Identifying the population to be targeted in a lung cancer screening programme in Denmark

Abstract

Introduction

We assessed the impact of recruitment criteria on lung cancer detection in a future Danish screening programme with low-dose CT.

Methods

We combined data from two Danish population-based health examination surveys with eligibility criteria from seven randomised controlled trials on lung cancer screening. Incident lung cancers were identified by linkage with the National Pathology Data Bank (Patobank). For an average of 4.4 years of follow-up, we calculated sensitivity, specificity, efficient frontier and number needed to screen (NNS) for lung cancer detection.

Results

When applying the different eligibility criteria to the 48 171 persons invited to the two surveys, the number of lung cancer cases identified in the target groups varied from 46 to 68. The National Lung Screening Trial (NLST) criteria had the highest sensitivity of 62.6% (95% CI 52.7 to 71.8) and the Dutch-Belgian NEderlands-Leuvens Screening ONderzoek (NELSON) criteria had the highest specificity 81.6% (95% CI 81.0 to 82.1). Sensitivity was higher for men than for women (NLST criteria 71.7% (95% CI 57.7 to 83.2) and 53.7% (95% CI 39.6 to 67.4), respectively). The NLST criteria identified the target population obtaining the lowest NNS with 46.3. The application of the NLST criteria showed that the higher the sensitivity, the lower the number of false-negative cases and, thus, the lower the NNS.

Conclusions

This study highlights the impact of the definition of the at-risk population on lung cancer screening efficacy. We found lower sensitivity among women regardless of screening criteria used. This should be carefully addressed in a possible screening programme.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• The definition of ‘risk population’ varies between the conducted randomised controlled trials on lung cancer screening with low-dose CT.

WHAT THIS STUDY ADDS

• When applying the definitions to follow-up of participants from Danish health examination surveys, we observed differences in the sensitivity and specificity of lung cancer detection, which were more pronounced in women.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Our results contribute to the optimisation of population-based lung cancer screening.

Introduction

Lung cancer is the most common cancer worldwide, with an estimated 2 480 308 new cases in 2022 and the most common cause of cancer-related death.¹ In 2022, 5710 new lung cancer cases were diagnosed in Denmark,² and in 2021, 3425 persons died from lung cancer.³ For those diagnosed with lung cancer in 2019–2021, the estimated 5-year relative survival was 25% in men and 31% in women, in contrast to 67% and 70%, respectively, for cancer patients in general.^{4 5} Most cases are non-small cell lung cancer, and curative treatment is offered only in case of local stage lung cancer.⁶

Early diagnosis of local stage lung cancer is challenging due to the asymptomatic presentation⁷; thus, most diagnoses occur at an advanced stage, leading to shorter survival.⁸ However, in the last 10 years, data from two large and several small randomised controlled trials (RCTs) on screening for lung cancer with low-dose CT, using inclusion criteria based on age and smoking demonstrated that detection of local stage lung cancer with CT screening reduces lung cancer mortality.

The American National Lung Screening Trial (NLST) and the Dutch-Belgian NEderlands-Leuvens Screening ONderzoek (NELSON) trial were the largest trials. NLST demonstrated a reduction in lung cancer mortality when comparing annual CT screening with annual chest radiology, rate ratio (RR) 0.80 (95% CI 0.73 to 0.93).^{9 10} NELSON demonstrated a reduction in lung cancer mortality for CT screening when comparing with no screening, RR 0.76 (95% CI 0.61 to 0.94) within men and 0.67 (95% CI 0.38 to 1.14) in women.¹¹ Results varied across the smaller trials but did overall not undermine the beneficial effect of CT screening on lung cancer mortality, see online supplemental table 1.

In December 2022, The European Council recommended to ‘explore the feasibility and effectiveness of low-dose CT to screen individuals at high risk for lung cancer, including heavy smokers and ex-smokers, and link screening with primary and secondary prevention approaches’.¹² As part of these explorative efforts, the present study aimed to assess the impact of the definition of the target group for a future lung cancer screening programme in Denmark. We focused on categorical inclusion criteria. As such criteria are used in the existing population-based lung cancer screening programmes in China, Croatia, Poland, South Korea, Taiwan and the USA,¹³ we found this also to be a relevant starting point to explore in Denmark. We assessed the predefined recruitment criteria used in the RCTs and measured the sensitivity and specificity of lung cancer detection in the total population, women and men. We used data from two population-based health surveys with follow-up for the incident and biopsy-verified lung cancer cases in the National Pathology Data Bank (Patobank).

Materials and methods

Study design

Cohort study.

Data sources

The Danish General Suburban Population Study (GESUS) is a health survey undertaken in 2010–2013 in the Næstved Municipality; 25% of persons aged 20–30 years and all persons aged 30+ years were invited.¹⁴ Participants provided information on risk factors, such as diseases, lifestyle and socioeconomic and family factors in self-administered questionnaires, underwent a simple health examination and provided biological samples, see online supplemental table 2.

The Lolland-Falster Health Study (LOFUS) is a health survey undertaken in 2016–2020 in the Lolland and Guldborgsund municipalities.¹⁵ A random sample of 18+ individuals and their entire household were invited. Participants provided information following procedures similar to those used in GESUS, see online supplemental table 2 .

Patobank is a Danish national database with data on all pathology specimens analysed in Denmark since 1 January 1990. Our data covered 1 January 1990 to 16 June 2022. Patobank uses a Danish version of the Systemized Nomenclature of Medicine (SNOMED) to record diagnoses.¹⁶

Definitions

Study population: all persons invited to either GESUS or LOFUS and aged 49–75 years at the time of invitation. The study population was split into the following subgroups:

Non-participants: persons invited to but not participating in GESUS or LOFUS.

Participants: persons invited to and participating in GESUS or LOFUS. This group was subdivided into:

Target group: participants meeting the specified inclusion criteria for lung cancer screening; see below.

Non-target group: participants not meeting the specified inclusion criteria for lung cancer screening.

Missing data group: participants without data on risk factors to define whether the person met the inclusion criteria for lung cancer screening or not, see figure 1.

Figure 1. Construction of study population using NELSON criteria, age group 50–74 years. Target group: person aged 50–74 years at first invitation to LOFUS or GESUS, participating and fulfilling the smoking criteria used in NELSON. Non-target group: person aged 50–74 years at first invitation to LOFUS or GESUS, participating but not fulfilling the smoking criteria used in NELSON. Missing data group: person aged 50–74 years at the first invitation to LOFUS or GESUS, participating without complete data for the smoking criteria used in NELSON. Non-participants: persons aged 50–74 years at the first invitation to LOFUS or GESUS but not participating. GESUS, Danish General Suburban Population Study; LOFUS, Lolland-Falster Health Study; NELSON, Dutch-Belgian NEderlands-Leuvens Screening ONderzoek.

The target group for lung cancer screening: the population invited to lung cancer screening according to inclusion criteria used in a lung cancer screening trial: NLST,^{9 10} NELSON,¹¹ Italian Lung Cancer Screening Trial (ITALUNG),¹⁷ German Lung Cancer Screening Intervention Trial (LUSI),¹⁸ Multicentric Italian Lung Detection Trial (MILD),¹⁹ Detection And screening of early lung cancer with Novel imaging TEchnology (DANTE)²⁰ and Danish Lung Cancer Screening Trial (DLCST),²¹ see online supplemental table 1 and figure 1. Trial exclusion criteria were not considered as these data were incomplete in GESUS and LOFUS.

Data variables

From GESUS and LOFUS, we retrieved the following variables: age, sex, smoking status (do you smoke?), smoking history (when did you start smoking?, when did you stop smoking?, how many cigarettes, pipes, cheroot or cigars do you smoke per day?) and per cent of the predicted value of the forced expiratory volume in the first second (FEV1% predicted). For GESUS current and former smokers and for LOFUS current smokers, the smoking history was calculated as pack-years: packs smoked per day multiplied by the number of years smoked. For LOFUS former smokers, only the number of years smoked was available, so we assumed one pack-year per year of smoking.²² For LOFUS participants, FEV1% predicted was available, while for GESUS, FEV1% predicted was estimated from FEV1 using the formula by Kuster et al.²³

Follow-up

We followed all persons up for incident cases of lung cancer in the Patobank thanks to the linkage of the Danish Central Person Register number (unique personal identification number given at birth or immigration to all Danish citizens) of the invited persons to the surveys with the lung cancer SNOMED codes, see online supplemental table 3. We used the first specimen registered after the invitation date to GESUS or LOFUS. For GESUS, the first invitation date was 11 January 2010, and for LOFUS, it was 29 January 2016. Persons diagnosed with lung cancer before the invitation date were excluded from the analysis. Patobank data were available until 31 August 2022, resulting in an average follow-up period of 4.4 years in LOFUS. To ensure equal weight of the GESUS and LOFUS data in the analysis, we restricted the average follow-up time of GESUS data to 4.4 years. As the study was run on the server in Region Zealand, we were not able to follow-up the population for deaths and emigrations.

Outcomes

Sensitivity: three calculations were undertaken: (persons with lung cancer in the target group) divided by (persons with lung cancer in all participants excluding the missing data group); or (persons with lung cancer in all participants including the missing data group); or (persons with lung cancer in the entire study population of participants and non-participants).

Specificity: three calculations were undertaken: (persons without lung cancer in the non-target group) divided by (persons without lung cancer in all participants excluding the missing data group); or (persons without lung cancer in all participants including the missing data group); or (persons without lung cancer in the entire study population of participants and non-participants).

Efficient frontier: The left and top-most points on the plot graph, where the X-axis represents 1-specificity and the Y-axis sensitivity, give us the efficient frontier, illustrating simultaneously the inclusion criteria with the highest sensitivity and the inclusion criteria with the highest specificity of a given set of inclusion criteria.^{24 25}

Number needed to screen (NNS) to detect one lung cancer case: (persons in target group) divided by (lung cancer cases in target group).

Statistical analysis

First, we merged GESUS and LOFUS data and excluded the LOFUS record from those participating in both surveys, see figure 1. Second, we calculated the distribution of persons and lung cancer cases in the study population subgroups based on the recruitment criteria from each lung cancer screening trial, see online supplemental figure 1. Third, we calculated the four study outcomes for each dataset, see online supplemental figure 2. In this study, we used the efficient frontier to illustrate sensitivity and specificity simultaneously, as you may want to prioritise differently between the two depending on health policy. The Youden index was not used since it does not allow the same prioritisation. Fourth, CIs were obtained by a procedure first given by Clopper and Pearson²⁶; this guarantees that the confidence level is at least 95%. Analyses were performed using R-V.4.2.2 and R Studio V.2023.03.0. Data handling was undertaken at a secure server in Region Zealand.

Patient and public involvement statement

Patients and public were not involved in the research process.

Results

In total, 48 171 persons aged 49–75 were invited to GESUS or LOFUS; evenly distributed between women and men, with numbers slightly decreasing by increasing age, see online supplemental table 4. Using the NELSON criteria including only the 44 973 persons aged 50–74 years, the participation rate was 47.3%, and among participants, 18.2% were current smokers, 40.0% former smokers, 37.7% never smokers, and data were missing for 4.1%. Among participants, 17.3% belonged to the target group; 75.3% to the non-target group and 7.4% to the missing data group, see table 1.

Table 1. Baseline characteristics of the study population aged 50–74 years, and participants divided into target/non-target/missing data groups according to NELSON criteria.

	Total		Women		Men
	Number	%	Number	%	Number	%
Total invited	44 973	100	22 696	100	22 277	100
Participants	21 287	47.3	11 355	50.0	9932	44.6
Non-participants	23 686	52.7	11 341	50.0	12 345	55.4
Women	22 696	50.5	22 696	100	0	0
Men	22 277	49.5	0	0	22 277	100
Age at invitation
50–54	9389	20.9	4708	20.7	4681	21.0
55–59	9432	21.0	4669	20.6	4763	21.4
60–64	9633	21.4	4878	21.5	4755	21.3
65–69	9037	20.1	4588	20.2	4449	20.0
70–74	7482	16.6	3853	17.0	3629	16.3
Participants only
Smoking status
Current	3882	18.2	1980	17.4	1902	19.1
Former	8509	40.0	4184	36.9	4325	43.6
Never	8024	37.7	4713	41.5	3311	33.3
Missing	872	4.1	478	4.2	394	4.0
Smoking history
For current smokers
Target	1898	48.9	896	45.3	1002	52.7
Non-target	1864	48.0	1000	50.5	864	45.4
Missing	120	3.1	84	4.2	36	2.0
For former smokers
Target	1779	20.9	862	20.6	917	21.2
Non-target	6145	72.2	3013	72.0	3132	72.4
Missing	585	6.9	309	7.4	276	6.4
Smoking status and history
Total participants
Target	3677	17.3	1758	15.5	1919	19.3
Non-target	16 033	75.3	8726	76.8	7307	73.6
Missing	1577	7.4	871	7.7	706	7.1

NELSONDutch-Belgian NEderlands-Leuvens Screening ONderzoek

In the NELSON age group 50–74 years, 346 lung cancer cases were diagnosed during follow-up; 19.6% in the target group; 13.9% in the non-target group; 2.9% in the missing data group and 63.6% in the non-participation group, see online supplemental table 5. Almost the same proportions belonged to these groups using the other trial definitions; apart from DANTE, which included only men aged 60–74 and identified only 12.8% of the lung cancer cases in the target group. For all sets of inclusion criteria, the proportions of lung cancer cases identified in the target group were somewhat lower for women; 17.2%–19.0%, than for men; 20.6%–25.3%.

For participants without missing data, the sensitivity varied from 56.3% to 62.6% across trial definitions, with DANTE as an outlier at 36.2%, see table 2. The specificity varied from 78.0% to 81.6% again, with DANTE as an outlier at 91.5%. For the total population of participants without missing data, the efficient frontier included NLST with a sensitivity of 62.6% (95% CI 52.7 to 71.8) and a specificity of 80.9% (95% CI 80.3 to 81.5) and NELSON with a sensitivity of 58.6% (95% CI 49.1 to 67.7) and a specificity of 81.6% (95% CI 81.0 to 82.1), see figure 2A.

Table 2. Persons in the target group as per cent of participants excluding missing data, and the total invited population; and sensitivity, specificity and number needed to screen to detect one lung cancer case during an average follow-up of 4.4 years.

	Participants excluding the missing data group				Study population
	Per centperson*	Sensitivity(95% CI)	Specificity(95% CI)	NNS†	Per centPerson*	Sensitivity(95% CI)	Specificity(95% CI)
Total
NELSON	18.7%	58.6 (49.1; 67.7)	81.6 (81.0; 82.1)	54.1	8.2%	19.7 (15.6; 24.2)	91.9 (91.7; 92.2)
NLST	19.4%	62.6 (52.7; 71.8)	80.9 (80.3; 81.5)	46.3	8.7%	21.3 (16.9;26.3)	91.4 (91.1; 91.7)
ITALUNG	22.2%	59.0 (47.3; 70.0)	78.0 (77.3; 78.7)	61.8	10.1%	21.1 (15.9; 27.1)	90.0 (89.6; 90.3)
LUSI	19.9%	56.3 (45.3; 66.9)	80.3 (79.7; 80.9)	66.9	8.7%	19.6 (14.9; 25.1)	91.3 (91.0; 91.6)
MILD	20.8%	61.2 (51.9; 69.9)	79.4 (78.8; 80.0)	58.9	9.1%	20.2 (16.2; 24.6)	91.0 (90.7; 91.3)
DANTE	8.7%	36.2 (22.7; 51.5)	91.5 (90.8; 92.2)	29.6	3.9%	12.8 (7.6; 19.7)	96.2 (95.8; 96.5)
DLCST	21.6%	59.8 (48.7; 70.1)	78.6 (77.9; 79.2)	66.5	8.8%	19.1 (14.6; 24.3)	91.2 (90.9; 91.5)
WOMEN
NELSON	16.8%	51.6 (38.6; 64.5)	83.4 (82.7; 84.1)	54.9	7.7%	17.2 (12.2; 23.4)	92.3 (92.0; 92.7)
NLST	16.4%	53.7 (39.6; 67.4)	83.8 (83.0; 84.6)	47.8	7.7%	17.7 (12.2; 24.4)	92.4 (92.0; 92.8)
ITALUNG	20.0%	54.8 (38.7; 70.2)	80.2 (79.2; 81.1)	59.1	9.6%	19.0 (12.4; 27.1)	90.5 (90.0; 90.9)
LUSI	17.8%	54.0 (39.3; 68.2)	82.4 (81.6; 83.2)	58.3	8.4%	18.9 (12.8; 26.3)	91.7 (91.3; 92.1)
MILD	19.3%	54.7 (41.7; 67.2)	80.9 (80.1; 81.6)	61.5	8.9%	17.8 (12.7; 23.8)	91.2 (90.8; 91.6)
DANTE	NA	NA	NA	NA	NA	NA	NA
DLCST	19.9%	56.2 (41.2; 70.5)	80.3 (79.4; 81.1)	64.0	8.8%	17.3 (11.7; 24.2)	91.3 (90.9; 91.7)
MEN
NELSON	20.8%	66.7 (52.5; 78.9)	79.5 (78.6; 80,3)	53.3	8.6%	22.5 (16.3; 29.8)	91.5 (91.1; 91.9)
NLST	22.6%	71.7 (57.7; 83.2)	77.7 (76.8; 78.7)	45.1	9.7%	25.3 (18.6; 33.1)	90.4 (90.0; 90.8)
ITALUNG	24.7%	63.9 (46.2; 79.2)	75.5 (74.4; 76.6)	64.5	10.6%	23.7 (15.7; 33.4)	89.5 (89.0; 90.0)
LUSI	22.3%	59.5 (42.1; 75.2)	77.9 (77.0; 78.8)	77.4	9.1%	20.6 (13.4; 29.5)	90.9 (90.5; 91.3)
MILD	22.6%	68.4 (54.8; 80.1)	77.7 (76.9; 78.5)	56.6	9.3%	22.9 (16.9; 30.0)	90.8 (90.4; 91.2)
DANTE	8.7%	36.2 (22.7; 51.5)	91.5 (90.8; 92.2)	29.6	3.9%	12.8 (7.6; 19.7)	96.2 (95.8; 96.5)
DLCST	23.6%	64.1 (47.2; 78.8)	76.6 (75.6; 77.6)	69.2	8.9%	21.6 (14.5; 30.1)	91.2 (90.7; 91.6)

^*Per cent person: the number of persons in the target group divided by the number of persons in the respective group indicated in each main column.

^†NNS: number needed to screen to detect one lung cancer case in the target group.

DANTEDetection And screening of early lung cancer with Novel imaging TEchnologyDLCSTDanish Lung Cancer Screening TrialITALUNGItalian Lung Cancer Screening TrialLUSIGerman Lung Cancer Screening Intervention TrialMILDMulticentric Italian Lung Detection TrialNA, not availableNELSONDutch-Belgian NEderlands-Leuvens Screening ONderzoekNLSTNational Lung Screening Trial

Figure 2. Efficient frontiers for detection of incident lung cancer based on follow-up of two Danish health examination surveys combined with recruitment criteria from six randomised controlled trials on lung cancer screening. (A) Including only participants in the relevant age group with full smoking data. (B) Including all persons in the relevant age group invited to the health examination surveys. DLCST, Danish Lung Cancer Screening Trial; ITALUNG, Italian Lung Cancer Screening Trial; LUSI, German Lung Cancer Screening Intervention Trial; MILD, Multicentric Italian Lung Detection Trial; NELSON, Dutch-Belgian NEderlands-Leuvens Screening ONderzoek; NLST, National Lung Screening Trial.

The sensitivity of screening for all trial recruitment criteria was lower for women than for men, while the specificity tended to be higher. For women, DLCST (sensitivity 56.2% (95% CI 41.2 to 70.5) and specificity 80.3% (95% CI 79.4 to 81.1)) and NLST (sensitivity 53.7% (95% CI 39.6 to 67.4) and specificity 83.8% (95% CI 83.0 to 84.6)) came out on the efficient frontier. For men, there was a wider distribution in the sensitivity outcomes with NLST (sensitivity 71.1% (95% CI 57.7 to 83.2) and specificity 77.7% (95% CI 76.8 to 78.7)) and NELSON (sensitivity 66.7% (95% CI 52.5 to 78.9) and specificity 79.5% (95% CI 78.6 to 80.3)) on the efficient frontier, see table 2 and figure 2A.

For all participants, including those with missing data, the sensitivity varied from 52.0% to 57.8%, and the specificity from 79.5% to 82.9%, again with lower sensitivity for women than for men, see online supplemental table 6. For the study population including participants and non-participants, sensitivity varied from 19.1% to 21.3%; specificity from 90.0% to 91.9%, see table 2. NLST (sensitivity 21.3 (95% CI 16.9 to 26.3) and specificity 91.4 (95% CI 91.1 to 91.7)) and NELSON (sensitivity 19.7% (95% CI 15.6 to 24.2) and specificity 91.9 (95% CI 91.7 to 92.2)) constituted the efficient frontier, see figure 2B.

Within the target population of participants without missing data, the NNS for the detection of one lung cancer case varied from 46.3 to 66.9, with DANTE as an outlier with 29.6; see table 2. The lowest NNS was found for NLST with 46.3 and NELSON with 54.1, see online supplemental figure 3. The NNS might vary depending on age range at recruitment, being lower in older groups.

Discussion

Main results

Using data from two Danish health examination surveys combined with recruitment criteria from lung cancer screening trials, we found that around 20% of study participants with full smoking data fulfilled criteria for lung cancer screening with low-dose CT (trial target populations). The optimal screening sensitivity and specificity combination was found for the NLST criteria (highest sensitivity, 62.6%) and the NELSON criteria (highest specificity, 81.6%), with considerable overlap in CIs. For both sets of trial recruitment criteria, about 6 in 10 lung cancer cases would be potentially detectable, and about 8 in 10 of those without lung cancer would not be invited for screening. The DANTE trial was an outlier with a lower sensitivity of 36.2% and higher specificity of 91.5%.

We found differences between women and men. For women, the highest sensitivity of 56.2% was seen for the DLCST, where the specificity was 80.3%. The highest specificity for women, 83.8%, was seen for NLST, where the sensitivity was 53.7%. For men, the sensitivity was in all six trials at the minimum 59.5%, going up to 71.7% for NLST, but the specificity was also somewhat lower for men than for women. Regarding these differences, we performed a sensitivity analysis. For all sets of inclusion criteria, the difference in sensitivity between men and women was not statistically significant, with the minor p value=0.08 when applying NLST criteria. The difference in specificity between men and women was statistically significant for all sets of inclusion criteria. This still may need to be addressed in clinical practice since our number of lung cancers was not very large, and the number of persons without lung cancer was quite large.

About 3%–4% of survey participants had missing smoking data. For all participants, the sensitivity therefore decreased slightly, and the specificity increased slightly, for instance, for NLST to 57.8% and 82.4%, respectively. More importantly, about half of the persons invited to the surveys did not participate. For all invited persons, the sensitivity decreased, and the specificity increased, for instance, for NLST to 21.3% and 91.4%, respectively. This means that if invitation to lung cancer screening had the same participation rates as seen in the health surveys, a screening programme following these criteria would potentially detect only 2 in 10 lung cancer cases, while 9 in 10 of those free from lung cancer would not be invited.

Other studies

Using smoking prevalence data from the Eurobarometer and relative risks of lung cancer by smoking history from a pooled analysis, Brenner and Krilaviciute²⁷ estimated the sensitivity and specificity of the NLST recruitment criteria for Denmark. They found, on average, for men and women together, sensitivity 53% and specificity 83%. The Eurobarometer data derived from the first 1000 personal interviews conducted in 2017 following a geographical selection procedure and weighted for missing data with the national population distribution.²⁸ Nevertheless, these crude data were relatively well per our findings on sensitivity and specificity. Across Europe, Brenner and Krilaviciute found that the sensitivity of the trial recruitment criteria was generally higher, and the specificity was generally lower in countries with a higher than in those with a lower smoking prevalence, and among men than among women.

Pinsky and Berg²⁹ estimated the proportion of US lung cancer cases that would be detected by screening according to the NSLT criteria. They used Surveillance, Epidemiology and End Results data for number of lung cancer cases; 2010 National Health Interview Survey for smoking data and relative risk (RR) of lung cancer by smoking from ‘statistical models’. They found that only 26.7% of lung cancer cases met the NLST criteria for screening.

Walter et al³⁰ identified 9481 patients with lung cancer registered in the German Center for Lung Research data warehouse. They analysed the 3588 patients with data on pack-years and age and studied sensitivity of eligibility criteria from NSLT, DLCST (Danish), US Preventive Service Task Force (USPSTF) 2013 recommendations, USPSTF 2021 recommendations; and two versions of the Prostate, Lung, Colorectal and Ovarian (PLCO) model (thresholds 1.7% in 6 years, and 1.0% in 6 years). Sensitivities were: NLST 48.5%; DLCST 48.7%, USPSTF 2013 57.0%, USPSTF 2021 70.0%; PLCO 1.7%/6 years 57.7%, PLCO 1.0%/6 years 72.4%. Sensitivity was systematically lower for women than for men. Using the NSLT and DLCST criteria, we found higher sensitivities of 62.6% and 59.8%, respectively.

Among lung cancer cases aged 55–74 diagnosed 2012–2015 at the Seoul National University Hospital, 29.6% fulfilled the NLST recruitment criteria.³¹ This sensitivity of 29.6% in the total population was higher than the 21.3% we found in our data.

The differences between these studies’ results and ours may be due to differences in the study design, population selection, patient characteristics or assumptions used to handle missing data.

Risk models can be an attractive option for identifying high-risk profiles. Bhardwaj et al³² used data from a German cohort; using the same cut-off as based on the trial criteria for number of persons screened, the authors calculated the expected number of lung cancer cases detected by three models. Using the NLST criteria, the sensitivity was 59.8%, while the three models predicted sensitivities of 71.6%–72.5%. Using the NELSON criteria, the sensitivity was 63.7%, while the models predicted sensitivities of 78.4–81.4%.

Risk models discriminate better than categorical criteria. However, they have certain limitations.³³ Therefore, they should be assessed when exploring optimal screening scenarios.

Participation

Optimal coverage of the target population is essential for successful screening results in trials and real-world programmes. Rankin et al,³⁴ reviewed recruitment strategies in trials and real-world programmes and found that 25%–52% of invited persons participated in trials, and 17%–40% in real-world programmes.

In our data, 47.3% of the persons aged 50–74 years and invited to either of the two health surveys participated. Furthermore, the participants represented a biased sample, as 0.6% of the participants (= ((68+48+10)/(3677+16 033+1577)) were diagnosed with lung cancer, while the percentage was 0.9% (= 220/23,686) in non-participants, see online supplemental table 5. The combined effect of low coverage and selection bias resulted for persons aged 50–74 years in a decrease in sensitivity from 58.6% for participants with full smoking data to 19.7% for all invited persons, see table 2.

Experiences from other countries have highlighted the importance of including high-risk areas in recruitment for lung cancer screening. For instance, an Australian study highlighted the increased prevalence of tobacco smoking in remote areas as compared with major cities.³⁵ In an evaluation of lung cancer screening from the USA, Lui et al noted that one of the limitations of their study was that the southeastern states of the USA with the highest lung cancer death rate were not included.³⁶ Using the US National Health Interview Survey, Jemal et al found more than 50% of current and former smokers who met the USPSTF selection criteria were uninsured, showing the importance of economic status and access to healthcare.³⁷ In the UK, a national targeted lung cancer screening programme, the Targeted Lung Health Check programme, has prioritised rolling out in areas of highest deprivation first.³⁸

Strengths and limitations

A strength of this study was that we used data from two population-based health surveys to evaluate the potential impact of lung cancer screening in Denmark using the recruitment criteria from the respective RCTs. It was also a strength that we could identify all incident lung cancer cases by the date of diagnosis. It was a limitation that the health surveys did not have national coverage. For LOFUS participants, the smoking history for former smokers was not available, so we assumed one pack per day. The average number of cigarettes smoked per day by former smokers in GESUS is lower than one pack per day. We performed sensitivity analyses regarding this, finding similar results when applying the two different smoking intensities. It was furthermore a limitation that only 47% of the persons invited to the health surveys participated and that 7% of the participants had missing data on smoking history. The different follow-up periods between both surveys were another limitation. As the linkage between LOFUS/GESUS and Patobank data was based on the unique personal identification numbers, we had complete ascertainment of histologically confirmed lung cancer cases. However, in the follow-up, we did not have access to data from the population register, which means that we could not to censor persons at death or emigration. We, therefore, calculated the lung cancer risks as the number of identified lung cancer cases divided by the number of persons under observation. Based on another LOFUS study with censoring at death and emigration, a 4% difference was found between the number of person-years and the number of persons times the length of the follow-up period.³⁹ This means that in the present study, we had the correct number of lung cancer cases, but we slightly underestimated the lung cancer risks.

Official lung cancer incidence data in Denmark are published with some time delay, typically 2 years. To ensure the inclusion of the latest diagnosed lung cancer cases in our cohort, we retrieved data on incident lung cancer cases from the daily updated Patobank. However, it should be noted that not all lung cancer diagnoses are histologically confirmed. A validation study of lung cancers diagnosed in 2014–2017 in patients below 80 years showed that 15% of the cases missed histological verification.⁴⁰

Clinical implications

Specificity is emphasised in screening programmes in Denmark because personal invitations are sent to all individuals in the target group. This means that among the two programmes on the efficient frontline, recruitment according to the NELSON criteria would be prioritised. Data from both this and other studies indicated differences in sensitivity and specificity between women and men, and this point needs to be carefully addressed before decisions are made on screening implementation. Screening criteria need to consider gender disparity to ensure equity in lung cancer screening outcomes.

Participation is relatively high in the organised cancer screening programmes in Denmark. Coverage in cervical screening is 74%,⁴¹ in breast screening 79%,⁴² and in colorectal screening 60%.⁴³ In the Danish trial on lung cancer screening, DLCST, participants volunteered,⁴⁴ and it is difficult to convert this finding to a population-based programme with personal invitations. Nevertheless, participation will be decisive for a screening programme’s outcome. In our study, participants with full smoking data constituted only 43.9% of people aged 50–74 years and invited to the health surveys. This means that with the same attendance in a national screening programme, less than half of the effect on lung cancer mortality observed in the trials would be seen.

Conclusion

We combined data from two Danish population-based health examination surveys with recruitment criteria from seven RCTs on lung cancer screening and 4.4 years of follow-up for incident lung cancer cases. Our results indicated that the outcome of a lung cancer screening programme in Denmark will, to some extent, depend on whether emphasis in the recruitment criteria is given to high sensitivity or to high specificity; that sensitivity of screening will be lower for women than for men and that the participation rate will have a decisive impact on the potential outcome of a national screening programme.

Data availability statement

Data may be obtained from a third party and are not publicly available.