|Home | About | Journals | Submit | Contact Us | Français|
Two randomized controlled trials of lung cancer screening initiated in the 1970's, the Johns Hopkins Lung Project and the Memorial Sloan-Kettering Lung Study, compared one arm which received annual chest x-ray and four-monthly sputum cytology (dual-screen) to a second arm which received annual chest x-ray only. Previous publications from these trials reported similar lung cancer mortality between the two groups. However, these findings were based on incomplete follow-up, and each trial on its own was underpowered to detect a modest mortality benefit.
We estimated the efficacy of lung cancer screening with sputum cytology in an intention-to-screen analysis of lung cancer mortality, using combined data from these trials (n=20,426).
Over one-half of squamous cell lung cancers diagnosed in the dual-screen group were identified by cytology; these cancers tended to be more localized than squamous cancers diagnosed in the x-ray only arm. After nine years of follow-up, lung cancer mortality was slightly lower in the dual-screen than in the x-ray only arm (rate ratio (RR) 0.88, 95% confidence interval (CI) 0.74-1.05). Reductions were seen for squamous cell cancer deaths (RR 0.79, 95% CI 0.54-1.14) and in the heaviest smokers (RR 0.81, 95% CI 0.67-1.00). There were also fewer deaths from large cell carcinoma in the dual-screen group, though the reason for this is unclear.
These data are suggestive of a modest benefit of sputum cytology screening, though we cannot rule out chance as an explanation for these findings.
Exfoliative cytology may be used to identify early-stage, curable cancers and thus prevent cancer mortality. The success of cytology is exemplified by Papanicolaou screening for cervical cancer, which is recommended based on dramatic (60-90 percent) decreases in cervical cancer mortality that were seen following the rapid introduction of screening into a number of populations in the 1960s-1970s1. But while the Pap technique has been successfully adapted by Saccomanno2 for the microscopic examination of sputum in order to identify those at higher risk of developing lung cancer3, 4, the utility of cytology as a lung cancer screening tool is far less clear-cut.
Two randomized trials of lung cancer screening initiated in the 1970s, the Johns Hopkins Lung Project5 and the Memorial Sloan-Kettering Lung Study6, compared two screening arms: one offered both annual chest x-ray and four-monthly sputum cytology examination (dual-screen), and the other offered annual chest x-ray alone. Thus, they evaluated the incremental impact of adding sputum cytology (specifically cytomorphology, the examination of the appearance of exfoliated, stained cells by light microscopy) to a chest x-ray screening regimen. Prior publications from these trials5-9 were either based on incomplete follow-up information5-7 and/or excluded any lung cancer deaths for cancers prevalent at baseline screening7, 8. Reported lung cancer mortality rates were similar between arms in both trials (3.4 and 3.8 per 1000 man-years in the dual-screen and x-ray only arms, respectively, at Hopkins8, and 2.7 per 1000 man-years in both arms at Sloan-Kettering9). However, because the full complement of lung cancer deaths was not considered in existing reports, because the original trials were powered to detect a large (50 percent) decrease in cancer mortality10, 11, and because lung cancer mortality rates in the dual-screen arm trended lower than those in the x-ray only arm6, 8, it remains possible that a true, modest benefit of cytology screening wasn't detected due to inadequate statistical power.
We reanalyzed the Johns Hopkins and Sloan-Kettering data using additional follow-up information from both trials. The very similar designs of the trials and similar data collection procedures coordinated by the NCI Cooperative Early Lung Cancer Group12-14 allowed us to combine these data for the purpose of calculating one joint estimate of screening efficacy. Further, since sputum cytology is better suited to detecting more central (usually squamous cell) cancers15, which may be less detectable by newer technologies such as spiral CT16, we also estimated efficacy according to specific histologic subtypes of lung cancer.
The designs of the Johns Hopkins Lung Project and the Memorial Sloan-Kettering Lung Study have previously been reported13. Briefly, both trials enrolled males aged 45 years and older who smoked at least one pack of cigarettes per day (or who had smoked this much within one year of enrollment) and who had no prior history of respiratory tract cancer. At Hopkins, volunteers were recruited in the Baltimore metropolitan area between 1973 and 1978; the Sloan-Kettering trial recruited participants in the New York City area between 1974 and 1978. Written informed consent was obtained for all subjects.
All eligible participants were randomized by computer to either a dual-screen or x-ray only group, and were invited to attend annual exams during which posterior-anterior and lateral chest x-rays were obtained. In the dual-screen group, sputum was collected at the annual exam after saline aerosol induction. Immediately following each annual exam and at four and again eight months later, dual-screen participants were instructed to collect at home their first sputum upon rising for three consecutive mornings, and return the pooled specimen in a postage-paid carton. Radiographs were read independently by at least two examiners, and from each sputum specimen preserved in Saccomanno's solution, four Papanicolaou-stained slides were read by qualified cytopathologists. Screening continued for 5-7 years at Hopkins and 5-8 years at Sloan-Kettering.
Chest radiographs were classified as unsatisfactory, negative, abnormal requiring further x-ray, and suspicious for cancer. Follow-up of positive x-rays (with or without abnormal sputum cytology) included physical examination, additional radiological investigation when necessary, review of previous available x-rays, and possibly needle aspiration cytology, bronchoscopy, mediastinoscopy/mediastinotomy, and/or thoracotomy. Sputum cytology exams were classified as unsatisfactory, normal, slight, moderate, or marked atypical metaplasia, or cancer. Slight atypia was considered to be a negative exam. A categorization of moderate or marked atypia led to immediate repeat cytology (with induction of sputum if possible). After two consecutive specimens demonstrated no or slight atypia, the subject was returned to the regular (4-monthly) cytology schedule. Two successive specimens with marked atypia or the discovery of any cancerous cells resulted in a recommendation of further procedures to localize the source of the abnormal cells; if two successive specimens had moderate atypia, the decision to recommend further diagnostic testing was based on clinical judgment. Localization procedures included radiography and possibly computed tomography, examination of the oral cavity, pharynx, and larynx by a head and neck surgeon, and fiberoptic bronchoscopy under general anesthesia, with biopsies and bronchial brushings/washings obtained from each segment if no endobronchial lesion was seen. All lesions were biopsied and individuals with localized cancers were advised to undergo thoracotomy. Any screening examinations coded as unsatisfactory were repeated whenever possible.
At study entry, all subjects completed a questionnaire that elicited information regarding demographics, smoking history, and exposure to other lung cancer risk factors. Response to follow-up questionnaires was sought annually thereafter. Lung cancer diagnoses and deaths were determined based on responses to the annual questionnaire (or lack of response with further investigation to determine whether death had occurred). Additional follow-up information was obtained in person, by telephone, from outside medical facilities, and from death certificates as required. The duration of active follow-up after cessation of screening was variable between subjects. Further, all living lung cancer cases at Sloan-Kettering were followed for four to eight months after all contact with non-cases had ceased. This additional follow-up of Sloan-Kettering cancer cases would lead to an artificial elevation of mortality rates if all available follow-up time were included. Therefore, for both trials we have chosen to consider only the subjects' first nine years after entry in these analyses; this time period was determined empirically based on a preliminary examination of cumulative hazard rates for lung cancer incidence and mortality. However, screening efficacy estimates were essentially unchanged if all follow-up time was considered. For the purposes of this manuscript, loss to follow-up was defined as occurring when a subject who was alive at last contact was not followed through the time of their expected fifth-year screening visit.
For all deaths identified during follow-up, death certificates were requested, as were relevant clinical records pertaining to cause of death. These records were reviewed by a mortality review committee composed of statisticians, clinicians, and pathologists who were blinded to screening group assignment. Initial review was conducted by two members of the committee, with final determination of cause of death made by the entire committee if the primary reviewers had discrepant opinions regarding the cause of death (as due to lung cancer or other causes). Deaths due to “lung cancer” included those occurring as a result of study-initiated procedures, regardless of whether the subject was ultimately found to have lung cancer; in this way, harms of screening were appropriately considered in measuring screening efficacy. The same mortality review committee evaluated deaths for both trials.
For all cases of lung cancer, tumor information including stage and histology was obtained from medical records. In the current manuscript, tumor stage (0 to 4) was defined using tumor, node, and metastasis (TNM) data, and was coded according to the current (sixth edition) AJCC staging manual17. “T” categories changed between the first (used during the trials)18 and sixth editions of the AJCC manual: tumors classified as T3 by the older version represented either T3 or T4 cancers according to the current version. This led to some uncertainty in defining stage. For these purposes, all T3N0M0 cancers (n=28 (Hopkins); n=18 (Sloan-Kettering)) were assigned a missing summary stage value. Additionally, at Hopkins, there were 16 presumptive lung cancers for which no primary was ever localized. These were also considered to be missing summary stage information, with the following exceptions: (1) three cytology-detected occult (T0N0M0) cancers were coded as stage 0, (2) one T0N1M0 tumor was coded as stage 2, and (3) six T0N2M0 tumors were coded as stage 3. In the last two categories, we believe that patterns of lymph node metastasis and absence of a distant tumor make assignment of stage appropriate.
Compliance was estimated as the proportion of living, lung cancer-free subjects who received screening during a particular time period. Incidence and mortality rates of lung cancer were calculated as the number of cancer cases/deaths divided by the total man-years at risk. To examine patterns of cancer incidence and mortality over time, the Nelson-Aalen estimator was used to plot the cumulative hazard for both incident and fatal cancers by screening arm. Screening efficacy was estimated as the ratio of lung cancer mortality rates in the dual-screen arm to those in the x-ray only arm; analysis was by intention-to-screen. Ninety-five percent confidence intervals for rates and rate ratios were calculated according to Rothman19 using STATA version 9.0.
The Hopkins trial randomized 10,386 eligible subjects (5,160 to the x-ray only arm and 5,226 to the dual-screen arm), while the Sloan-Kettering trial randomized 10,040 (5,072 and 4,968 into the x-ray only and dual-screen arms, respectively). Within each trial, the screening groups were similar with respect to age, race, smoking, asbestos exposure, and duration of follow-up. Between trials, these distributions were also similar except for a notable excess of asbestos exposure at Hopkins. (Table 1). Overall compliance with the screening invitations during the first five years of the study (during which exams were offered to all study participants) was high in both arms (Figures 1 and and2).2). There was, however, a tendency for reduced compliance over time. Compliance was well over 99 percent at baseline for all screening tests in both arms of both trials, but had decreased at Hopkins at the 5-year exams to 75, 68, and 67 percent for x-ray attendance in the x-ray only and dual-screen arms and cytology attendance in the dual-screen arm, respectively. Similar decreases in compliance were seen at Sloan-Kettering, with corresponding proportions of 62, 58, and 60 percent at five years. Attendance was under 50 percent in the sixth year and about 20 percent in the seventh year for both trials; though Sloan-Kettering did occasionally offer an eighth year of screening, less than one percent of living, lung cancer-free subjects attended this exam.
During follow-up (mean 7.2 years for both arms of both trials), there were 475 confirmed cases of lung cancer at Hopkins and 343 at Sloan-Kettering. Despite higher overall lung cancer incidence rates at Hopkins, adenocarcinoma was reported at higher rates at Sloan-Kettering. This was offset by a very low number of large cell cancers identified at Sloan-Kettering (Table 2). There was an excess of squamous cell cancers in the dual-screen as compared to the x-ray only arms early on during follow-up; however, by nine years, the number of squamous cancers was similar between the two arms (Figure 3A). The reverse pattern was noted for non-squamous cancers, with an excess of cases in the x-ray only arm early, and a coming together of the cumulative hazard plots by nine years (Figure 3B). This excess of non-squamous cases was driven largely by an increased number of large cell carcinomas diagnosed in the x-ray only arm at Hopkins (Table 2).
In the combined data there was a modest trend towards more cancers being diagnosed at earlier stages in the dual-screen as compared to the x-ray only group; however, this was driven by squamous cell cancers at Hopkins only; other histologies exhibited no clear stage shift (Table 3). A large majority of lung cancers detected by cytology only were squamous cell cancers (Table 4).
There were 1,533 deaths (329 due to lung cancer) at Hopkins and 1,208 deaths (224 due to lung cancer) at Sloan-Kettering. At Hopkins, eight of the lung cancer deaths were due to study-initiated procedures, including seven occurring in those with confirmed lung cancer and one in a subject with “probable” lung cancer. At Sloan-Kettering, there were ten deaths as a result of study-initiated procedures. Nine occurred in subjects with confirmed lung cancer and one in a subject with no indication of a lung cancer diagnosis; the latter was in the x-ray only arm.
Overall, lung cancer mortality rates were about 10 percent lower in the dual-screen than the x-ray only group, though this did not reach statistical significance. The relative difference in lung cancer mortality between screening arms was larger at Hopkins than at Sloan-Kettering. Reduced lung cancer mortality rates in the dual-screen arm were seen only for squamous cell and large cell carcinomas; adenocarcinoma and small cell lung cancer mortality were slightly higher in the dual-screen group. These patterns were seen for both the Johns Hopkins and Sloan-Kettering trials. Differences in screening efficacy estimates were also seen according to smoking history, with no apparent benefit of screening in those with less than 50 pack-years of smoking and an approximately 20 percent reduction in risk for those with 50 or more pack-years; again, this pattern was consistent across both trials (Table 5). When risks of lung cancer mortality were examined over time, there was a slight excess of squamous cancer deaths in the dual-screen as compared to the x-ray only arm early in follow-up, which was more than offset by a decrease in squamous cancer deaths in later years (Figure 4A); there was a small excess of non-squamous cancer deaths in the x-ray only arm which persisted over the entire length of follow-up (Figure 4B). Mortality due to causes other than lung cancer was similar between arms at both Hopkins (RR 0.99, 95% CI 0.88-1.11) and Sloan-Kettering (RR 1.06, 95% CI 0.93-1.20).
Despite the increased power obtained by combining data from the Johns Hopkins Lung Project and the Memorial Sloan-Kettering Lung Study and by considering all lung cancer deaths during nine years of follow-up, we were unable to demonstrate a statistically significant lung cancer mortality benefit associated with the addition of four-monthly sputum cytology screening to an annual chest x-ray regimen. However, there was a suggestion of both a modest benefit among the heaviest smokers and a moderate reduction in deaths due to squamous cell and large cell lung cancer.
The smoking-stratified efficacy estimates and the findings for squamous cell cancer were not surprising. Others have suggested that lung cancer screening would be expected to be most effective in the highest-risk groups (specifically heavy smokers with airflow obstruction)20. Also, sputum cytology is most suited for the detection of squamous cell cancers, which tend to be located centrally and to exfoliate early in their natural history15. Within these data, the findings for squamous cell carcinoma were remarkably consistent: over half of the squamous cancers in the dual-screen arm were detected by cytology, and there were larger numbers of squamous cancers detected earlier during follow-up and at earlier stages in the dual-screen arm (suggesting that these cancers were diagnosed earlier than they otherwise would have been, when they would potentially be more amenable to successful treatment). All of these findings are consistent with, though not conclusive of, the notion that the reduced squamous cell cancer mortality in the dual-screen arm could be due to a true, modest benefit of sputum cytology screening.
The large cell cancer results are more difficult to interpret. Although the efficacy estimates for large cell lung cancer mortality were similar for both trials, we did not see a stage shift for large cell cancers, and relatively few large cell cancers (less than 10 percent) were detected by cytology. At Hopkins, the observed difference in large cell cancer mortality seems likely to be because substantially fewer large cell cancers were diagnosed in the dual-screen group, which was unexpected. These findings suggest that the reduction in mortality observed for large cell lung cancers may be due to chance. This underscores the fact that we had limited power to conduct these histology-specific subanalyses, and therefore these findings (including those for squamous cell cancers) should be interpreted cautiously.
Because these findings are based on screening exams and lung cancer deaths from over 20 years ago, it is worthwhile to consider how changes in cancer incidence and technologic advances may impact the efficacy of lung cancer screening with sputum evaluation. Since the time of these trials, there has been a relative increase in the incidence of lung adenocarcinomas, and a resulting modest decrease in the proportion of squamous cell carcinomas21-23. If, in fact, any benefit of sputum cytology is largely restricted to a reduction in squamous cell carcinoma mortality, then a decrease in the proportion of squamous cell lung cancers would tend to reduce the efficacy of cytology in preventing lung cancer death. On the other hand, a number of techniques have been developed which may make the detection of cancers using sputum evaluation more effective, and therefore may make earlier detection more likely as compared to the available technology from the 1970s and 1980s. For example, molecular analysis of archived sputum samples has detected p53 and ras mutations24 and promoter hypermethylation25 in patients who were later diagnosed with lung cancer. Recent studies have also suggested that chronic tobacco smoke exposure induces a persistent change in gene expression throughout the respiratory epithelium (a field effect), and that individuals who develop cancer may be distinguished by the pattern and extent of these changes in exfoliated cells26. It should be emphasized, however, that none of these newer technologies has been tested in large-scale, well-designed randomized controlled trials.
There are a number of strengths and limitations to these analyses which should be considered. An obvious strength is that the efficacy estimates are based on populations that were randomized to a dual-screen versus x-ray only arm, and are therefore much less susceptible to bias than those of non-randomized studies of cancer screening. Additionally, the emphasis on mortality as a measure of screening efficacy eliminates the effects of lead-time, length, and overdiagosis biases, which are inherent to comparisons that examine case survival as a measure of screening efficacy27. However, there are a number of limitations that should also be acknowledged. Most notably, even combining data from the two trials, we had limited statistical power to detect a benefit of cytology screening. Also, the fact that there was no unscreened comparison group complicates the interpretation of these results, as it is unclear what lung cancer mortality would have been in the absence of screening. Finally, there were numerous challenges in performing these analyses many years following the conduct of the trials. For example, we noted a much lower incidence of large cell carcinoma at Sloan-Kettering than at Hopkins. This is likely to be at least partially related to differences in how adenocarcinoma and large cell carcinoma were classified at the two sites28. However, slides from the diagnosed lung cancers were not available, such that we could not base histologic subtype analyses on re-review of slides using a uniform classification scheme.
Despite these limitations, the Hopkins and the Sloan-Kettering trials represent the best information we have to date regarding the efficacy of lung cancer screening with sputum cytology. These results suggest that the use of sputum cytology may result in a modest (perhaps 10 percent) decrease in lung cancer mortality. However, given the uncertainty of any benefit, and given that, even if present, the benefit is likely to be small, the data presented here suggest that sputum cytomorphology has limited utility as a screening tool.
Financial Support: The trials were supported by the following grants and contracts from the National Cancer Insitute:NO1-CN-45007 (Memorial Sloan-Kettering Lung Study) and N01-CB-92172, N01-CN-45037, M01-RR-00035-21, and RR00722 (Johns Hopkins Lung Project).
Financial Disclosures: None of the authors have any financial disclosures to report.