Diffusing Capacity and Mortality in Chronic Obstructive Pulmonary Disease

Abstract

Rationale

Chronic obstructive pulmonary disease (COPD) mortality risk is often estimated using the BODE (body mass index, obstruction, dyspnea, exercise capacity) index, including body mass index, forced expiratory volume in 1 second, dyspnea score, and 6-minute walk distance. Diffusing capacity of the lung for carbon monoxide (Dl_CO) is a potential predictor of mortality that reflects physiology distinct from that in the BODE index.

Objectives

This study evaluated Dl_CO as a predictor of mortality using participants from the COPDGene study.

Methods

We performed time-to-event analyses of individuals with COPD (former or current smokers with forced expiratory volume in 1 second/forced vital capacity < 0.7) and Dl_CO measurements from the COPDGene phase 2 visit. Cox proportional hazard methods were used to model survival, adjusting for age, sex, pack-years, smoking status, BODE index, computed tomography (CT) percent emphysema (low attenuation areas below −950 Hounsfield units), CT airway wall thickness, and history of cardiovascular or kidney diseases. C statistics for models with Dl_CO and BODE scores were used to compare discriminative accuracy.

Results

Of 2,329 participants, 393 (16.8%) died during the follow-up period (median = 4.9 yr). In adjusted analyses, for every 10% decrease in Dl_CO percent predicted, mortality increased by 28% (hazard ratio = 1.28; 95% confidence interval, 1.17–1.41, P < 0.001). When compared with other clinical predictors, Dl_CO percent predicted performed similarly to BODE (C statistic Dl_CO = 0.68; BODE = 0.70), and the addition of Dl_CO to BODE improved its discriminative accuracy (C statistic = 0.71).

Conclusions

Diffusing capacity, a measure of gas transfer, strongly predicted all-cause mortality in individuals with COPD, independent of BODE index and CT evidence of emphysema and airway wall thickness. These findings support inclusion of Dl_CO in prognostic models for COPD.

Chronic obstructive pulmonary disease (COPD) is the fourth leading cause of death in the United States, with approximately 46.3 deaths per 100,000 people per year attributable to COPD in 2020 (1–3). The BODE index, commonly used to predict mortality risk in COPD, incorporates measures of body mass index, airflow obstruction, dyspnea, and exercise performance by 6-minute walk distance (4).

One key clinical physiologic measure that is frequently obtained, but not yet incorporated into routine prognostic models, is diffusing or transfer capacity of the lung for carbon monoxide (Dl_CO). Dl_CO is a measure of gas transfer across the alveolar-capillary membrane and provides independent and complementary physiologic information to the more frequently used prognostic measures, such as forced expiratory volume in 1 second (FEV₁) and imaging evidence of emphysema. Prior studies both in healthy and in COPD-specific populations have offered suggestions that Dl_CO is useful in predicting mortality (5–8), but with the advancement of deeper phenotyping in COPD, particularly imaging measures, the independent utility of Dl_CO beyond reflecting emphysema is less well understood.

This study seeks to evaluate diffusing capacity as a predictor of mortality among individuals with COPD and to compare Dl_CO with other common predictors of disease severity and prognosis. We hypothesize that diffusing capacity will be an independent predictor of mortality and will offer additional value to existing predictors of death among individuals with COPD. The COPDGene study offers a large, well-characterized cohort with a wide range of disease severity in which to assess Dl_CO as a predictor of mortality in COPD.

Methods

Study Population and Design

COPDGene is a multicenter prospective observational cohort study of individuals ages 45–80 years with a smoking history of ⩾10 pack-years from 21 clinical centers (9). Institutional review boards at all centers approved COPDGene, and all participants provided written informed consent. Enrollment occurred from 2007 to 2012, with subsequent in-person visits every 5 years. Dl_CO testing was initiated at the visit 5 years after initial enrollment. The current investigation is a time-to-event survival analysis of participants with COPD (defined as FEV₁/forced vital capacity < 0.7 and ⩾10 pack-year smoking history) who underwent Dl_CO testing.

Physiologic Testing

Physiologic testing was conducted on all participants including spirometry, Dl_CO, and 6-minute walk testing in accordance with American Thoracic Society/European Respiratory Society guidelines (10–12). Tests that were judged acceptable and reproducible were included in analyses; sensitivity analyses were conducted including individuals with failed Dl_CO measurements. Dl_CO and FEV₁ percent predicted values were calculated using Global Lung Initiative equations (13, 14), with Dl_CO values adjusted for site altitude and hemoglobin (13).

Survival Data

Vital status was obtained from the Social Security Death Index and the COPDGene longitudinal follow-up program. Survival time was calculated from the date of Dl_CO measurement to the date of death or censoring. Observations were administratively censored on December 18, 2016. For participants where vital status was searched using the Social Security Death Index, follow-up time was back-censored 3 months before the last Social Security Death Index search to account for lags in reporting, whereas for survival data obtained from longitudinal follow-up, censoring occurred at the time of most recent active participation. Cause of death was adjudicated by a committee using all available evidence of record, including death certificates. Towards a Revolution in COPD Health (TORCH) guidelines were used to classify cause of death into five categories: respiratory, cardiovascular, cancer, other, and unknown (15).

Computed Tomography (CT)

CT scans of the chest were performed as per previously published protocols on individual site CT scanners (16). Total percent emphysema (%LAA₋₉₅₀), defined as the percentage of voxels with attenuation equal to or below −950 Hounsfield units, and airway wall thickness (Pi10), defined as the square root wall area of an airway with a 10-mm internal perimeter, were calculated using Thirona software (Thirona, Nijmegen, the Netherlands; http://www.thirona.eu).

Statistical Analysis

Participant characteristics were described by vital status and by Dl_CO percent predicted. With Kaplan-Meier methods and log-rank testing, unadjusted survival was compared between individuals with Dl_CO above or below 50% predicted, a cutoff selected on the basis of previous work (17), and above or below the lower limit of normal (LLN). Dl_CO percent predicted was also treated as a continuous variable in Cox proportional hazard models. To verify linearity in the association between Dl_CO and log hazard of mortality, observed hazard ratios for bins of every 10% predicted Dl_CO were plotted alongside an unadjusted Cox model.

Multivariable Cox models were built with additional clinical predictors, including age, gender, body mass index, smoking status, pack-years, FEV₁ percent predicted, %LAA₋₉₅₀, Pi10, and comorbidities. Comorbidities included self-reported diabetes, chronic kidney disease, and cardiovascular disease, a composite variable encompassing any report of coronary artery disease, hypertension, or congestive heart failure. All covariates were selected a priori, on the basis of plausible mechanism and clinical relevance. With a relatively low rate of mortality in the population, covariates were minimized for parsimony and tested for multicollinearity using variance inflation factors (18, 19).

Models were constructed using BODE index scores in addition to the individual components. The BODE score was dichotomized above or below 4 points, based approximately on an estimated 50% survival at 4 years (4). One hundred ninety-four participants were missing %LAA₋₉₅₀ and Pi10 data; sensitivity analyses were conducted that categorized participants with emphysema into those with ⩽5%, with >5%, and data missing, and Pi10 into 92.5 mm (median value), >2.5 mm, and data missing. Additional analyses evaluating carbon monoxide transfer coefficient and smokers without COPD were also performed.

We compared the discriminative performance of Dl_CO percent predicted individually and in combination with other predictors using Harrell’s C statistics (20–22), calculated by somersd and estat concordance (STATA; StataCorp LLC, College Station, TX). The former uses a jackknife approach with conservative assumptions regarding right-censored individuals (23–25). C-statistics calculated using estat concordance are provided in the online supplement. To evaluate differences in cause of death, percentages within each TORCH category (15) were graphed among individuals with Dl_CO above or below 50% predicted and those who failed Dl_CO measurement.

Results

Participant Characteristics

The study population consisted of 2,329 COPD participants with an attempted Dl_CO and available follow-up survival data, 2,082 of whom had acceptable and reproducible Dl_CO measurements (Figure 1). There were 393 (16.8%) deaths during the observed follow-up time, with a median follow-up of 4.9 years (interquartile range, 3.6–5.8 yr). Baseline participant characteristics by vital status are noted in Table 1. Individuals who had died during follow-up were older, at a mean age of 71; had a lower average body mass index of 26.2 kg/m²; and had a higher number of smoking pack-years than those who were alive during follow-up. Sixty-one percent of those who died were male, compared with 54% male among those still alive at the end of follow-up. There was severe airflow obstruction (FEV₁, 49 ± 21% predicted), moderate to severe gas transfer impairment (Dl_CO, 53 ± 22% predicted), and increased CT evidence of emphysema (median, 10.9%) in deceased individuals, compared with milder impairments in airflow obstruction and diffusing capacity (FEV₁, 64 ± 22% predicted; Dl_CO, 69 ± 22% predicted), and less CT emphysema (median, 4.9%) in those still alive at the end of follow-up. Furthermore, comorbidities including coronary artery disease, hypertension, diabetes, congestive heart failure, and chronic kidney disease were more commonly reported among those who died.

Figure 1.

Study population. Selection of study population with description of excluded individuals. From the study population, 195 individuals were missing computed tomography emphysema or airway wall thickness measurements. Dl_CO = diffusing capacity of the lung for carbon monoxide; FEV₁ = forced expiratory volume in 1 second; FVC = forced vital capacity; GOLD 0 = global initiative for obstructive lung disease spirometry grade 0; P2 visit = phase 2 visit of the COPDGene study; PRISM = preserved ratio impaired spirometry.

Table 1.

Clinical characteristics at phase 2 visit by vital status

Characteristic	Died (n = 393)	Alive (n = 1,936)
Age, yr	71 ± 8	67 ± 8
Female, n (%)	154 (39)	886 (46)
African American, n (%)	85 (22)	444 (23)
BMI, kg/m2	26.7 ± 5.9	28.2 ± 6.2
Smoking status, n (%)
Former smoker	258 (66)	1,252 (65)
Current smoker	133 (34)	668 (35)
Pack-years, median (IQR)	51 (40–70)	45 (33–61)
FEV1, % predicted, mean ± SD	49 ± 21	64 ± 22
FVC, % predicted, mean ± SD	74 ± 19	85 ± 20
DlCO, % predicted, mean ± SD	53 ± 22	69 ± 22
KCO, % predicted, mean ± SD	62 ± 24	75 ± 21
Emphysema, %, median (IQR)	10.9 (3.5–24.3)	4.9 (1.6–13.8)
Airway wall thickness, mm, mean ± SD	2.7 ± 0.6	2.5 ± 0.6
Resting oxygen saturation, %, mean ± SD	94 ± 4	95 ± 3
Coronary artery disease, n (%)	70 (18)	186 (10)
Diabetes, n (%)	75 (19)	308 (16)
Hypertension, n (%)	244 (62)	1,009 (52)
Congestive heart failure, n (%)	42 (11)	75 (4)
Sleep apnea, n (%)	73 (19)	331 (17)
Kidney disease, n (%)	24 (6)	54 (3)
BODE index score, median (IQR)	5 (2–7)	2 (0–4)

Definition of abbreviations: BMI = body mass index, BODE = body mass index, obstruction, dyspnea, and exercise capacity; Dl_CO = diffusing capacity of the lung for carbon monoxide; FEV₁ = forced expiratory volume in 1 second; FVC = forced vital capacity; IQR = interquartile range; KCO = carbon monoxide transfer coefficient; SD = standard deviation.

Values are presented as mean ± SD, n (%), or median (IQR).

Additional participant characteristics by Dl_CO noted increased airflow obstruction, CT percent emphysema, oxygen use, preponderance of congestive heart failure, and BODE score among those with either a failed Dl_CO maneuver or a Dl_CO below 50% predicted as compared with those above 50% predicted (Table E1).

Diffusing capacity is a predictor of survival, independent of airflow obstruction and emphysema

Unadjusted 1-, 3-, and 5-year survival in the study population was 98%, 91%, and 82%, respectively. Participants with Dl_CO ⩽50% predicted had significantly worse mortality than those with Dl_CO >50% predicted, log-rank χ²(1, N = 127), P ⩽ 0.001) (Figure 2A). Using the LLN as a threshold, individuals with a Dl_CO < LLN also had significantly worse mortality than those with Dl_CO ⩾ LLN, log-rank χ²(1, N = 56), P ⩽ 0.001 (Figure 2B). Further, the 247 individuals who attempted a Dl_CO but were unable to generate acceptable and reproducible maneuvers had similar but slightly worse mortality than those with Dl_CO ⩽50% predicted (Figure E1).

Figure 2.

Severe diffusing capacity impairment is associated with mortality in COPD. Kaplan-Meier survival curves by categories of Dl_CO. (A) Solid line represents Dl_CO >50% predicted; dashed line refers to Dl_CO ⩽50% predicted. (B) Solid line represents Dl_CO ⩾ lower limit of normal (LLN); dashed line refers to Dl_CO < LLN. Log-rank test and at-risk table are provided. COPD = chronic obstructive pulmonary disease; Dl_CO = diffusing capacity of the lung for carbon monoxide.

In unadjusted Cox models, Dl_CO percent predicted was a significant predictor of mortality (hazard ratio, 1.37; 95% confidence interval [CI], 1.29–1.44; P < 0.001), along with age, sex, body mass index, pack-years smoked, FEV₁ percent predicted, %LAA₋₉₅₀, Pi10, cardiovascular comorbidities, and chronic kidney disease (Table 2). To verify that Dl_CO percent predicted as a continuous variable maintained a linear association with log hazard of death, without any obvious threshold effect, observed hazard ratios for every 10% predicted Dl_CO were plotted in conjunction with an unadjusted Cox model–predicted hazard ratio (Figure 3). The Cox model–predicted values closely approximated the observed hazard ratios. Notably, the observed data demonstrated a significant increase in hazard ratio below a Dl_CO of approximately 70% predicted as compared with the reference of Dl_CO 100% predicted.

Table 2.

Clinical predictors of mortality

Characteristic	Unadjusted			Adjusted Model 1 (n = 1,854)			Adjusted Model 2 (n = 1,799)
	HR	95% CI	P Value	HR	95% CI	P Value	HR	95% CI	P Value
Age	1.05	1.04–1.06	<0.001	1.06	1.04–1.08	<0.001	1.05	1.04–1.07	<0.001
Sex (female vs. male)	0.75	0.62–0.92	0.006	0.68	0.52–0.89	0.004	0.66	0.50–0.86	0.003
BMI	0.96	0.94–0.98	<0.001	0.99	0.97–1.01	0.409	—	—	—
Pack-years	1.01	1.01–1.01	<0.001	1.00	1.00–1.01	0.336	1.00	1.00–1.01	0.181
Smoking status (current vs. former)	1.12	0.91–1.38	0.28	1.62	1.17–2.23	0.003	1.64	1.20–2.24	0.002
DlCO (per 10% predicted reduction)	1.37	1.29–1.44	<0.001	1.28	1.17–1.41	<0.001	1.26	1.15–1.38	<0.001
FEV1 (per 10% predicted reduction)	1.31	1.25–1.38	<0.001	1.16	1.06–1.27	0.001	—	—	—
%LAA−950	1.03	1.02–1.04	<0.001	0.99	0.97–1.00	0.145	0.99	0.98–1.01	0.276
Pi10	1.68	1.42–1.98	<0.001	1.34	1.03–1.75	0.029	1.43	1.14–1.81	0.002
Cardiovascular disease	1.61	1.29–2.00	<0.001	1.36	1.04–1.79	0.026	1.35	1.02–1.78	0.034
Kidney disease	2.11	1.39–3.19	0.001	2.32	1.42–3.80	0.001	2.16	1.32–3.54	0.002
Diabetes	1.24	0.97–1.60	0.088	1.13	0.81–1.57	0.465	1.05	0.75–1.47	0.778
BODE index (:4 vs. <4)	3.98	3.25–4.89	<0.001	—	—	—	2.36	1.74–3.21	<0.001

Definition of abbreviations: BMI = body mass index; BODE = body mass index, obstruction, dyspnea, and exercise capacity; CI = confidence interval; Dl_CO = diffusing capacity of the lung for carbon monoxide; FEV₁ = forced expiratory volume in 1 second; HR = hazard ratio; %LAA₋₉₅₀ = percent low attenuation areas below −950 Hounsfeld units; Pi10 = airway wall thickness of 10-mm internal perimeter airways.

Unadjusted and adjusted hazard ratios for various clinical predictors of mortality. Adjusted model 1 includes age, sex, BMI, pack-years smoked, smoking status, Dl_CO percent predicted (per 10% predicted reduction), FEV₁ percent predicted (per 10% reduction), computed tomography percent emphysema, airway wall thickness, cardiovascular disease, kidney disease, and diabetes as covariates. Adjusted model 2 includes the predictors of adjusted model 1, but in lieu of BMI and FEV₁ percent predicted, the BODE index is included as a covariate. Bolded values refer to statistically significant results.

Figure 3.

DL_CO is associated with mortality across the range of observed values. Dashed line represents predicted hazard ratios from a Cox proportional hazard model with Dl_CO percent predicted as a continuous variable. Solid line and gray bars represent observed hazard ratios with 95% confidence intervals for categories every 10% predicted Dl_CO. For both predicted and observed hazard ratios, the reference was set at a Dl_CO 100% predicted. Dotted line represents a hazard ratio of 1. Dl_CO = diffusing capacity of the lung for carbon monoxide.

In adjusted models, Dl_CO percent predicted was an independent predictor of mortality, as were FEV₁ percent predicted, age, sex, smoking status, airway wall thickness, and chronic kidney disease (Table 2). Notably, %LAA₋₉₅₀ was no longer a significant predictor of mortality in multivariable analysis. In models including the BODE index, BODE scores were significantly associated with mortality (hazard ratio, 2.37; 95% CI, 1.74–3.22; P < 0.001) and Dl_CO percent predicted remained a significant independent predictor. Sensitivity analyses including individuals with missing %LAA₋₉₅₀ and Pi10 data did not alter the association between Dl_CO percent predicted and mortality, but they further demonstrated smoking status and cardiovascular disease as significant predictors (Table E2). Additional sensitivity analyses were conducted considering hemoglobin, 6-minute walk distance, and modified medical research council dyspnea scores, all covariates independently associated with mortality. These results continued to demonstrate the independent association between Dl_CO percent predicted and mortality with some observed attenuation of the effect magnitude in models including a 6-minute walk distance (Table E3). Similar results were also identified when carbon monoxide transfer coefficient percent predicted was used as a predictor in lieu of Dl_CO percent predicted (hazard ratio, 1.25; 95% CI, 1.15–1.35; P < 0.001) (Table E4).

Additional sensitivity analyses were conducted among smokers with normal spirometry (Global Initiative for Obstructive Lung Disease spirometry grade 0), and smokers without obstruction but with a reduced FEV₁ on spirometry, or preserved ratio impaired spirometry). Kaplan-Meier curves of unadjusted survival among those with a Dl_CO above and below 50% predicted (Figure E2) and adjusted models of survival demonstrated a similar association between survival and Dl_CO (Table E5).

Accuracy of Dl_CO compared with other common predictors in predicting survival

Dl_CO percent predicted was observed to have the highest C statistic of 0.68 (95% CI, 0.64–0.71) among individual predictors, including FEV₁ percent predicted, CT emphysema, and CT airway wall thickness (Figure 4). Additionally, there was no statistically significant difference in C statistics between Dl_CO percent predicted and BODE score (difference, 0.019; 95% CI, −0.01 to 0.05; P = 0.21). The addition of Dl_CO to BODE improved the Harrell’s C statistic to 0.71 (95% CI, 0.67–0.74), with statistically significant differences compared with Dl_CO alone (difference, 0.03; 95% CI, 0.01–0.05; P = 0.001) or BODE alone (difference, 0.01; 95% CI, 0.002–0.02; P = 0.04). No significant improvement in discriminative performance was noted with the addition of CT measures of emphysema or airway wall thickness to Dl_CO and BODE. Harrell’s C statistics were also calculated using more liberal assumptions regarding right-censored data with similar results (Table E6).

Figure 4.

Diffusing capacity performance compared with commonly used predictors of mortality in COPD. Harrell’s C-statistic estimates and 95% confidence intervals for various predictors of mortality, calculated using Somer’s D statistic. Predictors included in each model are listed along the y axis. P value comparisons are against the model of BODE (body mass index, obstruction, dyspnea, exercise capacity) as the sole predictor. *P < 0.05 and ***P < 0.001. COPD = chronic obstructive pulmonary disease; Dl_CO = diffusing capacity of the lung for carbon monoxide; FEV₁ = forced expiratory volume in 1 second; Pi10 = airway wall thickness of 10-mm internal perimeter airways; %LAA-950 = percent low attenuation areas below −950 Hounsfeld units.

Cause of Death

Adjudicated cause of death was available for 107 (28%) of the 378 individuals who died during follow-up. When separated into categories consisting of respiratory, cardiovascular, cancer, other, and unknown primary etiologies of death, approximately 45% (n = 19) of individuals with Dl_CO ⩽50% predicted had a respiratory cause of death as compared with 27% (n = 9) among those who died with a Dl_CO >50% predicted (Figure 5). Notably, individuals with unacceptable Dl_CO maneuvers had a similarly large proportion of deaths due to respiratory causes as compared with individuals with Dl_CO ⩽50% predicted.

Figure 5.

Cause of death by diffusing capacity impairment. Percentages of cause of death among TORCH categories (15), by Dl_CO category. Higher percentages of respiratory-related death were observed among individuals with Dl_CO ⩽50% predicted. Dl_CO = diffusing capacity of the lung for carbon monoxide.

Discussion

Among participants in the COPDGene study, diffusing capacity is a strong predictor of all-cause mortality in COPD, independent of airflow obstruction and CT evidence of emphysema and airway wall thickness. We observed that this measurement of gas transfer offered equal discriminative accuracy in modeling survival as the BODE index, and addition of Dl_CO to the BODE index offered even better performance in this population. As gas exchange is a primary function of the respiratory system, these findings emphasize the clinical relevance of this noninvasive, minimal-risk, and inexpensive tool and strongly support the inclusion of Dl_CO in prognostic models.

Our findings extend prior work that have demonstrated discordant results regarding diffusing capacity as a predictor of mortality. A study of the National Health and Nutrition Examination Survey cohort noted a significantly increased hazard of death, with reductions in Dl_CO beginning at a threshold of approximately 85% predicted (6). The National Health and Nutrition Examination Survey focused on the general population, specifically highlighting associations in those with an FEV₁ over 90% predicted and without clinical diagnosis of respiratory disease. As such, the population studied was younger, and diffusing capacity was overall preserved, with an average Dl_CO of approximately 95% (6). Recently, a study examining 360 individuals with COPD who had an FEV₁/forced vital capacity < 0.7 and an FEV₁ ⩾ 80% predicted demonstrated that Dl_CO less than 60% predicted was associated with increased mortality (26), similar to our own findings. Our study expands on the results from this study, with 22% of our study population including individuals with similar spirometric severity and the remaining individuals comprising participants with moderate-to-severe impairment in FEV₁. Numerous small studies conducted in COPD populations have demonstrated, on average, lower Dl_CO percent predicted values among nonsurvivors as compared with survivors (5, 7, 27–29). By way of contrast, the National Emphysema Treatment Trial, which was a cohort with severe emphysema and severe diffusing capacity impairment, demonstrated in univariate analyses that Dl_CO was associated with mortality, but that effect was no longer significant in multivariable analyses (8). Of note, median Dl_CO in this study was 28% predicted with minimal variability, which may have limited these findings. Additionally, a single-center study conducted in London of 604 outpatient COPD participants demonstrated that Dl_CO was a significant predictor of mortality, after adjusting for spirometry, but did not address further phenotyping such as CT evidence of emphysema (5). Our findings offer strong evidence in a large, well-phenotyped COPD cohort with a wide range of disease severity supporting diffusing capacity as a predictor of mortality.

Although FEV₁ has been well established as a predictor of mortality, our compelling results demonstrated that Dl_CO has an even stronger association with survival. This is a novel finding in the COPD population, where airflow obstruction has traditionally been the mainstay of prognostication. This result suggests that the drivers of mortality in obstructive lung disease may ultimately be related to pathways by which there is impaired gas transfer. Stated another way, as the primary function of the lung is gas exchange, pathology that limits diffusing capacity is also a primary determinant of survival. Studies examining the physiologic mechanisms associated with reductions in Dl_CO as it translates to increased dyspnea have noted reduced ventilatory efficiency, leading to higher inspiratory drive, reductions in inspiratory reserve, and lower peak oxygen uptake during exercise (30–32). Similar mechanisms may contribute to the associations with mortality, although further investigation is needed to elucidate these pathways. Ultimately, recent evidence has suggested that airflow obstruction may be insufficient in predicting morbidity outcomes in COPD (33, 34); our findings extend those results to mortality outcomes and offer diffusing capacity as an additional predictor that improves prognostication.

The present results do not offer a clear pathway by which impairment in diffusing capacity is associated with increased mortality. In light of the results being independent of airflow obstruction and emphysema, one possible explanation relates to additional comorbid conditions that may result in both a lower Dl_CO and increased mortality. One such possibility is the presence of pulmonary vascular disease, with many prior studies suggesting that pulmonary vascular involvement in COPD may be underappreciated (17, 30, 35, 36). It is well known that pulmonary hypertension is associated with increased mortality in COPD as compared with individuals with COPD alone (37, 38). Furthermore, studies have demonstrated impairment in Dl_CO associated with increased mortality among patients with COPD and pulmonary hypertension (39, 40). Similarly, gas transfer impairment is widely observed among individuals with heart failure, both with reduced and with preserved ejection fraction, and has been associated with mortality (41–43). Finally, the Lung Cancer Screening Score–COPD study identified diffusing capacity reduction as a predictor of lung cancer, raising another plausible pathway to explain the observed associations (44). Although we attempted to adjust for comorbidities within the models, there is likely some degree of residual confounding, especially with respect to cancer and pulmonary vascular disease. Further, although the evaluation of cause of death supported predominantly respiratory etiologies among those with severe Dl_CO reductions, nearly three-quarters of the deaths reported did not have adjudicated cause-of-death information, limiting these findings.

Another key finding was that even mild diffusing capacity impairment was found to be consequential with respect to survival. Our study noted reductions in Dl_CO roughly below 70% predicted to be significantly associated with mortality, similar to the National Health and Nutrition Examination Survey, which identified a threshold of 85% predicted in a general population (6), and the study of Global Initiative for Obstructive Lung Disease spirometry grade I individuals, which identified a threshold of 60% predicted (26). These results suggest that Dl_CO is a sensitive marker of early changes in the integrity of the alveolar-capillary interface. Common clinical practice in COPD is to use Dl_CO below 50% predicted as a marker of hypoxemia with exertion (45), knowing that severe hypoxemia is associated with increased mortality (46, 47). Our findings suggest that even mild impairment is relevant well before clinical evidence of hypoxemia. Furthermore, we observed that individuals who fail Dl_CO maneuvers either by acceptability or reproducibility criteria have outcomes similar to those with severe gas transfer impairment. Although traditional use of Dl_CO has focused on severe reductions, our findings highlight the value of more modest impairment and even an inability to obtain a Dl_CO measurement to clinical prognostication.

The present results have a few key limitations. First, the mortality rate in this cohort is lower than those observed in the cohorts used to develop and validate the BODE index score (4). This raises the possibility that these findings only pertain to a healthier population of COPD patients. External validation of these findings in a distinct prospective cohort would address this limitation as well as offer more robust evidence supporting the performance of Dl_CO as a predictor of mortality. There are only a few longitudinal studies that include Dl_CO measurements (48, 49), of which COPDGene has the largest sample size and the necessary power to adequately evaluate mortality. A similar large-scale COPD research study that includes Dl_CO measurements and mortality data is not yet available to serve as a validation cohort. Even a recent 10-year mortality risk score developed from the COPDGene cohort does not include Dl_CO, as measurement was unavailable at the initial enrollment visit (50). One concern that has traditionally been considered and has likely limited inclusion of Dl_CO in research studies is reproducibility. However, analysis of intrasubject variability in COPDGene has demonstrated a low coefficient of variation, at only 3.2% (51). The present findings strongly support the inclusion of Dl_CO in future studies to validate these results and study inclusion of Dl_CO in prognostic models of survival. The relatively infrequent deaths also limited the number of covariates included in multivariable models. As a result, there is likely residual confounding, especially as it pertains to comorbid conditions. Comorbidities were self-reported and noted to be less common than anticipated in a COPD population. Additionally, accurate measures of lung volumes to account for hyperinflation or air trapping were unavailable and may contribute to residual confounding, as a previous study noted associations between the ratio of inspiratory capacity to total lung capacity and mortality (52). Finally, the modest sample size for cause of death analysis limits the ability to draw conclusions regarding the underlying mechanism being described by Dl_CO impairment. However, the findings clearly demonstrate that Dl_CO offers distinct and relevant information beyond two prominent pathophysiologic characteristics of COPD: airflow obstruction and emphysema.

In conclusion, in this well-characterized cohort of COPD participants, we have identified diffusing capacity measurement to be a strong, independent predictor of mortality, similar in its predictive performance to the BODE index. Dl_CO measurement is often obtained in the clinical setting, but its utility has hitherto been poorly understood and therefore has been excluded from multidimensional assessment and prognostic tools in COPD. These findings add to the literature identifying Dl_CO as a marker of increased morbidity and mortality in COPD and further support consideration of Dl_CO in prognostic models. Future studies examining the underlying mechanism reflected by the impairment of gas transfer as it relates to both morbidity and mortality, factors conferring susceptibility to reductions in Dl_CO, and the impact of longitudinal trajectories of Dl_CO on clinical outcomes are necessary. In the context of a growing interest in deep phenotyping and precision medicine in COPD, a simple minimal-risk measurement of gas transfer offers important prognostic information that should be considered in assessing and managing COPD patients.