Lung Cancer 89 (2015) 1–3
Contents lists available at ScienceDirect
Lung Cancer journal homepage: www.elsevier.com/locate/lungcan
Editorial
Prediction of risk of lung cancer in populations and in pulmonary nodules: Significant progress to drive changes in paradigms a r t i c l e Keywords: Lung cancer Risk prediction Risk model Screening Early diagnosis Pulmonary nodule
i n f o
a b s t r a c t The ability to estimate the risk of lung cancer is important in three common clinical scenarios: the management of pulmonary nodules, the selection of people for screening with computed tomography and in the early identification of symptomatic disease. The risk prediction models that have been developed have similar themes owing to the strongest risk factors dominating the model. In the management of pulmonary nodules, there is a need to ensure that models reliably predict the chance of malignancy by performing validation studies in the population in which the models will be used. Two models stand out as the better ones in validation studies, one best used for smaller nodules and the other for larger ones. To maximise the cost effectiveness of screening with computed tomography, it is essential to select a population at high enough risk. A number of risk models have been developed, of varying complexity. Simpler models may be easier to use in practice but may miss a minority at high risk who have less common but important risk factors. Identification of early symptomatic lung cancer is important to improve early survival and reduce emergency presentations but single symptoms are non-specific. Risk prediction can improve the targeting of investigation and potentially identify patients early. Clinicians need to embrace the concept of estimating the risk of lung cancer in these three important areas because the evidence is strong enough to support a change in the clinical paradigm. © 2015 Elsevier Ireland Ltd. All rights reserved.
The ability to estimate the risk of lung cancer is important in three common clinical scenarios: the management of pulmonary nodules, the selection of people for screening with computed tomography and in the early identification of symptomatic disease. Risk prediction models have been developed that have similar themes owing to the dominant common risk factors of age and tobacco smoking. In this issue of Lung Cancer two papers add further to the knowledge base on risk prediction in pulmonary nodules and in selection for screening. 1. Predicting the risk of malignancy in pulmonary nodules The management of pulmonary nodules has always provided for a lively clinical debate. With the now widespread use of computed tomography (CT) this clinical problem is not only more common but further complicated by the fact that many nodules are small, with a lower risk of malignancy and therefore a more conservative approach has to be adopted. In case series describing incidentally detected nodules, an average of 15% of CTs detect nodules; in screening studies a quarter of CTs show nodules 4 mm in diameter or more. One of the key factors determining the optimal management of pulmonary nodules is the risk of them being malignant. Although opinion varies, it is generally agreed that as the risk of malignancy increases, the more aggressive should be the management. Recent ACCP guidelines suggest a surgical approach where the probability exceeds 65% in people with low to moderate surgical http://dx.doi.org/10.1016/j.lungcan.2015.05.004 0169-5002/© 2015 Elsevier Ireland Ltd. All rights reserved.
risk [1]. A number of studies have used case series of pulmonary nodules to develop risk prediction models [2–7]. The accuracy and clinical utility of predictive models depends on the case mix of the population in which it was derived and the prevalence of malignancy in that population. The applicability of the factors identified will depend on the methods used to identify the events (i.e. nodules) and the method of evaluation (in this case, CT). It is therefore essential to validate prediction models, if possible in a different clinical setting and ideally in a population similar to that in which the model is being used clinically. The paper by Al-Ameri et al., in this issue, has done just that [8]. The four models they validated were those of the Mayo Clinic [2], Veterans Administration (VA) [3], Herder et al. [9] and Brock University [7] (see Table 1). Previous external validation studies have shown area under the receiver operator curves (AUC) of between 0.79 and 0.80 for the Mayo Clinic model, and 0.73 for the VA model [9–12]. During a validation of the Mayo Clinic model, Herder et al. showed that by the addition of a 4-point ordinal scale of FDG uptake on PET, the AUC was increased to 0.92 [9]. These external validation studies showed that the models were not well calibrated for use in populations where the prevalence of malignancy is much lower or higher than that in which the model was developed [9–12]. The Brock University model was based on a screening population [7] so the prevalence of malignancy was only 5.5% in the development set. As would be expected, Al-Ameri et al. confirmed that the Brock model performed better for sub-centimetre nodules, with a lower
2
Editorial / Lung Cancer 89 (2015) 1–3
Table 1 Summary of studies that developed composite prediction models. Model
Brock University (full) [7]
Subjects
2961 (1871 development 1090 test)
Veterans Administration [3]
375
Mayo Clinic [2]
639
Herder et al. [9] 106
Study setting
Multi-centre screening study, Canada
Newly detected SPNs on CXR 7–30 mm. Multi-centre, USA Newly detected SPNs on CXR 4–30 mm, single centre, USA
Age mean (range) years
62 (50–75)
Male % Current/ former smokers %
53
100
Nodule size mean (range) mm
Benign
Malignant
4.1 (1–70)
15.7 (2–86)
Prevalence of malignancy %
4.8(1–19)
Development AUC (95% CI)
Al-Ameri validation AUC (95% CI)
0.97 0.93 for nodules
Contents lists available at ScienceDirect
Lung Cancer journal homepage: www.elsevier.com/locate/lungcan
Editorial
Prediction of risk of lung cancer in populations and in pulmonary nodules: Significant progress to drive changes in paradigms a r t i c l e Keywords: Lung cancer Risk prediction Risk model Screening Early diagnosis Pulmonary nodule
i n f o
a b s t r a c t The ability to estimate the risk of lung cancer is important in three common clinical scenarios: the management of pulmonary nodules, the selection of people for screening with computed tomography and in the early identification of symptomatic disease. The risk prediction models that have been developed have similar themes owing to the strongest risk factors dominating the model. In the management of pulmonary nodules, there is a need to ensure that models reliably predict the chance of malignancy by performing validation studies in the population in which the models will be used. Two models stand out as the better ones in validation studies, one best used for smaller nodules and the other for larger ones. To maximise the cost effectiveness of screening with computed tomography, it is essential to select a population at high enough risk. A number of risk models have been developed, of varying complexity. Simpler models may be easier to use in practice but may miss a minority at high risk who have less common but important risk factors. Identification of early symptomatic lung cancer is important to improve early survival and reduce emergency presentations but single symptoms are non-specific. Risk prediction can improve the targeting of investigation and potentially identify patients early. Clinicians need to embrace the concept of estimating the risk of lung cancer in these three important areas because the evidence is strong enough to support a change in the clinical paradigm. © 2015 Elsevier Ireland Ltd. All rights reserved.
The ability to estimate the risk of lung cancer is important in three common clinical scenarios: the management of pulmonary nodules, the selection of people for screening with computed tomography and in the early identification of symptomatic disease. Risk prediction models have been developed that have similar themes owing to the dominant common risk factors of age and tobacco smoking. In this issue of Lung Cancer two papers add further to the knowledge base on risk prediction in pulmonary nodules and in selection for screening. 1. Predicting the risk of malignancy in pulmonary nodules The management of pulmonary nodules has always provided for a lively clinical debate. With the now widespread use of computed tomography (CT) this clinical problem is not only more common but further complicated by the fact that many nodules are small, with a lower risk of malignancy and therefore a more conservative approach has to be adopted. In case series describing incidentally detected nodules, an average of 15% of CTs detect nodules; in screening studies a quarter of CTs show nodules 4 mm in diameter or more. One of the key factors determining the optimal management of pulmonary nodules is the risk of them being malignant. Although opinion varies, it is generally agreed that as the risk of malignancy increases, the more aggressive should be the management. Recent ACCP guidelines suggest a surgical approach where the probability exceeds 65% in people with low to moderate surgical http://dx.doi.org/10.1016/j.lungcan.2015.05.004 0169-5002/© 2015 Elsevier Ireland Ltd. All rights reserved.
risk [1]. A number of studies have used case series of pulmonary nodules to develop risk prediction models [2–7]. The accuracy and clinical utility of predictive models depends on the case mix of the population in which it was derived and the prevalence of malignancy in that population. The applicability of the factors identified will depend on the methods used to identify the events (i.e. nodules) and the method of evaluation (in this case, CT). It is therefore essential to validate prediction models, if possible in a different clinical setting and ideally in a population similar to that in which the model is being used clinically. The paper by Al-Ameri et al., in this issue, has done just that [8]. The four models they validated were those of the Mayo Clinic [2], Veterans Administration (VA) [3], Herder et al. [9] and Brock University [7] (see Table 1). Previous external validation studies have shown area under the receiver operator curves (AUC) of between 0.79 and 0.80 for the Mayo Clinic model, and 0.73 for the VA model [9–12]. During a validation of the Mayo Clinic model, Herder et al. showed that by the addition of a 4-point ordinal scale of FDG uptake on PET, the AUC was increased to 0.92 [9]. These external validation studies showed that the models were not well calibrated for use in populations where the prevalence of malignancy is much lower or higher than that in which the model was developed [9–12]. The Brock University model was based on a screening population [7] so the prevalence of malignancy was only 5.5% in the development set. As would be expected, Al-Ameri et al. confirmed that the Brock model performed better for sub-centimetre nodules, with a lower
2
Editorial / Lung Cancer 89 (2015) 1–3
Table 1 Summary of studies that developed composite prediction models. Model
Brock University (full) [7]
Subjects
2961 (1871 development 1090 test)
Veterans Administration [3]
375
Mayo Clinic [2]
639
Herder et al. [9] 106
Study setting
Multi-centre screening study, Canada
Newly detected SPNs on CXR 7–30 mm. Multi-centre, USA Newly detected SPNs on CXR 4–30 mm, single centre, USA
Age mean (range) years
62 (50–75)
Male % Current/ former smokers %
53
100
Nodule size mean (range) mm
Benign
Malignant
4.1 (1–70)
15.7 (2–86)
Prevalence of malignancy %
4.8(1–19)
Development AUC (95% CI)
Al-Ameri validation AUC (95% CI)
0.97 0.93 for nodules
