NEJM - The BMI, Airflow Obstruction, Dyspnea, and Exercise Capacity Index in COPD (Celli et al 2004)

Since I'm on the Pulmonary Service this month, thought I'll review some recent Pulmonary papers.

This prospective study consisted of 2 phases. In the first phase, 207 patients enrolled from 1995 to 1997 were assessed. Their baseline characteristics were studied for their ability to predict one-year mortality. A forward step-wise multiple logistic regression model was built, which contained 4 baseline variables (BODE):

1. BMI


2. Obstruction: FEV1 post-bronchodilator treatment


3. Dypnea: MMRC dyspnea sclae


4. Eercise tolerance: 6-minute walk distance

This model was then validated in a cohort of 625 patients rolled from 1997 to 2003. Their baseline BODE characteristics were recorded, and they were then followed for at least 2 years or until death. The ability of the BODE index to predict the hazard of death (either all-cause or from respiratory illnesses) was assessed with the Cox proportional hazards model. They conclude that:

1. For each 1-point increase in BODE index, a 34% increase in hazard of all-cause death (26%-42%, p<0.001)


2. For each 1-point increase in BODE index, a 62% increase in hazard of respiratory-cause death (48%-77%, p<0.001)

The BODE index was found to be a much better prognostic factor than FEV1 alone.

One can raise the reasonable objection that since FEV1 is a component of the BODE index, it is not surprising that BODE is a better predictor of death. Celli et al quantified the incremental improvement provided by BODE using the C-statistic, based on the area under the ROC, and found a substantial difference (0.74 vs 0.65). More on this later...

Another objection may be that the 3 variables besides FEV1 simply measured the overall state of health of the patient. As such, the worse one's health was to begin with, the higher risk of death was expected. Certainly this argument must to true in principle. After all, all physiological or functional assessments reflect some aspect of "overall state of health". Celli et al used regression to assess whether BODE provides information that quantitatively captures the important aspects of overall state of health. They made use of the Charlson index, which quantifies the degree of co-morbidity. In a Cox proportional hazards model, the Charlson index was not a significant predictor of death from respiratory illnesses when BODE score is also included. That is, the Charlson index does not provide prognostic information above and beyond that contained in the BODE index. The BODE index has successfully captured important prognostic imformation. Since BODE is focused on cardiopulmonary health, it is not surprising that the Charlson index did have small but marginally significant prognostic value for predicting all-cause mortality.

As the authors noted themselves, there are significant differences between patients from the US (N=348), Venezuela (N=54), and Spain (N=223). One would like to have seen an analysis restricted to the US patients and non-US patients to assess the rebustness of their conclusion.

The BODE Calculator from ICU Medicus is fun to play with.

ACP Journal Club review

medicine statistics pulmonology COPD