A number of risk prediction models have been developed in order to predict a person’s likelihood of developing lung cancer [35]. Use of such models provides the potential to target screening towards those at highest risk [36]. The majority of these risk models are based predominantly upon age and smoking; these include Bach [37], Spitz [38], LLP [39], Tammemagi [18], and Kovalchik [40].

However, the predictive accuracy of lung cancer risk models may be further improved by the addition of other epidemiological risk factors [41]. The Prostate Lung Colorectal and Ovarian (PLCO) cancer screening trial lung cancer risk model [42] was developed from the largest data set used to date in developing a lung cancer risk model. A revised version of this model has recently been applied to the NLST dataset and was able to select 81 additional persons for screening who received a diagnosis of lung cancer in follow-up that would have resulted in 12 fewer deaths [18].

Kovalchik et al. [40] calculated the number of lung-cancer deaths per 10,000 person-years that were prevented in the NLST CT-screening group, compared to the chest x ray group, and found that they increased according to the risk quintile (0.2 in quintile 1, 3.5 in quintile 2, 5.1 in quintile 3, 11.0 in quintile 4, and 12.0 in quintile 5. Sixty percent of the NLST participants at highest risk for lung-cancer death (quintiles 3 through 5) accounted for 88 % of the screening- prevented lung-cancer deaths, while the 20 % of participants at the lowest risk (quintile 1) accounted for only 1 % of prevented lung-cancer deaths. Thus, screening those with the highest risk prevented the greatest number of deaths from lung cancer.

In the UK, the Liverpool Lung Project (LLP) risk model has been developed from a large case-control study of the same name [39, 43]. The LLP risk prediction model incorporates age, sex, family history of lung cancer, smoking duration, personal history of other cancers and non-malignant respiratory diseases, and occupational exposure to asbestos [39, 43]. The LLP model is a robust algorithm that has been validated on two international case-control populations (Harvard and EUELC) and one independent cohort (LLP 7,500) [44]. The LLP risk model has distinctive strengths. Firstly, the predictor variables are all explicitly defined and can be readily assessed at the time of patient presentation, and, secondly, patients can be assigned to their appropriate risk class on the basis of information from the initial history alone. The utilisation of risk models, specifically the LLP risk model, has recently been highlighted by the National Cancer Institute (NCI) [45]. The LLP_v2 risk model [46] was utilised to select individuals in the UKLS trial; the first time a specific risk model was used in a RCT CT screening trial. Individuals with a 5 % absolute risk of developing lung cancer over a 5 year period were identified and recruited into the UKLS pilot screening trial. To date, the UKLS has already demonstrated 1.7 % prevalence of lung cancer at baseline, which is significantly higher than that seen in the NELSON and NLST trials [47].

There are harms and benefits associated with lung cancer screening and thus there is an obligation to select a population at sufficient risk in order to maximise the benefit/harm balance.

CT imaging is a highly sensitive test for lung cancer detection, however, it is non-specific. The NELSON and the UKLS trials employed a strategy to minimise the harmful effects of false positive tests by developing the ‘indeterminate’ CT screening identified nodule result. An indeterminate result is one where the probability that the finding represents malignancy, is sufficiently low enough to defer minimally invasive sampling at that time and a further CT scan is considered appropriate at either 3 or 12 months depending on the calculated nodule volume doubling time (VDT). Growth definition in volume should be discriminated from the definition in diameter. For example, a 25 % diameter growth from 8 to 10 mm reflects a volume growth from 268 mm [3] to 524 mm [3], almost a volume doubling (95 %). A 25 % volume growth from 80 mm [3] to 100 mm [3] reflects a diameter growth of 8 % from 5.35 to 5.78 mm, within the standard deviation of CT diameter measurement for this object range [48, 49].

In NLST, the cut off for nodules mandating further imaging or investigation was a diameter of 4 mm. This resulted in a very high number of false positive tests, although the vast majority of these participants were followed up with imaging techniques.

The NELSON screening interpretation is based on nodule volumetry. Nodules less than 50 mm [3], were classified as negative, greater than 500 mm [3] as positive; 50–500 mm [3] as indeterminate [27]. Indeterminate nodules underwent a 3-month follow-up LDCT for growth (VDT [50] were then used to distinguish between positive screens (VDT <400) requiring additional diagnostic procedures, and negative screens. NELSON reported a screen sensitivity of 94.6 % and a negative predictive value of 99.9 %. All malignant fast-growing lung nodules in the NELSON trial referred after the 3-month follow-up CT in the baseline lung cancer screening round had VDT of 232 days or less. From this, it seems that lowering the VDT cut-off may reduce false-positive referrals even further without impact on the lung cancer detection rate.

From the above, 3D volumetric measurements are likely to be superior to 2D diameter measurements in terms of accuracy and reproducibility, because the whole nodule is analysed. This is particularly informative for both size at a given time and growth when the nodule is non-spherical or if it grows asymmetrically.

The cut-off for slow-growing lung cancer varies, but most studies utilise a 400-day threshold proposed by the ELCAP investigators [51]. GGO’s cut-off VDT values are uncertain [52] but a mean VDT of 769 days has been reported [53].

It remains to be proven if the two approaches (diameter and volume) translate to differences in early lung cancer detection, morbidity, mortality, total radiation burden and costs. In addition, the side effects of the different protocols, such as recall CTs, invasive procedures, other diagnostic work-ups and referrals, should be taken into account when comparing the two approaches. In the near future, one would expect many of the international screening programmes to utilise volumetric analysis in their management of indeterminate nodules.

The USPSTF have recommended screening annually from 55 years of age to 80 years of age [17]. However, to date we only have evidence for 3 years of annual screening from the NLST trial. The longer term effect of annual screening for 25 years are unknown for mortality benefit, cost effectiveness and the psychosocial effects on the screenees. The recent data from the UKLS has indicated that the optimum age for screening would be 60–75 years based on the LLP_v2 risk profile [46].

Two trials have incorporated biennial screening, NELSON (final screen) and the MILD (Multicentric Italian Lung Detection) project [32]. Although no mortality benefit was shown in the MILD trial, there were more cancers detected in the annual group and this study does provide some insight into the effect of extending the screen interval on the detection of interval cancers. In the NELSON trial, which employed a cut-off of 50 mm [3] (4.6 mm diameter), the chance of detecting lung cancer on a CT scan after a baseline negative screen was 0.1 % at 1 year and 0.3 % at 2 years. In the same trial, the baseline cancer detection rate was 0.9 % and in the second annual screen the rate was 0.7 %. Thus extending the screening interval to 2 years might delay the diagnosis of a proportion of cancers, even where the threshold for further work up is as low as 50 mm [3].

There has to be a balance between the number of lives which can be saved and the cost of implementing a yearly screening programme. Cost effectiveness data have not yet been published by the NLST investigators but estimates based on modelling NLST data vary from $19,000 in one model to $126,000–$169,000/QALY in another [19, 20]. However, a judgement call may have to be made between providing an affordable biannual screening programme, which still saves lives or potentially no screening programme. The absolute impact of screening is very dependent on baseline risk as previously discussed in risk prediction modelling. Potentially, the screening interval and nodule workup threshold may be personalised to the individual risk of the screenees, in future programmes. This will require the development of the next generation of risk prediction models, which incorporates baseline CT characteristics based on volume doubling time (VDT).

It is possible to calculate the increase in mortality as a consequence of moving from an annual to a biennial screen. This has been modeled in a recent publication from the UKLS group [54]. Various estimates under different potential scenarios suggest that 20–40 % more lives might be saved with annual screening than biennial [54]. The outcomes suggest that biennial screening, while being less effective in absolute terms than annual screening, has the potential to be equally or even more cost-effective [54].

In lung cancer CT screening the benefits of CT screening outweigh the risks in high risk individuals. The risks which have been identified with lung cancer screening are: exposure to radiation and the consequences of identification of nodules that are benign, but may result in either unnecessary anxiety to the individual or an invasive procedure

The psychosocial and behavioral impact of lung cancer screening has been examined in controlled trials [55–59], which suggests, that adverse psychosocial consequences of lung screening do not persist at longer term follow-up and are unlikely to present a problem for the introduction of a screening programme.

Lung cancer CT screening trials provides a ‘teachable moment’ for smoking cessation, which would cascade into major beneficial health effects for all smoking related diseases. Smoking cessation is a cost effective intervention. In the NELSON trial, 14.5 % in the screened arm and 19.1 % in the control arm quit smoking compared with a background population quit rate of 6–7 % [57].