Cluster analysis of obstructive lung disease phenotypes: the Framingham heart study and the Rotterdam study.

BACKGROUND

The traditional classification of obstructive lung diseases into asthma and chronic obstructive pulmonary disease oversimplifies their complexity. A hypothesis-free classification using routine clinical and laboratory characteristics may identify clinically important subgroups.

METHODS

Hierarchical clustering was applied to 1473 Framingham Heart Study (FHS) participants with a history of asthma or airflow obstruction based on pre-bronchodilator spirometry. Variables included demographics, smoking history, body mass index, inflammatory markers, spirometry, and symptoms, excluding self-reported diagnoses. Classification rules were defined by Classification and Regression Trees (CART). Survival and polygenic scores were compared across clusters. The transportability analysis used 1080 Rotterdam Study (RS) participants.

RESULTS

Six clusters emerged in FHS using nine phenotypic variables. Cluster 1 showed preserved pulmonary function with minimal symptoms in never and former smokers. Clusters 2 and 3 had preserved pulmonary function with atopic symptoms but differed in airway symptom frequency. Cluster 4 had mild impairment of pulmonary function with frequent sputum production in smokers. Cluster 5 showed moderate pulmonary function impairment without exertional dyspnea, mainly in older former smokers. Cluster 6 had pulmonary impairment with dyspnea and obesity and included adults older than those in clusters 1-4. Clusters 2, 3, and 5 were genetically predisposed to asthma, while clusters 4 and 6 had higher mortality. Similar patterns were noted in RS.

CONCLUSIONS

Six clinically interpretable phenotypic subgroups based on smoking history, pulmonary function, and symptoms were identified and externally validated, highlighting the importance of classifying patients based on shared observable characteristics to inform on disease susceptibility and prognosis.