The National Lung Screening Trial: Results stratified by demographics, smoking history, and lung cancer histology
Abstract
BACKGROUND
The National Lung Screening Trial (NLST), which compared lung cancer screening with low-dose computed tomography (LDCT) versus chest radiography (CXR), demonstrated a statistically significant mortality benefit of LDCT screening. In the current study, the authors performed a post hoc analysis to examine whether the benefit was affected by various baseline factors, including age, sex, and smoking status, and whether it differed by tumor histology.
METHODS
Lung cancer death rates were computed as events over person-years of observation; the mortality risk ratio (RR) was defined as the lung cancer death rate in the LDCT versus CXR trial arms. Poisson regression was used to test for interactions of sex, age (< 65 years vs ≥ 65 years), and smoking status (current vs former) with trial arm. Mortality RRs were also computed for specific lung cancer histologies.
RESULTS
The overall mortality RR was 0.92 in men and 0.73 in women, with a P value for interaction of .08. RRs were similar for individuals aged < 65 years versus those aged ≥ 65 years (0.82 vs 0.87), and for current versus former smokers (0.81 vs 0.91). By tumor histology, mortality RRs were 0.75 for adenocarcinoma, 0.71 for all non-small cell lung cancers except squamous, 1.23 for squamous cell carcinoma, and 0.90 for small cell carcinoma. RRs were similar for men and women for nonsquamous non-small cell lung cancers (0.71 and 0.70, respectively); women were found to have lower RRs for small cell and squamous cell carcinoma.
CONCLUSIONS
A benefit of LDCT did not appear to vary substantially by age or smoking status; there was weak evidence of a differential benefit by sex. A differential benefit across lung cancer histologies may exist. Cancer 2013;119:3976–3983. Published 2013. This article is a U.S. Government work and is in the public domain in the USA.
INTRODUCTION
The National Lung Screening Trial (NLST) demonstrated a 20% relative reduction in lung cancer mortality for annual screening over 3 years with low-dose computed tomography (LDCT) compared with chest radiography (CXR).1 The eligibility criteria for NLST included an age range of 55 years to 74 years and a smoking history of at least 30 pack-years, with either current smoking status or having quit within the past 15 years; both men and women were included.
The NLST results prompted a discussion of how, and in which population groups, LDCT screening might be implemented in the United States.2-4 Addressing this question requires an understanding of how LDCT screening is performed in various subgroups within the NLST. Such subgroups include men and women, younger (aged < 65 years) and older subjects, and current and former smokers. Although the NLST was powered to find a statistically significant difference in lung cancer screening for the entire NLST population, there might be differences in the benefit of LDCT between subgroups.
For some lung cancer histologies, specifically small cell carcinoma, LDCT screening may not be effective.5, 6 It is also possible that the effectiveness of LDCT screening may differ among the histologies comprising non-small cell lung carcinoma (NSCLC). Again, although the NLST was not powered to find significant mortality differences for specific tumor histologies, trends can be examined, including differential survival rates across trial arms as well as mortality risk ratios (RRs). Note that the analysis of LDCT effectiveness by histology differs fundamentally from analyses of effectiveness by demographics or smoking history. In the latter, predefined subsets of the population may be identified and targeted to preferentially receive (or not receive) LDCT screening based on relative effectiveness; in contrast, for histology, it is not known a priori what histology subjects will be diagnosed with. However, whether LDCT screening is differentially effective by tumor histology is clearly an important scientific question. Similar analyses of the efficacy of screening tests for different disease histologies or subtypes provided insight regarding the differential ability of Papanicolaou tests to detect squamous versus adenomatous lesions of the cervix.7
In this article, we analyzed mortality RRs (LDCT vs CXR trial arms) for various NLST population subgroups to examine whether they might vary according to sex, age, or smoking status. In addition, we examined mortality rates and survival rates across trial arms for specific lung cancer histologies, including small cell carcinoma, adenocarcinoma, and squamous cell carcinoma.
MATERIALS AND METHODS
The NLST protocol has been described previously.8 Briefly, the NLST randomized subjects aged 55 years to 74 years to LDCT or CXR screening in a 1:1 ratio. Eligibility criteria included a ≥ 30 pack-year history of cigarette smoking and current smoking status or having quit within the past 15 years. Subjects were enrolled at 33 US screening centers between 2002 and 2004 and received either LDCT or CXR screens at baseline and then annually for 2 additional years. An LDCT scan that revealed a noncalcified nodule measuring ≥ 4 mm or a CXR that revealed any noncalcified nodule or mass was categorized as a positive screen. Diagnostic follow-up of positive screens was directed by subjects' individual health care providers; the NLST provided guidelines but no specific approach was mandated.
Incident cancers were tracked through the follow-up of individuals with positive screens and annual status update questionnaires. Deaths were tracked through the annual questionnaires as well as through National Death Index searches. Death certificates were obtained and an independent blinded review process for cause of death was conducted on deaths possibly due to lung cancer. Medical records, including pathology and tumor staging reports, were obtained for all suspected lung cancer cases and NLST coders abstracted lung cancer characteristics, including stage and histology (according to the third edition of the International Classification of Diseases for Oncology). The histological classifications shown in the current study are the same as those reported in the primary NLST outcome article1; because diagnosis and treatment in the NLST was performed outside of the auspices of the trial, histology was determined as part of routine patient care at the numerous clinical centers in which NLST subjects were evaluated.
Statistical Analysis
Subjects were followed for incident lung cancers and deaths through December 31, 2009. Incidence and mortality rates were calculated as the number of events divided by person-years (PY) of follow-up; RRs for the LDCT versus the CXR trial arm were computed as the ratio of rates in the 2 arms.
In the primary NLST outcome article, the final result for lung cancer deaths was derived using an event cutoff date of January 15, 2009, whereas results for incident lung cancers and total deaths were derived using a cutoff date of December 31, 2009.1 The earlier date for lung cancer deaths was used to allow adequate time for these events to undergo endpoint verification. With time passed since that publication, the endpoint verification process was extended to cover deaths through December 31, 2009, with 98% of all lung cancer deaths endpoint-verified. Therefore, for this analysis, we used all lung cancer deaths through December 31, 2009, resulting in some estimates herein that differ from those in the primary outcome article.1
To examine interactions of the mortality RR (lung cancer and overall) with the major covariates of sex, age at randomization (< 65 years vs ≥ 65 years), and baseline smoking status (current vs former), we used Poisson regression with a log link. PY of follow-up for each subject were computed as the time from randomization to either death, loss to follow-up, or December 31, 2009, whichever came first.
Mortality RRs for each tumor histology were computed by treating deaths from other lung cancer histologies as other-cause mortality. We used Kaplan-Meier curves to examine survival for specific histologies. Survival curves were generated both from the time of randomization and from the time of diagnosis; using randomization as the start time eliminates bias due to differential lead time across study arms (however, it should be noted that it does not remove the bias related to overdiagnosis in one trial arm relative to the other). Confidence intervals (95% CIs) on differences in RRs by histology were computed using bootstrapping.
For a global test of whether LDCT efficacy differed among histologies, we assumed outcomes in each trial arm were governed by a multinomial distribution, in which the outcomes were lung cancer death of a given histology or no lung cancer death. We used the likelihood ratio test to compare the model with a single common RR (for the LDCT vs CXR arm) across histologies to a saturated model with a fitted RR for each histology, with the histologies being squamous cell carcinoma, NSCLC excluding squamous cell, small cell, and other/unknown.
A potential bias in the analysis of histology-specific mortality RRs (and histology-specific survival) derives from the classification of NSCLC-not otherwise specified (NOS). In distinguishing between NSCLC histologic classifications (eg, between adenocarcinoma and squamous cell carcinoma), NSCLC-NOS constitutes missing data because the specific subtype of NSCLC is unknown.9 Due to the differential rate of NSCLC-NOS by trial arm, bias could result for specific NSCLC subtypes. To adjust for this bias, we performed simulations (N = 1000) in which, for each simulation, NSCLC-NOS cases were randomly assigned to squamous cell carcinoma (or nonsquamous NSCLC) according to the percentage of squamous cell (33.5%) noted among the cases with known NSCLC subtypes (bronchioloalveolar carcinoma [BAC] was excluded for this purpose because it is very different from other NSCLC, having substantially higher survival and overdiagnosis rates with LDCT). We then averaged over the simulations (N = 1000) to obtain adjusted mortality RRs and adjusted survival rates by NSCLC subtype (squamous and nonsquamous NSCLC).
RESULTS
Table 1 displays the number of enrolled subjects, by trial arm, for the major baseline covariates (sex, age, and smoking status), as well as incident lung cancer counts and rates. A total of 59% of participants were male, 27% were aged ≥ 65 years at the time of study enrollment, and 48% were current smokers. Among men and women, the percentages who were current smokers and who were aged ≥ 65 years were similar (eg, 47% of men and 50% of women were current smokers; data not shown in table).
| Covariate Subset | Trial Arm | Total Enrolled No. | No. of Incident Lung Cancers | Incidence Rate (per 100,000 PY) | Incidence RR (95% CI)a |
|---|---|---|---|---|---|
| All | LDCT | 26,722 | 1089 | 662 | — |
| All | CXR | 26,730 | 969 | 588 | — |
| No. (% Within Arm) | |||||
| Women | LDCT | 10,953 (41) | 434 | 638 | Referent |
| Men | LDCT | 15,769 (59) | 655 | 678 | 1.06 (0.94–1.20) |
| Women | CXR | 10,969 (41) | 395 | 580 | Referent |
| Men | CXR | 15,761 (59) | 574 | 594 | 1.02 (0.90–1.16) |
| Age <65 y | LDCT | 19,612 (73) | 620 | 508 | Referent |
| Age ≥65 y | LDCT | 7110 (27) | 469 | 1100 | 2.16 (1.92–2.44) |
| Age <65 y | CXR | 19,622 (73) | 549 | 450 | Referent |
| Age ≥65 y | CXR | 7108 (27) | 420 | 981 | 2.18 (1.93–2.49) |
| Former smoker | LDCT | 13,862 (52) | 452 | 522 | Referent |
| Current smoker | LDCT | 12,860 (48) | 637 | 816 | 1.56 (1.39–1.76) |
| Former smoker | CXR | 13,830 (52) | 365 | 421 | Referent |
| Current smoker | CXR | 12,900 (48) | 604 | 773 | 1.84 (1.62–2.09) |
- Abbreviations: 95% CI, 95% confidence interval; CXR, chest radiography; LDCT, low-dose computed tomography; PY, person-years; RR, risk ratio.
- a RR across covariate categories within the trial arm (eg, for men vs women within the LDCT arm).
Lung cancer incidence rates were similar between men and women within each trial arm. Incidence RRs were approximately 2.2 in each arm for those aged ≥ 65 years compared with those aged < 65 years; for current smokers (compared with former smokers), the incidence RR was 1.56 in the LDCT arm and 1.84 in the CXR arm.
Mortality results are shown in Table 2. Our updated analysis (using the later cutoff date of December 31, 2009) continued to demonstrate an overall reduction in lung cancer mortality (RR, 0.84; 95% CI, 0.75-0.95). For sex by trial arm, there was a borderline significant interaction (P = .08), with the mortality RR for women found to be more protective than that for men (RR of 0.73 vs 0.92). For both age and smoking status, there were no significant interactions noted for the lung cancer mortality RR, with P values for interaction of > .40. For all-cause mortality, there was no suggestion of interaction between the mortality RR and any of the 3 major covariates (P value for interaction range, 0.61-0.84).
| Covariate | Trial Arm | No. of Lung Cancer Deaths/Death Ratea | RRb | Pc | No. of Total Deaths/Death Ratea | RRb | Pc |
|---|---|---|---|---|---|---|---|
| Subset | |||||||
| All | LDCT | 469/280 | .84 | 1912/1141 | .931 | ||
| All | CXR | 552/332 | Referent | 2039/1225 | Referent | ||
| Women | LDCT | 158/228 | .73 | 574/828 | .921 | ||
| Women | CXR | 215/312 | Referent | 619/899 | Referent | ||
| Men | LDCT | 311/316 | .92 | 1338/1361 | .936 | ||
| Men | CXR | 337/345 | Referent | .08 | 1420/1454 | Referent | .84 |
| Age <65 y | LDCT | 253/205 | .82 | 1059/856 | .942 | ||
| Age <65 y | CXR | 307/250 | Referent | 1117/909 | Referent | ||
| Age ≥65 y | LDCT | 216/491 | .87 | 853/1943 | .918 | ||
| Age ≥65 y | CXR | 245/562 | Referent | .60 | 922/2116 | Referent | .67 |
| Current smoker | LDCT | 294/369 | .81 | 1146/1437 | .944 | ||
| Current smoker | CXR | 360/455 | Referent | 1206/1523 | Referent | ||
| Former smoker | LDCT | 175/199 | .91 | 766/872 | .914 | ||
| Former smoker | CXR | 192/220 | Referent | .40 | 833/954 | Referent | .61 |
- Abbreviations: CXR, chest radiography; LDCT, low-dose computed tomography; RR, risk ratio.
- a Per 100,000 person-years.
- b RR of LDCT versus CXR arm within covariate category.
- c Interaction of RR with given covariate.
Table 3 summarizes lung cancer cases and deaths by trial arm according to histology. Adenocarcinoma comprised approximately 30% of the lung cancer deaths in each arm (29% in the LDCT arm vs 33% in the CXR arm) and small cell approximately 20% in each trial arm. Squamous cell carcinoma comprised 22% of deaths in the LDCT arm and 15% of deaths in the CXR arm, whereas NSCLC-NOS comprised 10% of deaths in the LDCT arm and 14% of deaths in the CXR arm.
| Trial Arm | LDCT | CXR | LDCT | CXR | Lung Cancer Mortality, RR (95% CI) |
|---|---|---|---|---|---|
| Histology | Cases | Cases | Lung Cancer Deaths | Lung Cancer Deaths | |
| No. (%) | No. (%) | No. (%) | No. (%) | ||
| BAC | 111 (10) | 36 (4) | 13 (3) | 10 (2) | 1.3 (0.58–2.9) |
| Adenocarcinoma | 389 (35) | 337 (34) | 136 (29) | 181 (33) | 0.75 (0.60–0.94) |
| Large cell | 40 (4) | 44 (4) | 17 (4) | 24 (4) | 0.71 (0.38–1.3) |
| NSCLC-othera | 48 (4) | 49 (5) | 25 (5) | 34 (6) | 0.74 (0.44–1.2) |
| NSCLC-NOS | 89 (8) | 113 (11) | 45 (10) | 76 (14) | 0.59 (0.41–0.86) |
| All NSCLC minus squamous (excludes BAC) | 566 (51) | 543 (55) | 223 (48) | 315 (57) | 0.71 (0.60–0.84) |
| Squamous cell | 249 (22) | 214 (22) | 102 (22) | 83 (15) | 1.23 (0.92–1.64) |
| Small cell | 143 (13) | 163 (16) | 102 (22) | 113 (20) | 0.90 (0.69–1.18) |
| Carcinoid | 6 (0.5) | 3 (0.3) | 1 (0.2) | 0 | |
| Unknown | 34 (3) | 34 (3) | 28 (6) | 31 (6) | 0.90 (0.54–1.5) |
| All | 1109 | 993 | 469 | 552 | 0.84 (0.75–0.95) |
- Abbreviations: 95% CI, 95% confidence interval; BAC, bronchioloalveolar carcinoma; LDCT, low-dose computed tomography; CXR, chest radiography; NOS, not otherwise specified; NSCLC, non-small cell lung cancer; RR, risk ratio.
- a Excludes adenocarcinoma, large cell, squamous cell, BAC, and NSCLC-NOS.
Observed mortality RRs (LDCT arm vs CXR arm) were 0.75 (95% CI, 0.60-0.94) for adenocarcinoma, 1.23 (95% CI, 0.92-1.64) for squamous cell carcinoma, and 0.90 (95% CI, 0.69-1.18) for small cell carcinoma. For all NSCLC except squamous cell (and excluding BAC), the mortality RR was 0.71 (95% CI, 0.60-0.84); this RR was significantly different from the RR for squamous cell carcinoma (95% CI of log ratio of RRs, 0.18-0.83). The global test of differential LDCT efficacy demonstrated that the null hypothesis of equal RRs across histologies was rejected at the level of P = .01 compared with the alternative hypothesis of separate RRs for the major histologies (squamous cell, NSCLC excluding squamous cell, small cell, and other).
Survival rates by trial arm and histology are shown in Table 4. Analyzed by time from randomization, 3-year and 6-year survival rates (lung cancer-specific) among adenocarcinoma cases were significantly greater in the LDCT arm (89.4% and 71.6%, respectively) than in the CXR arm (81.9% and 54.5%, respectively). Survival rates were similar for all nonsquamous NSCLC cases as for adenocarcinoma. In contrast, for squamous cell cases, survival was similar between trial arms; 3-year and 6-year survival rates (from randomization) were 88.7% and 66.6%, respectively, in the LDCT arm versus 89.3% and 65.7%, respectively, in the CXR arm. NSCLC-NOS was found to have a worse survival than either squamous cell carcinoma or adenocarcinoma, irrespective of trial arm. The above results for both survival and mortality did not adjust for the bias due to the imbalance in NSCLC-NOS between arms (24 more cases in the CXR arm). After adjustment, which apportioned the NSCLC-NOS cases between squamous and nonsquamous NSCLC, the (estimated) numbers of deaths from squamous cell carcinoma was increased to 116.7 (LDCT arm) and 108.5 (CXR arm) compared with the original counts of 102 (LDCT arm) and 83 (CXR arm), thereby decreasing the mortality RR for squamous cell carcinoma to 1.08 (95% CI, 0.86-1.31) from the original 1.23. For nonsquamous NSCLC, the adjusted mortality RR changed only slightly, to 0.72 from the original (unadjusted) 0.71. The RR for squamous cell carcinoma remained significantly greater than that for nonsquamous NSCLC in the adjusted analysis (95% CI of log ratio of RRs, 0.14-0.65). For the adjusted analysis of survival (from randomization), survival rates for squamous cell carcinoma were still similar across arms, if slightly better for LDCT, with 3-year and 6-year survival rates of 87.8% and 65.4%, respectively, versus 87.1% and 61.7%, respectively, for the CXR arm. For all nonsquamous NSCLC, (adjusted) the 3-year and 6-year survival rates were 87.3% and 67.8%, respectively, for the LDCT arm versus 79.2% and 49.4%, respectively, for the CXR arm.
| Histologic Type | Start of Follow-Up for Analysisa | LDCT ArmSurvival Rateb | CXR ArmSurvival Rateb | p-value |
|---|---|---|---|---|
| 3 Years/6 Years | 3 Years/6 Years | |||
| Adenocarcinoma | Diagnosis | 69.7/59.1 | 43.5/33.2 | <.0001 |
| Randomization | 89.4/71.6 | 81.9/54.5 | <.0001 | |
| NSCLC-NOS | Diagnosis | 46.1/43.8 | 29.5/24.4 | .004 |
| Randomization | 80.9/55.4 | 75.0/39.5 | .034 | |
| All NSCLC minus squamous (excludes BAC) | Diagnosis | 64.4/54.3 | 38.4/30.7 | <.0001 |
| Randomization | 87.0/67.2 | 79.1/49.1 | <.0001 | |
| Squamous cell | Diagnosis | 59.5/50.7 | 58.6/48.5 | .82 |
| Randomization | 88.7/66.6 | 89.3/65.7 | .68 | |
| 1 Year/3 Years/6 Years | 1 Year/3 Years/6 Years | |||
| Small cell | Diagnosis | 56.2/15.8/14.4 | 49.9/21.3/11.5 | .72 |
| Randomization | 97.9/76.2/39.1 | 95.7/77.2/37.8 | .80 |
- Abbreviations: BAC, bronchioloalveolar carcinoma; NOS, not otherwise specified; NSCLC, non-small cell lung cancer.
- a Start of follow-up for survival analysis: either date of diagnosis or date of National Lung Screening Trial randomization.
- b Lung cancer-specific survival.
Table 5 shows lung cancer deaths by sex, histology, and study arm. For adenocarcinoma, and more broadly nonsquamous cell NSCLC, the mortality RRs were essentially equivalent for men and women (0.71 for men versus 0.70 for women). Mortality RRs were greater for men than women for both small cell carcinoma (1.10 for men vs 0.67 for women) and squamous cell carcinoma (1.31 for men vs 1.04 for women).
| Men | Women | |||||
|---|---|---|---|---|---|---|
| LDCT | CXR | Mortality RR (95% CI) | LDCT | CXR | Mortality RR (95% CI) | |
| Histologic Type | No. (%) | No. (%) | No. (%) | No. (%) | ||
| BAC | 10 (3) | 6 (2) | 1.7 (0.6–4.4) | 3 (2) | 4 (2) | 0.75 (0.2–3.0) |
| Adenocarcinoma | 85 (27) | 111 (33) | 0.77 (0.6–1.02) | 51 (32) | 70 (33) | 0.73 (0.51–1.05) |
| Large cell | 11 (3) | 18 (5) | 0.61 (0.3–1.3) | 6 (4) | 6 (3) | 1.0 (0.3–3.0) |
| NSCLC-othera | 15 (5) | 21 (6) | 0.71 (0.4–1.4) | 10 (6) | 13 (6) | 0.77 (0.4–1.7) |
| NSCLC-NOS | 26 (8) | 42 (12) | 0.62 (0.4–1.01) | 19 (12) | 34 (16) | 0.56 (0.3–0.98) |
| All NSCLC minus squamous (excludes BAC) | 137 (44) | 192 (57) | 0.71 (0.6–0.9) | 86 (54) | 123 (57) | 0.70 (0.5–0.9) |
| Squamous cell | 77 (25) | 59 (18) | 1.31 (0.9–1.8) | 25 (16) | 24 (11) | 1.04 (0.6–1.8) |
| Small cell | 68 (22) | 62 (18) | 1.10 (0.8–1.6) | 34 (22) | 51 (24) | 0.67 (0.4–1.03) |
| Carcinoid | 1 (0.3) | 0 | 0 | 0 | ||
| Unknown | 13 (4) | 7 (2) | 1.9 (0.8–4.5) | 14 (9) | 10 (5) | 1.4 (0.6–3.1) |
| All except small cell and squamous cell | 166 (53) | 216 (64) | 0.77 (0.6–0.9) | 99 (63) | 140 (65) | 0.71 (0.5–0.9) |
| All except small cell | 243 (78) | 275 (82) | 0.88 (0.7–1.05) | 124 (78) | 164 (76) | 0.76 (0.6–0.96) |
| All | 311 | 337 | 0.92 (0.8–1.08) | 158 | 215 | 0.73 (0.6–0.9) |
- Abbreviations: 95% CI, 95% confidence interval; BAC, bronchioloalveolar carcinoma; CXR, chest radiography; LDCT, low-dose computed tomography; NOS, not otherwise specified; NSCLC, non-small cell lung cancer; RR, risk ratio.
- a Excludes adenocarcinoma,large cell, squamous cell, BAC, and NSCLC-NOS.
DISCUSSION
In this post hoc analysis of the NLST mortality results according to the major demographic and smoking behavior covariates, we found no evidence of an interaction between the all-cause mortality risk ratio (LDCT arm vs CXR arm) and age, sex, or smoking status. For the lung cancer-specific mortality RR, which was the primary outcome for the current trial, we found no evidence of an interaction by age or smoking status; however, a borderline significant interaction by sex was observed, with women having a more protective effect from LDCT than men.
With respect to histology, we found differences in the mortality effect of LDCT screening by histologic subtypes in our exploratory analysis. For adenocarcinoma, and more broadly all NSCLC combined excluding squamous cell, the mortality RRs were significantly below 1 (0.75 and 0.71, respectively), thus demonstrating a substantial LDCT screening benefit. In contrast, for squamous cell carcinoma, mortality rates were higher in the LDCT arm, with an RR of 1.23 (95% CI, 0.92-1.64), which decreased modestly to 1.08 (95% CI, 0.86-1.31) when adjusting for the potential bias associated with the NSCLC-NOS classification. Although the lower 95% CI for the squamous cell mortality RR is consistent with a modest 10% to 15% benefit of LDCT screening, it should also be noted that the RR for all NSCLC excluding squamous cell (as well as for adenocarcinoma alone) was statistically significantly less than that for squamous cell carcinoma.
In the current analysis, we separated BAC from the adenocarcinoma cases, not including deaths from BAC in the adenocarcinoma relative risk estimate. However, after the NLST was completed, a new histologic classification scheme for adenocarcinoma was unveiled that eliminated the category of BAC and replaced it with adenocarcinoma in situ, minimally invasive adenocarcinoma, or invasive adenocarcinoma.10 For a subset of the NLST BAC cases, tissue specimens were retrospectively collected and a centralized pathology was performed using the new classification scheme. Of the subset reanalyzed, all of the fatal cases in each trial arm were classified as invasive adenocarcinoma. Based on these results, and the finding that adenocarcinoma in situ and minimally invasive adenocarcinoma cases are typically indolent, it is likely that most or all of the deaths in originally classified BAC cases were actually from invasive adenocarcinoma. If we thus add all the BAC deaths (13 in the LDCT arm and 10 in the CXR arm) to the adenocarcinoma category, the RR for adenocarcinoma would only increase slightly, from 0.75 to 0.78.
With relatively small numbers of deaths when broken down by histologic subtype, the mortality RR could be influenced by a chance imbalance in the numbers of incident cancers across trial arms. For that reason, and to compare our results with those of observational studies in the literature, we also analyzed lung cancer survival by histology. To control for lead-time bias, we computed survival from randomization as well as from diagnosis. We found that squamous cell carcinoma cases did not have better survival in the LDCT arm compared with the CXR arm, in contrast to the findings for adenocarcinoma. This provides further evidence that LDCT screening in NLST was not effective for squamous cell carcinoma. In contrast to the survival findings reported herein, a Japanese study demonstrated that both adenocarcinomas and squamous cell carcinomas had significantly greater survival when detected by LDCT than CXR.5 Note that this study did not adjust for lead-time bias; however, in the current analysis, even when analyzed from the time of diagnosis, squamous cell survival was not found to be better in the LDCT arm compared with the CXR arm. Other studies examining volume doubling time have generally found that volume doubling times were significantly shorter (indicating faster growth) for squamous cell carcinomas than for adenocarcinomas.11, 12 These findings are consistent with a smaller benefit of LDCT screening for squamous cell carcinoma compared with adenocarcinoma, because the window for screening effectiveness could thus be shorter for squamous cell carcinoma.
It has generally been assumed that small cell carcinoma is not amenable to LDCT screening due to its aggressive natural history and early metastases.5, 6 In the current study, the data were consistent with this assumption, with an RR that was not significantly different from 1 (RR, 0.9; 95% CI, 0.69-1.18) and no survival advantage for cases in the LDCT arm versus the CXR arm.
The analysis of mortality by histology sheds light on the possible greater benefit of LDCT screening in women versus men. For adenocarcinoma, the histology with the strongest case for a benefit of LDCT screening, the mortality RR was essentially the same for men as for women (RR, 0.77 and 0.73, respectively), as it was for all NSCLCs excluding squamous cell. The greater benefit for women comes almost exclusively from small cell and squamous cell carcinomas. If screening is ineffective for small cell, then the variations across trial arm and sex in deaths from small cell carcinoma may simply be chance occurrences, which should be ignored when evaluating the relative benefit of LDCT screening in men and women; RRs for men and women for all lung cancers excluding small cell were 0.88 versus 0.76, compared with 0.92 versus 0.73 for all lung cancers, a smaller differential. To the extent that screening for squamous cell carcinoma is also ineffective, then the same would be true for these deaths; excluding both squamous and small cell deaths gave similar RRs of 0.77 for men and 0.71 for women. Therefore, the true difference in LDCT screening effectiveness by sex, if one exists at all, is likely smaller than that observed in the NLST. Note that if screening was more effective for adenocarcinoma than for squamous cell carcinoma, then the finding that squamous cell carcinoma is relatively less prevalent in women compared with men would logically lead to the mortality benefit for all lung cancer being greater in women than in men. However, an analysis based on the NLST data indicated that this difference in effectiveness would most likely be slight. Using the observed LDCT mortality RR for NSCLC excepting squamous cell of 0.71, assuming true RRs of 1.0 for squamous cell and small cell, and given the histology distribution in the NLST (CXR trial arm) of 64% nonsquamous NSCLC in women versus 58% in men, the mortality RR (for all lung cancers) would be 0.81 in women and 0.83 in men, which is a quite modest differential.
As described above, in the current analysis we used a later cutoff date for lung cancer deaths than did the primary outcome artcle,1 resulting in the overall lung cancer RR increasing from 0.80 to 0.84. The effect of using the later cutoff date can also be examined in terms of other summary measures of screening efficacy, such as the number needed to screen (NNS) with LDCT to prevent 1 lung cancer death and the absolute difference in lung cancer death rates across trial arms. The NNS provides a rough measure of the harms-benefit tradeoff of screening, whereas the absolute difference in death rates provides a more direct measure (compared with the RR) of an individual's potential gain from screening. The NNS changed little, from 307 using the original (earlier) cutoff date to 322 using the later cutoff date (note an NNS of 320 was reported in the primary outcome article1; this was computed among those receiving at least 1 screen, whereas the numbers in the current study reflect an intent-to-screen analysis). The difference in death rates across trial arms decreased modestly from 62 per 100,000 PY originally to 52 per 100,000 PY. Therefore, the same qualitative conclusions regarding the efficacy of LDCT screening can be made based on the later compared with the earlier data.
An analysis that was conducted using the earlier cutoff date and also using weights (rising linearly from time 0 to 4 years after randomization and flat thereafter), as was done in the primary outcome article,1 did not change our finding of no statistically significant interaction between the randomization arm and any of the 3 major subgroups.
The question addressed in the current study of whether the mortality RR differs according to sex, age, or smoking status is different from that of whether there is a differential harms-benefit tradeoff of LDCT screening according to these characteristics.4 For example, if the mortality RR is similar for current versus former smokers, then the NNS would be greater for former smokers because the underlying death rates are lower. Although the observed RR was 0.81 for current smokers and 0.91 for former smokers, the lack of a significant P value for interaction leads to an assumption of equivalent RRs. With an (assumed) common RR of 0.84, the NNS would be 462 for former smokers versus 230 for current smokers, based on an approximately 2-fold higher background lung cancer death rate for current versus former smokers in the NLST. Thus, even with similar mortality RRs, differences in underlying death rates between population subgroups can modify the harms-benefit tradeoff of screening.
Conclusions
The benefit of LDCT did not appear to vary substantially by age or smoking status; there was weak evidence of a differential benefit by sex. A differential benefit across lung cancer histologies may exist.
FUNDING SUPPORT
This research was funded by the following National Institutes of Health grants and contracts: U01-CA-80098, U01-CA-79778, N01-CN-25511, N01-CN-20012, N01-CN-20013, N01-CN-20014, N01-CN-20015, N01-CN-20016, N01-CN-20018, N01-CN-25522, N01-CN-25524, N01-CN-75022, N01-CN-25476, and N02-CN-63300.
CONFLICT OF INTEREST DISCLOSURES
Dr. Church is supported by a National Cancer Institute contract grant and has received support from the National Cancer Institute for travel to meetings for the current study or other purposes.
