INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The single-breath diffusing capacity of the lungs for carbon monoxide (DL_CO) is frequently used in the differential diagnosis of patients with dyspnea in the clinical setting (1), for the characterization of subjects in epidemiological studies of obstructive and restrictive lung diseases (2), and for surveillance in the occupational setting (3). Among patients who have airflow limitation owing to cigarette smoking, diffusing capacity is highly correlated with the degree of emphysema on lung computed tomography (CT) scans (4). In recognition of the clinical importance of the DL_CO test, the American Thoracic Society has set standards for the performance of DL_CO tests (5, 6) and their clinical interpretation.

Changes in DL_CO over months to years may be important in following the course of chronic obstructive and interstitial restrictive lung diseases and the efficacy of interventions. Yet most population studies of the correlates of DL_CO, including smoking status, have been cross-sectional (7). A study of 159 steelworkers in France, with two DL_CO tests 5 yr apart, paradoxically found that the exposed workers had a smaller DL_CO decline (15%) than did 114 unexposed control workers (20%) (12). Changes in DL_CO measured twice 10 yr apart in 122 middle-aged men in London were associated with changes in smoking habit (13). Those who quit smoking had a mean increase in DL_CO, after correcting for their decrease in CO back pressure (carboxyhemoglobin [COHb] levels). The subjects from these previous longitudinal studies were not from a random population sample, and factors other than occupational exposures and changes in smoking status were not examined.

The objective of this study was to determine predictors of longitudinal change in DL_CO during an 8-yr interval of adult participants in the Tucson Epidemiological Study of Obstructive Lung Disease (14).

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The subjects were from a population sample of white men and women and their families living in Tucson, Arizona, followed since 1972 (14). The sampling was a random stratified scheme, which oversampled subjects into the older age categories. Good quality measurements of spirometry and diffusing capacity were obtained from adults (ages 20 to 59) during Survey 7 in 1982-1983 (described as the "baseline" examination in this report) and approximately 8 yr later during Survey 12, using spirometry and DL_CO techniques previously described (8, 9). Subjects were included in this analysis if they had at least one DL_CO measurement, FEV₁ tests results in both Surveys 7 and 12 (1990-1991) (used to compute change in FEV₁ [FEV₁]), and age greater than 20 yr. A standardized respiratory symptom questionnaire was administered at each examination (14). Smoking status at each examination (current, former, or never smoker) was determined by the answer to the question, "Have you ever smoked cigarettes regularly? 1) Yes, and I still smoke, 2) Yes, but I no longer smoke, or 3) No, I never smoked." Quitters were defined as those currently smoking at the baseline examination and self-reported as no longer smoking at the follow-up examination. Exact quit dates were not reported. Pre- and postbronchodilator spirometry was performed before DL_CO tests, using the same calibrated, heated pneumotachometer at each examination, as previously described (8). However, only the prebronchodilator spirometry values were used in the current analysis.

DL_CO Test Methods

The same W.E. Collins (Braintree, MA) DS model automated system was used during both surveys with no changes in the procedures. The same nurse technician was in charge. The washout volume was 1.0 L, except for participants with a vital capacity of less than 2.0 L for whom a washout volume of 500 ml was used. The test gas was 0.3% CO, 10% helium, balance nitrogen. The participants were seated, wearing nose-clips, and performed at least two DL_CO maneuvers separated by more than 5 min. Examinations were usually scheduled during early morning hours.

The mean DL_CO value from two maneuvers that matched within 2 ml/min/mm Hg was reported. Maneuvers with a breath-holding time < 9 or > 11 s, or with an inspiratory capacity less than 85% of the largest previously measured vital capacity were excluded. No corrections were made for hemoglobin (Hb) or CO back pressure in this analysis, but smokers were asked not to smoke for 4 h prior to their clinic visit. A two-point calibration of the CO and helium analyzers was done just prior to testing of each participant. Leak and volume calibration checks and biologic controls were done weekly.

Statistical Methods

The longitudinal DL_CO data were analyzed using repeated measures analysis. Because of missing observations, we used a random effects model (REM) (15). Jones and Boadi-Boateng (16) recently demonstrated how the REM procedure, which is generally used for unequally spaced continuous longitudinal data, can also be used for equally spaced data by implementing a "saturated" model. The saturated model fits polynomials that have order one less than the number of time points to each subgroup. For our data this meant fitting a linear or first-order polynomial because there were two time points. A saturated model allows both time points for each subgroup to have a different mean estimate. The likelihood ratio test, which was calculated as the change in 2ln(likelihood) between nested models, was used to test for significant differences between groups. For example, to test if the mean DL_CO estimates of current smokers had a different slope than that of the reference group or nonsmokers, two models would be fit. A model fitting both groups with linear polynomials would be compared with a reduced model where the reference group was fitted with a linear polynomial and the current smokers were fit with a model containing only an intercept term. Here, a significant increase in 2ln(likelihood) between these two models would indicate that current smokers have a significantly different group by time interaction or slope than the reference group. If the change in 2ln(likelihood) was not significant but the intercept coefficient estimated for current smokers was, this would imply that the two groups had similar slopes but that the current smokers differed by a constant amount at both time points. The intercept coefficient would thus be an estimate of this difference, which we call a significant group difference. For instance, if this latter intercept coefficient was not significantly different from zero, this would indicate no significant differences between the mean DL_CO values of current smokers and the nonsmokers, at either time point.

In the REM analyses initial FEV₁, FEV₁, sex, change in weight (weight), height, initial age, pack-years, and quit smoking variables were all entered as fixed covariables. Current and ex-smoking were included as time-dependent indicator variables. Two different forms of within-subject covariance structures were considered for each REM analysis: first-order autoregressive and uncorrelated or independent. The best-fitting structure was determined using Akaiki Information Criteria (AIC) with the model fully saturated for all covariables. For all analyses the uncorrelated covariance structure yielded the best overall fit.

Because we wanted an estimate of the true longitudinal slope, which based on plots of the raw data (not shown) appeared to differ from the cross-sectional slope with age, we also included DL_CO test dates as independent variables. The Survey 7 date was arbitrarily set to zero and the Survey 12 date was calculated as the change in time since the Survey 7 test. However, including initial age in the model allowed us to determine if the longitudinal slope estimates changed with age, an indication of acceleration in the DL_CO growth curves. Results from the REM analyses are presented as mean and standard errors of the means, instead of as actual coefficients because these are not directly interpretable. Statistical hypotheses tests were two-tailed comparisons at the  = 0.05 significance level.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

There were 543 subjects who met the inclusion criteria of having at least one DL_CO measurement, at least two FEV₁ and weight measurements (used to calculate the change variables), and age greater than 20 yr. These subjects provided a total of 959 DL_CO measurements. The selected population consisted of 41.3% males; 6.3% of the subjects had reported quitting smoking between surveys (Table 1). The mean age of participants was approximately 50 yr and the mean rate of decline of FEV₁ was 41.5 ml/yr. The percentage of current smokers declined between the initial and follow-up surveys and the number of ex-smokers demonstrated a corresponding increase. Current smokers had more smoke exposure (pack-years) than ex-smokers.

The results of fitting the REM model to the longitudinal DL_CO data are listed in Tables 234. Table 2 shows the mean DL_CO estimates for each survey and corresponding slope estimates stratified by sex, smoking, and pack-year categories, adjusted for the other variables in the model. The slopes were calculated as the difference between the Survey 12 and Survey 7 DL_CO mean estimates divided by the average time between surveys (8.0 yr). Males had statistically significantly higher levels of DL_CO and steeper rates of decline over time than females. Subjects who were currently smoking had significantly lower DL_CO levels than never smokers, which were reduced even further with increases in pack-years of exposure. However, smokers did not have significantly different slopes over time than never smokers. The effects of quitting smoking, changes in weight, and being an ex-smoker were not statistically significant; this implies that the mean DL_CO values for these factors were not significantly lower at either survey and thus it follows that the mean slope over time also did not differ from that of never smokers.

Females had significantly lower DL_CO levels and rates of decline than males, at all ages (Table 3). These results illustrate a very marked acceleration in DL_CO decline with age, with mean rates of decline in DL_CO per year more than doubling between ages 20 and 70 yr, for both sexes. Although females had lower levels of DL_CO, the fact that their line is parallel to that of males (Figure 1) indicates a similar rate of acceleration.

View larger version (11K): [in this window] [in a new window]   Figure 1.   Slope of DLCO [(ml/min/mm Hg)/yr] with age for never smokers assuming mean changes in FEV1 of 49 and 36 ml/yr, for males and females, respectively. Plots illustrate the acceleration in slope of DLCO that occurs in both male (circles) and female (squares) adults.

The effects of initial FEV₁ and changes in FEV₁ were statistically significant, independent of sex. The results for changes in FEV₁ are demonstrated for males (Table 4) assuming a range of changes between the mean ± 2 SD (mean FEV₁ = 0.049 and SD = 0.037 L/yr, for males). These means (Table 4) were calculated assuming the mean age of 50 yr. The DL_CO slope over time ranged from 0.29 to 0.70 for changes in FEV₁ from 27 up to 126 ml/yr, respectively. Initial FEV₁ was statistically significant with a positive coefficient, indicating that higher levels of FEV₁ are associated with higher levels of DL_CO. Higher levels of FEV₁ were not associated with different DL_CO slopes.

The REM procedure used in this analysis allowed us to include subjects who had one or more DL_CO measurements. There were 127 subjects who had only a single assessment. These individuals could not influence the slope estimates because a single observation contains no slope information; however, including their measures could influence the position of the fitted regression line (i.e., level) and it also improves the statistical power. To determine the influence of including these subjects, we reran the REM analysis excluding their data. This reduced the number of subjects from 543 to 416; however, none of the reported findings were significantly changed. We also wanted to test if including the change in FEV₁ in the DL_CO analysis influenced the estimated smoking effects on DL_CO, since smoking reduces FEV₁. For this evaluation we reran the REM analysis excluding FEV₁ and found no substantial changes (< 5%) in the estimated smoking coefficients.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

A previous report of baseline cross-sectional DL_CO results from this cohort showed that current smokers had lower mean DL_COs than former and never smokers, and that current smokers with airflow limitation had even lower DL_CO values than those with normal spirometry (9). DL_CO was inversely associated with pack-years of smoking in both current and former smokers. A history of smoking cessation in former smokers with normal spirometry was associated with DL_CO values close to those of never smoking men and women. Analysis of a few of the subjects who had been tested 5 yr previously showed mean declines in DL_CO, but no significant difference in the declines between the 15 never smokers and 13 continuing smokers (9).

Our longitudinal results from the larger cohort, now after 8 yr of prospective follow-up, show that the 34 adults who quit smoking during this period did not differ from never smokers in either level or slope during the 8-yr follow-up. Never smokers, current smokers, and ex-smokers had similar declines in DL_CO but only current smokers differed significantly in their mean levels. This suggests that the lung damage caused by smoking had occurred prior to initial DL_CO testing, thus not affecting the rate of decline. These results are similar to those reported by Watson and coworkers who measured DL_CO in 122 middle-aged men 10 yr apart (13). The 17 men who quit smoking in the first 2 yr of follow-up had only a negligible change in absolute mean DL_CO (corrected for changes in Hb, and estimated carbon monoxide back pressure), whereas the 42 men who continued to smoke and 42 never smokers had different levels of DL_CO, with smokers being lower, but the two groups having similar slopes over the 10-yr period.

The mechanism for the association of FEV₁ and DL_CO is unknown, but we suspect that it is due to susceptibility of smokers to develop both airways obstruction and emphysema. Another possible explanation for the association of the decrease in DL_CO with a decrease in FEV₁ is that the "effective" alveolar volume which is measured during the single-breath DL_CO test (VA_eff) is an underestimate of the subject's true TLC in smokers who are developing chronic obstructive pulmonary disease (COPD) (as defined by a low FEV₁/FVC or a rapid decline in FEV₁) (17, 18). Some investigators have suggested that in patients with airways obstruction, DL_CO be measured following a bronchodilator to minimize this error, or in patients with restrictive lung diseases and COPD that the alveolar volume (VA) from a body box measurement of TLC be substituted for the VA_eff from the single-breath helium dilution test (19). However, we did not measure the TLC by body plethysmography nor multiple breath methods for this study, which would have allowed us to determine how much of the association of change in DL_CO with change in FEV₁ was related to the underestimation of VA and how much was related to pathologic progression of a lung disease such as COPD.

Limitations of our study include the lack of Hb and COHb measurements, changes in which are known to affect DL_CO results (20). Aging is associated with small mean decrements in Hb which would slightly reduce the DL_CO. Some misclassification of smoking status may have occurred because smoking status was self-reported and not biochemically verified. We did not ask the female participants the date of their last menses, so we could not correct for the effect of the menstrual cycle on DL_CO (21). Some of the participants developed cardiovascular disease that may have caused a change in their DL_CO (22), but we did not consider these diseases in our current analyses. Also, we did not measure DL_CO on a third occasion, so unrecognized survey biases may have occurred as a result of subtle changes in the software.

A recent study by Burgess and colleagues (23) used the REM analysis to estimate the longitudinal declines in DL_CO among Seattle firefighters (n = 812), who had annual DL_CO assessments between 1989 and 1996. They reported that the raw DL_CO levels decreased with age and that current smoking and pack-years of smoking both resulted in significantly lower initial DL_CO levels, which are similar to our findings. In contrast, they reported a positive age by time interaction that suggests a decrease in the rate of decline in DL_CO over time, whereas our results showed an acceleration or increase in the decline of DL_CO with time (Table 3, Figure 1). They also reported a slight decrease in DL_CO by time associated with the average annual number of fires fought. This difference may be due to the fact that their study participants were not from a random population sample, instead the firefighters likely represent a healthier subgroup.

Short-term changes in DL_CO have been reported in a small group of middle-aged smokers (24), indicating a mean increase in DL_CO (about 2 units) within 1 d to 1 wk. The investigators corrected for both changes in Hb and COHb, and found that the DL_CO remained abnormal after 1 mo of smoking cessation in some subjects. Their results suggest that some, but not all of the improvement they saw in subjects who quit smoking was probably caused by a rapid reduction of COHb (CO back pressure) and could explain why we could not find any change in DL_CO related to smoking cessation.

Longitudinal changes in DL_CO were reported in 196 steelworkers with occupational exposure to polluted air and a control group of 186 unexposed workers, after 5 yr of follow-up (12). The mean DL_CO changes were significant: 15% of predicted in the steelworkers and 20% of predicted in the unexposed workers, but the difference between the two exposure groups was not statistically significant. Approximately 83% of the workers were cigarette smokers, but the results were not stratified by (or corrected for) smoking status or change in smoking status. The mean DL_CO changes in their workers were much larger than those noted among our current smokers. A survey bias or other factors could account for this difference, because an abstract of results from this same cohort shows a decrease from 99 to 96% predicted in 112 never smokers over 5 yr, while 426 continuing smokers experienced a DL_CO decline from 94 to 90% of predicted (25).

The changes in DL_CO that we have reported here, even in the subgroup of never smokers, should not be considered normative or "healthy." Many clinical and subclinical diseases that are known to lower the DL_CO, but were unrecognized by our limited exams, including pneumonia, interstitial lung diseases, congestive heart failure, vascular diseases, and anemia, were likely present during the baseline examination in some of our participants and may have developed during the 8 yr of follow-up in others. Some, but not all of the effects of these diseases on lung volumes and airways were measured by declines in the FEV₁ which we measured and which were associated with declines in DL_CO.

We conclude that more rapid declines in DL_CO may be noted in older adults and those who also have excessive FEV₁ declines.