Defining Optimal Lymph Node Yield in Gastrectomy: A Real-World Cohort Analysis

Abstract

Background

Gastric cancer (GC) has a high global mortality and incidence rate. Lymph node (LN) invasion is crucial in TNM staging, and an accurate LN staging system is vital for treatment decisions. However, the appropriate number of examined LNs remains uncertain.

Methods

We retrospectively analyzed consecutive GC patients who underwent gastrectomy at the First Medical Center of the Chinese PLA General Hospital from January 2010 to December 2023. A new statistical model based on the β-binomial distribution and maximum likelihood method in R software was employed to calculate false-negative probabilities.

Results

A total of 6463 GC patients were included. For cT1 patients, even with only five LNs excised, the likelihood of encountering occult positive LNs remained below 5%. For cT2 patients, 17 nodes were needed to rule out occult nodal disease with 90% confidence. While for cT3 and cT4 patients, even after the removal of 35 LNs, the likelihood of overlooking a positive node was still above 20%. Considering surgical extent, 25 nodes were required for patients who underwent proximal gastrectomy or distal gastrectomy to rule out occult nodal disease with 90% confidence, whereas those who received entire gastrectomy needed 59 nodes to achieve the same level of confidence.

Conclusion

Our study establishes a novel quantitative framework linking LN harvest thresholds to false-negative metastasis risk in GC, derived from real-world clinicopathological data.

Introduction

Gastric cancer (GC) ranks as the fifth leading cause of cancer-related mortality globally and has the fifth highest incidence rate worldwide [1–2]. Lymph node (LN) invasion, classified as the N stage, is a well-established prognostic factor correlating with recurrence and survival in patients who have undergone radical surgical resection for GC [3]. An accurate and standardized LN staging system is crucial for assisting physicians in making informed treatment decisions, evaluating various therapeutic options, and providing patients with clear insights regarding their prognosis.

The detection of positive LNs is crucial and requires an adequate number of examined LNs. The National Comprehensive Cancer Network recommended a minimum of 16 LNs is required for radical gastric resection to ensure accurate staging [4]. For the American Joint Committee on Cancer (AJCC) 8th edition stage system for GC, when more than 15 LNs are examined, the discriminatory performance is enhanced compared to the 7th edition [5]. Tumor invasion, classified by the T stage, is associated with the extent of N stage and has significant implications for prognosis. According to the report including 14,033 stage I–III GC patients from the Surveillance, Epidemiology, and End Results database, the minimum number of nodes examined for pT1 and pT2 GC were 7 and 24 [6]. While another indicated that the minimum number of nodes examined for primary tumor (pT1–pT4) were 6, 19, 40, and 66, respectively, to reach < 10% probability of missing nodal [7]. However, to our knowledge, no study has quantitatively analyzed the relationship between LN examination counts and false-negative nodal disease risk in GC using real-world surgical cohorts.

In this study, we aimed to present a novel statistical model designed to calculate the probability of occult nodal disease. This model is intended to assist in evaluating the risk of residual nodal disease and enhancing prognostic predictions.

Materials and methods

Patients’ selection

Consecutive GC patients who underwent R0 surgical resection for treatment at the First Medical Center of the Chinese People’s Liberation Army (PLA) General Hospital from January 2010 to December 2023 were screened. Related patients meeting the following criteria were selected: (1) an obvious pathologic diagnosis of GC; (2) primary cases without a history of stomach tumors; (3) undergo R0 resection; (4) one or more LNs surgically examined. The exclusion criteria were as follows: (1) patients receiving neoadjuvant therapy; (2) patients undergoing surgery for recurrent GC; (3) patients with evidence of distant metastatic disease. This study was conducted in accordance with the Declaration of Helsinki and approved by the Medical Ethics Committee of the First Medical Center of the Chinese PLA General Hospital (Approval No. S2021-022-01).

Data collection

The clinicopathological features were retrospectively studied, including the clinical T (cT) stage the surgical extent, examined LN number, positive LN number, and pathological N (pN) stage (The 8th Edition Staging System of AJCC). The condition of the LN was assessed by two experienced pathologists, and corresponding pathological reports were generated.

Statistical analyses

Utilizing the mathematical model outlined earlier, we proceed with the following assumptions and steps [8,9,10].

Assumption

All pathologically confirmed positive LNs were rigorously validated by two independent pathologists, ensuring diagnostic accuracy. Given the study’s focus on false-negative metastasis risk, we assumed no false positives in the pathological assessment. Patients can be classified into three distinct categories based on the pathological reports of examined LNs and their actual LN metastasis status: true negative (TN), true positive (TP), and false negative (FN). We can also derive measures such as positive predictive value and negative predictive value (NPV). The following notation and equations were used:

$${\text{ }}P(FN) = \frac{{\# FN}}{{\# FN + \# TP}}$$

$$\# TN = \# Observed\,Negative\,Patients - \# FN$$

$$\displaylines{\Pr evalence\,of\,LN\,Positive\,Disease \cr = \frac{{\# FN + \# TP}}{{\# FN + \# TP + \# TN}} \cr} $$

$$NPV{\text{ }} = {\text{ }}\frac{{\# TN}}{{\# TN + \# FN}}$$

Step 1: Assess the probability of FN findings regarding nodal status based on the number of examined LNs

Using the β-binomial distribution, we initially developed a mathematical model based on patients with positive LNs. These patients had at least two LNs examined and received at least one positive LN. This analysis was conducted utilizing a β-binomial distribution to establish a flexible probability model for the percentage of LNs that tested positive among patients with any known LN-positive disease. The maximum likelihood method in the VGAM package of R software was used to estimate the parameters of the model and the 95% confidence interval (CI) (including α and β). A unified set of α and β parameters was applied to all patients with confirmed LN-positive disease. We subsequently computed the check value for each potential number of nodes using the following equation (“m” was used to indicate the number of LNs examined. P (FN_m) indicated the probability of missing nodal disease when m nodes are examined):

$$\left( {\begin{array}{*{20}{c}} n \\ i \end{array}} \right){\text{ }}\frac{{B(i+\alpha,n+\beta - i)}}{{B(\alpha,\beta )}}$$

$$P({\text{F}}{{\text{N}}_{\text{m}}})=\frac{{{\text{Beta(}}\alpha {\text{,}}\beta {\text{+m)}}}}{{{\text{Beta(}}\alpha {\text{,}}\beta {\text{)}}}}$$

$$Where\,m = \# exa\min ed\,LNs\,from\,1\,to\,100$$

Step 2: Estimation of the number of patients with false LN-negative disease

Using the parameters obtained in step 1(including α and β), we further evaluated the number of false LN-negative patients in all patients. We calculated the number of false LN-negative patients at each value of m (the number of LNs examined) using the following equation (P(FN_m) indicated probability of missing nodal disease when m nodes were examined. #TP_m indicated the number of patients identified as node-positive when m nodes have been examined):

$$\# {\text{F}}{{\text{N}}_m}{\text{ = }}\frac{{{\text{P(F}}{{\text{N}}_m}{\text{)}} * {\text{\# T}}{{\text{P}}_m}}}{{\left[ {1 - {\text{P(F}}{{\text{N}}_m}{\text{)}}} \right]}}$$

Step 3: calculate the corrected prevalence of patients with true LN-positive disease

Due to the presence of false LN-negative, the observed prevalence of LN-positive was underestimated. The true prevalence of LN disease within the overall study population was estimated by taking into account the number of previously identified false LN-negative. The following equation was used:

$$\displaylines{ Corrected\,\Pr evalence = \Pr e\left( {True} \right) \cr = \frac{{\sum\nolimits_m {\left( {\# T{P_m} + F{N_m}} \right)} }}{{\sum\nolimits_m {\left( {\# T{P_m} + \# T{N_m} + \# F{N_m}} \right)} }} \cr} $$

Subsequently, we computed the corrected prevalence of patients across various cT stages and assessed the corrected prevalence associated with different surgical extent.

Step 4: compute the confidence in the NPV and false omission rate (FOR)

Lastly, we computed the NPV of estimated LNs for each value of m. The following equation was used:

$$\displaylines{ NP{V_m} = \frac{{\# T{N_m}}}{{\# T{N_m} + \# F{N_m}}} \cr = \frac{{1 - \Pr ev\left( {True} \right)}}{{1 - \Pr ev\left( {True} \right) + \left[ {\Pr ev\left( {True} \right)*P\left( {F{N_m}} \right)} \right]}} \cr} $$

NPV is referred to for confidence in a node-negative diagnosis given the absence of a positive LN. The NPV ranged from 0 to 100%, with higher values signifying increased confidence in the LN-negative diagnosis derived from the LN examination, thereby indicating that the patient was accurately classified as N0. The FOR was the complement of the NPV, with higher values signifying decreased confidence in the LN-negative diagnosis. The following equation was used:

$$FO{R_m} = 1 - NP{V_m} = \frac{{\# F{N_m}}}{{\# T{N_m} + \# F{N_m}}}$$

Follow-up

The follow-up was conducted through telephone interviews and outpatient visits at specific intervals: 3, 6, 9, 12, 18, 24, 30, 36, 48, and 60 months after surgery. The primary outcome measure was overall survival (OS). Patients were categorized into four groups based on the quartiles and median of the FOR. OS curves were generated utilizing the Kaplan-Meier method.

All analyses were conducted using R 4.4.1 (R Core Team, Vienna, Austria). The VGAM package was used to fit α and β parameters of the β-binomial distribution using a maximum likelihood approach. This study uses MathType (version 7.4.0.0) for equation editing. The production of OS curves was based on GraphPad Prism 8 software.

Results

Patient cohort

After a rigorous screening, a total of 6463 GC patients have undergone R0 surgical resection between January 2010 and December 2023 in our medical center. The steps of patient selection are shown in Fig. 1. The mean value of examined LN was 28.6 and the median was 26. The patient counts for cT1, cT2, cT3, and cT4 were 1609, 879, 2337, and 1638, respectively. According to postoperative pathology, 3718 patients (57.5%) were observed to have LN-positive disease. Of these, 951 patients were diagnosed with pN1, 1109 with pN2, and 1658 with pN3. These patients were analyzed to derive estimates of false-negative disease as a function of the number of LNs examined (Step 1). In addition, we observed that 1298 patients underwent proximal gastrectomy, while 3045 patients received distal gastrectomy, and 2120 patients underwent entire gastrectomy. (Table 1)

Fig. 1

The flow chart for patient selection

Table 1 No. of examined lymph nodes among 6463 patients

Step 1. Assess the probability of FN findings regarding nodal status based on the number of examined LNs

Among the 3718 LN-positive patients, at least one LN was positive and at least two LNs were evaluated. A β-binomial distribution model was fit from these patients. The parameters estimated by this model were α = 0.96 (95%CI, 0.93 to 0.99) and β = 2.20 (95%CI, 2.12 to 2.28). This parameter was subsequently utilized to estimate the probability of false-negative disease, based on the number of examined LNs. The probability of false-negative findings was estimated at 18.3%, 13.1%, 11.0%, 9.5%, 7.9%, 6.2%, and 4.9% for 5, 8, 10, 12, 15, 20, and 26 LNs examined, respectively. (Table 2; Fig. 2)

Table 2 Probability of missing nodal disease (false-negative) for selected values of examined lymph nodes

Fig. 2

The probability of false-negative findings on the basis of the number of lymph nodes examined. The horizontal solid reference lines correspond to 5% and 10%probability; the vertical dashed line corresponds to 12 and 26 nodes examined

Abbreviations: LN, lymph node

Step 2: estimate false LN-negative disease and true LN-positive disease

The true prevalence of LN-positive disease in these patients was then estimated for the overall population of 6463 patients. The observed prevalence and corrected prevalence (observed prevalence plus estimated false negative) rates were subsequently estimated (Table 3). The corrected prevalence of total patients was 63.7%. According to cT stage, the corrected prevalences were 14.9%, 48.5%, 81.5%, and 90.3% for the cT1, cT2, cT3, and cT4 stages. For the surgical extent, the corrected prevalences were 58.7%, 57.9%, and 74.6% for proximal gastrectomy, distal gastrectomy, and entire gastrectomy.

Table 3 Observed and corrected prevalence of nodal disease

Step 3. Risk of occult nodal disease

Finally, we integrated our refined estimates of the true prevalence of LN positivity with the likelihood of not detecting a positive LN in order to assess the risk of residual occult nodal disease following surgical resection. Totally, to rule out occult nodal disease with 90% confidence, 33 nodes were needed (excluding the consideration of the cT stage) (Fig. 3). For cT1 patients, even in cases where only five LNs are excised, the likelihood of encountering occult positive LNs remains below 5%. For cT2 patients, 17 nodes were needed to rule out occult nodal disease with 90% confidence. For cT3 and cT4 patients, even after the removal of 35 LNs, the likelihood of overlooking a positive node remains above 20% (Table 4; Fig. 4). Besides, when excluding the consideration of the cT stage, for patients who received proximal gastrectomy or distal gastrectomy, 25 nodes were needed to rule out occult nodal disease with 90% confidence. For patients who received entire gastrectomy, to rule out occult nodal disease with 90% confidence, 59 nodes were needed (Table 5; Fig. 5).

Fig. 3

The false omission rate (FOR) on the basis of the number of lymph nodes examined. The horizontal solid reference lines correspond to 5% and 10%probability; the vertical dashed line corresponds to 33 and 74 nodes examined

Table 4 The false omission rate on the basis of the number of lymph nodes examined, stratified by clinical T stage

Fig. 4

The false omission rate (FOR) on the basis of the number of lymph nodes examined, stratified by Clinical T stage. The horizontal solid reference lines correspond to 5%, 10%, and 20% probability

Table 5 The false omission rate on the basis of the number of lymph nodes examined, stratified by surgical extent

Fig. 5

The false omission rate (FOR) on the basis of the number of lymph nodes examined, stratified by surgical extent. The horizontal solid reference lines correspond to 5%, 10%, and 20% probability

Overall survival

Follow-up information was available for 2320 patients, with a median follow-up duration of 51 months. Among these patients, 1,048 were diagnosed with pN0 disease, and the median follow-up time for this group was 61 months. The OS probability among patients with a diagnosis of pN0 disease is shown in Fig. 6 (stratified by cT stage and surgical extent). Higher FOR corresponded to worse OS.

Fig. 6

Overall survival probabilities among patients diagnosed with node-negative disease, as a function of the quartiles of the estimated false omission rate (FOR), stratified by Clinical T stage and surgical extent

Discussion

Our study aimed to develop a novel statistical model to calculate the probability of occult nodal disease in GC. By retrospectively analyzing data from a large cohort of GC patients who underwent R0 surgical resection at the First Medical Center of the Chinese PLA General Hospital, a series of parameters and estimates were derived, providing valuable insights into the relationship between LN examination and the risk of false-negative nodal staging.

The cT stage is the crucial element of the AJCC staging system for GC, as an elevated cT stage correlates with advanced disease progression, increased LN metastasis, and poorer prognosis [11]. Regarding the minimum number of LNs to be dissected based on the cT stage, our study revealed significant differences. For cT1 patients, the risk of occult positive LNs remained relatively low even with only five LNs excised, suggesting that a relatively smaller number of dissected nodes might be sufficient to achieve a reasonable level of certainty in ruling out nodal involvement. Therefore, endoscopic resection, represented by endoscopic mucosal resection and endoscopic submucosal dissection was the curative treatment approach for early GC and its precursor lesions [12]. However, as the cT stage advanced, the situation changed dramatically. For cT2 patients, 17 nodes were required to rule out occult nodal disease with 90% confidence, indicating a need for more extensive lymphadenectomy. In cT3 and cT4 patients, even after removing 35 LNs, the likelihood of overlooking a positive node was still above 20%, highlighting the complexity and aggressiveness of the disease in advanced stages and the challenge of achieving accurate nodal staging. Our research aligns with the emphasis in the guidelines on the extent of lymph node dissection stratified by cT stage [13]. According to the Japanese Gastric Cancer Treatment Guidelines (6th edition), for cT2 + tumors, standard D2 dissection remains the gold standard. The term ‘personalization’ in our model refers to identifying subgroups where limited dissection (e.g., D1 + for select cT1 tumors) may suffice without compromising staging accuracy, not advocating for extended dissection beyond D2 in advanced cases. Prophylactic super-extended lymphadenectomy for staging alone is unjustified and potentially harmful. While our model identifies minimum LN thresholds, it does not supersede anatomical dissection standards. Overtreatment risks persist if quantitative targets are pursued without regard to surgical quality or tumor biology.

Similar patterns emerged when considering the surgical extent. For patients who underwent proximal gastrectomy or distal gastrectomy, 25 nodes were needed to rule out occult nodal disease with 90% confidence. In contrast, those who received an entire gastrectomy required a much larger number, 59 nodes, to reach the same level of confidence. According to previous studies [14,15,16], proximal gastrectomy was sufficient for proximal GC, and entire gastrectomy did not lead to longer overall survival or cancer-specific survival. This disparity could be attributed to differences in lymphatic drainage patterns and tumor spread characteristics associated with different surgical regions. Surgeons should be aware of these site-specific requirements to optimize the surgical procedure and improve patient outcomes.

The LN ratio (LNR) was another nodal staging system for postoperative decision-making [11, 17]. Clinically, a higher LNR indicated a greater likelihood of tumor spread and poorer survival prospects. It provided a more refined prognostic evaluation compared to simply counting the number of positive LNs. Nevertheless, LNR had its limitations. In cases where the number of examined LNs was insufficient, the calculated LNR could be inaccurate, leading to misjudgment of the disease severity [18]. Thus, our LN staging model presented in this paper accounts for false negative results arising from insufficient LN dissection. Particularly for patients classified as pN0, this model offered a more accurate evaluation of their postoperative status and staging.

Previous reports have also emphasized the significance of adequate LN retrieval for accurate staging. However, the specific numbers recommended varied. Some studies proposed different minimum numbers of nodes to be examined based on the cT stage, similar to our findings, while others focused on overall survival and recurrence rates in relation to lymphadenectomy extent [6–7]. Our model quantifies the false-negative probability of nodal metastasis based on clinicopathological variables, including tumor location and cT stage. This probabilistic framework may serve as a supplementary tool to guideline-recommended lymphadenectomy, aiding surgeons in balancing dissection extent against individual patient risks.

The models designed to identify false negative LNs and occult LN metastasis were extensively utilized in the management of various solid tumors, including colon cancer [8, 19], thyroid cancer [9, 20], lung cancer [10], pancreatic cancer [21], renal cell carcinoma [22], and ovarian cancer [23]. Similarly, in this study, the application of the β-binomial distribution played a pivotal role in assessing the probability of occult nodal disease in GC patients. It allows for a more accurate representation of the variability observed in real-world clinical data compared to simpler distribution models. Unlike some traditional methods that assume a fixed probability of LN-positivity, the β-binomial distribution can account for the heterogeneity in the lymphatic spread of cancer cells. This led to more precise estimates of the false-negative dissection probabilities for various numbers of examined nodes, providing clinicians with a more reliable tool for decision-making. Besides, our findings validate that inadequate LN dissection directly compromises survival, likely through false-negative nodal staging leading to inappropriate adjuvant therapy allocation. These data strongly support tailoring dissection extent to cT stage and anatomical location, as proposed in our risk-adapted model. Moreover, by using the maximum likelihood method in the VGAM package of R software to estimate the parameters (α and β) of the β-binomial distribution, we were able to obtain robust estimates with associated confidence intervals. This added a level of statistical rigor to our analysis, enhancing the credibility of our findings.

Despite the valuable insights provided by our study, several limitations should be acknowledged. Firstly, the data were retrospectively collected from a single medical center, which may introduce selection bias and limit the generalizability of our findings. Patient populations may not be fully representative of the global GC patient cohort, which may affect the accuracy of statistical models applied in different settings. Secondly, our β-binomial model operates under two key premises: (1) Uniform Probability Assumption: All LNs within a patient are treated as having equal metastasis risk. (2) Exchangeability Assumption: LNs are considered interchangeable regardless of station. We recognize these simplifications may not fully reflect biological reality, as proximal nodes inherently have higher metastatic likelihood than distant nodes. However, our approach was necessitated by two constraints: (1) Data Availability: Real-world surgical pathology reports rarely document LN station-specific counts, limiting station-stratified analysis. (2) We designed a simplified model that does not track the exact locations of LNs. Despite its limitations, the exchangeability hypothesis finds support in clinical evidence-Solitary Skip Metastasis. Approximately 1.0-2.8% of patients exhibit metastasis to “unexpected” stations without involvement of proximal nodes, suggesting non-hierarchical spread patterns in subsets [24]. Finally, external factors like differences in surgical techniques and pathological evaluation standards among different surgeons were not accounted for. These variations could influence the number of examined and positive LNs, further challenging the direct application of our results without considering local practice nuances.

Conclusion

Our study establishes a novel quantitative framework linking LN harvest thresholds to false-negative metastasis risk in GC, derived from real-world clinicopathological data. Future studies could further validate and expand on these findings, potentially leading to improved clinical guidelines and patient prognoses.