Chronic obstructive pulmonary disease (COPD) is a major cause of morbidity and mortality, recently becoming the third leading cause of death in the US. [1] Comorbidities are common in COPD [2] and have been shown to be associated with mortality, [3], [4] poor quality of life [5]–[7] and worse health status. [8]

Disease-specific instruments [9]–[11] for measuring comorbidity burden have more utility for predicting clinical events and mortality than general instruments. [12] The COTE (COPD specific comorbidity test) index was recently developed in a COPD population to use comorbidities for predicting mortality, [3] and another index was recently developed to describe general quality of life in COPD. [13] Currently, there is no externally validated disease-specific comorbidity index in COPD for predicting patient-reported clinical outcomes, particularly relevant given the importance of patient-centered outcomes in chronic disease populations such as COPD.

We hypothesized that the wealth of phenotypic data collected in the COPDGene (Genetic Epidemiology of COPD) [14] and SPIROMICS (Subpopulations and intermediate outcome measures in COPD study) [15] studies would enable development and external validation of a score to describe and quantify the impact of comorbidity burden on respiratory disease-specific quality of life. This score could help clinicians with prognostication and risk stratification of COPD in the clinic as well as serve as a valuable research tool.

Using COPD cohorts of two large multicenter studies, we have developed a COPD-specific comorbidity score which adds significant predictive ability to models predicting clinical outcomes in COPD. A simple count of number of comorbidities is a simplified way to quantify comorbidity in COPD and can be easily applied to other cohorts of COPD with different characteristics and spectrum of disease.

Accounting for comorbidity adds value to other known determinants of health status in COPD. After adding the comorbidity count to a model already including age, race, gender, lung function, current smoking status and cumulative smoking history, the model discrimination for SGRQ improved with p-value of <0.01 for statistical comparison with base model. The comorbidity score improved the prediction of health status, and three other outcomes of importance to patients with COPD: dyspnea, exercise capacity and exacerbation risk. The addition of comorbidities leads to changes in AUC values that may appear to be small. However, the AUC is rather insensitive to model improvements and additional predictors need to be strongly associated to change the AUC compared to a base model. Previous studies have found that even small increases in AUC of 0.02 correlate with clinically relevant changes in classification and diagnosis of individuals. [23] Indeed, the regression models (Table 2 and S4 Table) suggest that comorbidities independently and strongly impact quality of life, dyspnea, exercise capacity and exacerbation risk, when added to smoking, gender, lung function, age and race. The fact that prediction is improved with the addition of comorbidity is important and informative in this setting. Overall, it is the combination of the findings of the strong associations in regression models with the findings of the prediction steps of discrimination and calibration that underscores the importance of comorbidity burden on patient-centered outcomes in COPD.

Further, clinical studies of COPD are increasingly focused upon understanding risk for clinical outcomes. Because individuals with COPD have a significant burden of comorbidity [2], [3], [8] which appears to contribute to such poor clinical outcomes, quantifying comorbidity burden is of importance in epidemiologic studies and clinical trials. However, contribution of comorbidities to these outcomes, including outcomes such as dyspnea and quality of life that are not specific to COPD, may differ based upon the underlying chronic disease status. For example, obesity may impact individuals with COPD by causing more airways inflammation and change in lung volumes, whereas in a population with cardiovascular disease obesity may have more impacts through metabolic syndrome leading to endothelial dysfunction. Since it is likely that comorbidities act through different mechanisms in different chronic diseases, there is value in the development of a comorbidity score that is relevant to the specific population of individuals with COPD. Validation of comorbidity score was less significant for the predication of exacerbations because comorbidities likely have more impact on the general markers of health such as dyspnea and quality of life, whereas exacerbations are a disease-specific process related to COPD.

The burden of comorbidities can be established by a simple comorbidity count, which requires no equation or complicated mathematical algorithm. As a result, this kind of score is easy to apply in the clinical or research setting. Other general and specific comorbidity indices require calculations using weights. [3], [12] With internal and external validation, important steps in proving the applicability and reliability of prediction models, we have shown that a simple count of comorbidities can perform just as well as a weighted score or score based upon complicated statistical selection procedures.

Finally, we have shown that the comorbidity count not only predicts important clinical outcomes in the COPDGene study population, when applied to a different study population its predictive capability was equally strong. The SPIROMICS study population used in this analysis differs slightly from COPDGene in that participants are slightly older, with slightly better lung function and fewer current smokers. Given that both studies include individuals from a broad range of COPD severity with large percentage of African Americans and females, this scoring system is broadly applicable in the real world setting across a range of COPD severity, racial makeup, and gender. All of these characteristics demonstrate that this comorbidity count is generalizable. Notably, we tested our comorbidity index on the population of at-risk smokers without COPD who are enrolled in both the COPDGene and SPIROMICS cohorts and found that there may be utility of this index in describing comorbidity in the at-risk population of smokers without spirometric evidence of COPD, though our findings are predictably less robust in at risk populations than in the population with COPD (S5 Table and S6 Table).

Our study has limitations. First, we relied on self-report to define most comorbidities for this study. For large scale studies, assessment of comorbidity using medical records is cumbersome and costly. The validity of self-report does differ based upon the characteristics of the population studied and the comorbidity considered [24]–[29] and should be considered when interpreting the results of this study. In addition, we are limited by the lack of data on severity of comorbidities, and have assessed them simply as present or absent. More quantitative information on comorbidity severity including data on how comorbidities are treated could have the potential to enhance the utility or performance of such a score. Also, we did not include depression and anxiety as comorbidities since these data were not available. The other existing comorbidity index to predict generic quality of life, the COMCOLD index, [13] showed that depression and anxiety are among the strongest predictors of health-related quality of life. Additionally, if data on severity and treatment of comorbidities were available its possible that a weighted score would have been more useful in predicting outcomes. We also did not have data on cognitive impairment, an increasingly recognized comorbidity of the COPD population. [30] Further, though mental illness such as depression and anxiety were assessed in SPIROMICS, lack of data on this comorbidity in COPDGene limited our ability to include it in our index. However, when we incorporated these conditions for the external validation steps (requiring recalibration of the index specific to the SPIROMICS population), the comorbidity count continued to be comparable to the more complicated comorbidity index measures (data not shown). Further, lack of data on mental illness, as well as insufficient data on malignancies in COPDGene limited our ability to compare the performance of the comorbidity count to indices of comorbidity in COPD such as the COTE index. Finally, we have constructed and validated this score in a broad range of COPD patients. However, given the increasingly recognized heterogeneity of COPD and the likelihood that some comorbidities are more common in subsets of COPD patients [31], it is possible that our score may perform slightly better or worse in any one subset of COPD patients. In addition, because of the variability and inconsistency in the comorbidities assessed in these two study populations, it was impossible to formulate and validate an index with specific cutoffs for the prediction of clinical endpoints in COPD. Moving forward the next step would be to develop and externally validate such an index in study populations with consistent data on a broad range of comorbidities, testing cutoffs for the prediction of relevant clinical outcomes such as quality of life and dyspnea.

Our study did have several strengths. First, these results derive from two large, multicenter, well-characterized and fairly representative populations of individuals with COPD, making our results generalizable. Second, our finding that comorbidity in COPD can be quantified with such ease has the potential to improve and simplify the measurement of comorbidity in future studies and thus can have far-reaching effects. Next, we have focused upon clinically-relevant outcomes, which are important to patients with chronic disease such as COPD. Finally, by validating our score internally and externally we have undertaken an important step not demonstrated before in the COPD comorbidity literature.

We have shown that a comorbidity count in individuals with COPD is not only simplified, but when added to known risk factors is also predictive of important outcomes such as health status, dyspnea, exercise capacity and exacerbation risk. Moving forward, validation of a similar index in populations with a broad range of COPD severity could add generalizability and validity to our preliminary findings that a comorbidity count is a simple and relevant way to quantify comorbidity burden in COPD. Ultimately, such an index can have broad applicability to help quantify comorbidity burden in COPD clinical research. More importantly, such a score can assist clinicians to identify populations with COPD at risk for poor health related outcomes, so that interventions that improve quality of life, dyspnea and exercise capacity, such as rehabilitation, can be considered earlier.