JNCI J Natl Cancer Inst (2015) 107(6): djv063 doi:10.1093/jnci/djv063 First published online March 20, 2015 Brief Communication
brief communication
miR-Test: A Blood Test for Lung Cancer Early Detection Francesca Montani*, Matteo Jacopo Marzi*, Fabio Dezi, Elisa Dama, Rose Mary Carletti, Giuseppina Bonizzi, Raffaella Bertolotti, Massimo Bellomi, Cristiano Rampinelli, Patrick Maisonneuve, Lorenzo Spaggiari, Giulia Veronesi, Francesco Nicassio, Pier Paolo Di Fiore†, Fabrizio Bianchi†. Affiliations of authors: Molecular Medicine Program, Department of Experimental Oncology, European Institute of Oncology, Milan, Italy (FM, FD, ED, RMC, GB, PPDF, FB); Center for Genomic Science of IIT@SEMM, Istituto Italiano di Tecnologia, 20139 Milan, Italy (MJM, FN); Division of Epidemiology and Biostatistics, European Institute of Oncology, Milan, Italy (ED, PM); IFOM, The FIRC Institute for Molecular Oncology Foundation, Milan, Italy (RMC, PPDF); Division of Thoracic Surgery, European Institute of Oncology, Milan, Italy (RB, LS, GV); Division of Radiology, European Institute of Oncology, Milan, Italy (MB, CR); Department of Scienze della Salute, University of Milan, Milan, Italy (MB, LS, PPDF). * Authors contributed equally to this work. † Authors contributed equally to this work. Correspondence to: Pier Paolo Di Fiore, MD, PhD, European Institute of Oncology, Department of Experimental Oncology, Via Ripamonti, 435, 20141 Milan, Italy (e-mail:
[email protected]) or Fabrizio Bianchi, PhD, European Institute of Oncology, Department of Experimental Oncology, Via Ripamonti, 435, 20141 Milan, Italy (e-mail:
[email protected]).
Abstract
Single-arm and randomized studies have shown that low-dose computed tomography (LDCT) allows early detection of lung cancer in high-risk individuals and reduces disease-related mortality (1,2). Its cost-effectiveness has also been recently described (2–5). In principle, the refinement of the preselection criteria based on additional risk factors, such as blood tumor markers, might promote the implementation of widespread LDCT screening and improve its cost-effectiveness. MicroRNAs (miRNAs), short noncoding RNAs involved in cellular regulation (6,7), represent promising blood-borne tumor markers. The
expression of miRNAs is often deregulated in tumors, leading to alterations in miRNA profiles in bodily fluids, including serum and plasma (8–13). Thus, the detection of circulating miRNAs might be useful for lung cancer early detection, as recently suggested (14). We designed a multi-tiered study to validate a blood test based on serum miRNAs (12) in high-risk individuals (heavy smokers, older than age 50 years) enrolled in the Continuous Observation of Smoking Subjects (COSMOS) trial (15) and lung cancer patients diagnosed outside of the screening (Figure 1A).
Received: August 14, 2014; Revised: December 14, 2014; Accepted: February 19, 2015 © The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail:
[email protected].
1 of 5
brief communication
Lung cancer is the leading cause of cancer death worldwide. Low-dose computed tomography screening (LDCT) was recently shown to anticipate the time of diagnosis, thus reducing lung cancer mortality. However, concerns persist about the feasibility and costs of large-scale LDCT programs. Such concerns may be addressed by clearly defining the target “highrisk” population that needs to be screened by LDCT. We recently identified a serum microRNA signature (the miR-Test) that could identify the optimal target population. Here, we performed a large-scale validation study of the miR-Test in highrisk individuals (n = 1115) enrolled in the Continuous Observation of Smoking Subjects (COSMOS) lung cancer screening program. The overall accuracy, sensitivity, and specificity of the miR-Test are 74.9% (95% confidence interval [CI] = 72.2% to 77.6%), 77.8% (95% CI = 64.2% to 91.4%), and 74.8% (95% CI = 72.1% to 77.5%), respectively; the area under the curve is 0.85 (95% CI = 0.78 to 0.92). These results argue that the miR-Test might represent a useful tool for lung cancer screening in highrisk individuals.
2 of 5 | JNCI J Natl Cancer Inst, 2015, Vol. 107, No. 6
Informed consent was obtained from all participants after institutional review board approval of the study. Details on blood collection, serum analysis, expression profiling, and other methods are described in the Supplementary Methods (available online). Initially, we refined our original 34-miRNA signature (12) using a “Calibration set” of 24 subjects (Table 1; Supplementary Tables 1 and 2, available online). This refinement allowed a reduction of the signature to 13 miRNAs (henceforth, the miRTest: miR-92a-3p, miR-30b-5p, miR-191-5p, miR-484, miR-328-3p, miR-30c-5p, miR-374a-5p, let-7d-5p, miR-331-3p, miR-29a-3p, miR-148a-3p, miR-223-3p, miR-140-5p), which maintained the same performance as the original signature (Corr. ≥ 0.96) (Supplementary Figure 1, available online). The refinement of the signature reduces the costs and complexity of the test and increases its clinical translatability.
The miR-Test was then validated in an independent “Validation Set” of 1008 subjects enrolled in the COSMOS trial (Figure 1A; Supplementary Table 1, available online). Overall, risk scores between this set of lung cancer patients and healthy subjects were statistically significantly different (P < .001), as we observed in the Calibration Set (Figure 1B and Table 1; Supplementary Table 2, available online). In the Validation Set, the test displayed an area under the curve (AUC) of 0.85 (95% CI = 0.78 to 0.92) and an accuracy (ACC), sensitivity (SE) and specificity (SP) of 74.9% (95% CI = 72.2% to 77.6%), 77.8% (95% CI = 64.2% to 91.4%), and 74.8% (95% CI = 72.1% to 77.5%), respectively (Figure 1C and Table 1). Next, we assessed the ability of the miR-Test to distinguish between nonmalignant lung diseases (NMD) and lung cancer in an independent cohort of individuals with chronic obstructive pulmonary disease (COPD),
Table 1. Performance of the miR-Test in various cohorts Risk score Cohort
brief communication
Calibration Set Lung cancer Stage I Stage II-III Lung cancer deaths No lung cancer Validation Set Lung cancer Stage I Stage II-III Lung cancer deaths No lung cancer PN‡ COPD Specificity Set No lung cancer§ PN COPD NOD Benign All (CT screening) Lung cancer¶ Stage I Stage II-III Lung cancer deaths No lung cancer PN COPD NOD Benign Clinical Set (outside of screening) Lung cancer Stage I Stage II-III Lung cancer deaths
No.
miR-Test
AUC (95% CI)†
median (Q1;Q3)*
Pos (%)
Neg (%)
12 11 1 1 12
11.6 (4.0;17.8) 10.9 (3.7;17.9) 14.9 14.9 −11.0 (−14.8;−6.8)
10 (83.3) 9 (81.8) 1 (100.0) 1 (100.0) 1 (8.3)
2 (16.7) 2 (18.2) 11 (91.7)
0.98 (0.95 to 1)
36 31 5 3 972 22 59
13.1 (0.3;20.6) 12.2 (0.5;20.3) 20.0 (−2.3;24.6) 24.6 (0.1;60.5) −6.0 (−11.7;0.2) −6.7 (−11.7;−2.5) −2.5 (−8.5;6.2)
28 (77.8) 25 (80.6) 3 (60.0) 3 (100.0) 245 (25.2) 3 (13.6) 24 (40.7)
8 (22.2) 6 (19.4) 2 (40.0) 727 (74.8) 19 (86.4) 35 (59.3)
0.85 (0.78 to 0.92)
83 24 16 38 5
−6.9 (−11.8;−1.6) −4.6 (−13.5;0.4) −11.0 (−14.8;−4.7) −7.3 (−10.7;−2.9) −1.4 (−1.6;−1.1)
11 (13.3) 6 (25.0) 1 (6.2) 4 (10.5) -
72 (86.7) 18 (75.0) 15 (93.8) 34 (89.5) 5 (100.0)
0.89|| (0.82 to 0.96)
48 42 6 4 1067 46 75 38 5
12.2 (0.9;20.2) 12.2 (1.2;19.9) 17.4 (−2.3;24.6) 19.8 (7.5;42.6) −6.1 (−11.7;−0.4) −5.4 (−11.8;−1.4) −4.0 (−11.4;4.9) −7.3 (−10.7;−2.9) −1.4 (−1.6;−1.1)
38 (79.2) 34 (81.0) 4 (66.7) 4 (100.0) 257 (24.1) 9 (19.6) 25 (33.3) 4 (10.5) -
10 (20.8) 8 (19.0) 2 (33.3) 810 (75.9) 37 (80.4) 50 (66.7) 34 (89.5) 5 (100.0)
0.87 (0.82 to 0.92)
7.8 (−1.9;16.8) 3.6 (−1.9;12.8) 14.5 (−1.6;21.2) 18.4 (4.0;25.7)
52 (70.3) 29 (69.0) 23 (71.9) 10 (83.3)
22 (29.7) 13 (31.0) 9 (28.1) 2 (16.7)
74 42 32 12
NA
* The median risk score with first and third quartile in parentheses (Q1;Q3) are indicated. AUC = area under the curve; CI = confidence interval; COPD = patients with chronic obstructive pulmonary disease; CT = computed tomography; NOD = patients with low-dose computed tomography–detected, noncalcified, lung nodules stable in size at more than five years of follow-up; PN = patients with pneumonia. † The 95% confidence interval is reported in the parentheses.. ‡ Two individuals with both PN and COPD. § Individuals without lung cancer with more than five years of screening follow-up. || AUC was calculated using 83 individuals from the specificity set without lung cancer and 36 individuals from the validation set with lung cancer. ¶ Tumor stage according to the TNM Classification of Malignant Tumors, 7th Edition (UICC).
F. Montani et al. | 3 of 5
A
N=1115
N=74
COSMOS screening trial (asymptomatic cohort)
IEO thoracic surgery
Calibration Set N12/T12
P