Predicting survival from COPD exacerbations: DECAF score shows promise
DECAF Score Predicts COPD Exacerbation Mortality, But Needs Validation
Despite improvements in care, death during hospitalization for acute exacerbation of COPD (AECOPD) is not uncommon. In the UK in 2008, almost 1 in 12 people admitted with a COPD exacerbation died in-hospital. In the U.S. in 1996, about 1 in 40 people hospitalized with COPD exacerbations died, possibly reflecting a different threshold for hospital admission between the countries. Identifying upon admission those at higher risk of dying in the hospital could be useful for triaging patients to the appropriate level of care, determining aggressiveness of therapies, guiding goals-of-care discussions, and timing safe discharges. To address this need, John Steer, John Gibson, and Stephen Bourke derived the DECAF score—Dyspnea, Eosinopenia, Consolidation, Acidemia, and atrial Fibrillation—to predict in-hospital mortality in a prospectively enrolled community-based cohort of patients admitted to the hospital with acute COPD exacerbations in the UK.
What They Did
From 2008 to 2010 the authors enrolled a relatively unselected (i.e. real-life and generalizable) group of consecutive patients admitted to two hospitals in the UK with acute COPD exacerbations and collected demographic, clinical, radiographic, and laboratory data at or near the time of admission. Importantly, they included patients with evidence of pneumonia (i.e. consolidation on chest x-ray). Next, they screened potential predictors for their association with in-hospital mortality using a p-value threshold, and those selected were submitted to backward stepwise logistic regression to identify (after a bit more tweaking for convenience and simplification) their final model. Model discriminative performance (i.e. its ability to discriminate those with an outcome from those without) was evaluated by the area under the receiver operator curve (AUROC). Although they used bootstrap sampling for internal validation, and demonstrated their results in each individual hospital cohort, the model was derived from the entire cohort and so there was no true external validation cohort.
What They Found
A total of 920 patients were included in the study, of whom 96 (10.4%) died in the hospital. A simple point-score model consisting of 5 predictors cleverly named the DECAF score was very good at predicting the risk of in-hospital death with an AUROC of 0.86 (95% CI 0.82-0.89). Variable Points Dyspnea limiting the patient to home (MRCD 5) and: Independent in bathing and/or dressing (eMRCD 5a) Requires assistance with bathing AND dressing (eMRCD 5b) 1 2 Eosinopenia (<0.05 X 10^9/L) 1 Consolidation (on chest x-ray) 1 Acidemia (pH < 7.30) 1 Atrial Fibrillation (on admission EKG) 1 Here’s a link to the MRC dyspnea scale. The “e” means that category 5 is “extended” into the two subgroups “5a” and “5b” above. Using the model, patients can be categorized into three risk groups:
Low (0-1 point), in-hospital mortality 1.4%
Intermediate (2 points), in-hospital mortality 8.4%
High (3-6 points), in-hospital mortality 34.6%
The DECAF score performed better than the APACHE II, COPD and Asthma Physiology score, BAP-65 score, and the CURB-65. But remember, a model almost always looks the best in the cohort from which it was derived, whereas this is a validation cohort for those other scores.
What It Means
A successful clinical prediction model needs three things: 1) practicality, 2) validity, and 3) utility. Most of the predictors in the DECAF score make sense as to why they might predict mortality in AECOPD, meaning they carry “face validity” (although the eosinopenia concept was new to me), and they are largely objective and reliable. Further, calculating the score requires only one history question, routine labs, a chest x-ray, and an ECG – which most patients admitted with AECOPD should have anyway. So the score seems practical for the clinical setting. However, one major limitation of this study is the lack of validation in a separate cohort of patients (external validation), which is vital to demonstrate generalizability of the model and achieve clinician confidence that it can be used to guide clinical care. Last, a model should have utility, meaning, applying it should change something about what we do. For example, should high risk DECAF patients go to the ICU or be transitioned to comfort care based on patient/family goals? Could low risk DECAF patients safely go to the floor and be discharged after relatively short hospital stays? The authors elaborate on these potential uses, but again, more studies would be needed to support them, and only after the score is well validated. So for now, I’ll keep in mind that it exists (with late-night visits to my hospital’s cafe a reminder), and I might even calculate the score for patients I admit with AECOPD out of interest, but I’ll wait for further validation before making management decisions based on the DECAF score. John Steer, John Gibson, Stephan C Bourke. The DECAF Score: predicting hospital mortality in exacerbations of chronic obstructive pulmonary disease. Thorax 2012;67:970-976.