DALLAS--Results of 3 studies presented at the American Thoracic Society meeting this week focus on the predictive role of disease exacerbations, machine learning, and Hgb levels in COPD.
New research on predicting outcomes in patients with chronic obstructive pulmonary disease (COPD) was presented in a session titled “COPD: Mortality and Risk Prediction” at the American Thoracic Society Meeting 2019 held May 17 – 22, 2019, in Dallas, Texas.
Here we highlight studies on the role of respiratory exacerbations; the predictive accuracy of a new machine learning tool; and, hemoglobin as a biomarker of clinical outcomes.
Exacerbation frequency predicts mortality
An analysis of the COPDGene cohort, which included more than 10 000 current and former smokers with or without COPD, revealed possible predictors and causes of mortality in this patient population. Study authors posited that prediction and causality would depend on severity of airflow restriction.
The most common causes of death among smokers without COPD fell into the miscellaneous category, Other (33%), followed by cardiovascular (25%) and lung cancer (10%). By contrast, the most common causes of death among smokers with GOLD stage 1 or stage 2 COPD were cardiovascular (22%) and respiratory (20%) and most smokers with GOLD stage 3 or stage 4 COPD died from respiratory disease (61%).
In an adjusted analysis, having at least 2 respiratory exacerbations during the past year was predictive of all-cause mortality for GOLD stage 1 or stage 2 COPD (hazard ratio [HR]=1.44; P=0.05) and GOLD stage 3 or 4 COPD (HR=1.61; P<.001), but not for smokers without COPD (HR=0.89; P=0.70).
Chronic bronchitis was predictive of all-cause mortality for smokers without COPD (HR=1.47; P=0.009) and smokers with GOLD stage 3 or stage 4 COPD (HR=1.35; P=0.003), but not for smokers with GOLD stage 1 or stage 2 COPD (HR=1.11; P=0.50).
“We affirmed that reducing the exacerbation rate remains an important clinical and research priority for patients with COPD,” said the study presenter Wassim Walid Labaki, MD, Division of Pulmonary and Critical Care Medicine, University of Michigan.
Machine learning beats the BODE index. Results of this new study found that a machine learning algorithm combined with clinical and quantitative CT imaging features predicted mortality among COPD patients better than the traditionally used BODE (Body-mass index, airflow Obstruction, Dyspnea, and Exercise) index and other prognostic indices.
“While the BODE index may have some limitations, despite many people trying, no one has actually successfully outperformed the BODE index consistently,” said the study presenter Matthew Moll, MD, Channing Division of Network Medicine, Brigham and Women’s Hospital, Boston, Massachusetts.
To develop the machine learning model, patients with GOLD 2 to 4 COPD from the COPDGene and the ECLIPSE cohorts were used. Three-quarters of the COPDGene cohort (n=1,974) were used for the training set and the remaining quarter (n=658) were used for the testing set. The algorithm was then externally validated in the ECLIPSE cohort (n=1,645).
The most important predictors of mortality in the model were 6-minute walk distance, FEV1, and age. The machine learning model outperformed 4 other prognostic indices, including BODE, updated BODE, e-BODE, and ADO (age, dyspnea and FEV1), for both the testing set (P<.05) and externally validated set (P<.05).
“To put this in clinical context, if you use our model compared to BODE, for every 26 patients that you use our model in, 1 additional person would be predicted correctly to die at 5 years,” said Dr. Moll.
Anemia linked to worse outcomes. Among patients with COPD, anemia was independently associated with worse outcomes, including symptoms, quality of life, and functional performance, according to results of a cross-sectional study presented during the session.
“Because there are new therapies that are available for the treatment of chronic inflammation-related anemia, understanding hemoglobin as a biomarker across its continuous range is pretty important,” said study presenter Aparna Balasubramanian, MD, Pulmonary and Critical Care Medicine, Johns Hopkins University, Baltimore, Maryland.
The cross-sectional study used data from the COPDGene cohort for which hemoglobin data were available. On the basis of hemoglobin level, patients were categorized as normal, anemic (Hgb <12 g/dL for females and Hgb <13 g/dL for males), or polycythemic (Hgb >16 g/dL for women, Hgb>16.5 g/dL for men).
In an adjusted analysis, compared with a normal hemoglobin level, anemia was associated with worse scores on St. George's Respiratory Questionnaire (which assesses quality of life), worse scores on the Short Form (36) Health Survey for physical and general function, and decreased distance on the 6-minute walk test, whereas polycythemia was not. Continuous analyses showed that the relationship between hemoglobin and COPD morbidity is likely nonlinear.