Does the evidence support the implementation of lung cancer screening with low-dose computed tomography?

KEYWORDS:

• Lung cancer
• low dose CT
• screening
• RCTs
• benefits to harm ratio

In attempting to evaluate whether the current evidence supports low-dose computed tomography (LDCT) lung cancer screening, in the appropriate high-risk population, it is useful to consider, sequentially, the following three questions:

1. Does evidence from randomized controlled trials (RCTs) show that LDCT screening reduces lung cancer mortality?

2. Does evidence from RCTs show that the benefits of LDCT screening outweigh the harms?

3. What direct or indirect evidence is there that LDCT screening performed in the U.S. as part of routine clinical care (and outside of a research setting) can attain the same benefits to harms ratio as seen in RCTs, or at least attain a minimally acceptable benefits to harms ratio?

With respect to the first question, the U.S.-based National Lung Screening Trial (NLST) comparing LDCT to chest radiograph screening enrolled over 53,000 persons aged 55–74 at high risk for lung cancer based on smoking history (30+ pack-years and current smoker or quitting within 15 years) [1]. The initial 2011 NLST report showed a lung cancer mortality decrease of 20% (95% CI: 7–27%), which after more complete follow-up decreased slightly to 16% (95% CI: 8–24%) [1,2]. Importantly, there was also a statistically significant decrease in all-cause mortality of 6.7%.

Four other trials, all small (2500–4000 subjects) and conducted in Europe, and all comparing LDCT screening to usual care (no screening), have reported mortality outcomes to date [3–6]. Three reported a relative risk (RR) for lung cancer mortality of 1.0 or greater, though with wide confidence intervals, and a fourth reported a RR of 0.70 (95% CI: 0.47–1.03). All were under-powered as individual trials for the outcome of lung cancer mortality. An intermediate-sized trial, NELSON, with about 15,000 enrolled, has been under way for some time in Europe, and is expected to report mortality findings in 2018 [7].

Assessing all the trials with reported mortality endpoints, the evidence does support a reduction in lung cancer mortality with LDCT screening. Further, as long as NELSON shows some evidence of a mortality reduction with LDCT screening, even if not statistically significant, say a reduction of 10% or more, this assessment would not change upon publication of its results. If NELSON shows no evidence at all of a benefit, then a careful assessment of all the randomized trial data, including a formal meta-analysis, would be in order.

Given the affirmative answer to question 1, we can next evaluate the benefits to harms trade-off of LDCT screening. A commonly used metric for assessing this trade-off in cancer screening is the number needed to screen (NNS), or the number of persons needing to be screened (a given number of times) to prevent one death from the cancer of interest [8]. For NLST, the NNS was around 300, indicating that 300 persons meeting the NLST eligibility criteria need to undergo three annual screens to prevent one lung cancer death [1]. The higher the background mortality rate of the cancer of interest (M0) and the greater the mortality reduction (1-RR) of screening, the lower the NNS (note NNS = 1/(M0 × (1-RR))). Since lung cancer has the highest mortality rate of any cancer in the U.S., and since a high-risk population is readily identifiable and targeted for screening, the background (lung cancer) mortality rate in the population recommended for screening is very high, roughly 8-fold higher than for breast cancer in the population recommended for mammography [9,10]. Further, for comparable screening regimens, annually for several years, trial evidence shows a generally similar, but perhaps slightly larger, percentage mortality reduction (i.e. 1-RR value) for mammography as for LDCT [11]. Even granted a somewhat greater (percentage) mortality reduction for mammography, however, the much smaller background mortality rate (M0) for breast versus lung cancer still results in an NNS value approximately five times larger for mammography compared to LDCT screening in their respective recommended screening populations.

Although the NNS provides a metric to assess the benefits versus harms and efficiency of screening, not all screens are alike in terms of their potential for harm (or costs). Therefore, a lower NNS for one screening modality over another could be counter-balanced by a higher potential for harms per screen. This potential for harms includes the rate of false positive screens, the frequency and type of diagnostic follow-up procedures of positive screens and their risks of complications, and the risk of overdiagnosed cases.

A major caveat of the NLST findings was the high false positive rate (FPR). Since screening involves a healthy population, screened multiple times over their lifetime, it is desirable that the FPR (per screen) be low, optimally below 10% or even 5%. In NLST, the FPR in the first two rounds was quite high, around 25%, and somewhat lower, 16%, in the final round due to a modified screen positivity definition considering nodule stability over time [1]. This high FPR muted the enthusiasm over the trial’s demonstrated reduction in lung cancer mortality.

A mitigating factor, however, of the high FPR was the relatively low rate of subsequent invasive procedures and complications thereof; approximately 2.5% of false positives resulted in an invasive procedure, and only 0.06% of false positives resulted in a major complication from an invasive procedure [1]. Most diagnostic follow-up consisted of only repeat CT scans. In the European trials, the FPR was generally lower than in NLST [3–6]. There was some evidence of overdiagnosis, but the rate was low compared to breast or prostate cancer screening [12]. There is also radiation exposure, but the associated risk is low [13]. The U.S. Preventive Services Task Force (USPSTF), in evaluating the evidence, largely from the NLST, concluded that LDCT screening is of moderate net benefit in persons at high risk of lung cancer (NLST criteria with age modified to 55–80) [14,15]. That assessment seems reasonable, at least based on the evidence from RCTs. Due to limited evidence at the time, however, the USPSTF assessment did not address the issue of LDCT screening as practiced in real-world settings.

That brings us to the last question. In translating from a trial setting to clinical practice, the benefits of many interventions are attenuated due to various factors. Therefore, if the net benefit from the trial evidence was only moderate, in the task force’s words, any substantial attenuation of the benefit, or increase in harms, could tip the balance.

The MEDCAC advisory panel to the Center for Medicare and Medicaid Services (CMS), in discussing Medicare coverage of LDCT screening in 2014, recommended against such coverage, voting an average of 2.2 on a scale of 1–5 on whether the evidence was adequate that the benefits of LDCT screening outweighed the harms [16]. The panel’s major concerns related to how LDCT screening would be implemented in clinical practice, and specifically, whether the appropriate persons, in terms of general health status and lung cancer risk, would be screened.

CMS did decide to cover LDCT screening, but with conditions imposed to address some of MEDCAC panel’s concerns. These included requiring a shared decision-making visit for the initial screen and a written order from a physician (or equivalent) for all screens, and also requiring that the screen be entered into a CMS-approved LDCT screening registry.

The intent of requiring the registry was to enable monitoring of screening in practice, and thus the registry is a good place to evaluate current LDCT performance. The sole CMS-approved LDCT registry is maintained by American College of Radiology (ACR); summary data is available on the registry website on approximately 160,000 screens performed in 2016 and 110,000 performed in the first half of 2017 [17]. Note the registry includes data on all persons screened, not just those who are Medicare-eligible. Some of the data are encouraging. Almost 90% of screened subjects meet the USPSTF criteria for screening, which suggests that the appropriate persons are being screened in terms of lung cancer risk, although this is based on self-reported smoking history. A more objective criterion for this assessment, though, is the cancer detection rate of screening. For calendar year 2016 screens, it was 4.8 per 1000 for baseline and 2.1 per 1000 for post-baseline screens, or about half the rate at baseline and one fourth the rate post-baseline of NLST. However, the registry is not set up to obtain complete follow-up of subjects, so some of this decrease could be the result of under-reporting. Linkage of the ACR registry with the NCI SEER database (in SEER regions) is currently being considered; this will allow better assessment of not only the cancer detection rate, but also of cancer characteristics, including stage, histology, and survival, in relation to screen results. For example, comparison of the stage distribution and survival of screen detected cancers in the registry with those in NLST can help assess the performance of LDCT screening in practice.

A factor that could potentially reduce the harms of LDCT screening is the development in 2014 of the Lung-RADS evaluation system, which is currently in wide use [18]. Analogous to BIRADS for mammography, the purpose of Lung-RADS was to standardize reporting for LDCT screening but also to reduce the FPR. A retrospective analysis of NLST showed that applying Lung-RADS criteria to the reported LDCT findings reduced the FPR to about 13% at baseline and 5% post-baseline (while modestly reducing test sensitivity) [19]. In practice, though, the FPR appears to be higher than predicted above, around 19% for baseline and 12% for post-baseline screens based on the registry data, which is still higher than desirable for a screening test [17]. Note that in population screening, with annual exams for up to 25 years, the vast majority of screens eventually will be post-baseline (incident); thus, effective management of new and pre-existing nodules found on incident screens will be critical.

Other ‘real-world’ evidence of LDCT screening in clinical practice in the U.S. comes from the Lung Cancer Screening Demonstration Project (LCSDP), which was initiated by the Veterans Health Administration (VHA) in 2013 to assess the feasibility and implications of implementing LDCT screening in the VHA system [20]. Eight VHA hospitals participated, out of 35 facilities who applied. The design of the project was first to identify, through electronic health records, patients meeting age (55–80), comorbidity (no history of lung and certain other cancers), and minimum life expectancy criteria. Next, patients were assessed for smoking history based on clinical reminders developed specifically for the LCSDP; these included assessments of pack-years and of certain clinical factors, including symptoms suggestive of possible lung cancer. Finally, those meeting the LCSDP eligibility criteria (30+ pack-years and either current smoking or quitting within 15 years, plus some clinical appropriateness factors) underwent shared decision-making on LDCT screening. For the LDCT scan itself, the Fleischner Society guidelines were used; accordingly, the presence of any solid or mixed nodule regardless of size (or a ground glass nodule ≥5mm) was considered a positive screen, in the sense that the nodule should be either tracked or evaluated.

Of 93,033 primary care patients initially identified as meeting age, comorbidity, and life-expectancy criteria, 18,083 met the smoking history criteria, of which 5035 (28%) were assessed for appropriateness of screening; many of the remainder were not assessed for logistical reasons due to the phased roll-out of the project and variability in implementation procedures across sites [20]. Of 4246 patients assessed as appropriate, 2106 (50%) completed LDCT screening. The overall screen positivity rate was 60%, with considerable variability across the eight sites (range 31–85%). Incidental findings that would likely require follow-up were found in 41% of patients (range across sites 20–63%). The overall lung cancer yield was 1.5% (31 total cancers). The VHA estimates that 900,000 of a total 6.7 million VHA patient population meet the eligibility criteria for LDCT screening and that the implementation of LDCT screening in the VHA system will require substantial clinical effort.

Outside of the U.S., other approaches are being considered with respect to various aspects of LDCT screening implementation, including eligibility. The recently released European Union (EU) position statement on lung cancer screening advises using a risk prediction model to select those eligible for screening [21]. Such models generally include as inputs demographic and smoking history variables, and may also utilize family history of lung cancer and aspects of medical history (e.g. history of COPD). This contrasts with approaches in the U.S., which generally set an age range, minimum pack-year level, and maximum time since quit (for former smokers) for eligibility criteria.

The EU position statement also recommended that management of screen-detected solid nodules utilize semi-automatically derived volume measurement and volume-doubling time [21]. In the U.S., a volumetric approach is not generally recommended; for example, neither the National Comprehensive Cancer Network clinical guidelines, the American Thoracic Society/American College of Chest Physicians policy statement, nor the Lung-RADS criteria recommend a volumetric approach to nodule management [18,22,23].

It is estimated that approximately 8 million Americans are eligible for lung cancer screening, yet based on the ACR registry data, only 2–3% of those are currently being screened. This low rate of uptake may be due in part to continued uncertainties about how LDCT screening will perform in the general population setting. Based on the available evidence, it was reasonable to recommend LDCT screening for the high-risk population, given their burden of lung cancer. However, continued careful assessment of how LDCT screening is being carried out in practice is required to make sure the benefits to harm trade-off in those being screened remains substantially positive.

Declaration of interest

The author has no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties. A reviewer on this manuscript has disclosed that they are the principal investigator of one of the studies listed in the manuscript. Peer reviewers on this manuscript have no other relevant financial or other relationships to disclose.

Additional information

Funding

This manuscript has not received funding.