Clinical Characteristics
A total of 2376 patients aged 65 years or older were enrolled in our study after excluding patients who met the exclusion criteria. The median age was 70 years old (interquartile range (IQR) [67.00–74.00]), with the proportion of males and females being 68.2% and 31.8%, respectively. The percentage of patients with lung cancer was the largest at 34.2%, followed by upper gastrointestinal cancer 21.9% and colorectal cancer 19.1%. There were only 6.4% patients with hepato-biliary and pancreatic cancer and 18.4% patients with other cancer types. Majority of the patients had advanced stage disease, with the percentage of stage Ⅳ disease being 49.6% and stage Ⅲ being 24.5%. The median BMI in all population was 22.19 (IQR [19.79–24.39]). During the median follow-up of 17.43 months, we observed 1112 cases of death.
Comparison of performance and discrimination of the BMI-based inflammation index
To select a robust indictor for survival in older patients with cancer, we compared the C-index in 13 variables related to immunity, inflammation, and nutrition (Table S1). BCLR showed the largest C-index of 0.636, with a 95% CI ranging from 0.619 to 0.654. Similarly, the area under the curve (AUC) of BCLR was higher than that of other biomarkers (1-year AUC=0.658, 3-year AUC=0.668) (Figure 2). The level of BCLR in different cancer types is shown in Figure S2. Liver cancer patients had the lowest BCLR level while breast cancer patients had the highest. There was a weak positive correlation between BCLR and albumin (Males: R=0.36, P<0.001; Females: R=0.37, P<0.001) and a weak negative correlation between BCLR and neutrophils (Males: R=-0.22, P<0.001; Females: R=-0.27, P<0.001). In men, a very weak positive association was observed between BCLR with QLQ-C30 (R=0.21, P<0.001) and hemoglobin (R=0.20, P<0.001) (Figure S3).
Patient baseline characteristics stratified by BCLR
We set a cut-off point of 6.42 for BCLR using standardized log-rank statistics, based on which all older patients with cancer were divided into two groups (Figure S4). There were 1262 patients with low BCLR and 1114 patients with high BCLR. Statistical results indicated that patients with low BCLR tended to be male, with older ages, advanced TNM stages, a history of diabetes, smoking, drinking, and loss of weight, and with lower BMI, CC, HGS, albumin, hemoglobin, lymphocyte, and QLQ-C30. Additionally, the patients with low BCLR were more likely to have increased WBC, neutrophil, platelet, CRP, NRS 2002 scores, and ECOG grade. More detailed demographic information, tumor-related characteristics, and laboratory data are presented in Table 1. Moreover, the median summary score of quality of life in low BCLR population was 84.87 (IQR [75.22–93.14]), notably lower than patients with high BCLR (92.05, IQR [84.20–97.44], P<0.001). Scores of every domain in two groups are also listed in Table S2.
Association between BCLR and all-cause mortality
Restricted cubic spline plot exhibited a significant L-shape association between BCLR and HR for all-cause mortality, with P-value for nonlinearity <0.001 (Figure 3). As the value of BCLR decreased per standard deviation, the risk of mortality rose to 1.17 (95%CI=1.07–1.28, P<0.001) after adjusting for gender, smoking, drinking, diabetes, hypertension, tumor type, tumor stage, treatment, NRS 2002, ECOG, HGS, CC, albumin, hemoglobin and platelet levels. We divided the BCLR into quartiles and set the fourth quartile as the reference. Compared with the fourth quartile (Q4, >15.65), the third quartile (Q3, 5.48 to 15.65), the second quartile (Q2, 1.23 to 5.48) and the first quartile (Q1, ≤1.23) were all associated with a worse prognosis. The HR for mortality in Q3, Q2, and Q1 were 1.22 (95%CI=1.01–1.48, P=0.042), 1.49 (95%CI=1.24–1.80, P<0.001), and 1.88 (95%CI=1.56–2.28, P<0.001), respectively after adjusting for confounding factors (P value for trend<0.001) (Table 2).
Survival analysis and prognostic significance stratified by BCLR
Kaplan–Meier curve analysis showed that compared to high BCLR, low BCLR had shorter OS in older patients with cancer (Log-rank P<0.001) (Figure 4). Multivariate Cox regression confirmed low BCLR was an independent risk factor for prognosis, after adjusting for gender, smoking, tumor type, tumor stage, surgery, radiotherapy, chemotherapy, NRS 2002, ECOG, HGS, CC, albumin, hemoglobin and platelet counts (HR=1.51, 95%CI=1.32–1.73, P<0.001) (Table 3). We also performed a sensitivity analysis to confirm these results by excluding patients who died within 6 months after the beginning of our study. Similarly, low BCLR was still negatively related to survival and was an unfavorable factor for prognosis after adjusting for confounding factors (HR=1.40, 95%CI=1.20–1.64, P<0.001) (Figure S5).
Further we plotted Kaplan–Meier curves according to tumor types and found that older patients with low BCLR had shorter OS in lung cancer, upper gastrointestinal cancer, hepato-biliary and pancreatic cancer, and colorectal cancer (Figure S6). Moreover, low BCLR was an independent negative prognostic factor in older patients with lung cancer (HR=1.40, 95%CI=1.13–1.73, P=0.002), upper gastrointestinal cancer (HR=1.32, 95%CI=1.01–1.73, P=0.041), hepato-biliary and pancreatic cancer (HR=1.78, 95%CI=1.10–2.88, P=0.020) and colorectal cancer (HR=2.24, 95%CI=1.57–3.18, P<0.001). In addition, low BCLR had an adverse impact on survival in I, II, III and IV TNM stages among older patients with cancer (Figure S7). When combined with low BCLR, the prognostic value of TNM stage was also improved. The 1-year and 3-year AUC of TNM were 0.666 and 0.705, respectively, and the AUC for TNM plus BCLR were 0.706 and 0.745, respectively (Figure S8). Notably, the prognostic value of low BCLR among older cancer patients was also confirmed in underweight (BMI<18.5 ), normal (18.5 ≤BMI<24.0 ) and overweight and obesity (BMI≥24.0 ) populations (Figure S9). We also performed subgroup analyses based on various clinical characteristics, including gender, HGS, CC, NRS 2002, ECOG grade, tumor type, TNM stage, albumin and hemoglobin. The results confirmed the predictive value of low BCLR for worse prognosis in all subgroups (Figure 5).
Randomized internal validation of low BCLR in the elderly cancer patients
To evaluate the performance of BCLR in predicting prognosis in older patients with cancer, we further randomly assigned the total population into two validation groups, with a seven to three ratio based on computer-generated random numbers. There were 1664 patients in validation group A and 712 patients in validation group B. The baseline characteristics are displayed in Table S3. The low BCLR group included 874 (52.5%) patients and 790 (47.5%) patients were classified into high BCLR group in validation cohort A. The low BCLR group comprised 388 (54.5%) patients and 324 (45.5%) patients were in high BCLR group in validation cohort B. Multivariate analysis confirmed that low BCLR was an independent, unfavorable factor for survival in both validation groups after adjusting for confounding factors (Validation cohort A, HR=1.44, 95%CI=1.24–1.69, P<0.001; Validation cohort B, HR=1.69, 95%CI=1.30–2.20, P<0.001) (Table S4).
Relationship between BCLR and secondary outcomes
We also performed Logistic regression to explore the relationship between BCLR and secondary outcomes in older patients with cancer (Table S5). Low BCLR was an independent risk factor of short-term survival after adjusting confounder factors (OR=3.58, 95%CI=2.47–5.33, P<0.001). We divided the BCLR into quartiles and set the fourth quartile as the reference. Compared with patients in high BCLR (Q4, >15.65), the risk of mortality within 90 days was more than 3 times higher (OR=5.45, 95%CI=3.30–9.49, P<0.001) in patients with low BCLR (Q1, ≤1.23). In similar trend, low BCLR was an independent risk factor associated with malnutrition (OR=2.66, 95%CI=2.18–3.26, P<0.001) and cachexia (OR=2.12, 95%CI=1.77–2.55, P<0.001). The risk of developing malnutrition and cachexia rose to more than 4 (OR=6.01, 95%CI=4.33–8.45, P<0.001) and 2 times (OR=2.64, 95%CI=2.04–3.43, P<0.001) in patients with low BCLR (Q4 vs Q1).