Robotic-Assisted Bronchoscopy for Histopathologic Subtyping of Primary Lung Adenocarcinoma

Abstract

Background:

The identification of high-grade patterns and mucinous features of invasive primary lung adenocarcinoma on biopsy specimens can have implications on therapeutic decisions, across all stages of disease. Shape sensing robotic-assisted bronchoscopy (ssRAB) is an emerging modality for the concomitant diagnosis and staging of lung cancer. We evaluated the performance of ssRAB for adenocarcinoma pattern identification, and particularly high-grade patterns, as well as the histopathologic concordance between biopsy and surgical resection specimens.

Methods:

Patients with lung adenocarcinoma diagnosed via ssRAB forceps or cryobiopsy specimens between October 2019 and December 2023 were included in the analysis. Biopsy specimens were evaluated for the identification of histopathologic patterns and mucinous features. A generalized linear mixed model quantified the association between pre- and intraoperative factors and successful pattern identification on biopsy. The concordance between high-grade patterns and mucinous features on ssRAB-acquired biopsy and poorly differentiated grade and mucinous features on subsequent surgical resection was determined.

Results:

A total of 242 ssRAB-acquired specimens were included in the final analysis. The biopsy specimen was sufficient to identify adenocarcinoma histopathologic patterns in 71%. In a multivariable analysis, sampling by cryobiopsy was positively associated with pattern identification (OR 3.54, CI: 1.02-12.30; P=0.04), as compared with forceps biopsy. A corresponding surgical resection specimen was available in 66 cases. The sensitivity, specificity, positive, and negative predictive values of biopsy were 63, 72, 61, and 74%, respectively for the presurgical detection of poorly differentiated adenocarcinoma, and 87, 100, 100, and 96%, respectively for the presurgical detection of mucinous features.

Conclusion:

This study is the first to report the performance of ssRAB-acquired biopsy for identification of adenocarcinoma patterns and its concordance with surgical resection. Our findings align with those previously reported for percutaneous lung biopsy. ssRAB emerges as a viable tool for the identification of adenocarcinoma patterns. Future studies are needed to confirm these findings in larger patient cohorts.

Despite advances in the early detection and treatment, lung cancer remains the deadliest cancer worldwide (1). Among primary lung neoplasms, adenocarcinoma is the most prevalent histological type and harbors much morphologic heterogeneity. Invasive nonmucinous adenocarcinoma is divided into three main histopathologic grades: low-grade, which includes lepidic (LEP), intermediate grade, which includes acinar (ACI) and papillary (PAP), and high-grade, which includes micropapillary (MIP), and solid (SOL) growth patterns (2). Cribriform (CRIB) and fused glandular (GLAN) were subsequently recognized as additional high-grade patterns (3). A single tumor can harbor one or more patterns in variable proportions. The International Association for the Study of Lung Cancer (IASLC) has further classified the different adenocarcinoma patterns into grades of differentiation, based on the degree of pattern involvement in surgical resection specimens (4). In this system, tumors with <20% high-grade patterns are classified as well- or moderately differentiated while tumors with ≥20% high-grade patterns are classified as poorly differentiated. The IASCL system was shown to correlate well with prognosis (5–10) and was later adopted by the World Health Organization (WHO) (2). Invasive mucinous lung adenocarcinoma (IMA) is considered a separate entity and tends to carry a worse prognosis, compared with its nonmucinous counterpart (11–13).

Prior data suggest that patients with poorly differentiated adenocarcinoma may have lower rates of recurrence after a lobectomy, compared with a sublobar resection (12, 14–16) as well as higher rates of recurrence after curative-intent radiation therapy (17). As such, in early-stage lung adenocarcinoma, pre-operative identification of high-grade and mucinous histopathologic features may assist in guiding curative-intent interventions. The presence of MIP or SOL patterns was also shown to have adverse prognostic implications in patients with more advanced-stage lung adenocarcinoma, who are being considered for neoadjuvant therapy or nonsurgical management (6, 7, 18–20).

Shape-sensing robotic-assisted bronchoscopy (ssRAB) is emerging as a high-yield and safe tool for the sampling of pulmonary parenchymal lesions (21–24). Prior studies have explored the accuracy of percutaneous biopsy specimens in the prediction of adenocarcinoma histopathologic patterns on surgical resection specimens (15, 18, 25–28); however, no such data is available for biopsies obtained via ssRAB.

Prior studies indicated that ssRAB-acquired small tissue biopsies provide adequate material for diagnostic and molecular profiling purposes (22, 23). We therefore hypothesized that primary lung adenocarcinoma biopsy specimens acquired via ssRAB would provide sufficient tissue for histopathologic pattern identification. We further hypothesized that in lung adenocarcinoma, ssRAB-acquired small biopsy specimen histopathology will sufficiently reflect the tissue composition of the surgically resected lesion. The purpose of this study was to evaluate the accuracy of ssRAB in defining histopathologic subtypes of primary lung adenocarcinoma and identifying high-grade patterns. We further described the concordance rates between the pre-operative biopsy and the surgical resection specimen in terms of high-grade patterns and mucinous features, thus establishing the accuracy of ssRAB for the pre-operative identification of poorly differentiated and mucinous primary lung adenocarcinomas.

Methods

Memorial Sloan Kettering Cancer Center (MSK) maintains a prospectively curated research electronic data capture (REDCap^™) database (29) of all ssRAB procedures performed across the institutional Interventional Pulmonology and Thoracic Surgery services. All cases of ssRAB performed at MSK between October 2019 and December 2023 were retrospectively reviewed. Cases in which a primary lung adenocarcinoma was diagnosed via ssRAB-guided transbronchial forceps biopsy (TBFB) and/or transbronchial cryobiopsy (TBCB) were included in the final analysis (Figure 1). Cases in which only a transbronchial needle aspiration (TBNA) specimen was obtained were excluded as this tool is not sufficiently reliable for histopathologic pattern identification (30, 31). Abstracted variables included patient-level demographic, clinical, and procedure data, lesion-level radiographic, sampling, and histopathologic data, and surgical resection data when available. This Health Insurance Portability and Accountability Act-compliant study was approved by the MSK institutional review board (protocol #20-166), and informed consent was waived.

Figure 1.

Study Case Selection Flow Chart

Robotic-Assisted Bronchoscopy

All procedures were performed using the Ion Endoluminal System^™ (Intuitive Surgical Inc., Sunnyvale, CA) by one of 10 users and under general anesthesia, as previously described (21). The use of intraoperative imaging, including radial probe endobronchial ultrasound (EBUS; UM-S20-17S or UM-S20-20R, Olympus Corp., Tokyo, Japan), pulsed fluoroscopy, and/or mobile cone-beam computed tomography (CBCT) imaging (Cios-Spin^™ Mobile 3D C-Arm, Siemens Healthineers Inc., Erlangen, Germany) as well as the choice of sampling tools were left to the discretion of the operator. Linear EBUS-guided mediastinal lymph node staging was performed when appropriate. Rapid on-site evaluation (ROSE) was available during all procedures.

Histopathology Specimen Evaluation

Histopathology interpretation reports of ssRAB-acquired TBNA, TBFB, and TBCB specimens of patients diagnosed with primary lung adenocarcinoma were individually reviewed. Pattern subtyping was considered successful if the biopsy specimen was sufficient to identify histopathologic patterns or mucinous features as defined by the 2021 WHO Classification of Thoracic Tumours (2). The presence of any amount of MIP, SOL, CRIB, and/or GLAN patterns defined the biopsy histology as high grade. Biopsies showing mucinous features were classified as compatible with IMA vs. incompatible with IMA based on the 2021 WHO criteria (2). Cases of “NSCLC, favor adenocarcinoma” were revisited by a board-certified thoracic pathologist (MKB) to identify tumors with SOL architecture and to distinguish them from those without discernable growth patterns.

When surgical resection was pursued, histopathologic reports were abstracted for characteristics of the resected tumor, including neoplasm type, quantification of adenocarcinoma patterns, and the presence and nature of mucinous features. As per the IASLC and WHO statements, the dominant pattern was defined as the pattern with highest representation in the surgical resection, while primary nonmucinous lung adenocarcinoma was considered poorly differentiated if involved by ≥20% of high-grade patterns, including any combination of MIP, SOL, CRIB, and GLAN (4, 32). The predominant histologic growth pattern in the surgically resected tumors was defined as the pattern constituting the largest proportion of the tumor volume. IMA and mixed nonmucinous-IMA adenocarcinomas were defined as per the 2021 WHO criteria (2). All biopsy and resection specimens displaying mucinous features as well as discordant cases were individually revisited by a board-certified thoracic pathologist (MKB) to confirm or revise the findings.

Concordance analyses were performed to explore the accuracy of biopsy specimens in identifying poorly differentiated adenocarcinoma, IMA, and dominant growth pattern in surgical resection specimens. All discordant cases were revisited by a board-certified thoracic pathologist (MKB) to confirm or revise the findings as well as identify the likely reason for discrepancy.

Statistical Analysis

Categorical variables are presented as counts and percentages and continuous variables are presented as medians and interquartile ranges (IQR). Comparisons between successful and unsuccessful biopsy pattern subtyping groups were conducted using the Wilcoxon rank-sum test for continuous variables and the Pearson’s Chi-square or the Fisher’s Exact tests for categorial variables. For the biopsy pattern identification analysis, a generalized linear mixed model with logit link was utilized to quantify associations between patient, lesion, and procedure variables and successful adenocarcinoma pattern subtyping. This model included patient-level random effects to account for potential correlation among separate lesions from the same patient. The univariable models included age, sex, body mass index (BMI), smoking status, prior cancer history, American Society of Anesthesiologists score, lesion size, radiographic consistency, lung zone location, lung centrality, number of traversed airway generations, radial EBUS view, use of intraoperative imaging, TBFB sampling, and TBCB sampling. A multivariable model was then constructed using factors that had a P-value of <0.05 in the univariable analysis and other factors determined a priori to be clinically relevant. For the biopsy-surgical resection concordance analysis, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated using standard definitions (33). All statistical analyses were two-tailed with P-value <0.05 considered statistically significant and were performed using R version 4.3.1 (R Core Development Team, Vienna, Austria).

Results

Of 1,652 ssRAB procedures performed during the study period, 488 specimens provided a diagnosis of a lung adenocarcinoma. Of these, in 242 samplings performed on 210 individual patients, a biopsy specimen was obtained by TBFB and/or TBCB (Table 1, Figure 1). The median patient age was 71 years (IQR: 64-78) and 65% were female. A prior history of cancer was documented in 64% (Supplementary Table 1). More than one lesion was sampled in 12% of the cases and the median ssRAB procedure time was 49 minutes (IQR: 35-69). Linear EBUS-guided mediastinal lymph node staging was concomitantly performed in 130 cases (62%). Five instances of procedure-related adverse events were recorded (Supplementary Table 2). There were 3 instances of pneumothorax (1.4%), one of which required tube thoracostomy (0.4%), and one instance of grade 3 bleeding (0.4%). Lung resection was subsequently performed in 99 (41%) cases (Figure 1).

Table 1.

Patient-Level Characteristics (n=210)

Variable			Median (IQR) or n (%)
Age			71 (64–78)
Female Sex			136 (65)
BMI (kg/m2)			25.4 (22.5–29.5)
Smoking Status
	Never			60 (29)
	Ever			150 (71)
		Current			17 (11)
		Former			133 (89)
ASA Score
	1			3 (2)
	2			49 (23)
	3			148 (70)
	4			10 (5)
Number of Sampled Pulmonary Lesions
	1			185 (88)
	2			19 (9)
	3			5 (2)
	4			1 (1)
Procedure Time (min)			49 (35–69)

ASA, American Society of Anesthesiologists; BMI, body mass index; IQR, interquartile range.

Pattern Identification by Biopsy

Table 2 shows lesion-level data for the 242 adenocarcinomas sampled via ssRAB-guided TBFB and/or TBCB. The median lesion diameter was 23.0 mm (IQR: 16.0-32.8), 66% of lesions were radiographically solid, and 67% were beyond the 6^th generation airway. Tissue acquisition and staging data are shown in Supplementary Tables 3 and 4, respectively. TBNA, TBFB, and TBCB were utilized in 97, 95, and 21% of samplings, respectively. The individual tool diagnostic yield rates for the diagnosis of adenocarcinoma were 86, 86, and 96% for TBNA, TBFB, and TBCB, respectively. The overall and individual sampling tool performance success rates for pertinent immunohistochemistry and molecular assays along with adequacy thresholds are shown in Supplementary Table 5. Immunohistochemistry stains for anaplastic lymphoma kinas (ALK) and programmed death ligand 1 (PD-L1) were performed successfully in 98 and 97% of lesions, respectively, and 89% of the samplings were deemed adequate for predictive marker and driver mutation assays. The overall assay success rates for epidermal growth factor receptor (EGFR) polymerase chain reaction (PCR), Ki-ras2 Kirsten rat sarcoma viral oncogene homolog (KRAS) PCR, and next-generation sequencing were 92, 92, and 93%, respectively. A histopathologic adenocarcinoma pattern was successfully identified in 172 (71%) biopsy specimens. Of diagnostic TBFB samplings, pattern was identified in 76%, while of diagnostic TBCB samplings, pattern was identified in 92% (Supplementary Table 3). The distribution of identified patterns is shown in Supplementary Table 6. A high-grade pattern was identified in 44% of successful samplings, while mucinous features were identified in 19%. Review of representative biopsies in which no pattern was identified revealed that the primary reasons for pattern nonidentification were crush artifact (39%), scant tumor cells (21%), or nondiagnostic sampling (40%). In the univariable models, lower BMI, prior history of cancer, peripheral location, and TBCB sampling were associated with successful pattern identification in biopsy specimen. In the multivariable model, TBCB sampling remained independently associated with successful pattern recognition (Table 3 and Supplementary Table 7).

Table 2.

Lesion-Level Characteristics and Small Biopsy Specimen Pattern Identification Analysis^*

Variable			Full Cohort (n=242)	Cohort Subgroups
				Pattern Identified (n=172)	Pattern Not Identified (n=70)	P-Value
Largest Dimension (mm)			23.0 (16.0–32.8)	23.2 (1.60–32.9)	22.1 (15.6–32.8)	>0.9
Radiographic Consistency						>0.9
	Solid		159 (66)	113 (71)	46 (29)
	Non-Solid**		83 (34)	59 (66)	24 (34)
Lung Zone***						0.9
	Upper		164 (68)	117 (71)	47 (29)
	Lower		78 (32)	55 (70)	23 (30)
Centrality						0.007
	Inner 2/3		126 (52)	80 (47)	46 (66)
	Outer 1/3		116 (48)	92 (53)	24 (34)
Number of Traversed Airway Generations						0.2
	<7		80 (33)	53 (63)	27 (37)
	≥7		162 (67)	119 (73)	43 (27)
Radial Probe EBUS
	Not Used		33 (14)	22 (67)	11 (33)
	Used		209 (86)	150 (72)	59 (28)	0.6
		Concentric	96 (46)	66 (69)	30 (31)
		Eccentric	80 (38)	60 (75)	20 (25)
		No View	33 (16)	24 (73)	9 (27)
Intraoperative Imaging						0.8
	Pulsed Fluoroscopy Only		62 (26)	45 (73)	17 (27)
	Pulsed Fluoroscopy and CBCT		180 (74)	127 (71)	53 (29)
Forceps Biopsy Performed			229 (95)	161 (70)	68 (30)	0.4
Cryobiopsy Performed			51 (21)	45 (88)	6 (12)	0.002

CBCT, cone-beam computed tomography; EBUS, endobronchial ultrasound.

^*Summaries are presented as median (interquartile range) or n (%).

^**A composite variable including semi-solid, pure groundglass, and cavitary lesions.

^***Upper lung zone includes the upper and middle lobes while the lower lung zone includes the lower lobes.

Table 3.

Patient, Lesion, and Procedure Factors Associated with Successful Adenocarcinoma Pattern Identification

Variable		Univariable Logistic Regression Model		Multivariable Logistic Regression Model
		OR (95% CI)	P-Value	OR (95% CI)	P-Value
BMI (1.0-kg/m2 increment)		0.92 (0.86–0.99)	0.033	0.94 (0.88–1.01)	0.086
Prior History of Cancer			0.033		0.2
	No	1.0		1.0
	Yes	2.17 (1.06–4.42)		1.68 (0.82–3.45)
Lung Centrality			0.013		0.062
	Inner 2/3	1.0		1.0
	Outer 1/3	2.35 (1.20–4.60)		2.09 (0.96–4.55)
Number of Traversed Airway Generations			0.3		>0.9
	<7	1.0		1.0
	≥7	1.43 (0.72–2.83)		1.04 (0.48–2.28)
TBFB Sampling			0.3		>0.9
	No	1.0		1.0
	Yes	0.40 (0.07–2.24)		1.04 (0.13–12.3)
TBCB Sampling			0.006		0.046
	No	1.0		1.0
	Yes	4.54 (1.56–13.30)		3.54 (1.02–12.30)

BMI, body mass index; CI, confidence interval; OR, odds ratio; TBCB, transbronchial cryobiopsy; TBFB, transbronchial cryobiopsy.

Concordance Analysis with Lung Resection Specimens

A total of 99 lesions (41%) were surgically resected. Of these, a pattern was identified on prior biopsy in 69. Three cases of surgically confirmed pleomorphic carcinoma were subsequently excluded. The remaining 66 resected adenocarcinomas comprised the accuracy analysis cohort (Figure 1 and Supplementary Table 8). Of these, 51 were nonmucinous, 6 were nonmucinous with mucinous features, 5 were pure IMA, and 4 were mixed nonmucinous and IMA. Nine (13%) patients received pre-operative neoadjuvant therapy.

The accuracy, sensitivity, specificity, PPV, and NPV rates of the biopsy specimen to identify poorly differentiated adenocarcinoma [≥20% high-grade pattern in the resection specimen (4)], were 68, 63, 72, 61, and 74%, respectively (Table 4). Examples of histopathologic concordance between ssRAB-acquired biopsy and surgical resection are illustrated in Figure 2. Discordant false negative (n=10) and false positive (n=11) cases are described in detail in Supplementary Table 9. Comparison of concordant and discordant cases across different lesion- and procedure-related factors did not reveal statistically significant differences (data not shown). After exclusion of neoadjuvant cases, the accuracy, sensitivity, specificity, PPV, and NPV rates were 72, 61, 79, 67, and 75%, respectively (Supplementary Table 10). The accuracy, sensitivity, specificity, PPV, and NPV rates of the biopsy specimen to identify IMA or mixed nonmucinous-IMA adenocarcinoma in the surgical resection specimen were 95, 89, 96, 80, and 98%, respectively (Table 4). The extent of surgical resection – lobar vs. sublobar – did not differ in a statistically significant manner between cases in which the pre-operative biopsy showed a high-grade pattern or mucinous features and cases in which no high-grade patterns or mucinous features were identified pre-operatively (data not shown).

Table 4.

Diagnostic Accuracy Analysis for Identification High-Grade Histopathologic Patterns of Adenocarcinoma Across Small Biopsy and Surgical Resection Specimens (n=66):

		Surgical Resection Specimen
		Poorly Differentiated	Non-Poorly Differentiated
Biopsy Specimen	High Grade Pattern Identified	17 (63%)	11 (47%)	28 (43%)
	High Grade Pattern Not Identified	10 (27%)	28 (73%)	38 (57%)
		27 (41%)	39 (59%)	66 (100%)
Accuracy: 68% Sensitivity: 63% Specificity: 72% Positive predictive value: 61% Negative predictive value: 74%
		Surgical Resection Specimen
		Pure IMA or mixed nonmucinous-IMA	Nonmucinous Adenocarcinoma
Biopsy Specimen	Compatible with IMA	8 (89%)	2 (4%)	10 (15%)
	Incompatible with IMA	1 (11%)	55 (96%)	56 (85%)
		9 (14%)	57 (86%)	66 (100%)
Accuracy: 95% Sensitivity: 89% Specificity: 96% Positive predictive value: 80% Negative predictive value: 98%

CI, confidence interval; IMA, invasive mucinous adenocarcinoma.

Figure 2.

Histopathologic Illustration of Biopsy-Surgical Resection Concordance

(A) Robotic-assisted bronchoscopy-acquired cryobiopsy specimen revealing a lung adenocarcinoma with micropapillary growth pattern (arrowheads). (B, C) Low and intermediate magnification images of the corresponding resection specimen revealing micropapillary tufts within the tumor bulk. (D) Robotic-assisted bronchoscopy-acquired cryobiopsy specimen revealing mucinous features compatible with an invasive mucinous adenocarcinoma. (E, F) Low and intermediate magnification images of the corresponding resection specimen confirming invasive mucinous adenocarcinoma.

The overall concordance rate between the biopsy identified pattern and the surgical resection predominant pattern was 83% with 93-100% concordance for ACI, PAP, and MIP patterns and 55% concordance for LEP and SOL patterns (Supplementary Table 11). Concordance rates for three cases of pleomorphic carcinoma are shown in Supplementary Table 12.

Discussion

Identification of high-grade histopathologic subtypes of primary lung adenocarcinoma has prognostic implication for both early- and late-stage disease (11, 14, 18–20, 25, 26). This is the first study to evaluate the yield and accuracy of biopsy specimens obtained via ssRAB in the subtyping of primary lung adenocarcinoma, including mucinous variants, and correlate these with surgical resection specimens across all stages of the disease. Overall, our results indicate that ssRAB-guided biopsy can adequately classify adenocarcinoma patterns in 71% of cases with an overall accuracy of 68% for poorly differentiated histology on resection.

Primary lung adenocarcinoma harbors morphologic heterogeneity comprised by multiple histopathologic growth patterns (3, 4, 32). Classification of these patterns into grades was shown to correlate with long term prognosis (7, 9, 17, 25, 34). Many cases of lung adenocarcinoma are initially diagnosed and staged by biopsy specimens, obtained via the percutaneous or transbronchial routes (35, 36). Identification of the various histopathologic subtypes of adenocarcinoma prior to definitive therapy carries a prognostic value that allows more informed decision making and alignment of patient expectations, regardless of clinical stage (16, 34, 35).

Prior studies have described the utility of image-guided transthoracic needle biopsy (TTNB) in the subtyping of adenocarcinoma (15, 17, 25, 37); however, when compared with bronchoscopy, this sampling modality is associated with a higher rate of complications, primarily pneumothorax and bleeding (38, 39). Furthermore, TTNB does not include concurrent mediastinal lymph node staging in early-stage disease. Although bronchoscopy is considered a safer diagnostic modality, for many years, legacy bronchoscopic platforms were plagued by suboptimal diagnostic yield rates (40, 41). ssRAB is emerging as a bronchoscopic platform that combines a diagnostic yield, comparable to that of TTNB with a superb safety profile (42, 43). Furthermore, mediastinal lymph node staging can be combined with ssRAB to provide concomitant diagnosis and staging in a single procedure. The acquisition of biopsy specimens for adenocarcinoma pattern identification via ssRAB can be performed via TBFB or TBCB (23, 44, 45). TBNA provides cytology material which is not considered sufficiently reliable for pattern identification.

In this study, we have demonstrated that ssRAB-guided TBFB or TBCB allowed the identification of adenocarcinoma patterns in 71% of all samplings and 79% of diagnostic samplings. In a study by Zhang et al, the rate of adenocarcinoma pattern identification on TTNB specimens by multiple pathologists was in the range of 31-62% with discrepancies attributed to the observer’s level of experience (37). Indeed, interobserver variability was previously acknowledged in adenocarcinoma pattern identification and strategies for standardization have been proposed (30). In a TTNB study from a cancer center (MSK), Kim et al were able to identify a histopathologic adenocarcinoma pattern in 74% of TTNB specimens (15). This rate is aligned with that reported here for ssRAB. It is reasonable to assume that pathologists from dedicated cancer centers would be more experienced in adenocarcinoma pattern identification. This may explain the higher pattern identification rates seen in our study and in that of Kim et al compared with the study by Zhang et al. Current guidelines do not favor a certain sampling modality in the evaluation of parenchymal lung lesions and frequently the decision regarding bronchoscopy vs. image-guided biopsy remains dependent on provider preference, local expertise, and availability. Experience with ssRAB is expanding worldwide. As ssRAB provides diagnostic yield rates which are comparable to TTNB (42, 43), allows for concurrent mediastinal staging, and is overall safer, we anticipate that a shift towards bronchoscopic sampling, when available, will occur in the future.

We demonstrate that the choice of biopsy tool plays a role in pattern identification. The pattern diagnostic yield for TBFB was 76% as compared with 92% for TBCB. Furthermore, in a multivariable analysis, TBCB emerged as the only factor that was independently associated with pattern identification on the biopsy specimen. The difference in yield between the two biopsy tools can be attributed to a) larger specimen size obtained by TBCB (44, 46) and b) improved preservation of tissue architecture on TBCB by avoiding crush and shear artifacts (44, 47). These attributes ultimately explain the improved tissue preservation of TBCB and its ability to provide sufficient material for ancillary molecular and genomic studies (23, 44–46, 48, 49). Nevertheless, due to difficulty in predicting the extent of tissue freezing during TBCB sample acquisition, this method imposes a higher than traditional risk for severe bleeding and/or pneumothorax when performed by inexperienced providers. TBCB should therefore be reserved to centers with experience in the performance of the procedure as well as the management of its complications (50, 51). Within this context, it is noteworthy that despite our study’s exceedingly low overall complication and pneumothorax rates of 2.3% (n=5) and 1.4% (n=3), respectively, two instances of pneumothorax and one of grade 3 bleeding occurred in patients who underwent TBCB. Of note, the proportion of cases in which TBCB was performed in our study was significantly lower when compared with TBFB; 21 vs. 95%. We therefore advise caution when interpreting these data. Studies are underway to further characterize the role of ssRAB-guided TBCB in the evaluation of pulmonary lesions in cancer patient populations.

We subsequently evaluated the concordance between pattern identification by ssRAB-acquired biopsy and surgical resection specimens. Expectedly, most patients included in this analysis had earlier stage disease. The overall concordance rate between the biopsy pattern and the surgical resection predominant pattern was 83%. In comparison, the predominant pattern concordance rate for TTNB was reported in the range of 59-77% (18, 26–28, 52). The concordance rate reported here was >90% for ACI, PAP, and MIP patterns and 55% for LEP and SOL patterns. While these rates may be biased by the relatively low number of observations, the lower LEP and SOL rates can also be attributed to sampling limitations. LEP pattern typically involves a lesion’s groundglass periphery while sampling is usually directed at the lesion center or solid component, if present. Additionally, SOL pattern identification on small biopsy requires highly preserved, non-fragmented tissue (30), which may be difficult to obtain with TBFB. TBCB holds a promise of providing larger and more preserved tissue biopsies compared with TBFB (23, 45). Although this carries a potential to improve tissue diagnosis, the proportion of TBCB in our study cohort was relatively low and therefore does not allow direct comparison with TBFB. Studies dedicated for TBCB may be able to clarify its role in subtyping of lung adenocarcinoma (53).

The accuracy analysis of ssRAB-guided biopsy in predicting poorly differentiated adenocarcinoma on resection, revealed accuracy, sensitivity, and PPV rates of 68, 63, and 60%, respectively; however, review of the false positive cases revealed that in 91% (10/11) the high-grade pattern was indeed identified on the biopsy but ultimately comprised <20% of the resection specimen histopathologic landscape, classifying it as moderately differentiated. Four of these were neoadjuvant cases. Prior studies have shown that neoadjuvant therapy has a potential to induce favorable morphologic and molecular changes in the tumor (54, 55). This may explain the discrepancy between the pre-neoadjuvant biopsy and post-neoadjuvant resection specimen. Elimination of neoadjuvant cases from this analysis increased the accuracy to 72%. Moreover, most discordant cases in our cohort can be attributed to sampling and treatment effect with only a minor contribution from interpretation errors. Collectively, these data indicate that ssRAB-guided biopsy may be sufficiently sensitive to identify high grade patterns even when these are not dominating the overall histopathologic landscape of the surgical resection. Furthermore, there are data to suggest that even minor involvement by an aggressive pattern is associated with adverse outcomes and higher recurrence rates, which may justify a more extensive surgical resection and rigorous post-operative surveillance (14, 17, 56–58). Reanalysis of our data based on this classification scheme generated accuracy, sensitivity, and PPV rates of 67, 56, and 96%, respectively. Our results are also aligned with prior TTNB studies, which report the sensitivity and PPV of core needle biopsy for high grade patterns to be in the ranges of 16-48% and 64-96%, respectively (15, 26, 27, 52). While some TTNB studies have correlated biopsy sample size and radiographic consistency with degree of concordance with surgical resection (26, 52), our study did not reveal such an association.

IMA are biologically distinct from their nonmucinous counterparts regardless of stage (11, 59–62). Identifying these entities on biopsy can inform therapeutic decisions and patient expectations. We found pure IMA or mixed nonmucinous-IMA adenocarcinoma in 11% of surgical resections, which is aligned with the prevalence reported in the literature (11, 12, 61, 62). Our findings indicate that ssRAB-guided biopsies harbor accuracy, sensitivity, and PPV rates of 98, 89, and 80%, respectively, for IMA and mixed nonmucinous-IMA adenocarcinoma. In a retrospective analysis of 105 cases of surgically resected IMA, Lee et al reported the diagnostic accuracy of TTNB for IMA was 59% (31). Data derived from percutaneous sampling studies also indicate that tissue biopsy may be superior to fine needle aspiration for the identification of mucinous features (31, 63). Our study corroborates this finding; of 32 specimens with mucinous features, only 6 were identified by TBNA (Supplementary Table 5).

The translation of adenocarcinoma grades of differentiation into management decisions is a matter of ongoing research. While the role of neoadjuvant chemo-immunotherapy for resectable, stage 2, lymph node-positive, driver mutation-negative lung adenocarcinoma is well-established, the best management strategy for stage 2, lymph node-negative tumors remains a matter of debate (64, 65). Within this context, high-risk histology on pre-operative biopsy may inform the choice of peri-operative therapy over upfront resection (66, 67). As for pre-operative biopsy to guide the extent of surgical resection, there is data to support the extension of the resection margins when certain high-grade patterns are identified (14, 16, 68); however, this has yet to materialize into established guidelines and the decision remains at the discretion of the surgeon. In our cohort, the presence of high-grade patterns or mucinous features did not seem to translate into more extensive resection; however, the relative low number of observations may limit the interpretation of these data.

Several limitations of our study warrant attention. This is a retrospective analysis of a prospectively curated database. Our study cohort represents a patient population referred and managed in a single, high-volume, quaternary cancer center. Together these two limitations may affect prevalence rates and introduce a selection bias resulting in skewing of the disease severity indices towards more advanced disease. Generalization of these results to lower volume or less experienced centers should be therefore performed with caution. It is noteworthy that our ssRAB program includes providers of interventional pulmonology and thoracic surgery backgrounds along a range of procedural experience, which, on one hand, introduces a higher degree of diversity and generalizability to our results, but on the other hand results in heterogeneity in procedure performance and sampling tool selection. Interobserver variability was previously described in relation to adenocarcinoma pattern identification (30). Although not all biopsy specimens were revisited for the purpose of this study, in our institution, lung biopsy specimens are interpreted by dedicated thoracic pathologists, who are considered experts in the field. Indeed, the rate of pattern identification in our study was higher than that reported in the literature (27, 37).Moreover, biopsy-surgical resection discordant cases in the study were reviewed by an expert thoracic pathologist to provide a finalized interpretation. Of the full study cohort, 47% of patients underwent surgical resection and 31% were eligible for the concordance analysis. While this rate calls for caution in the interpretation of the concordance analysis, it can be explained by the inclusion of both early- and late-stage cases of adenocarcinoma in the study cohort, therefore inherently attenuating the number of patients who would be eligible for surgery. Additionally, since the concordance analysis cohort included patients at earlier stages of disease, the concordance rates presented here cannot be directly applied to patients with more advanced stages of disease. Furthermore, we excluded cases in which the biopsy did not disclose a pattern from the denominator for this analysis; however, this methodology appears to be accepted and reported in prior studies of TTNB (15, 26, 52). Our data reflect ssRAB procedures performed in our institution since the inception of the ssRAB Program in 2019; as expertise in the field proliferated over the years, the performance of ssRAB procedures has evolved in terms of accuracy, reliability, and technique. These dynamics introduce heterogeneity into the data, which is reflected in the number of excluded cases due to TBNA sampling-only and in the relatively low number of TBCB in our cohort. Moreover, while this study highlights TBCB as a high-yield biopsy tool in adenocarcinoma subtyping, the risk associated with the performance of TBCB by inexperienced users cannot be over-emphasized (50, 51) and appropriate training and supervision are fundamental to any TBCB program.

Conclusion

This study is the first to quantify the performance of ssRAB-guided biopsy in the accurate identification of primary lung adenocarcinoma growth patterns across all stages of disease and its concordance with surgical resection. Our study identifies ssRAB as a viable tool for adenocarcinoma subtyping with diagnostic yield and performance rates that are comparable to those of TTNB, while maintaining a superior safety profile. Future studies are needed to generalize these results across more diverse patient populations and expand patient cohorts to guide the diagnostic evaluation and management of patients with a variety of lesions, morphologic patterns, and stages of disease.

Highlights.

• Lung adenocarcinoma growth patterns have prognostic implications across all stages of disease and can affect management decisions.

• Robotic-assisted bronchoscopy-guided lung biopsy allowed for the subtyping of lung adenocarcinoma in 71% of cases.

• Transbronchial cryobiopsy was associated with a higher likelihood of adenocarcinoma pattern identification.

• The concordance between bronchoscopic biopsy and surgical resection with regards to poorly differentiated pattern identification was 68%.

• Robotic-assisted bronchoscopy is viable tool for lung adenocarcinoma subtyping with a yield that is aligned with that of percutaneous needle biopsy.

List of Abbreviations:

ACI

acinar (LEP)

ASA

American Society of Anesthesiologists

BMI

body mass index

CRIB

cribriform

CBCT

cone-beam computed tomography

EBUS

endobronchial ultrasound

GLAN

fused glandular

IASLC

The International Association for the Study of Lung Cancer

LEP

lepidic

MIP

micropapillary

MSK

Memorial Sloan Kettering Cancer Center

NPV

negative predictive value

NSCLC

non-small cell lung cancer

PAP

papillary

PPV

positive predictive value

ROSE

rapid on-site evaluation

SOL

solid

ssRAB

shape-sensing robotic-assisted bronchoscopy

TBCB

transbronchial cryobiopsy

TBFB

transbronchial forceps biopsy

TBNA

transbronchial needle aspiration

WHO

World Health Organization