absence of unadjusted replication of the relationship between cluster and mortality signal in the PROFILE cohort compared with the replication cohort.

The authors concluded that “Using a data-driven unsupervised approach, we identified four clusters of lung function trajectory with distinct clinical and biochemical features. Enriching or stratifying longitudinal spirometric data into clusters might optimise evaluation of intervention efficacy during clinical trials and patient management” in individuals with IPF.

Enriching or stratifying longitudinal spirometric data into clusters might [optimize] the evaluation of intervention efficacy during clinical trials and patient management [in individuals with IPF].”

Disclosure: Some of the study authors have declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.