RESULTS

Section:

ChooseTop of pageAbstractMETHODSRESULTS <<DISCUSSIONReferencesCITING ARTICLES

Demographic findings, pulmonary function variables, and CT features are tabulated against smoking status in Table 1. Significant subgroup differences in disease severity across the three groups were largely due to milder disease in current smokers, as judged by pulmonary function variables and the extent of disease on CT. Paired comparisons showed that FVC levels (P < 0.0005) and Dl_CO levels (P = 0.03) were lower, and CT disease extent (P < 0.005) and CPI scores were higher (P < 0.0005) in ex-smokers than in current smokers. Similarly, FVC levels (P < 0.0005) and Dl_CO levels (P = 0.05) were lower, and CT disease extent (P = 0.02) and CPI scores were higher (P < 0.0005) in nonsmokers than in current smokers. There were no significant or marginal differences between former smokers and nonsmokers for the three cardinal measures of disease severity (Dl_CO, CPI, and CT extent), although FVC levels were higher in former smokers (P < 0.005).

Follow-up was complete to death (n = 189, 76%) or to 3 years in 232 of 249 patients (93%). The median survival was 26 months and 5-year survival was 23%.

Survival Compared between Current and Former Smokers

As shown in Table 2 and illustrated in Figure 1, unadjusted survival was higher in current smokers than in ex-smokers (hazard ratio [HR] = 0.44; 95% confidence interval [CI] = 0.24, 0.80; P = 0.007). However, survival differences were reduced with the adjustment (in separate models) for Dl_CO levels (P = 0.03) and the extent of disease on CT (P = 0.08), and were abolished when the CPI was used as the cardinal measure of baseline severity (P = 0.39). Age, sex, and the CT coarseness score had no significant or marginal link with survival in any model.

Figure 1. Mortality compared between current, former, and never-smokers. Current smokers had a lower mortality than former smokers (P = 0.007), but this difference was abolished (P = 0.39) with adjustment for disease severity, using the composite physiologic index (CPI). Former smokers had a higher mortality than nonsmokers (P = 0.008), and this difference was amplified (P < 0.0005) with adjustment for baseline disease severity, using the CPI.

Download Figure | Download Powerpoint

[More] [Minimize]

TABLE 2. SURVIVAL COMPARED BETWEEN CURRENT SMOKERS AND FORMER SMOKERS

	HR	95% CI	Statistical Significance
Unadjusted survival
Current smoker	0.44	0.24, 0.80	P = 0.007
Model containing DlCO
Current smoker	0.51	0.27, 0.95	P = 0.03
DlCO	0.96	0.95, 0.97	P < 0.0005
Model containing CT extent
Current smoker	0.57	0.30, 1.08	P = 0.08
CT disease extent	1.02	1.01, 1.03	P < 0.0005
Model containing CPI
Current smoker	0.75	0.40, 1.43	P = 0.39
CPI	1.05	1.03, 1.06	P < 0.0005

Definition of abbreviations: CI = confidence interval; CPI = composite physiologic index; CT = computed tomography; Dl_CO = diffusing capacity for carbon monoxide; HR = hazard ratio.

Data were analyzed as unadjusted survival, and with adjustment for percent-predicted Dl_CO, the extent of disease on high-resolution CT (HRCT), and CPI levels in three separate models. Age, sex, and the HRCT coarseness score were also included in all models but were not significantly or marginally predictive of survival (P > 0.10).

Findings were unchanged, with no major differences in the statistical strength of trends, with the exclusion of known smokers with (1) a pack-year smoking history less than 10 (n = 14) or (2) no precise pack-year smoking history recorded in the case notes. Findings were also unchanged when the presence of emphysema (current smokers, n = 16; former smokers, n = 65) was taken into account in survival analysis, with the presence of emphysema having no influence on survival.

On the basis of these findings, survival of never-smokers was compared with survival of (1) former smokers and of (2) the combined group of current/former smokers.

DISCUSSION

Section:

ChooseTop of pageAbstractMETHODSRESULTSDISCUSSION <<ReferencesCITING ARTICLES

We observed higher mortality in former smokers than in current smokers, as in a previous report (10). However, severity-adjusted survival did not differ significantly between the two groups with the use of the CPI as a severity variable. This finding justified separate analyses comparing (1) former smokers with nonsmokers and (2) combined current/former smokers (viewed as a single group with respect to outcome) with nonsmokers. In both analyses, the significant survival advantages enjoyed by nonsmokers increased strikingly when functional and morphologic severity was taken into account.

Adjustment for baseline disease severity is essential in mortality studies, especially in disorders in which a healthy smoker effect is possible. Smoking is generally viewed as a risk factor for IPF, based on an association between cigarette smoking and fibrotic lung disease in familial fibrotic lung disease (20), and associations between smoking and IPF in case-control studies grouped in a meta-analysis (21). Thus, the observation of higher survival in IPF among current smokers is, at first sight, surprising. However, current smokers were characterized by strikingly better preservation of pulmonary function indices at presentation, both in the present study and in the cohort of King and coworkers (10). The healthy smoker effect, most studied in chronic obstructive pulmonary disease (11, 12), applies to comparisons between current and former smokers. From our findings, it appears possible that unadjusted survival in IPF is lower among former smokers because of a healthy smoker effect (i.e., smoking cessation may be a marker of more advanced disease). This problem needed to be addressed by adjusting for disease severity in the present study, before meaningful outcome comparisons could be made between smokers and nonsmokers.

Thus, the choice of severity variables was pivotal. In preliminary analyses, the CPI and, to a lesser degree, Dl_CO levels predicted mortality across the whole cohort better than other lung function indices, probably reflecting better correlations with CT morphologic extent (12, 22, 23). However, smoking-related lung damage and, especially, concurrent emphysema have a major confounding effect in IPF, giving rise to a relative preservation of lung volumes and a disproportionate reduction in gas transfer (12, 14, 15). Estimation of CT disease extent partially overcomes the problem, but imprecision is unavoidable in some lung regions because emphysema can be difficult to distinguish from honeycomb change in IPF when the processes are admixed (18, 24, 25). The CPI is the only current severity variable to take smoking-related damage completely into account.

In effect, the CPI quantifies functional impairment specifically due to pulmonary fibrosis, while excluding that due to emphysema (16). This rationalization of the CPI was confirmed in a study of 42 patients with IPF, including 21 with concurrent emphysema: the CPI correlated with the extent of IPF on CT more strongly than other pulmonary function variables but had no significant or marginal correlation with the extent of emphysema (25). This is likely to explain the prognostic advantages of this index in some survival analyses (16) and the fact that the use of the CPI was more influential than other severity variables, amplifying the survival advantages associated with nonsmoking status, but reducing survival differences between current and former smokers.

However, the small size of the current smoker subgroup should be stressed. Although the better survival in this group was, at least partially, ascribable to less severe disease at presentation, an additional smaller protective effect from current smoking, compared with former smoking, cannot be wholly excluded. Interestingly, in models containing the CPI, current smokers had marginally lower mortality than do former smokers, but marginally higher mortality compared with never-smokers, although neither comparison was statistically significant. This suggests that in terms of severity-adjusted survival, current smokers may lie somewhere between the other two groups, but much larger numbers of current smokers with IPF will need to be studied to resolve this question.

The divergence in our findings from those reported by King and colleagues (10) is deceptive. In both studies, current smokers were a relatively small subgroup with less functional impairment at presentation and better survival, and the different conclusion reached in the present study was purely a function of the method used to adjust for baseline severity. More importantly, longer term survival, as shown by survival curves, was somewhat better in nonsmokers than in ex-smokers in the earlier study (10), although the authors did not emphasize this point and the effect was more pronounced in our cohort, with adjustment for disease severity using the CPI.

It is important to stress the striking similarities between the observations in the two studies because inclusion criteria differed and this, in turn, led to different biases in the two cohorts. The patients with IPF studied by King and colleagues all underwent a surgical biopsy. Although a histologic diagnosis improved diagnostic confidence, an important selection bias was thereby introduced: patients undergoing biopsy are younger and have less functional impairment at presentation (26, 27). By contrast, most of our cohort consisted of consecutive patients with typical clinical and HRCT features. This allowed us to study a wider spectrum of disease severity, including patients too compromised to undergo a diagnostic surgical procedure. In theory, absence of a diagnostic surgical biopsy might have led to the inclusion of a small subgroup of patients with nonspecific interstitial pneumonia. However, an HRCT appearance typical of IPF, as in the present study, has a positive predictive value for IPF of more than 95%, even before clinical features are considered (19, 28), and the overall survival was typical of patients with biopsy-proven IPF. Our study and that of King and colleagues are complementary, with different methodologic limitations but similar subgroup trends.

There are several other issues that merit consideration. We have never, as a matter of policy, analyzed outcome in IPF according to cause of death. Opinion differs on this point but we argue that even when cardiac disease or malignancy is the obvious immediate fatal event, it is impossible to exclude a major contribution from IPF, due to hypoxia (leading to cardiac stress) or linkage between more extensive pulmonary fibrosis and neoplasia. Furthermore, it is impossible to determine the cause of death in some patients (29). For these reasons, we continue to favor all-cause mortality analyses, in the knowledge that most deaths among patients with IPF tend to be IPF related (30). A further limitation was the retrospective nature of the study, but it should be stressed that all patients were investigated according to a prospective clinical protocol, with the expectation of a full smoking history and routine baseline HRCT and pulmonary function testing. This was, therefore, a prospective data set from consecutive patients, analyzed retrospectively in response to the provocative observations of King and colleagues, and not subject to the selection biases seen in some retrospective cohorts, in which investigations are not determined according to a rigorous protocol.

The better survival among nonsmokers was the major positive finding in our study, but it is not clear whether this was linked to pathogenesis. Cigarette smoke may prime the lung toward a fibrotic response when other injuries are encountered (31, 32) and may injure lung fibroblasts, partly by means of oxidative stress, which may have a critical role in the gene expression of a variety of profibrotic factors (31, 33, 34). It is also possible that some of the excess mortality in smokers represents smoking-related comorbidity, including cardiac disease and lung cancer. However, the absence of any relationship between mortality in smokers and the pack-year smoking history provides limited indirect evidence against a smoking comorbidity hypothesis.

In conclusion, we report that life-long nonsmokers have a better outcome than former smokers, and the combined group of current and former smokers in IPF, although pathogenic differences have yet to be elucidated. Our findings provide support for the hypothesis that the better outcome previously reported for current smokers with IPF is compatible with a healthy smoker lead-time effect.