Symptom phenotyping in people with cystic fibrosis during acute pulmonary exacerbations using machine-learning K-means clustering analysis

Introduction

People with cystic fibrosis (PwCF) experience frequent symptoms associated with chronic lung disease. A complication of CF is a pulmonary exacerbation (PEx), which is often preceded by an increase in symptoms and a decline in lung function. A symptom cluster is when two or more symptoms co-occur and are related; symptom clusters have contributed meaningful knowledge in other diseases. The purpose of this study is to discover symptom clustering patterns in PwCF during a PEx to illuminate symptom phenotypes and assess differences in recovery from PExs.

Methods

This study was a secondary, longitudinal analysis (N = 72). Participants at least 10 years of age and being treated with intravenous antibiotics for a CF PEx were enrolled in the United States. Symptoms were collected on treatment days 1–21 using the CF Respiratory Symptom Diary (CFRSD)-Chronic Respiratory Symptom Score (CRISS). K-means clustering was computed on day 1 symptom data to detect clustering patterns. Linear regression and multi-level growth models were performed.

Results

Symptoms significantly clustered based on severity: low symptom (LS)-phenotype (n = 42), high symptom (HS)-phenotype (n = 30). HS-phenotype had worse symptoms and CRISS scores (p < 0.01) than LS-phenotype. HS-phenotype was associated with spending 5 more nights in the hospital annually (p < 0.01) than LS-phenotype. HS-phenotype had worse symptoms over 21 days than LS-phenotype (p < 0.0001).

Conclusion

Symptoms significantly cluster on day 1 of a CF-PEx. PwCF with HS-phenotype spend more nights in the hospital and are less likely to experience the same resolution in symptoms by the end of PEx treatment than LS-phenotype.