Malignancy estimation of Lung-RADS criteria for subsolid nodules on CT: accuracy of low and high risk spectrum when using NLST nodules

In 2014, the ACR published Lung-RADS for screen-detected pulmonary nodules. Similarly to other management recommendations, it differentiates nodule types and uses nodule diameter and growth as major input parameters [1, 8, 14, 18]. Nodule management is adapted to estimated malignancy probability with category 2 corresponding to the finding of a nodule with benign appearance and very low risk of being a malignancy, and category 4B corresponding to the finding of a suspicious nodule with a high malignancy risk, requiring more intense diagnostic work-up including biopsy or resection. In Lung-RADS, thresholds for nodule diameters and corresponding risk estimates had been determined by a consensus panel of experts and were based on publications of data on a summary-level of various screening trials including NLST, ELCAP, and NELSON [19]. Recently, Pinsky et al. [19] published a retrospective analysis of the Lung-RADS performance in the NLST. Lung-RADS scores 1 and 2 were defined as negative and scores 3 to 4B as positive screening results. Performance was analyzed on participant level with respect to the presence of diagnosed malignancy. However, no individual nodules were tracked and no differentiation was made between part-solid nodules with various sizes of solid components (categories 3-4A-4B), since this information is not automatically available by the NLST-database. The fact that a subset of nonsolid nodules <20 mm in diameter is malignant was well discussed, but no further information on distribution of diameter with respect to histology was provided. We, therefore, focused our analysis on the two Lung-RADS categories 2 and 4B, the first with respect to the presence of malignancies despite of low risk categorization, the latter because of its potential overcall of transient infectious lesions as high-risk nodules triggering substantial management consequences.

In our study, we found a malignancy rate of 2.5% in Lung-RADS 2 subsolid lesions and 23.9% for Lung-RADS 4B subsolid lesions. Malignancy-rates published by Lung-RADS that include all nodule types are <1% for Lung-RADS 2 and > 15% for Lung-RADS 4B.

Lung-RADS chose for a relatively large cut-off of 20 mm for nonsolid nodules in the lowest risk category. Both the Fleischner Society and, if there is no previous imaging, also the British Thoracic Society, recommend an initial 3-month follow-up for nonsolid lesions ≥5 mm to assess persistence. The NCCN further differentiates between nodules <5 mm, 5–10 mm, and larger and propose a generally closer follow-up for nonsolid nodules: ≤5 mm yearly LDCT, and for lesions >5-10 mm an LDCT in 6 months, and for lesions >10 mm an LDCT in 3–6 months. Uniformly, all guidelines recommend more invasive diagnostics in subsolid nodules with new or growing (solid) components [1, 14, 20]. Pinsky et al. [19] reported in their analysis of NLST data that Lung-RADS underestimates the likelihood of lung cancer in subjects with nonsolid lesions <20 mm or nodules <6 mm. In line with the findings reported by Pinsky et al. [19] our analysis of NLST nodules also showed a slight underestimation of the Lung-RADS 2 probability score in subsolid nodules (2.5% versus <1%).

A recent publication by I-ELCAP reported that nonsolid nodules can be safely monitored with a 1-year follow-up scan [21]. Similarly, Scholten et al. [22] found that long-term follow-up appears to be safe to check for changes in SSNs. There are, however, studies that have reported a non-negligible percentage of nonsolid lesions representing invasive carcinomas [15; 16]. Hence, it is not yet clarified whether indeed all nonsolid nodules represent indolent malignancies or whether other texture features such as border characteristics or spatial attenuation characteristics beyond diameter or lack of solid component need to be considered to correctly assess malignancy risk [23, 24]. Part of this controversy is most likely related to the variability of differentiating nonsolid from part-solid nodules [25, 26]. A general lowering of the threshold to 15 or 10 mm seems not advisable given the disadvantage of inducing large numbers of false-positive follow-ups and the fact that I-ELCAP could not observe a stage shift/deterioration of patient outcome applying a 1-year follow-up strategy [21]. Interestingly 18% (8/44) of category 2 malignancies were resected within 1 year of the screening CT, another 55% (24/44) within the following 2 screening years, supporting the idea that—although classified as category 2—they displayed some morphological features that made them more suspicious. In addition to the size and nodule-type based categories Lung-RADS includes an additional category 4X triggering more intense diagnostic work up. This category is meant for categories 3, 4A, and 4B nodules that based on visual assessment of the radiologist demonstrate suspicious malignant imaging features justifying an upgrade to a more intense work-up. Currently, category 2 lesions are not included in such a procedure. Therefore, instead of lowering diameter thresholds, it could be an option to also allow for a subset of category 2 lesions an upgrade to category 4X, when radiologists find suspicious malignant imaging features. In that context it is noteworthy that according to Pinsky et al. [19] proportion of stage 1 cancers and 5-year lung cancer-specific survival were not found to be significantly different between malignancies in Lung-RADS category 2 and malignancies in Lung-RADS categories 3, 4A, or 4B subjects, indicating that category 2 malignancies cannot all be considered as indolent [19]. Their results are thus at least partially controversial to the findings published by Yankelevitz et al. [21]. It has to be noted that we do not know how many of the Lung-RADS 2 subsolid lesions in our study group represented adenocarcinoma in situ (AIS) minimally invasive adenocarcinoma (MIA) or indeed invasive adenocarcinomas. AIS and MIA have a (near) 100% disease-free survival after resection and usually grow slowly [27]. Several studies have investigated which morphological features would allow for differentiating pre-invasive from invasive nonsolid lesions [15, 28]. More research would be beneficial to reveal more insight in predictive morphological features of pre-invasive and invasive nonsolid nodules.

Quantification of the potential overestimation of category 4B SSNs was the other purpose of our study. This phenomenon actually has a larger impact on patient management, since category 4B results in a more intense and potentially invasive work-up including a “clinical chest CT, PET-CT, and/or tissue sampling depending on the probability of malignancy, comorbidities, and the radiologist’s final decision” [8].

Lung-RADS published an overall malignancy rate for all nodule types of >15% for category 4B, which is in line with the malignancy rate of 23.9% we found in our analysis for SSNs. This indicates that from a statistical point-of-view there is no overestimation. However, more importantly, one third (30.3%) of the benign 4B SSNs were transient as confirmed by follow-up. Although we conclude that the statistical malignancy risk of Lung-RADS 4B SSNs is in agreement with the Lung-RADS guidelines, our results represent a strong argument to include a short-term follow-up after three months to confirm persistency before considering more invasive management strategies, as also has been suggested by other guidelines. The NCCN suggests to perform a LDCT in 3 months to check for growth when there is low suspicion of lung cancer in their highest risk group for part-solid nodules (>8 mm lesion size). As for nonsolid nodules, the BTS recommends a 3-month CT repeat in cases when there is no previous imaging to assess persistence, growth, or change in morphology for part-solid nodules as well. Similarly the Fleischner Guidelines, designed for incidental nodule findings, recommend always to do a short-term at 3 months to confirm persistence in all part-solid nodules [1, 14, 20].

In both categories, we found that about one third of the lesions were transient. This is an important finding because it eliminates any need for further follow-ups. Previous studies report a wide range of numbers for the rate of transient SSNs (ranging between 12-70%) [11, 21, 29‐31]. Lee et al. [30] found a rather high rate of 70% of 126 transient part-solid lesions seen in an Asian population. On the other hand, Yankelevitz et al. [21] who analyzed screen-detected nonsolid nodules in a population comparable to the NLST, reported results more similar to our findings. Their study found that 26% (628/2392) of the nonsolid nodules seen at baseline screening subsequently resolved or decreased. The Multicentric Italian Lung Detection trial found similar results with 31% (15/48) of nonsolid nodules to be transient but a lower percentage of 12% (3/26) for part-solid nodules with a solid core <5 mm [31]. Although our finding generally matches previous literature, rate of persistence in SSNs appears to be influenced by type and study group inclusion.

Our study has some limitations. First, the most important limitation is the fact that SSNs not diagnosed as malignancies within a median period of 6.5 years were considered benign though no histological proof is available. However, this assumption seems to be acceptable based on current data knowledge and mirrored by the fact that current guidelines of the Fleischner society recommend a maximum follow-up of SSNs of 5 years [1]. Second, a small subset of SSNs could not be localized as described in the methods section and were excluded. However, this concerned only a small percentage of nodules and therefore it seems unlikely to have substantially influenced results. The NLST does not provide any information which nodule actually represented the resected malignancy. The standard of which nodule was considered malignant was determined by an experienced radiologist having anatomic information and follow-up scans available. To exclude any indeterminate or doubtful lesions a scoring system was defined and only lesions with high confidence scores were included in the final analysis. Third, pathologic subtypes were not available in the NLST database at the time of analysis. Thus, further differentiation between AIS, MIA and invasive adenocarcinomas was not possible. Last, we used the nodule type classification of the NLST-database, which is based on the assessment of the local screening radiologist. Given the observer variability for the assessment of presence and size of the solid core, it is likely that a subset of SSNs might have been classified differently by other observers [25, 26]. Taken together, the limitations might have slightly influenced the percentages described, but are considered unlikely to have any impact on our conclusions.

In conclusion, malignancy probabilities for Lung-RADS 2 and 4B generally match the malignancy risks in SSNs. The slight statistical underestimation of nonsolid nodules appears to be clinically less relevant given the generally indolent character of these lesions and the fact that the management difference only refers to the time interval of low-dose follow-up CT scans. An option to include also category 2 SSNs for possible upgrade to category 4X designed for more suspicious nodules might be considered in the future; however, more research is needed to define which imaging features qualify for such a procedure. Our finding that one third of the benign 4B SSNs were transient and thus falsely categorized as relatively high risk nodules is of significant clinical importance. Integration of a short-term low-dose follow-up to differentiate persistent from transient lesions would avoid unnecessary invasive work-up in Lung-RADS category 4B SSNs, and should, therefore, be considered in future upgrades of Lung-RADS.