Comparison of three lymph node staging schemes for predicting the outcome in patients with small bowel adenocarcinoma: A population-based cohort and international multicentre cohort study.

BACKGROUND: The prognostic roles of three common lymph node staging schemes, number of positive lymph nodes (pN), lymph node ratio (LNR) and log odds of positive lymph nodes (LODDS) in small bowel adenocarcinoma (SBA) are unclear. We assessed their prognostic ability in SBA.
METHODS: A total of 2128 patients diagnosed with SBA between 1988 and 2010 from the Surveillance, Epidemiology, and End Results (SEER) database and 186 patients from 15 hospitals in France and China were identified. We evaluated the prognostic ability of the schemes in both continuous and stratified patterns using R2, Harrell's C, and time-dependent receiver operating characteristic curve analyses.
FINDINGS: For continuous pattern, the LODDS had a better capacity of discrimination and higher accuracy of prognosis than pN and LNR. Similarly, the stratified LODDS classification had a better performance of discrimination and higher accuracy of prognosis than the pN and LNR classification. The multivariable model using the LODDS classification also showed superiorly predictive accuracy and discriminatory capacity to those of the 7th and, 8th TNM node and LNR classification. These results were fully validated in an independent international multicentre cohort.
INTERPRETATION: The LODDS scheme showed a better prognostic performance than the LNR or pN schemes in patients with SBA regardless of continuous or stratified pattern. The LODDS scheme could serve as an auxiliary to lymph node staging systems in future revisions of the American Joint Committee on Cancer (AJCC) manual. FUND: This work was funded by the Zhejiang Province Natural Science Fund of China.

Evidence before this study

We search the literature on the treatment and survival of small bowel adenocarcinoma and evaluation of lymph node metastasis in patients with small bowel adenocarcinoma in PubMed database. We found that lymph node metastasis has been considered as one of the most robust prognostic markers for small bowel cancer using multivariate analysis. However, the role of the number of positive lymph nodes, lymph node ratio and log odds of positive lymph nodes in small bowel adenocarcinoma are ill-defined.

Added value of this study

This large study involved two different cohorts to compare the predictive capacity of three lymph node staging schemes for the survival of small bowel adenocarcinoma patients. We found a superior prognostic ability of the log odds of positive lymph nodes scheme in patients with small bowel adenocarcinoma from the Surveillance, Epidemiology, and End Results database, and uniquely verified this finding in a cohort of patients from France and China.

Implications of all the available evidence

Lymph node metastasis is an important prognostic indicator for cancers. The rarity of small bowel adenocarcinoma hinders multidisciplinary management of patients with surgery. The log odds of positive lymph nodes scheme could provide surgical guidelines to improve the prediction of prognosis of small bowel adenocarcinoma and could be included in the lymph node staging system in future revisions of the American Joint Committee on Cancer (AJCC) manual.

Alt-text: Unlabelled Box

Introduction

Small bowel cancers are rare malignant tumors, comprising around 3·2% of cancers of the digestive system [1]. They include tumors derived from diverse histology, mainly carcinoids, adenocarcinomas, lymphomas, and sarcomas [2]. Among them, small bowel adenocarcinoma (SBA) is the most common histology of tumors in the United States, with a rising incidence not only in North American and Western Europe, but also in Asia [3,4]. Surgical excision remains the foundation of therapy for SBA manifesting as locoregional disease [5]. The rarity of SBA hinders further understanding of its molecular mechanism, resulting in current bottlenecks in the multidisciplinary management of patients after surgery [2,6]. Thus, there is an urgent need to identify prognostic performance metrics.

In addition to the presence of distant metastasis [7], the most prognostic factor for patients with SBA, lymph node metastasis was also shown to be an important independent prognostic factor for SBA in multivariate analysis [8,9]. Adequate lymph node histopathological assessment is critical for accurate pN staging [10]. In order to avoid stage migration effects, adequate evaluation of the lymph node status is important. However, 82% of patients with SBA did not have adequate lymph nodes examination in our previous study based on a minimal of 17 retrieved lymph nodes. In this context, we tried to identify new measures of lymph node status with the combination of the number of retrieved lymph nodes.

To the best of our knowledge, two measures, namely lymph node ratio (LNR), log odds of positive lymph nodes (LODDS) have been proposed. LNR is calculated by dividing the number of positive lymph nodes by the total number of excised lymph nodes. LNR provides important guidance regarding the survival of patients with SBA [11]. Few intensive studies on LNR have shown its superiority in guiding the prognosis of SBA over that of the numbers of positive lymph nodes (pN). LODDS, calculated as the log of the ratio between the number of positive and negative lymph nodes, has been applied to predict the prognosis of several tumors [[12], [13], [14], [15]]. There is little clinical evidence demonstrating the role of LODDS in SBA and the most suitable scheme to describe prognosis of SBA remains ill-defined.

Therefore, this study compared the prognostic ability of the lymph node schemes in SBA by means of the population-based Surveillance, Epidemiology, and End Results (SEER) database with patients diagnosed between 1988 and 2010 and an international multicentre cohort of 15 hospitals that was built to validate our findings.

Materials and methods

Data source and patients

Data of patients with SBA during 1988–2010 were obtained from the SEER database. The patient eligibility criteria were as follows: (1) primary tumor located in the small intestine; (2) ≥18 years of age; (3) treatment with primary tumor surgery; (4) without radiotherapy for the first round of therapy; (5) with close follow-up; (6) at least one lymph node was retrieved; (7) histologically-confirmed SBA. Another independent international multicentre cohort of 186 patients with SBA from 15 hospitals in France and China from 1998 to 2010 with the same inclusion and exclusion criteria was developed for the validation of the LNR and LODDS schemes. In this study, cause-specific survival (CSS) was defined as death caused by SBA and overall survival (OS) as death regardless of any causes. The primary outcome was CSS with OS and CSS considering competing death due to non-SBA death as the secondary outcome.

Definitions of classification

To describe lymph node status more precisely, we attempted to evaluate the prognostic performance of the American Joint Committee on Cancer (AJCC) 7th edition N stage (7th pN classification), AJCC 8th edition N stage (8th pN classification), LNR classification, and LODDS classification in patients with SBA. According to the 7th edition of AJCC TNM staging, N0 was defined as no regional lymph node metastasis, N1 as metastasis in one to three regional lymph nodes and N2 as metastasis in four or more regional lymph nodes. However, the definition of regional lymph nodes changed in 8th edition. N1 was redefined as one or two regional lymph nodes and N2 as more than two positive nodes [16]. LODDS was defined as loge[(pN + 0·5)/(nN + 0·5)] [17], where pN is the number of positive lymph nodes and nN is the number of negative nodes retrieved.

Statistics

The correlation of continuous LNR, pN, and LODDS were calculated using Spearman coefficients. The relationship between LNR, pN or LODDS and SBA survival were determined by univariate Cox regression model with a restricted cubic spline function for each variable. The cut-off points to define LNR or LODDS lymph node staging classification were determined using X-tile software [18], which provides a single, global assessment of every possible way of dividing a population into low-, medium-, and high-level marker expression. The three optimal divisions of the data were identified by selecting the highest X2 value with statistical significance assessed by using the cutoff point derived from a training set to parse a separate validation set with standard log-rank tests.

The prognostic performances for CSS of the lymph node staging schemes were compared. Regarding the overall model performance assessment, R2 demonstrated the survival variability that could be explained by a predictive model [19], and high R2 value indicated superiority of prognostic model. Harrell's C statistic with corresponding 95% confidence intervals (95%CIs), as a measure of discriminatory power, was used to evaluate the proportion of positive predictive value [20]. The bootstrap method (N = 1000) was used to calculate bias-corrected Harrell's C statistic [21]. To evaluate the discrimination power of the three lymph node schemes for time-dependent survival, the area under receiver operating characteristics (AUROC) of time-dependent receiver operating characteristic curve (tdROC) analyses over 12 years were made for CSS, OS, and competing risk model, which dynamically displayed the prognostic superiority of the LODDS scheme [22].

For the extracted patients, each incomplete variable was imputed by a separate model based on fully conditional specifications [23]. Multivariable Cox regression models were built to jointly assess the prognostic ability of the three lymph node staging schemes and other possible prognostic indicators. Each candidate variable with a P-value <0·1 in the univariate model was included in the multivariate model.

Sensitivity analyses were performed to assess our results. Firstly, the prognostic performance for OS using a Cox regression model and CSS using regression modeling of subdistribution functions in consideration of competing risk of non-SBA death for the three lymph node staging schemes were compared. Secondly, the established novel LNR or LODDS classification schemes were validated in the cohort of patients diagnosed from 2011 to 2014. Thirdly, a multivariable Cox regression model was used to analyze prognostic performance of the three node staging schemes and other possible prognostic predictors after excluding patients with missing values. All data analyses and drawings were completed using R software 3·4·3 (http://www.r-project.org).

Results

Characteristics of patients

A total of 2128 qualified participants with SBA diagnosed from 1988 to 2010 in the SEER database were enrolled in this study (Supplementary Fig. 1). The clinical characteristics and distributions of different lymph node staging systems are described in Table 1. In SEER database, 172 (8·10%) patients had SBA of stage I, 699 (32·93%) stage II, 830 (39·10%) stage III, 422 (19·88%) stage IV with the five-year CSS rates of stage I to IV being 82·4%, 58·7%, 31·3%, and 7·85%, respectively (Supplementary Fig. 2). Meanwhile, a total of 186 patients from an international multicentre cohort was included for the independent validation of our findings (Table 1).

Table 1

Clinical characteristics of for patients with small intestine adenocarcinoma from the Surveillance, Epidemiology, and End Results database and multicentre cohort.

Factor		SEER database	Multicentre cohort
		N = 2128	N = 186
Sex	Male	1102 (52%)	89 (48%)
	Female	1026 (48%)	97 (52%)
Race	White	1659 (78%)	/
	Black	344 (16%)	/
	Others	125 (6%)	/
Age (Years)	≤ 60	907 (43%)	84 (45%)
	> 60	1221 (57%)	102 (55%)
Marriage	Yes	1259 (61%)	/
	No	806 (39%)	/
Size	≤ 2	233 (13%)	/
	≤ 2	365 (20%)	/
	≤ 3	619 (35%)	/
	>5	574 (32%)	/
T categorya	T1	103 (5%)	4 (2%)
	T2	118 (6%)	23 (12%)
	T3	900 (47%)	84 (45%)
	T4	777 (41%)	74 (40%)
M stage	M0	1706 (80%)	165 (89%)
	M1	422 (20%)	21 (11%)
Tumor site	Duodenum	839 (46%)	126(68%)
	Ileum	411 (23%)	25(13%)
	Jejunum	539 (30%)	35(19%)
	Others	34 (2%)	/
Grade	I/II	1201 (61%)	125 (67%)
	III/IV	778 (39%)	61 (33%)
Total no. of nodes retrieved	Median (IQR)	8 (4, 14)	6 (4,11)
No. of positive nodes (pN)	Median (IQR)	1 (0, 3)	0 (0,2)
LNR	Median (IQR)	0·083 (0, 0·454)	0 (0,0·362)
LODDS	Median (IQR)	−1·609 (−2·617, −0·167)	−1·807 (−2·565,−0·511)
7th pN classification	N0	980 (46%)	98 (53%)
	N1	753 (35%)	57 (31%)
	N2	395 (19%)	31 (17%)
8th pN classification	N0	980 (46%)	98 (53%)
	N1	604 (28%)	45 (24%)
	N2	544 (26%)	43 (23%)
LNR classification	LNR1 (≤0·02)	981 (46%)	98 (53%)
	LNR2 (0·02–0·47)	628 (30%)	48 (26%)
	LNR3 (>0·47)	519 (24%)	40 (21%)
LODDS classification	LODDS1 (≤ − 1·89)	925 (43%)	38 (36%)
	LODDS2 (−1·89– −0·51)	589 (28%)	33 (31%)
	LODDS3 (> − 0·51)	614 (29%)	36 (34%)

Among all 2128 patients in SEER database, the number of missing items for marriage, tumor site, size, tumor category and grade were 63 (3%), 305 (14%), 337 (16%), 230 (10.8%) and 149 (7%), respectively.

LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes.

Tumor category and 8th pN were graded according to the 8th edition of the tumor node metastasis (TNM) classification of malignant tumors proposed by the American Joint Committee on Cancer/International Union Against Cancer (AJCC/UICC), whereas 7th pN was graded according to the 7th edition of the TNM staging manual. pM indicates metastatic disease pathologically coded 85 in “EOD 10-extent (1988–2003)” of SEER data.

Characteristics of the lymph node staging schemes

The relationship between LNR and pN, LODDS and LNR, and LODDS and pN are displayed as scatterplots in Supplementary Fig. 3. The LODDS was more highly correlated with LNR than with pN (r = 0·885 versus r = 0·757). The relationship between log hazard ratio and LODDS and between LNR and pN were shown in Supplementary Fig. 4a-c. In the respective lymph node schemes, the mortality risk rose as the value of LODDS, LNR or pN increased, while the trends of function were not linear simply. Moreover, it was suggested from restricted cubic spline analyses that the mortality risks also increased as LODDS increased among patients with no positive lymph node involved (Supplementary Fig. 4d).

Prognostic abilities of LODDS, LNR, and pN in the SEER database

As shown in Table 2, the overall performance of the LODDS had a higher R2 value (0·184) than LNR (0·158) or pN (0·068) schemes, indicating its superiority as a prognostic factor. Measured by Harrell's C statistic, the LODDS (0·673; 95%CI: 0·656–0·691) could provide a better discriminatory capacity than those of the pN (0·629; 95%CI: 0·614–0·645) and LNR (0·655; 95%CI: 0·640–0·671).

Table 2

Univariable analysis for overall prognostic performance of node staging schemes for small intestine adenocarcinoma for cause-specific survival.

	SEER database				Multicentre cohort
	R2	Harrell's C	Bootstrap	AUC (5 year)	R2	Harrell's C	Bootstrap	AUC (5 year)
pN	0·068	0·629 (0·614–0·645)	0·629	69·70 (67·42–71·97)	0·049	0·599 (0·545–0·653)	0·599	64·83 (56·11–73·56)
LNR	0·158	0·655 (0·640–0·671)	0·655	71·88 (69·70–74·06)	0·110	0·624 (0·570–0·678)	0·623	67·06 (58·87–75·25)
LODDS	0·184	0·673 (0·656–0·691)	0·674	74·61 (72·33–76·89)	0·131	0·647 (0·587–0·707)	0·646	69·09 (59·88–78·30)
7th pN classification	0·129	0·626 (0·611–0·642)	0·627	69·21 (66·96–71·46)	0·077	0·604 (0·550–0·657)	0·600	66·05 (57·71–74·39)
8th pN classification	0·130	0·626 (0·611–0·642)	0·626	69·51 (67·25–71·77)	0·069	0·593 (0·540–0·646)	0·583	64·83 (56·11–73·56)
LNR classification	0·157	0·644 (0·629–0·660)	0·644	70·64 (68·47–72·82)	0·090	0·615 (0·561–0·668)	0·611	65·98 (57·59–74·37)
LODDS classification	0·178	0·656 (0·640–0·671)	0·656	72·30 (70·13–74·46)	0·120	0·629 (0·574–0·683)	0·627	67·61 (59·12–76·10)

LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes.

Analyses based on AUC measures (Table 2, Supplementary Table 1) showed LODDS (five-year: 74·61%; 95%CI: 72·33%–76·89%) provided better discriminatory capacity than those of pN (five-year: 69·70%; 95%CI: 67·42%–71·97%) and LNR (five-year: 71·88%; 95%CI: 69·70%–74·06%), which was consistent with results measured by the Harrell's C indices. As shown in Fig. 1, the LODDS had higher one-, three-, five-, seven-, and ten-year AUROC for CSS, indicating its better performance for determining the prognosis of patients with SBA. The sensitivity analysis using OS as outcome instead of CSS still identified the superiority of the LODDS over LNR or pN (Supplementary Table 1, Supplementary Table 2 and Supplementary Fig. 5) as measured by R2, Harrell's C, and AUC. Meanwhile, we also compared the R2, Harrell's C, and AUC values of the three lymph node staging schemes in the competing risk model in consideration of a competing risk of non-SBA death with similar results (Supplementary Table 1, Supplementary Table 3 and Supplementary Fig. 6). Besides, the sensitivity analysis was also performed after eliminating SBA patients with systemic metastases. Results indicate that LODDS had a higher R2 (0·164), Harrell's C (0·665; 95%CI: 0·647–0·684), and five-year AUC (69·20%, 95%CI: 66·83%-71·58%) than those of pN and LNR, reflecting its better discriminatory capacity and prognostic performance (Supplementary Table 4).

Fig. 1

ROC curve of the LODDS, LNR and pN in prediction of prognosis of patients with adenocarcinoma at 1 (a), 3 (b), 5 (c), 7 (d), 10 (e) year point for cause-specific survival in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Prognostic abilities of LODDS, LNR, and pN classification in the SEER database

To comprehensively and reasonably compare the lymph node staging schemes, we grouped continuous variables of the LODDS and LNR schemes into three classification levels using the minimal P-value approach in the X-Tile software. LNR was classified into three group: 981 (46%) in LNR1 (≤0·02), 628 (30%) in LNR2 (0·02–0·47), and 519 (24%) in LNR3 (>0·47). A novel three-subgroup LODDS classification was determined using two LODDS cut-off points: 925 (43%) in LODDS1 (≤ − 1·89), 589 (28%) in LODDS2 (−1·89– − 0·51), and 614 (29%) in LODDS3 (> − 0·51) (Table 1).

As shown in Supplementary Fig. 7, increased LODDS or LNR values were associated with shorter survival times and higher mortalities. To compare the prognostic abilities of the three lymph node staging schemes, we plotted and compared the survival rates stratified by LODDS, LNR, and 7th and 8th pN classifications. The results (Fig. 2) showed that groups stratified by LODDS and LNR classifications had more discriminated survival rates than those stratified by the 7th and 8th pN classifications, which was consistent with the results from analyses using OS as outcome instead of CSS (Supplementary Fig. 8) and cumulative SBA death probability (Supplementary Fig. 9). The sensitivity analysis also supported the rationality of our LODDS and LNR classifications and the superiority of these classifications to those of the 7th and 8th pN based on Kaplan-Meier plots and comparison in another group of SBA patients diagnosed from 2011 to 2014 in the SEER database with similar results (Supplementary Fig. 10).

Fig. 2

Kaplan-Meier survival analysis according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN(d) for cause-specific survival, respectively in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

We next compared the prognostic performance of the lymph node schemes in classification patterns using similar measures; namely, R2 value for overall performance and Harrell's C statistic for discriminatory capacity. As shown in Table 2, as measured by Harrell's C statistic, the LODDS classification (0·656; 95%CI: 0·640–0·671) still showed a better discriminatory capacity than those of the LNR (0·644; 95%CI: 0·629–0·660) and 7th pN (0·626; 95%CI: 0·611–0·642), and 8th pN (0·626; 95%CI: 0·611–0·642) classification schemes. Analyses based on AUC measures (Table 2, Supplementary Table 1 and Fig. 3) also identified the superior discriminatory capacity of the LODDS classification (five-year: 72·30%; 95%CI: 70·13%–74·46%) over those of the LNR (five-year: 70·64%; 95%CI: 68·47%–72·82%), 7th pN (five-year: 69·21%; 95%CI: 66·96%–71·46%), or 8th pN (five-year: 69·51%; 95%CI: 67·25%–71·77%) classification schemes. When the patients with systemic metastases were excluded from the cohort (Supplementary Table 4 and Supplementary Fig. 11), we could also draw the conclusion that stratified LODDS classification had a better capacity of discrimination and higher accuracy of prognosis than the pN and LNR classification for a higher R2 (0·1584), Harrell's C (0·648; 95%CI: 0·632–0·665), and five-year AUC (70·75%, 95%CI: 68·44%-73·06%) than those of pN and LNR. Meanwhile, sensitivity analyses also identified LODDS classification as the best lymph node scheme using OS instead of CSS as the outcome (Supplementary Table 1, Supplementary Table 2 and Supplementary Fig. 12) or when the competing risk model was used in consideration of competing risk of non-SBA death (Supplementary Table 1, Supplementary Table 3 and Supplementary Fig. 13) .

Fig. 3

ROC curve of the LODDS classification, LNR classification, 7th pN stage and 8th pN stage in prediction of prognosis of patients with adenocarcinoma at 1 (a), 3 (b), 5 (c), 7 (d), 10 (e) year point for cause-specific survival in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; SEER: Surveillance, Epidemiology, and End Results.

The five-year survival rate for different groups decreased with increased LODDS classification for both CSS and OS (Table 3). For the 8th pN category, the five-year OS and CSS rates were 50·9% and 58·3% respectively, in N0 patients. The five-year OS and CSS rates decreased to 14·7% and 16·4%, respectively, in N2 patients. However, in the respective subgroup of pN category, the five-year CSS and OS rates were dramatically reduced with increasing LODDS classification. For instance, in N1 patients, the five-year OS rate was 48% in patients with LODDS1 and decreased to 11·4% in patients with LODDS3 disease.

Table 3

Five-Year Survival Rates for patients with small bowel adenocarcinoma stratified by LODDS and 8th pN category in SEER database.

Groups	Overall	LODDS classification			LNR classification
		LODDS1 (≤ − 1·89)	LODDS2 (−1·89– −0·51)	LODDS3 (> − 0·51)	LNR1 (≤0·02)	LNR2 (0·02–0·47)	LNR3 (>0·47)
N0 patients
OS (%)	50·9	55·6	31·6	NA	50·9	NA	NA
CSS (%)	58·3	62·0	42·1	NA	58·3	NA	NA
N1 patients
OS (%)	26·4	48·0	29·6	11·4	NA	43·2	25·4
CSS (%)	30·5	53·2	33·3	13·9	NA	47·8	29·1
N2 patients
OS (%)	14·7	NA	22·4	10·7	NA	20·0	9·8
CSS (%)	16·4	NA	24·6	12·1	NA	22·1	11·1

OS: overall survival; CSS: cause-specific survival; LODDS: log odds of positive lymph nodes; LNR: lymph node ratio; NA: Not Available.

The LODDS had a natural advantage in assessing patients without positive lymph node involvement. As shown in Fig. 4a, the LODDS scheme could reflect the prognosis of patients without positive lymph node involvement (P < 0·001).

Fig. 4

Kaplan-Meier survival analysis for cause-specific survival according to LODDS classification in patients with no lymph node involvement in SEER database (a) and the international multicenter cohort (b), respectively. LODDS: log odds of positive lymph nodes.

In multivariable analysis of models with lymph node classifications in cohorts with multivariate imputation used for missing variables, the LODDS classification model had an R2 value of 0·313, which was higher than those for the LNR (0·295) and 7th pN (0·284) and 8th pN (0·284) classifications (Table 4). In addition, the highest value of Harrell's C statistic observed in the LODDS classification model represented its superiority in discriminatory performance compared to those of the other classification schemes. Similar results were obtained in a sensitivity analysis in the group of patients without any missing clinical variables (Supplementary Table 5).

Table 4

Multivariable analysis for prognostic performance of models with different node classifications for small intestine adenocarcinoma in SEER database with multivariate imputationused for missed variable.

Factor	LODDS classification		LNR classification		7th pN classification		8th pN classification
	HR	P	HR	P	HR	P	HR	P
Node Stagea
LODDS1/LNR1/pN0	1·00	<0·001	1·00	<0·001	1·00	<0·001	1·00	<0·001
LODDS2/LNR2/pN1	1·82 (1·58–2·11)		1·62 (1·41–1·86)		1·82 (1·59–2·08)		1·76 (1·53–2·03)
LODDS3/LNR3/pN2	2·95 (2·55–3·40)		2·64 (2·28–3·05)		2·32 (1·99–2·71)		2·26 (1·95–2·60)
T categoryb
T1	1·00		1·00		1·00		1·00
T2	1·43 (0·82–2·51)	0.204	1·40 (0·80–2·46)	0.234	1·43 (0·82–2·50)	0.206	1·44 (0·82–2·52)	0.200
T3	2·22 (1·36–3·61)	0.002	2·06 (1·26–3·38)	0.005	2·05 (1·26–3·34)	0.005	2·05 (1·26–3·33)	0.005
T4	3·11 (1·91–5·06)	<0.001	2·89 (1·77–4·74)	<0.001	2·81 (1·74–4·54)	<0.001	2·81 (1·74–4·56)	<0.001
M stage
M0	1·00	<0·001	1·00	<0·001	1·00	<0·001	1·00	<0·001
M1	2·45 (2·14–2·81)		2·56 (2·23–2·93)		2·82 (2·47–3·26)		2·82 (2·46–3·22)
Age
≤ 60	1·00	<0·001	1·00	<0·001	1·00	<0·001	1·00	<0·001
>60	1·56 (1·39–1·76)		1·54 (1·37–1·73)		1·57 (1·40–1·76)		1·57 (1·39–1·76)
Marriage
Yes	1·00	<0·001	1·00	<0·001	1·00	<0·001	1·00	<0·001
No	1·22 (1·09–1·38)		1·23 (1·09–1·39)		1·25 (1·11–1·41)		1·25 (1·11–1·41)
Sex
Male	1·00	0·024	1·00	0·032	1·00	0·007	1·00	0·006
Female	0·87 (0·78–0·98)		0·88 (0·78–0·99)		0·85 (0·76–0·96)		0·85 (0·75–0·95)
Grade
I/II	1·00	<0·001	1·00	<0·001	1·00	<0·001	1·00	<0·001
III/IV	1·30 (1·16–1·46)		1·31 (1·17–1·47)		1·28 (1·14–1·44)		1·27 (1·14–1·43)
Site
Duodenum	1·00		1·00		1·00		1·00
Ileum	0·92 (0·78–1·07)	0·283	0·91 (0·78–1·07)	0·244	0·87 (0·74–1·01)	0·075	0·87 (0·75–1·02)	0·087
Jejunum	0·76 (0·66–0·88)	<0·001	0·78 (0·68–0·91)	0·001	0·78 (0·67–0·90)	0·009	0·77 (0·67–0·89)	0·009
Othersc	0·99 (0·66–1·50)	0·963	1·05 (0·69–1·59)	0·832	1·04 (0·67–1·61)	0·877	1·04 (0·67–1·61)	0·870
Model performanced
SEER database
Mean R2	0·309		0·291		0·280		0·280
Mean with range of Harrell's C	0·730 (0·708–0·742)		0·723 (0·701–0·745)		0·717 (0·695–0·739)		0·717 (0·695–0·739)
Multicenter cohort
R2	0·191		0·155		0·142		0·140
Harrell's C	0·663 (0·602–0·724)		0·654 (0·593–0·715)		0·642 (0·581–0·703)		0·649 (0·588–0·710)

LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes.

Represent the 7thpN, 8thpN, LODDS classifications and LNR classifications for corresponding multivariable models.

Tumor category were graded according to the 8thtumour node metastasis (TNM) classification of malignant tumors proposed by the American Joint Committee on Cancer/International Union Against Cancer (AJCC/UICC), whereas M stage was recorded metastatic disease pathologically coded 85 in extent of disease “EOD 10-extent (1988–2003)” of Surveillance, Epidemiology, and End Results (SEER) data.

Others were recorded because tumors were located in the Meckels diverticulum or overlapping lesion of small intestine.

Indicates the joint prognostic performance of models with different node staging schemes.

Validation of the prognostic abilities of the LODDS, LNR, and pN in a multicentre cohort

In the international multicentre cohort, the LODDS had better a R2 value (0·131) than those of the LNR (0·110) and pN (0·049) and the LODDS (0·647; 95%CI: 0·587–0·707) also showed a better discriminatory capacity than those of the pN (0·599; 95%CI: 0·545–0·653) and LNR (0·624; 95%CI: 0·570–0·678) (Table 2).

In addition to the continuous pattern, the LODDS classification had higher R2 (0·120) and better discriminatory capacity (0·629; 95%CI: 0·574–0·683) than those of the LNR and 7th and 8th pN classifications in the multicentre cohort (Table 2). Analyses based on AUC measures (Table 2, Supplementary Table 1, Supplementary Table 2 and Supplementary Table 3) showed that the LODDS and LODDS classification provided a better discriminatory capacity than those of the pN and LNR for CSS, OS, and competing risk model in the multicentre cohort. As shown in Supplementary Fig. 14, there appeared to be a positive correlation between LODDS or LNR and survival time. However, the correlation between LNR and survival was weak at the poles.

Fig. 5 shows that groups stratified by LODDS and LNR classifications had better discrimination than those stratified by the 7th pN and 8th pN classification schemes for CSS, consistent with the results of OS (Supplementary Fig. 15) and cumulative SBA death probability (Supplementary Fig. 16) in the multicentre cohort. For patients without positive lymph node involvement, those with LODDS2 disease had a worse outcome than those with LODDS1, a finding consistent with that in the SEER database (Fig. 4b). In multivariable analysis, the model of LODDS classification also had higher R2 and Harrell's C values than those of other classifications in the multicenter cohort (Table 4).

Fig. 5

Kaplan-Meier survival analysis according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN (d) for cause-specific survival in the international multicentre cohort, respectively. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes.

Discussion

To our knowledge, this was the largest study involving two different cohorts to compare the predictive ability of three lymph node staging schemes in the survival of patients with SBA. Our results showed that the LODDS had better prognostic evaluation and discriminatory capacities in SBA than those of the pN and LNR staging schemes for both continuous and stratified patterns. Furthermore, multivariable analysis revealed that the LODDS classification also had better predictive accuracy and discriminatory capacity than those of the 7th and 8th editions of the TNM node and LNR classification schemes. The prognostic superiority of the LODDS and LODDS classification was also validated in an international multicenter cohort using the same inclusion and exclusion criteria.

Lymph node status is a vital prognostic indicator of SBA [11,24,25]. The AJCC TNM staging currently dominates lymph node staging systems; however, the reliability of the pN staging scheme has recently been questioned. Overman et al. assessed total lymph nodes, positive lymph nodes, and LNR in 1991 patients with SBA. They found that survival after surgical resection for stage I, II, and III SBA was related to the total number of lymph nodes retrieved. In addition, among patients with stage III SBA, the LNR had prognostic performance than stratification by the number of positive lymph nodes [26]. Tran et al. evaluated the number of lymph nodes retrieved and the impact of LNR on SBA survival in the SEER database. They concluded that the total number of lymph nodes retrieved and LNR were important prognostic factors of survival in SBA and recommended that the current lymph node staging system should be updated [11].

In the present study, a median of around 6–8 lymph nodes was retrieved, with more than half of the patients with SBA lacking an adequate number of examined lymph nodes, indicating the scarcity of lymph node harvest and poor overall surgical quality [27]. SBAs are often diagnosed due to local complications such as obstruction or bleeding, leading to surgery for symptom relief but not timely recognition of oncologic resection with more lymph nodes retrieved for N staging [28,29]. This may cause poor surgery quality for SBA and a need for increased awareness. Thus, surgeons should assess lymph nodes as much as possible and surgery quality should be strongly advocated to increase the number of retrieved lymph nodes for accurate N staging. However, if surgery quality cannot be guaranteed with inadequate lymph nodes harvested, LODDS and LNR may be surrogates for pN and the LODDS scheme has better discrimination than that of LNR.

Studies have shown a positive association between an increased number of lymph nodes retrieved and a better prognosis [10,26,30,31]. To ensure the quality of surgery measured by the number of lymph nodes retrieved [27], the proposed cutoffs for the optimal number of lymph nodes varied in different studies, ranging from 8 to 15 [26,30,31]. In our previous study, we determined the optimal number of examined lymph nodes to be at least 17. In the multivariate Cox regression model, a cutoff of 17 as an adequate number of examined lymph nodes was an independent factor for better survival [10].

Since an adequate number of examined lymph nodes was associated with a better survival and was used to assess the quality of surgery, we next tried to include the adequate number of examined lymph nodes and other factors to build a predictive model for the survival of SBA patients using backward Cox analysis using Akaike information criteria (AIC) selection criteria [10]. However, the factor was excluded from the predictive model since the model that included the factor did not have the lowest AIC. The number of examined lymph nodes should be considered since nearly half of cancers did not meet this criterion. Hence, we assessed the performance of the three lymph node schemes most commonly used to determine prognosis, namely, the pN, LNR, and LODDS schemes.

It stands to reason that the LODDS is better than the pN and LNR staging schemes. The pN staging scheme only records the absolute number of positive lymph nodes without considering the numbers of total and negative lymph nodes retrieved. Visually, both LODDS and LNR are more rational than the pN staging scheme for consideration of the number of positive or negative and total lymph nodes. Some researchers have questioned the prognostic accuracy of LNR [32,33], especially when all or no retrieved lymph nodes are metastatic.

As shown in Supplementary Fig. 7 and Supplementary Fig. 14, patients with the same LNR may have different outcomes. For example, the LNR may be the same for patients with one metastatic node out of one retrieved as for forty metastatic nodes out of forty retrieved, which is a fatal drawback of the LNR scheme. In contrast, the LODDS can further stratify patients with no (Fig. 4) or all metastatic nodes retrieved. Close analysis of the data in Table 3 revealed that some patients with N0 disease had worse survival than those with N1 and that some patients with N1 disease had a poorer prognosis than those with N2, which could be shown by LODDS instead of LNR classification. According to the definition of LNR, almost all of the N0 patients belonged to the subgroup of LNR1 and most of the N1 and N2 patients were LNR2 and LNR3. In any LNR subgroup, the five-year OS and CSS rates decreased with increasing pN category, in contrast to the LODDS classification, which indicated the flaws of pN category in evaluating OS and CSS of patients with SBA. The LODDS classification had greater power of discrimination compared to that of the LNR scheme.

We found that the OS and CSS of duodenal carcinoma were worse than those of jejunoileal tumors when we assessed the five-year survival rate and number of retrieved lymph nodes from different locations (Supplementary Table 6), consistent with the findings of a previous study [26]. Comparison of the number of retrieved lymph nodes between the duodenum and jejunum or between the ileum and jejunum revealed that the OS and CSS improved with more lymph nodes being assessed. However, the finding was contrary to the survival rate of carcinoma located in the duodenum and ileum; one explanation for this phenomenon may be that the number of examined lymph nodes is not the only factor affecting the outcome and other risk factors such as different pathogenesis could also contribute to the outcome [26].

This study has some limitations. First, the retrospective study design had an inherent bias. The predictive performance of lymph node staging schemes cannot be completely illuminated without a prospective and randomized clinical trial. The SEER database lacked information on adjuvant therapy, genetic status, etc. [2,34,35], which may be confounding factors for prognosis. However, the sample capacity of this study, which we believed to be sufficient, and the long duration of follow up could make up for these drawbacks and provide a comprehensive view of the prognostic ability of three common lymph node staging schemes in SBA. We found a superior prognostic ability of the LODDS in patients with SBA in the SEER database and uniquely verified our finding in a cohort from Europe and Asia. The approaches were methodologically sound and the results were statistically compelling and convincing.

Conclusion

In conclusion, LODDS scheme showed a better prognostic ability than pN or LNR scheme in patients with SBA. LODDS could serve as a significant auxiliary of lymph node staging system in the future revisions of the AJCC manual and surgical guidelines to improve prediction for prognosis of SBA.

The following are the supplementary data related to this article

Supplementary Fig. 1

Flowchart of patient selection in SEER database for this study. SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 2

Kaplan-Meier survival analysis of patients with small intestine adenocarcinoma according to 8th TNM staging in SEER database. SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 3

Distribution characteristics of association between LODDS and LNR (a), LODDS and pN (b), LNR and pN (c) in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 4

Log hazard ratio as a function of LODDS (a), LNR (b), and pN (c) for all 2128 patients and log hazard ratio as a function of LODDS individually for patients with no lymph nodes involved (n = 980) (d) in SEER database. Ninety-five percent confidence intervals are indicated. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 5

ROC curve of the LODDS, LNR and pN in prediction of prognosis of patients with adenocarcinoma at 1 (a), 3 (b), 5 (c), 7 (d), 10 (e) year point for overall survival in SEER database.LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 6

ROC curve of the LODDS, LNR and pN in prediction of prognosis of patients with adenocarcinoma at 1 (a), 3 (b), 5 (c), 7 d), 10 (e) year point for competing risk model in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 7

Distribution of survival time, survival status and lymph node staging scheme value among different lymph node staging groups according to the LODDS classification (a) and LNR classification (b) in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 8

Kaplan-Meier survival analysis according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN (d) for overall survival in SEER database, respectively. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes. SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 9

Cumulative probability of death due to small intestine adenocarcinoma according to according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN (d) in SEER database, respectively. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 10

Kaplan-Meier survival analysis according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN (d) for cause-specific survival in patients with small bowel adenocarcinoma diagnosed from 2011 to 2014 in SEER database, respectively. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 11

Kaplan-Meier survival analysis according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN (d) for cause-specific survival after eliminating patients with systemic metastases in SEER database, respectively. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes.

Supplementary Fig. 12

ROC curve of the LODDS classification, LNR classification, 7th pN and 8th pN stage in prediction of prognosis of patients with adenocarcinoma at 1 (a), 3 (b), 5 (c), 7 (d), 10 (e) year point for overall survival in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 13

ROC curve of the LODDS classification, LNR classification, 7th pN stage and 8th pN stage in prediction of prognosis of patients with adenocarcinoma at 1 (a), 3 (b), 5 (c), 7 (d), 10 (e) year point for competing risk model in SEER database. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; SEER: Surveillance, Epidemiology, and End Results.

Supplementary Fig. 14

Distribution of survival time, survival status and lymph node staging scheme value among different lymph node staging groups according to the LODDS classification (a) and LNR classification (b) in the multicentre cohort. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes.

Supplementary Fig. 15

Kaplan-Meier survival analysis according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN (d) for overall survival in the international multicenter cohort, respectively.LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes.

Supplementary Fig. 16

Cumulative probability of death due to small intestine adenocarcinoma according to according to LODDS classification (a) and LNR classification (b), 7th pN (c) and 8th pN (d) in the international multicenter cohort, respectively. LNR: lymph node ratio; LODDS: log odds of positive lymph nodes; pN: number of positive nodes.

Supplementary Table 1

Areas under the receiver operating characteristic curve for survival of patients with small intestine adenocarcinoma according to different sensitivity analyses.

Supplementary Table 2

Univariable analysis for overall prognostic performance of node staging schemes for small intestine adenocarcinoma for overall survival.

Supplementary Table 3

Univariable analysis for overall prognostic performance of node staging schemes for small intestine adenocarcinoma for competing risk model.

Supplementary Table 4

Univariable analysis for overall prognostic performance of node staging schemes for small intestine adenocarcinoma without systemic metastases for cause-specific survival.

Supplementary Table 5

Multivariable analysis for prognostic performance of models with different node classifications for small intestine adenocarcinoma in patients with all clinical variable available.

Supplementary Table 6

Five-year survival rate and number of retrieved lymph nodes for small bowel adenocarcinoma according to different tumor sites.