Brain Metabolic Activity Measured by [¹⁸F]FDG PET/CT Predicts Survival in Patients with Advanced Non–Small Cell Lung Cancer

Visual Abstract

Abstract

[¹⁸F]FDG PET/CT images play a key role in the management of patients with non–small cell lung cancer (NSCLC). In these scans, the focus is on detected tumors and their characteristics, neglecting information from other organs or tissues. We investigated whether the mean brain [¹⁸F]FDG uptake (brain SUV_mean) is associated with overall survival (OS) in patients with advanced NSCLC. Methods: This retrospective study included patients with advanced NSCLC who underwent pretreatment [¹⁸F]FDG PET/CT scans between 2010 and 2023. Clinical and biologic data, tumor radiomic features, and brain SUV_mean were collected. The ability of these features to predict OS was evaluated using univariable and multivariable Cox regression models. The correlation between brain SUV_mean and clinical, imaging, and blood biomarkers was investigated using Spearman correlation coefficients. Results: Patients were chronologically divided into a discovery set (n = 234; mean age, 64 ± 11 y) and test set (n = 146; mean age, 66 ± 11 y). In the discovery set, univariable analysis showed that high brain SUV_mean (greater than or equal to the median) was associated with longer OS (hazard ratio [HR], 0.83; 95% CI, 0.76–0.92; P < 0.001). Brain SUV_mean was significantly lower in patients who died within 1 y compared with those who were still alive at the same time point (median brain SUV_mean, 4.9 ± 1.4 vs. 5.7 ± 1.5, respectively; P < 0.001). Multivariable analysis revealed that brain SUV_mean was an independent prognostic factor for OS (HR, 0.89; 95% CI, 0.80–0.98; P = 0.02), which was confirmed in the test set (P < 0.001). Brain SUV_mean was independent of the radiomic features quantifying tumor involvement (r < 0.24, n = 380) and significantly correlated but complementary to several blood biomarkers including C-reactive protein (r = −0.37, n = 110 patients). The prognostic significance of brain SUV_mean persisted in patients without brain metastases (P < 0.001). Conclusion: Low brain metabolic activity was associated with increased mortality in patients with advanced NSCLC. Brain SUV_mean was an independent prognostic factor that may aid in patient stratification, although its interpretation requires further investigation.

Lung cancer was the leading cause of cancer-related deaths worldwide in 2022, with a total of 1.8 million deaths (18%) (1). Despite promising new therapies, the prognosis for patients is relatively poor, with overall survival (OS) ranging from 37.5 mo in early-stage disease to 4.8 mo in the metastatic stage (2). In clinical practice, the management of metastatic lung cancer remains challenging, as some patients are resistant to treatments or experience only short-term benefits without any significant impact on their survival (3,4).

Numerous studies have shown that advanced analysis of whole-body [¹⁸F]FDG PET/CT images could provide valuable information regarding tumor invasion to predict survival in patients with lung cancer. Total metabolic tumor volume (TMTV), total lesion glycolysis (TLG), SUV_max, and the maximum distance between 2 lesions (D_max) are PET-derived biomarkers reflecting metabolic tumor burden, tumor activity, and disease dissemination and have a strong prognostic value for patient survival (5–7).

Recent studies have suggested that metabolic activity and density in various organs and tissues measured from whole-body [¹⁸F]FDG PET/CT images also provide valuable information for predicting patient outcomes in lung cancer by reflecting the patient’s general condition (8,9). For example, [¹⁸F]FDG uptake measured in lymphoid organs may reflect systemic inflammation and has been associated with prognosis in patients with advanced non–small cell lung cancer (NSCLC) (10,11).

In parallel, several publications suggest the existence of a lung–brain axis (12), indicating that diseases affecting the pulmonary system can lead to alterations in brain structure and function. Brain alterations have been observed in patients with lung cancer when compared with healthy controls, even before the initiation of therapy (13). Recent studies have also suggested a link between neuroinflammation and several prognostic factors, such as pathogenesis, cachexia (14), tumor growth (15), and systemic inflammation (16), calling for further investigations regarding the role of the brain function in patients with cancer.

The aim of this study was to evaluate the relationship between brain metabolic activity, measured using [¹⁸F]FDG PET, and survival in patients with advanced NSCLC.

MATERIALS AND METHODS

This study was approved by the institutional review board of Institut Curie (DATA220207). Informed consent was provided by all patients. All related data were deidentified, collected, and stored in compliance with General Data Protection Regulation and the Declaration of Helsinki.

Patients

This retrospective study included 455 patients with advanced NSCLC treated at Institut Curie between 2010 and 2023 who underwent baseline [¹⁸F]FDG PET/CT before any treatment. Patients were treated with the standard of care at the time of their management. Patients with no baseline [¹⁸F]FDG PET/CT before surgery or neoadjuvant therapy, a follow-up duration of less than 12 mo, missing clinical data, a PET/CT scan that did not fully include the brain in the axial field of view, and no lesion with an SUV of greater than 4 were excluded from the study. Clinical data, including age, sex, body mass index (BMI), smoking history, performance status (PS), stage, and treatment administered, were collected. Blood parameters, including albumin, C-reactive protein (CRP), monocytes, leukocytes, lymphocytes, neutrophils, and cortisol, were collected when available within 30 d of the PET scan and prior treatment initiation. The prognostic nutritional index (PNI), defined as albumin (grams per liter) + 5 × lymphocyte count (10⁹/liter), was calculated. Glycemic control was performed immediately before the PET scan. Patients’ history of depression was retrospectively collected from their medical records when the information was available.

The cohort was divided into 2 datasets: patients who underwent their PET/CT scan between January 2010 and December 2018 comprised the discovery set, whereas those scanned between January 2019 and December 2023 comprised the test set (Fig. 1).

FIGURE 1.

Flow diagram of study patients with advanced NSCLC.

OS was defined as the time between the baseline PET/CT scan acquisition and the date of the patient’s death or the last time the patient was known to be alive. Follow-up was determined from the date of the baseline PET/CT scan to the date of the last clinical consultation.

Imaging and Image Analysis

[¹⁸F]FDG PET/CT scans were performed in various centers using different PET/CT systems and acquisition and reconstruction protocols (Supplemental Table 1, available at http://jnm.snmjournals.org). The mean blood glucose level before [¹⁸F]FDG PET/CT acquisition was 6.0 ± 1.4 mmol/L. An injected dose of 235 ± 75 MBq of [¹⁸F]FDG was administered 66 ± 11 min before image acquisition, with 1.8 ± 0.7 min per bed position. PET images were analyzed using SUV normalized to patient body weight.

All lesions were automatically segmented on PET images using LION version 0.14 (17). Lesion delineation was automatically refined by only including voxels with an SUV of 4 or greater. The whole brain was automatically delineated on the CT with TotalSegmentator version 2.0.5 (18), and the resulting volume of interest (VOI) was copied to the corresponding PET image (VOI_brain). In addition, segmentation of 8 cerebral subregions, corresponding to the largest brain structures (lobes, ventricle, cerebellum, and brainstem), was performed.

For patients with brain metastases identified by any imaging modality (e.g., CT, MRI) at the time of the PET/CT scan or within 3 mo thereafter, we manually delineated brain lesions on the PET images, using the anatomic modality if necessary, and excluded these brain lesions from VOI_brain, resulting in a volume corresponding to the whole brain without metastatic lesions (VOI_{brain_noM}).

PET images were then resampled to a fixed voxel size of 2 × 2 × 2 mm³, and 5 tumor-related radiomic features were automatically extracted across all tumor lesions using LIFEX (version 26.3.8; http://www.lifexsoft.org) (19): D_max, maximum SUV_max, and total tumor SUV_mean, TMTV, and total TLG (equal to tumor SUV_mean × TMTV). Cerebral metabolic activity was calculated as the mean [¹⁸F]FDG uptake within VOI_brain, within the 8 cerebral subregions and also within VOI_{brain_noM}, normalized by patient body weight (brain SUV_mean and brain SUV_mean without metastases, respectively) or normalized by lean body mass (brain SUL_mean), using the following formulas (20):

$$\mathrm{Brain}\mathrm{ }{\mathrm{SUL}}_{\mathrm{mean}}\mathrm{ }\left(\mathrm{Female}\right)=\frac{{\mathrm{SUV}}_{\mathrm{mean}}}{\mathrm{Weight}\mathrm{ }\left(\mathrm{g}\right)}\times (1.07\times \mathrm{Weight}\mathrm{ }\left(\mathrm{kg}\right)-148\times \frac{{\mathrm{Weight}}^2\mathrm{ }\left(\mathrm{kg}\right)}{{\mathrm{Height}}^2\mathrm{ }\left(\mathrm{cm}\right)})\times 1000.$$ Eq. 1

$$\mathrm{Brain}\mathrm{ }{\mathrm{SUL}}_{\mathrm{mean}}\mathrm{ }\left(\mathrm{Male}\right)=\frac{{\mathrm{SUV}}_{\mathrm{mean}}}{\mathrm{Weight}\mathrm{ }\left(\mathrm{g}\right)}\times (1.10\times \mathrm{Weight}\mathrm{ }(\mathrm{kg})-120\times \frac{{\mathrm{Weight}}^2\mathrm{ }\left(\mathrm{kg}\right)}{{\mathrm{Height}}^2\mathrm{ }\left(\mathrm{cm}\right)})\times 1000.$$ Eq. 2

Statistical Analysis

Patient Characteristics

Baseline patient characteristics were analyzed using Wilcoxon tests for continuous variables and χ² tests for categoric variables. Spearman correlation coefficients were calculated to assess the correlation between imaging features.

Evaluation of Automatic Segmentation

The reliability of the automatic segmentation performed by LION was assessed by comparison with the manual segmentation performed using an SUV threshold of 4 by an expert physicist under the supervision of a nuclear physician, in 100 randomly selected patients. The Dice score was calculated to assess agreement between expert and automated segmentation, and Bland–Altman analysis was performed to compare the 5 feature values as a function of the delineation method.

Survival Analysis

Associations between feature values and OS in the discovery set were studied using univariable Cox regression. All variables associated with OS at a P value of <0.05 in univariable analysis were included in a Cox multivariable model, without and with the brain [¹⁸F]FDG uptake (models 1 and 2, respectively). ANOVA was conducted to compare the 2 models. A risk score was generated from the multivariable models and then binarized by the median on the discovery set to distinguish between high- and low-risk patients by OS. We then applied the same models (same coefficients and same cutoff) to the test set. For both the discovery and test sets, the association of low-and high-risk groups with OS was evaluated using Kaplan–Meier curves with the log-rank test and a Cox proportional hazards regression model, including calculation of adjusted hazard ratios (HRs) with 95% CIs and Wald test P values.

Identification of Factors Affecting Cerebral Metabolic Activity

The influence of brain metastases on the prognostic value of cerebral metabolic activity was investigated by comparing the results obtained with VOI_brain and VOI_{brain_noM}. The prognostic value of cerebral metabolic activity was also assessed in the subgroup of patients with advanced NSCLC without known brain metastases. The impact of the treatment on the prognostic value of cerebral metabolic activity was studied by categorizing patients by the treatment received after PET/CT (chemotherapy, radiotherapy, immunotherapy or targeted therapy). We used the Wilcoxon test and Spearman correlation coefficient to study the relationship between cerebral uptake and the presence of cerebral metastases, the technical PET/CT acquisition parameters, the value of biologic parameters, and patients’ depression history. We explored whether integrating cerebral metabolic activity with blood biomarkers could enhance prognostic stratification. For each prognostic biologic parameter, patients were stratified into 3 risk groups: low risk (0 risk factor), intermediate risk (1 risk factor), and high risk (2 risk factors). A P value of less than 0.05 was defined as statistically significant. All statistical analyses were conducted using R software version 4.2.2 (R Project for Statistical Computing).

Our study followed the criteria of the methodological radiomics score (21) and achieved a score of 84%, indicating high adherence to recommended radiomic research practices (Supplemental Fig. 1).

RESULTS

Patient Characteristics

In total, 380 patients (mean age, 65 ± 11 y) were included in this retrospective study (Fig. 1; Table 1). Our study cohort was divided into 2 groups by scan date: the discovery set (n = 234, 62%) and the test set (n = 146, 38%). Patients in the test set had a higher BMI (P = 0.004), 126 (86%) had a PS of 0 of 1 (P = 0.04), and 116 (79%) were treated with immunotherapy (P < 0.001). The test set also included a higher proportion of metastatic patients (99% had stage IV disease) compared with the discovery set (89%, P < 0.001).

TABLE 1.

Demographic and Clinical Characteristics of Patients

Characteristic	Eligible cohort	Discovery set	Test set	P *
n	380	234	146
Age at diagnosis (y)	65 ± 11	64 ± 11	66 ± 11	0.15
Sex				0.96
Female	162 (43)	99 (42)	63 (43)
Male	218 (57)	135 (58)	83 (57)
BMI (kg/m²)	23.8 ± 4.0	23.4 ± 4.0	24.4 ± 3.9	0.004
Histology				0.43
Adenocarcinoma	279 (73)	167 (71)	112 (77)
Squamous cell carcinoma	52 (14)	36 (15)	16 (11)
Other	49 (13)	31 (14)	18 (12)
Stage				<0.001
III	29 (7)	27 (11)	2 (1)
IVa	128 (34)	70 (30)	58 (40)
IVb	223 (59)	137 (59)	86 (59)
PS				0.04
0 or 1	307 (81)	181 (77)	126 (86)
≥2	73 (19)	53 (23)	20 (14)
Smoking history				0.57
Never	53 (14)	35 (15)	18 (12)
Current or former	327 (86)	199 (85)	128 (88)
Depressive disorders				0.84
Absence/not specified	362 (95)	222 (95)	140 (96)
Presence	18 (5)	12 (5)	6 (4)
Brain metastasis				0.31
Absence	293 (77)	185 (79)	108 (74)
Presence	87 (23)	49 (21)	38 (26)
Treatment received
Chemotherapy	325 (86)	200 (85)	125 (86)	>0.99
Immunotherapy	222 (58)	106 (45)	116 (79)	<0.001
Radiotherapy	247 (65)	154 (66)	93 (64)	0.76
Targeted therapy	122 (32)	80 (34)	42 (29)	0.32
1-y survival				<0.001
Alive	233 (61)	131 (56)	102 (70)
Dead	147 (39)	103 (44)	44 (30)
Follow-up duration (mo)	24.6 ± 22.3	25.1 ± 25.5	23.7 ± 15.9	0.16

^*P values from Wilcoxon and χ² tests between discovery and test sets are shown.

Categoric data expressed as number, followed by percentage in parentheses; continuous data expressed as mean ± SD.

Evaluation of Automated Segmentations

The tumor regions segmented by LION and by the experts were very similar, with an average Dice score of 0.90 ± 0.16 (range, 0.0–1.0) and no substantial differences in the derived radiomic feature values were found by Bland–Altman analyses (Supplemental Fig. 2). The prognostic performance of the 5 radiomic features was similar for both segmentation approaches, except for D_max (Supplemental Fig. 3). To ensure replicability, the results presented below were obtained using LION segmentation refined with a threshold of 4 SUV.

Survival Analysis

Among all 380 patients, brain SUV_mean showed a significant but weak negative correlation with TMTV (r = −0.24) and total TLG (r = −0.21) (Supplemental Fig. 4). There was no significant correlation between brain SUV_mean and D_max (r = −0.09), total tumor SUV_mean (r = 0.03), and maximum SUV_max (r = 0.06) (P > 0.05). Total TLG was excluded from subsequent analyses because of its high correlation with TMTV (r = 0.98).

Kaplan–Meier curves for each feature are shown in Supplemental Figure 5 (continuous variable dichotomized by the median). In the discovery set, these univariable analyses identified 6 variables statistically associated with OS (Table 2): PS, smoking history, stage, D_max, TMTV, and brain SUV_mean. In the multivariable analysis without brain SUV_mean (model 1), all of these features were significantly associated with OS except D_max. Using model 1, patients from the discovery set were divided into high- and low-risk groups on the basis of the score calculated from the multivariable Cox model dichotomized by the median value (1.56 [interquartile range, 0.5–42.7]). OS differed significantly between the 2 groups (P < 0.001) (Figs. 2A and 2B). Median OS was 25.4 mo (95% CI, 19.7–32.4 mo) and 8.2 mo (95% CI 6.8–10.1 mo) in the low-and high-risk groups, respectively.

TABLE 2.

Cox Proportional Hazards Regression Analysis of Association of Clinical and Radiomic Features and Brain SUV_mean with OS in Discovery Set (n = 234)

			Multivariable analysis
	Univariable analysis		Model 1*		Model 2†
Variable	HR (95% CI)	P ‡	HR (95% CI)	P ‡	HR (95% CI)	P ‡
Age (y)	1.01 (1.00–1.02)	0.08
Sex
Female	—
Male	1.19 (0.90–1.57)	0.21
BMI (kg/m²)	0.99 (0.95–1.03)	0.56
PS
0 or 1	—		—		—
≥2	2.20 (1.60–3.03)	<0.001	1.96 (1.41–2.71)	<0.001	1.88 (1.35–2.61)	<0.001
Smoking history
Never	—		—		—
Current or former	1.63 (1.10–2.42)	0.01	1.71 (1.13–2.59)	0.01	1.53 (1.00–2.35)	0.049
Stage
III	0.91 (0.60–1.38)	0.64	1.01 (0.62–2.59)	0.97	0.97 (0.59–1.58)	0.89
IVa	0.62 (0.45–0.84)	0.002	0.65 (0.46–0.92)	0.01	0.65 (0.46–0.91)	0.01
IVb	—		—		—
Dmax/10	1.07 (1.02–1.13)	0.009	1.03 (0.96–1.10)	0.35	1.03 (0.96–1.10)	0.42
TMTV/100	1.12 (1.06–1.19)	<0.001	1.12 (1.06–1.18)	<0.001	1.11 (1.04–1.17)	<0.001
Total TLG/100	1.02 (1.01–1.02)	<0.001
Total tumor SUVmean	1.05 (0.99–1.12)	0.11
Maximum SUVmax	1.00 (1.00–1.01)	0.06
Brain SUVmean	0.83 (0.76–0.92)	<0.001			0.89 (0.80–0.98)	0.02

^*Includes all variables significantly associated with OS in univariable analysis (except brain SUV_mean).

^†Includes features from model 1 and brain SUV_mean.

^‡Determined using Wald test.

FIGURE 2.

Kaplan–Meier curves for 2 multivariable models in discovery (n = 234) and test (n = 146) sets. Model 1 includes PS, smoking history, stage, D_max, and TMTV for discovery (A) and test (B) sets. Model 2 includes features from model 1 and brain SUV_mean for discovery (C) and test (D) sets. Patients were stratified into risk groups using median value (1.56 for model 1 and 1.37 for model 2) derived from discovery set. Log-rank test P values are shown.

When brain SUV_mean was used to build multivariable model 2, a significantly better patient stratification was observed when compared with model 1 (ANOVA, P = 0.02).

When model 1 was applied to the test set, OS significantly differed between the 2 risk groups (P = 0.03), with a median OS of 24.4 mo (95% CI 20.7–NA months) for 84 low-risk patients and 17.1 mo (95% CI 13.4–29.6 mo) for 61 high-risk patients, respectively. The addition of brain SUV_mean (model 2) improved the distinction between low- and high-risk patients in that test set (P < 0.001), with a difference in median OS of 14.3 mo between the 2 patient groups, compared with 7.3 mo with model 1. Very similar results were obtained when using brain SUL_mean (Supplemental Table 2; Supplemental Fig. 6).

Patients who died within the first year of follow-up (61%) had a significantly lower median brain SUV_mean than did patients who were alive after 1 y (4.9 ± 1.4 vs. 5.7 ± 1.5, respectively; P < 0.001) (Fig. 3B; Supplemental Figs. 7B–7D). Similar results were found for brain SUL_mean (Supplemental Fig. 8). When stratifying patients by treatment received (Supplemental Fig. 9), the prognostic value of brain SUV_mean was confirmed for patients who received chemotherapy (86%), radiotherapy (65%), and immunotherapy (58%), but did not reach statistical significance in patients treated with targeted therapies (32%, P = 0.16). We also demonstrated that the combination of brain SUV_mean (or brain SUL_mean) with TMTV enables the identification of 3 distinct risk groups associated with patient survival (Supplemental Fig. 10).

FIGURE 3.

Relationship between brain SUV_mean and OS in patients with advanced NSCLC in eligible cohort (n = 380). (A) Kaplan–Meier curves (P value determined using log-rank test) for patients with NSCLC stratified into risk groups by median brain SUV_mean (median = 5.5). (B) Box plot representation of brain SUV_mean according to 1-y vital status in full cohort (P value determined using Wilcoxon test).

Identification of Factors Affecting Cerebral Metabolic Activity

Among the 87 patients with brain metastases, only 29 (33%) had either metastases or perilesional edema visible on PET images. Low cerebral metabolic activity was significantly associated with poor OS (P < 0.001), regardless of the use of VOI_brain or VOI_{brain_noM} (Fig. 3A; Supplemental Figs. 7A–7C). When the whole-brain was automatically segmented by TotalSegmentator (VOI_brain), we observed no significant difference in brain SUV_mean distribution between patients with brain metastases and those without (P = 0.68). Low [¹⁸F]FDG uptake in the brainstem, cerebellum, and each of the 5 lobes was significantly associated with poor OS (P < 0.04; Supplemental Fig. 11), but no more than the mean [¹⁸F]FDG uptake over the whole brain (brain SUV_mean).

Brain SUV_mean was significantly but weakly correlated with BMI (r = 0.26; Supplemental Fig. 12) and negatively correlated with age (r = −0.20). Moreover, brain SUV_mean was significantly lower in current smokers than in nonsmokers (P = 0.002) as well as in those with a PS of 2 or greater compared with those with a PS of 0 or 1 (P < 0.001). There was no significant correlation between brain SUV_mean and technical parameters influencing uptake, namely the injected dose, time from injection to acquisition, and acquisition duration (Fig. Supplemental 13A).

Regarding biologic data, blood biomarkers were available for 110 patients (29%) (Supplemental Figs. 13B and 14). Brain SUV_mean was positively correlated with albumin (r = 0.31), lymphocytes (r = 0.15), and PNI (r = 0.35) and negatively correlated with blood glucose levels (r = −0.43), neutrophils (r = −0.38), CRP (r = −0.37), leukocytes (r = −0.37), and monocytes (r = −0.32). The correlation between brain SUV_mean and cortisol was not significant (P = 0.26). Kaplan–Meier curves of each blood parameter, enabling patient stratification into 2 groups of risk on the basis of normal biologic values, are shown in Supplemental Figures 15A–15F. Patients who died within the first year (46/110, 43%) had higher levels of CRP, leukocytes, monocytes, and neutrophils and lower levels of lymphocytes compared with patients who were still alive after 1 y (Supplemental Figs. 15G–15L). However, there were no significant differences in albumin levels between risk groups (P = 0.30). Blood glucose levels were significantly associated with survival (P < 0.001), with higher values observed in patients who died within the first year (P = 0.02) (Supplemental Fig. 16). Kaplan–Meier curves showed significantly different OS among patients stratified by 0, 1, or 2 risk factors, defined by the association of brain SUV_mean with blood biomarkers (P ≤ 0.04) (Supplemental Figs. 17 and 18), except for cortisol (P = 0.42, data available for 42 patients). Low PNI was significantly associated with poorer survival (P = 0.02), and its combination with brain SUV_mean identified patients with a poorer prognosis than those with low PNI alone (PNI < 46.9 and brain SUV_mean < 5.5) (Supplemental Fig. 19).

Finally, patients with known depressive disorders (n = 18) had lower median brain [¹⁸F]FDG uptake (brain SUV_mean, 4.9 ± 1.3; brain SUL_mean, 3.9 ± 1.0) than those without these disorders (brain SUV_mean, 5.5 ± 1.5; brain SUL_mean, 4.2 ± 1.1), but the difference was not statistically significant (Supplemental Fig. 20). The presence of depressive disorders was not significantly associated with OS.

Figure 4 shows patients with similar clinical and radiomic features but differing brain SUV_mean and their associated survival.

FIGURE 4.

Examples of baseline [¹⁸F]FDG PET images of 4 patients with metastatic NSCLC. Primary tumor and metastatic lesions are delineated in red. TMTV, brain SUV_mean, and 1-y survival status are shown.

DISCUSSION

Recent studies have suggested that [¹⁸F]FDG metabolic activity measured across various organs and tissues outside the tumor site could provide valuable information for predicting patient outcomes in lung cancer (8,9). We found that brain FDG uptake was an independent prognostic factor in both univariable and multivariable analyses and complemented radiomic features reflecting tumor invasion. Brain FDG uptake, combined with clinical and tumor-related radiomic features measured on baseline [¹⁸F]FDG PET/CT scans, stratified patients with advanced NSCLC into low- and high-risk groups according to their OS (P < 0.001). Low brain FDG uptake was significantly associated with a poor prognosis, regardless of the treatment received after PET/CT, except for targeted therapy (122 patients).

When studying the metabolism of healthy organs, the “tumor sink effect” should be considered. In this phenomenon, increased tumor burden, whether attributable to lesion size, number, or aggressiveness, reduces the availability of the tracer to healthy tissues, thereby reducing their uptake (22). We found only a weak correlation between brain [¹⁸F]FDG uptake and TMTV (r = −0.24; Supplemental Fig. 4) and showed that these 2 parameters were complementary to predict OS in NSCLC (Supplemental Fig. 10), demonstrating that the prognostic value of brain SUV_mean cannot be explained by the tumor sink effect.

Several assumptions can be made to explain our observations. One is a possible relationship between brain metabolism and systemic inflammation. A recent study demonstrated the existence of a circuit capable of detecting inflammation in the blood and regulating the antiinflammatory response through the brain, revealing a bidirectional communication between the brain and the immune system (23). Inflammation is one of the inevitable consequences of tumorigenesis, leading to the recruitment of inflammatory cells, such as monocytes, to tumor sites via the bloodstream (24). In our study, we observed moderate but significant negative correlations between brain SUV_mean and CRP or monocytes. The combination of brain SUV_mean with these features stratified patients into 3 groups with significantly different OS (P ≤ 0.002; Supplemental Fig. 17), supporting their complementary prognostic value. Overall, these findings suggest that cerebral glucose metabolism measured on PET images may reflect mechanisms involved in regulating the antiinflammatory response, which is associated with immune cell counts and inflammatory response proteins, such as CRP.

Another possible explanation could be a correlation between cerebral [¹⁸F]FDG uptake and the patient’s functional status, encompassing physical or cognitive components. Our analysis demonstrated that cerebral [¹⁸F]FDG uptake is weakly but significantly positively or negatively correlated with age, BMI, smoking status, PS, serum albumin levels, and blood glucose levels, consistent with previous studies that have individually reported associations between some of these parameters and cerebral uptake (25,26). Taken together, these findings suggest that brain SUV_mean may reflect the patient’s overall functional and nutritional statuses, which are known to be associated with clinical prognosis (27,28). Other studies using [¹⁸F]FDG PET/CT have reported a correlation between decreased cerebral glucose metabolism and psychologic conditions in patients with lung cancer (29). Patients with cancer face a higher risk of depression, with a prevalence influenced by sex, age, and cancer type (30). It has been reported that patients with lung cancer have lower brain [¹⁸F]FDG uptake than do patients with other cancer types and a higher incidence of depression and anxiety, particularly patients with metastases (29). In our study, patients with documented depressive disorders (5%) had lower brain [¹⁸F]FDG uptake than those without such disorders (95%), although the difference was not statistically significant (P = 0.33). Moreover, several studies have reported a link between patients with depressive syndromes and blood biomarkers associated with systemic inflammation (31–33) and between nutritional status and mental condition (34). In addition, advanced arteriosclerosis of the carotid and cerebral vessels could influence the availability of [¹⁸F]FDG to the brain and explain a low brain SUV_mean. This hypothesis cannot be excluded, as 86% of patients in our cohort were smokers; however, this could not be tested because of the lack of information regarding their vascular status. Further studies are needed to elucidate the main biologic mechanisms reflected by brain metabolism.

Furthermore, we investigated whether [¹⁸F]FDG uptake in individual brain structures provided additional prognostic information; however, regional brain metabolism of 8 major regions did not outperform average brain uptake for predicting prognosis (Fig. 3; Supplemental Fig. 11).

This study had several limitations. One such limitation was the significant difference between the composition of the discovery and test sets. We divided the cohort by PET scan date, introducing bias in the treatment selection and OS. Indeed, patients in the test cohort had access to more innovative treatments, such as immunotherapy. Yet, the models could stratify patients in the test set without adjusting the cutoff values. Because of the heterogeneity of treatments and the limited number of patients in most subgroups, the effect of combined therapies on cerebral metabolism could only be assessed in the subgroup receiving chemotherapy, radiotherapy, and immunotherapy—the largest subgroup with 108 patients—where the prognostic value of cerebral metabolism for OS remained significant (data not shown). Another limitation was the incomplete availability of biologic data, which may have limited the statistical power of some analyses. Finally, we retrospectively collected information on depression from patient medical records, which is incomplete and imperfect, as patients may not necessarily report a depressive disorder or may be unaware of it. To investigate the link between brain SUV_mean, patient’s functional status and survival, we will need to prospectively collect a score measuring depression at the time of baseline [¹⁸F]FDG PET/CT scan, such as the Hamilton Depression Rating Scale (35), possibly use connected devices to measure the patient’s physical condition and activity, and investigate the impact of concomitant medications. Our results warrant further studies to investigate the relationship between cerebral [¹⁸F]FDG uptake and the patient’s functional status in oncology. Depending on the results of these investigations, brain SUV_mean could potentially be used to identify patients who should be prioritized for supportive care interventions, such as psychologic support, nutritional counseling, tailored physical activity programs, or the prescription of a comprehensive cardiovascular evaluation.

CONCLUSION

Low brain [¹⁸F]FDG uptake was associated with an increased risk of mortality in patients with advanced NSCLC. Accounting for brain [¹⁸F]FDG uptake in addition to clinical and tumor-related radiomic features improved OS prediction compared with relying only on clinical and tumor-related radiomic features. Further studies investigating the role of brain [¹⁸F]FDG uptake as a prognostic biomarker for OS and its association with systemic inflammation and the patient’s functional status are warranted.

DISCLOSURE

This work was supported by the French National Research Agency (ANR-22-CE45-0001 NEMO-PET). Nicolas Girard reports a consulting or advisory role for Abbvie, Amgen, AstraZeneca, BeiGene, Bristol-Myers Squibb, Daiichi Sankyo/AstraZeneca, Gilead Sciences, Ipsen, Janssen, LEO Pharma, Lilly, MSD, Novartis, Pfizer, Roche, Sanofi, and Takeda Pharmaceuticals. Lalith Sundar and Thomas Beyer are cofounders of Zenta GmbH. No other potential conflict of interest relevant to this article was reported.

KEY POINTS

QUESTION: Could brain metabolic activity measured by [¹⁸F]FDG PET/CT serve as an independent prognostic factor for survival in patients with advanced non–small cell lung cancer?

PERTINENT FINDINGS: Low pretreatment brain [¹⁸F]FDG uptake was associated with an increased risk of mortality in patients with advanced NSCLC. Brain [¹⁸F]FDG uptake predicts survival regardless of the presence of brain metastases; it is independent of tumor volume and metabolic activity and is significantly correlated, yet complementary, to several clinical and blood biomarkers, including CRP.

IMPLICATIONS FOR PATIENT CARE: Brain [¹⁸F]FDG uptake is a prognostic biomarker for survival in patients with advanced NSCLC and may help identify patients who would benefit most from prioritized supportive care interventions.