Lung cancer risk following previous abnormal chest radiographs: A 27-year follow-up study of a Chinese lung screening cohort

INTRODUCTION

Lung cancer is the most common cancer and the leading cause of cancer death globally, with an estimated 2.1 million new lung cancer cases and 1.8 million deaths in 2018.¹ The relative 5-year survival of this lethal disease is still less than 20% in China despite of the improvement in treatment techniques.²

The National Lung Screening Trial (NLST) demonstrated a 20% lung cancer mortality reduction in high-risk individuals for low-dose computed tomography compared with Chest radiograph (CXR) screening.³ Recently, several randomized trials conducted in Europe also displayed the mortality benefit from low-dose tomography (LDCT) lung cancer screening.^4-6 Based on the NLST results, some US clinical guidelines recommend LDCT screening in high-risk populations.^{7, 8} However, LDCT is also associated with several potential harms including high false positives, radiation and economic burden.⁹ In addition, fewer than 5% of Americans who met the U.S. Preventive Services Task Force criteria for lung cancer screening were screened in 2015, a lower prevalence compared with chest CXR.¹⁰

CXR is considered ineffective because no randomized controlled trial has shown a lung cancer mortality reduction.¹¹ However, CXR is still one of the most commonly used diagnostic tools for chest diseases in clinical practice. In China, CXR was used for diagnosis in about a third of the lung cancer patients.¹² In the United Kingdom, chest X-ray is still used for initial evaluation in all patients, apart from those aged >40 years with unexplained symptoms.¹³

Selecting the population with the highest lung cancer risk is the first step of lung cancer screening. Variation of lung cancer risk will influence the false-positive levels, efficacy, and cost-effectiveness of LDCT screening. In NLST, noncalcified nodules were associated with increased lung cancer risk up to a decade.¹⁴ In a prospective study, a 8% of lung cancer risk was observed in those who had solitary pulmonary nodules in routine CXR. In addition, a recent chest X-ray was incorporated into the lung cancer risk model.¹⁵ Accordingly, Abnormal CXR results might be useful to lung cancer risk stratification. However, little is known about the temporal pattern of lung cancer risk associated with abnormal CXR findings. The aim of this study was to investigate the long-term lung cancer risk following abnormal CXR findings based on an extended follow-up of an occupational screening cohort in Yunnan, China.

METHODS

Statistical analysis

Three sets of statistical analyses were performed according to the result of CXR screening (Figure 1). First, the analysis was based on the baseline CXR screening (T0) results, and person-years of follow-up were calculated from the date of baseline screening to the date of lung cancer confirmation, death or censoring as of December 31, 2018 (whichever came first). Lung cancer incidence and incidence rate ratios according to personal characteristics and baseline CXR screening result were calculated. Descriptive statistics were used to show the distribution of baseline CXR screening results by characteristics of participants.

Second, the analysis was restricted to participants who received the first four consecutive rounds of screening (T0–T3), and person-years of follow-up were calculated from the date of T3 to the date of lung cancer confirmation, death or censoring as of December 31, 2018. The lung cancer risks associated with abnormal CXR findings only in the first two rounds (T0 or T1), only in the last two rounds (T2 or T3), or in both T0-T1 and T2-T3 were compared to those with all negative results in T0–T4 rounds.

The association between abnormal CXR findings and subsequent lung cancer risk was analyzed with time-varying coefficient Cox regression model since the proportional hazards assumption was violated based on Schoenfeld residuals test results. In the time-varying coefficient Cox regression model, a CXR screening result*log of time, i.e., ln (t), was added. The effect of abnormal CXR screening results on lung cancer was also analyzed according to the different intervals of the follow-up period. To avoid the confounding effect from the changes in various kinds of exposure during the long-term follow-up, age, cumulative exposure of radon, arsenic and smoking (for current smokers), years since last exposure of radon, arsenic and smoking (for former smokers) were adjusted as time-varying covariate. Age at first exposure of radon and arsenic, gender, prior lung disease were also adjusted in the time-varying covariate Cox regression model.

Finally, two kinds of sensitivity analyses were conducted to assess the robustness of the effect of abnormal CXR findings on subsequent lung cancer risk. First, competing-risks regression analysis was conducted in considering of the increased risk of death from a cause other than lung cancer accompanied by aging, and the direction, the magnitude of this association was estimated with subhazard ratios (HR). Second, stratified analysis according to different levels of age, smoking, occupational radon and arsenic was performed. A two-tailed p-value of 0.05 or less was considered statistically significant.

RESULTS

Of 9274 participants who received at least one CXR screening and had satisfactory results, 1345 lung cancer cases were confirmed during the study period. However, 32 lung cancer cases lacked a definite date of diagnosis. Therefore, the current analysis is restricted to 9242 participants, in which 1313 lung cancer cases were confirmed, with a cumulative lung cancer incidence of 773.4/10⁵ person-years.

The characteristics of 9242 participants and characteristic-specific lung cancer incidence are presented in Table 1. Two thirds (66.3%) of participants were 40–59 years old at the time of enrollment. Nearly one quarter of participants (23.7%) never attended school. Most participants were males. A large majority had a smoking history and occupational arsenic exposure. Nearly 80% of participants had occupational radon exposure. Significantly increased lung cancer incidence rate ratios and more abnormal baseline CXR findings were observed in males, smokers, participants with low education level, high radon or arsenic exposure and prior lung disease including asthma, chronic bronchitis and silicosis.

The associations between abnormal CXR findings with long-term risk of lung cancer are presented in Table 2. For baseline screening, abnormal results significantly increased lung cancer risk with an adjusted hazard ratio (HR) of 26.74 (95% CI: 18.33–39.01), and this relative hazard significantly decreased with time. A total of 4556 participants received the first four consecutive rounds of CXR screening (T0–T3). Compared with solely negative screening results, the adjusted HRs for at least one abnormal result, abnormal results only in T2–T3 and in both T0–T1 and T2–T3 were 13.60 (95% CI:8.69–21.28), 11.49 (95% CI: 7.10–18.58) and 29.50 (95% CI:13.38–56.56), and both demonstrated a significant decreasing trend with time. Of 1313 lung cancer cases, 747 (56.9%) had definite pathology results. The two most common lung cancer cell types were squamous carcinoma (51.3%) and adenocarcinoma(19.7%). The relationship between abnormal CXR findings and lung cancer risk by histology is also presented in Table 2. For both squamous and adenocarcinoma, their risks were significantly increased following abnormal results from baseline and the first four consecutive screening round. In addition, their risks were also decreased over time.

Figure 2 intuitively shows the time-varying lung cancer risk associated with abnormal CXR results. Significantly increased lung cancer risks for at least one abnormal CXR result of baseline screening and the first four screening rounds were observed in the first 10 years since the follow-up, with adjusted HRs in 5–10 years after the beginning of follow-up of 4.13 (95% CI: 1.53–11.14) and 4.06 (2.12–7.77).

Finally, sensitivity analyses were conducted to assess the robustness of above results (Tables 3 and 4). Both competing-risk regression analysis and stratified analysis suggest that the association between long-term lung cancer risk and abnormal CXR screening results was independent of other aging-related competing risks and confounding effect from other exposures including age, smoking, occupational radon and arsenic.

DISCUSSION

In this prospective study, up to 10 years increase of lung cancer risk associated with previous abnormal CXR findings was observed, although there was a decreasing trend. This association was stronger for persistent abnormal findings, abnormal findings identified in the most recent screening rounds and in terms of histology, for both squamous carcinoma and adenocarcinoma. These results suggest that abnormal radiographic findings might be a potential risk stratification tool for lung cancer screening.

Selecting the most appropriate target candidates for lung cancer screening is critical to maximize benefits and minimize adverse effects. LDCT lung cancer screening is recommended for heavy smokers (≥30 pack-years and ≤15 years since quitting) older than 50 or 55 in some countries in North America and Asia. However, from 2010 to 2015, the percentage of eligible smokers who reported LCDT screening in the past 12 months was lower than 5% in the USA, which is still lower than the percentage undergoing chest X-ray.^{10, 18} A study in the USA found that the proportion of lung cancer patients meeting the USPSTF screening criteria decreased significantly between 1984 and 2011, suggesting that more sensitive criteria may be needed to identify the most suitable candidates for LDCT screening. Risk prediction models have been reported to more accurately predict lung cancer risk.^{19, 20} In these models, risk factors other than age and smoking were incorporated, such as prior lung disease and family history of cancer.

Lung cancer risk was different according to prior imaging findings. In the NELSON trial, the risk for detecting lung cancer in the fourth round was 3.7% for those with indeterminate third round results compared with 0.6% of those with negative results.²¹ In the NLST, significantly increased lung cancer risk associated with abnormal noncalcified nodules following screening was also observed.^{14, 22} Similarly, this study also found a higher lung cancer risk following abnormal CXR findings. Due to the radiographic nature, the increased lung cancer risk associated abnormal CXR screen might merely be a reflection of long-term exposure to other risk factors. First, in this study, most participants were smokers, and had occupational radon or arsenic exposure, and the long-term effect of field cancerization from these carcinogens might lead to an abnormal imaging result.^{14, 23} Second, prior lung disease was reported previously as an independent risk factor for lung cancer.²⁴ Accordingly, an abnormal CXR screen might also be a marker of long-term inflammation in the lung. Third, the highest risk of lung cancer in the first 5 years following abnormal CXR screen might occur due to the existing lung cancers. However, both multivariate analysis and stratified analysis demonstrated a significant relationship between abnormal CXR findings and increased lung cancer risk, and suggested that at least some abnormal screens were precursors of lung cancer, similar to the results of a study based on NLST.¹⁴

Few studies have reported the temporal trend of lung cancer risk following abnormal radiographic findings. In the NLST, lung cancer risk associated with noncalcified nodules in the LDCT arm was 5.6 for period between 0–4 years following the baseline screen, and was decreased to 1.5 for 8–12 years.¹⁴ This study observed similar results. However, compared with the results of LDCT, the magnitude of the lung cancer risk in the first 5 years following abnormal CXR findings was much larger, and correspondingly, the decreasing trend was sharper during the follow-up period. The main reason for these differences might be the differences in screening modality and the definition of a abnormal screen between the NLST and YTC study. In NLST, abnormal screen was any noncalcified nodule measuring at least 4 mm in any diameter, which resulted in high positive rate. In contrast, with the abnormal screen definition, the baseline positive rate was less than 1% in this study.

To the best of our knowledge, this study is the longest long-term evaluation of lung cancer risk concerning previous abnormal CXR findings. In addition to the increased lung cancer risk in the first decades of follow-up, an insignificant decreased risk 10 years after abnormal CXR findings was also observed. The dynamic changes of lung cancer risk following abnormal CXR findings might help identify high risk individuals for LDCT lung cancer screening. Currently, most lung cancer screening guidelines focus on heavy smokers,²⁵ which would exclude a large number of non or light smokers from LDCT screening. For example, it has been previously reported that nearly half of lung cancer cases in China are nonsmokers.¹² Should non- or light smokers receive screening or intensive surveillance if they have abnormal CXR within 5 or 10 years in the real world? Alternatively, should persons with an abnormal CXR result 10 or more years before but do not develop lung cancer receive annual LDCT screening, even among smokers? If they had received LDCT screening, can we lengthen their screening interval?

In the YTC cohort, sputum cytological screening was also conducted annually. Our previous study found that sputum atypia significantly increased the risk of squamous cell carcinoma and small cell lung cancer, but it was not related to the risk of adenocarcinoma.²⁶ In this study, increased risks of both squamous carcinoma and adenocarcinoma were observed following abnormal CXR findings. The reason for these differences might be that CXR is more sensitive than sputum cytological screening, especially for adenocarcinoma. Accordingly, the atypical adenomatous hyperplasia, the precancerous lesion of adenocarcinoma is also likely to be found by CXR. The decreasing trend of the squamous cell carcinoma risk in relative to prior abnormal CXR results during the long-term follow-up might be that squamous cell carcinoma is developed in a stepwise pattern where the epithelium changes from normal to hyperplasia, metaplasia, mild, moderate, and severe dysplasia and then carcinoma in situ. High-grade lesions are more likely to progress to invasive cancer than low-grade lesions.^{27, 28} Similarly, the large majority of adenocarcinoma precursor lesions regress spontaneously.²⁹However, The regression rate of adenocarcinoma precursor was hard to determine through specific radiographic features. Characterizing the molecular alterations that are associated with progression of premalignant lesions to invasive squamous carcinoma or adenocarcinom will reveal molecular mechanisms of progression and advance the field of precision chemoprevention and lung cancer risk stratification.^{30, 31}

This study has several strengths. It was a large, prospective study that prospectively collected detailed covariates at baseline. The extended follow-up and a large number of lung cancer cases which increased the statistical power provided an opportunity to explore the temporal trend of lung cancer risk related to prior screening results of CXR. Also, sensitivity analysis confirmed the stability and robustness of the current findings. However, limitations can still be found in this study. First, occupational radon/arsenic exposure and smoking were the main cause of extreme high lung cancer incidence in the YTC cohort,¹⁶ thus their confounding effects might still exist. Second, when the relationship between prior abnormal CXR findings and lung cancer risk was analyzed according to lung cancer histological type, instable results were concerns since nearly half of lung cancer cases lacked histological information.

In conclusion, new techniques, including computer-aided detection and deep learning, were found to be helpful for the identification of high-risk smokers for LDCT lung cancer screening.^{32, 33} In consideration of the wide utility of CXR in clinical practice, the results of this study imply that the magnitude and the temporal pattern of lung cancer associated with abnormal CXR screens might be helpful for quantifying lung cancer risk in the real world.

CONFLICT OF INTEREST

The authors confirm that there are no conflicts of interest.