Introduction
Pulmonary nodules (PNs) are small (≤ 3 cm) lesions surrounded by lung parenchymal tissue that are not transparent and not the results of atelectasis, mediastinal lymphadenopathy, or pleural effusion [
1‐
3]. In cases where these nodules are > 6 mm in size, computed tomography (CT)-based routine follow-up is warranted [
4], with a 1.1-fold increase in the risk of PN malignancy with each 1 mm increase in diameter [
5]. Analyses of patient clinical data and CT imaging findings are the most commonly used approach to PN diagnosis [
6‐
8].
CT features often indicative of PN malignancy include CT bronchus sign, vascular convergence sign, pleural retraction, lobulation, and spiculated sign [
6‐
8]. Clinical risk factors for PN malignancy include more advanced age, elevated serum levels of tumor marker proteins, and a history of smoking [
6,
9]. Researchers have devised an array of predictive models based on these clinical and imaging features with the goal of more reliably identifying malignant PNs [
6‐
8]. Most CT-derived imaging features, however, are classified as binary variables that can be inconsistently identified based on the experience level of the attending physician. More reliable quantitative imaging strategies are thus needed to minimize this potential for bias, thereby increasing the odds of accurately diagnosing PNs.
18F-fludeoxyglucose (
18F-FDG) positron emission tomography (PET)/CT scans have emerged as a powerful approach to PN diagnosis, with standardized maximum uptake values (SUV
max) serving as a proxy for radiotracer uptake on imaging scans [
10]. Given these advantages, researchers have also incorporated PET/CT imaging parameters into predictive models designed to diagnose PNs in an effort to achieve superior accuracy [
11‐
23]. However, there has been substantial variability among studies with respect to the purported diagnostic performance of these individual PET/CT-based models [
11‐
23]. There thus remains the pressing need for large-scale analyses capable of systematically clarifying the diagnostic utility of the models developed to date.
Accordingly, the present meta-analysis was conducted to clarify the diagnostic performance of PET/CT-based models when used for the differential diagnosis of potentially malignant PNs.
Materials and methods
Study selection
Studies of potential relevance were identified by searching the Web of Science, PubMed, and Wanfang databases for all articles published through September 2023 based on the following search strategy: (((((positron emission tomography) OR (PET/CT)) AND (model)) AND ((lung) OR (pulmonary))) AND (nodule)) AND ((((differential) OR (diagnosis)) OR (probability)) OR (predictive)). This meta-analysis was registered at
https://inplasy.com/ (No. INPLASY2023100042).
To be eligible for inclusion, studies had to be: (1) focused on the differential diagnosis of malignant or benign PNs, (2) centered on the development or testing of PET/CT-based models that were provided within the study, and (3) transparent with respect to the true positive (TP), true negative (TN), false positive (FP), and false negative (FN) values associated with the tested models. provided. Case reports, non-human studies, and reviews were excluded from this study.
Data extraction and quality analyses
Two investigators were responsible for independently extracting pertinent data from these studies, including baseline study data, baseline patient data, and the results of diagnostic analyses. Any discrepancies were resolved by a third investigator. The QUADAS-2 tool was used to gauge risk of bias [
24].
Definitions
TP results were those for which both PET/CT-based models and final diagnoses were indicative of PN malignancy, whereas FP results were those for which PET/CT-based models predicted that a given lesion was malignant but it was ultimately found to be benign. Conversely, TN results were those for which both PET/CT-based models and final diagnoses indicated that a PN was benign, whereas FN results were those for which PET/CT-based models predicted that a given lesion was benign but it was ultimately found to be malignant.
Stata v 12.0 (Stata Corporation, TX, USA) was used to compute pooled sensitivity, specificity, diagnostic score, negative likelihood ratio (NLR), positive likelihood ratio (PLR), and summary receiver operating characteristic (SROC) curves for this study. A given predictive model was considered to exhibit high diagnostic performance if it exhibited an NLR < 0.2 or a PLR > 5. An area under the SROC curve (AUC) value greater than 0.8 was also considered to indicate a high degree of diagnostic utility [
3]. RevMan v 5.3 was used to compare pooled SUV
max values between benign and malignant PNs. I
2 values were employed to gauge the degree of heterogeneity, with I
2 > 50% indicating that such heterogeneity was significant. The possibility of publication bias was assessed with Deeks’ funnel plots, and
P < 0.05 served as the threshold for defining statistical significance.
Discussion
The present meta-analysis explored the performance of PET/CT-based models as tools for the differential diagnosis of PNs. The overall pooled AUC value of 0.91 was indicative of excellent predictive performance in this context, while the low NLR value (0.15) demonstrates that these PET/CT-based models can satisfactorily diagnose benign PNs when predictive scores fall below the established cut-off value. As the pooled PLR value of 4.10 was less than 5, however, this suggests that the diagnostic ability of these PET/CT-based models for malignant PNs is only moderate when predictive scores fall above the established cut-off value.
PET/CT imaging can yield both CT images that offer morphological insight regarding a given lesion, as well as PET images capable of quantifying glucose metabolism rates. PET scans thus enable the detection of malignant lesions composed of highly metabolically active cells, given that they take up
18F-FDG and glucose at higher rates than do benign cells [
25,
26]. In the present meta-analysis, a significantly higher pooled SUV
max value was exhibited by malignant PNs as compared to benign PNs.
The diagnostic utility of individual CT features is relatively limited when evaluating PNs. In prior meta-analyses assessing the diagnostic performance of lobulation sign, calcification, and spiculation as approaches to differential diagnosis of PNs, the AUC values were between 0.65 and 0.76 [
1‐
3]. The AUC for the diagnostic utility of PET alone in the present study was 0.82, but the pooled specificity was just 51%. High levels of
18F-FDG uptake can also be observed for benign inflammatory, infectious, or granulatomous disease-associated lesions [
27], contributing to a relatively low PLR of 1.89. The comparison of diagnostic performance between the predictive model and PET alone suggests that the diagnostic ability of PET alone is limited when evaluating PNs, emphasizing the need to combine multiple signs in an effort to improve the performance of diagnostic models.
There are many advantages to utilizing mathematical models when diagnosing PNs. Notably, these models can ensure that patients can be assessed in a more objective manner, yielding a predictive score reflective of the odds of PN malignancy. In addition, these models can provide risk coefficients for all predictive factors incorporated therein, allowing researchers to directly establish the relative risk associated with each incorporated factor.
The Mayo model was the first predictive model designed to distinguish between benign and malignant PNs [
28]. Herder et al. [
29] combined the Mayo model with PET results to establish the first PET/CT-based model, which exhibited an AUC of 0.92 in line with the pooled AUC measured in the present meta-analysis. This AUC value was also higher than that of the Mayo model (0.79) or PET scanning results alone (0.88) [
29].
In addition to imaging features, predictive models can also incorporate levels of tumor markers or particular clinical features [
3]. More advanced age and higher serum concentrations of carcinoembryonic antigen have both been linked to a greater risk of PN malignancy [
3,
9]. While age was a factor that was included in most predictive analyses analyzed herein, none incorporated tumor markers. Additional research focused on developing new PET/CT-based predictive models incorporating clinical characteristics, imaging features, and tumor marker levels are thus warranted to improve diagnostic accuracy.
This meta-analysis is subject to certain limitations. For one, as all included studies were retrospective in design, these findings are subject to a high risk of bias. Moreover, many of the included studies failed to indicate whether patients were recruited consecutively, and this oversight may have influenced the diagnostic performance of the models developed in individual studies. Next, different models contained different predictive factors, and the diagnostic results were not only influenced by PET/CT, but also influenced by other factors. However, different models also have the similarity that the predictive models can provide the comprehensive and quanitative analysis for the PNs. Lastly, the included studies did not utilize consistent reference standards, again potentially impacting the resultant diagnostic accuracy.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.