Respiratory Function as a Prognostic Factor for Lung Cancer in Screening and General Populations

Abstract

Rationale

Despite advancements in screening, lung cancer remains the leading cause of cancer-related mortality globally.

Objectives

To investigate respiratory function as a prognostic factor for survival in the UK Biobank, a population-based cohort of more than 500,000 participants, and the NLST (National Lung Screening Trial), a high-risk screening population of more than 50,000 screenees.

Methods

Participants with an incident lung cancer diagnosis and spirometry-assessed lung function were included. Lung cancer was measured as the ratio of forced expiratory volume in 1 second (FEV₁) and forced vital capacity and percentage of predicted FEV₁. Multivariable Cox proportional hazards models were fitted to estimate the impact of lung function on 5-year overall survival in populations with different baseline lung cancer risks.

Results

A total of 2,690 and 609 patients were included in the analysis from the UK Biobank and the NLST, respectively. In the UK Biobank, a higher percentage of predicted FEV₁ and ratio were associated with better survival after lung cancer diagnosis, with hazard ratios of 0.97 (95% confidence interval [CI], 0.95–1.00 per 10% increase) and 0.95 (95% CI, 0.90–1.00 per 10% increase), respectively. No statistically significant results were found when assessing the data from the NLST study.

Conclusions

Impaired lung function was associated with poorer survival for patients with lung cancer in the general population, although this was less clear in a high-risk, screening-eligible population. This highlights the potential clinical importance of respiratory function as a prognostic factor in lung cancer in the general population and presents a possibility for personalized cancer management.

Lung cancer is the leading cause of cancer-related death globally, with more than 2 million deaths estimated for 2025 (1). Screening with low-dose computed tomography was shown to reduce lung cancer mortality in high-risk individuals because of shifts to earlier stage at diagnosis, when treatment outcomes are favorable (2–5). However, the prognosis for advanced-stage disease remains poor, with a 5-year survival rate of 50–55% for localized and 4–5% for metastatic lung cancers, respectively (6). Age at diagnosis, tumor stage, histology, somatic mutations (e.g., EGFR, KRAS), and several comorbidities are known prognostic factors for overall survival in patients with lung cancer, but they do not fully account for the wide range of observed survival rates (6, 7).

The identification of prognostic factors that can be easily measured in a clinical setting is a critical aspect of optimal patient management at the time of diagnosis. Spirometry can rapidly assess respiratory function, with standardized and accurate quantitative measures not susceptible to self-report biases. The forced expiratory volume in 1 second (FEV₁) and forced vital capacity (FVC) are common absolute measures of lung function, whereas the FEV₁/FVC ratio and percentage of predicted FEV₁ (observed FEV₁ relative to predicted FEV₁) are important relative performance measures (8–10). Furthermore, it has been identified that consideration of spirometry-measured chronic obstructive pulmonary disease (COPD) may be important in identifying individuals who are at a high risk of lung cancer and who may otherwise not be identified by current eligibility guidelines (11). Despite the clear relevance, the role of respiratory function in lung cancer prognosis has yet to be well elucidated (12–14).

Few studies with selected populations, specifically screening-eligible or under specific treatment protocols, have examined lung function and overall survival (15–18). No studies have thoroughly investigated lung function as a quantitative measurement and its potential dose–response relationship with overall survival. With the goal to identify prognostic factors that can be easily measured in clinics, we investigated the association between lung function and 5-year overall survival using two large-scale populations: a general population, represented by the UK Biobank (UKB), and a low-dose computed tomography screening–eligible population, represented by the NLST (National Lung Screening Trial).

Methods

Study Populations

This study used data from the UKB and the NLST to investigate the associations between lung function and overall survival in populations with different baseline risks (19, 20). Both study populations have been previously described. In brief, the UKB is a population-based cohort of more than 500,000 participants aged 40–69 years (19). The NLST is a randomized screening trial that included 53,454 individuals aged 55–74 years with a heavy smoking history (2). The Mount Sinai Hospital Research Ethics Board approved conduct of analyses on deidentified data for the UK Biobank (REB: 17-0020-E) and the NLST (REB: 17-0119-E).

Outcome Ascertainment

In the UKB, incident lung cancer cases were identified using International Classification of Diseases, 10th Revision codes and were limited to those diagnosed at least 2 years after enrollment to avoid potential reverse-causation bias from subclinical cancers present at baseline. Patients with lung cancer were followed until September or October 2021 (region dependent) or the date of death, whichever came first. Deaths were identified using linkage to national mortality registers. In the NLST, lung cancers diagnosed during the study follow-up period (i.e., within 6 yr of enrollment) were included. Deaths were identified by ascertainment of death certificates. Survival time was measured from the date of lung cancer diagnosis to the date of death or date of last follow-up.

Respiratory Function

Spirometry measurements were recorded at cohort enrollment for UKB participants and within 1 year of enrollment and preceding lung cancer diagnosis for NLST participants. Participants were excluded if either FEV₁ or FVC were missing. Cohort flow charts are shown in Figure E1 in the data supplement. The primary lung function values of interest were percentage of predicted FEV₁ and the FEV₁/FVC ratio. FEV₁/FVC ratio was derived using maximum FEV₁ and FVC values. We used the predicted FEV₁ (i.e., the expected FEV₁ based on a participant’s demographics) provided from the UKB and NLST, and when the reference was missing (21% of UKB cases and 6% of NLST cases), it was estimated using reference equations from the Global Lung Function Initiative (9). The Global Initiative for Chronic Obstructive Lung Disease guidelines were applied to derive lung impairment variables comparable to a clinical diagnosis of COPD (21). COPD was defined as FEV₁/FVC < 70%, with severity categorized by percentage of predicted FEV₁: mild (FEV₁ > 80% predicted), moderate (50%< FEV₁ < 80% predicted), severe (30%< FEV₁ < 50% predicted), and very severe (FEV₁ < 30% predicted). To enhance interpretability and present clinically meaningful results, the FEV₁/FVC ratio and the percentage of predicted FEV₁, when included in models, were scaled to reflect a 10% change in respiratory function.

Statistical Analysis

Cox proportional hazards models were applied to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) for the effect of lung function on 5-year overall survival. The dose–response relationship with survival was assessed based on FEV₁/FVC and the percentage of predicted FEV₁, as well as COPD severity. Penalized smoothing splines were applied to assess the presence of nonlinearity in the relationship between lung function and overall survival. Covariates assessed in models were chosen a priori. For the UKB, models were adjusted for age at diagnosis, sex, and smoking status, but not tumor stage as it was unavailable. For the NLST, models were adjusted for age at diagnosis, sex, and tumor stage, but not smoking status as all participants were heavy smokers, per trial inclusion criteria. Contour plots were generated to present the differences in probability of overall survival over time by variations in lung function.

A series of sensitivity analyses were performed to assess the robustness of the study findings. First, we estimated 5-year lung cancer–specific mortality wherein non–lung cancer deaths were censored at the time of death. Second, to generate a second high-risk population, a subset analysis was performed including only UKB participants meeting the NLST screening criteria, with the same statistical approaches described above. Third, as the UKB did not report lung cancer stage at diagnosis or treatment, the analysis was repeated in the NLST population without adjustment for stage to assess stage as a potential confounder. Similarly, to assess the possible impact of treatment on the association between lung function and prognosis, treatment was added to the model in the NLST cohort, where data were available. Finally, to assess any differences in the association between lung function and survival based on lead time from spirometry measurement to lung cancer diagnosis, a sensitivity analysis was performed in the UKB cohort, stratified by lead time as less than versus more than 5 years after enrollment. These results were compared with the estimates from the main analysis. All statistical analyses were performed using R (v4.1.1).

Results

In total, 3,299 patients with lung cancer, including 2,690 from the UKB and 609 from the NLST, were included in this analysis. Patient characteristics and follow-up time by cohort are described in Table 1. In both populations, more deaths occurred among males and current smokers. Late-stage diagnosis resulted in more deaths in the NLST. Figure E2 displays the increased mean respiratory function in the general population (UKB) compared with the screening-eligible population (NLST). The measures of central tendency for each lung function measurement are summarized by cohort (Table E1), showing that lung function is generally better in the UKB cohort, but the two cohorts display similar variability in their distributions.

Table 1.

Demographics and patient characteristics by vital status as of date of last follow up and study population (N = 3,299)

	UKB		NLST
	Dead (n = 1,876)	Alive (n = 814)	Dead (n = 278)	Alive (n = 331)
Follow up time, yr	1.86 (1.72)		2.42 (1.87)
Age, yr	69.45 (6.13)	69.56 (6.40)	63.46 (5.17)	63.61 (5.25)
Sex
Male	1,010 (53.84)	335 (41.15)	166 (59.71)	171 (51.66)
Female	866 (46.16)	479 (58.85)	112 (40.29)	160 (48.34)
Smoking status*
Never	208 (11.09)	142 (17.44)	—	—
Former	863 (46.00)	419 (51.47)	81 (29.14)	120 (36.25)
Current	805 (42.91)	253 (31.08)	197 (70.86)	211 (63.75)
Stage†
I	—	—	43 (15.47)	227 (68.58)
II	—	—	18 (6.47)	21 (6.34)
III	—	—	80 (28.78)	51 (15.41)
IV	—	—	131 (47.12)	25 (7.55)
Unknown	—	—	6 (2.16)	7 (2.11)
Randomization group‡
CT scan	—	—	161 (57.91)	236 (71.30)
X-ray	—	—	117 (42.09)	95 (28.70)

Definition of abbreviations: CT = computed tomography; NLST = National Lung Screening Trial; UKB = UK Biobank.

Means and standard deviations are reported for numeric variables and frequencies and percentages are reported for categorical variables.

^*Only heavy current and former smokers were eligible for enrollment into the NLST.

^† Tumor staging was not reported in the UKB and therefore could not be assessed in this study.

^‡ Randomization group for screening was reported in the NLST but was not applicable in the population-based UKB cohort study.

On average, those in the UKB cohort with mild to moderate COPD (HR, 1.15; 95% CI, 1.05–1.27) and severe to very severe COPD (HR, 1.46; 95% CI, 1.20–1.78) had significantly worse unadjusted overall survival than those with no COPD (Figure 1). In the UKB cohort, we observed better overall survival for patients with lung cancer with increased percentage of predicted FEV₁ (HR, 0.97; 95% CI, 0.95–1.00 per 10% increase), as well as those with increased FEV₁/FVC ratio (HR, 0.95; 95% CI, 0.90–1.00 per 10% increase) after adjusting for age, sex, and smoking status (Table 2). Severe or very severe COPD was associated with worse survival compared with those without COPD (HR, 1.31; 95% CI, 1.07–1.60). Figure 2 shows that there is a nonlinear association between the percentage of predicted FEV₁ and the overall survival time; however, this was not observed for the FEV₁/FVC ratio. To provide information regarding the variability of overall survival dependent on different combinations of key patient characteristics, Figure 3 illustrates the probability of overall survival by respiratory function measurement (FEV₁/FVC ratio and percentage of predicted FEV₁) and time since diagnosis separated by sex. For example, the probability of survival at 5 years for a 65-year-old patient with COPD who is currently smoking is 20–30% for females, with an FEV₁/FVC ratio of 0.7; however, it decreases to 10–20% if their FEV₁/FVC ratio is 0.4.

Figure 1.

Kaplan-Meier curves of 5-year overall survival by study cohort: (A) UK Biobank and (B) National Lung Screening Trial. CI = confidence interval; COPD = chronic obstructive pulmonary disease; HR = hazard ratio.

Table 2.

HRs and 95% CIs for 5-year overall survival in the UKB and NLST cohorts

	UKB					NLST
	n	Deaths	HR*	95% CI	P Value	n	Deaths	HR†	95% CI	P Value
FEV1% predicted (per 10%)	2,690	1,876	0.97	0.95–1.00	0.033	609	278	1.00	0.99–1.00	0.535
FEV1/FVC ratio (per 10%)	2,690	1,876	0.95	0.90–1.00	0.043	609	278	1.04	0.38–2.90	0.937
COPD severity‡
None	1,466	979	1.00	Ref	—	300	131	1.00	Ref	—
Mild to moderate	1,092	788	1.04	0.94–1.14	0.482	221	103	1.07	0.82–1.41	0.604
Mild	439	320	1.07	0.94–1.21	0.337	63	36	1.51	1.02–2.23	0.037
Moderate	653	468	1.02	0.91–1.14	0.769	158	67	0.93	0.69–1.27	0.659
Severe to very severe	132	109	1.31	1.07–1.60	0.008	88	44	1.31	0.92–1.86	0.134
Severe	117	94	1.26	1.02–1.56	0.033	72	34	1.24	0.84–1.83	0.280
Very severe	15	15	1.76	1.06–2.95	0.030	16	10	1.68	0.88–3.22	0.117

Definition of abbreviations: CI = confidence interval; COPD = chronic obstructive pulmonary disease; FEV₁ = forced expiratory volume in 1 second; FVC = forced vital capacity; HR = hazard ratio; NLST = National Lung Screening Trial; UKB = UK Biobank.

HRs and CIs are estimated using Cox proportional hazards models.

^*Models adjusted for age, sex, and smoking status.

^† Models adjusted for age, sex, stage, and NLST randomization group.

^‡ Separate models were fit comparing COPD severity (mild vs. none, moderate vs. none, severe vs. none, very severe vs. none) and grouped COPD severity (mild to moderate vs. none, severe to very severe vs. none).

Figure 2.

Examination of the presence of nonlinearity in hazard ratio for 5-year all-cause mortality across a range of values for percentage of predicted forced expiratory volume in 1-second (FEV₁) in the UK Biobank cohort (i.e., FEV₁ relative to a predicted FEV₁ based on age, sex, height, and ethnicity). Generated using a penalized smoothing spline. FVC = forced vital capacity.

Figure 3.

Contour plots presenting the estimated probability of 5-year overall survival for variations in (A) the FEV₁/FVC ratio and (B) the percentage of predicted FEV₁ over time in the UK Biobank cohort. Plots are presented by sex based on a 65-year-old current smoker with a history of chronic obstructive pulmonary disease. FEV₁ = forced expiratory volume in 1 second; FVC = forced vital capacity.

In contrast, we did not observe a significant relationship between COPD status and overall survival in the NLST cohort, comprising high-risk individuals (Figure 1). We also did not observe a clear association between lung function and overall survival after adjustment for covariates, although there was a tendency of positive HRs for those with impaired lung function (Table 2).

The sensitivity analysis based on the 5-year lung cancer–specific mortality showed results in a similar direction, albeit with diminished statistical significance (Table E2). Based on the sensitivity analysis performed in a subset of the UKB cohort that was high risk (meeting NLST inclusion criteria), results showed a tendency toward higher hazards of all-cause mortality for those with suboptimal lung function; however, results lacked statistical significance (Table E3). This concurred with the results we observed from the NLST population. Based on the sensitivity analyses performed in the NLST population with and without adjusting for stage or treatment as a confounder, the results remained largely the same (Tables E4 and E5). This suggested that stage and treatment were not major confounders between lung function and lung cancer prognosis. In the sensitivity analysis stratifying by lead time between spirometry measurement to lung cancer diagnosis in the UKB cohort (<5 yr vs. ≥5 yr), the two sets of stratified results were generally in agreement, indicating that the lead time to diagnosis did not play a major role in our results (Table E6). Furthermore, the key results of both stratified analyses remained in the same direction as the estimates of our primary analysis.

Discussion

To our knowledge, this is the first study to examine the dose–response relationship between respiratory function and survival after lung cancer diagnosis. Based on two large populations, we found that percentage of predicted FEV₁ and FEV₁/FVC ratio are associated with 5-year overall survival of patients with lung cancer in the general population but not in those high-risk populations eligible for lung cancer screening. Our findings indicate that lung function can be a useful predictor of overall survival among patients with lung cancer in general populations.

Previous studies have demonstrated that respiratory comorbidities are linked to poor outcomes in lung cancer (7, 11–13). Furthermore, researchers identified that when respiratory function is defined as below normal, this results in decreased survival time and worse survival prognosis, compared with normal respiratory function (22–24). In addition, increased lung cancer mortality is seen in nonsmokers with emphysema and chronic bronchitis (25). This is compatible with our findings from the UKB sample, representing unselected patients with lung cancer in the general population. The possible reasons for this observed association could be suboptimal physiological state of the lung with increased inflammation or limited treatment options for those with poor lung function.

In contrast, the lack of associations in the NLST study may be due to the highly selected nature of the screening population, with heavy smoking history and lower lung function measures. Our results indicate there is similar variability in lung function measurements within both cohorts. Therefore, there are two possible explanations for the lack of association seen in the NLST: 1) although the NLST cohort does have worse lung function, there may be a threshold effect occurring where lung function below a certain point does not continue to confer worse survival; or 2) the lower sample size in the NLST cohort, compared with the UKB cohort, may be less powered to detect modest effect sizes. In this case, given the tendency of positive HR estimates observed, the possibility of the adverse impact of poor lung function on survival cannot be entirely precluded in the high-risk population.

In this paper, the primary endpoint was 5-year overall survival. When considering lung cancer–specific mortality, the results were similar to the primary findings, albeit with diminished statistical significance, mainly because of a lower number of outcome events. A majority of the deaths in both the NLST (91%) and the UKB (85%) were reported to be due to lung cancer, potentially explaining the lack of major differences seen when we examine all-cause mortality and lung cancer–specific mortality.

Strengths and Limitations

One limitation is that treatment and stage information were not available for the UKB cohort. Our sensitivity analyses showed that these two variables did not have a major role in the association between lung function and prognosis in the NLST; however, it does not preclude the possibility of acting as confounder in the UKB. It is worthwhile to point out that the lung function measurement in UKB was obtained at enrollment (2–13 yr before cancer diagnosis) and, thus, not expected to be highly correlated with stage of diagnosis. Finally, it is important to acknowledge the possibility for healthy volunteer bias, whereby individuals who chose to participate in the UK Biobank cohort may be more health conscious than those who do not participate, introducing differences between the study sample and the target population. This has been documented previously and may affect the generalizability of our findings (26). However, by nature of our study design and through the inclusion of two different types of populations, we aim to mitigate this limitation.

This is the largest study to examine the association between respiratory function and survival, based on two distinct populations. This allowed for analysis and interpretation in both general and heavy-smoking populations, increasing the generalizability of our findings. Specifically, we were able to assess detailed dose–response relationships between lung function and overall survival. Studies have shown that treating COPD in patients with lung cancer can improve survival (27), and our research may contribute to increasingly personalized and informed approaches toward lung cancer management.

Conclusions

We observed worse overall survival in those with poorer lung function within a set of patients with lung cancer from a general population cohort. These findings did not extend to patients from a high-risk screening-eligible population. Future studies with larger sample sizes and tumor stage information are warranted to corroborate these findings.