Diagnostic accuracy of 18F-FDG PET/CT scan in distant metastasis staging of non-small cell lung cancer: A meta-analysis

Abstract

Background:

Accurate detection of distant metastasis (M stage) is pivotal in treatment planning for non-small cell lung cancer (NSCLC). 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG PET/CT) combines metabolic and anatomic information and may outperform conventional imaging for baseline M-staging. Hence, this review was done to determine the diagnostic accuracy of 18F-FDG PET/CT for distant metastasis of NSCLC.

Methods:

Following Preferred Reporting Items for Systematic Review and Meta-Analysis-Diagnostic Test Accuracy, we included studies of adults with NSCLC that reported per-patient diagnostic accuracy of 18F-FDG PET/CT for any distant metastasis against histology and/or composite reference standard. Two reviewers independently screened, extracted data, and appraised risk of bias with Quality Assessment of Diagnostic Accuracy Studies-2. Summary sensitivity and specificity were pooled using a bivariate random-effects model with summary receiver operator characteristics curve and likelihood ratios (LRs); heterogeneity, threshold effects, and small-study bias (Deeks test) were assessed.

Results:

Fifteen studies (298 reference-positive, 1915 reference-negative; pretest probability 13%) were included. Pooled sensitivity and specificity were 0.91 (95% confidence interval: 0.70–0.98) and 0.98 (0.96–0.99), area under the receiver operating characteristic 0.99 (0.97–0.99). LR⁺ and LR⁻ were 39.0 (21.5–70.8) and 0.09 (0.02–0.35), respectively; heterogeneity was substantial (I² = 87%). Deeks test showed no asymmetry (P = .48). With contrast-enhanced computed tomography (8 studies), sensitivity and specificity were 0.99 (0.86–1.00) and 0.98 (0.95–0.99), area under the receiver operating characteristic 1.00, with negligible heterogeneity.

Conclusion:

18F-FDG PET/CT provides substantial discrimination for distant metastasis in NSCLC, strongly informing treatment intent. Incorporating diagnostic contrast-enhanced computed tomography within PET/CT materially improves and stabilizes sensitivity; non-contrast protocols retain rule-in value but are less reliable to rule out metastasis.

Key Points.

Question: Can integrated 18F-FDG PET/CT – particularly with contrast-enhanced CT – reliably detect distant metastases at initial NSCLC staging, overcoming limitations of conventional imaging and minimizing futile curative-intent therapy?

Findings: Across 15 studies (n = 2213), pooled sensitivity 0.91 and specificity 0.98 (AUROC 0.99); with diagnostic CECT, sensitivity 0.99 and specificity 0.98, markedly reduced heterogeneity.

Clinical relevance statement: Baseline PET/CT preferably with contrast-enhanced CT confidently rules in or out extrathoracic disease, aligns patients to appropriate systemic versus curative-intent pathways, and reduces unnecessary procedures; non-contrast protocols require supplemental organ-specific imaging.

1. Introduction

Lung cancer remains the leading cause of cancer mortality worldwide, accounting for an estimated 1.8 million deaths annually; non-small cell lung cancer (NSCLC) constitutes majority of cases.^[1] Accurate definition of distant metastatic spread (M category) is pivotal because it determines prognosis and therapeutic intent. In the eighth TNM edition, the M descriptors were refined to distinguish single extrathoracic metastasis (M1b) from multiple extrathoracic metastases (M1c), reflecting clinically meaningful survival differences and potential for metastasis-directed strategies in strictly oligometastatic disease.^[2]

Conventional contrast-enhanced computed tomography (CECT) surveys the chest and abdomen efficiently but is limited by size and morphologic criteria alone for detecting small or metabolically indolent lesions. By combining metabolic information from 18F-fluorodeoxyglucose (18F-FDG) with anatomic localization, integrated positron emission tomography/computed tomography (PET/CT) improves overall staging accuracy and reduces discordance relative to PET or CT alone, thereby mitigating futile local therapies when unsuspected stage IV disease is present.^[3] Contemporary reviews aligned with the eighth TNM framework emphasize PET/CT’s central role in M-staging while underscoring site-specific strengths (e.g., bone, adrenal, liver) and known limitations (e.g., brain, where physiologic cortical uptake necessitates magnetic resonance imaging [MRI]).^[4]

Quantitative evidence supports high diagnostic performance of FDG PET/CT for detecting distant metastases in lung cancer. Meta-analysis across histologies reported pooled sensitivity and specificity of 0.93 and 0.96, respectively (area under the curve [AUC]: 0.98) for identifying extrathoracic disease, indicating substantial rule-in capability with strong overall accuracy.^[5] Focusing specifically on NSCLC at initial staging, subsequent meta-analysis of 10 studies found pooled sensitivity 0.81 and specificity 0.96 (AUC: 0.97), confirming consistent high specificity and management impact at time treatment plans are formulated.^[6]

Comparative evidence suggests that whole-body MRI (with or without diffusion-weighted imaging) can achieve diagnostic performance similar to PET/CT for M-staging, yet PET/CT remains the most widely available, standardized, and workflow-efficient whole-body modality in routine NSCLC pathways.^[7] Previous review reported pooled PET/CT sensitivity and specificity of 0.83 and 0.93 versus 0.92 and 0.93 for whole-body MRI, with no significant difference between techniques.^[8] More recent head-to-head analyses likewise show comparable sensitivity and specificity between PET/CT and PET/MRI for distant metastasis detection in NSCLC, indicating that MRI-based hybrids may be reasonable alternatives where available.^[9]

Recent evidence further extends this landscape to PET/MR. A 2024 diagnostic meta-analysis of whole-body 18F-FDG PET/MR reported high M-staging accuracy in NSCLC (pooled sensitivity: ~0.92; specificity: ~1.00), suggesting PET/MR can perform at the highest tier in advanced centers.^[10] Nevertheless, PET/CT remains the most broadly accessible hybrid platform in routine NSCLC pathways; therefore, defining modifiable protocol elements within PET/CT, particularly whether the CT component is diagnostic and contrast-enhanced, has immediate relevance for protocol standardization and for real-world multidisciplinary decision-making.

However, an important unresolved issue is that prior syntheses largely combined PET/CT studies performed with different CT protocols, ranging from low-dose, non-contrast CT used primarily for attenuation correction to diagnostic, CECT acquisitions. As a result, although the high accuracy of PET/CT is established, the incremental impact of incorporating diagnostic CECT on sensitivity and on between-study heterogeneity has not been systematically quantified. Clarifying this protocol-dependent effect is clinically relevant because the CT component is modifiable and may determine whether PET/CT can be relied upon not only to rule in, but also to confidently rule out, extrathoracic disease.

Precise M-staging is indispensable in NSCLC, and FDG PET/CT integrating whole-body metabolic and anatomic assessment consistently identifies extrathoracic disease with high specificity, influences treatment selection, and guides verification strategies. Building on established eighth-edition TNM principles and a robust evidence base across general lung cancer and NSCLC-specific cohorts, focused evaluation of PET/CT diagnostic accuracy for distant metastasis remains clinically salient and directly relevant to optimizing contemporary management.^[2,4,6,8] Hence, this review was done to determine the diagnostic accuracy of 18F-FDG PET/CT for distant metastasis of NSCLC.

2. Methods

2.1. Eligibility criteria

Population: adults with histologically or cytologically confirmed NSCLC undergoing initial staging before any systemic or local curative-intent therapy.

Index test: whole-body integrated 18F-FDG PET/CT performed as a single combined examination. Studies evaluating stand-alone PET only, PET with side-by-side CT (not integrated), or PET/MRI were excluded from the primary analysis.

Target condition (clinical role): presence of distant metastasis (M stage) as defined by the 8th TNM classification (e.g., adrenal, liver, bone, brain, extrathoracic nodal, and other organs). The intended clinical role of PET/CT was “replacement/triage” to detect M1 disease and avoid futile curative local therapy.

Reference standard: histopathological confirmation of at least 1 suspected metastatic site through biopsy. We anticipated that, in real-world staging studies, histopathologic confirmation would be more common after a positive/equivocal PET/CT result, whereas PET/CT-negative patients would more often be verified by imaging/clinical follow-up. This pattern represents differential verification (work-up) bias, which we considered during Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 appraisal and in interpreting pooled accuracy, particularly “rule-out” inferences.

Study designs: prospective or retrospective cohort/cross-sectional studies that permitted construction of a 2 × 2 table at the per-patient level (preferred). Case reports/series (<10 patients), reviews, editorials, and conference abstracts without extractable accuracy data were excluded.

Outcomes: sensitivity and specificity for detecting any distant metastasis at initial staging.

2.2. Search strategy

We searched MEDLINE (via PubMed), Embase, Web of Science Core Collection, Scopus, and Cochrane Library from inception to the search date, without language restrictions. Grey literature sources included ClinicalTrials.gov/ICTRP, ProQuest Dissertations, and major oncology/nuclear medicine conference proceedings. Reference lists of eligible papers and topic reviews were scanned for additional studies. A professional medical librarian refined the strategy using both controlled vocabulary and free-text terms for NSCLC, PET/CT, FDG, and metastasis. Complete strategies and date of final search will be reported in Appendix, Supplemental Digital Content, per Preferred Reporting Items for Systematic Review and Meta-Analysis-Diagnostic Test Accuracy.^[11]

2.3. Study selection

Two reviewers independently screened titles/abstracts, then full texts, using pre-piloted forms. Disagreements were resolved by consensus or third-reviewer adjudication. When multiple reports addressed overlapping cohorts, we preferentially retained the most complete or recent dataset to avoid double-counting. Reasons for exclusion at full text were documented, and the Preferred Reporting Items for Systematic Review and Meta-Analysis-Diagnostic Test Accuracy flow diagram will summarize selection.^[11]

2.4. Data extraction

Two reviewers independently extracted data using piloted template: study characteristics (year, country, design, sample size), patient demographics, tumor characteristics, details of the index test (scanner type, tracer dose, uptake time, CT contrast use, reconstruction, reading method, SUV thresholds or qualitative criteria, reader expertise/blinding), reference standard (histology vs composite reference standard; follow-up duration), timing between index and reference, and outcomes. Uninterpretable/indeterminate scans were extracted explicitly; when needed, corresponding authors were contacted for clarifications or 2 × 2 data. In the included primary studies, indeterminate/uninterpretable PET/CT results were infrequently reported as a separate outcome category. Where equivocal findings were described, we extracted the final binary classification used by the study after applying its prespecified interpretive criteria and/or verification process. If indeterminate/uninterpretable results were reported separately and could not be allocated to true positive/false positive/false negative/true negative cells, they were treated as missing for 2 × 2 construction (complete-case extraction); authors were contacted when clarification was required.

2.5. Risk of bias and applicability

Methodological quality was appraised with QUADAS-2 across 4 domains (Patient Selection, Index Test, Reference Standard, Flow/Timing) and 3 applicability domains, using signaling questions tailored to this review (e.g., consecutive recruitment, prespecified positivity criteria, blinding, acceptable reference standard, short index-to-reference interval). Each domain was judged low, high, or unclear risk of bias by 2 independent reviewers.^[12]

2.6. Outcomes and summary measures

Primary outcomes were study-level sensitivity and specificity for detecting any distant metastasis. Secondary outcomes included positive/negative likelihood ratios (LRs), and diagnostic odds ratio (DOR). Confidence intervals (95% CIs) for proportions used the exact/binomial method.

2.7. Synthesis of results

We fitted a bivariate random-effects model to jointly pool logit-transformed sensitivity and specificity, accounting for their correlation and between-study heterogeneity.^[12] Summary operating points with 95% CIs and 95% prediction regions were displayed on the summary receiver operator characteristics curve (SROC) plot. For robustness, we also estimated the hierarchical summary receiver operating characteristic model (Rutter & Gatsonis).^[4] From the bivariate model, we have reported between-study variances (τ²_sens and τ²_spec) and the between-study correlation (ρ), as requested by the authors’ heterogeneity-reporting preference. Where cells contained zeros, continuity correction of 0.5 was applied only when required by estimation routine.^[13,14] All pooled estimates were based on study-level 2 × 2 tables reflecting the study’s final binary read of PET/CT, rather than a separate indeterminate category. A LR scattergram was also plotted for determining clinical usefulness of the imaging technique.

Heterogeneity was examined using statistical measures such as bivariate boxplot and I² statistic for heterogeneity. We assessed threshold effects using the Spearman correlation between logit(sensitivity) and logit(1 – specificity) and by visual inspection of SROC shape. Intraclass correlation coefficient (ICC) values are reported to summarize the proportion of total variability attributable to between-study heterogeneity for sensitivity and specificity; pooled sensitivity/specificity from the bivariate model remains the primary summary accuracy estimates. We performed separate subgroup analysis for use of CECT within PET/CT. Results are presented as relative DORs and shifts in summary sensitivity/specificity with 95% CIs.^[13,14]

When ≥ 10 studies are pooled, we have examined small-study effects with Deeks funnel-plot asymmetry test (effective sample size [ESS]–weighted regression of ln[DOR] against 1/√ESS). P value < .10 suggests potential asymmetry; interpretation will consider heterogeneity/threshold effects.^[15] In addition, to formally explore prespecified sources of heterogeneity, we performed univariable meta-regression/subgroup analyses within the bivariate random-effects framework for the following study-level covariates: study design (prospective vs retrospective), overall risk of bias (QUADAS-2 overall judgement: low vs high), and index-test category (PET/CT with diagnostic CECT vs non-contrast/low-dose CT). Meta-regression for reference standard type was planned a priori; however, all included studies used histopathologic/cytologic confirmation as the reference standard, precluding meaningful between-study comparisons on this covariate. Analyses will be conducted in the Stata “metandi/midas” package. Graphical outputs (forest plots of sensitivity/specificity, coupled forest plots, SROC with prediction regions, Deeks plot) will be generated accordingly.

3. Results

3.1. Search results

We identified 3273 records from databases and removed 928 duplicates, leaving 2345 unique records for screening. After title/abstract screening, 2192 were excluded, and 153 full texts were assessed. Of these, 138 were excluded (not NSCLC populations, n = 61; not 18F-FDG PET/CT, n = 49; no diagnostic accuracy measures, n = 28). Ultimately, 15 studies met the eligibility criteria and were included in the review (Fig. 1).^[16–30]

Figure 1.

PRISMA flowchart. 18F-FDG = 18F-fluorodeoxyglucose, PET/CT = positron emission tomography/computed tomography, PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

3.2. Characteristics of the included studies

Fifteen studies (Table 1) published between 2003 and 2020 were included, spanning Asia (n = 6; Korea and Japan), Europe (n = 7; Germany, Switzerland, Poland, Belgium, Spain), North America (n = 1; USA), and Africa (n = 1; Egypt). Most were conducted in tertiary university hospitals; 1 was multicenter. Study designs were predominantly prospective comparative or cohort evaluations, with 2 retrospective comparative series and 1 cross-sectional analysis. Across studies, participants were adults with suspected or histologically confirmed NSCLC undergoing baseline staging; 2 reports focused on specific clinical spectra (1 cohort enriched for subsolid tumors ≤ 3 cm and 1 restaging cohort of stage IIIa/IIIb disease). Sample sizes ranged from 27 to 855 patients (median 103; total N = 2493), with a male predominance of ~56% and median ages typically in the 6th to 7th decade. All studies assessed integrated, dual-modality 18F-FDG PET/CT as the index test. Eight examinations incorporated a diagnostic CECT component, while 7 used non-contrast CT primarily for attenuation correction and localization.

Table 1.

Characteristics of the included studies (N = 15).

Study (yr)	Region	Study design	Characteristics of participants	Median age (yr)	Sample size	Gender (M/F)	Study setting	Index test	Risk of bias
Antoch (2003)	Essen, Germany	Comparative accuracy study	Patients with pathologically confirmed NSCLC	Men: 57 yr (range: 39–70); women: 48 yr (range: 40–57)	27	23 M/4 F	University Hospital Essen	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	High
Cerfolio (2004)	Birmingham, Alabama, USA	Prospective blinded comparative trial	Consecutive patients with NSCLC (indeterminate pulmonary nodule or biopsy-proven), aged 19 yr or greater, undergoing staging/restaging	Median 66 (range: 24–87)	129	77 M/52 F	University of Alabama at Birmingham and Birmingham Veterans Administration Hospital	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	Low
Huellner (2015)	Zurich, Switzerland	Prospective comparative study	Patients with biopsy-proven or suspected NSCLC for initial staging	Median 65 (range: 35–89)	42	29 M/13 F	University Hospital Zurich	Integrated/dual-modality 18F-FDG PET/CT (with non-contrast-enhanced CT component)	Low
Lee (2009)	Seoul, Korea	Prospective study	Patients with histopathologically proven lung adenocarcinoma	54 (range: 23–88)	442	238 M/204 F	Samsung Medical Center, Sungkyunkwan University School of Medicine	Integrated/dual-modality 18F-FDG PET/CT (with non-contrast-enhanced CT component)	Low
Lee (2016)	Seoul, Korea	Prospective comparative study	Patients with proven or radiologically suspected lung cancer appearing resectable on CT, ultimately confirmed as NSCLC	62.9 (range 35–79)	45	26 M/19 F	Seoul National University Hospital	Integrated/dual-modality 18F-FDG PET/CT (with non-contrast-enhanced CT component)	Low
Ohno (2015)	Kobe, Japan	Prospective comparative study	Patients with pathologically confirmed NSCLC	Overall: 72.0; men: 71.6; women: 72.4.	140	75 M/65 F	Kobe University Hospital	Integrated/dual-modality 18F-FDG PET/CT (with non-contrast-enhanced CT component)	Low
Ohno (2008)	Kobe, Japan	Prospective comparative study	Patients with pathologically confirmed NSCLC	Overall: 72; men: 72; women: 72	203	109 M/94 F	Kobe University Hospital	Integrated/dual-modality 18F-FDG PET/CT (with non-contrast-enhanced CT component)	High
Plathow (2008)	Germany (multiple centers)	Prospective comparative study	Patients with histologically proven NSCLC, stage IIIa or IIIb, for restaging	62 (range: 49–71)	52	36 M/16 F	Multiple University Hospitals	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	Low
Yi (2008)	Seoul, Korea	Prospective comparative study	Patients with pathologically confirmed NSCLC	61 (range: 34–82)	165	125 M/40 F	Samsung Medical Center, Sungkyunkwan University School of Medicine	Integrated/dual-modality 18F-FDG PET/CT (with non-contrast-enhanced CT component)	High
Opoka (2013)	Warsaw, Poland	Prospective comparative study	Patients with NSCLC (tumor size > 10 mm)	64.2 (range: 41–88)	99	70 M/29 F	National Tuberculosis and Lung Diseases Research Institute, Warsaw; and affiliated PET-CT centers	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	Low
Suh (2020)	Seoul, Korea	Retrospective comparative study	Patients with pathologically confirmed NSCLC; subsolid nodules with solid portion ≤3 cm	61.0 (IQR: 54.3–68.0)	855	335 M/520 F	Seoul National University Hospital	Integrated/dual-modality 18F-FDG PET/CT (with non-contrast-enhanced CT component)	High
El-Hariri (2012)	Egypt (Zagazig)	Prospective comparative study	Patients with lung cancer undergoing staging	64 (range: 34–76)	33	28 M/5 F	Zagazig University, Egypt	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	Low
De Wever (2006)	Leuven, Belgium	Retrospective comparative study	Patients with suggestive lung lesions for staging, primarily NSCLC, some benign	64 (range: 26–83); women 60 (range: 46–72)	50	44 M/6 F	University Hospitals Gasthuisberg, Leuven	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	High
Gómez León (2014)	Madrid, Spain	Cross-sectional study	Patients with histologically confirmed NSCLC, unknown disease stage, and no prior treatment	68 (SD: 10, range: 46–83)	103	90 M/13 F (87.4% M/12.6% F)	La Princesa University Hospital, Madrid, Spain	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	High
Rodríguez Fernández et al (2007)	Granada, Spain (Eastern Andalusia)	Prospective cohort study	Patients with suspected potentially resectable NSCLC	63	108	101 M/7 F	University Hospital, Granada, Spain	Integrated/dual-modality 18F-FDG PET/CT (with diagnostic contrast-enhanced CT component)	Low

18F-FDG = 18F-fluorodeoxyglucose, CT = computed tomography, F = female, IQR = interquartile range, M = male, NSCLC = non-small cell lung cancer, PET/CT = positron emission tomography/computed tomography, PET/MR = positron emission tomography/magnetic resonance imaging, PET = positron emission tomography, SD = standard deviation.

3.3. Risk-of-bias assessment

Risk-of-bias appraisal at the study level was low in 9 studies and high in 6 (Fig. 2). High-risk judgments most commonly reflected concerns about patient selection (enrichment or restaging spectra), partial/differential verification (histology more often triggered by positive/equivocal findings, with negatives verified primarily by follow-up), or insufficient reporting of index-test thresholds and blinding. Notably, the group employing diagnostic CECT contained both low- and high-risk studies but, as shown in the quantitative synthesis, displayed markedly less performance variability than the non-contrast group.

Figure 2.

Quality assessment of the included studies (traffic light plot).

3.4. Diagnostic accuracy of 18F-FDG PET/CT for distant metastasis of NSCLC

Across 15 studies (298 reference-positive, 1915 reference-negative; pretest probability = 0.13), pooled sensitivity was 0.91 (95% CI: 0.70–0.98) and specificity 0.98 (0.96–0.99), with area under the receiver operating characteristic (AUROC) = 0.99 (0.97–0.99; Fig. 3). The model yielded LR⁺ = 39.0 (21.5–70.8), LR⁻ = 0.09 (0.02–0.35) and DOR = 423 (85–2103), indicating substantial discrimination and strong rule-in/rule-out capability. Heterogeneity was substantial (likelihood ratio test [LRT]_Q = 15.49, P < .001; I² = 87% [74–100]) and this was further confirmed by bivariate boxplot (Fig. 4).

Figure 3.

SROC curve for diagnostic accuracy of 18F-FDG PET/CT of distant metastasis for non-small cell lung cancer. 18F-FDG = 18F-fluorodeoxyglucose, AUC = area under the curve, PET/CT = positron emission tomography/computed tomography, SROC = summary receiver operator characteristics curve.

Figure 4.

Bivariate boxplot to check the heterogeneity.

Correlation between sensitivity and specificity was small (0.09), and the proportion attributable to threshold effects was minimal (0.01). Interstudy variation was larger for sensitivity than for specificity (ICC_SEN = 0.58 vs ICC_SPE = 0.15), indicating that heterogeneity was driven primarily by variability in sensitivity, while specificity was comparatively stable across studies.

LR scattergram (Fig. 5) showed that the point estimates were positioned in the left upper quadrant, indicating that 18F-FDG PET/CT can be used for both confirmation and exclusion of distant metastasis among patients with NSCLC. With cohort pretest probability of 13%, Fagan nomogram (Fig. 6) shows that positive FDG PET/CT (LR⁺ = 39) raises posttest probability of distant metastasis to ~85%. Conversely, negative scan (LR⁻ = 0.09) lowers posttest probability to ~1%. Thus, PET/CT provides strong rule-in capability and substantial rule-out value for distant metastasis at this baseline risk; however, these posttest probabilities should be interpreted within the full staging pathway, and do not remove the need for confirmatory testing in scenarios where management would change decisively (like denial of curative-intent therapy). Deeks funnel plot (Fig. 7) showed no significant asymmetry (P = .48), suggesting little evidence of publication bias or small-study effects. Distribution of ln(DOR) versus 1/√ESS appears roughly symmetric around shallow regression slope, indicating pooled accuracy estimates are unlikely to be inflated by selective reporting.

Figure 5.

Likelihood ratio scattergram. LLQ = left lower quadrant, LRN = likelihood ratio negative, LRP = likelihood ratio positive, LUQ = left upper quadrant, RLQ = right lower quadrant, RUQ = right upper quadrant.

Figure 6.

Fagan’s nomogram.

Figure 7.

Funnel plot with Deeks test to check publication bias. ESS = effective sample size.

3.5. Subgroup analysis

PET/CT with CECT (8 studies; pretest = 17%): pooled sensitivity was 0.99 (95% CI: 0.86–1.00) and specificity 0.98 (0.95–0.99), yielding LR⁺ 54.4 (20.5–144.4), LR⁻ 0.01 (0.00–0.16), and DOR 3930 (233–66,327); AUROC was 1.00 (0.99–1.00), reflecting rounding of a near-ceiling estimate rather than literal perfect discrimination. Statistical heterogeneity was not detected (LRT_Q = 0.31; I² ≈ 0%), but precision was limited given the small subgroup (95% CI for I²: 0–100); ICCs suggested residual between-study variation (ICC_SEN = 0.34; ICC_SPE = 0.16).

PET/CT without CECT (7 studies; pretest = 12%): pooled sensitivity was 0.62 (95% CI: 0.42–0.79) and specificity 0.97 (0.94–0.98), with AUROC 0.95 (0.93–0.97), LR⁺ 20.5 (12.1–34.6), LR⁻ 0.39 (0.24–0.64), and DOR 53 (28–99). Heterogeneity was substantial (LRT_Q = 8.87, P = .006; I² = 77%), driven largely by threshold effects (correlation = −0.91; proportion due to threshold = 0.83).

3.6. Meta-regression

Univariable meta-regression indicated that index-test category significantly moderated diagnostic performance and explained a substantial proportion of between-study heterogeneity, primarily through effects on sensitivity (sensitivity: 0.31 [95% CI: 0.13–0.57], P < .001; joint model LRT χ² = 14.11, P < .001). In contrast, study design (sensitivity: 0.90 [0.21–1.00], P = .92; joint model P = .18) and overall risk of bias (sensitivity: 0.98 [0.70–1.00], P = .35; joint model P = .38) were not significant moderators. For specificity, estimates remained uniformly high across covariate strata and did not show statistically significant moderation (all P > .05). These findings are summarized in Figure 8.

Figure 8.

Univariable meta-regression assessing the impact of study design, overall risk of bias, and index-test category on pooled sensitivity and specificity (bivariate random-effects framework). CI = confidence interval.

4. Discussion

This meta-analysis shows that 18F-FDG PET/CT has substantial overall diagnostic performance for detecting distant metastasis in NSCLC at initial staging. Using 15 studies with more than 2000 reference-verified participants, the pooled summary point sits very close to the upper-left corner of the SROC space, with high sensitivity and very high specificity.^[16–30] LR framework confirms clinical utility of these estimates: positive scan moves typical patient with modest pretest probability into decisively metastatic posttest range, while negative scan drives the posttest probability down to level that effectively excludes disseminated disease in most scenarios.^[31] Fagan nomogram constructed from summary LRs makes this tangible, converting 13% baseline risk into ~85% after a positive result and ~1% after negative result. Fagan nomogram constructed from summary LRs makes this tangible, converting a 13% baseline risk into ~85% after a positive result and ~1% after a negative result. Clinically, these figures are best interpreted as probabilistic decision-support: a positive PET/CT that would upstage disease and preclude curative-intent therapy typically warrants confirmatory evaluation (biopsy when feasible and/or targeted anatomic imaging), and a negative PET/CT does not replace mandatory organ-specific staging investigations where PET/CT has known limitations.

In particular, intracranial metastases are not reliably excluded by FDG PET/CT due to physiologic brain uptake; therefore, brain MRI remains necessary where recommended by contemporary NSCLC staging pathways, irrespective of PET/CT findings. Likewise, when PET/CT identifies lesions that would change intent from curative to palliative, residual uncertainty (even at high posttest probability) means that histologic confirmation and/or problem-solving imaging is often required before denying potentially curative surgery or definitive radiotherapy.

Despite tight summary performance, correlation between study-level sensitivity and specificity is near zero, and proportion of heterogeneity attributable to threshold effects is very small in the overall model, suggesting that variation is not primarily driven by different positivity cutoffs or interpretive thresholds at study level. Instead, it points toward differences in patient spectrum, reference standards, imaging protocols, and technology as dominant contributors.^[32] Consistent with this, ICCs show greater variability in sensitivity than in specificity, implying that case mix and technical/readout factors more strongly influence the ability to “find what is there” than ability to correctly dismiss what is not.

Prespecified subgroup analysis by CT technique clarifies 1 major driver of that heterogeneity. When PET/CT incorporated CECT, diagnostic performance became nearly ideal, with sensitivity and specificity both very high and the SROC area essentially maximal. Heterogeneity in this subgroup was negligible, supporting the consistency of that effect across contributing studies. In contrast, studies using non-contrast CT within PET/CT framework showed markedly lower sensitivity while maintaining high specificity, and heterogeneity was substantial. Moreover, negative correlation between sensitivity and specificity and large fraction of heterogeneity explained by threshold behavior in non-contrast subgroup suggest that interpretive rules and lesion-calling criteria were more variable across these studies. From clinical vantage point, these subgroup findings matter: adding contrast to the CT component converts PET/CT from an already strong test into test that is both uniformly consistent and near-perfect for identifying distant spread; omitting contrast preserves the ability to confidently confirm metastasis when present but compromises the test’s capacity to safely exclude it.^[33] Accordingly, a negative result on a non-contrast/low-dose PET/CT should not be used as the sole basis to rule out metastatic disease when committing a patient to curative-intent therapy. In such cases or when clinical suspicion remains, supplemental diagnostic imaging should be strongly considered, typically with CECT (where feasible) and/or organ-directed MRI (with brain MRI remaining mandatory where intracranial disease is a concern).

Nevertheless, these subgroup estimates should be interpreted cautiously. The CECT subgroup comprised only 8 studies, and when multiple studies report near-boundary accuracy (very high sensitivity/specificity), the SROC area can approach 1.00 and round to 1.00 without implying flawless discrimination in all real-world settings. Likewise, an I² estimate near zero in a small subgroup primarily indicates limited detectable heterogeneity and low power to identify between-study variance, consistent with the very wide I² CI and non-zero ICC values. Accordingly, the CECT findings should be framed as evidence of very high and comparatively consistent performance under diagnostic CECT protocols, rather than universal “perfect” accuracy.

The near-uniform, near-maximal performance observed when PET/CT incorporates diagnostic CECT in contemporary NSCLC management. Reliable identification of true stage IV disease is essential to prevent futile radical local therapy in patients with occult multi-metastatic burden (M1c), while equally enabling confident recognition of carefully selected patients with limited metastatic spread (e.g., M1b/oligometastatic patterns) who may be considered for metastasis-directed approaches within multidisciplinary pathways.^[34] The clinical value of protocol optimization therefore lies not only in improving average accuracy, but in tightening the decision boundary between curative-intent strategies, systemic therapy-first approaches, and integrated metastasis-directed treatment where appropriate.

Positioning these findings alongside evolving hybrid platforms is informative. In a meta-analysis by Mirshahvalad et al, whole-body FDG PET/MR demonstrated high diagnostic performance for NSCLC (sensitivity 0.92; specificity ~1.00).^[10] Our CECT-PET/CT subgroup achieved similarly high and in sensitivity, near-ceiling performance with very low heterogeneity, reinforcing that PET/CT can remain a “best-available” staging tool in most settings when the CT component is diagnostic and contrast-enhanced. In practice, optimizing PET/CT protocols may deliver PET/MR-like staging confidence while leveraging PET/CT’s wider availability and workflow integration.

Importantly, our univariable meta-regression further supports this interpretation: differences in study design and overall QUADAS-2 risk-of-bias classification did not materially explain heterogeneity, whereas index-test category remained a statistically significant moderator, driven by sensitivity differences. This reduces concern that the observed contrast-related performance differences are primarily attributable to confounding by design or study quality. Although our protocol allowed composite clinical follow-up as a reference standard, all included studies relied on histopathologic/cytologic confirmation, which limited our ability to evaluate reference-standard effects as a heterogeneity source.

Our findings align with and extend existing evidence based on the role of FDG PET/CT in lung cancer staging. Earlier meta-analyses that pooled lung cancer cohorts (often mixing NSCLC and small-cell lung cancer or not isolating the M-component) consistently reported very high specificity and high sensitivity for PET/CT when assessing extrathoracic disease, with SROC areas approaching unity.^[35] Present analysis restricted to NSCLC and focused explicitly on distant metastasis at baseline confirms those broad conclusions while providing more clinically precise estimate for population and decision point most relevant to thoracic oncology practice. Magnitude of our pooled LRs is comparable with, or higher than, those reported in earlier syntheses, reaffirming that PET/CT meaningfully shifts posttest probabilities and alters management decisions.^[36]

Compared with earlier syntheses,^[35,36] the principal incremental contribution of the present review is that we did not treat PET/CT as a single uniform modality; instead, we prospectively separated protocols by CT technique and demonstrated that this protocol choice is a major determinant of both sensitivity and heterogeneity. In other words, prior meta-analyses establish that PET/CT is accurate on average, whereas our results clarify when and why performance becomes near-uniform (diagnostic CECT) versus variably sensitive (non-contrast/low-dose CT). This distinction has direct implications for how staging PET/CT should be specified and interpreted in routine NSCLC pathways.

Where this review adds information toward handling technology and protocol differences that earlier work rarely separated. Historically, studies spanned multiple PET generations (from non–time-of-flight systems to modern digital detectors) and used CT largely for attenuation correction and localization, with optional contrast at the discretion of the center. As a result, meta-analytic outputs blended fundamentally different PET/CT executions.^[37] By stratifying on contrast use, our results demonstrate that much of variability seen across literature can be explained by CT technique. Near-perfect and highly consistent performance observed when contrast is administered supports the growing view in contemporary practice that staging PET/CT should, whenever feasible, include diagnostic-quality CECT from skull base to mid-thigh. This resonates with institutional series and guideline commentaries that have long argued for the added value of contrast in liver, adrenal, and peritoneal evaluation while acknowledging test’s inherent limitation in brain where cortical FDG uptake masks small metastases and dedicated MRI remains necessary.^[38]

Finally, the implications of improved M-staging accuracy extend beyond diagnosis alone. PET-derived quantitative metrics also inform risk stratification; for example, a 2017 meta-analysis showed that higher pretreatment SUVmax predicts poorer overall risk of distant metastases in early-stage NSCLC patients treated with stereotactic body radiotherapy.^[39] In this context, ensuring that baseline staging is maximally sensitive particularly through diagnostic CECT integration helps align biologically aggressive disease to appropriate systemic or combined-modality strategies and reduces the risk that patients with occult metastatic disease are inadvertently managed with definitive local therapy alone.

Our non-contrast subgroup showed mixed results reported by older cohorts in which the CT portion was low-dose, non-enhanced, and optimized primarily for attenuation correction. Those studies frequently reported on false-negative hepatic or peritoneal metastases and on challenges differentiating benign adrenal adenomas from metastases when relying on PET uptake alone.^[40] Pattern we observe, high specificity with lower sensitivity, parallels those narratives: metabolic information continues to be exceptionally effective for ruling in disease when avid foci are present, but absence of contrast degrades lesion conspicuity and anatomical characterization, leading to missed or indeterminate findings. Strong evidence of threshold-driven heterogeneity in the non-contrast subgroup is also consistent with earlier reports that varied in how equivocal uptake was handled (e.g., SUV cutoffs vs purely qualitative reads) and whether small subcentimeter lesions were called positive without morphological corroboration.^[41]

Further previous research work is on the relationship between disease prevalence and clinical utility. Earlier series often originated from tertiary surgical centers with low metastatic prevalence, which reduced positive predictive value and risked spectrum bias.^[42] By reporting LRs and using Fagan analysis anchored at realistic baseline probability, our synthesis maintains generalizability across settings. Resulting posttest probabilities are similar to decision thresholds embedded in many multidisciplinary pathways, in which clearly positive PET/CT redirects patients away from futile thoracotomy or definitive chemoradiation and toward systemic therapy or clinical trials, whereas convincingly negative study supports proceeding to curative-intent local treatment once brain MRI excludes intracranial disease.^[43]

The performance profile observed for FDG PET/CT reflects complementary strengths of metabolic and anatomic imaging in the biological context of NSCLC dissemination. FDG uptake is driven by overexpression of glucose transporters and hexokinase activity in most lung cancer clones, enabling detection of metabolically active deposits that may still be morphologically subtle. CT component then anchors these foci to anatomy, allowing confident attribution to organ compartments and differentiation from physiologic uptake.^[44] This synergy is most pronounced when CT is diagnostic quality and contrast-enhanced. Intravenous contrast improves parenchymal lesion conspicuity in liver, delineates adrenal morphology and wash-in patterns, outlines peritoneal and omental soft-tissue thickening, and sharpens borders of small pleural or chest wall deposits. In those settings, faintly avid or borderline-size metastases that would otherwise remain equivocal on low-dose non-contrast CT become visible, explaining sharp rise in sensitivity and the disappearance of heterogeneity when contrast is used.^[45]

Sites where contrast makes the largest difference are also those with frequent NSCLC spread. The liver and adrenals are common destinations for hematogenous metastasis; both organs are susceptible to interpretive ambiguity on PET alone because benign entities, steatosis, hemangiomas, and lipid-rich adenomas, may be non-avid or mildly avid. Contrast resolves part of this ambiguity by supplying morphologic criteria and, when multiphasic techniques are used clinically, vascular behavior that favors or argues against metastasis.^[46] In peritoneum and mesentery, where small nodules can be metabolically heterogeneous, enhancement accentuates serosal outlines against surrounding fat. Conversely, FDG PET is intrinsically limited in the brain by intense cortical uptake; therefore, whole-body advantage of PET/CT never extended to intracranial metastasis, where dedicated brain MRI remains mandatory regardless of PET/CT protocol.^[47,48] These organ-specific considerations map closely to performance gradients observed between contrast-enhanced and non-contrast examinations.

Several limitations warrant emphasis. First, although we performed a priori subgrouping by CT technique and additional univariable meta-regression, the number of studies within each subgroup was modest, which reduces statistical power to detect effect modification, yields wider CIs, and limits the stability of heterogeneity estimates. Accordingly, subgroup differences and moderator tests should be interpreted as supportive rather than definitive, and additional large, protocol-stratified datasets would improve precision. Second, most included studies reported distant metastasis as a composite (any-site) outcome and did not provide organ-specific or lesion-level 2 × 2 tables; therefore, we could not estimate site-specific accuracy (e.g., adrenal vs bone vs liver) or assess whether the incremental value of diagnostic CECT varies by metastatic site. Future prospective studies should report standardized organ-level results and prespecify handling of equivocal findings to enable clinically granular accuracy estimates. Subgroup summaries, particularly for the CECT subgroup, may be optimistic or unstable because several included studies reported extremely high accuracy and some 2 × 2 tables contained sparse/zero cells, which can push model-based estimates toward boundary values and inflate metrics such as DOR and AUROC. We therefore emphasize CIs and interpret these results as indicating near-ceiling performance under diagnostic CECT protocols, not literal perfect discrimination.

Histopathologic confirmation for every suspected metastatic site is rarely feasible; many cohorts used composite reference standards with imaging follow-up, which risks partial verification and misclassification bias. More specifically, several studies exhibited differential verification (work-up) bias, wherein PET/CT-positive or equivocal findings preferentially underwent biopsy, while PET/CT-negative patients were verified mainly through clinical/imaging follow-up. If follow-up is incomplete in duration or intensity, occult metastases in PET/CT-negative patients may be missed, which can inflate specificity and the apparent “rule-out” performance; conversely, selective verification of only the most suspicious lesions can also distort sensitivity estimates. We were unable to quantitatively adjust for this bias because most primary studies did not report verification pathways in sufficient detail (e.g., biopsy rates by PET/CT result, organ-specific verification, and standardized follow-up intervals) to reconstruct alternative 2 × 2 tables or perform bias-adjusted sensitivity analyses.

5. Conclusion

Current evidence demonstrates that FDG PET/CT is a highly effective modality for detecting distant metastasis in NSCLC at presentation. When paired with diagnostic CECT, performance becomes both exceptionally high and remarkably consistent across settings, allowing clinicians to confirm or exclude extrathoracic disease with confidence and to align treatment intent accordingly. Non-contrast PET/CT retains value as rule-in test but cannot reliably rule out metastasis, underscoring the need for supplemental organ-specific imaging or biopsy when clinical pretest probability is nontrivial.

Author contributions

Data curation: Guangda Wang, Monica Karunakaran, Suganya K, Rajan Rushender, Qi Lv.

Formal analysis: Guangda Wang, Monica Karunakaran, Suganya K, Rajan Rushender, Qi Lv.

Investigation: Guangda Wang, Monica Karunakaran, Rajan Rushender, Qi Lv.

Methodology: Guangda Wang, Monica Karunakaran, Suganya K, Rajan Rushender, Qi Lv.

Software: Guangda Wang, Suganya K, Qi Lv.

Supervision: Guangda Wang, Monica Karunakaran, Suganya K, Qi Lv.

Visualization: Guangda Wang, Monica Karunakaran, Rajan Rushender, Qi Lv.

Conceptualization: Monica Karunakaran.

Project administration: Suganya K, Rajan Rushender, Qi Lv.

Validation: Rajan Rushender.

Resources: Qi Lv.

Writing – original draft: Guangda Wang.

Writing – review & editing: Guangda Wang, Monica Karunakaran, Suganya K, Rajan Rushender, Qi Lv.