Baseline characteristics
A total of 25955 participants (weighted population of 156,150,8941) were included, with mean (SD) aged 47.0 (31.2, 59.0) years, 13210 (51.3 %) women and 11,878 (70.2%) non-Hispanic white participants. SII was grouped into Q1 (1.53≤ SII <337.87), Q2 (337.87≤ SII <476.00), Q3 (476.00≤ SII <676.42) and Q4 (SII ≥676.42) categories, and the participant characteristics were presented by the SII groups in Table 1. Participants with higher SII were more likely to be female, white people, with an education background of more than high school, married, never smoker but current drinker, BMI > 30 and with hypertension and diabetes. During 90.46 (47.99) months of follow-up, 3047 of 25955 participants died. A total of 781 deaths occurred due to cardiovascular diseases and 717 due to cancer (Table S1 in the Supplement).
Association between SII and mortality
The relationships between SII and mortality were presented in Fig.2. In minimally adjusted model, the adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs) in Q4 of SII were 1.34 (1.19, 1.50) for all-cause mortality and 1.52 (1.33, 1.75) for non-cancer mortality compared with Q1. The magnitudes of the association attenuated when the model was fully adjusted but remain statistically significant,with the aHRs and 95% CIs in Q4 of SII were 1.24 (1.09, 1.41) for all-cause mortality and 1.41 (1.23, 1.63) for non-cancer mortality compared with Q1. Notably, for cancer mortality in fully adjusted model, the aHRs and 95% CIs in Q2 of SII were 0.70 (0.50, 0.99) and in Q3 were 0.68 (0.52, 0.87) compared with Q1 (Fig.2). All trends remained similar after imputing the missing data or after excluding deaths that occurred within the initial 2-year follow-up (Fig.S1 and Fig.S2 in the Supplement).
Subgroup analysis
The subgroup analysis was shown in Fig.3. Participants who were more than 60 years , non-Hispanic black, other races, with an education background less than high school, married, never drinker, with hypertension or diabetes showed statistically significant association between SII and all-cause mortality. In addition, participants who were male, more than 60 years, non-Hispanic black, other races, with an education background less than high school, married, never smoker and never drinker, BMI <25 or >30, with hypertension, with diabetes, with and without dyslipidemia showed positive association between SII with non-cancer mortality (Fig.3). Interaction tests revealed no significant difference in the correlation between SII and mortality for age, gender, race, education attainment, marital status, PIR, smoking status, alcohol drinking status, the prevalence of hypertension and dyslipidemia (P for interaction > 0.05). However, BMI and the prevalence of diabetes may influence the positive correlation between SII and mortality (P for interaction < 0.05).
Analysis of RCS model
After adjusting all the covariates, non-linear relationships of J-shaped curves were observed between SII with all-cause and non-cancer mortality in RCS regression (P < 0.001, Fig.4). Higher SII was significantly associated with higher all-cause mortality or non-cancer mortality after SII reaching 406.00 or 410.58, respectively. Besides, a U-shaped association between SII and cancer mortality was observed. The risk of cancer mortality decreased with SII below the threshold of 445.22, and the decreasing trend reversed after SII exceeding the threshold.
Kaplan-Meier survival analysis
Participants in the Q4 of SII had a higher cumulative hazard for all-cause mortality and non-cancer mortality than those in the Q1 of SII (Q4 vs. Q1, P < 0.001, Fig.5A, 5C). For cancer mortality, although there is no statistically significant difference cumulative hazard between participants of Q1 and Q4, the two populations have higher cumulative hazards than the counterparts of Q2 or Q3 (Q4 vs. Q1, P > 0.05, Q1 vs. Q2 or Q3, P < 0.001, Q4 vs. Q2 or Q3, P < 0.001, Fig.5B). Furthermore, SII of Q2 and Q3 were combined into one group. The participants with SII of Q1 had a higher cumulative hazard (Q1 vs. Q2 + Q3, P = 0.0019, Fig.5D), and those with SII of Q4 had the similar trend (Q4 vs. Q2 + Q3, P < 0.001, Fig.5D) than counterparts with median SII values (Q2 + Q3).