Vascular endothelial growth factor C complements the ability of positron emission tomography to predict nodal disease in lung cancer

Abstract

Objective

Vascular endothelial growth factors (VEGFs) C and D are biologically rational markers of nodal disease that could improve the accuracy of lung cancer staging. We hypothesized that these biomarkers would improve the ability of positron emission tomography (PET) to predict nodal disease among patients with suspected or confirmed non–small cell lung cancer (NSCLC).

Methods

A cross-sectional study (2010–2013) was performed of patients prospectively enrolled in a lung nodule biorepository, staged by computed tomography (CT) and PET, and who underwent pathologic nodal evaluation. Enzyme-linked immunosorbent assay was used to measure biomarker levels in plasma from blood drawn before anesthesia. Likelihood ratio testing was used to compare the following logistic regression prediction models: Model_PET, Model_PET/VEGF-C, Model_PET/VEGF-D, and Model_{PET/VEGF-C/VEGF-D}. To account for 5 planned pairwise comparisons, P values<.01 were considered significant.

Results

Among 62 patients (median age, 67 years; 48% men; 87% white; and 84% NSCLC), 58% had fluorodeoxyglucose uptake in hilar and/or mediastinal lymph nodes. The prevalence of pathologically confirmed lymph node metastases was 40%. Comparisons of prediction models revealed the following: Model_PET/VEGF-C versus Model_PET (P = .0069), Model_PET/VEGF-D versus Model_PET (P = .1886), Model_{PET/VEGF-C/VEGF-D} versus Model_PET (P = .0146), Model_{PET/VEGF-C/VEGF-D} versus Model_PET/VEGF-C (P = .2818), and Model_{PET/VEGF-C/VEGF-D} versus Model_PET/VEGF-D (P = .0095). In Model_PET/VEGF-C, higher VEGF-C levels were associated with an increased risk of nodal disease (odds ratio, 2.96; 95% confidence interval, 1.26–6.90).

Conclusions

Plasma levels of VEGF-C complemented the ability of PET to predict nodal disease among patients with suspected or confirmed NSCLC. VEGF-D did not improve prediction.

Nodal staging is a key determinant of appropriate treatment selection and therefore outcomes among patients with nonmetastatic non–small cell lung cancer (NSCLC). Practice guidelines recommend the routine use of computed tomography (CT) and positron emission tomography (PET) to initially stage patients with suspected or confirmed NSCLC.^1,2 The diagnostic limitations of noninvasive imaging tests have been well characterized and have led to dependence on invasive staging modalities such as mediastinoscopy and endobronchial ultrasound-guided biopsy.¹ In the absence of novel imaging modalities, 2 ways to improve lung cancer staging is through better use of the existing radiographic information and adoption of new risk factors for nodal disease. Prediction models may make better use of imaging information leading to higher accuracy and therefore better use of invasive staging procedures. Biomarkers measured in plasma and correlating with nodal disease may complement radiographic predictors of nodal involvement.

Advances in lymphangiogenesis research show that epithelial tumors—including NSCLC—metastasize to lymph nodes via the formation of de novo peritumoral lymphatic channels.³ Experimental animal models demonstrate that expression of vascular endothelial growth factors (VEGF) C and D by tumor cells promote nodal metastases, and blocking the VEGF-C/VEGF-D receptor eliminates lymphatic metastases.^3–7 Studies in humans reveal higher tissue (both in primary tumor and lymph nodes) and plasma levels of VEGF-C and VEGF-D among node-positive NSCLC patients compared with node-negative patients, individuals with benign nodules, and healthy volunteers.^8–14 Markers of lymphangiogenesis appear to be biologically rational predictors of nodal disease.

The ultimate goal of our research is to develop and validate a prediction model based on multiple radiographic and molecular predictors of nodal disease. As a first step toward this goal, we sought to determine whether plasma levels of VEGF-C and/or VEGF-D improve the ability of PET to predict nodal disease in patients with suspected or confirmed nonmetastatic NSCLC. Five randomized trials show that PET improves the diagnostic accuracy of nodal staging.^15–19 A biomarker would ideally add diagnostic value to this dominant predictor of nodal disease. We hypothesized that VEGF-C and/or VEGF-D would improve the ability of PET to predict nodal disease.

MATERIALS AND METHODS

Patients and Study Design

The source of patient information for this study was a lung nodule biorepository maintained by the Fred Hutchinson Cancer Research Center. Since 2008, all patients referred to the Seattle Cancer Care Alliance Lung Cancer Early Detection and Prevention Clinic have been approached to be prospectively enrolled in the biorepository. The Fred Hutchinson Cancer Research Center institutional review board approved this biorepository (file #6663, protocol #2242). Patients are required to provide written consent. The biorepository is strictly observational. It was designed for multiple research purposes and is not associated with a parent study involving interventions. Therefore, patient selection and treatment reflect usual care. Study variables are ascertained from patient records by trained abstractors using standard definitions. A board-certified thoracic surgeon resolved uncertainties over study eligibility and variable classification, and verified the accuracy of data collection for all patients included in the analysis.

A retrospective cross-sectional study was conducted of patients consecutively enrolled in the biorepository between October 2010 and December 2013. The start date was chosen based on the availability of plasma specimens preserved in ethylenediaminetetraacetic acid (EDTA). Patients eligible for study included adults with suspected or confirmed primary NSCLC who had been staged with CT and PET, were without evidence of metastatic disease, and had undergone endobronchial ultrasound guided biopsy, mediastinoscopy, and/or intraoperative nodal assessment at the time of pulmonary resection.

Plasma Sample Preparation and Storage

Blood samples were collected before the administration of anesthesia using a standardized protocol—temporary storage at 2°C to 8°C, centrifuging at 4°C for 10 minutes at 1300 × g, preservation using EDTA, and long-term storage of plasma specimen aliquots in a −80°C freezer within 4 hours of blood draw.

Biomarker Measurement

Plasma levels of VEGF-C and VEGF-D were measured using a commercially available enzyme-linked immunosorbent assay (R&D Systems, Minneapolis, Minn). The manufacturer reported sensitivity limit and range for VEGF-C was 48.4 pg/mL and 109 to 7000 pg/mL, respectively, and the sensitivity limit and range for VEGF-D was 31.3 pg/mL and 125 to 4000 pg/mL. Assay performance was validated in serum, heparinized plasma, saliva, cell culture supernates, EDTA-preserved plasma, and platelet-poor plasma. Measurements were performed in triplicate for each patient and the mean result was recorded for each individual. The coefficient of variation was calculated for each patient as a measure of within-subject intraassay variation.²⁰

Analysis

STATA (special edition 12.1; Statacorp, College Station, Tex) was used to conduct all statistical analyses. Independent variables selected a priori included PET findings and plasma levels of VEGF-C and VEGF-D. A binary indicator of fluorodeoxyglucose (FDG) uptake within hilar and/or mediastinal lymph nodes was created to summarize PET findings. FDG uptake above background levels as reported by the radiologist was considered a positive PET result. Plasma levels of VEGF-C and VEGF-D were log-transformed to achieve a normal distribution and were modeled as continuous variables. The primary dependent variable was a binary indicator of nodal disease based on pathologic examination of tissue obtained by invasive staging procedures and/or intraoperatively. There were no missing data.

Logistic regression was used to determine whether VEGF-C and VEGF-D improve the predictive performance of PET. Four models were created using different sets of predictors to estimate the probability of nodal disease. For descriptive purposes, we also report associations (using odds ratios), discriminatory accuracy (using c-statistics), and calibration (using goodness-of-fit tests). The study was not powered to detect differences in these parameters. A likelihood ratio test was used to formally test the hypothesis that 1 or more biomarkers would improve the predictive performance of PET.²¹ Our sample size was informed by a study indicating that a logistic regression analysis is reasonably powered by 5 to 10 events per covariate.²² We estimated needing between 15 and 30 patients with nodal disease to have sufficient power to conduct our a priori-specified aims.²² Because 5 pairwise comparisons were planned, a P value<.01 was considered statistically significant to account for multiple comparisons.

An exploratory analysis was planned to better understand the potential clinical influence of using lymphangiogenesis markers in lung cancer staging. Specifically, we were interested in contrasting the accuracy of varying approaches to risk stratification for the purposes of selecting patients for invasive staging. To create risk strata based on PET findings and lymphangiogenesis marker levels, logistic regression was used to predict the probability of nodal disease for each patient. An empirically derived cutoff maximizing the sensitivity and specificity of the model was determined (42% probability of nodal disease) using the Youden Index and used to create high-risk and low-risk groups.²³ We assumed a simple binary decision-making framework where high-risk patients would undergo invasive staging and low-risk patients would proceed to local therapy. This risk stratification was used to estimate the expected frequency of invasive staging had providers adhered to using the model. To provide contrasts, we described the accuracy of PET and practice guidelines in selecting patients to undergo invasive staging. We estimated the expected frequency of invasive staging based on PET alone by assuming that all patients with FDG uptake in any node to undergo invasive staging. Using the American College of Chest Physicians and the National Comprehensive Cancer Network recommended indications for invasive nodal staging: (FDG uptake within hilar and/or mediastinal lymph nodes, lymphadenopathy, central tumor, and tumors ≥3cm in size^1,2), we calculated the proportion of patients with 1 or more of these indications to determine the expected frequency of invasive staging.

RESULTS

Among 566 patients seen in clinic during the study period, 448 (79%) consented to participate in the lung nodule biorepository. Most patients (86%) enrolled in the biorepository did not meet study inclusion criteria (Figure 1). The 2 most common reasons for exclusion were patients recommended short-interval imaging follow-up for a lung nodule, and patients with a nonconcerning nodule recommended no further follow-up. Table 1 summarizes the characteristics of the study cohort. Most patients (84%) were ultimately diagnosed with NSCLC (final pathologic stage: 44% stage I, 15% stage II, 23%stage IIIA, and 17% stage IIIB). Thirty-six patients (58%) had FDG uptake within lymph nodes (N1: n = 4, 16%; N2: n = 8, 32%; and N3: n = 6, 24%). The median number of nodal stations sampled by invasive staging was 3 (range, 0–6). For those undergoing resection (55% lobectomy, 38% wedge, 5% bilobectomy, and 2.5% segmentectomy), the median number of nodal stations sampled at the time of lung resection was 3 (range, 0–6). Overall, a median of 4 (range, 1–9) nodal stations were sampled by invasive staging and/or intraoperatively. Twenty-five patients (40%) had pathologically confirmed nodal disease.

FIGURE 1.

Patient accrual and cohort selection. SCLC, Small cell lung cancer; AIS, adenocarcinoma in situ; PET, positron emission tomography; NSCLC, non–small cell lung cancer.

TABLE 1.

Patient characteristics (n = 62)

Characteristic	Result
Age (y)	67 (39–84)
Men	48
Race
White	87
Black	0
Asian	8.1
Native Hawaiian/Pacific Islander	3.2
American Indian/Alaska Native	1.6
Comorbid conditions*
Hypertension	48
Coronary artery disease	24
Cerebrovascular disease	1.6
Peripheral vascular disease	1.6
Chronic obstructive pulmonary disease	36
Diabetes mellitus	8.1
Smoking status
Former	47
Current	42
Never	11
Pack-years†	42 (0.5–180)
Forced expiratory volume in 1 s	83 (31–132)
Clinical stage
I	40
II	16
IIIA	26
IIIB	18
Nodal disease by positron emission tomography	58
Invasive staging
None	9.7
Mediastinoscopy	53
Endobronchial ultrasound guided biopsy	26
Endobronchial ultrasound guided	11
biopsy/mediastinoscopy
Diagnosis
Non–small cell lung cancer	84
Small cell lung cancer	1.6
Metastasis (nonlung)	1.6
B-cell lymphoma	3.2
Inflammatory nodule	8.1
Benign nodule	1.6
Pathologic nodal disease	40
Therapy
Resection alone	3
Resection/adjuvant therapy	19
Radiation only	4.8
Chemoradiation	16
Chemotherapy	1.6
Antibiotics/antifungal	1.6
None	10
Unknown	10
Vascular endothelial growth factor C
Nontransformed
Mean ± standard deviation	1191 ± 965
Median (range)	861 (218–4241)
Log-transformed
Mean ± standard deviation	6.8 ± 0.7
Median (range)	6.8 (5.4–8.4)
Coefficient of variation
Median (range)	3.3 (0–19)
Proportion ≤ 5.0	74
Vascular endothelial growth factor D
Nontransformed
Mean ± standard deviation	263 ± 129
Median (range)	235 (91–919)
Log-transformed
Mean ± standard deviation	5.5 ± 0.4
Median (range)	5.5 (4.5–6.8)
Coefficient of variation
Median (range)	4.2 (0.4–39)
Proportion ≤5.0	60

Values are presented as median (range), %, or mean ± standard deviation.

^*Columns may total more than 100% because of multiple conditions. No patients had congestive heart failure, pulmonary hypertension, interstitial pulmonary fibrosis, or dialysis dependence.

^†Calculated among current and former smokers (n = 55).

VEGF-C and VEGF-D levels correlated with nodal disease. Median levels of VEGF-C were higher among node-positive patients (1179 pg/mL) compared with node-negative patients (733 pg/mL). Similarly, median levels of VEGF-D were higher among node-positive patients (273 pg/mL) compared with node-negative patients (229 pg/mL). Figure E1 shows an increasing trend in the prevalence of nodal disease across increasing levels of VEGF-C and VEGF-D. A test for a linear trend was statistically significant for VEGF-C (P = .017) but not VEGF-D (P = .123). Examination of the models used for hypothesis testing revealed that increasing VEGF-C levels were associated with a significantly higher risk of nodal disease independent of PET findings and/or VEGF-D levels (Table 2). Although increasing VEGF-D levels were also associated with higher risks of nodal disease in these models, the relationship was not statistically significant.

TABLE 2.

Summary of model coefficients, discrimination, and calibration

Model	Odds ratio (95% confidence interval)	c-Statistic	Goodness-of-fit test*
ModelPET		0.617	—†
PET	2.71 (0.92–8.03)
ModelPET/VEGF-C		0.725	0.376
PET	2.56 (0.80–8.12)
VEGF-C	2.96 (1.26–6.90)
ModelPET/VEGF-D		0.671	0.387
PET	2.45 (0.81–7.39)
VEGF-D	2.35 (0.63–8.67)
ModelPET/VEGF-C/VEGF-D		0.738	0.360
PET	2.34 (0.72–7.57)
VEGF-C	2.87 (1.22–6.77)
VEGF-D	2.04 (0.55–7.52)

PET, Positron emission tomography; VEGF, vascular endothelial growth factor.

^*A nonsignificant result (P >.05) indicates a model that fits the data well.

^†It was not possible to calculate a goodness-of-fit test for Model_PET because there were too few degrees of freedom to perform the test.

Table 3 summarizes the primary findings from this study. After correcting for multiple comparisons, Model_PET/VEGF-C was significantly better at predicting nodal disease than Model_PET. Likewise, Model_{PET/VEGF-C/VEGF-D} was significantly better at predicting nodal disease than Model_PET/VEGF-D. As shown by a higher c-statistic value, models with VEGF-C had better discriminatory accuracy than models without VEGF-C (Table 2). The addition of VEGF-D did not significantly improve prediction for any of the models. All models fit the data well (Table 2).

TABLE 3.

Comparison of prediction models inclusive of lymphangiogenesis markers and positron emission tomography (PET)

Comparison	Likelihood ratio test P value*
ModelPET/VEGF-C vs ModelPET	.0069
ModelPET/VEGF-D vs ModelPET	.1886
ModelPET/VEGF-C/VEGF-D vs ModelPET	.0146
ModelPET/VEGF-C/VEGF-D vs ModelPET/VEGF-C	.2818
ModelPET/VEGF-C/VEGF-D vs ModelPET/VEGF-D	.0095

PET, Positron emission tomography; VEGF, vascular endothelial growth factor.

^*Statistical significance was defined by P<.01 to account for multiple comparisons.

An analysis was planned to explore how VEGF-C might influence clinical practice (Table 4). Compared with PET, a prediction model inclusive of PET findings and VEGF-C levels would increase the specificity, positive predictive value, and overall accuracy of selecting patients for invasive staging without influencing sensitivity or negative predictive value (NPV). The expected use of invasive staging procedures would have been 47% for a model with PET findings and VEGF-C levels rather than 58% based on PET findings alone. However, in clinical practice, practice guidelines inclusive of multiple radiographic predictors are used to select patients to undergo invasive staging procedures. The sensitivity and NPV of practice guidelines indications for invasive staging are 100%; however, specificity is low (35%) and the expected rate of guideline-concordant use of invasive staging procedures would have been 79%.

TABLE 4.

Exploratory analysis of the accuracy of risk stratification for invasive nodal staging using positron emission tomography (PET) alone, Model_PET/VEGF-C, and practice guidelines

Risk stratification	PET	ModelPET/VEGF-C	Practice guidelines
Sensitivity	72	72	100
Specificity	51	70	35
Positive predictive value	50	62	51
Negative predictive value	73	78	100
Overall accuracy	60	71	61
Expected rate of invasive staging*	58	47	79

Values are presented as %. PET, Positron emission tomography; VEGF, vascular endothelial growth factor.

^*The expected rate of invasive staging based on PET use alone was calculated assuming that all patients with fluorodeoxyglucose uptake within any nodal station would undergo invasive staging. For Model_PET/VEGF-C, the expected rate of invasive staging was calculated by assuming that all patients classified as high risk would undergo invasive staging. To risk-stratify patients, the predicted probability of nodal disease was estimated for each patient based on PET findings and VEGF-C levels. An empirically defined cutoff (42%) maximizing the sensitivity and specificity of the model was determined by the Youden Index.²³ Patients with a probability of nodal disease higher than the cutoff were considered high risk and those with a probability below the cutoff were considered low risk. The expected rate of invasive staging among guideline-adherent providers was calculated assuming that patients with any 1 or more of the following radiographic findings would undergo invasive staging^1,2: fluorodeoxyglucose uptake within hilar and/or mediastinal lymph nodes, lymphadenopathy, central tumor, and tumors ≥3 cm in size.

We conducted several post-hoc analyses to evaluate concerns that VEGF-C may not add value to a model with multiple radiographic predictors of nodal disease (Tables E1–E3). Node-positive patients tended to have larger tumor size by CT (P = .016), higher maximized standardized uptake value (SUVs) measured from the primary tumor (P = .006), and lymphadenopathy on CT (P = .092), but only the relationships with tumor size and SUV were statistically significant. Central location was not associated with nodal disease (P = .204). VEGF-C levels tended to be higher among patients with larger and central tumors, lymphadenopathy by CT, and higher primary tumor SUVs, but none of these associations were statistically significant (P >.05). A model inclusive of PET findings, primary tumor size, location, and SUV, lymphadenopathy by CT, and VEGF-C predicted nodal disease better than a similar model without VEGF-C (P = .034). The VEGF-C-inclusive model also had better discriminatory accuracy (c-statistic 0.760 vs 0.731). Despite concerns over sample size, both models appeared to fit the data well (P = .281 and P = .298, respectively). In the full model, VEGF-C remained the predictor most strongly associated with nodal disease (odds ratio, 2.58; 95% confidence interval, 1.03–6.43).

DISCUSSION

A growing body of basic research and epidemiologic evidence motivates the use of lymphangiogenesis biomarkers to predict nodal disease in patients with suspected or confirmed nonmetastatic NSCLC. Previously, a group of investigators demonstrated that VEGF-C could improve the diagnostic performance of CT.⁹ Our study is the first to provide evidence of the added diagnostic value of VEGF-C in predicting nodal disease in the modern era of PET.

It is unclear why VEGF-D did not improve prediction. In mice models of embryonic development, homozygous deletion of VEGF-C leads to complete absence of lymphatic vasculature, whereas homozygous deletion of VEGF-D does not.⁶ However, exogenous VEGF-D rescues vessel sprouting in VEGF-C-null mice, suggesting that VEGF-C is essential for lymphangiogenesis in normal development, whereas VEGF-D is not. It is possible that VEGF-C and VEGF-D similarly have essential and nonessential roles, respectively, in the pathogenesis of lymphatic metastases. Also, VEGF-D levels were lower and more variable than VEGF-C levels, both of which would be expected to unfavorably influence prediction.

By virtue of evaluating a population encountered in everyday practice—patients with suspected or confirmed NSCLC—our estimates of the added diagnostic value of VEGF-C are both conservative and generalizable. Unlike other primary tumor sites, diagnosis and staging of lung cancer can occur simultaneously rather than sequentially. One consequence of this approach is that some patients will undergo staging but will not be diagnosed with lung cancer. Accordingly, the predictive performance of diagnostic tests may be diminished in patients with suspected or confirmed NSCLC compared with patients with confirmed NSCLC. Nodal staging among patients with suspected NSCLC represents guideline-concordant practice^1,2 and is increasingly recognized as a marker of high-quality, high-value oncologic care.^24,25 Our study provides an understanding of the diagnostic value of VEGF-C in a real-world setting.

An important limitation of our study is that we were unable to validate the added value of VEGF-C in an independent dataset because of our small sample size. Several methods—change in the area under the curve and net reclassification index—have been used to evaluate the added value of biomarkers in cancer diagnostics when validation in an independent cohort is not possible because of feasibility and/or expense. A recent critical review of these methods recommended a different statistical approach—likelihood ratio testing—on the basis of theoretical and empirical evidence in support of its superiority over other methods, particularly in terms of avoiding overly optimistic results.²¹ Nonetheless, our study cannot provide definitive evidence of the diagnostic value VEGF-C without independent validation. The intent of this study was to provide evidence of the potential diagnostic value of VEGF-C to demonstrate feasibility and justify the resources and expense of carrying out larger-scale investigations.

One perceived limitation of this study is that the accuracy of combined PET findings and VEGF-C levels was lower than the reported pooled accuracy of PET for mediastinal nodal disease.¹ Traditionally, the diagnostic performance of PET has been described in terms of its ability to identify mediastinal nodal disease (ie, N2 or N3) rather than any nodal disease (ie, N1-N3), and thus direct comparisons to prior pooled estimates of PET performance may not be valid. Similar to prior studies evaluating lymphangiogenesis markers in patients with lung cancer,^8–14 we evaluated diagnostic performance in terms of any nodal disease (ie, N1-N3) because lymphangiogenesis markers are not expected to necessarily discriminate between metastases in specific nodal stations. This assumption was supported by a post-hoc analysis showing no clear relationship between nodal station positivity and VEGF-C or VEGF-D levels (Table E4). VEGF-C levels appeared to be highest in patients with pathologically confirmed N1 disease (followed by N3 and then N2), but this apparent relationship was not statistically significant and was most likely due to chance. Some may appropriately argue that there is no value in predicting N1 disease because management will not change and there is no clear benefit of induction versus adjuvant therapy for stage II disease. Nonetheless, if predicting any nodal disease results in more accurate staging overall and better use of invasive staging procedures among patients with suspected or confirmed nonmetastatic NSCLC, then clinical prediction will increase the value of thoracic oncologic care.

Our exploratory analysis highlights performance gaps that future prediction models will need to address and provides insights into how VEGF-C may beneficially influence clinical staging. Practice guidelines would have perfectly selected all patients with true nodal disease to undergo invasive staging (100% sensitivity), and would have ensured that all patients who did not undergo staging did not have nodal disease (100% NPV). However, a high proportion of patients (79%)—nearly double the prevalence of true pathologic nodal disease (40%)—would have to undergo invasive staging to achieve these results. Invasive staging procedures are associated with rare but potentially life-threatening risks, increased costs, and possibly decreased efficiency of care.^26–28 Greater accuracy in predicting nodal disease should result in better use of invasive staging procedures. Although VEGF-C does not appear to improve upon the sensitivity and NPV of PET, it does appear to improve specificity and positive predictive value. Accordingly, one would expect VEGF-C to help reduce the number of invasive procedures performed among patients who do not have nodal disease. Clearly, a prediction model inclusive of only PET findings and VEGF-C levels performs inadequately and is therefore unsuitable for clinical use. Practice guidelines recommend using several radiographic predictors of nodal disease to select patients for invasive staging—including FDG uptake within hilar and/or mediastinal nodes, lymphadenopathy, centrally located tumors, and tumors ≥3 cm.^1,2 Combining these factors with VEGF-C and other reported radiographic predictors of nodal disease—such as histologic information (eg, adenocarcinoma) from a transthoracic or transbronchial biopsy of the primary tumor, the SUV of the primary tumor, and patterns of FDG uptake across nodal stations^29–31—may result in a highly accurate prediction model. The relative contribution of each predictor remains uncertain and a larger study will need to be performed assess the value of VEGF-C, although our post-hoc analyses suggest that it is a dominant predictor of nodal disease. A funded investigation is currently underway to develop a risk-prediction model for nodal disease based on multiple radiographic predictors of nodal disease, and to determine if VEGF-C improves the model’s ability to accurately predict nodal metastases. Until a higher level of evidence is available, VEGF-C should not be used in clinical practice outside the context of a clinical trial.

CONCLUSIONS

Without new imaging technology, a clinical prediction model inclusive of radiographic and molecular risk factors for nodal disease may allow us to improve the accuracy of lung cancer staging and make better use of invasive diagnostic procedures. The model would be used to stratify patients into 2 groups. Those at low risk for nodal disease would proceed directly to treatment without invasive staging, whereas those at high risk would be selected to undergo invasive staging. The clinical prediction model would not be a substitute for invasive staging in high-risk patients or intraoperative nodal sampling or dissection in any operative patient. Recently, the potential benefits of using a prediction models based on radiographic risk factors for nodal disease was demonstrated in a subset of patients with a negative mediastinum by PET.^31,32 Prediction models may lead to better diagnostic accuracy, fewer invasive staging procedures, and less provider-level variability in staging practices while ensuring personalized cancer care.

Central Message.

Vascular-endothelial growth factor C complements positron emission tomography in predicting nodal disease for lung cancer staging.

Perspective.

Our work justifies and motivates development and validation of a prediction model for nodal disease in lung cancer based on multiple radiographic predictors of nodal disease and vascular endothelial growth factor C. Prediction models may lead to better staging accuracy, fewer invasive staging procedures, and less provider-level variability in staging practices while ensuring personalized cancer care.

Abbreviations and Acronyms

CT

computed tomography

EDTA

ethylenediaminetetraacetic acid

FDG

fluorodeoxyglucose

NPV

negative predictive value

NSCLC

non–small cell lung cancer

PET

positron emission tomography

SUV

standardized uptake value

VEGF

vascular endothelial growth factor