The advisability of population screening for lung cancer is currently a topic of active debate. Because of the potential high cost of lung cancer screening programmes, cost-effectiveness questions are an important component to be evaluated. This article attempts to inform the debate by a multivariable analysis model. We incorporated a range of optimistic and pessimistic input variables taken from literature sources on both sides of this debate. In addition, this analysis presents an important method for evaluating a lung cancer-screening programme in contrast to analyses using QALYs, which can be misleading in terms of the efficacy of treatment regimens, particularly when there is little chance of long-term survival, as in the treatment of stage IIIB and IV NSCLC with chemotherapy. Owing to the frequent use of the QALY metric in the literature, we have provided for the reader a cost-per-year of life calculation for all 729 scenarios in the Appendix Table A3 and Table A4 to allow a rough comparison to other studies evaluating cost-effectiveness on a per-year basis. It should be emphasised, however, that the data in Table A3 and Table A4 is calculated assuming that these lung cancer survivors live only 5 years. It is therefore a conservative estimate and overstates the cost-per-year of life, as many lung cancer survivors detected by screening will live longer than 5 years (Marcus et al, 2000
; Sobue et al, 2002
; Henschke et al, 2006a
The combination of decreased CP5YS and increased 5-year survivors in all 729 scenarios for the cohort ages 60–79 years and in the majority of scenarios for the cohort ages 40–79 years suggests that health-care dollars spent on proactively screening for lung cancer may achieve higher cost-effectiveness. These data may also suggest advantages in focusing a screening programme on patients aged 60 years or older as the marginal costs for implementing screening in the cohort ages 60–79 years were between 46 and 60% of the marginal cost for the cohort ages 40–79 years, with greater cost effectiveness in the 60–79 age group when compared with the STID method (see ). These findings are consistent with a 2008 study by Whynes (2008
), analysing the potential cost effectiveness of lung cancer screening in the United Kingdom.
The I-ELCAP study results have shown the ability of CT scanning to detect asymptomatic lung cancers. By definition, a lung cancer programme that detects symptomatic and asymptomatic malignancies will diagnose more tumours at the onset of the programme than a system that detects symptomatic tumours alone, as illustrated in . The magnitude of this increase is unknown, and sufficient information to evaluate the reasonableness of the 24–66% increase in diagnosed tumours under the EDCTS method presented in this model is currently unavailable. Additional long-term data on the prevalence of lung cancer in an asymptomatic at-risk population annually screened over multiple years are needed to enable further estimates in this area.
Figure 5 Illustration demonstrating the theory for an increase in the number of tumours identified by screeingfor asymptomatic malignancies in addition to identifying symptomatic tumours. ‘Tumour crop' refers to a cohort of lung cancers that clinically (more ...)
Increased total costs associated with a successful proactive screening programme are to be expected, not only due to the cost of screening procedures, but also more importantly due to the increased patient survival and related increase in long-term treatment costs and follow-up care. The available data for cost of treatment utilised in this study (see , inputs M, N, and O) all indicate that the treatment for early stage cancer is more expensive than cancers detected in stages IIIB and IV. This is counter-intuitive, as stage I NSCLC is treated by surgery alone, whereas advanced stage lung cancers require multimodality treatment, adding the cost of expensive radiation and chemotherapy. Furthermore, treatment of advanced stage lung cancer is changing, which may potentially shift this balance, making the treatment of late-stage cancers increasingly more expensive. The percentage of patients in higher stages who are receiving expensive radiation therapy and chemotherapy appears to be increasing (Ramsey et al, 2004
; Langer et al, 2005
). In addition, new molecular ‘targeted' medications used in second- and third-line treatment may markedly increase treatment costs (Adis International Ltd. Erlotinib, 2003
). As I-ELCAP data report increased actuarial 10-year survival of 80% with screen-detected tumours (Henschke et al, 2006a
), which is dramatically different than the survival of symptom-detected tumours which are disproportionately detected in later stages, it would follow that increased costs of long-term care would be expected. The cost of follow-up and care in the majority of survivors, however, adds little to routine health-care expenditures, and only a small minority would require downstream salvage or palliative radiation therapy and chemotherapy. In addition, the cost of screening programmes and treatment may reasonably be expected to decrease as a result of economies of scale, if lung cancer screening were implemented at a state or nationwide level.
It is important to consider several potential sources of bias when evaluating lung cancer screening:
(1) ‘Overdiagnosis bias
': small, slow-growing lesions are detected by screening for intervention that would never become symptomatic within a patient's lifetime in the absence of screening (Black et al, 2006
). This could be caused by an improper pathological diagnosis; however, all lung cancers in the I-ELCAP study are vetted by a panel of prominent international pathologists (Flieder et al, 2006
). With regard to the theoretical possibility of screening the detection of very slow growing malignant neoplasms that do not cause symptoms during the patient's anticipated normal lifespan, for ethical reasons, a randomised trial comparing surgery with no surgery for stage I NSCLC is not possible. However, data on untreated screen-detected NSCLC from screening studies including the Johns Hopkins study, the Memorial Sloan-Kettering study, the Mayo Clinic study, and the I-ELCAP study indicate that almost all of these untreated patients die within 5 years (Flehinger et al, 1992
; Henschke et al, 2006a
). A study by Raz et al (2007)
, examined the natural history of patients with stage I NSCLC and concluded that long-term survival with untreated stage I NSCLC is uncommon with an overall survival of 6% for untreated stage I NSCLC. In addition, a 2003 study by Henschke et al
analysed data from 885 cases of stage IA lung cancer and concluded that almost all diagnosed cases of Stage IA lung cancer have a malignant natural course, fatal if not treated, thus representing genuine cancer. Although some speculate that the explanation for the paradoxical result in the Mayo Lung Trial of increased survival but no reduction in mortality with chest radiograph screening was due to, in part, overdiagnosis bias (Patz, 2006
), no empirical evidence to support this theory exists. Furthermore, Yankelevitz et al, (2003
) analysed the doubling time of stage I tumours in the Mayo Lung Trial and concluded that ‘the hypothesis that early-stage lung tumours diagnosed on chest radiography during lung carcinoma screening may frequently be overdiagnosed, indolent cases needs to be rejected'.
Another form of overdiagnosis could exist in instances where comorbid disease would kill the patient before symptoms of lung cancer were experienced. This form of overdiagnosis bias can be reasonably assumed to be rendered largely irrelevant in medical environments, where patients are managed with good, sensible clinical judgment by physicians adept in identifying comorbid disease and reasonably accurately estimating the anticipated survival of their patients. Furthermore, in this study we have carefully taken into account the percentage of patients eligible for screening who have serious comorbid disease and excluded them from our analysis (refer to the ‘Population Screened' section of the Materials and Methods above). We have been conservative in this approach by additionally excluding patients aged 80 years and above from screening, although many 80-year-olds are perfectly capable of undergoing minimally invasive surgical resection of lung cancers or being treated with radiationtherapy.
Finally, overdiagnosis could result if spontaneous remission of a preclinical cancer were to occur. Case reports of this phenomenon are extremely rare (Kappauf et al, 1997
(2) ‘Length bias
': detection of more patients with less aggressive disease and fewer of those with more aggressive disease, because the duration of asymptomatic disease is longer in less aggressive tumours (Black et al, 2006
). This also could result in ‘overdiagnosis', discussed above. The baseline round of screening is inherently different from the repeat rounds because cancers with a longer latent (asymptomatic) phase are more frequently identified in the baseline round, whereas cancers found in repeat rounds are found earlier in their latent phase than in the baseline round (Morrison, 1982
; Henschke and Yankelevitz, 2008
). Cancers that are diagnosed at baseline, thus, tend to grow more slowly than does the subtype in general; they also grow more slowly than do tumours that are diagnosed in repeated screenings. As noted by the I-ELCAP researchers, this fact does not introduce a bias, but it may call for making a distinction between baseline screening and repeated screening (Henschke et al, 2007
). ‘Length bias' also implies that the faster growing tumours may present symptomatically between screening exams, however, the I-ELCAP data showed a very low incidence of such cases.
(3) ‘Lead-time bias
': screening-detected patients are accorded extended survival times solely because cancer was detected earlier owing to screening, although death occurred at the same time as would have happened without screening (Black et al, 2006
). To address this potential source of bias, the estimated percentage of stage I 5-year survivors for all scenarios using input variable ‘Q' (see ) are based on 10-year survival percentages reported by I-ELCAP (Henschke et al, 2006a
). If lead-time bias was evident in this study, a large number of individuals who survived 5 years would be expected to die before 10 years; the I-ELCAP survival curve shows no decrement in survival between 5 and 10 years (Henschke et al, 2006a
Although some emphasise that the risks and complications of lung cancer screening may be considerable (Bach et al, 2007
), the minutes of National Cancer Institute's National Lung Screen Trial (NLST) Data Safety Monitoring Board (DSMB), which we have reviewed, indicate that during the first 5 years of the study, which began in 2002, no unanticipated complications or risks have been recognised. The DSMB, which is responsible for the safety of NLST research subjects, has neither terminated the study nor reported information on new complications to study subjects.
Input variables based on SEER data may not be representative of actual nationwide lung cancer incidence or staging. Our model does not account for costs associated with complications from biopsies or other screening procedures, nor does it account for the increase in capital equipment and resources necessary to implement large-scale comprehensive CT screening programmes. In addition, our study design does not consider the indirect cost of lost productivity attributable to lung cancer morbidity and mortality. Of the estimated $167 billion costs of all diseases caused by tobacco products, indirect costs ($92 billion) are substantially higher than direct costs (Centers for Disease Control and Prevention, 2005
). As survival increases, indirect costs attributable to loss of patient income, spousal income, and other factors may reasonably be expected to diminish substantially. This study is based on a representative subset of the US population and it is beyond the scope of this analysis to extrapolate the results to a state or nationwide level. The model does not factor in any additional benefit conferred by survival beyond 5 years. On the basis of the results from the Mayo Lung Trial (Marcus et al, 2000
), Japanese Anti-Lung Cancer Association (ALCA) trial (Sobue et al, 2002
), and I-ELCAP (Henschke et al, 2006a
), there is a strong evidence to suggest that the majority of 5-year survivors will continue to survive for 10, 15, and even 20 years following diagnosis and treatment. Finally, there is no accurate method to calculate a dollar value for not dying of lung cancer in an individual or group of individuals with lung cancer.