Prognostic significance of a five-factor modified frailty index in patients with gastric cancer undergoing curative-intent resection: A cohort study

Abstract

The 5-factor modified frailty index (mFI-5) evaluates frailty based on variables including functional status, diabetes, chronic obstructive pulmonary disease, congestive heart failure, and hypertension requiring medication. Despite its effectiveness in predicting surgical risk, the potential of mFI-5 as a predictor of long-term survival in patients with gastric cancer (GC) has not been investigated. This study aims to assess the prognostic significance of mFI-5 in patients with GC who have undergone curative-intent gastric resection. Among the 494 patients diagnosed with stage I to III GC, multivariate analysis revealed that age, tumor–node–metastasis (TNM) stage, geriatric nutritional risk index, mFI-5, and the type of gastrectomy were significant predictors for both overall survival (OS) and disease-free survival (DFS). We assessed 3 models: Baseline model (BM, TNM stage only), interim model (IM, all significant variables except mFI-5), and full model (FM, all significant variables including mFI-5). FM outperformed BM for OS (C-index 0.818 vs 0.683; P < .001) and DFS (C-index 0.805 vs 0.687; P < .001). Similarly, IM outperformed BM for OS (C-index 0.811 vs 0.683; P < .001) and DFS (C-index 0.797 vs 0.687; P < .001). Multiple metrics consistently supported the improved discriminative capacity of FM and IM compared to BM. However, while FM exhibits enhanced predictive capacity over IM, this improvement lacks statistical significance across key metrics. In conclusion, our study highlights the clinical significance of the mFI-5, along with age, TNM stage, geriatric nutritional risk index, and type of gastrectomy, as valuable predictors of long-term survival in GC patients. The FM consistently demonstrates enhanced predictive accuracy compared to the BM. However, it is important to note that while the FM improves predictive power over the IM, this enhancement does not achieve statistical significance across multiple metrics. These findings collectively emphasize the potential clinical value of the FM as a robust tool for surgeons in predicting long-term survival outcomes before surgery in patients with GC.

1. Introduction

Gastrectomy is the standard treatment for stage I to III gastric cancer (GC); however, the risks of relapse and death remain substantial. To improve survival outcomes, biomarkers that can accurately predict prognosis before gastrectomy need to be established.

Although the tumor–node–metastasis (TNM) staging system is commonly used to predict prognosis in patients with cancer, it has limitations, such as different prognoses for patients with the same TNM stage.^[1,2] Inflammatory markers, including the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), absolute monocyte and lymphocyte count prognostic score, and lymphocyte-to-monocyte ratio (LMR), have been identified as determinants of survival outcomes.^[3–5] However, the optimal cutoff points for these markers have not been universally agreed upon, limiting their clinical utility. Recently, the geriatric nutritional risk index (GNRI) has emerged as a potential determinant of survival outcomes; however, further external validation is needed.^[6] Furthermore, the identification of minimal residual disease through the measurement of circulating tumor DNA (ctDNA) levels has the potential to aid personalized adjuvant therapy and enhance survival outcomes for patients with GC. Nevertheless, the use of ctDNA in clinical settings for GC remains restricted and requires further investigation.^[7] Considering the limitations of existing biomarkers, additional research is essential to develop novel and reliable biomarkers that are both simple and accurate in predicting survival outcomes for patients with GC.

Frailty is a clinical syndrome that describes decreased reserves and functional capacity across multiple physiological systems, making individuals more vulnerable to stressors and adverse health outcomes.^[8,9] Panayi et al^[10] found that frail patients who underwent surgery for various causes were more likely to experience complications and readmission, with a 4.19 times higher risk of mortality. Ethun et al^[9] also reported that frail patients with malignant tumors were at increased risk of postoperative complications, chemotherapy intolerance, disease progression, and death. Similarly, Lee et al^[8] showed that frailty following gastrectomy for GC resulted in higher mortality, longer hospital stays, and increased costs. Therefore, identifying frailty in cancer patients is crucial as it helps to recognize individuals at high risk of adverse health outcomes, such as hospitalization, functional decline, disability, and death, irrespective of age.^[10]

Frailty is often associated with aging, but it can also occur in younger patients with multiple chronic health conditions, physical limitations, and/or cognitive impairments, creating discrepancies between chronological and physiological age.^[10,11]

Various methods, such as the Rockwood frailty index, Fried frailty criteria, frailty screening tools, and geriatric assessment, are available to recognize frailty in older adults with cancer, with the Rockwood frailty index being recognized as one of the gold standards.^[12,13] The Rockwood frailty index is based on a cumulative deficit model and includes 70 items, proposing that the accumulation of medical, social, and functional deficits over a person’s lifetime leads to nonspecific, age-associated vulnerability or frailty.^[9] Obeid et al^[14] developed an 11-factor modified Frailty Index (mFI-11) that matches 11 comorbidity and deficit variables to the 70 variables from the original Frailty Index.

A simplified version of the mFI-11, called the 5-factor modified frailty index (mFI-5), includes only 5 components, namely functional status, diabetes, chronic obstructive pulmonary disease (COPD), congestive heart failure (CHF), and hypertension requiring medication.^[15,16] By focusing on the most important factors, mFI-5 may be equally or more effective in predicting surgical risk in certain patients.^[16] Although most studies on mFI-5 are available in the orthopedic field, several studies have evaluated its predictive value for survival outcomes in patients with malignancies and found that it appears to be a useful tool for predicting survival outcomes and postoperative complications.^[17–19] In patients with GC, mFI-5 is a determinant of adverse outcomes, such as hypoproteinemia following surgical resection.^[20] However, there are no reports on the clinical value of mFI-5 as a predictor of long-term survival outcomes in patients with GC.

Therefore, the present study aimed to evaluate the clinical significance of mFI-5 as a determinant of overall survival (OS) and disease-free survival (DFS) in patients of all ages with stage I to III GC.

2. Methods

2.1. Patients

This retrospective study analyzed patients who underwent curative-intent gastrectomy at Kyung Hee University Hospital at Gangdong between June 2006 and July 2018. This study included patients with primary GC (based on the Lauren classification of gastric carcinoma),^[21] stage I to III GC (based on the 8th edition of the American Joint Committee on Cancer staging system for GC),^[22] and microscopically negative resection margins. Patients with concurrent malignancies or malignancies within the past 5 years, prior anticancer treatment, active infection, or autoimmune diseases were excluded.

The Institutional Review Board of Kyung Hee University Hospital at Gangdong approved the study (2023-04-030) and the requirement for informed consent was waived because of the retrospective design of the study.

2.2. Baseline clinical characteristics

Clinicopathological parameters included age, sex, body mass index (BMI), tumor site, tumor size, extent of the primary tumor (T stage), presence or absence of cancer cells in nearby lymph nodes (N stage), TNM stage, lymphatic invasion (LI), vascular invasion, perineural invasion (PI), histological classification,^[21] NLR,^[4] LMR,^[4] GNRI,^[6] American Society of Anesthesiologists physical status (ASA-PS) classification,^[23] mFI-5, and type of gastrectomy (TOG). Age and BMI were dichotomized using standard criteria. Tumor size, NLR, LMR, and GNRI were analyzed as continuous variables, given the absence of consensus regarding cutoff values for these variables.

2.3. Measurement of mFI-5

The mFI-5 includes 5 components: functional status, diabetes, history of COPD, history of CHF, and hypertension requiring medication. The mFI-5 assesses the functional status using a single item related to the patient’s ability to perform activities of daily living (ADLs), including bathing, feeding, toilet use, mobility, and dressing. The ADL component was scored as 1 if the patient required assistance with ADLs and 0 otherwise. For diabetes, a score of 1 was assigned if the patient had been diagnosed with the condition, irrespective of its control status, and 0 if the patient did not have diabetes. Similarly, for a history of COPD and CHF, a score of 1 was given if the patient had a history of the condition, regardless of their current status, and 0 if they did not. For hypertension, a score of 1 was given if the patient had a history of hypertension that required medication, and 0 if they did not have hypertension or did not require medication. The scores for each component were summed to obtain a total score ranging from 0 to 5.^[15,16]

2.4. Statistical analysis

OS was calculated as the interval between the date of gastrectomy and the date of death from any cause, whereas DFS was measured as the time between the date of gastrectomy and the date of either disease recurrence or death from any cause, whichever came first.

Continuous variables were presented as medians with interquartile ranges (IQRs) in parentheses. The Kruskal–Wallis test or chi-square test was used for between-group comparisons of variables, and the Bonferroni method was used for multiple comparisons.

Kaplan–Meier curves were used to analyze OS and DFS based on the mFI-5. Cox regression analysis (CR) was used to determine the hazard ratios (HRs) for variables including age, sex, BMI, tumor size, T stage, N stage, TNM stage, LI, vascular invasion, PI, histology, NLR, LMR, GNRI, ASA-PS classification, mFI-5, and TOG.

Univariate CR was initially performed, and variables with a significance level of P < .05 were subsequently included in the multivariate CR. To assess multicollinearity, variance inflation factor (VIF) was employed. Using all significant variables, full model (FM) was constructed for predicting both OS and DFS. Nomograms for OS and DFS prediction based on FM were developed. To assess the discriminative ability of FM, we created calibration curves employing 1000 bootstrap samples to mitigate overfitting.

We assessed 3 models: Baseline model (BM, TNM stage only), interim model (IM, all significant variables except mFI-5), and FM. The discriminative capacity of these models was evaluated using the concordance index (C-index). Furthermore, we calculated the C-index for OS and DFS over a decade using bootstrap cross-validation with 1000 resamples, replacing the original datasets.

Additionally, we conducted area under the curve (AUC) comparisons to evaluate OS and DFS risk between the models at 36 and 60 months post-surgery, employing the Delong method. Furthermore, continuous net reclassification index analysis (cNRI) and integrated discrimination improvement (IDI) analyses were carried out to compare OS and DFS risk between the models at 36 and 60 months after surgery using bootstrap with 1000 resamples. To further assess the predictive performance of the models in estimating OS and DFS risk at 36 and 60 months after surgery, decision curve analysis (DCA) was conducted across a range of threshold probabilities from 0% to 40% using 1000 bootstrap samples.

All statistical analyses used 2-sided p-values, with statistical significance defined as P < .05. The R package (r-project.org) was utilized for the statistical computations.

3. Results

3.1. Patients clinical characteristics

Among the initial 501 patients evaluated for eligibility, 3 individuals with concurrent malignant tumors, 2 with microscopic residual disease, and 2 with stage IV disease were excluded. Consequently, this study encompassed 494 patients with GC, with 391 (79.1%) undergoing partial gastrectomy and 103 (20.9%) undergoing total gastrectomy. The majority of patients were of Asian ethnicity (96.2%, n = 475), with a minority identified as Caucasian (3.8%, n = 19). Patient age exhibited a median of 60.5 years (IQR, 52.0–70.0 years), and the median tumor size was 3.0 cm. The distribution of cancer stages showed that the largest group had stage I cancer (61.5%, n = 304), followed by 18.4% (n = 91) with stage II cancer, and 20.0% (n = 99) with stage III cancer (Table 1).

Table 1.

Patients characteristics stratified by the five-factor modified frail index.

Variables	Median (IQR) or n (%) according to mFI-5 score						P value
	Overall	0 (n = 200)	1 (n = 182)	2 (n = 80)	3 (n = 26)	4 (n = 6)
Age	60.5 (52.0–70.0)	55.0 (47.5–65.0)	62.0 (53.0–70.0)	67.0 (59.5–75.0)	68.0 (63.0–74.0)	76.0 (67.0–82.0)	<.001
Sex							.324
Male	332 (67.2%)	124 (62.0%)	128 (70.3%)	56 (70.0%)	20 (76.9%)	4 (66.7%)
Female	162 (32.8%)	76 (38.0%)	54 (29.7%)	24 (30.0%)	6 (23.1%)	2 (33.3%)
BMI	23.7 (21.4–26.0)	23.2 (21.3–25.4)	23.8 (21.3–26.2)	24.0 (21.9–26.1)	25.4 (22.4–7.7)	24.6 (22.1–30.2)	.168
Tumor site							.889
Upper	49 (9.9%)	16 (8.0%)	24 (13.2%)	5 (6.2%)	3 (11.5%)	1 (16.7%)
Middle	169 (34.2%)	66 (33.0%)	65 (35.7%)	28 (35.0%)	8 (30.8%)	2 (33.3%)
Lower	268 (54.3%)	114 (57.0%)	90 (49.5%)	46 (57.6%)	15 (57.7%)	3 (50.0%)
Diffuse	8 (1.6%)	4 (2.0%)	3 (1.6%)	1 (1.2%)	0 (0.0%)	0 (0.0%)
Tumor size	3.0 (2.0–5.5)	3.0 (1.9–5.5)	3.0 (2.0–5.0)	3.2 (2.2–5.0)	4.6 (3.0–6.6)	5.2 (1.1–6.5)	.166
T stage							.624
1–2	336 (68.0%)	134 (67.0%)	131 (72.0%)	50 (62.5%)	17 (65.4%)	4 (66.7%)
3–4	158 (32.0%)	66 (33.0%)	51 (28.0%)	30 (37.5%)	9 (34.6%)	2 (33.3%)
N stage							.343
0	322 (65.2%)	131 (65.5%)	125 (68.7%)	50 (62.5%)	13 (50.0%)	3 (50.0%)
1–3	172 (34.8%)	69 (34.5%)	57 (31.3%)	30 (37.5%)	13 (50.0%)	3 (50.0%)
TNM stage							.968
I–II	395 (80.0%)	159 (79.5%)	145 (79.7%)	66 (82.5%)	20 (76.9%)	5 (83.3%)
III	99 (20.0%)	41 (20.5%)	37 (20.3%)	14 (17.5%)	6 (23.1%)	1 (16.7%)
Histology							.634
Intestinal	243 (49.2%)	92 (46.0%)	92 (50.5%)	42 (52.6%)	13 (50.0%)	4 (66.6%)
Diffuse	118 (23.9%)	57 (28.5%)	41 (22.6%)	13 (16.2%)	6 (23.1%)	1 (16.7%)
Mixed	111 (22.5%)	45 (22.5%)	38 (20.9%)	20 (25.0%)	7 (26.9%)	1 (16.7%)
Unknown	22 (4.4%)	6 (3.0%)	11 (6.0%)	5 (6.2%)	0 (0.0%)	0 (0.0%)
LI							.160
No	327 (66.2%)	140 (70.0%)	123 (67.6%)	48 (60.0%)	13 (50.0%)	3 (50.0%)
Yes	167 (33.8%)	60 (30.0%)	59 (32.4%)	32 (40.0%)	13 (50.0%)	3 (50.0%)
VI							.111
No	468 (94.7%)	193 (96.5%)	174 (95.6%)	72 (90.0%)	23 (88.5%)	6 (100.0%)
Yes	26 (5.3%)	7 (3.5%)	8 (4.4%)	8 (10.0%)	3 (11.5%)	0 (0.0%)
PI							.944
No	444 (89.9%)	180 (90.0%)	163 (89.6%)	72 (90.0%)	23 (88.5%)	6 (100.0%)
Yes	50 (10.1%)	20 (10.0%)	19 (10.4%)	8 (10.0%)	3 (11.5%)	0 (0.0%)
NLR	1.9 (1.4–2.7)	1.8 (1.4–2.4)	2.0 (1.4–2.7)	2.1 (1.5–2.8)	2.2 (1.8–2.8)	2.3 (1.2–3.2)	.065
LMR	4.3 (3.2–5.4)	4.6 (3.5–5.6)	4.1 (3.2–5.2)	3.7 (2.7–5.4)	3.7 (3.0–4.6)	4.1 (2.9–5.4)	.009
GNRI	102.8 (98.3–105.7)	102.8 (99.6–105.7)	102.8 (98.3–105.7)	101.1 (95.5–104.2)	101.3 (93.8–104.2)	94.6 (79.0–101.3)	.004
TOG							.519
Patial	391 (79.1%)	152 (76.0%)	150 (82.4%)	65 (81.2%)	20 (76.9%)	4 (66.7%)
Total	103 (20.9%)	48 (24.0%)	32 (17.6%)	15 (18.8%)	6 (23.1%)	2 (33.3%)
ASA-PS							<.001
1–2	438 (88.7%)	192 (96.0%)	160 (87.9%)	65 (81.2%)	19 (73.1%)	2 (33.3%)
3	56 (11.3%)	8 (4.0%)	22 (12.1%)	15 (18.8%)	7 (26.9%)	4 (66.7%)

ASA-PS = American Society of Anesthesiologists physical status, BMI = body mass index, GNRI = geriatric nutritional risk index, IQR = interquartile range, LI = lymphatic invasion, LMR = lymphocyte-to-monocyte ratio, mFI-5 = five-factor modified frail index, N stage = presence or absence of cancer cells in nearby lymph nodes, NLR = neutrophil-to-lymphocyte ratio, PI = perineural invasion, T stage = extent of the primary tumor, TNM = tumor–node–metastasis, TOG = type of gastrectomy, VI = vascular invasion.

3.2. Five-factor modified frail index and its association with clinical variables

The mFI-5 was constructed using the following 5 variables: functional status, CHF, COPD, DM, and hypertension with treatment. Of these, 28.5% (n = 141) had a functional status score of 1, whereas only 3.2% (n = 16) had a CHF score of 1. Additionally, 8.9% (n = 44), 15.6% (n = 77), and 33.6% (n = 166) had scores of 1 for COPD, DM, and hypertension with treatment, respectively. The scores for each variable were summed to obtain the mFI-5 score, which range from 0 to 5. In total, 200 patients had an mFI-5 score of 0, 182 had a score of 1, 80 had a score of 2, 26 had a score of 3, and 6 had a score of 4. None of the patients had an mFI-5 score of 5.

The median age varied significantly based on the mFI-5 score, with median ages of 55, 62, 67, 68, and 76 years for scores of 0, 1, 2, 3, and 4, respectively (P < .001). Furthermore, there was a significant association between mFI-5 and ASA-PS classification (P < .001), LMR (P = .009), and GNRI (P = .004). However, no significant correlations were found between the mFI-5 and sex, BMI, tumor site, tumor size, T stage, N stage, TNM stage, histology, LI, vascular invasion, PI, NLR, or TOG (Table 1).

3.3. Kaplan–Meier curve analysis of survival outcomes stratified by 5-factor modified frail index

The median follow-up time was 64.7 months (IQR, 21.2–94.9 months). The Kaplan–Meier curve analysis revealed that the 5-year estimated OS rates for mFI-5 scores were 84.4% for a score of 0, 81.2% for a score of 1, 72.4% for a score of 2, 69.7% for a score of 3, and 41.7% for a score of 4. There was a significant difference in OS based on the mFI-5 (P = .003) (Fig. 1A).

Figure 1.

Kaplan–Meier curve analysis of survival outcomes stratified by the five-factor modified frail index. (A) Overall survival, (B) Disease-free survival.

Similarly, the estimated 5-year DFS rates by mFI-5 scores were 83.1% for a score of 0, 77.6% for a score of 1, 68.0% for a score of 2, 70.3% for a score of 3, and 33.3% for a score of 4. There was a significant difference in DFS based on the mFI-5 (P = .001) (Fig. 1B).

3.4. CR of the variables for overall and disease-free survival

Regarding OS, univariate CR showed that age, BMI, tumor size, T stage, N stage, TNM stage, vascular invasion, NLR, LMR, GNRI, ASA-PS classification, mFI-5, and TOG, were significant variables. Multivariate CR revealed that age (HR, 1.99; P = .002), TNM stage (HR, 3.15; P < .001), GNRI (HR, 0.94; P < .001), mFI-5 (HR, 1.28; P = .012), and TOG (HR, 1.62; P = .038) remained significant variables after adjustment for other variables, and the VIFs were 1.13, 1.10, 1.09, 1.15, and 1.02, respectively (Table 2).

Table 2.

Univariate and multivariate Cox regression analyses of factors affecting overall and disease-free survival.

Covariates	OS		DFS
	HR (95% CI)	P value	HR (95% CI)	P value
Univariate analysis
Age, yr (≥65 vs <65)	2.71 (1.80–4.07)	<.001	2.42 (1.65–3.54)	<.001
Sex (female vs male)	0.78 (0.50–1.22)	.276	0.70 (0.46–1.07)	.099
BMI, kg/m2 (<18.5 vs ≥18.5)	2.21 (1.18–4.15)	.013	1.94 (1.04–3.62)	.038
Tumor size, cm*	1.19 (1.14–1.24)	<.001	1.19 (1.14–1.23)	<.001
T stage (3–4 vs 1–2)	5.06 (3.32–7.70)	<.001	4.81 (3.25–7.12)	<.001
N stage (1–3 vs 0)	3.78 (2.51–5.70)	<.001	4.14 (2.80–6.12)	<.001
TNM stage (III vs I–II)	5.55 (3.71–8.29)	<.001	5.67 (3.88–8.29)	<.001
Histology (intestinal vs others)	0.88 (0.59–1.31)	.536	0.93 (0.64–1.36)	.721
Lymphatic invasion (yes vs no)	3.49 (2.32–5.21)	<.001	3.27 (2.24–4.78)	<.001
Vascular invasion (yes vs no)	3.21 (1.67–6.19)	<.001	3.59 (1.97–6.57)	<.001
Perineural invasion (yes vs no)	2.55 (1.49–4.37)	<.001	2.37 (1.41–3.98)	.001
NLR*	1.17 (1.10–1.25)	<.001	1.17 (1.10–1.24)	<.001
LMR*	0.80 (0.70–0.92)	.002	0.76 (0.67–0.87)	<.001
GNRI*	0.91 (0.89–0.92)	<.001	0.90 (0.89–0.92)	<.001
ASA-PS†	1.98 (1.28–3.06)	.002	2.12 (1.41–3.18)	<.001
mFI-5†	1.45 (1.20–1.76)	<.001	1.45 (1.21–1.73)	<.001
TOG (total vs partial)	2.65 (1.75–4.00)	<.001	2.42 (1.63–3.59)	<.001
Multivariate analysis
Age, years (≥65 vs <65)	1.99 (1.29–3.07)	.002	1.61 (1.07–2.42)	.022
TNM stage (III vs I–II)	3.15 (1.93–5.14)	<.001	3.35 (2.12–5.31)	<.001
GNRI*	0.94 (0.92–0.96)	<.001	0.94 (0.92–0.96)	<.001
mFI-5†	1.28 (1.06–1.55)	.012	1.27 (1.06–1.53)	.009
TOG (total vs partial)	1.62 (1.03–2.56)	.038	1.57 (1.01–2.42)	.044

ASA-PS = American Society of Anesthesiologists physical status, BMI = body mass index, CI = confidence interval, DFS = disease-free survival, GNRI = geriatric nutritional risk index, HR = hazard ratio, LMR = lymphocyte-to-monocyte ratio, mFI-5 = five-factor modified frail index, N stage = presence or absence of cancer cells in nearby lymph nodes, NLR = neutrophil-to-lymphocyte ratio, OS = overall survival, T stage = extent of the primary tumor, TNM = tumor–node–metastasis, TOG = type of gastrectomy.

^*Continuous variable.

^†Ordinal variable.

Univariate CR showed that age, BMI, tumor size, T stage, N stage, TNM stage, vascular invasion, PI, NLR, LMR, GNRI, ASA-PS classification, mFI-5, and TOG, were significant determinants of DFS. Multivariate CR revealed that age (HR, 1.61; P = .022), TNM stage (HR, 3.35; P < .001), GNRI (HR, 0.94; P < .001), mFI-5 (HR, 1.27; P = .009), and TOG (HR, 1.57; P = .044), remained significant determinants of DFS after adjusting for other variables, and the VIFs were 1.14, 1.09, 1.16, 1.17, and 1.01, respectively (Table 2).

3.5. Establishment and validation of prognostic models for overall and disease-free survival

In this study, we compared 3 models: the BM, which included only TNM stage, the IM, consisting of all significant variables except mFI-5, and the FM, which incorporated all significant variables including mFI-5.

FM demonstrated significantly higher C-index values than BM for both OS (0.818 vs 0.683, P < .001) and DFS (0.805 vs 0.687, P < .001). Similarly, IM exhibited significantly higher C-index values than BM for both OS (0.811 vs 0.683, P < .001) and DFS (0.797 vs 0.687, P < .001). Importantly, this improvement in predictive accuracy remained consistent for up to 10 years post-surgery. However, the enhanced predictive capacity of FM compared to IM did not reach statistical significance (Table 3 and Fig. 2).

Table 3.

Performance metrics for mortality prediction in patients with gastric cancer.

Metric	Models			Baseline vs Interim model		Baseline vs Full model		Interim vs Full model
	Baseline model*	Interim model*	Full model*	Difference*	P value	Difference*	P value	Difference*	P value
C-index
OS	0.683 (0.635–0.731)	0.811 (0.772–0.849)	0.818 (0.782–0.855)	-	<.001	-	<.001	-	.166
DFS	0.687 (0.641–0.732)	0.797 (0.755–0.840)	0.805 (0.764–0.845)	-	<.001	-	<.001	-	.172
AUC
OS (36 mos)	0.719 (0.652–0.786)	0.824 (0.766–0.881)	0.829 (0.773–0.884)	0.104 (0.058–0.151)	<.001	0.110 (0.829–0.155)	<.001	NA	.370
OS (60 mos)	0.712 (0.655–0.770)	0.834 (0.788–0.880)	0.835 (0.789–0.881)	0.122 (0.082–0.162)	<.001	0.123 (0.082–0.163)	<.001	NA	.681
DFS (36 mos)	0.720 (0.660–0.780)	0.816 (0.759–0.873)	0.820 (0.764–0.877)	0.096 (0.051–0.141)	<.001	0.100 (0.055–0.146)	<.001	NA	.375
DFS (60 mos)	0.711 (0.657–0.764)	0.814 (0.764–0.864)	0.818 (0.769–0.867)	0.103 (0.064–0.142)	<.001	0.107 (0.068–0.146)	<.001	NA	.275
cNRI
OS (36 mos)	-	-	-	0.287 (0.174–0.466)	<.001	0.412 (0.239–0.530)	<.001	0.116 (−0.005 to 0.268)	.178
OS (60 mos)	-	-	-	0.378 (0.259–0.496)	<.001	0.374 (0.313–0.512)	<.001	0.041 (−0.182 to 0.169)	.474
DFS (36 mos)	-	-	-	0.308 (0.162–0.447)	<.001	0.370 (0.221–0.485)	<.001	0.092 (−0.152 to 0.232)	.196
DFS (60 mos)	-	-	-	0.306 (0.202–0.448)	<.001	0.365 (0.229–0.472)	<.001	0.066 (−0.185 to 0.184)	.328
IDI
OS (36 mos)	-	-	-	0.077 (0.025–0.148)	.008	0.086 (0.040–0.155)	.002	0.009 (−0.005 to 0.038)	.267
OS (60 mos)	-	-	-	0.096 (0.051–0.158)	<.001	0.108 (0.064–0.165)	<.001	0.011 (−0.005 to 0.040)	.178
DFS (36 mos)	-	-	-	0.084 (0.037–0.143)	<.001	0.093 (0.046–0.158)	<.001	0.010 (−0.006 to 0.035)	.192
DFS (60 mos)	-	-	-	0.079 (0.034–0.142)	<.001	0.092 (0.045–0.157)	.002	0.013 (−0.004 to 0.042)	.164

AUC = area under the curve, BM = Baseline model, C-index = concordance index, cNRI = continuous net reclassification index analysis, DFS = disease-free survival, FM = model consisting of age, stage, geriatric nutritional index, five-factor modified frail index, and type of gastrectomy, IDI = integrated discrimination improvement, IM = model consisting of age, stage, geriatric nutritional index, and type of gastrectomy, mos = months, NA = not assessable, OS = overall survival.

^*95% confidence interval in parenthesis.

Figure 2.

Comparison of concordance indices between models for survival outcomes. (A) Overall survival, (B) Disease-free survival.

The analysis of the AUC demonstrated that, overall, the risk of both OS and DFS for patients in the FM was significantly higher than that of the BM at both 36 and 60 months after surgery (P < .001 overall). Similarly, the overall risk of OS and DFS for patients in the IM was significantly higher than that for BM at 36 and 60 months post-surgery (P < .001 overall). Nonetheless, while FM exhibits enhanced predictive capacity over IM, this improvement lacks statistical significance (Table 3).

Furthermore, the assessment of cNRI and IDI revealed a notable difference between the 2 models, FM and BM, in predicting both OS and DFS at 36 and 60 months (P < .05 overall). Similarly, the analysis of cNRI and IDI indicated a significant distinction between the 2 models, IM and BM, for predicting OS and DFS at 36 and 60 months (P < .05 overall). However, while FM exhibits improved predictive capacity over IM, this improvement does not reach statistical significance (Table 3 and Fig. 3).

Figure 3.

Continuous net reclassification index and integrated discrimination improvement for comparing the baseline and full model for survival outcomes. (A) Overall survival (OS) risk at 36 months, (B) OS risk at 60 months, (C) disease-free survival (DFS) risk at 36 months, (D) DFS risk at 60 months.

DCA consistently reinforced the improved predictive capabilities of both FM and IM compared to BM for forecasting both OS and DFS risks at 36 and 60 months post-surgery, encompassing threshold probabilities ranging from 0% to 40%. However, when examining DCA for both OS and DFS risks at both 36 and 60 months post-surgery, it revealed no significant difference in predictive capacities between FM and IM (Fig. 4). In this study, direct statistical comparisons between the 2 models in terms of differences and P values were not conducted within the context of DCA.

Figure 4.

Decision curve analysis comparing the models for survival outcomes. (A) Overall survival (OS) risk at 36 months, (B) OS risk at 60 months, (C) disease-free survival (DFS) risk at 36 months, (D) DFS risk at 60 months.

3.6. Establishment of nomograms for overall and disease-free survival

We developed nomograms to predict OS and DFS using a FM that incorporated significant variables, including age, TNM stage, GNRI, mFI-5, and TOG. Notably, a point span ranging from 0 to 30 points suggests that mFI-5 covers a range of values, indicating that the variable can capture different levels of frailty among patients, which can be clinically relevant, underscoring its clinical significance (Fig. 5). The calibration curves displayed a close alignment between predicted and actual survival outcomes (Fig. 6).

Figure 5.

Nomograms for predicting 3- and 5-year survival based on the full model. (A) Overall survival, (B) disease-free survival.

Figure 6.

Calibration curves predicting survival. (A) Overall survival (OS) at 36 months, (B) OS at 36 months, (C) disease-free survival (DFS) at 36 months, (D) DFS at 60 months.

4. Discussion

Our study highlights the clinical significance of the mFI-5, along with age, TNM stage, GNRI, and TOG, as valuable predictors of long-term survival in patients with GC. The FM consistently demonstrates enhanced predictive accuracy compared to the BM. However, it is important to note that while the FM improves predictive power over the IM, this enhancement does not achieve statistical significance across multiple metrics. These findings collectively emphasize the potential clinical value of the FM as a robust tool for surgeons in predicting long-term survival outcomes before surgery in patients with GC.

The mFI-5 is a simplified and abbreviated index consisting of 5 components, such as functional status, diabetes, chronic obstructive pulmonary disease, congestive heart failure, and hypertension requiring medication. Therefore, it has the advantage of being a quick and easy tool for assessing surgical risk. Numerous studies have examined the ability of mFI-5 to predict surgical outcomes in patients with malignant tumors. Hermann et al^[17] found that mFI-5 is a potent predictor of 30-day mortality and severe postoperative complications in patients who underwent hysterectomy for benign and oncological indications. In the study published by Mah et al^[18], mFI-5 had substantial predictive power for complications in patients with gynecologic cancer aged > 70 years who underwent surgery. Lee et al^[19] discovered in their investigation that the mFI-5 was an independent predictor of postoperative complications and administrative outcome categories in patients with lung cancer undergoing surgical resection. Ding et al^[20] previously identified mFI-5 as a determinant of adverse outcomes, such as hypoproteinemia following surgical resection in patients with GC.

The results of present study underscore the clinical significance of the mFI-5 as a predictor of long-term survival in patients with GC. Kaplan–Meier analysis and multivariate CR established mFI-5 as an independent determinant of both OS and DFS. The wide point span of mFI-5 in the nomogram, ranging from 0 to 30 points, highlights its clinical relevance by enabling the assessment of various frailty levels among patients. This versatility is crucial for tailoring interventions and care plans to individual patient needs. Furthermore, the calibration curves demonstrate the precision and reliability of the nomograms, indicating a close alignment between predicted and actual survival outcomes. While the inclusion of mFI-5 in the FM enhanced predictive capacity compared to the IM, it’s important to note that statistical significance was not reached across various metrics. Nevertheless, the consistent improvement observed across multiple metrics suggests that mFI-5 significantly contributes to the FM’s predictive capacity. This study, being the first to establish the role of mFI-5 as a determinant of OS and DFS in patients with GC, emphasizes its potential clinical importance in risk assessment and tailoring interventions for patients with varying levels of frailty. These findings provide valuable insights for clinicians in the management and treatment of patients with GC.

In addition to mFI-5, this study found that age, TNM stage, GNRI, and TOG were independent predictors of OS and DFS. Age is often regarded as a marker of physiological reserve and has been shown to predict survival outcomes in patients with GC.^[6] In the current study, we found a significant difference in median age across various mFI-5 scores, ranging from 55 to 76 years (P < .001). Although mFI-5 was associated with age, multivariate CR indicated that both variables were independent predictors of OS and DFS without collinearity, as indicated by the VIFs.

The TNM staging system is widely considered the gold standard for the prognosis of malignant tumors.^[1] While previous studies have reported the prognostic value of the TNM stage as a determinant of survival in patients,^[24–27] the inability to incorporate other variables and variable outcomes from the same tumor stage are considered some of their limitations.^[1,2] However, in the multivariate CR in our study, TNM stage remained a significant determinant of both OS (HR, 3.15; P < .001) and DFS (HR, 3.33; P < .001) after adjusting for other covariates, indicating its continued importance as a prognostic factor for GC survival outcomes.

Malnutrition contributes to tumor recurrence through tumor immunosuppression and is associated with poor survival outcomes.^[28] However, there is no consensus on optimal nutritional markers. Recently, the GNRI was introduced as a nutritional index that includes serum albumin level, body weight, and height, and has been reported to be a determinant of OS in patients with stage I to III GC.^[25] In the present study, we observed a significant association between the mFI-5 and GNRI (P = .004). However, multivariate CR showed that the 2 covariates were independent predictors of OS and DFS without collinearity, as indicated by the VIFs.

Additionally, prior research has highlighted the nutritional challenges faced by patients undergoing total gastrectomy, which can lead to high rates of 30-day morbidity and readmission.^[29,30] Total gastrectomy has also been identified as a determinant of OS in patients with GC.^[3,27] Our multivariate CR in this study affirmed that the TOG remained a significant predictor of both OS (HR, 1.62; P = .038) and DFS (HR, 1.57; P = .044) even after accounting for other covariates.

This study rigorously compared 3 models: the BM, IM, and FM to assess their predictive capabilities in patients with GC. Our findings highlight the importance of using multiple metrics, including the C-index, AUC, cNRI, IDI, and DCA, to comprehensively evaluate model performance rather than relying solely on a single metric. The C-index reveals the model’s ability to discriminate between event occurrences, while the AUC measures overall model performance in ranking patients based on event risk. The cNRI and IDI metrics indicate the net improvement in risk prediction when transitioning between models, emphasizing the added value of incorporating significant variables like mFI-5 in the FM. DCA assesses the practical utility of a model in clinical decision-making, providing valuable insights into its real-world applicability. Therefore, by considering a range of metrics, our study provides a robust assessment of these models.

Regarding the IM, findings of present study emphasize the potential value of the IM as a valuable tool for surgeons in predicting long-term survival outcomes before surgery in patients with GC. The IM consistently demonstrated a high C-index for both OS and DFS, maintained impressively over a remarkable 10-year post-surgery period. When compared to the BM, the IM significantly outperformed it across various metrics, including C-index, AUC, cNRI, IDI, and DCA, providing strong evidence of its superior predictive capabilities. These results collectively underscore the IM’s potential as a robust tool for enhancing clinical decision-making in the management of patients with GC, particularly in predicting long-term survival outcomes.

Regarding the FM, our findings highlight the substantial clinical value of the FM as a robust tool for surgeons in predicting long-term survival outcomes before surgery in patients with GC. The FM consistently demonstrated impressive predictive accuracy with high C-index values for both OS and DFS, maintained over a 10-year post-surgery period. The use of FM in developing nomograms further emphasized its clinical utility, aligning closely with actual survival outcomes. When compared to the BM, the FM significantly outperformed it across various metrics, including C-index, AUC, cNRI, IDI, and DCA, providing strong evidence of its superior predictive capabilities. However, it’s essential to note that when directly comparing the FM to the IM, although multiple metrics indicated improved predictive capacity in favor of FM, this enhancement did not reach statistical significance. While the FM consistently showed improved predictive capacity compared to BM and IM across various metrics, it is important to note that statistical significance was not always achieved. Nevertheless, these findings emphasize the potential clinical value of the FM, especially when it comes to its practical utility in guiding clinical decisions for patients with GC.

This study presents several noteworthy strengths and findings. Firstly, it pioneers the exploration of mFI-5 as a predictor of long-term survival in patients with GC. Through Kaplan–Meier analysis and multivariate Cox regression, mFI-5 emerges as a robust determinant of both OS and DFS, highlighting its clinical relevance. The inclusion of mFI-5 within the FM significantly enhances its predictive capacity compared to the IM, although statistical significance is not consistently achieved across various metrics. The nomogram further underscores the significance of mFI-5 by assigning it a broad point range, signifying its ability to capture a wide spectrum of frailty levels among patients, thus emphasizing its clinical relevance. Secondly, the FM consistently exhibits a high C-index for both OS and DFS, maintaining strong performance indicators for up to 10 years post-surgery. Compared to the BM, the FM significantly outperforms it, as evidenced by markedly higher C-indices for both OS and DFS. A comprehensive analysis involving metrics such as the AUC, cNRI, IDI, and DCA further validates the FM’s superiority in predicting OS and DFS risks compared to the BM. These findings underscore the potential value of the FM as an invaluable tool for surgeons in forecasting long-term survival outcomes before surgery. Additionally, the development of nomograms using the FM results in calibration curves that closely align with actual survival outcomes, reinforcing the clinical utility of these predictive models. Thirdly, this study employs a diverse array of metrics, including the C-index, AUC, cNRI, IDI, and DCA, providing a comprehensive evaluation of model performance. This approach recognizes the importance of considering various metrics rather than relying solely on a single metric, ensuring a robust assessment of the models. Different metrics offer unique insights into a model’s performance, enhancing the overall reliability of the study. In conclusion, this study highlights the clinical significance of mFI-5 as a predictor of long-term survival in patients with GC. The FM, which incorporates mFI-5 along with other significant variables, consistently demonstrates improved predictive accuracy compared to the BM and the IM. However, it is important to note that statistical significance is not consistently reached across all metrics. These findings emphasize the potential clinical value of the FM as a tool for forecasting long-term survival outcomes in patients with GC. The inclusion of mFI-5, in conjunction with other variables, enhances the model’s predictive power, although statistical significance varies. Despite this, the FM’s practical utility in guiding clinical decisions for GC patients is underscored. It provides clinicians with valuable insights into risk assessment and tailored interventions, ultimately contributing to improved patient management.

However, this study has some limitations that should be acknowledged. First, the retrospective nature of the study means that the data were collected from existing medical records, which may have introduced biases and confounding factors that were not fully considered in the analysis. Although efforts were made to control for potential confounders through multivariate analysis, residual confounding remains a possibility. Second, the sample size was relatively small, which may limit the generalizability of our results to other populations and settings. While the study’s findings provide valuable insights into the prognostic value of the mFI-5 in patients with GC undergoing curative-intent gastric resection, larger studies involving diverse patient populations are warranted to validate these results. Lastly, the study was conducted at a single center, which may have limited the external validity of the findings. Multi-center studies involving multiple institutions could provide more robust and generalizable evidence on the predictive utility of the mFI-5 and other variables in GC outcomes. While acknowledging these limitations, this study lays the foundation for further research in this area, and its findings have potential implications for optimizing patient care and surgical decision-making in GC management. Despite these limitations, this research serves as an important foundation for further investigation in this field. Larger, multi-center studies involving diverse patient populations are needed to validate and expand upon these results. The findings of this study have the potential to impact patient care and surgical decision-making in the management of GC, emphasizing the need for continued research in this area to improve patient outcomes.

Author contributions

Conceptualization: Soomin An, Wankyu Eo, Sookyung Lee.

Data curation: Soomin An, Wankyu Eo, Sookyung Lee.

Formal analysis: Soomin An, Wankyu Eo, Sookyung Lee.

Investigation: Soomin An, Wankyu Eo, Sookyung Lee.

Methodology: Soomin An, Wankyu Eo, Sookyung Lee.

Supervision: Wankyu Eo, Sookyung Lee.

Writing – original draft: Soomin An.

Writing – review & editing: Soomin An, Wankyu Eo, Sookyung Lee.