Validation of a blood-based autoantibody test to assess lung cancer risk in 4-30 mm pulmonary nodules: a retrospective pooled analysis of four cohort studies.

1.
Introduction
Risk stratification of indeterminate pulmonary nodules (IPN) is an increasing clinical challenge with the rising use of chest computed tomography (CT), leading to an estimated 1.6 million new nodules identified annually [1]. Previous studies have shown that approximately 31% of chest CT scans identify a lung nodule, with the probability of malignancy for these nodules spanning a broad, heterogeneous range dependent on factors like size, morphology, and patient characteristics (e.g., age, smoking history) [1,2]. Given their frequency and potential clinical significance, accurately risk-stratifying IPNs across the size spectrum is critical for guiding appropriate follow-up and management strategies. In practice, referring all patients with pulmonary nodules for specialist evaluation or nodule clinic review is rarely feasible, and doing so would strain already limited pulmonary and radiology resources. Clinicians must therefore balance the need to identify cancers promptly while avoiding both unnecessary invasive procedures and overuse of specialist services. Additional noninvasive tools, such as blood-based tests that refine malignancy risk estimates, may help physicians evaluate pulmonary nodules more accurately and consistently, while focusing specialist resources on the highest-risk patients [3–5].
In clinical practice, pulmonary nodule management relies on estimating the probability of malignancy (pCA) by physician assessment or using validated risk models, such as the Mayo Clinic calculator, which integrates clinical and radiographic features [6,7]. Guidelines recommend surveillance for low-risk nodules and invasive evaluation for high-risk nodules, with intermediate-risk nodules often requiring a combination of additional imaging with 18F-fluorodeoxyglucose positron emission tomography (FDG-PET), serial CT imaging, and/or biopsy [7,8]. However, real-world management frequently deviates from guidelines, and decisions remain challenging when estimated risk is intermediate or when small nodules have concerning features [9,10]. A key limitation of current approaches is that nodule diameter is a commonly used feature for risk stratification and is a primary driver of follow-up recommendations in many lung nodule clinical guidelines, including Fleischner, Lung-RADS, and the NCCN lung cancer screening; however, size alone fails to capture the full heterogeneity of malignancy risk across the nodule spectrum [8,11,12]. Smaller nodules are often surveilled due to a lack of high-risk features, whereas larger nodules more readily trigger invasive evaluation. With the advent of robotic bronchoscopy and structured nodule programs, the lower size limit for lung nodules considered for biopsy has decreased in real-world care. As such, the total number of nodules considered for biopsy will continue to grow. In this setting, a biomarker that performs consistently across nodule sizes could help prioritize high-risk patients for expedited workup even when nodules are small, while maintaining high specificity to avoid unnecessary procedures. For such a test to be clinically useful across diverse practice settings, its performance must remain robust.
A blood-based autoantibody test (AAT) has been developed to detect autoantibodies to lung cancer-associated antigens via enzyme-linked immunosorbent assay (ELISA) and has demonstrated high specificity for distinguishing lung cancer from benign conditions [13–16]. The AAT evaluates for the presence of autoantibodies targeting the tumor antigens p53, NY-ESO-1, MAGE-A4, GBU4-5, CAGE, HuD, and SOX2, which have been shown to be elevated in patients with lung cancer [13,17–19]. The test has been clinically validated in patients with 4–30 mm nodules across the full spectrum of malignancy risk, where positive results were associated with increased lung cancer risk when combined with clinical risk models [14–16]. These studies show that the AAT can reclassify intermediate risk nodules (pCA 5–65%) into the high-risk category (pCA >65%), potentially accelerating the time to lung cancer diagnosis. For patients with a pretest pCA >65%, a positive AAT result may help support a shared decision to proceed with an invasive procedure, particularly when patients are frail or have significant comorbidities.
The goal of this pooled analysis is to further validate the performance of the AAT in a large, diverse clinical cohort of patients with 4–30 mm pulmonary nodules across the full spectrum of pretest malignancy risk. We evaluated 1,164 patients from four clinical study cohorts, including nodules as small as 4 mm, and assessed test performance. We hypothesized that the AAT would maintain high specificity across the entire 4–30 mm size range and that nodule diameter would not significantly affect test characteristics, supporting use of the test across the full spectrum of nodule sizes encountered in clinical practice.

2.
Methods
2.1.
Study design and cohorts
This pooled analysis integrated data from four observational pulmonary nodule studies – PANOPTIC, FHCC, ORACLE, and CLARIFY, each contributing harmonized clinical, imaging, and outcome data.
PANOPTIC (NCT01752114), The Pulmonary Nodule Plasma Proteomic Classifier, was a prospective observational, multicenter study enrolling unselected patients aged ≥40 years with CT-detected pulmonary nodules 8–30 mm [16,20]. Enrollment required a baseline CT within 60 days and specialist evaluation. Key exclusions included any prior diagnostic biopsy, earlier CT documenting the same nodule >60 days prior, and current or recent cancer within 2 years. The study received ethical approval from Quorum IRB (IRB #27115), and all participants provided written informed consent.
The Fred Hutchinson Cancer Center (FHCC) study was a prospective observational cohort enrolling consecutive patients evaluated for pulmonary nodules (solid and subsolid) at the FHCC outpatient clinic or University of Washington Thoracic Surgery clinic (2010–2020) [16]. For this pooled analysis, eligible patients with 4– <8 mm nodules were also included. The study received ethical approval from the FHCC Institutional Review Board (IRB#2242), and all participants provided written informed consent.
ORACLE (NCT03766958), An Observational Registry Study to Evaluate the Performance of the Nodify XL2™ Test, was a prospective observational registry across 15 community and academic pulmonary practices (2018–2020) [21]. Eligible patients were ≥40 years old with pulmonary nodules 8–30 mm and Mayo pretest malignancy probability ≤50%. Exclusions included any prior biopsy, physician-determined high cancer risk, or inability to complete followup. ORACLE received approval from Advarra IRB (IRB #Pro00034556), and all participants provided written informed consent.
CLARIFY (NCT06728319), A Multicenter, Retrospective, Chart Review Study Evaluating the Impact and Utility of the Blood Based Proteomic Integrated Classifier and AutoAntibody Tests in the Real World, is a retrospective observational study across clinical sites with extensive AAT/IC utilization. Eligible patients were ≥40 years old and had undergone AAT and IC testing at least one year prior to analysis to ensure adequate followup. CLARIFY received approval from Advarra IRB (IRB #Pro00081099), with a waiver of informed consent granted due to its retrospective design.
Across all studies, research activities were conducted in accordance with the Declaration of Helsinki.

2.2.
Clinical and imaging variable harmonization
Imaging, clinical, and outcome data were standardized to ensure full comparability across studies, enabling integration into a single pooled analysis dataset. Clinical variables, smoking history, and nodule characteristics were abstracted from case report forms, medical records, and radiology reports and harmonized using uniform definitions. Age was recorded in whole years, and smoking history was standardized to Current/Former or Never, based on documentation of ≥100 lifetime cigarettes in the medical record. Prior cancer history was coded as present or absent according to medical history. Imaging variables were aligned by recording the maximum nodule diameter in millimeters, spiculation status as present or absent based on explicit mention in the radiology report, and lobe location as upper lobe or other.

2.3.
Reference standard for nodule diagnosis
The reference standard for determining AAT performance was the final classification of each index nodule as malignant or benign. Malignant nodules (primary lung cancer) were definitively established by histopathology from an invasive biopsy and/or surgical resection. Tumor histology was standardized into three categories: non – small cell lung cancer, small cell lung cancer, or carcinoid tumor based on pathology reports. Benign nodules were defined by either (i) specific benign histopathology from an invasive procedure (e.g., granuloma, hamartoma, infection, fibrosis) or (ii) radiographic stability of nodule size and morphology on serial chest CT for at least 1 year or (iii) complete nodule resolution. Pulmonary nodules without histologically proven diagnoses or sufficient clinical follow-up to determine an outcome were excluded from the analysis.

2.4.
Eligibility criteria
Eligible patients were aged ≥40 years with at least one pulmonary nodule with a maximal diameter between 4 and 30 mm on baseline chest CT, a valid AAT result, with baseline clinical data needed for calculation of the Mayo model, and a benign or malignant outcome for the index nodule. When multiple nodules were present, a single index nodule was selected as the most suspicious lesion and used for all analyses. Patients were excluded if they had a prior history of any cancer, nodule diagnosis as extra-thoracic cancer; empirical treatment with radiation without definitive tissue diagnosis; nodules outside the 4–30 mm range; missing key clinical or imaging data; inadequate follow-up to confirm benignity for nodules without tissue diagnosis; or an invalid AAT result.

2.5.
Autoantibody test
The AAT (the Nodify CDT® test; Biodesix, Inc.) is an indirect ELISA that measures circulating autoantibodies against a panel of seven lung cancer-associated antigens (p53, NY-ESO-1, MAGE A4, GBU4-5, CAGE, HuD, and SOX2), performed at a central Clinical Laboratory Improvement Amendments certified (CLIA) and College of American Pathologists (CAP)-accredited laboratory. The technical and clinical validations of this assay have been reported previously [13–16,22]. Test results were categorized as positive if at least one autoantibody exceeded one of two pre-defined positivity thresholds (Moderate or High Level). The Moderate Level threshold defines the overall positive category, while the High Level threshold represents a more stringent cutoff and represents a subset of positive results within the Moderate Level category, associated with higher specificity and greater risk of malignancy.
A negative result, reported as “No Significant Level of Autoantibodies Detected” (NSLAD), indicates that none of the seven autoantibodies exceeded either positivity threshold. For PANOPTIC, FHCC, and ORACLE, AAT results were performed on banked plasma by personnel masked to clinical outcomes; in CLARIFY, AAT results were available to treating clinicians as part of routine clinical care.

2.6.
Baseline variables and risk models
Baseline variables included age, gender, smoking history, nodule diameter, lobe location, and spiculation, abstracted from case report forms and radiology reports. Pretest pCA was calculated using the validated Mayo Clinic model, which is widely utilized in clinical practice and calibrated for a similar cancer prevalence to this study [6,23]. To determine the AAT-adjusted post-test pCA, the AAT results were used to adjust the Mayo pretest pCA using previously reported positive likelihood ratios to generate post-test pCA based on the combination of clinical risk and AAT results [13,14].

2.7.
Statistical analysis
Analyses were performed by evaluating test performance across several subgroups. All statistical evaluations used a complete case dataset. Participants were excluded if they lacked variables required for Mayo pretest pCA computation, had invalid or missing AAT results, or did not meet the minimum follow-up criteria necessary to classify benign nodules.
The resulting complete case dataset was then used to evaluate test performance across several subgroups: the overall 4–30 mm cohort, two size strata (4– <20 mm and 20–30 mm), and pretest pCA strata (≤65% vs >65%), consistent with prior publications and established clinical guidelines [7,15]. Test performance was evaluated using three distinct thresholds: 1) the Moderate Level (defined as a positive result at either the Moderate Level or High Level thresholds), 2) the High Level threshold (defined as a positive result at the High Level threshold), and 3) a clinically relevant threshold defined by AAT results (either Moderate or High Level) that lead to a post-test >65% following application of the test positive likelihood ratios to the Mayo pretest pCA.
Performance metrics were calculated using the following definition: “test positive” as AAT positive results (Moderate or High Level) that resulted in a post-test pCA >65%; “test negative” included all NSLAD results and AAT positive results that resulted in a post-test pCA ≤65%. The post-test pCA >65% risk threshold was selected to align with the American College of Chest Physicians (ACCP) guidelines, which identify risk exceeding 65% as “High Risk,” warranting an invasive biopsy or surgical resection [7]. Sensitivity analyses were conducted to examine the impact of nodule diameter on test performance by defining nested cohorts with progressively higher minimum nodule sizes (e.g., ≥4, ≥6, ≥8, ≥10 mm, etc).
Continuous variables are reported as medians with interquartile ranges (IQR) and were compared using the Kruskal – Wallis or Wilcoxon Rank Sum tests as appropriate. Categorical variables are reported as counts and percentages and were compared using Fisher’s exact test or chi-square test. Sensitivity, specificity, and positive predictive value were calculated with Wilson 95% confidence intervals (CI).
Risk reclassification was assessed among patients with intermediate pretest probability (5–65%) by comparing the proportion of malignant versus benign nodules reclassified above the ACCP high-risk threshold ( >65% post-test probability). Risk ratios (RR) with 95% confidence intervals were calculated using the Wald method. Net Reclassification Improvement (NRI) was also computed as the difference in proportions of correctly versus incorrectly up-classified cases.
To evaluate the impact of cohort heterogeneity on pooled performance estimates, we first conducted a leave-one-cohort-out (LOCO) analysis. For each iteration, one cohort was omitted, and pooled sensitivity and specificity were recalculated for the remaining cohorts at both positivity thresholds (Moderate Level and High Level) using a random-effects model for proportions. Between-study heterogeneity was quantified using Higgins’ I2 statistic, which represents the proportion of variability attributable to true heterogeneity rather than chance (0–40% = low, 30–60% = moderate, >75% = substantial) [24]. We then fit multivariable logistic regression models using fixed-effects to assess whether age, nodule size, spiculation status, smoking history, or study cohort influenced sensitivity or specificity. Continuous variables (age and nodule diameter) were centered on their median values to improve interpretability. Odds ratios with 95% confidence intervals were reported. All analyses were conducted in R version 4.4.1 or later (R Foundation for Statistical Computing), with two-sided P <0.05 considered statistically significant. Because this was a retrospective pooled study, no a priori power calculation was performed; all available complete cases were included.

2.8.
STROBE compliance
This study adheres to the STROBE reporting guidelines; a completed checklist is included in the supplementary materials.

3.
Results
3.1.
Patient characteristics and case mix
A total of 1164 patients from 48 clinical practices with 4–30 mm pulmonary nodules and valid AAT results met the inclusion criteria and formed the combined analysis cohort (Figure 1). Overall cancer prevalence was 35% (411/1,164), with a breakdown of cancer type 90% non-small cell lung cancer (NSCLC) (368/411), 6% small cell lung cancer (SCLC) (25/411), and 4% (18/411) with lung carcinoid tumors. Median age was 66 years (IQR 59–72), 52% were female, and 81% currently or formerly smoked (Table 1). The median nodule diameter was 12 mm (IQR 9–17), with 81% of nodules <20 mm, 25% spiculated, and 54% located in an upper lobe. Median Mayo pretest pCA was 21% (IQR 11–42%), with 91% (1,064/1,164) of patients having a Mayo pCa ≤65% vs 9% (100/1,164) with a Mayo pCA >65%. The AAT was positive in 11% of patients.

Patient mix varied across cohorts (Supplement Table 1). Cancer prevalence ranged from 13% in the ORACLE study cohort to 54% in the PANOPTIC study cohort (p < 0.001). PANOPTIC patients had larger and higher-risk nodules, with a median diameter of 16 mm (IQR 12–22) versus 11–12 mm in the other cohorts (p < 0.001) and a median Mayo pCA of 39% (IQR 19–60) versus 13–22% elsewhere (p < 0.001). Spiculation was also more frequent in PANOPTIC (41%) and CLARIFY (30%) than in FHCC (7.8%) and ORACLE (9.4%) (p < 0.001). The distribution of AAT positivity by either threshold was broadly similar across cohorts, ranging from 10% in CLARIFY up to 13% in PANOPTIC (p value 0.7).
Of the 1164 patients, 945 (81%) had small (4– <20 mm) nodules and 219 (19%) had large (20–30 mm) nodules. Cancer prevalence was 29% (277/945) in the 4– <20 mm group and 61% (134/219) in the 20–30 mm group. Age, sex, and smoking history were similar between size categories. Larger nodules (20–30 mm) had substantially higher malignancy pCA and more concerning imaging features: the median diameters for the two groups were 11 mm (IQR 9–14) and 23 mm (IQR 21–26), for the small and large nodules, respectively. Spiculation varied from 23% to 33%, and upper-lobe location from 52% to 63%. Median Mayo pretest probability was 17% (IQR 10–31) in 4– <20 mm nodules and 57% (IQR 41–72) in 20–30 mm nodules. The distribution of AAT result categories did not differ significantly by size, with approximately 5–7% High Level, 5–7% Moderate Level, and 86–90% NSLAD results in each group. Figure 2 shows the distribution of nodule size and Mayo pCA, with malignant nodules enriched at larger diameters and higher risk estimates, but substantial overlap between benign and malignant nodules across the 4–30 mm range.

3.2.
AAT performance in the combined 4–30 mm cohort
The AAT showed similar performance across the four clinical study cohorts for the Moderate Level and High Level test thresholds (Supplement Table 2). To assess residual heterogeneity, we performed a leave one cohort out analysis (Supplemental Table 3). At the Moderate Level threshold, sensitivity ranged from 15% to 17% and specificity from 87% to 93%, with I2 = 0% for both metrics. At the High Level threshold, sensitivity ranged from 8% to 11% and specificity from 95% to 97%, with moderate heterogeneity (overall I2 47% for sensitivity and 45% for specificity). These results indicate that removing any single cohort did not meaningfully alter performance estimates.
We also used multivariable logistic regression to evaluate whether study cohort, clinical or imaging characteristics influenced AAT performance (Supplemental Table 4). At the Moderate Level threshold, none of the assessed variables, including cohort, age, nodule size, smoking history, spiculation, lobe location, or gender, were significantly associated with sensitivity or specificity (all p >0.20). At the High Level threshold, most associations were non-significant, although isolated effects were observed for age (p = 0.032 for sensitivity) and selected cohort comparisons (FHCC vs PANOPTIC, p = 0.049; ORACLE vs PANOPTIC, p = 0.037). These effects were small, with wide confidence intervals, and did not materially influence overall performance. Consequently, no consistent predictors of sensitivity or specificity were identified, and subsequent analyses used the combined cohort.
In the pooled 4–30 mm cohort (cancer prevalence 35%), overall performance was as follows (Table 2). At the Moderate Level threshold, specificity was 91% (95% CI 89–93%), sensitivity was 16% (95% CI 13–20%), and PPV was 50% (95% CI 41–58%) (Table 2). At the High Level threshold, specificity was 96% (95% CI 95–97%), sensitivity was 9% (95% CI 7–12%), and PPV was 57% (95% CI 45–68%). For pulmonary nodules with post-test pCA >65% (n = 71), sensitivity was 12% (95% CI 9–15%) and specificity 97% (95% CI 96–98%), with PPV 69% (95% CI 58–79%). Test performance at each threshold was stratified by Mayo pre-test pCA (≤65% vs >65%). Sensitivity and specificity for the Moderate or High Level thresholds were similar across groups (p values ≥0.207, Table 2), whereas PPV did differ between groups due to differences in cancer prevalence (p values ≤0.013).

3.3.
Size-stratified performance and effect of nodule size
Performance was assessed in small (4– <20 mm) and large (20–30 mm) nodules (Table 3). Sensitivity was similar across nodule sizes: 16% at the Moderate Level and 9% at the High Level in both groups. Specificity was consistently high (≥88%) regardless of nodule size. At the Moderate Level threshold, specificity was 91% in 4– <20 mm nodules and 88% in 20–30 mm nodules (p = 0.31); at the High Level threshold, specificity was 96% versus 95% (p = 0.55). In contrast, PPV increased markedly with nodule size. At the Moderate Level threshold, PPV was 44% in the small (4– <20 mm) nodules compared with 69% in the large (20–30 mm) nodules (p = 0.02). At the High Level threshold, PPV was 51% for small nodules and 75% for large nodules. Among nodules with post-test pCA >65%, PPV ranged from 64% for small nodules to 78% for large nodules.
To assess whether these findings depended on the categorical size stratification, we evaluated test performance across progressively higher minimum nodule diameter cutoffs from 4 to 30 mm (Figure 3 and Supplemental Table 5). Across all three thresholds, specificity remained consistently high with broadly overlapping 95% CIs across the entire size spectrum (≥87%), with the High Level and post-test pCA >65% thresholds maintaining specificity >91% regardless of the minimum size cutoff (Figure 3(B) and Supplemental Table S5). Sensitivity varied modestly across nodule-size thresholds, ranging from 9% to 25% at the High Level threshold, 16–25% at the Moderate Level threshold, and 12–25% when using a post-test pCA >65% criterion, with overlapping 95% CI across nodule sizes (Figure 3(A) and Supplemental Table S5). PPV increased steadily at all three thresholds, from 50% to 69% when including all nodules ≥4 mm to >99% when restricted to 30 mm nodules (Figure 3(C) and Supplemental Table 5), mirroring the rise in cancer prevalence from 35% in the full cohort to 89% among 30 mm nodules. Taken together, the minimal variation in sensitivity and consistently high specificity across all minimum size cutoffs indicate that the discriminatory performance of the AAT is stable across the 4–30 mm nodule diameter range and does not materially depend on nodule size.

3.4.
Risk reclassification
To assess risk reclassification of intermediate-risk patients (pCA 5–65%) to the high-risk category (pCA >65%), we examined how positive AAT results shifted patients above the 65% pCA threshold (Figure 4). In the overall cohort, 34 of 330 intermediate-risk cancers (10%) were reclassified to high risk (post-test pCA >65%) after a positive AAT, compared with 19 of 656 benign nodules (3%) incorrectly up-classified (supplemental Figure S1), yielding a net reclassification improvement (NRI) of 7% and a risk ratio of 3.6 (95% CI 2.1–6.1), indicating malignant nodules were over three times more likely than benign nodules to be reclassified. This pattern was similar by size grouping: in the small (4– <20 mm) nodules, 25 of 262 cancers (10%) and 15 of 587 benign nodules (3%) were reclassified (NRI = 7%), corresponding to a risk ratio of 3.7 (95% CI 2.0–7.0). In the large (20–30 mm) nodules, 9 of 68 cancers (13%) and 4 of 69 benign nodules (6%) were reclassified (NRI = 7%), with a risk ratio of 2.3 (95% CI 0.7–7.1). Among patients reclassified into the high-risk >65% pCA category, malignancy rates were higher, with PPVs of 64% (34/53) overall, 62% (25/40) in the small nodules, and 69% (9/13) in the large nodules (Table 3).

Similar to patients reclassified into the high-risk ( >65% pCA) category, those with an initial risk >65% pCA and a positive AAT result had enriched malignancy rates: 83% overall (15/18), 75% for small nodules (4– <20 mm; 3/4), and 86% for large nodules (20–30 mm; 12/14). Overall, these data show that patients with a post-test pCA >65% have elevated PPVs of 69% (95% CI 58–79%), 64% (95% CI 49–76%), and 78% (95% CI 59–89%) for the overall, small-nodule, and large-nodule groups, respectively.

4.
Discussion
In this pooled analysis of 1,164 patients with 4–30 mm pulmonary nodules, we evaluated the autoantibody test (AAT) across nodule size and pretest pCA spectrum. The AAT maintained high specificity (91–97%) at both the Moderate and High Level thresholds and for high-risk results (post-test pCA >65%) with consistent performance across pre-test risk groups (≤65% and >65%). This stability extended across the entire 4–30 mm size range: specificity remained above 91% regardless of size, and sensitivity varied only modestly (9–16%) as shown in Table 3 and Figure 3. These findings confirm that the AAT performs consistently across size and risk strata, supporting its use as a robust rule-in tool for pulmonary nodules. Although the absolute effect size is modest, with ~10% of intermediate-risk cancers reclassified above the >65% threshold, this focused impact can still be clinically meaningful by helping prevent diagnostic delays in patients who might otherwise remain in surveillance pathways.
Patients with post-test pCA >65% formed a group with enriched malignancy risk, with PPVs of about 60% in 4– <20 mm nodules and 80% in 20–30 mm nodules. Among intermediate-risk nodules (pCA 5–65%), malignant nodules were 3.6 times more likely than benign nodules to be reclassified above the >65% threshold following AAT, with a net reclassification improvement (NRI) of 7%. These metrics underscore the test’s ability to enrich the high-risk category with cancers while minimizing false-positive up-classification. A negative (NSLAD) result, however, does not exclude malignancy and should be considered non-informative; such nodules should continue to follow standard guideline-based management without deescalation of surveillance or delaying diagnostic evaluation.
This independent signal of elevated malignancy risk is especially important in the context of current clinical practice. Many patients with high-risk pulmonary nodules are still managed with CT surveillance alone; one study reported that 46% of patients were followed with serial imaging, including individuals later shown to have malignant disease [9]. Delays in lung cancer care have been associated with worse survival outcomes, and more recently, one-third of malignant nodules were not diagnosed until more than 90 days after initial identification [16,25]. Together, these findings underscore the need for additional tools such as the AAT biomarker to help physicians more rapidly and confidently identify patients who may benefit from earlier evaluation and treatment. These findings confirm and expand upon prior validation studies of the AAT. Massion et al. 2017 first established the AAT’s high-specificity rule-in profile in pulmonary nodules and showed its added value when combined with clinical risk models [15]. That study demonstrated that AAT positive patients had a 2.7-fold higher risk of lung cancer when used alone, and when combined with the Mayo Clinic model-based positivity thresholds, improved test specificity and positive predictive value. Similarly, subsequent validations focused on cohorts of 8–30 mm nodules reported consistent rule-in performance across the continuum of pretest pCA [16]. Our pooled analysis, which includes a substantially larger analysis population across the entire 4–30 mm range, demonstrates that this rule-in performance persists consistently across the full clinical spectrum of nodule size and confirms that the test is independent of pretest risk. Because AAT results provide information independent of clinical risk models, a positive result indicates increased malignancy risk even when the post-test probability is below the ACCP guideline threshold for high risk ( >65% pCA). This independence highlights its value as a complementary tool to imaging and clinical models.
For clinicians managing pulmonary nodules, the central question is how to integrate AAT results within existing nodule management pathways. Despite guideline frameworks and risk models such as the Mayo calculator, real-world care is complex, and many patients remain in an intermediate-risk zone where management is uncertain [3,7,9]. In these patients, particularly those with an intermediate pCA and worrisome CT features, a positive AAT result that raises post-test pCA above the 65% high-risk threshold may support an expedited invasive workup (e.g., biopsy or surgery) rather than continued surveillance. In case of nodules too small to biopsy, an AAT result may shift from expediting an invasive workup to justifying more aggressive surveillance per the Fleischner guidelines [8]. For instance, if a nodule is <6 mm and would typically require no additional follow-up in patients with low-risk factors, a positive AAT result may act as a risk modifier, leading clinicians to follow the higher risk surveillance pathway with a follow-up CT scan. Similarly, in 6–8 mm nodules requiring standard surveillance, a positive AAT result may justify placing the patient on an accelerated surveillance protocol (e.g., 3–6-month follow-up) to ensure any nodule growth is caught as early as possible.
The role of AAT should be considered in the context of other rule-in tools such as FDG-PET. PET is widely used for nodule evaluation, but its sensitivity declines in small lesions, and its specificity is significantly reduced in regions with endemic infectious or inflammatory lung disease [26–28]. Current guideline recommendations and technical limitations mean that PET is primarily used for nodules ≥10 mm in diameter, and its diagnostic value is limited for smaller nodules. In contrast, the AAT can be applied across the 4–30 mm range, including 4– <8 mm nodules where PET is often not recommended or is non-informative. Prior work from Long et al. indicates that AAT is more specific but less sensitive than PET and that the two provide complementary information, with combined use improving overall risk discrimination compared with either test alone [16].
This study has several strengths, including its large, multi-cohort design that includes three studies with prospective blood collection and centralized testing, as well as a postmarket real-world cohort. In total, the analysis evaluated more than 1,000 pulmonary nodules from 48 clinical practices across the United States, representing a diverse patient population reflective of real-world clinical practice. This combined cohort constitutes one of the largest and most geographically diverse clinical validations of a blood-based biomarker for pulmonary nodule risk assessment to date. We systematically examined test performance across nodule sizes from 4 to 30 mm and within clinically important pretest pCA categories (≤65% vs >65%), demonstrating the importance of identifying post-test pCA values >65% to align with guideline-based risk stratification. Important limitations should also be acknowledged: the retrospective design precludes direct assessment of how AAT-guided management would influence procedural use, diagnostic delays, or stage at diagnosis; and differences in referral patterns, imaging protocols, and follow up across cohorts introduce the potential for residual heterogeneity. Although pooling data from multiple studies can introduce variability related to patient mix and cancer prevalence, we addressed this by a conducting-leave-one-cohort-out analysis and a multivariable logistic regression analysis, which demonstrated stable AAT performance and suggested that heterogeneity did not materially affect the findings.
Future work should focus on prospective studies embedding the AAT within structured nodule programs to measure its impact on decision-making, resource use, and patient outcomes. In clinical practice, the AAT is often paired with the rule-out biomarker, the Integrated Classifier (IC), to provide complementary rule-in and rule-out information for overall nodule risk stratification [20,21]. The IC was developed as a “rule-out” test, optimized with high sensitivity and modest specificity, thus complementing the combined use with a rule-in biomarker optimized with high specificity. Utilization of these two tests together may further enhance pulmonary nodule risk stratification. The ongoing CLARIFY study, a large retrospective evaluation of real-world AAT use in clinical practice that plans to collect data on up to 4,000 patients, will elucidate how the AAT alone and in combination with the IC influences patterns of care and downstream testing. Additionally, future integration of the AAT with other advanced imaging approaches, such as radiomics, may yield multimodal models with better discrimination than any single component alone [29].

5.
Conclusion
In conclusion, the AAT provides consistent high specificity across more than 1000 pulmonary nodules from 48 diverse U.S. clinical practices, spanning in sizes from 4 to 30 mm. Positive results at the High level threshold or those that raise post-test probability >65% identify a subset of patients whose malignancy risk may justify expedited evaluation. These findings support the AAT as a high-specificity rule-in blood test that may enhance pulmonary nodule risk assessment.

Supplementary Material

Supplement Data V2.docx

AAT STROBE_checklist_cohort .docx