Predictive value of remnant cholesterol inflammatory index for stroke risk: Evidence from the China health and Retirement Longitudinal study

Study population

This study conducted a secondary analysis of CHARLS, a nationwide cohort study examining the economic, social, and health conditions of Chinese adults aged 45 and older. The baseline survey, conducted from June 2011 to March 2012, initially enrolled 17,708 individuals. For our analysis, we selected data from the 2011, 2013, 2015, and 2018 waves, excluding participants with incomplete follow-up data, a history of stroke or cancer, or unavailable baseline lipid and hs-CRP data. This resulted in a final cohort of 9898 individuals. To investigate the association between cumulative RCII and stroke risk, we further included 5704 participants with complete lipid and hs-CRP data at both the first and third waves, who also had no history of stroke between these periods. The participant selection process is detailed in Fig. 1.

Ethics statement

The CHARLS study involving human participants was approved by the Biomedical Ethics Review Board of Peking University (IRB00001052-11015), and all participants provided written informed consent [23].

Exposure and outcome

Fasting venous blood samples were collected and analyzed by the Chinese Center for Disease Control and Prevention's medical staff at a central laboratory following a standard protocol. Blood lipid levels, including triglycerides (TG), LDL-C, high-density lipoprotein cholesterol (HDL-C), and total cholesterol (TC), were measured using enzymatic colorimetric tests. Hs-CRP (mg/L) was determined using an immunoturbidimetric assay on a Hitachi 7180 chemistry analyzer. RC (mg/dL) was calculated as: RC = TC − (HDL-C + LDL-C). RCII was calculated by multiplying RC by hs-CRP: RCII = RC (mg/dL) × hs-CRP(mg/L)/10. The cumulative RCII (CumRCII) was calculated as: CumRCII = [(RCII2012 + RCII2015) / 2] × (2015 − 2012). The cumulative RC (CumRC) and hs-CRP (CumCRP) were also calculated using the same algorithm: CumRC = [(RC2012 + RC2015) / 2] × (2015 − 2012), CumCRP = [(hs-CRP2012 + hs-CRP2015) / 2] × (2015 − 2012).

The primary outcome of our study was the incidence of new-onset stroke events. Data on stroke occurrence were systematically collected through a questionnaire that queried participants about physician-diagnosed strokes, the timing of diagnosis, and treatment status. Affirmative responses regarding stroke diagnosis during follow-up were classified as first-time events, with the reported date marking the onset. The stroke's timing was ascertained by calculating the interval from baseline assessment to the reported onset. For participants without reported strokes, the follow-up duration was calculated based on the interval between baseline assessment and their final survey date.

Covariates

At the baseline assessment, trained interviewers administered a structured questionnaire to collect sociodemographic and health-related data. The covariates considered included age, sex, residence (urban or rural), self-reported education level (no formal education, primary school, junior high school, or high school and above), marital status (married or other), smoking and drinking status, body mass index (BMI), self-reported physician-diagnosed hypertension, diabetes, heart disease, and history of dyslipidemia medication use. Hypertension was defined by systolic blood pressure (SBP) of ≥ 140 mmHg, diastolic blood pressure (DBP) of ≥ 90 mmHg, the use of antihypertensive medication, or a reported history. Diabetes was identified by fasting plasma glucose (FBG) level of ≥ 126 mg/dL, glycated hemoglobin (HbA1c) level of ≥ 6.5 %, the use of antidiabetic medication, or a reported history.

Statistical analyses

Continuous variables were reported as medians and interquartile ranges (IQR) for non-normal distributions and as means ± standard deviation (SD) for normal distributions. Categorical variables were presented as frequencies (percentage, %). RCII levels were categorized into quartiles for group comparisons. Categorical group differences were assessed using the Pearson Chi-squared test, while continuous variables were compared using one-way ANOVA for normally distributed data and the Kruskal-Wallis test for non-normally distributed data. The distribution of missing data for the included participants was detailed in Table S1. We followed the Missing-at-Random (MAR) assumption to ensure the validity of the imputation process, and missing values were imputed using multiple imputation by chained equations (MICE) method with the “mice” package.

We compared cumulative stroke risks across RCII levels using Kaplan-Meier curves, with the log-rank test assessing differences. The relationship between RC, hs-CRP, RCII, and the risk of new-onset stroke was quantified through multivariable Cox proportional hazards regression, yielding hazard ratios (HR) and 95 % confidence intervals (CI). The incremental predictive value of these biomarkers was evaluated using the C-statistic, net reclassification index (NRI), and integrated discrimination improvement (IDI), and likelihood ratio test (LR). Mediation analysis investigated the role of impaired blood glucose and blood pressure as potential mediators of the association between RCII levels and stroke risk. We also examined hs-CRP and RC as potential mediators in the stroke risk, utilizing the “mediation” package for analysis. Using the “survminer” package, we identified the optimal cut-off values for CumRCII, CumRC, and CumCRP (as shown in Figure S1) to distinguish between high and low exposure groups. Multivariable Cox regression analysis evaluated the impact of CumRCII on stroke risk. Additionally, we utilized Mplus software for a cross-lagged panel model to explore the temporal relationship between RC and hs-CRP.

We assessed the impact of baseline RCII and CumRCII on stroke risk through stratified analyses according to sex, age, BMI, LDL-C levels, and the presence of diabetes, hypertension, and heart disease. To ensure the robustness of our findings, sensitivity analyses were conducted: examining the full dataset without imputation, excluding participants with less than 2 years of follow-up to address attrition bias, and incorporating the fifth wave (2020) data into our longitudinal analysis despite missing specific stroke onset times. Statistical analyses were performed using Stata 17 and R 4.2, considering a two-sided P < 0.05 as statistically significant.

Population characteristics

A total of 9,898 participants in CHARLS from 2011 to 2018 were included in the final analyses. During a median follow-up of 7.0 years, 560 participants (5.7 %) experienced a stroke. The baseline characteristics of these participants, categorized by stroke status, were detailed in Table S1. The mean (SD) age was 58.8 (9.3) years, with females representing 53.0 % of the cohort. Participants who developed stroke exhibited significantly higher baseline levels of RC, hs-CRP, and RCII compared to those without stroke (P < 0.001).

Table 1 categorized the study population into quartiles based on baseline RCII levels. Participants in higher RCII quartiles were older and displayed elevated levels of BMI, SBP, DBP, FBG, HbA1c, TC, LDL-C, TG, hs-CRP, and RC, along with decreased HDL-C (P < 0.001). The prevalence of comorbidities, including hypertension, diabetes, and heart disease, was significantly higher in participants with higher RCII levels (P < 0.001). Moreover, these individuals were more likely to receive lipid-lowering medications, indicating a link between elevated RCII and adverse health outcomes.

Association between RCII and incident stroke

Fig. 2 presented the Kaplan-Meier survival analysis, estimating the cumulative incidence of stroke across RCII quartiles as 3.5 %, 5.1 %, 6.4 %, and 7.6 % for Q1 to Q4, respectively. The Log-rank test revealed a significant difference in stroke incidence rates among the quartiles (P < 0.001), indicating a graded increase in stroke risk with higher RCII levels.

The association between RCII and stroke risk was further examined using Cox proportional hazards regression models, as shown in Fig. 3. All the adjusted variables were evaluated for collinearity, and no clear evidence of multicollinearity was detected (the variance inflation factor of the all included variables was less than 5) (Table S2). The unadjusted model showed a 14.1 % increase in stroke risk for each SD increase in RCII (HR = 1.141, 95 % CI: 1.086–1.199). The fully adjusted Model 3 indicated that RCII was associated with a 10.6 % increased risk per SD increase (HR = 1.106, 95 % CI: 1.048–1.167), which was lower than the 14.2 % increased risk for RC (HR = 1.142, 95 % CI: 1.067–1.223), but higher than the 9.2 % increased risk for hs-CRP (HR = 1.092, 95 % CI: 1.030–1.158). The fully adjusted HR (95 % CI) for RCII in quartiles Q2, Q3, and Q4 were 1.388 (95 % CI: 1.056–1.824), 1.607 (95 % CI: 1.234–2.093), and 1.884 (95 % CI: 1.453–2.442), respectively, compared with those in Q1. The HR for RC and hs-CRP in their corresponding quartiles were consistently lower than those for RCII. Specifically, the highest quartile HR were 1.693 for RC and 1.404 for hs-CRP, both lower than RCII's highest quartile HR of 1.884 (Figure S2). These findings suggest that RCII may be a more effective predictor for stroke risk stratification than RC and hs-CRP.

Incremental predictive value of RCII

Table 2 demonstrated the incremental predictive value of RC, hs-CRP, and RCII when added to the basic model for stroke incidence. The C-statistic improved from 0.662 for the basic model to 0.666 with RC and 0.664 with hs-CRP, and resulted in the highest C-statistic of 0.670 with the addition of RCII (P < 0.001 for all). The NRI improved to 0.1462 with RC, 0.1867 with hs-CRP, and achieved the highest NRI of 0.1986 with RCII (P < 0.001 for all). The IDI also showed the greatest improvement with RCII, increasing to 0.0024, compared to 0.0019 with RC and 0.0008 with hs-CRP (P < 0.001 for RCII and RC, P = 0.029 for hs-CRP). The likelihood ratio test indicated the highest model fit with the addition of RCII, with a statistic of 20.219 (P < 0.001), compared to 14.467 (P < 0.001) with RC and 5.793 (P = 0.020) with hs-CRP. These results suggest that RCII may provide a relatively higher incremental predictive value for stroke incidence compared to RC and hs-CRP.

Furthermore, to assess the predictive value of RC, hs-CRP, and RCII, we conducted a Receiver Operating Characteristic (ROC) analysis (Figure S2). The results revealed that the Area Under the Curve (AUC) for RCII was 0.581, for RC it was 0.566, and for hs-CRP it was 0.560. RCII demonstrated the highest predictive value among the three biomarkers. Although the difference between RCII and RC did not achieve statistical significance (P = 0.166), the trend suggest that RCII may offer a more comprehensive assessment of stroke risk when considering the combined effects of RC and hs-CRP.

Mediation Analysis

Mediation analyses, as detailed in Fig. 4A, indicated that the association between RCII and incident stroke was partially mediated by FBG, SBP, and DBP, with respective mediation proportions of 6.2 %, 20.9 %, and 19.8 % (P < 0.05 for all). Moreover, TyG (Triglyceride and glucose index) also mediated a significant proportion (29.6 %, P = 0.028) of the effect of RCII on stroke risk, as shown in Figure S3. Fig. 4B revealed a complex interplay between RC, hs-CRP, and stroke risk. RC accounted for 10.5 % of the mediating effect between hs-CRP and stroke (P = 0.028), while hs-CRP mediated 3.3 % of the effect between RC and stroke (P = 0.036). This reciprocal mediation suggests a synergistic effect between RC and hs-CRP, implying that their combined influence on stroke risk might be more significant than their individual impacts.

Association between cumulative RCII and incident stroke

In our analysis of a subset from CHARLS between 2011 and 2018, we investigated the potential association between cumulative exposure to RCII, RC, and hs-CRP with the incidence of stroke, as presented in Table 3. The study involved 5,704 participants with a mean age of 58.3 years, of whom 5.0 % (288 participants) suffered a stroke during the 3.0-year follow-up period, according to Table S3.

The fully adjusted model showed a significant increase in stroke risk associated with higher levels of CumRCII. For each SD increase in CumRCII, the risk of stroke rose by 14.1 % (HR = 1.141, 95 % CI: 1.047–1.242). This increase was intermediate between that of CumRC, which showed a higher risk increase of 15.6 % per SD, and CumCRP, which had a lower increase of 8.5 % per SD. Participants with CumRCII levels at or above the optimal cut-off of 36.14 had a significantly increased risk of new-onset stroke (HR = 1.463, 95 % CI: 1.103–1.940). Consistent with these findings, those with CumRC above 73.03 and CumCRP above 5.42 also exhibited significantly increased risks, with HR of 1.552(95 % CI: 1.218–1.979) and 1.301 (95 % CI: 1.028–1.648), respectively. The concurrent presence of high CumRC and CumCRP further elevated the risk, with an adjusted HR of 1.846 (95 % CI: 1.351–2.524), as shown in Table S4 and Figure S4. Furthermore, cross-lagged panel analysis (as shown in Figure S5) showed that baseline hs-CRP significantly influenced the path coefficient of RC during follow-up in the fully adjusted Model (Model2, β2 = 0.127, 95 % CI: 0.093–0.161; P < 0.001). The findings indicate that CumRCII emerges as a potentially valuable composite index for the long-term assessment of stroke risk, reflecting the temporal correlation and chronic interplay observed between RC and hs-CRP.

Subgroup and sensitivity analysis

Subgroup analyses (Fig. 5, Fig. 6) showed that baseline and cumulative RCII were associated with stroke risk in specific groups. Baseline RCII was linked to higher stroke risk in males under 60, those with high BMI, and LDL-C under 100 mg/dL, with significance maintained across individuals with or without chronic conditions like hypertension, diabetes, or heart disease. CumRCII showed similar associations, with the risk being especially pronounced in participants with hypertension, diabetes, or heart disease. No significant interactions were found among subgroups, suggesting a consistent effect of RCII on stroke risk. Sensitivity analyses, as detailed in Tables S5 and S6, demonstrated the robustness of our findings. Our complete-case analysis of 8203 participants showed an 8.1 % increase in stroke risk for each SD increase in RCII (HR = 1.081, 95 % CI: 1.017–1.150). After excluding participants with less than 2 years of follow-up, the risk for each SD increase in RCII was 10.5 % higher (HR = 1.105, 95 % CI: 1.039–1.175). The inclusion of data from the fifth wave (2020) further supported these results, with a 9.1 % higher risk for each SD increase in RCII (HR = 1.091, 95 % CI: 1.039–1.146). Consistent with these findings, higher CumRCII levels were also associated with an increased stroke risk in the sensitivity analyses. These results confirmed the significant association between elevated RCII levels and stroke risk.