The UK Lung Cancer Screening Trial: a pilot randomised controlled trial of low-dose computed tomography screening for the early detection of lung cancer

Chest radiography and sputum cytology lung cancer screening

The earliest lung screening trial was undertaken in London in the 1960s: > 55,000 individuals were randomised to either chest radiography every 6 months for 3 years, or chest radiography at the beginning and end of the 3-year period. 26 No mortality difference was found between the two groups. Three major trials in the USA, and one in Czechoslovakia, were developed in the 1970s; these trialled chest radiography with or without sputum cytology, but none showed any reduction in lung cancer mortality. A more recent trial is the lung component of the NCI Prostate, Lung, Colorectal and Ovarian Screening Trial (PLCO) of 150,000 individuals. In this trial, those in the intervention arm were offered chest radiography at baseline and then annually for 3 subsequent years (smokers) or 2 subsequent years (never-smokers); those in the control arm were given ‘usual care’ only. No mortality benefit was observed for the chest X-ray arm. 27,28

Low-dose computed tomography lung cancer screening

Low-dose computed tomography, which was introduced in the late 1990s, offers a major advance in imaging technology. 29 LDCT is more sensitive than chest radiography, and has enabled detection of lung tumours that are < 1 cm. Early studies using this technology include the USA-based Early Lung Cancer Action Project (ELCAP)30 in 1000 high-risk smokers; the Mayo Clinic project in 1520 individuals, including annual sputum cytology,31 the Milan study32 and a 3-year mass screening programme using a mobile computed tomography (CT) unit in Japan. 33 The ELCAP (observational) study was later expanded to an international collaboration including 30,000 subjects. The European Union–US Spiral CT Collaboration was initiated in 2001 in Liverpool. Subsequent meetings throughout Europe resulted in the development of collaborative protocols to provide a mechanism for different trial groups to work together, and for outcome data to be comparable between studies. The ultimate aim – pooling of results – was formalised in The Liverpool Statement 2005. 34

Rationale for UK Lung Cancer Screening Trial pilot trial

By combining results of the UKLS pilot with those from NELSON37 and other European studies, it will be possible to model any likely mortality benefit of LDCT screening and treatment of early lesions. The pilot trial will also enable assessment of screening uptake, and factors that are unique to the UK population and the NHS. This will include testing the intervention against the criteria outlined by the UK National Screening Committee, especially some of those criteria concerning cost-effectiveness. The potential cost-effectiveness of UKLS has been maximised by the trial design (i.e. selection of a population at sufficiently high risk to yield a substantial number of tumours in return for the screening activity). 14 However, as the pilot trial is insufficiently powered to demonstrate a reduction in mortality, and a decision has been made not to fund the full UKLS, cost-effectiveness will have to be modelled.

Overall aim of the UK Lung Cancer Screening trial

The overall aim of the UKLS pilot is to contribute to the data required for an informed decision to be made regarding the introduction of population screening for lung cancer. This involves determining the best recruitment and screening strategies, introducing a screening protocol for the management of CT-detected nodules, establishing that the individuals selected for screening are at a sufficiently high risk for the benefits of the screening to outweigh the harms, and assessing the physical and psychological consequences and the health-economic implications of screening. Further objectives are to create a resource for future improvements to screening strategies and provide data for the European Union lung cancer pooling to ascertain European mortality.

Ethical approval and research governance

The study received approval from the National Information Governance Board. Ethical approval for the study was given by Liverpool Central Research Ethics Committee in December 2010 (reference number 10/H1005/74). Site-specific assessments were obtained from the PCTs covering Liverpool, Sefton and Knowsley in the north, and Bedfordshire, Cambridgeshire and Peterborough in the south. The trial was registered with the International Standard Randomised Controlled Trial Register under the reference number 78513845.

A summary of the changes made to the original protocol is given in Appendix 1. Additionally, as fewer people than anticipated were at high risk of lung cancer (and therefore eligible for the trial), two alterations were made to the original plan. First, even though the response rate to the initial UKLS questionnaire was at the expected level, the anticipated proportion of responders with a risk of ≥ 5% over a 5-year period was far lower than expected. This observation may indicate that lower-risk people are more health conscious or simply more willing to participate in trials. Thus, in order to achieve the target of 4000 high-risk individuals, the number of population approaches was increased from 82,000 to 250,000 (this is expanded further below; see Sample size). Second, the LLP risk assessment model (used to determine eligibility for the trial) was adapted to include in the risk calculation a history of respiratory diseases (COPD, tuberculosis, bronchitis and emphysema) as well as pneumonia. Additionally, for the purposes of calculating lung cancer risk, a history of pipe or cigar smoking was classified in the same way as cigarette smoking. The refined model is known as the Liverpool Lung Project lung cancer risk prediction algorithm, version 2 (LLP_v2; see Appendix 2).

Eligibility criteria

The eligibility criteria for trial participants are shown below.

Setting and population approaches

As UKLS is a population study, its starting point was NHS PCT records covering regions in the vicinity of two thoracic hospitals specialising in lung cancer imaging, pathology and surgery: the LHCH and Papworth Hospital. Records relating to 249,988 individuals, aged 50–75 years, residing in specific PCT areas (Liverpool, Knowsley, Sefton, Cambridgeshire, Peterborough and Bedfordshire) were provided by the PCTs to a third-party data management company (DMC: Radar). Following exclusions by the PCTs, 247,354 of these individuals were approached by post by Radar. Approaches took place over two time periods: August 2011 to March 2012, and May to August 2012.

The initial contact from the DMC took the form of an invitation letter (on the respective PCT-headed notepaper), a UKLS participant information sheet and questionnaire 1 (see Appendix 3). This questionnaire covered smoking history and duration, personal history of non-malignant lung diseases (e.g. pneumonia, COPD, emphysema, bronchitis and tuberculosis) and previous malignancy, exposure to asbestos, and family history of lung and other cancers; in addition, it enquired whether or not the individual would be interested in participating in a screening study. For those who were unwilling to complete the entire questionnaire, and who were not interested in participating further, there was a shorter non-participation questionnaire (covering smoking status, lung cancer prior experience and concern, and educational level), which they were asked to return instead (see Appendix 3).

Response categorisation

Approached subjects were categorised as follows, based on their response to the first invitation: positive responders, individuals who returned questionnaire 1 and agreed to participate in UKLS; negative responders, individuals who declined to participate in UKLS but supplied some basic information by completing the shorter, non-participation questionnaire; and non-responders, individuals who did not respond to the first invitation.

Risk assessment/eligibility for trial

Completed questionnaire 1s (from positive responders) were returned to the DMC, scanned, and the data analysed automatically in order to identify individuals who were at high risk of developing lung cancer over the next 5 years (LLP risk score of ≥ 5). 20 A modified version of the LLP risk algorithm (LLP_v2) was utilised for risk calculations (see Appendix 2). This incorporated additional respiratory parameters (i.e. COPD, emphysema, bronchitis and tuberculosis), as well as pneumonia, and also included both pipe and cigar usage within the smoking criteria. (Risk assessment was not carried out for negative responders, as this was not possible from the limited information that they supplied.) High-risk positive responders were contacted with a further questionnaire (questionnaire 2; see Appendix 3) to establish eligibility for the RCT (i.e. no exclusion criteria met)53 and sent a detailed patient information booklet: these people were also asked to consent to release their personal information to the UKLS research team. Non-responders to this second invitation were sent a reminder letter (see Appendix 3).

UK Lung Cancer Screening research clinics

Individuals responding to, and eligible on the basis of, the second questionnaire were invited to one of the recruitment centres (LHCH or Papworth Hospital). They were shown a digital versatile disc outlining the UKLS study57 and provided with an opportunity to ask any questions in a discussion group (n = 6–8) with the recruitment clinic administrator or on a one-to-one basis with a research nurse. Following informed consent, subjects underwent spirometry (forced expiratory volume in 1 second/forced vital capacity ratio), and provided blood, buccal swab, nasal brushings and sputum specimens. Recruits also completed a touchscreen questionnaire; this consisted of follow-up epidemiological and clinical questions, and psychosocial and quality-of-life questions, including the Hospital Anxiety and Depression Scale (HADS)58 and the Cancer Worry Scale-Revised (CWS-R)59,60 (adapted for lung cancer). All smokers (both CT screened and non-screened) were offered smoking cessation advice sheets and a list of local NHS stop smoking services.

Sequence generation, allocation concealment mechanism and implementation

The randomisation method was developed by the UKLS database designers (Artex, The Netherlands). Upon upload of subjects to the database, each subject was given a randomly generated unique code, consisting of eight characters (0–9; A–Z). All subjects who had given fully informed signed consent were available for randomisation; a minimum of two subjects from the same site (Liverpool or Papworth) were required in order for randomisation to be implemented. Each time randomisation took place, the computer generated a random shift number so that the order of characters in each participant’s unique code was shifted by this number. The rearranged codes were then ordered alphanumerically, and split on a 1 : 1 basis into A (intervention) or B (control) groups. The outcome was thus dependent both on the unique codes of the participants available for randomisation and the random shift number. Randomisation was carried out once per week by the trial manager or data manager, using a button function within the database. As the process was fully automated, the operator was unable to influence the process. Once participants had been randomised, it was not possible for their allocation to change. A more detailed example of the randomisation sequence is given in Appendix 4.

Follow-up

Any lesions identified on LDCT screening were treated as per the planned UKLS care pathway in the study protocol53 [e.g. follow-up scan or referral to multidisciplinary team (MDT)] (see Figure 12). Health and mortality outcomes of UKLS participants in both study arms will be followed for 10 years, via the Office for National Statistics (ONS), the Hospital Episode Statistics database and the National Cancer Registration Service. The full protocol for the UKLS study is available online. 61

Initial contact with general practitioners

The UKLS Project Management team wrote to the lead GP and the practice manager in the GP practice to inform them that individuals in their practice were being approached regarding the UKLS. They were provided with a copy of the 14-page UKLS patient information booklet and a poster to place in their surgery to inform patients that the UKLS was recruiting in their area. This enabled the GPs to respond to questions that their patients may have concerning the trial.

Spirometry result

Once a participant had been consented at the pilot site recruitment centre and the lung function (forced expiratory volume in 1 second/forced vital capacity) had been measured as part of the recruitment process, a letter was sent to the GP, with a copy of the lung function report, to enable them to act on lung function readings that indicated COPD.

Computed tomography scan results letter

Copies of letters sent to the UKLS recruits regarding CT scan results were also sent to their GPs, with details on the follow-up procedures.

Referral to multidisciplinary team

Details of participants referred to the MDTs with category 4 nodules, or participants reported to have a significant other finding (SOF), were reported to the GPs.

General practitioner questionnaire on significant other findings

A follow-up letter with a short questionnaire was sent to all the GPs at the end of the pilot, requesting the outcome of the SOFs.

General practitioner queries

The UKLS developed a mechanism for dealing with concerns and feedback from primary care. Like all studies of this kind, patients will often discuss their participation with their GPs; furthermore, GPs were informed about the involvement of their patients in the trial. As a consequence, although the trial was conducted largely independently of primary care (in terms of recruitment of patients and follow-up of patients with ‘positive’ screening results), GPs were made aware of the study and were able to contact the study team with any concerns.

We received a small number of written communications from GPs. They were read first by the ‘core’ study team (led by John Field) and a draft response was formulated. Then the UKLS primary care lead (David Weller) was sent the draft response for further input. There were no complaints of adverse clinical outcomes from the screening process within the correspondence received. On the whole, complaints related to (1) misunderstandings about who was responsible for follow-up investigations (although the study materials indicated that all such follow-up would be co-ordinated by the study team, in a couple of instances the GP thought he/she was responsible for ordering the investigations) and (2) concerns over patient anxiety that was generated through involvement in the study (again, through a misunderstanding of written materials and communications).

In all cases, the issues appeared to be addressed and resolved through our letters back to GPs – we did not hear further from the GPs concerned – and there are no ongoing unresolved issues.

Primary outcomes

1. Population-based recruitment based on risk stratification Compliance rate percentage (response to initial questionnaire); percentage with LLP_v2 risk ≥ 5%; percentage response to questionnaire 2 ‘wishing to participate’.

2. Management of trial through web-based database The UKLS recruitment, CT images and clinical pathway were successfully managed through the bespoke web-based UKLS database.

3. Optimal characteristics of CT scan readers (radiologists vs. radiographers) Radiologists and radiographers were assessed at each pilot site and also at the second read site.

4. Characterisation of CT-detected nodules utilising volumetric analysis The proportions of the CT-screened individuals receiving a repeat CT scan (at 3, 6, 9 or 12 months) are reported.

5. Prevalence of lung cancer at baseline The baseline for the UKLS was found to be 1.7%. In addition, details are provided for those individuals who were diagnosed at the 3-month repeat scan).

Secondary outcomes

1. Sociodemographic factors affecting participation.

2. Psychosocial measures (cancer distress, anxiety, depression and decision satisfaction): likely psychosocial impact of participating in a lung cancer screening trial (HADS; CWS-R; decision satisfaction); preliminary data on reasons for non-participation/barriers to participation.

3. Cost-effectiveness modelling: modelling of health economics data regarding the likely consequences of undertaking a larger trial or population screening (using European Quality of Life-5 Dimensions questionnaire).

Tertiary outcome

Future pooling of pertinent data with similarly designed European lung cancer screening trials, such as NELSON. 37 Data to be pooled are those relating to lung cancer detection, incidence and mortality. (The pooling of the data, and subsequent interpretation, lies outside the remit of this report.)

Socioeconomic data analysis

Index of Multiple Deprivation (IMD) rank62 was obtained from postcode data in an anonymised form for all 249,988 individuals whose details were provided by the PCTs (including non-responders). UK IMD data were analysed and reported as ranks within quintiles based on England-wide population data: quintile 1 (Q1; most deprived) = IMD ranks 1–6496; quintile 2 (Q2; above average deprivation) = 6497–12,993; quintile 3 (Q3; average) = 12,994–19,489; quintile 4 (Q4; below average deprivation) = 1490–25,986; quintile 5 (Q5; least deprived) = 25,987–32,482.

The English Indices of Deprivation 201062 constitute the official measure of deprivation in England. The most commonly used of these 10 indices is the IMD, which is calculated from 38 separate indicators, based on weighted data from the following seven domains: income deprivation (22.5%); employment deprivation (22.5%); health deprivation and disability (13.5%); education, skills and training deprivation (13.5%); barriers to housing and services (9.3%); crime (9.3%); and living environment deprivation (9.3%). For the purposes of measuring deprivation, England is divided into 32,482 ‘lower layer super output areas’, with each area being carefully defined by its local geography and containing in the region of 1500 residents (minimum 1000 residents). IMD ranks, which provide a relative measure of deprivation in small areas across England, range from 1 (most deprived lower layer super output area) to 32,482 (least deprived lower layer super output area).

UK Lung Cancer Screening pilot trial statistical analysis plan

As this is a pilot trial, not powered for the end point of lung cancer mortality, analysis of the end point features is secondary in this plan, as the main UKLS was not funded. However, at the appropriate time, lung cancer mortality data will be analysed and pooled with the NELSON37 trial data. Risk scores were analysed based on a modification of the LLP risk model as published20,23 (see Appendix 2). The modifications took account of other smoking materials with the same projected risk as cigarette smoking and other respiratory diseases with the same projected risk as pneumonia.

Objectives and outcomes

To document overall response rate to the initial invitation to participate in UKLS, percentage of positive responders at high risk of lung cancer, and trial participation rate among high-risk responders. To assess the impact of specific sociodemographic factors (age, gender, socioeconomic status, region and smoking status) on trial uptake. To use these data to suggest particular demographic groups to be targeted to maximise screening uptake in any future lung cancer screening programme.

Overall response rate and risk

Figure 2 and Appendix 5 show the overall response rate, numbers at high risk and clinic attendance. Details of 249,988 individuals in the 50- to 75-year age range were provided by the PCTs to the DMC. The PCTs sent additional information requesting that 2634 of these not be contacted (e.g. because they were recently deceased). The remaining 247,354 individuals were therefore approached by the DMC. Out of these, 148,608 (60.1%) were non-responders (no questionnaire returned), 22,788 (9.2%) were negative responders (non-participation questionnaire returned) and 75,958 (30.7%) were positive responders (questionnaire 1 returned; willing to participate). Of the positive responders, 8729 (11.5%) were classified by the LLP_v2 risk prediction model to be at high risk (≥ 5% over the next 5 years) of developing lung cancer (mean LLP_v2 risk score 8.77% vs. 1.00% for the low-risk group). Of the high-risk responders, 5967/8729 (68.4%) returned the second questionnaire and agreed to participate in the RCT. A total of 1291 individuals were subsequently excluded because they (1) did not meet the inclusion criteria (e.g. because they had had a previous CT scan of the chest); (2) had not completed the eligibility questionnaire correctly; (3) replied after maximum trial recruitment numbers had been reached; or (4) were unable to give fully informed consent. A total of 582 individuals either changed their mind or failed to attend clinic, and a further 33 attended the recruitment clinic but declined to consent. In total, therefore, 4061 individuals (5.3% of all positive responders and 46.5% of all high-risk positive responders) consented and were recruited into the RCT (see Figure 2).

FIGURE 2.

A Consolidated Standards of Reporting Trials diagram showing participant flow from initial contact to CT screening. DNA, did not attend; OFS, off study; Q, questionnaire.

Factors influencing response rate and risk

Gender, age, socioeconomic status, region (north vs. south) and smoking status were analysed with respect to (1) response to the initial invitation letter; (2) risk status; and (3) participation of high-risk people in the UKLS screening trial. The actual numbers involved are tabulated in Appendix 5; trends are discussed in this chapter.

Age

Positive response rate ranged from 26.6% in the 50–55 years age group to 35.0% in the 61–65 years age group, before dropping to 27.6% in the eldest (71–75 years) age group (Figure 3). Age is a major component of the LLP risk model, so unsurprisingly had a major impact upon risk classification: only 82 out of 16,273 positive responders (0.5%) in the 50–55 years age group were classified as high risk compared with 2046 (24.8%) in the 71–75 years group. In addition to the youngest (50–55 years) age group having the lowest positive response rate and the least likelihood of being at high risk, high-risk individuals in this age group were also less likely than those in any other age group to consent to participate in the RCT: only 29 out of 82 (35%) consented. Thus, overall in the 50–55 years age group, just 29 of 61,168 individuals originally approached (0.05%) were recruited to the RCT (Figure 4). The different age profiles of the original approached group compared with high-risk clinic attendees are shown in Figure 4. A total of 94.2% of high-risk clinic attendees were aged 61–75 years (compared with 58.1% of the 247,354 subjects originally approached).

FIGURE 3.

Impact of age on initial response rate, LLP risk and trial consent rate.

FIGURE 4.

Age distribution of the original approached population sample, positive responders, high-risk responders and recruits to the RCT. In the last two categories, percentages for the youngest age group are shown to the left of the y-axis. Nearly 95% of trial recruits were aged ≥ 61 years.

When lung cancer risk was considered by individual year of age, there was a steady increase in the percentage of people at high risk, with a progressive increase starting from age 59 years (6.1% at high risk; Figure 5).

FIGURE 5.

Percentage of UKLS positive responders (n = 75,958) with a LLP risk of ≥ 5%, by individual year of age. (There are eight individuals aged 76 years; these are included with the 75-year-old group.) See Appendix 5 for primary data.

Socioeconomic status

The English IMD rank62 was available on 247,328 individuals. Analysis of IMD data showed that the social demographic characteristics of the two recruitment areas (Liverpool and Cambridgeshire) were markedly different. Almost 45% of the Liverpool area residents approached for UKLS fell into the most deprived quintile of the English population, compared with < 10% in Cambridgeshire. However, as planned in the study design, the total UKLS approached sample was similar in socioeconomic distribution to the entire English population (Figure 6).

FIGURE 6.

Social demographics of UKLS sample. Distribution of individuals within IMD national quintiles, by area. Much more deprivation is seen in the Liverpool areas (Liverpool, Knowsley and Sefton) than in the Cambridgeshire areas (Cambridgeshire, Peterborough and Bedfordshire). However, the overall socioeconomic distribution of the UKLS sample is similar to that of the entire English population: this was an inherent feature of the study design. Q, quintile.

The positive response rate increased steadily with higher socioeconomic status: 21.7% of individuals in the lowest (most deprived) IMD quintile gave a positive response compared with 39.7% in the highest quintile (p < 0.001) (Figure 7). The proportion of individuals with a high LLP risk score decreased with higher socioeconomic status, ranging from 18.2% in the most deprived quintile to 8.3% in the least deprived quintile (p < 0.001) (Figure 8). As the social gradient of response and the social gradient of LLP risk were offset by each other, the sociodemographic spectrum of the individuals attending clinic was in proportion to that of the original approached sample (see Figure 8). People attending clinic therefore spanned all IMD quintiles in roughly equal numbers, including a representative proportion from more deprived postcodes (see Figure 8). However, of the high-risk individuals invited for screening, there was a trend towards individuals of higher socioeconomic status being more likely to consent to participate in the trial (see Figures 7 and 8).

FIGURE 7.

Impact of socioeconomic status on initial response rate (lower line), LLP risk (bars) and trial consent rate (upper line). Q, quintile.

FIGURE 8.

Socioeconomic distribution of the original approached population sample, positive responders, high-risk responders and recruits to the RCT. The social demographic spectrum of the high-risk individuals participating in the trial is roughly in proportion to that of the original approached sample. Q, quintile.

Region

There was a higher response rate from individuals in the southern region (Cambridgeshire, Peterborough and Bedfordshire) compared with the northern region (Liverpool, Knowsley and Sefton) (35.0% vs. 26.4%). Much, but not all, of this difference could be accounted for by the different social demographic spectrum between the two regions. When stratified by IMD quintile, there was still a tendency for a better response rate from the southern region than the northern region (Figure 9). Within each IMD quintile, individuals from the north were slightly more likely than those from the south to be at high risk of lung cancer (see Appendix 5). In the southern region, the likelihood of high-risk people consenting to the RCT increased with IMD quintile (from 18% to 60%), whereas in the northern region the likelihood of consent was static (around 45%) for the lowest four IMD quintiles, but dropped to 35% in the highest IMD quintile.

FIGURE 9.

Positive response rate to initial approach, stratified by both region (north/Liverpool and south/Cambridgeshire) and IMD quintile. A slightly better response rate was observed in the southern region in each IMD quintile. Q, quintile.

Smoking

Of the 75,958 positive responders, 73,934 (97.3%) gave information about their smoking habits. A total of 43.4% were never-smokers, 14.7% current smokers and 39.3% ex-smokers (Figure 10). Of the 22,788 negative responders, 22,024 (96.6%) provided information about their smoking habits. Never-smokers totalled 51.3%, 9.1% were current smokers and 36.2% ex-smokers (see Figure 10). National smoking data (from the ONS) for the 50+ years age group give an expected distribution of 49.5% never-smokers, 17.5% current smokers and 33% ex-smokers. 66 Thus, the observed values in our sample of both positive and negative responders are significantly different from population values (chi-squared test; p < 0.0001). The data suggest that ex-smokers are more likely to respond positively to a screening invitation, with current and never-smokers being less likely to respond positively. Negative responders are enriched with never-smokers, but comparatively few current smokers respond negatively. From expected population figures (and accounting for known figures from responders) it was possible to estimate the smoking status of the non-responders, of whom approximately 20% are probably current smokers (see Figure 10).

FIGURE 10.

Smoking status among UKLS positive and negative responders, and likely smoking status among non-responders, based on expected values from age 50+ years population figures less known figures from responders. In this case, ‘non-responders’ includes 2024 positive responders and 764 negative responders who did not provide their smoking status.

Unsurprisingly, smokers and ex-smokers were much more likely than never-smokers to have a LLP risk of ≥ 5%. Of the 75,958 positive responders, 3724 (33.5% of 11,130) current smokers and 4995 (16.7% of 29,855) ex-smokers were designated as high risk. However, only 10 (0.03% of 32,949) never-smokers had a high LLP risk score, and only two of these consented to participate in the RCT. In total, therefore, 21.3% of current and ex-smoking positive responders were high risk, and 9.9% participated in the RCT. Of the high-risk individuals, ex-smokers were more likely than current smokers to consent to participate in the RCT (49.9% vs. 42.1%, χ² = 27.744; p < 0.0001).

Interaction between smoking and socioeconomic status

As people in more deprived socioeconomic groups are more likely to smoke, it is important to control for any possible confounding effects. Data were therefore stratified by both IMD quintile and smoking status. For all IMD national quintiles, the proportion of ex-smokers among positive responders was around 40%. Of the positive responders in the most deprived quintile, 27.6% were current smokers, decreasing to 9.1% in the least deprived quintile (see Appendix 5). Comparing positive and negative responders, there were proportionally more current smokers among positive responders, and proportionally more never-smokers among negative responders, across all IMD quintiles. There were proportionally more ex-smokers in positive responders in all but the most deprived IMD quintile (see Appendix 5).

Based on known smoking prevalence rates for 50- to 75-year-olds in each IMD quintile,67 it was possible to calculate the expected number of smokers in each IMD quintile for the 247,354 approached subjects (see Appendix 5). By subtracting the known number of smokers among positive and negative responders, this allowed an estimation of smoking status for non-responders in each IMD quintile. It was then possible to estimate the percentage of current smokers, ex-smokers and never-smokers who respond positively to the first screening invitation (Figure 11). The calculated response rate is generally lower among current smokers than the overall response rate for the relevant quintile (i.e. there is a smaller proportion of smokers among responders than would be expected from quintile-matched population figures). However, in the most and least deprived quintiles it appears that the response rate among smokers is similar to, or greater than, the overall response rate for that quintile (see Figure 11). For ex-smokers and never-smokers, positive response rate increases with socioeconomic status across all IMD quintiles, whereas the positive response rate among current smokers only rises in the upper two quintiles of IMD. The group least likely to respond positively are never-smokers in the most deprived IMD quintile (calculated as a 17.2% positive response rate), followed by current smokers in the three most deprived quintiles (calculated as 21.0–22.2% positive response rate). The highest positive response rate is seen in current smokers from the least deprived IMD quintile (calculated as a 45.2% positive response rate). The general trend across all IMD quintiles is consistent with the overall trend discussed above: ex-smokers are the most likely group to accept an invitation to be screened for lung cancer.

FIGURE 11.

Interaction between smoking status and socioeconomic status in determining initial response to the UKLS invitation. Calculated positive response rate, per cent, by smoking status and IMD quintile. The line (and percentages) denotes the known positive response rate of all individuals within that IMD quintile. Calculations were based on Health Survey for England population smoking figures for each IMD quintile, for an age group of 50–75 years (see Appendix 5). Q, quintile.

Gender

Although women are slightly more likely than men to respond positively to an invitation to take part in UKLS, more men than women are designated as being at high LLP risk (6141 males vs. 2588 females, giving a gender ratio of 2.4 : 1). However, national lung cancer incidence figures for the 50–74 years age group in the UK give a much smaller male–female ratio of 1.34 : 1. 1 In addition, as high-risk women are significantly more likely than high-risk men to opt out of the screening trial, it is possible that women are under-represented in the trial. It will be interesting to consider the longer-term gender-related outcomes in the UKLS participants in future follow-up, as data become available through the ONS, Cancer Registry and Hospital Episode Statistics.

Age

Response rate was highest in the middle age group (61–65 years) and lowest in the youngest group (50–55 years). There was a fall-off in response in the oldest age group (71–75 years), who conceivably are less concerned about their risk of lung cancer or who anticipate more practical difficulties with participating in the study. Nevertheless, as the oldest age group has the greatest percentage at high LLP risk (24.8% of positive responders aged 71–75 years were high risk) this age group was reasonably well represented in terms of trial participation (22.4%, of clinic attendees were aged 71–75 years compared with 12.1% of the originally approached population). However, as around one-third of lung cancers in the UKLS age range occur in individuals of 71–75 years,1 it could be argued that it would be desirable to screen more people from this oldest group. The best represented age group in the RCT itself was the 66–70 years group, which accounted for 39.8% of trial participants.

Only 234 (5.8%) clinic attendees were aged ≤ 60 years, of whom only 29 (0.7%) were in the 50–55 years age group. Thus the youngest (50–55 years) age group yielded a trial participation rate of just 0.05% of the 61,168 who were originally invited to participate in UKLS. This has implications for cost-effectiveness, and, under the LLP model of risk prediction, suggests that it would not be prudent to include this age group in any future population-based lung screening studies or programmes.

In order to determine more accurately when response rate and risk increase to a point where screening is viable from a detection and cost-effectiveness point of view, we analysed LLP risk by individual year of age. The results suggest a fairly sharp increase at 59 years of age in the percentage of positive responders at high LLP risk, from around 3.1% at age 58 years to 6.1% at age 59 years, and 7.5% at age 60 years. This suggests that setting the lower cut-off point for eligibility at 60 years old would be a reasonable strategy for future studies. By considering data only from individuals aged between 60 and 75 years, 8339 out of the 49,468 positive responders in this age group (16.9%) are at high LLP risk (compared with just 1.5% for the 50–59 years group).

Socioeconomic group

There was a strong positive correlation between higher socioeconomic group (less deprived quintile of IMD) and positive response to the screening invitation. Similar trends have also been observed in other screening studies, and the lower uptake is considered to relate to barriers including fear and fatalistic beliefs about cancer68,69 and poorer self-rated health in people from lower socioeconomic groups. 70 Unlike with other cancers (e.g. breast cancer) there are marked sociodemographic differences in lung cancer risk, with individuals from lower socioeconomic groups being at greater risk of developing the disease: this largely relates to disparate tobacco use in different socioeconomic groups. It is therefore not ideal that those who are at highest risk are the least likely to take up the offer of screening. Our data suggest that this works at two levels: individuals at highest risk of lung cancer (i.e. from the lower socioeconomic groups) are less likely to respond to the initial screening invitation and also less likely to attend clinic after having been identified as at high risk. Consideration will have to be given to addressing this in any future screening programme. Within each socioeconomic quintile, although there is a significant difference in screening uptake between Liverpool and Cambridgeshire, the difference is not large. Most of the difference in uptake between the two regions therefore relates to the different social demographics of the two areas.

Smoking status

Analysis suggested that ex-smokers are the most responsive to a screening invitation, and the most likely to participate in the trial once identified as high risk. It could be argued that consciously deciding to stop smoking and being motivated to participate in screening are related decisions, perhaps made by individuals who are more health aware, or perceive greater risks from smoking. The initial response rate from current smokers was lower than would be expected based on age-matched population figures, possibly suggesting that current smokers are less likely to want to consider their cancer risk, or feel more threatened by the prospect of lung cancer screening. However, it was observed that, if smokers take the trouble to respond, they are more likely in percentage terms to be positive rather than negative responders. The converse is true for never-smokers, who perhaps (correctly) view their own risk as low and hence are over-represented among negative responders. Having been identified as at high risk, current smokers were significantly more likely than ex-smokers to opt out of the trial; conceivably, this could also be related to their perception of high risk or threat. Only two never-smokers participated in the RCT; it is clear that this is not a group to be targeted in a population-based screening approach.

Interaction between smoking and Index of Multiple Deprivation

The likelihood of a positive response to a CT screening invitation is lower both in more deprived socioeconomic groups and in current smokers. As smoking status and socioeconomic status are closely related, it is important to establish whether both independently affect response rate or if there is any confound. Predicted population smoking figures stratified by IMD quintile and age adjusted to the UKLS sample67 were therefore compared with the smoking prevalence seen among UKLS positive responders in each IMD quintile. This suggests that, in general, socioeconomic deprivation and current smoking status both act independently to lower the positive response rate. However, in the highest and lowest IMD quintiles, the impact of smoking status on response rate is much less marked. In the highest IMD quintile (least deprived), the calculated positive response rate among smokers is higher than the overall positive response rate. In the case of the results from the lowest (most deprived) IMD quintile, the minimal impact of smoking status partially offsets the trend for high-risk individuals being less likely to respond. As a general trend across all socioeconomic groups, ex-smokers are the most likely to respond positively (except in the least deprived quintile, in which a greater proportion of current smokers respond positively). It is clear that smoking status and socioeconomic status interact when predicting response rate, so in a possible future UK National lung Cancer CT screening programme, strategies may therefore need to be devised to target both current smokers and individuals from lower socioeconomic groups.

Rationale

Accurate and efficient CT scan reading is crucial to lung cancer screening. In this regard, one of the aims of the UKLS pilot study was to investigate the reliability of various methods of CT scan reading and CT scan reading strategies. The method of reading CT scans has potential implications not only for successful detection of lung cancers in a screening programme, but also for the cost-effectiveness of such a programme.

Objectives

To investigate the performance of:

1. UKLS thoracic radiologists in nodule detection and interpretation (i.e. nodule reporting), having undergone initial training

2. UKLS radiographers in nodule reporting, having undergone initial training, and the comparison of radiographer performance with that of radiologists both in the trial and in the wider literature

3. radiologists in nodule reporting when using radiographers as part of a ‘concurrent’ reading strategy.

Computed tomography acquisition, storage and transfer

Computed tomography scans were acquired using a Siemens Definition Flash 128 slice scanner (Siemens, Erlangen, Germany) at Papworth Hospital and LHCH. In the initial phase of the study, from November 2011 to December 2011, a Philips (Best, The Netherlands) Brilliance 65 scanner was utilised at LHCH.

Thoracic CT images were obtained (craniocaudally from lung apices to bases) during suspended maximal inspiration, in a single breath hold and without the administration of intravenous contrast. The field of view selected was the smallest diameter as measured from the widest point of the outer rib to outer rib (usually not > 35 cm). Thin detector collimation (0.5–0.625 mm) was used with a pitch of 0.9–1.1. Exposure factors were tailored to patient height and weight, with the aim of ensuring that CT dose index was kept at < 4 milliGray (mGy), with the effective radiation dose of < 2 milliSieverts. For a 70-kg patient, a peak kilovoltage of 120 was used and the milliampere-second was tailored to achieve a CT dose index of 1.6 mGy. Images were reconstructed at 1-mm thickness with 0.7 increment, using a moderate spatial frequency kernel.

Thin-section images were transmitted to the following locations: (1) a dedicated local Syngo Siemens (Forcheim, Germany) workstation (so that nodule volumetry could be performed); (2) local picture archiving and communication system servers (so that images could be stored and subsequently retrieved to the workstation); and (3) a central site [Royal Brompton Hospital, London) via the NHS Image Exchange Portal (Burnbank Systems, Ipswich, England) for the purposes of double reading.

Quality assurance

All local sites were required to have in place daily quality assurance practices for the CT scanner, using a water and body phantom. All doses were recorded at the time of acquisition on the UKLS database by radiographers.

During the process of CT scan reading, any reader (radiologist or radiographer) could highlight defects in the acquisition quality of any CT scan, such as image degradation due to breathing artefact. All such cases were reviewed and discussed by the local and central radiologist, with respect to diagnostic quality and a decision was made as to whether or not a repeat CT scan was required.

Additionally, every month for the first 6 months, and then quarterly, 10 randomly selected cases from both sites were reviewed at the central site by two radiographers. CT scans were assessed with respect to adequacy of craniocaudal coverage, adequacy of field of view, satisfactory degree of inspiration, correct reconstruction algorithms, presence of motion artefacts and the recorded radiation dose. Outliers were highlighted to the central site radiologist and feedback was provided to local sites if any of the acquisition parameters were deemed to be out of range.

Reading methods

To optimise sensitivity and specificity, all baseline CT scans were read by two thoracic radiologists at a local (LHCH or Papworth Hospital) and central (Royal Brompton Hospital) site (see Reading selection, training and testing). All discrepancies were reviewed by a third thoracic radiologist at the central site, who was the final arbiter. Once consensus had been achieved for all nodules, a letter was sent to the participant and his/her GP, outlining the results of the scan. The letter also detailed whether or not further follow-up CT scans within the UKLS were required. No computer aided detection software was used.

Nodule classification and management

Nodules were classified into one of four categories as per the UKLS nodule management protocol53 (Figure 12).

FIGURE 12.

The UKLS nodule care pathway management protocol. 53 VDT, volume doubling time.

Category 1 Benign nodules fulfilling one of the following criteria: a benign pattern of calcification, fat, measuring < 3 mm in diameter or volume of < 15 mm³; or intrapulmonary lymph nodes fulfilling the following criteria: they lie within 5 mm of the pleura, are < 8 mm in diameter, smooth bordered and ovoid, and have at least one interlobular septum radiating from their surface.

Category 2 If solid and intraparenchymal, a maximum diameter of 3.1–4.9 mm or a volume of 15–49 mm³. If solid and pleural or juxtapleural, a maximum diameter of 3.1–4.9 mm. If non-solid or part solid, a maximum diameter of 3.1–4.9 mm. The solid component has a diameter of < 3 mm and/or volume of < 15 mm³. All non-solid/ground glass opacities, independent of diameter.

Category 3 If solid and intraparenchymal, a volume of 50–500 mm³. If solid and pleural or juxtapleural, a diameter 5–9.9 mm. If non-solid or part solid, a diameter of the ground-glass component of > 5 mm. If part solid, the solid component has a volume of 15–500 mm³ or a maximum diameter of 3.0–9.9 mm.

Category 4 If solid and intraparenchymal, a volume of > 500 mm³. If solid and pleural or juxtapleural, a diameter of ≥ 10 mm. If part solid, the solid component has a diameter of ≥ 10 mm or has a volume of > 500 mm³.

Nodules were managed as follows (see Figure 12):

No nodules or Category 1 nodules No further action required.

Category 2 nodules Follow-up CT scan at 12 months.

Category 3 nodules Follow-up CT scan at 3 months.

Category 4 nodules Referral to MDT.

When follow-up scans (at 3 or 12 months) were performed, the volume doubling time (VDT) of the nodule was calculated. VDTs were designated as < 400 days or ≥ 400 days.

Reading of follow-up computed tomography scans

A key part of the UKLS protocol was the assessment of VDTs to identify significant nodule growth with a follow-up CT scan. If nodules were deemed to have been reliably segmented at baseline and follow-up then VDTs were automatically calculated in the UKLS database. For this to happen, CT scan readers were required to match the data of follow-up nodules with baseline nodules, by ‘dragging and dropping’ the data into the same row as the baseline nodule in the UKLS database (Figure 13). From October 2013, the UKLS protocol was changed such that follow-up CT scans were read by a single reader. This was done because the reading of follow-up CT scans largely consisted of the task of matching nodules on the follow-up scan to existing nodules that had already been identified. (This was in contrast with all baseline scans, for which double reading was used because of the theoretical benefits of using two readers to improve lung nodule detection and interpretation.)

FIGURE 13.

Screenshot from UKLS database. Nodules at baseline, 3-month and 12-month CT scans are matched in a row. The second figure in each box refers to the slice number and the third figure represents nodule volume. By hovering over the nodule, a figure for VDT is generated automatically. All nodules with a significant VDT of < 400 days are highlighted in purple (not shown in this example).

Computed tomography acquisition

From 29 November 2011 to 8 June 2013, baseline CT scans were performed on 1994 participants. Seven of these 1994 scans (0.4%) required repeating (six because of breathing or motion artefact and one because of inadequate craniocaudal coverage).

The range of doses for baseline CT scans was 0.54–3.93 mGy, with a median dose of 1.62 mGy.

Overall performance of radiographers and radiologists

Radiographers 1–4 had sensitivities of 67.6%, 77.8%, 79.4% and 61.6%, respectively (mean sensitivity 71.6 ± 8.5%). Radiologists A–C had sensitivities of 88.9%, 87.0% and 74.0%, respectively (mean sensitivity 83.3 ± 8.1%).

The average numbers of false-positives per case for radiographers 1–4 were 1.2 (SD 2.1), 2.9 (SD 2.8), 0.6 (SD 1.0) and 1.1 (SD 1.3), respectively, whereas those of radiologists A–C were 0.5 (SD 0.8), 0.7 (SD 1.0) and 0.2 (SD 0.5), respectively.

Direct comparison of radiographer and radiologist performance

The sensitivities of each radiographer compared with those of the corresponding radiologists within a particular radiographer–radiologist combination are illustrated in Table 2.

Radiographer sensitivity was significantly lower than radiologist sensitivity in 7 of 10 radiographer–radiologist combinations (range of difference 9.7–32.8%; p < 0.05), and not significantly different in 3 of 10 combinations. Radiographers had significantly higher average false-positives per case than radiologists in 8 of 10 combinations (range of difference 0.4–2.6; p < 0.05), and there was no significant difference in the remaining two combinations (Table 3).

Reader performance in the first 10 weeks (period 1) compared with second 10 weeks (period 2)

The two radiographers with the lowest overall sensitivity (radiographers 1 and 4) showed a significant improvement in sensitivity between the first and second 10-week period (sensitivity 50.0% in period 1 vs. 74.1% in period 2 for radiographer 1; 41.8% in period 1 vs. 67.2% in period 2 for radiographer 4; p < 0.005), but their sensitivity in period 2 still did not reach the level of radiographers 2 and 3, who showed no significant difference in their sensitivity between the two periods.

Radiologists’ sensitivity did not significantly differ between the two periods. No radiographer or radiologist demonstrated a significant difference in average false-positives per case between the two periods.

Comparisons of sensitivity using alternate diameter thresholds

There were 236 reference standard nodules that were ≥ 5 mm in diameter. When considering only these nodules, the number of radiographer–radiologist combinations with significantly lower radiographer sensitivity decreased to 4 of 10 combinations (range of difference 17.2–37.3%; p < 0.05). No significant difference was seen in the remaining 6 of 10 combinations.

Comparison of radiologists’ performance when reporting alone and concurrently with radiographers

The overall sensitivity of each radiologist for the different reading methods is detailed in Table 4. The mean sensitivity for radiologists reading independently was 77.5 ± 11.2%, increasing to 90.8 ± 5.6% with the use of concurrent reading. For all but one radiologist (radiologist D), statistically significant higher sensitivity was achieved with concurrent reading compared with independent reading.

There was a wide variation in the average false-positives per case. Although the overall mean of average false-positives per case increased from 0.33 ± 0.20 with independent reading to 0.60 ± 0.53 with concurrent reading, average false-positives per case ranged between 0.06 and 1.38, increasing with concurrent reading for radiologists A–C (and statistically significant for radiologists B and C) but decreasing for radiologist D (Table 5).

Reading times using independent and concurrent reading methods

The mean reading times per case for concurrent reading ranged from 6.2 to 8.6 minutes compared with 7.0–12.4 minutes for independent reading (Table 6). Concurrent reading was faster than independent reading for all radiologists, but this increase in reading speed was not statistically significant for radiologist C. Furthermore, the maximum decrease in mean reading time was just < 4 minutes (radiologist A).

Outcomes

The outcomes were the number of baseline and follow-up scans performed, categorisation of any lung nodules identified on scan, number of lung cancers diagnosed (including their staging and treatment) and any significant other findings. Mortality is not reported, as data are not mature.

Quality control for pathology of resected nodules

Quality control for histopathological examination of resected involved exchange between the reference thoracic pathologists at Liverpool (Professor John Gosney) and Papworth (Dr Doris Rassl) of a representative haematoxylin and eosin-stained section from all cases. This was accompanied, where necessary, by any immunolabelled sections used in diagnosis and/or classification of the lesion. Sections were reviewed ‘blind’ and responses were exchanged, with appropriate discussion, in cases of discordance.

Statistical analysis

Logistic regression was utilised in order to determine which factors were associated with requirement for further scans or investigations. Stata version 12 was used for analysis. The following variables (data obtained from questionnaire 1) were used in both univariate and multivariate analyses. For analyses including smoking status, the two never-smokers in the CT screening arm of the trial were removed from the analysis.

Diagnostic work-up and false-positives

In the UKLS, we defined false-positives or rate of recall as those requiring further diagnostic investigation more immediately than a repeat annual screen, but who transpired on such investigation not to have lung cancer.

Overall, 951 of 1994 (47.7%) of subjects underwent at least one further CT scan after the initial screen. This comprised 479 subjects in category 2 who underwent a 12-month interval CT scan and 472 in category 3 (see Table 10). In addition, at the time of reporting, a further 414 CT scans have been carried out as per protocol (see Figure 14).

It should be noted that a repeat CT scan at 3 months for category 3 nodules was mandated by the protocol. There were a total of 536 subjects (i.e. 472 category 3, 64 category 4) with nodules requiring a repeat scan. Of the 64 category 4 individuals, 41 were found to have lung cancer.

Owing to our failsafe policy reflecting the single-screen design (i.e. follow-up of category 2 individuals), there were a further 479 individuals for whom a repeat screen was recommended at 12 months, only one of whom was found to have a confirmed cancer.

The referral rate to the MDT was low and there were relatively few people with benign disease who had invasive tests. One hundred and fourteen (5.7%) were referred to the MDT of whom 72 (3.6% of the total; 72/1994) did not have cancer.

For complete clarity, the proportion of false-positive tests is now provided in two ways, which allows an appreciation, in a patient-centred approach, of the variable impact on the subject in a trial or the patient in a programme. A ‘false-positive’ that mandates referral to the lung cancer MDT will usually be associated with significant psychological distress, and additional more or less invasive investigations with, in some cases, definitive treatment. An individual with a false-positive so defined is thus more likely to suffer harm than one defined in a different way; that is, those subjects who are recalled solely for further CT imaging to clarify the nature of a nodule. The latter is best termed ‘interval imaging rate’ and may, in screening programmes, merely mean continuing in the programme rather than referral to the MDT. For this reason, all category 3 lesions without cancer are reported separately as false-positives warranting interval imaging. Category 2 findings are not classified as false-positives warranting recall, as the cancer rate was found to be so low in this study that interval imaging would not be recommended.

Thus, on examining the number of UKLS participants referred to the MDT clinic, the false-positive rate is 3.6% (114 – 42/1994 = 3.6), whereas the interval imaging rate for the category 3 nodules is 23.2% (472 – 9/1994).

In total, 114 of 1994 (5.7%) participants were referred to the MDT, of whom 42 (2.1% of all screened) had lung cancer.

Details of cancer diagnoses

Of the 42 cancers with a confirmed diagnosis, 27 were diagnosed at LHCH, and 15 at Papworth Hospital. Twenty-five diagnoses were adenocarcinoma, 12 were squamous cell carcinoma, three were small cell carcinoma, one was a typical carcinoid and one was of unknown type (this patient was treated with palliation alone).

Pathological staging of cancers was completed for 35 of 42 cancers: 17 were T1aN0 (one Nx), six were T1bN0, two were T2aN0, two were T2bN0, two were T1aN1, three were T2aN1, two were T1–2N2 and one was T3N0. Clinical staging was recorded for a further seven cancers (Table 10). Thus, 67% of the screen-detected cancers that have so far been staged (28/42) were identified at stage 1. Details of all individuals referred to the MDT with a cancer diagnoses are shown in Table 10.

Incidental findings

To date, over the course of the UKLS screening trial, 128 SOFs have been identified on the CT scans by the trial radiologists. Most of these findings related to thoracic disease, but 17 extrathoracic SOFs were also identified. Individuals with a SOF on scan were referred for further investigations and treatment to a relevant specialist or MDT, or their own GP. Details of the thoracic and extrathoracic SOFs are shown in Tables 12 and 13, respectively.

Significant other findings were managed locally in Liverpool or Papworth. For any incidental findings that were identified before 9 December 2013, the participants’ GPs were contacted by UKLS and asked to provide follow-up and outcome data. The results from both Liverpool and Papworth are detailed below. Follow-up information for incidental findings arising after 9 December 2013 has not yet been requested, hence the figures below do not include all of the incidental findings detailed above (see Tables 12 and 13).

In the Liverpool cohort, 55 SOFs were identified, and the patients’ primary care physician informed. Seventeen different conditions were identified, including eight abdominal conditions. Thoracic conditions included pneumonia (n = 18), pulmonary fibrosis (n = 8), pleural thickening (n = 5), bronchiectasis (n = 3), lymphadenopathy (n = 3), severe emphysema (n = 3), lobar collapse (n = 2), oesophageal thickening (n = 2), mediastinal mass (n = 2), thyroid mass (n = 1) and thoracic aortic aneurysm (n = 1). Abdominal abnormalities included adrenal lesion (n = 2), abdominal aortic aneurysm (n = 2), cirrhosis of the liver (n = 1), splenomegaly (n = 1), renal mass (n = 1) and pancreatic cyst (n = 1). Further information on these patients was identified from secondary care records and 31 responses from primary care for follow-up data. Although the vast majority of pneumonias were managed in primary care, at least 22 of the others were referred to secondary and tertiary care for further investigation and management.

There were 48 SOFs in the Papworth cohort. The GPs of patients with SOFs were contacted by letter and asked to take appropriate action. For five cases with a potentially life-threatening condition the letter was followed up with a telephone call a few days later to ensure that action had been taken. Thoracic conditions included interstitial lung disease (n = 10), severe emphysema (n = 2), pneumonia (n = 16), atypical pneumonia (n = 4), lymphadenopathy (n = 3), posterior mediastinal mass (duplication cyst) (n = 1), thymoma (n = 1) and aortic calcification (n = 4). Abdominal conditions included renal mass (n = 2), adrenal mass (n = 1), hepatic mass (n = 1), abdominal aortic aneurysm (n = 1), biliary dilatation (n = 1) and hydronephrosis (n = 1). In 29 cases, outcomes are known from secondary care records or via the primary care follow-up questionnaire. Of these, the UKLS identified a diagnosis that was already known to the GP in three cases and no further action was taken. In 12 cases the GP treated the patient and in 14 cases the patient was referred to secondary or tertiary care.

Overall, about 5% of patients had a SOF identified. Nearly all of these were previously undiagnosed conditions. Their identification empowered primary care physicians to manage or appropriately refer on, for the patient’s benefit.

The importance of assessing psychosocial impact of lung cancer screening

Alongside clinical evidence, an evaluation of psychosocial harms and benefits is essential when considering the introduction of a new screening programme. Studies of cancer screening in the general population have highlighted adverse psychosocial effects, in particular those associated with abnormal, false-positive or inconclusive test results. 83,84 The psychosocial impact of LDCT screening for lung cancer has been examined in controlled trials outside the UK, in terms of effects of both trial allocation and screening results. 85–91 Overall, evidence from these trials suggests that LDCT screening does not produce long-term anxiety or other adverse effects that could deter high-risk individuals from further help-seeking or future participation in a lung screening programme. However, it is important to examine psychosocial effects within the UKLS context to ensure that any future lung screening programme is tailored to the specific needs of the UK high-risk population.

Barriers to uptake of lung cancer screening

Previous studies have highlighted some of the demographic and psychosocial barriers to lung screening uptake. Silvestri et al. 92 reported that smokers perceived fewer benefits and were less likely to consider lung cancer screening than non-smokers. Non-participants in NELSON88 were more likely to be female and former smokers, and to have a lower perceived risk of lung cancer and less-positive beliefs about the benefits of lung screening than participants. 93 In the first 88,897 individuals approached to take part in UKLS, participation was more likely in ex-smokers and those in higher socioeconomic groups. 94 Other barriers may include fears about lung screening tests and radiation exposure, and fatalistic beliefs about lung cancer. 95,96

Psychosocial impact of lung screening trial allocation

The psychosocial impact of trial group assignment has been examined in both the NELSON trial88 and the DLCST. 85 In NELSON,88 the intervention and control groups were equivalent in a range of psychosocial outcomes, both general and cancer-specific. Similarly, the DLCST found no differential effects of trial allocation on psychosocial outcomes at 1-year follow-up. 85 Although those in the control group reported slightly worse psychosocial outcomes on some measures at follow-up, the difference between groups was not significant when examining the mean increase in scores on each scale. The NCI’s PLCO87 reported poorer health-related quality of life (HRQL) in the intervention groups on one measure (mental component of the Short Form questionnaire-12 items), but no difference on any other psychosocial measure.

Psychosocial impact of lung screening results

Recent trials that have examined the psychosocial effects of lung screening results indicate that the negative impact of receiving an unfavourable result, especially in relation to cancer-specific distress, tends to diminish over time. The NELSON trial88 reported poorer quality of life, increased general anxiety and clinically elevated levels of cancer-specific distress at 2 months follow-up in high-risk individuals receiving an indeterminate CT scan result than in those receiving a negative (normal) result. However, these effects did not persist at 6 months’ follow-up (i.e. after the second screening round). 88,89

A similar pattern was observed in the Pittsburgh Lung Screening Study86 with a temporary increase in stated anxiety in those with an indeterminate screening result, although cancer-related fear did not return to baseline levels at 1-year follow-up. Participants in the PLCO who received at least one abnormal result across any of the cancers being screened, reported higher cancer-specific distress in the short term than in those who had received all normal screening results. 87 No significant differences were found on the mental or physical components of the Short Form questionnaire-12 items or were there differences found on any of the measures at 9 months’ follow-up, suggesting that negative psychosocial consequences of lung screening results are not long lasting.

Participants

A random sample of 247,354 individuals, aged between 50 and 75 years, from six PCTs (three from the Liverpool area and three from the Cambridge area) were initially invited to take part in the UKLS by completing a risk assessment questionnaire based on the LLP_v2. Individuals who were identified as being at high risk of lung cancer were then sent further information about the trial and invited to participate. Following completion of a further screening questionnaire to identify trial eligibility, people who met the criteria were invited to attend the recruitment centre. High-risk individuals who attended the recruitment centre were confirmed as eligible, and who consented to take part in the trial were included in the psychosocial outcomes component of the study. Individuals who actively declined to take part by completing an optional non-participation questionnaire are referred to as ‘negative responders’.

Procedure

At the UKLS recruitment centre, participants completed a baseline touchscreen questionnaire, which included a number of psychosocial measures (referred to as T0). Participants were sent a follow-up psychosocial questionnaire (T1) approximately 2 weeks after receiving either (1) the letter detailing that they had been assigned to the control arm of the trial (control group) or (2) the baseline CT scan result letter (intervention group). A freepost envelope was included with the questionnaire pack for the participant to use to return his/her completed questionnaire to the DMC. Between September 2012 and January 2013, a subset of just over 600 participants who did not return their T1 questionnaire received a reminder. A further follow-up questionnaire (T2) and freepost envelope were sent in January 2014 (10–27 months after attending the recruitment centre) to all participants who were still alive and not ‘off study’.

Measures

Statistical analyses

Analyses were conducted using SPSS version 20 (IBM Corporation, Armonk, NY, USA). A mean replacement strategy was used when participants were missing data on the psychosocial variables. Scale scores were assessed for normality and log-transformed in cases of non-normal distribution (cancer distress, anxiety and depression). Statistical associations between individual characteristics and trial participation were examined using the univariable and multivariable regression modelling. Attrition bias was assessed using the chi-squared test and independent t-tests.

Trial participation

A flow diagram of participation in the trial is shown in Figure 15. In total, 4061 individuals (5.3% of 75,958 positive responders; 46.5% of all high-risk positive responders) attended the recruitment clinic and were consented into the trial. Of these, 4039 trial participants completed baseline T0 psychosocial questionnaires: 4037 were randomised (n = 2018 intervention, n = 2019 control) and 3232 of the randomised participants completed T1 psychosocial questionnaires [n = 1653 (84%) intervention, n = 1579 (78%) control]. At T2, 2855 participants completed psychosocial questionnaires [n = 1553 (82%) intervention, n = 1302 (65%) control].

FIGURE 15.

Trial participation flow diagram.

Attrition bias

Short-term outcomes analysis

The final sample for short-term psychosocial outcomes analysis (i.e. trial participants who had completed both T0 and T1 psychosocial questionnaires) consisted of 3232 individuals (see Figure 1). The mean age of participants was 67.7 years (see Table 16). The majority of participants were male (76%), married or cohabiting (75%), and of white ethnic origin (99%). One-third (31%) of the sample were educated up to and including General Certificate of Secondary Education, Ordinary level or equivalent, and 40% were educated beyond (data on educational level were missing or non-informative for 28% of the sample). Most of the sample comprised ex-smokers (63%). There was a spread of different levels of deprivation within the sample, with just over one-quarter being in the highest quintile and just over one-quarter being in the lowest quintile. Low mean levels of cancer distress, anxiety and depression were observed at baseline (see Tables 16 and 19). When categorised, 10% of the sample showed high levels of cancer distress at baseline, 8% showed moderate to severe levels of anxiety and 3% showed moderate to severe levels of depression. Just over 40% were very satisfied with their decision to take part in UKLS. Comparisons between trial allocation groups showed no significant differences in baseline characteristics (Table 19).

Hypothesis 1: intervention arm participants will report higher short-term cancer distress than those in the control arm

As shown in Table 20, different effects of trial allocation on T1 lung cancer distress were found for participants who scored above and below the distress threshold (12.5) at baseline. For those with low baseline distress, T1 distress scores were significantly higher in the intervention group, although not to clinical levels, and with a very small effect size. For those with high baseline distress (10%, 326/3225), the effect of trial allocation on T1 cancer distress was not significant: mean levels of cancer distress remained high and bordered on clinical levels regardless of trial allocation (47%, 153/326). The effect of trial allocation on T1 general anxiety was not statistically significant. Significantly higher log-transformed depression scores were found in the control group, but the absolute difference was very small. When converted back to the original scale (0–21 range), depression scores were low and within the normal range for both control and intervention groups.

A significantly greater proportion of control arm participants were not very satisfied with their decision to take part (66%, 953/1451) than those in the intervention arm (58%, 875/1499) (χ²₍₁₎ = 16.7; p < 0.001).

Sensitivity analyses showed small changes in F-values but not in statistical significance levels, therefore further exclusions were not made.

Hypothesis 2: intervention arm participants with a positive baseline computed tomography scan result will report higher short-term cancer distress than those with negative results

As shown in Figure 15, 1653 participants who completed a T1 questionnaire were eligible for inclusion in the impact of baseline scan result analyses. Thirteen participants were excluded because of discrepancies in classification of their test result.

Preliminary analyses were conducted to test for potential covariates and are shown in Appendix 7, Table 35. Baseline differences between the screening result groups were not significant, hence only the baseline scores of each dependent variable were included as covariates.

Table 21 shows a statistically significant difference in T1 lung cancer distress among the different result groups. Participants who were positive for MDT referral (major lung abnormality) were significantly more distressed than each of the other result groups (negative mean difference 0.36; p < 0.001; negative with incidental finding mean difference 0.33; p < 0.001; positive for repeat scan mean difference 0.24; p < 0.001). Lung cancer distress scores for the MDT group approached clinical thresholds, with upper CIs crossing into clinically significant levels. Participants who were positive for a repeat scan also reported significantly greater T1 cancer distress than those receiving a negative (normal) result (mean difference 0.12; p < 0.001).

Differences in T1 general anxiety were found between groups. Those referred to MDT reported significantly greater anxiety than those receiving any other result (negative mean difference 0.36; p < 0.001; incidental finding mean difference 0.37; p = 0.022; positive for repeat scan mean difference 0.31; p = 0.003), although their scores were in the low/normal range. The difference in anxiety between the MDT referral group and the incidental finding group was not significant when the sensitivity analysis accounted for test timing issues. When participants who had completed their T1 questionnaires at more than a month after they had been sent their result were excluded (n = 174), the difference in anxiety for these groups was no longer significant (mean difference 0.28; p = 0.18). Differences in depression scores were not statistically significant for any of the screening result groups.

There was a significant association between screening result group and decision satisfaction (Table 22). A greater proportion of screening participants who were positive for MDT referral were very satisfied with their decision compared with the other three groups. Those who were positive for repeat scan had the greatest proportion not very satisfied with their decision.

Long-term outcomes analysis

Data were available for 2855 trial participants for the long-term outcomes analysis (see Figure 15). Demographic and psychosocial characteristics of the sample at T2 were very similar to those of the sample included in T1 outcome analyses (see Table 19 and Appendix 7, Table 36).

Hypothesis 3: there will be no difference in long-term cancer distress between trial arms

Analyses of T2 cancer distress scores were split by high and low baseline scores because of differential short-term effects of trial allocation on these groups. These effects were not evident at long-term follow-up: in both T0 low- and high-scoring participants, the difference between trial arms in T2 cancer distress levels was not statistically significant (Table 23). Control group participants had significantly higher T2 anxiety and depression scores than those receiving the intervention; however, the absolute differences between trial arms were minimal and not of clinical significance. When converted to raw scores, all three measures for both trial arms were within the normal range.

A significantly greater proportion of those in the control arm (74%, 883/1189) were not very satisfied with their decision to take part in screening compared with the intervention arm (60%, 855/1422, χ²₍₁₎ = 58.2; p < 0.001).

Hypothesis 4: there will be no difference in long-term cancer distress between screening outcome groups

The impact of screening outcome was assessed at long-term follow-up. Participants were excluded if they had no baseline scan, had a MDT referral but no lung cancer diagnosed, were awaiting their scan or results, or had results that were not classified (n = 267). Figure 16 indicates the final number of participants in each T2 screening outcome group: 23 true-positive (i.e. lung cancer diagnosed), 445 false-positive, 740 true-negative and 78 incidental screening results.

FIGURE 16.

Screening outcome groups at T2. Note: participants were categorised into screening outcome groups based on screening history and outcome of the final CT scan (where relevant) at the time of T2 questionnaire completion.

As shown in Appendix 7, Table 37, differences between T2 screening outcome groups on baseline measures were not observed.

Differences between screening outcome groups in T2 cancer distress, anxiety and depression were not statistically significant (Table 24). The raw scores of all psychosocial variables in all groups were observed to be within the normal range and not clinically relevant. Similarly, no differences were found in satisfaction with decision to take part in screening between the four groups (Table 25).

Secondary analysis: linear mixed-effects modelling

Impact of trial allocation

Short-term negative consequences were observed in individuals allocated to receive LDCT screening, with higher cancer distress in the intervention arm at 2–4 weeks’ follow-up. However, this was a temporary effect that did not persist at long-term follow-up. Linear mixed-effects modelling confirmed that, despite the notional increase in cancer distress for those in the intervention arm in the short term (but not in the long term), there was little evidence of a meaningful difference between trial allocation groups once effect modifiers had been taken into account. Similarly, both the NELSON88 and PLCO87 trials found minimal long-term psychosocial effects of allocation to LDCT screening or no screening control. Although UKLS participants who were assigned to the control group reported slightly higher long-term anxiety and depression, the absolute differences were again small and not clinically significant. The latter trend is likely to reflect the control group’s disappointment and frustration at having been identified as high risk but denied the intervention, supported by the finding that a greater proportion were less satisfied with their decision to participate compared to the intervention arm. This finding mirrors the DLCST, in which control arm participants reported more negative consequences,85 perhaps perceiving a missed opportunity to gain reassurance from screening.

Overall, scores on measures of distress, anxiety and depression in the present trial were within the normal range in each arm and at each time point. For the minority of participants with pre-existing high cancer distress, short-term distress levels were high, regardless of trial allocation. These individuals could be identified for additional psychosocial support during routine lung screening. 97 Individual difference variables that adversely influenced levels of cancer distress over time, regardless of trial allocation, included female gender, younger age (< 65 years), smoking, deprivation and having prior experience of lung cancer. In addition, psychosocial outcomes were poorer in those recruited at the Liverpool site, over and above the effect of deprivation, suggesting that supportive interventions to improve the quality of care and minimise distress should be implemented alongside routine CT lung screening.

Impact of computed tomography screening results

The initial impact of trial allocation largely reflects transient adverse effects of positive screening results, which disappeared at up to 2 years’ follow-up. Within the intervention arm, there was a significant graded effect of screening result on short-term lung cancer distress. Unsurprisingly, participants who were referred to MDT because of a suspected major lung abnormality on their baseline scan reported higher cancer distress than any other screening result group, with levels close to threshold scores. Participants who required a repeat scan reported higher distress than those receiving an immediate ‘all-clear’ result. Differences in general anxiety (but not depression) were observed, with higher levels of short-term anxiety in participants who were referred to MDT. However, MDT group scores were within the normal range and they were also more at ease with their decision to take part in UKLS than any other screening outcome group. The latter finding has been reported in other screening evaluation studies, reflecting decision consolidation and the perception that further diagnostic tests have been carried out thoroughly and for personal benefit. 99

Adverse effects of screening result were not seen in the long term, when the final outcome of the screening episode was examined, although type II error cannot be discounted because of the relatively small number of participants (within the psychosocial sample) who eventually received a lung cancer diagnosis. Similarly, the absence of observed psychosocial effects in the small group of participants with incidental findings may reflect the heterogeneity of clinical diagnoses in this group. A number of participants were excluded from screening outcomes analyses, for example those referred to MDT but in whom lung cancer was not diagnosed; longer-term effects in these individuals are therefore unknown. The potential for distress caused by unfavourable or unexpected findings of CT lung screening should not be ruled out, and people who receive such results may benefit from further psychosocial support. On balance, however, the pattern of findings supports those of previous studies,87–89 which indicate long-term resolution of adverse screening effects.

Trial non-uptake

Although the overall trend towards minimal psychosocial consequences of the UKLS is encouraging, it is important to acknowledge the possibility of sample selection bias, which reduces the trial’s external validity and may mean that adverse effects were underestimated. A profile of potential risk factors for non-uptake of lung screening was revealed: high-risk individuals who were older (> 70 years of age), female, smokers, from a lower socioeconomic group or with a higher affective risk perception were less willing to participate in the trial. These findings are consistent with barriers to uptake reported in previous lung screening studies, including female gender,100 perceived threat associated with lung cancer and lung screening tests,95,96 and low perceived benefit among smokers. 92 Individuals from more affluent backgrounds may have a better understanding of the benefits of screening and face fewer barriers than those from poorer backgrounds. 101 Furthermore, these groups continued to be under-represented as the trial progressed, with greater attrition seen in women, smokers, less affluent individuals, those from the Liverpool area and those with higher baseline psychosocial scores. The same groups were also vulnerable to experiencing higher cancer distress when they did take part in the trial.

UK Lung Cancer Screening evaluation model

The UKLS was very much smaller and very much briefer than the cited colorectal trial. Although the original intention had been to recruit more than 30,000 subjects and to follow them for at least 5 years,46 UKLS did not proceed beyond its pilot phase, whereby approximately 2000 subjects were screened and observed over 12–15 months. This short follow-up period precludes adopting the conventional approach to trial evaluation, namely, the measurement of long-term costs and outcomes in both the test and the control arms, and the comparison thereof. Of necessity, the observational element of the economic evaluation was restricted to those events and findings that occurred within the active trial period. Observable costs that accrued in the active period were the costs of (1) screening the target population; (2) rescreening or investigating patients with suspicious nodules according to the trial protocol;53 and (3) diagnostic work-up and treatment for the detected cancers. In that which follows, all costs are expressed in UK pounds sterling at 2011–12 prices. Where costs originally denominated in currencies other than sterling are reported, values have been converted to 2011–12 pounds sterling at purchasing power parity exchange rates.

Although the numbers and type of abnormality detected as a result of screening were also observable, the possible consequences of detection were not. Any survival benefits would accrue far beyond the current (observation) period. Being vital to the evaluation, these benefits were therefore modelled on the basis of observational data from other contexts. We presumed that the survival benefits of screening would be confined to those patients in whom cancers had been detected, and that such benefits would constitute the only source of health gain in the screening programme. For each of the cancers detected, the patient’s health gain would equal the life-years to be expected following screen detection and treatment less the life-years that the patient might expect as a result of treatment following eventual symptomatic presentation, assuming the patient had survived that long. In addition to future benefits, we modelled the future costs of the diagnostic work-up and treatment of the same cancers which would present eventually, had they not been screen detected. These were to be offset against the current period costs of the screening programme.

The UKLS recorded incidental findings, principally cases of pneumonia, bronchiectasis, interstitial lung disease and severe emphysema. Such findings are regularly reported in CT screening studies. 107 As the collection of cost and outcome data for individual non-cancer cases lay outside the trial protocol, the consequences of incidental findings have not been included in the UKLS economic evaluation. We note that only the effect of including costs and consequences of incidental findings would be indeterminate. Although it is probable that incidental findings would generate additional work-up and imaging costs (studies in both Canada108 and Italy109 have reported costs equivalent to around £9 per participant), it is equally likely that the earlier diagnosis of these non-cancer conditions could potentially generate compensating outcome gains.

Invitation and selection

Our calculations thus far have excluded the costs of selecting and inviting screening subjects. Obviously, subjects were invited to the UKLS, but the sequential invitation procedure was devised to target and recruit particular participants to a clinical trial, and to obtain both clinical and sociodemographic data. An invitation protocol addressing the needs of research does not provide a template for a pure screening protocol; including the actual recruitment costs for UKLS would therefore prove misleading when judging the likely cost-effectiveness of a screening programme. Were CT screening for lung cancer to be implemented in the UK, it seems probable that it would be organised in a fashion similar to other national cancer screening programmes (i.e. via a centralised, computerised call system). We constructed, accordingly, a hypothetical recruitment process to represent the invitation of the UKLS subjects who were actually screened.

We took our lead from UK colorectal screening, for which the current cost of inviting subjects to flexible sigmoidoscopy screening has been estimated at around £6 per invitation. 112 A programme of CT colonography in the Netherlands has been the subject of a very detailed audit,113 and the combined costs of invitation, reminders and distribution of results convert to a similar figure, around £5.50 per person invited. An additional requirement for lung cancer would be the restriction of potential subjects to a defined high-risk category, as cost-effectiveness is positively associated with disease prevalence. 52 Risk profiling following individual person-to-person contact is likely to prove expensive, but the presence of many propriety risk calculators freely available via the internet suggests the possibility of developing a cheaper alternative to judge eligibility. In our model we assumed that pure invitation and selection costs would have amounted to £10 per person invited. We assumed that, of those invited and selected, only 30% would agree thereafter to be screened. In other words, nearly 7000 individuals would have to be approached in an actual programme to achieve the numbers screened in the trial. This proportion is similar to that of those expressing an interest in screening following receipt of the initial UKLS questionnaire. It should be noted, however, that real-world cancer screening typically achieves a higher uptake rate when implemented as a formal programme. The uptake rate for the initial phases of the national colorectal screening programme, for example, was of the order of 50%. 114 Factoring these invitation/selection costs into the total caused gross current costs to increase by £67,260, or 10%, to £754,877 (95% CI £544,824 to £966,304).

Age-, gender- and stage-specific survival

The original, general, model52 failed to distinguish different cancer stages or different ages at screen detection. The detailed findings of UKLS, however, permit a more rigorous specification and enable us to simulate screen-detecting each of those cancers that are actually detected. We solved the survival model for each cancer individually, using life tables specific to the patient’s gender and his/her age at screening, assuming post-detection survival rates specific to the cancer stage at detection. As noted above, this required the assumption of lead times and two sets of mortality rate adjustments (screen detection vs. symptomatic presentation) for each individual cancer detected.

The UKLS detected 10 cancers among women and 32 among men, at ages ranging from 56 to 77 years. The numbers by stage were 28, 8, 3 and 3 for stages 1–4, respectively. Survival rates by stage and gender were assigned by assumption, informed by the literature, as follows.

The data available on which to build our survival assumptions were limited, because only the earliest of the clinical studies have accumulated sufficient follow-up time. With respect to post-treatment survival of screen-detected cancers, the principal source remains the US ELCAP, which reported 92% 10-year survival for patients in which operable stage 1 tumours were identified, and 80% survival for all cancers detected and treated. 115 A Japanese programme116 reported 90% 5-year survival for all screen-detected cancers, with 97% for those detected at stage 1. An earlier mobile programme in Japan117 reported 10-year survival at 88% for stage 1 and 50% for cancers at more advanced stages. Survival projections from ongoing trials have suggested 90% 2-year survival for cancers detected at early stages (stages 1 and 2) and 47% for cancers at advanced stages. 118 Although not trial based, audit studies from Alabama, USA,119 and from Taipei, Taiwan,120 indicated 5-year survival of > 70% for those treated for stage 1 cancers. As screening trials tend to focus on detecting early-stage disease, relatively little attention has been paid to the impact of screening on the survival of late-stage cancers. A recent review121 suggested that survival following treatable stage 3 cancer may be similar to that of early-stage disease and, with appropriate therapy, incurable stage 3 cancer could produce up to 23% 5-year survival. Although there is evidence that early chemotherapy improves survival for stage 4 cancers, the expected life-year gains are modest. 122,123

We used recent stage distributions and survival estimates for the UK124,125 as the basis for modelling post-presentation survival of the trial’s cancers, assuming that they had not been screen detected. We assumed that, were the patient to survive, a cancer which had been screen detected at stage S would have eventually presented at stage S or later (i.e. cancers would possibly progress but would never regress). For example, a stage 1 cancer could eventually present at any of the four stages according to the UK distribution, but a stage 3 cancer could present only at stages 3 or 4. Table 31 displays the survival rates assumed, based on the literature cited, and the implied values for the parameters for the life table modifications described above. It should be noted that survival rates for women in the absence of screening are superior to those of men, although the absence of gender-specific data required us to assume the same screen detection survival rates for both genders.

Lead time

The final assumption required for outcome estimation is the lead time of screening. Although some modellers have presumed that lung cancer develops sufficiently rapidly to produce lead times of around 2 years,126–128 those actually engaged in screening have been more circumspect. Development times of 6 and 8 years between stages 1 (if screen detected) and four (if presenting) have been suggested for Japan129 and ELCAP,130 respectively, and these figures set the upper limits on lead times. A Danish CT trial has posted lead times of 4–5 years for early-stage cancers. 44 Analysis of the US NLST data suggests that the majority of, but by no means all, cancers would present within 5 years. 131

A possible approach to estimating lead time in the case of UKLS subjects would be to compare the mean subject ages at detection of the cancers by stage and the ages of symptomatic presentation currently observed in the UK. 124 The integer age differences amount to 3 years, 2 years and 1 year, for stages 1–3, respectively. In other words, the stage 1 cancers in UKLS were detected, on average, at an age 3 years lower than cancers of stage 1 or later, which present symptomatically. Although it is reasonable to expect lead times to fall by stage at detection, shorter lead times per se raise expected gains from screening in the survival model. To err on the side of caution, therefore, we assumed, first, the detection of stage 4 cancers would secure no survival advantage: stage 4 lead times, as it were, would correspond to the modest gains in earlier initiation of treatment. Second, lead times for stages 1–3 would be 6 years, 4 years and 2 years, respectively. Doubling our previous estimate ensured consistency with the more conservative published opinions. We note, in passing, that incorporating lead time in a life table model corrects for overdiagnosis, as it allows subjects to die of other causes before lead time elapses.

As the survival gain for each individual cancer depends on stage, age and gender, each gain accrues at a different time. For example, the model calculates that a 70-year-old male with stage 1 cancer would have died at 78.1 years in the absence of screening but would, as a result of early detection, gain 4.1 further years of life beyond that age. A 70-year-old male with stage 3 cancer would gain an extra 0.6 years of life beyond the age at which he would have died without screening, in 4.3 years’ time. To enable summation of survival gains accruing at different times we discounted future life-year gains to present values at 3.5% annually.

Predicted outcomes

The model predicted total life-year gains of 137.2 (discounted 89.4) from detecting and treating the 42 cancers. This translated to an average gain of 3.3 (95% CI 2.6 to 3.9) life-years per cancer, undiscounted, and 2.1 (95% CI 1.7 to 2.5) life-years, discounted. The average gain per person screened was 0.07 (discounted 0.05) life-years [i.e. 25.1 (16.4) life-days]. Most of the total life-year gains from screening (86% discounted) accrued as a result of the early detection and treatment of stage 1 cancers. It follows that model outcomes are especially sensitive to the parameters that are associated with stage 1 survival. Were we to assume a 7-year lead time for stage 1 cancers instead of a 6-year lead time, total gains would be 111.8 (discounted 73.3) life-years and the average gain per cancer would be 2.7 (95% CI 2.1 to 3.2) life-years per cancer, undiscounted, and 1.7 (95% CI 1.4 to 2.1) life-years, discounted. The assumption of a 5-year lead time increases total gains to 151.8 (discounted 98.7) life-years, and average gains to 3.6 (95% CI 2.8 to 4.4) life-years per cancer, undiscounted, and 2.4 (95% CI 1.9 to 2.8) life-years, discounted.

These predicted survival gains appear similar to that of a US state-transition simulation model,132 which used national epidemiological and clinical data. 133 The estimated survival gain from a single screen was 0.01–0.04 life-years per person screened, depending upon the assumptions made. 134 A more recent synthesis of five independent US models, again founded on the most recent clinical evidence, yielded predicted gains of between 0.02 and 0.09 life-years per person screened, depending on assumptions relating to screening frequency and subjects’ smoking history. 135

Quality-adjusted life-years

Although the HRQL of UKLS subjects was measured using the European Quality of Life-5 Dimensions instrument prior to screening, an equivalent assessment post screening was precluded by the time constraint. In order to translate life-year gains into the more commonly quoted QALY gains, we followed the practice of other recent economic evaluations (see below). Most of the gains from a screening programme accrue to those treated successfully for early-stage cancers, with the result that the HRQL pre and post screening must be essentially similar to population norms. For the UK, HRQL norms for the genders/ages at which simulated deaths from cancer would have occurred lie in the range 0.71–0.78 relative to perfect health. 137 Adjusting life-year gains for each cancer detected by the HRQL coefficient for the patient’s expected age at death transforms the predicted total gain of 89.4 life-years into 66.8 QALYs, or 0.03 QALYs per person screened. The mean discounted gain per cancer detected was 1.6 QALYs. At 2.2 QALYs, expected mean gains for stage 1 cancers were considerably higher than QALY gains from detecting stage 2 cancers (0.6) and cancers at more advanced stages (0.2). With outcomes defined in terms of QALYs as opposed to life-years, the ICER equalled £8466 per QALY gained (95% CI £5516 to £12,634).

A modified protocol

The vast majority of cancers detected during the UKLS came via the category 3 or category 4 nodule management routes. Around 24% of trial subjects followed the category 2 route after their initial CT scan, yet that route yielded only one cancer. The question arises whether or not including the category 2 classification was, in itself, cost-effective. Our calculations thus far enable us to infer what would have happened had (1) only category 3 and 4 subjects proceeded to further investigation and treatment, and (2) both category 1 and 2 subjects immediately exited the trial. Under this revised scenario, the UKLS protocol would have been less costly, at the expense of one cancer remaining undetected.

Not rescreening the subjects classified as category 2, as the original protocol had required, would have saved £49,816 in CT scan costs. One fewer cancer detected would have avoided a MDT-directed work-up and treatment. Pro rata savings would amount to £1728 for work-up and £7917 for treatment. The single cancer would have presented symptomatically, at an expected future cost of £4508, discounted. Therefore, the net savings from excluding the category 2 arm would have been £54,953, making the net cost of the modified protocol £510,545. We make the conservative assumption that the cancer not detected would have been at stage 1. We have already estimated the expected gain from detecting a stage 1 cancer at 2.2 QALYs, discounted, making the outcome of the modified protocol 64.6 QALYs. It follows that the ICER of the modified protocol would be £7903 per QALY gained.

This revised ICER is marginally lower than the original baseline value, indicating that, as a result of the change in protocol, cost savings outweigh the reduction in outcome. The significance of this is best understood in reverse, by taking the modified protocol, rather than the original, as the reference case. The move from modified (no repeat CT scans for category 2 nodules) to original would entail the addition of £54,953 in costs for an outcome gain of 2.2 QALYs, implying an ICER of £24,978.

Comparison with other estimates

Until around 10 years ago, cost-effectiveness evaluations of CT screening consisted of technically sophisticated mathematical models, usually of the Markov stage transitions type. These models were based almost entirely on unsubstantiated assumptions, as was perhaps inevitable, given the paucity of clinical evidence at the time. The models were typically opaque, thereby frustrating attempts to discern reasons for differences in performance. It is probable, however, that differing assumptions (e.g. about sojourn time, lead time, interstage transition probabilities, the effect of screening on a subject’s cigarette smoking behaviour and the probabilities of harm owing to repeated screening) accounted for the unhelpfully wide range of ICERs predicted. 138

The publication of experimental data has enabled recent modellers to build on more robust foundations, yet the range of reported ICERs remains substantial. Consider just three examples. First, a US simulation;139 based on the state-transition model informed by national epidemiological and Mayo Clinic CT trial data132 estimated QALY gains from a single screen at 0.01–0.02 per person, producing an ICER in excess of £106,000 per QALY. The ICER for the annual screening of ever-smokers was at least £85,000 per QALY gained. Second, a US stage-shift model,140 using ELCAP protocols and outcomes data, Medicare tariffs and national survival rates by stage, produced baseline ICER estimates in the range £7900–17,560 (close to the UKLS estimate cited above). The model was re-estimated for annual screening and yielded baseline estimates of £19,060 or £31,800, depending on whether the model’s cancer stage predictions derived from the ELCAP- or the NLST-reported stage shift. 141 Finally, a study in Israel128 reported QALY gains of around 0.06 per person screened, along with an ICER of only £1005 per QALY gained.

Incremental cost-effectiveness ratios calculated from real data tend to be more transparent than those derived from mathematical models. Transparency facilitates comparison, and comparison reveals several likely explanations for variability in cost-effectiveness. To begin with, unit costs differ significantly by health-care system. Differences in unit cost enable the corresponding ICERs to differ, even when the same quantity of resources is being used to produce the same outcomes. Unit costs are substantially higher in the USA than they are in the UK and, for that matter, in virtually any other country. By way of example, our UKLS model included full investigation and surgical costs at < £10,000 per cancer (see Table 30), yet the US models included a cost of around £19,000 in the month of surgery142 or around £55,000 for diagnosing and treating localised cancer. 141 Although smaller than US costs, UK unit costs were significantly higher than those reported in the Israel study. For example, Table 30 costs for bronchoscopy and for lobectomy were, respectively, four and three times greater than their counterparts in Israel. Other things remaining equal, interventions appear the least (most) cost-effective in countries where the unit costs are highest (lowest).

The cost-effectiveness of screening is a function of disease prevalence. With a higher prevalence in the target population, more cases will be detected and more health gains will accrue for the same cost of screening. 52 In screening programmes and models, prevalence impacts in two ways. First, prevalence will be progressively raised by the gradual restriction of targets to individuals with higher and higher risk statuses. All things remaining equal, therefore, any programme involving a more stringent risk criterion for selection will record a greater yield of significant findings per person screened, and a lower ICER, than one with less strict criteria. Beyond the initial screen, costs and outcomes of screening depend on the positive predictive value of subject selection, and the different risk models used to identify targets in different studies have been shown to possess different positive predictive values. 143 Furthermore, it has been shown18 that a more accurate targeting of subjects in the NLST could have reduced considerably the number needed to be screened to prevent a cancer death, suggesting that the trial’s protocol was less efficient (and, implicitly, less cost-effective) than it might otherwise have been.

Second, the UKLS engaged a single screen, although other programmes, and their evaluations, have involved multiple screens. Unless disease incidence is especially high, the prevalence of disease in a population falls between the initial screen and any rescreen. A succession of screens will progressively increase total yield, although the increase at the margin would be expected to be lower than the increase in the costs of additional screening. This proposition can be illustrated by a simple simulation. In one of the US models,141 the proportion of positive results on rescreening was judged to be one-third of positives detected at the initial screen. Were we to assume similarly that a rescreen of UKLS subjects using the UKLS protocol would yield one-third of the outcomes detected in the actual screen, cancer management costs would be correspondingly lower, but screening costs would be the same as for the first screen. The ICER for the rescreen would be, at £19,096 per QALY gained, more than double the ICER for the single screen. Were the yield to be one-quarter of the original outcomes, the ICER would be even higher, at £24,173.

The USA’s long-running NLST has recently published an estimate of US$81,000 per QALY as its mean ICER. 144 This figure is nearly seven times greater than the corresponding estimate for the UKLS. Fortunately, the transparency of reporting makes cost-effectiveness comparisons straightforward, and Table 32 lists the principal similarities and differences between the two trials’ economic calculations. As is evident, all of the factors noted earlier (local unit costs, intensity of resource use, disease prevalence in the target population, multiple screens) contribute to an explanation of the difference in the ICERs.

Work-up of computed tomography-detected nodules

The UKLS incorporated nodule volumetric software as part of its protocol; this provides a more accurate and reliable measurement of lung nodule size and growth than diameter measurements. Despite this, however, a major implication of the findings from the UKLS is the need to lessen the burden of diagnostic work-up of indeterminate nodules. There were a total of 536 subjects (i.e. categories 3 and 4), with nodules requiring a repeat scan. Of the 64 category 4 nodules, 41 were found to have lung cancer. On examining the number of UKLS participants referred to the MDT clinic, the false-positive rate was 3.6% (114 – 42/1994 = 3.6), whereas the interval imaging rate for the category 3 nodules was 23.2% (472 – 9/1994).

As this work was undertaken within the context of a trial, we erred on the side of ‘over-reporting’ by incorporating category 2 nodules in the care pathway, and this has provided valuable data. However, in a UK national screening programme, one would consider a higher cut-off point for the classification of indeterminate nodules, most likely 100 mm³. The nodule management algorithms are likely to adopt a similar approach to that recently recommended by the British Thoracic Society for incidental detected nodules, although there may be a slight difference for a screening population. 145 Work-up of screen positives was meticulous, and was delivered according to strict protocol. However, sensitivity of the screen can be reliably estimated only on follow-up of screen negatives for subsequent cancer. This will be the subject of future research in the UKLS cohort.

Figure 17 shows an example of a possible pathway for management of nodules detected by LDCT in lung cancer screening; it is based on the successful NELSON37 and UKLS model, with modifications from more recent publications on size threshold and risk of malignancy.

FIGURE 17.

Pathway for management of indeterminate and positive findings detected by LDCT. GGN, ground glass nodules; PET, positron emission tomography. Notes: Volume measurements should be used in preference to diameter. Part solid module has both ground glass and solid components. GGNs have no solid component.

To date, the international lung cancer screening trials, including the UKLS, have not stipulated in their protocols the clinical work-up or the treatment interventions of CT-detected nodules that were referred to the MDT. The UKLS pilot was based at thoracic centres of excellence, where all of the clinical specialties were represented; multidisciplinary management of lung cancer is firmly established in the UK. The policies and pathways for managing suspected or proven lung cancer are summarised in the updated NICE146 and British Thoracic Society147 guidelines and the most recent British Thoracic Society guidelines on lung nodule management. 145 Subjects who enter a CT screening programme and have a possible cancer detected would have access to standard management based on the MDT.

Screening age range/duration

The US Preventive Services Task Force has recently recommended screening yearly from 55 to 79 years of age, based on their microsimulation modelling, for people with a smoking history of at least 30 pack-years who had smoked within 15 years. However, the trials did use this age range, and the trial evidence necessarily pertains to shorter periods of intervention. The approach taken in UKLS was to use an individual risk score to assess eligibility. The results of UKLS suggest that screening might reasonably be offered from ages 60 to 75 years. As so few people aged < 60 years were at high risk of lung cancer, the recruitment activity below this age range was not productive. Screening at 70–75 years might be expected to prevent deaths up to the age of 79 years.

Screen interval

A far more difficult problem will be deciding on the screen interval. An annual screen was used in the NLST but the cost will be higher than a biennial or longer interval. It is possible to calculate the increase in mortality as a consequence of moving from an annual to a biennial screen. This has been modelled in a recent publication from the UKLS group. Various estimates under different potential scenarios suggest that 20–40% more lives might be saved with annual screening rather than biennial (Figure 18). 148 Follow-up of the UKLS population, to assess the time elapsed until the lung cancer incidence and mortality in the intervention and control groups converge, will give valuable information in terms of a safe interval. We suggest a strictly applied protocol for screen interval, based on our cost-effectiveness modelling given above.

FIGURE 18.

Screening flow chart.

Selection of high-risk individuals

There is clearly an imperative to offer screening specifically to those individuals who are most likely to benefit from the intervention. Therefore, a robust risk score should be used to select people for screening and this can be done at any point of recruitment, provided a validated score is available. Some scores select on smoking and age at a population level (as used in the NLST),36 whereas others consider a combination of factors (such as previous respiratory disease, previous cancer, family history of lung cancer and exposure to asbestos) in addition to smoking duration and age (as used in the UKLS). 23,94 A number of these risk parameters are readily obtainable in a general practice setting. 149 The utilisation of a risk prediction model to refine the selection process can also be successfully undertaken by an agency such as Radar, who worked very successfully with the UKLS.

The UKLS data also provide an opportunity to develop a risk prediction model that is based on the patient’s clinical data, integrated with imaging characteristics and, in the future, validated biomarkers.

Response to initial invitation letter

Risk status

Participation of high-risk people in the screening trial