A homogenized approach to classify advanced gastric cancer patients with limited and adequate number of pathologically examined lymph nodes

Gastric cancer (GC) is the second most common cancer in China [1] and the third leading cause of cancer-related death worldwide [2]. Its high mortality rate can be largely attributed to the early spread of cancer cells via the lymphatic system [3]. As a result, the number of surgically retrieved lymph nodes (LNs) is positively correlated with the likelihood of complete removal of all perigastric metastatic LNs and is among the most determining factors for assessing the quality of gastrectomies [4, 5].

Studies have shown that patients with ≥ 16 pathologically examined LNs (eLNs) have better prognoses as compared to those with < 16 eLNs [6, 7]. The two major acknowledged bodies in the gastric oncological field, the National Comprehensive Cancer Network (NCCN) [8] and the American Joint Committee on Cancer (AJCC) [9], both advocate the retrieval of at least 16 LNs for optimizing the radicality of D2 lymphadenectomies and enabling proper staging of the disease.

However, performing such gastrectomies are challenging as they necessitate a high level of expertise due to the risky anatomical surrounding structures. In Asian countries (such as China, Japan and Korea), these are comparatively easier as high-volume institutions refer to centers performing at least 200 gastrectomies annually in which high-quality lymphadenectomies are routinely achieved [10], whereas in Western settings, these might prove arduous due to their smaller number of cases and where “high-volume” institutions often refer to centers with 15–20 GC cases per year [11, 12]. Correspondingly, an analysis of 18,043 GC patients from the Surveillance, Epidemiology, and End Results Program (SEER) database [13], studies from 691 hospitals in the United States [14] and 18 hospitals in the United Kingdom [15] found that only 33%, 40% and 31% of their respective gastrectomies met the minimum recommended number of surgically retrieved LNs.

In addition, patient-related factors such as elderly patients presenting with comorbidities [16], obese patients in whom adipose tissues often obscure surgical visibility for retrieval of LNs intermingled with major blood vessels [17–19] and/or surgery-related factors such as risky tumor locations (esophagogastric or gastro-duodenal junctions) and large tumor size may also complicate surgeries [20]. These commonly observed factors in the day-to-day practice may at times compel even experienced surgeons to opt for a limited number of LNs retrieved (< 16 eLNs).

Consequently, the worldwide annual total number of gastrectomies with limited eLNs may agglomerate to a substantial proportion, yet neither the AJCC nor the NCCN committees have proposed a standard classification for this category of patients. Clinically, this often leads to therapeutic confusion, coercing an unstandardized approach by oncologists with no other choice than to rely on their own clinical judgment for the counseling [21] and planning of treatments for this category of patients [22].

Thus, the aim of this study was to develop an easy-to-adopt and clinically reliable stratifying approach for homogenizing the nodal classification of patients with limited eLN in cohorts of GC patients comprising of both patients with limited and adequate number of pathologically eLNs.

In this study, we have concisely demonstrated the significant heterogeneous correlation in survival existing between the corresponding nodal subgroups of the Limited and Adequate eLN cohort when adhering to the customary 8th AJCC pN and pTNM stratification system, which resulted in substantial prognostic difference between these two cohorts of patients. The proposed N′ classification was able to largely compensates for this significant stage migration/misclassification by providing a simple, alternate and more homogenous stratification approach.

The N′ classification also displayed superior stratification and 5-year OS prognostic reliability compared to the two other most studied nodal classification in gastric oncology. Nevertheless, one important hindrance of the N′ classification was that it demonstrated a lower discriminatory ability (lower value of linear trend χ²) as compared to the LNR. We hypothesize that this may have been resulted due to the LNR’s algorithm that groups patients with similar nodal ratio irrespective of their total number of LNs retrieved. For instance, patients with 1/10 and 3/30 eLNs (no. of metastasized LNs/total no. of eLNs) would be grouped within the same nodal subgroup if using the LNR but when in fact they are classified as pN1 and pN2, respectively, using the 8th AJCC N classification. Therefore, when such sub-groups were correlated with OS, they tended to merge instead of differentiating their survival differences, thereby resulting in a better discriminatory ability but at the expense of prognostic accuracy. In addition, the LNR has other inherent drawbacks when compared to the N′ classification. First, it tended to demonstrate significant stage migration resulting in an under-/over-estimation of prognoses, particularly for less advanced cases as illustrated by the juxtaposed survival curves in Fig. 2 (N′0 vs. LNR 1, P = 0.017). Second, grouping patients with marked differences in overall number of eLNs together within the same group is an analytical bias as these patients have different survival outcomes, making the LNR clinically less reliable and may be one of the contributing factors for its reluctant global acceptance by professional committees such as the AJCC and NCCN.

The AJCC pN classification allows patients’ grouping uniformity in terms of survival by providing a stable stratification method, unlike the LNR, which is hindered by its unstable varying cut-off values between different populations [24, 25], similar populations at different institutions [26, 27], or even similar institutions but with different sample sizes analyzed at different time periods [28, 29]. Furthermore, the LNR and other similar strategies [30, 31] has cumbersome applicability in the daily busy clinical settings as they demand a high level of complicated calculations and do not comply with the easy-to-remember, straightforward AJCC nodal classification criteria. Moreover, it has been demonstrated in an Italian study by Pedrazzani et al. [32]. that the LNR approach has limited utility for patients who have had few eLNs. In contrast, since the N′ classification uses similar stratification analogy to the AJCC N classification, these hurdles are easily overcome, giving it greater potential to be more widely adopted.

Further, multivariate analyses of the N′ classification found that the number of eLNs was not an independent prognostic factor for survival. Also, we noticed that as the prognostic performance of the nodal classification increases, the P-value representing its independent correlation with survival increases as well (N, P < 0.001; LNR, P = 0.041; N′, P = 0.374). We, therefore, hypothesize that as the homogeneity in survival between the Limited and Adequate eLN cohort is increased, this decreases the impact of the eLNs category as an independent factor since more emphasis was placed on subgroups’ classification.

Regarding the stage classification, the data from Table 3 show that using the 8th AJCC classification, the 5-year OS of substages IB–IIIB of the Limited eLN cohort approximated those of IIA–IIIC of the Adequate eLN cohort, respectively, with significant intersection observed between their survival curves (all P > 0.05). By implementing the TN′M classification, the heterogeneity between these two cohorts was observed to decrease, resulting in significant improvement in the stratification (wider survival range and HR values) and prognostication of the patients (amelioration in the 5-year OS prognostic estimation and superior AIC values). Prior to nodal subgroups re-classification, although there was no significant difference observed between the corresponding substages IB and IIA of the Limited and Adequate eLN cohorts, however, after re-arrangement using the TN′M classification, improved results were obtained as shown by the substantial increase in P values from 0.136 to 0.677 and 0.169 to 0.995, respectively. This, therefore, demonstrates an enhanced prognostic estimation approximating to that of the actuarial 5-year OS of the patients (Table 3). Of note, considering that the pT and pM categories were kept constant in all the analyses and that the LNR performance was inferior compared to the N′ classification, for ease of interpretation and avoiding repetition of data, only the prognostic performance for the TN′M and 8th AJCC pTNM classification were provided for this study.

If the stratifying technique proposed in this study can be widely validated, we expect that the proposed novel, optimized and homogenized classification to significantly impact treatment decisions as not only the survival prognostication would be more accurately determined, but most importantly, this approach does not significantly affect the prognostication of patients with Adequate eLNs. Therefore, to a certain extent, we predicate that the proposed classification could facilitate the enrollment of patients, on a more individualized basis, in clinical trials comprised of both categories of patients and improve their obtained results. As an annotation, based on ethical practices, the results of this study are not to be considered for restricting the extent of LN retrieval but, to be used post-gastrectomy for improving the prognostic estimation of patients with limited eLNs, which were due to unexpected circumstances (patient’s or surgical related factors). Thereby, providing oncologists with an unbiased, easy to use, more standardized, and individualized approach for selecting treatment modalities and follow-up evaluations for this category of patients rather than relying on their personal judgment. Subsequently, patients with limited eLNs, especially those staged as IB when using the proposed classification, would have a lesser risk of being under-treated or under-followed.

The limitations of this study are worth mentioning. Patients with early GC could not be analyzed for the following reasons. First, the number of early GC cases were limited in our database as compared to our advanced cases. Second, in our institution, the surgical treatment for patients with T1 disease varied considerably from endoscopic resection (endoscopic submucosal dissection or endoscopic mucosal resection) [33] to D1 and/or D1+ gastrectomy [34] and many of these patients were not treated in our department. Therefore, early GC cases were omitted in the calculations since they could have biased the findings of this study. Second, because of the retrospective and mono-institutional nature of this study, the findings need validation from large multicentered-cohorts and/or in different population settings before being widely applied.

In conclusion, we have developed an easy-to-adopt, reliable and practical stratification approach which uses similar analogy as the 8th AJCC nodal classification to homogeneously classify cohorts of advanced GC patients comprising of both limited and adequate numbers of pathologically eLNs.