Abnormalities on Chest Radiograph Reported in Subjects in a Cancer Screening Trial^* FREE TO VIEW

Diagnostic chest radiographs (CXRs) are commonly performed in patients with a variety of symptoms and conditions. Although a CXR is not currently recommended for lung cancer screening, two randomized trials of lung cancer screening with CXR are ongoing in the United States, the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial¹ and the National Lung Screening Trial.² CXRs are also used to screen for tuberculosis and certain other diseases in specialized populations, and are used for the surveillance of workers who are at risk for occupational respiratory diseases.³⁴

Many abnormalities are potentially detectable on a CXR. The purpose of the present analysis was to describe abnormalities that were seen on CXRs in a large population of men and women who were enrolled in a cancer screening trial. The emphasis here was on abnormalities other than those that are suspicious for lung cancer, specifically, on such abnormalities as COPD/emphysema, cardiac abnormalities, scarring or pulmonary fibrosis, granulomas, pleural fluid, and pleural fibrosis. There is relatively little current literature, especially in the United States, on the prevalence of and risks associated with these abnormalities. In the current article, we describe the prevalence and persistence over time of these abnormalities, examine their relationship with age, sex, and smoking status, and analyze the risks for lung cancer incidence and overall mortality that are associated with these abnormalities.

The Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial⁵ is a multicenter, randomized, controlled trial that was designed to test the effect of screening for four types of cancer in persons 55 to 74 years of age at baseline.⁵ Randomization to a screened or control arm occurred between November 1993 and July 2001, with almost 155,000 persons randomized. Screening was performed at 10 centers located in the following cities: Birmingham, AL; St. Louis, MO; Detroit MI; Pittsburgh, PA; Washington, DC; Honolulu, HI, Marshfield, WI; Denver, CO; Salt Lake City, UT; and Minneapolis, MN. Screened arm subjects received posteroanterior view CXRs at baseline and then annually for 3 years; beginning with those subjects who were randomized in 1996, never-smokers received only two postbaseline lung screens. Subjects with a history of prostate, lung, ovarian, or colorectal cancer were ineligible for the study. Around the time of randomization, subjects filled out a self-administered demographic and medical/screening history questionnaire.

CXR findings were interpreted as follows: abnormal and suspicious for cancer (AS); abnormal and not suspicious for cancer (AN); negative; and inadequate. Each CXR was interpreted in isolation without knowledge or review of prior screening results, and without knowledge of subjects’ age, smoking status, or other characteristics. A nodule, mass, or specific other findings (eg, hilar/mediastinal lymph nodes) qualified an examination as AS. Radiologists also reported various other abnormalities, including the following seven, which were specifically listed on the screening form: granuloma; scarring/pulmonary fibrosis/honeycombing; cardiac abnormalities; COPD/emphysema; bone/soft tissue lesion; pleural fibrosis; and pleural fluid. Other noncancer abnormalities could be reported under an “other specify” category. In the absence of an AS finding, the presence of any of these noncancer abnormalities qualified an examination as AN; otherwise, the results of the examination were denoted as negative. For AN examinations, the radiologist indicated whether clinical referral was recommended. Quality control was performed by periodic checks on radiograph equipment and technicians, and through rereads of a sample (approximately 4%) of the examinations by an alternate radiologist.

Study subjects were sent annual study update forms that inquired about cancer diagnoses. In addition, positive screen results (ie, AS) were tracked for diagnostic procedures and cancer diagnoses. Reported lung cancers were confirmed through medical record review. Suspected deaths were ascertained by various means (eg, reports from next of kin, local searches, and national death index searches) and were confirmed with death certificates. All subjects with at least one adequate CXR screen that had been performed by December 31, 2002, were included in this analysis.

We performed multiple logistic regression to examine the relationship of sex, age, cigarette smoking, and specific medical history (where relevant) on the prevalence of various major abnormalities, where prevalence was defined as having at least one examination with the reported condition. Multivariate Cox proportional hazards models were used to examine the association between examination findings and subsequent lung cancer incidence and non-lung cancer mortality.⁶ Specifically, we modeled the hazards for incidence and mortality as functions of the abnormalities that were found at the latest screening examination; examination findings were thus time-varying covariates. The following baseline covariates were included in each model: age; sex; history of emphysema; smoking status (ie, current, former, or never); and (log) number of pack-years. The mortality models also included a history of heart disease, diabetes, hypertension, and stroke. To avoid possible confounding with the effect of a positive (AS) screen, subjects were (right) censored at their first AS screen for the analysis of lung cancer incidence. The models included separate indicator variables for the different major AN abnormalities, so that the hazard ratio (HR) for each abnormality was calculated by assessing the risk of cancer (or death) in those with that abnormality compared to those without it, controlling for all other abnormalities and baseline covariates. In addition, we also ran models utilizing a single indicator variable of an AN result or not. The time period of observation was from the initial screen to 3 years after the final scheduled screen or the cutoff date of December 31, 2002, whichever came first.

We analyzed the persistence of abnormalities over time by calculating intrasubject correlation coefficients for each abnormality. Correlations were calculated separately for pairs of examinations with the same radiologist and pairs with different radiologists. A correlation of 0 indicates that a past report of the condition does not predict a future report; a perfect correlation of 1 indicates that each subject has the condition reported at all visits or at none. κ coefficients of interreader agreement were calculated for each abnormality using the approximately 8,000 examinations that were reread as part of the quality control effort.

Table 1 gives the demographics of the population. A total of 236,183 adequate CXRs were performed in 70,602 subjects (average number of examinations per subject, 3.3). Altogether, 300 different radiologists read examinations; however, 99% of examinations were read by 114 radiologists, each of whom read ≥ 100 examinations. Of all the examinations, 34.5% had a result of AN (8% were AS and the remainder negative). By the study year, AN percentages were 32%, 34%, 35%, and 38% for years 0 through 3, respectively. A total of 32% of examinations in women and 37% of examinations in men had AN results.

Table 2 displays reported findings on AN examinations. The most common finding was granuloma (29%), followed by scarring/pulmonary fibrosis (22%) and bone/soft tissue lesion (15%). Over all examinations, granuloma was reported 10.7% of the time, scarring 8.2% of the time, and bone/soft tissue lesions 5.5% of the time (Table 3 ). A total of 17.5%, 16.8%, and 11.4% of subjects, respectively, had granuloma, scarring, and bone/soft tissue lesions reported on at least one examination. About 55% of subjects had at least one AN examination finding, and 18.5% of subjects had at least one AS examination finding.

The intrasubject correlation was highest for granuloma (r = 0.61 for the same radiologist; r = 0.56 for a different radiologist), followed by pleural fibrosis and cardiac abnormalities (Table 3). Nodule/mass had a relatively low correlation (0.24/0.22, respectively), as did pleural fluid (0.13/0.13, respectively). The intrasubject correlation for one condition, COPD/emphysema, varied significantly by smoking status; correlations (for the same radiologist) increased from 0.22 for never-smokers to 0.38 for former smokers to 0.51 for current smokers. Table 3 also displays κ-statistics for interreader agreement; these ranged from 0.27 to 0.66 for the specific AN abnormalities.

Recommendations for referral were relatively uncommon for examinations with AN findings. Only the relatively rare finding of pleural fluid prompted a significant rate of referral (56%) [Table 3].

Table 4 gives odds ratios (ORs) for age, sex, and smoking history for major abnormalities. All abnormalities had significantly increased prevalence in older subjects, all abnormalities except granuloma were significantly more prevalent in men than in women, and all abnormalities except for cardiac and bone/soft tissue lesions were significantly more prevalent in current or former smokers compared to never-smokers. Smoking had the greatest effect on COPD/emphysema, with ORs of 4.7 for current smokers and 1.8 for former smokers. A reported history of emphysema at baseline was also associated with findings of COPD/emphysema abnormalities on CXRs (25% of subjects) compared to findings of COPD/emphysema abnormalities (5% of subjects).

Table 5 shows HRs associated with various examination findings. Scarring (HR, 2.0) and other AN findings (HR, 1.5) were significantly associated with lung cancer incidence. COPD showed a nonsignificant HR of 1.5 (95% confidence interval [CI], 0.9 to 2.5). An overall AN finding of an examination gave a significantly increased HR of 1.8.

With respect to overall mortality (ie, non-lung cancer mortality), cardiac abnormalities (HR, 2.1), COPD (HR, 1.7), pleural fluid (HR, 2.3), scarring (HR, 1.4), and other AN finding (HR, 1.4) all gave significantly elevated HRs, as did the overall finding of an AN examination (HR, 1.6). Controlling for AN abnormalities, an AS screening result was also associated with a significantly increased mortality risk (HR, 1.6). For cardiovascular mortality, significant excess risk was found for cardiac abnormalities (HR, 2.7), scarring (HR, 1.4), pleural fluid (HR, 2.5), other AN finding (HR, 1.3), and AS finding (HR, 1.4), while for respiratory mortality significant excess risk was found for COPD (HR, 3.7), scarring (HR, 3.3), cardiac abnormality (HR, 2.1), other AN finding (HR, 2.0), and AS finding (HR, 2.3). The general finding of AN for an examination gave a very large increased risk for respiratory mortality (HR, 5.5; 95% CI, 3.4 to 8.8).

Noncancer abnormalities on CXRs are quite prevalent. The most prevalent condition reported, granuloma, did not convey excess risk of lung cancer or mortality. However, other conditions that were relatively common, such as COPD/emphysema, cardiac abnormalities, and scarring/pulmonary fibrosis, did convey an excess risk for lung cancer and/or overall mortality.

We found an HR for lung cancer associated with scarring/pulmonary fibrosis of 2.0, and an HR of 1.8 for AN examination findings in general. The HR for COPD/emphysema (1.5) was not statistically significant; however, this estimate was derived from a model that controlled for self-reported history of emphysema. When history of emphysema was not included in the model, the HR for COPD/emphysema increased to 1.8, which was statistically significant (p = 0.02). For perspective, the lung cancer HR associated with an AS screening result was approximately 14; as reported in the article describing the baseline findings in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial,¹ approximately 9% of subjects had an AS examination finding, and lung cancer was diagnosed in 2% of these subjects within a year of the screen.

Previous studies have also shown associations between pulmonary fibrosis and lung cancer. Epidemiologic and autopsy studies have reported prevalence rates of lung cancer of between 10% and 31% in patients with nonspecific pulmonary fibrosis.⁷ Hubbard et al⁸ found a relative risk for lung cancer, controlling for smoking, of 7.4 among subjects with idiopathic pulmonary fibrosis.

The locations of abnormalities were recorded as being in the upper, middle, or lower right or left hemithorax, or diffuse in the right and/or left hemithorax. For scarring, the location was marked as diffuse 10% of the time. We found a significant interaction between scarring location and the risks of lung cancer incidence and overall mortality. For lung cancer, the HR was 2.1 for diffuse scarring vs 1.4 for nondiffuse scarring, while for mortality the HR was 1.7 for diffuse scarring vs 1.15 for nondiffuse scarring. We also found a significant correlation between the scarring location and the location of lung tumors. Among subjects with scarring in a single lung who received diagnoses of lung cancer within 3 years, 72% of subjects with right lung scarring vs 42% of subjects with left lung scarring had a tumor in the right lung (p = 0.005).

In general, no specific finding was given when cardiac abnormality was reported. However, about 10% of the baseline screens utilized an early version of the screening form on which cardiomegaly could be specified. Data from these early forms suggest that overall about 90% of the cardiac abnormalities were cardiomegaly. Reported cardiomegaly here was usually based on estimates of the cardiothoracic ratio. Hemingway et al⁹ examined the association between mortality and cardiothoracic ratio in the Whitehall Study of male British civil servants. The age-adjusted rate ratio for the highest/lowest quintile of the cardiothoracic ratio was 1.38 for all-cause mortality, 2.11 for cardiovascular mortality, and 1.73 for cardiovascular mortality adjusted for various cardiovascular risk factors (eg, BP and total cholesterol). In our multivariate model, which controlled for history of heart disease, age, sex, and other factors, the HRs for overall and cardiovascular mortality associated with cardiac abnormalities were 2.1 and 2.7, respectively, with the risks being generally similar for men and women. The usefulness of CXR for cardiac abnormalities was recently investigated by Rayner et al,¹⁰ who concluded that it provided important predictive information of associated target organ damage in hypertensive subjects. Pleural fluid here was found to have a significantly elevated risk for cardiovascular mortality (HR, 2.5); this is probably due to the fact that pleural fluid is a symptom of congestive heart failure.

In addition to AN abnormalities being associated with increased mortality, we also found that, controlling for smoking and other covariates, a positive (AS) CXR finding was independently associated with overall mortality (ie, non-lung cancer mortality) [HR, 1.6], respiratory mortality (HR, 2.3), and cardiovascular mortality (HR, 1.4). Part of this excess risk for overall mortality may be due to metastases from nonlung sites that are interpreted as positive screen findings. The HR for cancer mortality (excluding lung cancer mortality) associated with an AS screen finding was 1.95 (95% CI, 1.5 to 2.5), while the HR for noncancer mortality was 1.45 (95% CI, 1.24 to 1.70). It is not clear why positive screen findings were associated with excess noncancer mortality or specifically with respiratory mortality.

The screening center in St. Louis, MO, reported granulomas at a greatly elevated rate (52% of all examinations) compared to other centers (range, 2 to 12%). Since Missouri is known for a high background rate of histoplasmosis, this could be largely a geographic effect.¹¹

The intrasubject correlations for all noncancer abnormalities except pleural fluid ranged from about 0.35 to 0.6, indicating an intermediate level of reproducibility over time. Correlations were only marginally increased when the same radiologist interpreted each examination. Pleural fluid is known to be a transient finding, so the low correlation is not surprising. Interestingly, the finding of a nodule/mass, which indicates a screen finding that is suspicious for cancer, showed a lower intrasubject correlation (approximately 0.23) than did the noncancer abnormalities. Data from the Lung Screening Study¹² over two rounds of screening showed intrasubject correlations of 0.26 for a positive CXR screen finding (essentially, a nodule/mass) and 0.41 for a positive spiral CT scan screen finding. The κ interreader agreement values reported here for the various AN abnormalities (0.27 to 0.62) are in the range reported in the literature for radiologic interpretations. κ values of 0.45 to 0.65 have been reported for radiologists interpreting screening mammograms.¹³¹⁴ A recent study¹⁵ of spiral CT scanning for lung cancer screening showed a κ-statistic of approximately 0.35.

In conclusion, several abnormalities commonly identified on CXRs convey excess risk for lung cancer incidence and for premature death from respiratory or cardiovascular diseases. Surveillance and treatment of these abnormalities may therefore be clinically important.