Secondary Analysis of the NLST data with simpler comparative methods between CXR and LDCT
WISSAM S.A. AL-JANABI, MD, MPH/Biostat Wayne State University, Detroit, Michigan
ZIAD DM FAWZI, MD, University of Baghdad Collage of Medicine
YASIR DURAID MOHAMMED FAWZI, MD, Al-Mustansiriya University
Editor: Abdalla H. Sadoon
Conflicts of Interest: The author has no conflicts of interest to declare.
Keywords: Low Dose Computed Tomography (LDCT), National Lung Screening Trial (NLST).
Abbreviations: Danish Lung Cancer Screening Trial (DANTE); International Early Lung Cancer Action Program (I-ECLAP); Lung Screening Study (LSS).
The death rate from lung cancer is highest amongst all cancers; it comprises approximately 20% of all cancer death. After decades of striving to find a screening tool similar to Chest x-ray (CXR) and blood biomarkers for the deadliest cancer in the world, three decades ago, the screening with Low Dose Computed Tomography (LDCT) began. Unless the patient becomes symptomatic with a cough, hemoptysis, weight loss, this cancer was hard to detect.
Even though smoking cessation is the best way to reduce mortality and morbidity from lung cancer, LDCT showed its ability to identify lung cancer earlier and thus decrease the death rate from lung cancer in countries that can afford to use this tool. LDCT can decrease all-cause mortality by approximately 7% and lower lung cancer mortality by about 20%. LDCT has high sensitivity when compared to the CXR. In addition to detecting late-stage cancer, LDCT can also detect early-stage lung cancer (stage I), which can decrease mortality as well as morbidity. When first introduced as a screening tool for lung cancer, clinicians and scientists raised concerns about radiation exposure, cost, psychological effects, and high false-positive rates. Due to these concerns, countries like the USA and some European countries were hesitant to approve LDCT as a screening tool for two decades. Notwithstanding, in 2013, the United State Preventive Services Task Forces (USPSTF) gave the LDCT a B recommendation as a screening tool for lung cancer.
For both men and women, lung cancer has the highest mortality rate amongst all cancers1. The International Agency for Research on Cancer estimated deaths from lung cancer to be 1.18 million in 2007, which will rise to 10 million deaths in 2030 1,2.
There are two main types of lung cancer: nonsmall cell lung cancer (NSCLC), which is the most common type, and small cell lung cancer, which is about 20%2. In all kinds of lung cancers, cigarettes were suspected of being the main culprit2.
The increase in both cigarette smoking and lung cancers in the early 1950s propelled a British doctor to conduct a retrospective study which subsequently identified a link between lung tumors and cigarette smoking2.
Lung cancer is associated with the number of cigarettes smoked per day3. The risk of lung cancer increases about 60-70 fold for a person who smokes two packs/day compared to a nonsmoker3. In the United States, smoking is inversely related to education3. The prevalence of cigarette smoking is highest (41%) among adults with a General Educational Development (GED), followed by 24.1% for individuals with less than a high-school education. Smoking is lowest among adults with graduate degrees (4.5%)3. According to the Centers for Disease Control (CDC), in the United States alone, there are about 38 million individuals who smoke and over 16 million who live with a smoker3. However, 15% of lung cancers occur in non-smokers, which is usually due to a mutation in the cytosine kinase domain of epidermal growth factor receptor (EGFR)3,4.
As lung cancer is the most prevalent cause of death amongst all other cancers for both males and females, it is important to find a useful screening tool that can detect the tumor in its early stage (stage I) and thereby reducing its mortality and morbidity rate4. Finding an appropriate screening tool will improve early detection and treatment, and limits morbidities before death from this incurable cancer4. Although researchers have developed many screening methods for different diseases and cancers, few were effective before the National Lung Screening Trial (NLST) has identified a screening strategies effective in diagnosing lung cancer at an early stage4,5. Many studies have sought to identify an effective tumor marker or screening tool (e.g. chest x-ray (CXR) and sputum cytology) to detect lung cancer, but unfortunately, none of these methods have supplied convincing data for lowering the mortality rate5. Until the NLST Trial, CXR was the first step for screening symptomatic patients5. Recent studies by The National Lung Screening Trial (NLST) tested the efficacy of low dose CT-scans (LDCT) for screening high-risk patients for lung cancer and established the significance of LDCT in lowering the morbidity and mortality rate and improving the quality of life5,6.
Review of Literature:
In 2010 the NLST reported that CXR screened about 3.4 million persons and LDCT screened about 1.8 million individuals 4. The NLST study found LDCT screening can decrease mortality by 20%4. They also found that this screening method can decrease other mortality rates by 7% because this tool can catch other diseases that could affect organs, such as the liver, adrenal glands, and kidneys4. Initially, many scholars raised serious concerns about using LDCT as a screening tool4. Concerns were raised about cost-effectiveness, negative psychological impacts from false positive results, and radiation exposure4. As the use of LDCT has been studied, each of these concerns has been addressed5. Despite all of these hurdles, the United State Preventive Services Task Forces (USPSTF) approved LDCT as a screening tool for lung cancer in high-risk patients and gave it a type B recommendation5. Currently, most health company insurances cover the cost of this screening tool5,6.
Emphasizing on the importance of this proposal in figure one simplify and show the number of patients who were detected as a positive result during the screening period of NLST trial for both CXR and LDCT. The numbers below are per the entire sample (53,454 subjects). 26,722 were randomly assigned to LDCT screening test, and 26,732 to the CXR screening test. The T symbol used in the below graph represents a screening time; subjects were screened yearly during the NLST trial for three successive years.
Since the results of the NLST were reported, many clinicians have supported its findings5,6. Consequently, this has led many researchers to lunch studies to scrutinize the LDCT as a screening tool for lung cancer. Unsurprisingly, most of these studies demonstrated results similar to the NLST findings5,6. By replicating the results of the NLST, these studies provide additional support for the use of LDCT5,6.
A study in 2006 found that LDCT can detect lung cancer in its earliest stage (stage I), which is a curable stage, and that LDCT has improved the 10-year survival of lung cancer dramatically7,8,9. Therefore, curing cancer like this will, without any doubt, decrease the morbidity of this cancer and improve the lives of those patients7,8. Noteworthy, the NLST study found about 40% of those who were diagnosed with lung cancer by LDCT have stage IA of that cancer9. Although those people who were diagnosed with an early stage of lung cancer needed surgery, and some of them needed chemotherapy, they regained their health and functioning ability soon after the surgery10. However, people who are diagnosed with this cancer in later stages get only chemo and radiotherapy as palliative therapy because they are at a metastatic and inoperable stage5,9,11. The 5-year survival for patients with stage I is over 50% compared to patients with stage III/IV who have a 5-year survival rate of lower than 5%10,12.
Project Rationale: The primary aim of this project is to provide information about the importance of LDCT as a method for screening for lung cancer in a way that is easily understood by laypersons. The parent study for the NLST team used survival analysis to compare the mortality rate between LDCT and CXR. Even though the use of survival analysis for comparing mortality is completely appropriate, the results of these analyses are not always easily understood by persons who lack appropriate statistical training. This lack of understanding can limit the dissemination of the benefits of LDCT to non-academic audiences. A method that can compare LDCT and CXR for detecting lung cancer in a way that is more easily understood is Receiver Operating Characteristic (ROC) curves. ROC can be used to compare the ability of each method to detect lung cancer by calculating the accuracy of the two diagnostic tools and then testing for differences in accuracy. By providing a more easily understood method for comparing the LDCT and CXR, this project will aid the layperson to understand the importance of the LDCT.
Mathematical advantages of ROC: the formula of the ROC is Sensitivity plotted against 1-specifity. The sensitivity of a test means the ability to correctly detect diseased patients. Whereas the specificity refers to the ability of the same test to identify the non-diseased people. The sensitivity is inversely related to the specificity. Additionally, ROC accounts for false positive numbers in each test.
This project utilized data from the NLST. Complete methodological details of the NLST study can be found at (NLST research team trial, 2012) Data from the LDCT was obtained through the nih.gov website (https://biometry.nci.nih.gov/cdas/nlst/).
The parent study: NLST recruited participants from August 2002 through September 2004 through more than 30 participating medical institutions nationwide. Participants eligible for the study were between 55 and 74 years of age at the time of randomization and had a history of cigarette smoking of at least 30 pack-years. If former smokers, they had to have quit within the previous 15 years. Participants excluded had a previous diagnosis of lung cancer, had undergone chest CT in past 18 months, had hemoptysis, or an unexplained weight loss of more than 15 lbs. in the past 12 months. A total of 53,456 participants were recruited and randomly assigned to either CXR (N=26,733) or LDCT (26,723) screening. Each subject then was screened yearly for three consecutive years (2004-2007) to check for developing lung cancer. Following that, subjects were followed prospectively and cohortly until December of 2009 for any events.
The study measures the rate of incidence of lung cancer per person-years at risk for event7. The follow up started from the time of randomization until the time of diagnosis or censoring (whichever comes first)7 the latest date of follow up was December 31, 2009. The NLST team calculated the Confidence Interval (C.I) and consider a poisson distribution7. The number needed to be treated (NNT) was calculated by the reciprocal of the absolute relative risk7.
Providing a survival analysis results will not give a clear image on benefits of LDCT to clinicians or to laypersons because this type of analysis is sophisticated to be comprehended unless you have a strong background in biostatistics7. Further, survival analysis has many weaknesses. Lead-time bias is a well-known weakness in survival analysis. Lead time bias is the time frame between diagnosing a disease and emerging of the symptoms. Other weakness in survival analysis is overly diagnosing patients, the higher the sensitivity of any test the more people which can get falsely diagnosed with disease/condition and this subsequently causes some psychological distress7. Another noteworthy about the NLST trial is that they used the poisson distribution to calculate the C.I, and the poisson distribution is not a normal distribution and it uses an exponential formula for calculate the mean7. With poisson distribution the variance is equal to the mean which creates a suspicion about the how realistic is this result7.
Positive cases were defined as any participant with a lung nodule of 0.4 cm or larger or a radiographic image that revealed any noncalcified nodule or mass.
Current study: The same inclusion criteria were applied but with stretching age to 80 years old, and cases with missing or uncertain data were excluded. The total subjects in the current study were 37,170.
Statistical Analyses: SAS 9.4 (TSIM3) was used to perform secondary statistical analysis. Initially, results from the three screening periods for each subject, T0, T1, T2, (each T contains the observations for subjects who participated in the baseline, first and second screening periods respectively) were merged into a single variable indicating whether the subject was negative or positive for lung cancer. This was done for subjects in the CXR and LDCT screening groups. This resulted in a single observation (confirmed lung cancer Y/N) for each subject. The main outcome for this study was confirmed lung Cancer (Y/N).
To assess the ability of each method (LDCT and CXR) to detect lung cancer, Receiver Operator Curve (ROC) with chi-square statistic and 95% Confidence Interval (C.I.) for the accuracy of each method were performed. Mathematically, there is an inverse relationship between sensitivity and specificity. Hence, the plot of sensitivity against 1- specificity forms what is called the Receiver Operator Curve (ROC) and the area under the curve. To compare the two screening methods, a graph, as well as a chi-square statistic, were performed. Following the initial analyses, subgroup analysis was performed to compare the two screening methods across age groups (55-59; 60-64; 65-69; and ≥70) and gender. Nonetheless, for avoiding redundancy of the graphs, only chi-square results for the subsequent analysis were reported in this study. The follow-up analysis showed that age is not a confounder for the detection of lung cancer.
The ROC method that can compare LDCT and CXR for detecting lung cancer in a way that is more easily understood than survival analysis. ROC can be used to compare the ability of each screening method to detect lung cancer by calculating the accuracy of the two diagnostic tools and then testing for differences in accuracy. There are a number of mathematical advantages of ROC. First, the formula of the ROC is Sensitivity plotted against 1-specifity. The sensitivity of a test means the ability to correctly detect diseased patients. Whereas the specificity refers to the ability of the same test to identify the non-diseased people. The sensitivity is inversely related to the specificity. Additionally, ROC accounts for false positive numbers in each test. Take a glimpse at the below 2x2 table, mathematically someone can understand how does ROC account for false positive tests. The specificity formula is comprised True negative/False positive plus True negative. Hence, it accounts for false positive by dividing the sensitivity of a test by specificity which includes the false positive result in its denominator. And this tackles two issues, the false positive result and the exaggerated sensitivity that burden the LDCT itself.
The ROC analytic method is simple to understand as it gives a lot of information just based on look to the graph, and it provides a chi-square as well as a confidence interval. ROC is devoid of lead time and length time biases. Additionally, ROC uses a normal distribution not a poisson distribution; hence, it gives more realistic results.
The results of the Receiver Operator Curves for LDCT and CXR predicting lung cancer are presented in Figure 6 and Tables 1 and 2. As can be seen in Figure 6, the difference between the sensitivity of CXR when compared to LDCT is large and significant (p < 0.0001). Area for LDCT is quite significant compared to CXR area. There is a difference of 0.4447 between the two modalities. Moreover, the table below shows the statistical significance when contrasting both screening tools. The χ 2 = 4834.3 with P = <.0001, and 95% Wald C.I (0.43-0.45).
The results of the analysis comparing LDCT and CXR across age are presented in table 3 below. As can be seen in Table 3, LDCT is a better screener than CXR for all age levels.
Table 1 below displays some of the characteristic features for the sample that was used particularly for this analysis and in this study.
The burden of lung cancer mortality is very high in Western countries; hence, finding a screening tool for those who are at high risk for developing lung cancer is a crucial step in the era of medical revolution13,15.
Despite the deterrent actions taken by a large number of countries around the globe against smoking, the below bar charts information are gleaned from the World Health Organization (WHO) website. The left one reflects the mortality rate per 100,000 in males and females in each continent, and the right bar charts display the median quotidian cigarette smoking in each continent for both males and females13.
In over five cohort studies, the sensitivity of LDCT was reported to be over 90%, and specificity ranges between 28-100%12,14. These results conform to the ROC results in this study. The sensitivity of LDCT reaches 99% compared to the ordinary CXR, which carries a sensitivity of about 55%14,15. The further stratification of age into four subgroups in this study failed to reveal any confounding factor, and the ROC of each subgroup came identical to the original ROC of the entire sample size. This indicates that the benefit of the LDCT over CXR is consistent for all ages14,15.
The NLST study was conducted from 2002-2006; after 2006, the patients were followed up until 200916,17. Over six years of follow up in NLST, the lung cancer mortality was 249 per 100,000 in LDCT group compared to 309 per 100,000 in CXR group, with a reduction in lung cancer mortality of 20% and all-cause mortality by about 7% 16-21. Only 16% of lung cancer is caught during stage I, whereas with LDCT the rate of lung cancer caught at stage I rises to 70%16,22. NLST found that the number needed to be screened via LDCT in order to prevent one lung cancer is 320 16,23-25. Consequently, USPSTF updated their recommendation in 2013 for LDCT16,24. Interestingly, the 10-year survival for lung cancer detected via LDCT goes over 80% and reaches 88% with stage I16,24.
The rest of the discussion will aim to address the four main concerns from using the LDCT as a screening tool. One of the major concerns in using the LDCT for lung cancer screening is the high false positive rate. Recent research has shown that using lung Imaging Reporting and Data System criteria (which is abbreviated as Lung-RAD) could substantially reduce the false positive rate in LDCT. The caveat is that this technique also reduces the sensitivity of the test25. Given the much higher sensitivity of LDCT compared to CXR, as demonstrated in the current ROC analyses, LDCT would still be a better screening tool even with a reduction in sensitivity.
Furthermore, the ROC in this analysis demonstrated the higher sensitivity of the LDCT while accounting for the false positive number. The ROC analysis is assessing the sensitivity of a screening test while accounting for the false positive rate. ROC formula utilizes the sensitivity of the test as a function of the false positive rate (100-specificity) 25.
Another concern is the cost of LDCT. Initially, scientists were concerned with the cost of LDCT and whether insurances will cover the procedure. A new German study was published in July of 2018, estimated the cost-effective of LDCT in high-risk population between the age of 55-75 years old. The German study had three outcomes: Costs, life years saved, and quality-adjusted life years (QALYs). The study concluded that the use of LDCT in high risk-populations is cost-effective, and shows an incremental gain in life years (0.06 per person) and QALYs (0.04 per person)30,21,26. Further, currently, most of the insurance companies in the United States have approved the coverage for the LDCT in high-risk populations7,26. In 2018 surprisingly, Green et al. found no statistical significance between the cost of LDCT compared to CXR31.
Another concern was the psychological stress on the patient while he/she is waiting for the result or if the result was a false positive. In 2016, the result of a large randomized clinical trial was published. This study was launched to assess the psychosocial impacts of the LDCT in patients who participated in the LDCT screening trial. A questionnaire was sent on three time-intervals: T0 (baseline screening), T1 (after the second screening), and T3 (after the third screening) 26-29. The questionnaire was to assess a patient’s depression, anxiety, and distress or satisfaction from the LDCT screening program. Of a sample over 4,000 subjects, the results revealed no clinically significant psychosocial impacts on those patients 1-9, 26-29.
The last concern about the LDCT was the amount of radiation exposure from this tool. Table 3 compares the amount of radiation between screening tools that emit radiation and have already been approved by the Food and Drug Administration (FDA). The amount of radiation from LDCT is almost negligible and is lower than the amount of radiation of other screening tools, like a mammogram17,27-28. Although the LDCT carries about 1.4 mSv radiation exposure compared to CXR which is 0.02 mSv, the later did not prove its
Summary of the results for all randomized clinical trials in USA & Europe:
Both NLST and ELCAP had the same conclusion that the LDCT has a higher sensitivity than CXR.
Nonetheless, the DANTE trial did not reach the same conclusion. Reasons for this included small sample size, including only males, and a control group based on yearly clinical exam, sputum exam and baseline CXR. Noteworthy, in the DANTE study, there was more detection of patients with stage I compared to other studies16,23,32.
Although the LSS study was a pilot study, it results supported the NLST conclusion. LSS found that positive rate for LDCT was 25.3% compared to 8.7% forCXR16-19. In addition, the LSS concluded that LDCT has five times more powerful than the CXR in detecting non-calcified nodules16.
Italian Lung Study (ITALUNG) was a small-scale randomized clinical trial (RCT), and it was EU-US collaboration, its results are the same results of the NLST and ELCAP17,20-23.
NELSON is a Dutch-Belgian lung cancer screening trial, and it is considered the largest clinical trial that compares LDCT to the usual care. It showed that the positive predictive value for the LDCT is 40.4%, which is significantly greater than what the NLST trial found (3.8%)16. Further, the NELSON study revealed that annual screening is as effective as biannual screening, which in other words, the biannual screening is redundant11-16. Overall, LDCT is considered better than other screening tools that have been approved for other cancer, such as fecal occult or blood test for colon cancer, mammogram for breast cancer25-28.
Finally, noteworthy, a meta-analysis that was done by Dr. Gopal et al. showed that for every 1,000 patients screened via LDCT, there are nine will have stage I lung cancer, and 320 with a benign lung nodule16.
Reducing the risk of death by 20% if the high-risk patients use the LDCT screen test annually has become a fact. A solution to one of the biggest issue in public health is found; most health insurances cover this kind of screening test1-9,22. United State Preventive Services Force Tasks (USPSFT) considered screening for lung cancer in a high-risk patient using LDCT is type B recommendation, which means according to their data, there is a moderate certainty that using LDCT as a screening test for lung cancer yields moderate or substantial benefits to the high-risk patients10,11,17-20
LDCT has shown tangible improvement in the patient quality of life as well as reduced mortality rate from lung cancer and other diseases, yet still, some scholars and insurance companies endorsing this new technique with scans.7-9,16The Ongoing study of LDCT in the United Kingdom, Germany, and other European countries will divulge more data to address issues such as cost, psychological anxiety, and repeated radiation exposure25-28. By far, LDCT is still the best method to screen and detect lung cancer in its early stages 1Collecting more data regarding this method and running more studies about this technique will help us find good solutions and defeat the number one cause of death in men and women worldwide19-28.
Acknowledgments to Abdalla Sadoon in finalizing the manuscript.