Advertisement

Stage Migration and Lung Cancer Incidence After Initiation of Low-Dose Computed Tomography Screening

Published:September 07, 2022DOI:https://doi.org/10.1016/j.jtho.2022.08.011

      Abstract

      Introduction

      Despite evidence from clinical trials of favorable shifts in cancer stage and improvements in lung cancer-specific mortality, the effectiveness of lung cancer screening (LCS) in clinical practice has not been clearly revealed.

      Methods

      We performed a multicenter cohort study of patients diagnosed with a primary lung cancer between January 1, 2014, and September 30, 2019, at one of four U.S. health care systems. The primary outcome variables were cancer stage distribution and annual age-adjusted lung cancer incidence. The primary exposure variable was receipt of at least one low-dose computed tomography for LCS before cancer diagnosis.

      Results

      A total of 3678 individuals were diagnosed with an incident lung cancer during the study period; 404 (11%) of these patients were diagnosed after initiation of LCS. As screening volume increased, the proportion of patients diagnosed with lung cancer after LCS initiation also rose from 0% in the first quartile of 2014 to 20% in the third quartile of 2019. LCS did not result in a significant change in the overall incidence of lung cancer (average annual percentage change [AAPC]: −0.8 [95% confidence interval (CI): −4.7 to 3.2]) between 2014 and 2018. Stage-specific incidence rates increased for stage I cancer (AAPC = 8.0 [95% CI: 0.8–15.7]) and declined for stage IV disease (AAPC = −6.0 [95% CI: −11.2 to −0.5]).

      Conclusions

      Implementation of LCS at four diverse health care systems has resulted in a favorable shift to a higher incidence of stage I cancer with an associated decline in stage IV disease. Overall lung cancer incidence did not increase, suggesting a limited impact of overdiagnosis.

      Keywords

      Introduction

      Lung cancer represents a substantial portion of the overall burden of cancer worldwide and accounted for approximately 235,000 deaths in the United States in 2021.
      • Siegel R.L.
      • Miller K.D.
      • Fuchs H.E.
      • Jemal A.
      Cancer statistics, 2021.
      Survival is strongly associated with stage of disease at time of diagnosis, but historically, most lung cancers are diagnosed at the late stage when curative treatment options are expensive and survival probabilities are limited.
      • Lu T.
      • Yang X.
      • Huang Y.
      • et al.
      Trends in the incidence, treatment, and survival of patients with lung cancer in the last four decades.
      Five-year stage-specific survival rates range from greater than 53% to 69% for patients diagnosed with stage I disease to less than 10% for patients with distant organ metastasis.
      • Chansky K.
      • Detterbeck F.C.
      • Nicholson A.G.
      • et al.
      The IASLC lung cancer staging project: external validation of the revision of the TNM stage groupings in the eighth edition of the TNM classification of lung cancer.
      ,
      • Ganti A.K.
      • Klein A.B.
      • Cotarla I.
      • Seal B.
      • Chou E.
      Update of incidence, prevalence, survival, and initial treatment in patients with non–small cell lung cancer in the US.
      In 2011, the National Lung Screening Trial (NLST) revealed a 20% relative reduction in lung cancer mortality with annual low-dose computed tomography (LDCT) among high-risk individuals.
      • Aberle D.R.
      • Adams A.M.
      • et al.
      National Lung Screening Trial Research Team
      Reduced lung-cancer mortality with low-dose computed tomographic screening.
      The Dutch–Belgian lung cancer screening trial (Nederlands–Leuvens Longkanker Screenings Onderzoek [NELSON]) recently confirmed the mortality benefit of annual LDCT screening.
      • de Koning H.J.
      • van der Aalst C.M.
      • de Jong P.A.
      • et al.
      Reduced lung-cancer mortality with volume CT screening in a randomized trial.
      Efforts in many countries are underway to either initiate or accelerate implementation of LCS given these recent findings. Both the NLST and NELSON trials had a significant shift in the distribution of cancer stage between LDCT and control arms, with more early stage and fewer advanced-stage cancers identified with screening. The presence of this “stage shift” coupled with timely evaluation, diagnosis, and treatment likely represents the primary mechanism through which annual screening results in improved lung cancer-specific mortality. The trial results also support our understanding of the natural history of lung cancer which suggests a progression of disease from undetectable lesions to localized tumors, followed by locoregional involvement, and later development of distant metastases. Nevertheless, concerns persist regarding the detection of a large proportion of indolent cancers resulting in unnecessary diagnosis and overtreatment.
      • Brodersen J.
      • Voss T.
      • Martiny F.
      • Siersma V.
      • Barratt A.
      • Heleno B.
      Overdiagnosis of lung cancer with low-dose computed tomography screening: meta-analysis of the randomised clinical trials.
      Despite the evidence from NLST and NELSON on stage shift and lung cancer-specific mortality, the effectiveness of LDCT screening in clinical practice has not been clearly revealed. Both trials were well conducted, achieving high rates of adherence (>95%) and were performed largely in centers of excellence that may have resulted in optimal outcomes from diagnostic evaluation and subsequent cancer treatment. Enrolled participants were also younger, more educated, and had fewer comorbidities than the population of individuals that are potentially eligible for screening in real-world settings.
      • Burnett-Hartman A.N.
      • Carroll N.M.
      • Honda S.A.
      • et al.
      Community-based lung cancer screening results in relation to patient and radiologist characteristics: the PROSPR consortium.
      ,
      • Iaccarino J.M.
      • Steiling K.A.
      • Wiener R.S.
      Lung cancer screening in a safety-net hospital: implications of screening a real-world population versus the National Lung Screening Trial.
      Concerns have also been raised that harms from screening which may be greater in clinical practice have persisted and may be contributing to the slow uptake of lung cancer screening that has been observed in the United States.
      • Zhao H.
      • Xu Y.
      • Huo J.
      • Burks A.C.
      • Ost D.E.
      • Shih Y.C.T.
      Updated analysis of complication rates associated with invasive diagnostic procedures after lung cancer screening.
      ,
      • Huo J.
      • Xu Y.
      • Sheu T.
      • Volk R.J.
      • Shih Y.C.T.
      Complication rates and downstream medical costs associated with invasive diagnostic procedures for lung abnormalities in the community setting.
      Emerging evidence also suggests that adherence to annual screening is considerably lower than the rates observed in the NLST and NELSON trials, which may diminish the mortality benefit observed in trial settings.
      • Lin Y.
      • Fu M.
      • Ding R.
      • et al.
      Patient adherence to lung CT screening reporting & data system-recommended screening intervals in the United States: a systematic review and meta-analysis.
      • Kunitomo Y.
      • Bade B.
      • Gunderson C.G.
      • et al.
      Racial differences in adherence to lung cancer screening follow-up: a systematic review and meta-analysis.
      • Lopez-Olivo M.A.
      • Maki K.G.
      • Choi N.J.
      • et al.
      Patient adherence to screening for lung cancer in the US.
      Although the assessment of screening benefits, such as lung cancer mortality, is difficult to measure and may only be observed a decade after implementation of LCS in clinical practice, the distribution of cancer stage among screened individuals may allow for an early indication of the impact of screening on lung cancer outcomes. Cancer registry-based analyses and individual lung cancer screening programs have started to report initial outcomes, and many, though not all, have reported favorable distributions of cancer stage among their population of patients diagnosed after screening initiation.
      • Handy J.R.J.
      • Skokan M.
      • Rauch E.
      • et al.
      Results of lung cancer screening in the community.
      • Copeland A.
      • Criswell A.
      • Ciupek A.
      • King J.C.
      Effectiveness of lung cancer screening implementation in the community setting in the United States.
      • Kennedy M.P.T.
      • Cheyne L.
      • Darby M.
      • et al.
      Lung cancer stage-shift following a symptom awareness campaign.
      • Potter A.L.
      • Rosenstein A.L.
      • Kiang M.V.
      • et al.
      Association of computed tomography screening with lung cancer stage shift and survival in the United States: quasi-experimental study.
      Although promising, these studies have been limited by their inability to identify the source population and meaningfully compare stage outcomes among individuals with lung cancer diagnosed among screened and unscreened groups. The aim of this study was to evaluate the impact of LDCT-based screening on shifting the stage distribution toward earlier stage disease across a well-characterized source population, assess the effect of screening on stage-specific incidence over time, and identify factors, in addition to screening, which increase the likelihood of an early stage diagnosis.

      Materials and Methods

      Study Setting and Data Sources

      The Population-based Research to Optimize the Screening Process (PROSPR) Lung Consortium is a collaboration of five diverse health care systems, including Henry Ford Health System, Kaiser Permanente Colorado (KPCO), Kaiser Permanente Hawaii, Marshfield Clinic Health System, and the University of Pennsylvania Health System. The current retrospective cohort analysis excluded Marshfield Clinic Health System owing to incomplete cancer stage information for the full study period. PROSPR-Lung developed a standardized Common Data Model containing data on patients aged 35 to 89 years who were affiliated with any of these five health care systems from January 1, 2010, to September 31, 2019. This Common Data Model includes harmonized data derived from administrative, electronic health record (EHR), and claims systems that are supplemented with limited chart review. Two of the health care systems (KPCO and Kaiser Permanente Hawaii) operate under an integrated care delivery model and the other two (Henry Ford Health System and University of Pennsylvania Health System) limited the cohort to individuals who received primary care within their systems. Variables include patient demographics, details of lung cancer screening, procedures, diagnoses, census-based measures of socioeconomic status, and cancer registry variables collected in a manner consistent with the standards of the North American Association of Central Cancer Registries. Cancer registry data are obtained from manual review by certified tumor registrars and include date of diagnosis, cancer stage, tumor characteristics, and first-course therapy. The study was reviewed and approved by the Institutional Review Board at KPCO, the Institutional Review Board of record for PROSPR-Lung, which waived the informed consent requirement because this observational study presented minimal risks to the participants whose data were analyzed.

      Study Population and Variables

      We restricted the cohort to adults diagnosed with primary in situ or invasive lung cancer between January 1, 2014, and September 30, 2019. Participants were excluded if they had a previous diagnosis of lung cancer, were younger than age 55 years or older than age 80 years, or had a tobacco use history documented as “never” or was missing. The primary outcome variable was cancer stage on the basis of the American Joint Commission on Cancer (AJCC) system in place in the year of diagnosis and was extracted by the local cancer registry at each institution. The primary exposure variable was the performance of at least one LDCT for lung cancer screening before lung cancer diagnosis. LCS-LDCT scans were identified by resulted radiology scans under one of the following CPT or HCPC codes: G0297, S8032, or 71250. We collected sociodemographic and clinical variables, including age, sex, previous malignancy, comorbid conditions, chronic obstructive pulmonary disease (COPD) in the year preceding lung cancer diagnosis, self-reported race and ethnicity, and smoking behavior. We ascertained tobacco use and body mass index at the last date with available data in the EHR before lung cancer diagnosis. We used the patient’s home address mapped to census tract level information to determine the Yost Index as a measure of socioeconomic status.
      • Yu M.
      • Tatalovich Z.
      • Gibson J.T.
      • Cronin K.A.
      Using a composite index of socioeconomic status to investigate health disparities while protecting the confidentiality of cancer registry data.

      Statistical Analysis

      Descriptive statistics were computed using frequencies and proportions to describe the distribution of baseline patient and tumor characteristics among individuals with and without at least one LCS-LDCT before diagnosis and by year of diagnosis. Age-adjusted incidence rates, including both invasive and in situ cancers, were calculated by year between 2014 and 2018 using the age distribution of the U.S. 2000 population as the standard. Patients of each health care system with a recorded encounter within the year of interest were included in the person-years-at-risk denominator calculation. Trends in incidence rates over time were analyzed using joinpoint regression analysis and summarized with average annual percentage change (AAPC) and 95% confidence interval (CI).
      • Kim H.J.
      • Fay M.P.
      • Feuer E.J.
      • Midthune D.N.
      Permutation tests for joinpoint regression with applications to cancer rates.
      Differences in the distribution of categorical variables between screened and unscreened groups and by year of diagnosis were compared using chi-squared tests. Risk differences (and corresponding 95% CIs) were calculated as the difference in the proportion of cancer stage and histologic subgroups by screening history.
      For multivariable analyses, we assessed factors associated with a greater likelihood of an early stage diagnosis. For this analysis, cancer stage was categorized as early stage, defined as an AJCC stage 0, I, or II, and late stage, defined as AJCC stage III or IV. For this component of the analysis, we excluded lung cancer cases with unknown stage, carcinoid tumors, or a diagnosis in 2014 (Fig. 1). The proportion of early and late-stage lung cancer and corresponding 95% CI intervals were calculated overall and across subgroups of a priori identified variables. Generalized estimating equations with a generalized logit distribution and unstructured covariance structure with robust standard errors were used to estimate adjusted ORs and 95% CIs.
      • Hanley J.A.
      • Negassa A.
      • Edwardes MD. deB.
      • Forrester J.E.
      Statistical analysis of correlated data using generalized estimating equations: an orientation.
      Models were clustered on health care system and included the following factors: age at time of lung cancer diagnosis, sex, race and ethnicity, year of diagnosis, cancer subtype (small versus nonsmall cell, excluding carcinoid), smoking history, receipt of previous lung cancer screening, COPD, a previous nonlung cancer malignancy, body mass index, and socioeconomic status as measured by the Yost Index. Customary residual and effect statistics were evaluated to assess model fit and evaluate for outliers.
      Figure thumbnail gr1
      Figure 1Screening volume and lung cancer diagnoses, 2014 to 2019. Bars illustrate the proportion of incident lung cancer diagnoses after at least one previous LCS-LDCT by quarter between 2014 and 2019. The gray line illustrates the total volume of LCS-LDCT scans performed in that quarter. LCS, lung cancer screening; LDCT, low-dose computed tomography; Q, quarter.
      To address missing data for BMI (6%) and Yost Index (3%), we used multiple imputation by chained equations.
      • Azur M.J.
      • Stuart E.A.
      • Frangakis C.
      • Leaf P.J.
      Multiple imputation by chained equations: what is it and how does it work?.
      All variables (outcome, exposure, and covariates) in the outcome model were included as independent variables in the imputation models. We performed 20 iterations of imputation and combined them.
      • Azur M.J.
      • Stuart E.A.
      • Frangakis C.
      • Leaf P.J.
      Multiple imputation by chained equations: what is it and how does it work?.
      ,
      • White I.R.
      • Royston P.
      • Wood A.M.
      Multiple imputation using chained equations: issues and guidance for practice.
      To assess the quality of our imputed data, we compared distributional characteristics preimputation and postimputation. Analyses were performed using SAS version 9.4M6 (SAS Institute Inc., Cary, NC) and Joinpoint Regression Program version 4.8.0.1.

      Results

      Study Population

      A total of 6012 individuals between the ages of 35 and 89 years were diagnosed with a primary lung cancer during the study period. For the primary analysis, we excluded patients with a previous diagnosis of lung cancer (n = 155), age less than 55 years or more than 80 years (n = 1383), unknown/missing tobacco history (n = 322), and individuals who never smoked (n = 474), resulting in a cohort of 3678 individuals with an incident lung cancer; 404 (11%) individuals were diagnosed with lung cancer after screening (Supplementary Fig. 1). The median age of the cohort was 69 years (interquartile range: 64–74 y) and was composed of 1922 female (52%) and 1756 male (48%) patients. There were more previous (2299 [63%]) than current (1379 [37%]) smokers. Most patients were non-Hispanic white (2442 [66%]); non-Hispanic black (713 [19%]); or Asian, Native Hawaiian, or Pacific Islander (276 [8%]) (Table 1). Compared with patients diagnosed with lung cancer in the absence of screening, patients diagnosed after screening were more likely non-Hispanic white, individuals who currently smoke, have greater comorbid illness, and more likely to have a diagnosis of COPD.
      Table 1Baseline Characteristics of Patients Diagnosed With Lung Cancer, 2014 to 2019
      CharacteristicLung Cancer Diagnoses, n (%)p Value
      Overall (N = 3678)No Previous LCS (n = 3274)LCS (n = 404)
      Age at diagnosis0.0020
       55–64 y1025 (28)908 (28)117 (29)
       65–74 y1741 (47)1526 (47)215 (53)
       75–80 y912 (25)840 (26)72 (18)
      Sex0.0889
       Female1922 (52)1727 (53)195 (48)
       Male1756 (48)1547 (47)209 (52)
      Race and ethnicity0.0098
       Non-Hispanic white2442 (66)2147 (66)295 (73)
       Non-Hispanic black713 (19)657 (20)56 (14)
       Asian, Native Hawaiian, or Pacific Islander276 (8)254 (8)22 (5)
       Hispanic112 (93)98 (3)14 (3)
       American Indian, other, or unknown135 (4)118 (4)17 (4)
      Smoking history<0.0001
       Current1379 (37)1161 (35)218 (54)
       Former2299 (63)2113 (65)186 (46)
      Health system<0.0001
       Site 11368 (37)1227 (37)141 (35)
       Site 21031 (28)851 (26)180 (45)
       Site 3414 (11)384 (12)30 (7)
       Site 4865 (24)812 (25)53 (13)
      Modified CCI
      Modified CCI: excludes AIDS diagnosis.
      0.0028
       0589 (16)544 (17)45 (11)
       1514 (14)450 (14)64 (16)
       2563 (15)483 (15)80 (20)
       3+2012 (55)1797 (55)215 (53)
      COPD
      COPD diagnosis code (ICD 9/10: 491, 492, 496, J41, J42, J43, J44) in year before lung cancer diagnosis.
      1683 (46)1438 (44)245 (61)<0.0001
       Previous cancer445 (12)403 (12)42 (10)0.2659
       BMI0.0983
       <251437 (39)1267 (39)170 (42)
       25–291197 (33)1058 (32)139 (34)
       30+991 (27)903 (28)88 (22)
       Missing53 (1)46 (1)7 (2)
      Socioeconomic status (Yost Index)0.4173
       Quintile 1 (lowest)894 (24)810 (25)84 (21)
       Quintile 2603 (16)534 (16)69 (17)
       Quintile 3668 (18)587 (18)81 (20)
       Quintile 4613 (17)535 (16)78 (19)
       Quintile 5 (highest)769 (21)690 (21)79 (20)
       Missing131 (4)118 (4)13 (3)
      AIDS, acquired immunodeficiency syndrome; BMI, body mass index; CCI, Charlson comorbidity index; COPD, chronic obstructive pulmonary disease; ICD, International Classification of Diseases; LCS, lung cancer screening.
      a Modified CCI: excludes AIDS diagnosis.
      b COPD diagnosis code (ICD 9/10: 491, 492, 496, J41, J42, J43, J44) in year before lung cancer diagnosis.

      Lung Cancer Screening Activity and Lung Cancer Incidence

      Lung cancer screening activity, measured by the combined volume of baseline and annual scans, increased steadily over the study period, increasing from a total of 1238 screenings in the first quarter (Q1) of 2014 to 3059 screenings in the third quarter (Q3) of 2019. As screening volume increased, the proportion of patients diagnosed with lung cancer after screening also rose from 0% in Q1 of 2014 to 20% in Q3 of 2019 (Fig. 1). Initiation of LCS did not result in a significant change in the overall incidence of lung cancer (AAPC = −0.8 [95% CI: −4.7 to 3.0]) between 2014 and 2018 (Table 2). Stage-specific incidence rates increased for stage I (AAPC = 8.0 [95% CI: 0.8–15.7]), declined for stage IV (AAPC = −6.0 [95% CI: −11.2 to −0.5]), and were not significantly altered for stage II (AAPC = −7.5 [95% CI: −28.7 to 20.1]) or stage III (AAPC: −3.3 [95% CI: −14.2 to 9.0]) lung cancer (Fig. 2 and Table 2). Very few stage 0 cancers (n = 11) were diagnosed during this time frame, limiting the assessment of incidence rate for in situ cancers.
      Table 2Age-Standardized Overall and Stage-Specific Lung Cancer Incidence Rates, 2014 to 2018
      Year
      Total person-years by year: 413,511.3 (2014); 440,530.4 (2015), 468,131.5 (2016), 490,317.4 (2017), 520,511.2 (2018).
      OverallStage 0Stage IStage IIStage IIIStage IVUnknown Stage
      Lung Cancer DiagnosesLung Cancer RateLung Cancer DiagnosesLung Cancer RateLung Cancer DiagnosesLung Cancer RateLung Cancer DiagnosesLung Cancer RateLung Cancer DiagnosesLung Cancer RateLung Cancer DiagnosesLung Cancer RateLung Cancer DiagnosesLung Cancer Rate
      2014560151.600.0013837.966116.4010026.6324065.01215.6
      2015650164.441.0717844.915915.3812631.4027168.54123.1
      2016652158.500.0020550.23379.0612930.9326063.23215.0
      2017690153.720.4821147.176815.3013228.3525958.06184.3
      2018739150.350.9726354.315211.1511923.6725951.85418.4
      AAPC−0.8 [95% CI: −4.7 to 3.2]N/A
      The AAPC is not reported for stage 0 given the small number of cases which does not allow for an accurate calculation.
      8.0 [95% CI: 0.8–15.7]−7.5 [95% CI: −28.7 to 20.1]−3.3 [95% CI: −14.2 to 9.0]−6.0 [95% CI: −11.2 to −0.5]12.0 [95% CI: −22.1 to 60.9]
      Note: A total of 132 individuals who were aged 80 years were combined with the individuals aged 75 to 79 years for the age-adjustment calculations.
      AAPC, average annual percent change; CI, confidence interval; N/A, not applicable.
      a Total person-years by year: 413,511.3 (2014); 440,530.4 (2015), 468,131.5 (2016), 490,317.4 (2017), 520,511.2 (2018).
      b The AAPC is not reported for stage 0 given the small number of cases which does not allow for an accurate calculation.
      Figure thumbnail gr2
      Figure 2Stage-specific trends in lung cancer incidence, 2014 to 2018. Stage 0 lung cancer excluded given the small number of in situ lung cancer diagnoses (n = 11).

      Impact of Screening on Cancer Histology and Stage Migration

      The distribution of tumor histologic subtype between screened and unscreened patients did not differ significantly between groups (Table 3). In particular, there were no differences in the proportion of two most common histologic subtypes, adenocarcinoma (47.0% versus 42.8%; risk difference [RD] = −4.2 [95% CI: −9.3 to 1.0]) and squamous cell carcinoma (23.1% versus 26.5%, RD 3.4 [95% CI: −1.1 to 8.0]) between the unscreened and screened groups. Screening was, however, associated with an increase in the proportion of stage I cancer (54.7% versus 27.9%, RD 26.8 [95% CI: 21.7–31.9]) with a concomitant decrease in stage IV cancer (17.6% versus 41.7%, RD −24.1 [95% CI: −28.2 to −20.0]) compared with the population diagnosed with lung cancer detected without previous screening. In contrast to the observed rates with stages I and IV, the proportion of stages II and III lung cancer were similar between screened and unscreened groups (Table 2).
      Table 3Differences in Lung Cancer Histological Sub-type and Stage by Previous LCS, 2014 to 2019
      Cancer CharacteristicNo LCS (n = 3274)LCS (n = 404)Risk Difference (95% CI)
      Tumor histology, n (%)
       Adenocarcinoma1538 (47.0)173 (42.8)−4.2 (−9.3 to 1.0)
       Squamous cell755 (23.1)107 (26.5)3.4 (−1.1 to 8.0)
       Large cell22 (0.7)7 (1.7)1.1 (−0.2 to 2.4)
       Non-small cell/other440 (13.4)58 (14.4)0.9 (−2.7 to 4.5)
       Small cell456 (13.9)47 (11.6)−2.3 (−5.6 to 1.1)
       Carcinoid63 (1.9)12 (3.0)1.1 (−0.7 to 2.8)
      AJCC stage, n (%)
       0/I915 (27.9)221 (54.7)26.8 (21.7–31.9)
       II268 (8.2)36 (8.9)0.7 (−2.2 to 3.7)
       III601 (18.4)66 (16.3)−2.0 (−5.9 to 1.8)
       IV1365 (41.7)71 (17.6)−24.1 (−28.2 to −20.0)
       Unknown/missing/occult/NA125 (3.8)10 (0.5)−1.3 (−3.0 to 0.3)
      AJCC, American Joint Commission on Cancer; CI, confidence interval; LCS, lung cancer screening; NA, not applicable.

      Factors Associated With a Diagnosis of Early Stage Lung Cancer

      Among those with lung cancer, we used multivariable regression to identify factors associated with a diagnosis of early stage disease (stages 0, I, and II) independent of lung cancer screening. Compared with patients diagnosed without previous lung cancer screening, a diagnosis of lung cancer after any screening had an increased odds of early stage diagnosis in both unadjusted and multivariable models (adjusted OR [aOR] = 3.61, 95% CI: 2.81–4.65) (Fig. 3 and Supplementary Table 1). Additional factors associated with the increased likelihood of an early stage diagnosis included the following: older age (aOR = 1.36, 95% CI: 1.08–1.70, for ages 75–80 y compared with ages 55–64 y), female sex (aOR = 1.49, 95% CI: 1.27–1.75), former tobacco use (aOR = 1.29, 95% CI: 1.08–1.54), a diagnosis of COPD (aOR = 1.61, 95% CI: 1.37–1.90) or previous malignancy other than lung cancer (aOR = 1.82, 95% CI: 1.43–2.33), and an elevated BMI (aOR = 1.27, 95% CI: 1.04–1.55 for BMI > 30 kg/m2 compared with BMI < 25 kg/m2) (Table 3). Patients with small-cell lung cancer had a decreased odds of early stage disease compared with those with non-small cell lung cancer (aOR = 0.12, CI: 0.09–0.18). There was no association of socioeconomic status or race and ethnicity with early stage diagnosis.
      Figure thumbnail gr3
      Figure 3Factors associated with a diagnosis of early stage lung cancer. Early stage defined as stages 0, I, and II. BMI, body mass index; COPD, chronic obstructive pulmonary disease; Pac Isl, Pacific Islander.

      Discussion

      Although trials have revealed that lung cancer screening with LDCT reduces lung cancer-specific mortality, the effectiveness of LDCT in clinical practice has not been clearly revealed. In this multicenter cohort analysis, we evaluated the impact of screening on the distribution of cancer stage and lung cancer incidence and found several important findings during this relatively early phase of lung cancer screening implementation. First, among individuals who were diagnosed with lung cancer after being screened during this time period, most (64%) were identified at an early stage, and this was accompanied by a concomitant decrease in patients with metastatic lung cancer. Second, at the population level across the four health care systems evaluated in this analysis, the overall annual incidence of lung cancer was relatively stable; however, there were notable changes in the incidence of stage-specific disease as the rate of screening increased over time. The annual rate of stage I lung cancer increased by an average of 8.4% and was accompanied by an average decline of 6.6% in stage IV disease. By 2018, these changes in incidence resulted in a higher rate of stage I compared with stage IV cancers. This migration to early stage disease with no change in the overall incidence of lung cancer suggests that implementation of screening was achieving the desired effect of identifying early stage lung cancers that were destined to progress to more advanced stages of disease, and without resulting in a significant rate of overdiagnosis. Although overdiagnosis is difficult to explicitly define, our findings stand in contrast to the evidence of overdiagnosis from LDCT screening in a largely nonsmoking population of women in Taiwan noted by Gao et al.
      • Gao W.
      • Wen C.P.
      • Wu A.
      • Welch H.G.
      Association of computed tomographic screening promotion with lung cancer overdiagnosis among Asian women.
      who reported a sixfold increase in the incidence of early stage lung cancer without change in the incidence of late-stage lung cancer.
      Differences between efficacy and effectiveness with respect to the benefit of lung cancer screening may exist if implementation varies in important aspects from how it was administered in the randomized trials. The NLST and NELSON trials were conducted primarily in urban academic centers and had resources and processes in place to optimize study procedures, evaluation of positive findings, including timely evaluation, and minimizing harms from invasive evaluation. Whether the results observed in these trials will result in similar outcomes with screening in community-based settings as a part of standard clinical care remains unclear. For example, study procedures used in the NLST and NELSON trials resulted in very high adherence to annual screening, but observed rates in community settings have been considerably lower, including at the health systems included in this analysis.
      • Lin Y.
      • Fu M.
      • Ding R.
      • et al.
      Patient adherence to lung CT screening reporting & data system-recommended screening intervals in the United States: a systematic review and meta-analysis.
      ,
      • Lopez-Olivo M.A.
      • Maki K.G.
      • Choi N.J.
      • et al.
      Patient adherence to screening for lung cancer in the US.
      ,
      • Kim R.
      • Rendle K.A.
      • Neslund-Dudas C.
      • et al.
      Community-based lung cancer screening adherence to Lung-RADS recommendations.
      Despite lower adherence across the four systems, the distribution of cancer stage was nearly as favorable as the distribution observed in the NLST and NELSON trials. For example, when compared with NLST, we observed a slightly lower rate of stage I disease (55% versus 61%) diagnosed after screening, but this was considerably higher than the rate of stage I disease in the nonscreened population (28%). Our findings build on the results of previous analyses from community and academic settings which have revealed a favorable shift in stage distribution among screened populations, including a previous report from one health care system within our multicenter cohort.
      • Carroll N.M.
      • Burnett-Hartman A.N.
      • Joyce C.A.
      • et al.
      Real-world clinical implementation of lung cancer screening—evaluating processes to improve screening guidelines-concordance.
      In the first year of screening (2013–2014) in a Veterans Affairs medical system, Okereke et al.
      • Okereke I.C.
      • Bates M.F.
      • Jankowich M.D.
      • et al.
      Effects of implementation of lung cancer screening at one Veterans Affairs Medical Center.
      reported an early stage disease rate of 67% compared with 35% in the prescreening period. An eighth-site Veterans Affairs demonstration project and a multisite program within a large integrated health system both reported a 71% rate of early stage disease.
      • Handy J.R.J.
      • Skokan M.
      • Rauch E.
      • et al.
      Results of lung cancer screening in the community.
      ,
      • Kinsinger L.S.
      • Anderson C.
      • Kim J.
      • et al.
      Implementation of lung cancer screening in the Veterans Health Administration.
      In addition to LCS-LDCT, we identified factors that were associated with a diagnosis of early stage lung cancer. A personal history of extrathoracic cancer and a history of COPD were associated with a diagnosis of early stage lung cancer, likely reflecting higher utilization of thoracic imaging in these individuals. Patients with previous cancer generally undergo several years of surveillance which can identify lesions suggestive of lung cancer before clinical presentation. Patients with COPD also undergo more frequent chest imaging to establish the diagnosis for ongoing assessment and management of symptoms and exacerbations which can similarly identify suspicious lung lesions. Individuals with an increased BMI of more than 30 were also more likely to be diagnosed with early stage disease, which may also reflect greater health care utilization, including higher rates of imaging.
      Our study has limitations. First, our results are derived from a retrospective study of four diverse health systems in the United States, but they may not reflect the stage distribution and stage-specific rates observed in other settings or more broadly across the United States. Second, our study uses health system-level EHR data allowing accurate identification of screening events, but we could not differentiate between screen-detected and interval cancers. Third, although previous studies have revealed that those who elect to receive screening are healthier in ways that are difficult to measure,
      • Kramer B.S.
      • Croswell J.M.
      Cancer screening: the clash of science and intuition.
      given that a 30 pack-year smoking history is an LCS eligibility requirement, we believe that healthy-user bias is minimal in the PROSPR-Lung screened population. Last although cancer case ascertainment relied on local cancer registry information, individuals may have sought care outside of one of the four health systems included in our analysis. We believe that this was a relatively rare event given that two of the health care systems operate under an integrated care delivery model and the other two limited the cohort to individuals who received primary care within their systems and retain patients diagnosed with cancer given their role as comprehensive cancer centers. Furthermore, any out migration before lung cancer diagnosis would be anticipated to occur at similar rates for unscreened and screened individuals because there was no socioeconomic difference between the two groups.
      In conclusion, to the best of our knowledge, this is the first study to determine the impact of LCS on cancer stage migration using a population-based multicenter cohort. Our results suggest that LCS results in a shift to early stage disease coupled with a decline in the proportion diagnosed with metastatic lung cancer when compared with the unscreened population. The distribution of stage was similar to rates observed in previous clinical trials despite limitations, such as lower adherence to annual screening, which have been observed outside of trial settings. By the end of the study, approximately 20% of those diagnosed with lung cancer had received at least one previous LCS-LDCT. Although overdiagnosis remains a concern, at this rate of screening, we did not observe an increase in the overall rate of lung cancer. As screening implementation progresses, future population-based studies are needed to evaluate the impact of screening on other effectiveness outcomes, including rates of harms related to screening and the impact on lung cancer mortality.

      CRediT Authorship Contribution Statement

      Anil Vachani: Conceptualization, Methodology, Data curation, Formal analysis, Funding acquisition, Writing—original draft, Writing—review and editing.
      Nikki Carroll: Conceptualization, Methodology, Data curation, Formal analysis, Software, Writing—review and editing.
      Michael Simoff, Christine Neslund-Dudas, Stacey Honda, Robert T. Greenlee, Katharine A. Rendle, Andrea Burnett-Hartman: Conceptualization, Writing—review and editing.
      Debra P. Ritzwoller: Conceptualization, Methodology, Data curation, Formal analysis, Funding acquisition, Writing—review and editing.

      Acknowledgments

      This work was supported by the National Cancer Institute of the National Institutes of Health under award number UM1CA221939. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

      References

        • Siegel R.L.
        • Miller K.D.
        • Fuchs H.E.
        • Jemal A.
        Cancer statistics, 2021.
        CA Cancer J Clin. 2021; 71: 7-33
        • Lu T.
        • Yang X.
        • Huang Y.
        • et al.
        Trends in the incidence, treatment, and survival of patients with lung cancer in the last four decades.
        Cancer Manag Res. 2019; 11: 943-953
        • Chansky K.
        • Detterbeck F.C.
        • Nicholson A.G.
        • et al.
        The IASLC lung cancer staging project: external validation of the revision of the TNM stage groupings in the eighth edition of the TNM classification of lung cancer.
        J Thorac Oncol. 2017; 12: 1109-1121
        • Ganti A.K.
        • Klein A.B.
        • Cotarla I.
        • Seal B.
        • Chou E.
        Update of incidence, prevalence, survival, and initial treatment in patients with non–small cell lung cancer in the US.
        JAMA Oncol. 2021; 7: 1824
        • Aberle D.R.
        • Adams A.M.
        • et al.
        • National Lung Screening Trial Research Team
        Reduced lung-cancer mortality with low-dose computed tomographic screening.
        N Engl J Med. 2011; 365 (doi:10.1056/NEJMoa1102873): 395-409
        • de Koning H.J.
        • van der Aalst C.M.
        • de Jong P.A.
        • et al.
        Reduced lung-cancer mortality with volume CT screening in a randomized trial.
        N Engl J Med. 2020; 382: 503-513
        • Brodersen J.
        • Voss T.
        • Martiny F.
        • Siersma V.
        • Barratt A.
        • Heleno B.
        Overdiagnosis of lung cancer with low-dose computed tomography screening: meta-analysis of the randomised clinical trials.
        Breathe (Sheff). 2020; 16200013
        • Burnett-Hartman A.N.
        • Carroll N.M.
        • Honda S.A.
        • et al.
        Community-based lung cancer screening results in relation to patient and radiologist characteristics: the PROSPR consortium.
        Ann Am Thorac Soc. 2022; 19: 433-441
        • Iaccarino J.M.
        • Steiling K.A.
        • Wiener R.S.
        Lung cancer screening in a safety-net hospital: implications of screening a real-world population versus the National Lung Screening Trial.
        Ann Am Thorac Soc. 2018; 15: 1493-1495
        • Zhao H.
        • Xu Y.
        • Huo J.
        • Burks A.C.
        • Ost D.E.
        • Shih Y.C.T.
        Updated analysis of complication rates associated with invasive diagnostic procedures after lung cancer screening.
        JAMA Netw Open. 2020; 3e2029874
        • Huo J.
        • Xu Y.
        • Sheu T.
        • Volk R.J.
        • Shih Y.C.T.
        Complication rates and downstream medical costs associated with invasive diagnostic procedures for lung abnormalities in the community setting.
        JAMA Intern Med. 2019; 179: 324
        • Lin Y.
        • Fu M.
        • Ding R.
        • et al.
        Patient adherence to lung CT screening reporting & data system-recommended screening intervals in the United States: a systematic review and meta-analysis.
        J Thorac Oncol. 2022; 17: 38-55
        • Kunitomo Y.
        • Bade B.
        • Gunderson C.G.
        • et al.
        Racial differences in adherence to lung cancer screening follow-up: a systematic review and meta-analysis.
        Chest. 2022; 161: 266-275
        • Lopez-Olivo M.A.
        • Maki K.G.
        • Choi N.J.
        • et al.
        Patient adherence to screening for lung cancer in the US.
        JAMA Netw Open. 2020; 3e2025102
        • Handy J.R.J.
        • Skokan M.
        • Rauch E.
        • et al.
        Results of lung cancer screening in the community.
        Ann Fam Med. 2020; 18: 243-249
        • Copeland A.
        • Criswell A.
        • Ciupek A.
        • King J.C.
        Effectiveness of lung cancer screening implementation in the community setting in the United States.
        J Oncol Pract. 2019; 15: e607-e615
        • Kennedy M.P.T.
        • Cheyne L.
        • Darby M.
        • et al.
        Lung cancer stage-shift following a symptom awareness campaign.
        Thorax. 2018; 73: 1128-1136
        • Potter A.L.
        • Rosenstein A.L.
        • Kiang M.V.
        • et al.
        Association of computed tomography screening with lung cancer stage shift and survival in the United States: quasi-experimental study.
        BMJ. 2022; 376e069008
        • Yu M.
        • Tatalovich Z.
        • Gibson J.T.
        • Cronin K.A.
        Using a composite index of socioeconomic status to investigate health disparities while protecting the confidentiality of cancer registry data.
        Cancer Causes Control. 2014; 25: 81-92
        • Kim H.J.
        • Fay M.P.
        • Feuer E.J.
        • Midthune D.N.
        Permutation tests for joinpoint regression with applications to cancer rates.
        Stat Med. 2000; 19: 335-351
        • Hanley J.A.
        • Negassa A.
        • Edwardes MD. deB.
        • Forrester J.E.
        Statistical analysis of correlated data using generalized estimating equations: an orientation.
        Am J Epidemiol. 2003; 157: 364-375
        • Azur M.J.
        • Stuart E.A.
        • Frangakis C.
        • Leaf P.J.
        Multiple imputation by chained equations: what is it and how does it work?.
        Int J Methods Psychiatr Res. 2011; 20: 40-49
        • White I.R.
        • Royston P.
        • Wood A.M.
        Multiple imputation using chained equations: issues and guidance for practice.
        Stat Med. 2011; 30: 377-399
        • Gao W.
        • Wen C.P.
        • Wu A.
        • Welch H.G.
        Association of computed tomographic screening promotion with lung cancer overdiagnosis among Asian women.
        JAMA Intern Med. 2022; 182: 283
        • Kim R.
        • Rendle K.A.
        • Neslund-Dudas C.
        • et al.
        Community-based lung cancer screening adherence to Lung-RADS recommendations.
        JCO. 2021; 39 (10540–10540)
        • Carroll N.M.
        • Burnett-Hartman A.N.
        • Joyce C.A.
        • et al.
        Real-world clinical implementation of lung cancer screening—evaluating processes to improve screening guidelines-concordance.
        J Gen Intern Med. 2020; 35: 1143-1152
        • Okereke I.C.
        • Bates M.F.
        • Jankowich M.D.
        • et al.
        Effects of implementation of lung cancer screening at one Veterans Affairs Medical Center.
        Chest. 2016; 150: 1023-1029
        • Kinsinger L.S.
        • Anderson C.
        • Kim J.
        • et al.
        Implementation of lung cancer screening in the Veterans Health Administration.
        JAMA Intern Med. 2017; 177: 399
        • Kramer B.S.
        • Croswell J.M.
        Cancer screening: the clash of science and intuition.
        Annu Rev Med. 2009; 60: 125-137