Introduction
The prognostic assessment of patients with suspected cardiac dysfunction or systemic inflammatory disease remains a clinical challenge.1,2 B-type natriuretic peptide (BNP) and N-terminal pro BNP (NT-proBNP) are well-established cardiac biomarkers for the diagnosis and monitoring of heart failure.3,4 Inflammatory markers, including C-reactive protein (CRP), interleukin (IL)-6 and tumour necrosis factor (TNF)-α, reflect systemic inflammatory and immune activation, and have been implicated in adverse cardiovascular outcomes.5–9 In addition, markers of immunological activity, including immunoglobulin free light chains (FLCs), are gaining recognition for their roles in chronic inflammatory conditions.10
In prior clinical research, we identified elevated circulating FLC λ levels and a decreased κ/λ ratio in patients with myocarditis-associated heart failure compared with healthy controls.11 These findings suggested that FLC λ and the κ/λ ratio may have diagnostic utility in myocarditis and are prognostically relevant. We also observed significant associations between mortality and elevated NT-proBNP and FLC λ levels, and identified NT-proBNP and the κ/λ ratio as independent predictors of survival.11 A tree-based risk stratification approach further delineated three distinct prognostic groups.11
Despite growing interest in multimodal biomarker profiling, the relative contributions of these markers to risk stratification in advanced heart failure remain incompletely defined. In particular, a unified prognostic model has not comprehensively assessed the interplay between immune dysregulation and cardiac stress biomarkers.
Building on previous observations, the aim of the present study was to evaluate whether a panel of inflammatory, cardiac and immunological biomarkers (including FLCs, BNP and NT-proBNP) can serve as prognostic indicators in patients with advanced heart failure. Using Cox proportional hazards models and survival analyses, we sought to determine which biomarkers provide the most robust association with clinical outcomes and explore their utility in integrated risk assessment.
Methods
Study Design and Patient Population
This was a retrospective cohort study conducted at Hayama Heart Center, a tertiary referral cardiovascular centre in Japan. Consecutive patients hospitalised for advanced chronic heart failure between October 2002 and November 2004 were identified from the medical records and enrolled in this study. All patients underwent comprehensive evaluation.
Heart failure was diagnosed according to standard clinical criteria based on symptoms, physical examination findings and objective evidence of cardiac dysfunction. Patients were required to have New York Heart Association functional class II–IV at the time of enrolment. The underlying aetiologies of heart failure included ischaemic heart disease and dilated cardiomyopathy, as determined by clinical history, imaging studies and coronary angiography when indicated.
Patients were excluded from the study if they had known plasma cell dyscrasias (including multiple myeloma or monoclonal gammopathy), active malignant disease, acute infectious or inflammatory conditions at the time of enrolment or advanced renal failure that could substantially influence circulating FLC levels.
In all, 78 consecutive patients (65 men, 13 women; mean [± SD] age 57.5 ± 12.4 years) met the inclusion criteria and were included in the final analysis.
The study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Hayama Heart Center. The reference number was not retained in the archived documentation. Informed consent was obtained from all participants prior to enrolment, in accordance with the approved protocol.
Frozen serum samples from all participants were stored at −80°C until use in FLC assays.
Biomarker Measurement
Nephelometry and ELISA techniques were used to determine serum FLC κ and λ concentrations.12 Briefly, microtitre wells were coated with 1 µg monoclonal antibody 4E6 for FLC λ and blocked with 0.5% skim milk. Standard FLC solutions (κ: 20–320 µg/l; λ: 17.5–280 µg/l) and serum samples diluted 1:400 were incubated in wells for 2 h, followed by incubation with horseradish peroxidase-conjugated monoclonal antibodies. Absorbance at 492 nm was determined using o-phenylenediamine.
Other biomarkers, namely BNP, CRP, IL-6, NT-proBNP and TNF-α, were measured by BML, Inc. (Tokyo, Japan).
Follow-up and Clinical Endpoints
Patients were followed longitudinally after enrolment through regular outpatient visits and a review of hospital medical records. Survival status was confirmed by medical records and, when necessary, by direct contact with patients or their families.
The primary endpoint of this study was all-cause mortality. Patients were followed from the date of biomarker sampling until death or the last confirmed follow-up. The median follow-up duration was 1,960 days (interquartile range 520–3,553 days).
Statistical Analysis
Prior to statistical analysis, we assessed whether the variables were normally distributed using distribution patterns and Shapiro–Wilk tests. Variables with a skewed distribution were normalised by natural log transformation before analysis. Detailed data visualisations and distributions have been reviewed to ensure the robustness of these transformations.
Given the limited data available for analysis, we carefully constructed a parsimonious model to avoid overfitting, which could reduce statistical power and destabilise various models. This approach allowed us to define the maximum number of analyses. Therefore, we used a parsimonious model.
Multivariable Cox proportional hazards models were used to assess associations between baseline variables and survival outcomes. We calculated HRs with 95% CIs.
Univariable logistic regression was used to identify optimal cut-off values for predicting survival from continuous variables. Patients were subsequently dichotomised based on these thresholds into two groups: those with values equal to or above the cut-off value and those with values below the cut-off value. Kaplan–Meier survival curves were generated for these groups.
We constructed diverse models to maximise the available information from as many biomarkers as possible for each patient. For the final analysis, we selected the model yielding the largest analysable sample size.
To enhance the analysis, the integrated discrimination improvement (IDI) and the net reclassification improvement (NRI) were calculated to quantify the clinical utility of the expanded model.13,14
To evaluate the incremental predictive value of the FLC k/λ ratio for mortality, we assessed the improvement in model discriminative performance after adding the FLC k/λ ratio to the base model (BNP or NT-proBNP alone). In addition to comparing the areas under the receiver operating characteristic curve (AUCs), the NRI and the IDI were calculated to further quantify the clinical utility and reclassification power of the expanded model. The NRI was used to determine the proportion of patients correctly reclassified into higher- or lower-risk categories, whereas the IDI was used to measure the mean increase in the difference between predicted probabilities for events and non-events. Statistical significance was defined as two-tailed p<0.05.
Statistical analyses were performed using JMP version 14 (SAS Institute) and R version 4.3.1 (R Foundation for Statistical Computing) with EZR (version 1.6.8; Jichi Medical University Saitama Medical Center).
Results
In all, 78 patients were included in the study cohort. During the follow-up period, 31 patients died, corresponding to an overall event rate of 40%. Biomarker measurements were not available for all participants, and so the number of individuals included in each analysis varied according to biomarker availability.
FLC κ/λ ratio measurements were available for all 78 patients (31 deaths), whereas BNP and NT-proBNP measurements were available for 76 patients, among whom 30 patients died (event rate 39%). Consequently, the number of events informing each statistical model differed slightly depending on the biomarker included in the analysis.
The exact number of events contributing to each model is shown in Figure 1.
We conducted a Cox proportional hazards regression analysis to assess the association between clinical and biochemical variables and the risk of adverse outcomes. The analysis included 78 patients with available biomarker data, with results presented in Table 1.
BNP was associated with an increased risk of mortality in this cohort, with an HR of 53.02 (95% CI [9.43–253.27]; p<0.0001); NT-proBNP was also associated with an increased risk of mortality, with an HR of 30.26 (95% CI [5.78–128.03]; p<0.0001). The FLC κ/λ ratio was inversely associated with increasing mortality, with a unit-based HR of 0.0066 (95% CI [0.00056–0.064]; p<0.0001). FLC κ and λ levels showed trends towards significant associations with mortality (p=0.0542 and p=0.0426, respectively). Although IL-6 appeared to be associated with mortality risk (HR 4.63; 95% CI [0.93–16.13]; p<0.03), the CIs were wide and the number of events informing the model was limited. CRP and TNF-α were not significantly associated with increasing mortality (p=0.466 and p=0.0786, respectively).
Kaplan–Meier survival analysis suggested a difference in survival between groups stratified according to the FLC κ/λ ratio (cut-off value=0.6). Survival tended to be longer in patients with an FLC κ/λ ratio ≥0.6 than in those with a ratio <0.6 (Figure 1). The median survival for patients with an FLC κ/λ ratio <0.6 was 1,823 days, compared with 6,372 days for those with an FLC κ/λ ratio ≥0.6, indicating significantly longer survival in the latter group. The failure plot supported this finding, showing a higher cumulative failure rate in the group with an FLC κ/λ ratio <0.6. Statistical comparisons using the log-rank and Wilcoxon tests confirmed the survival difference (log-rank χ2=20.29, p<0.0001; Wilcoxon χ2=20.5, p<0.0001).
Comparison of single-marker and expanded models (i.e. BNP versus BNP+FLC κ/λ, and NT-proBNP versus NT-proBNP+FLC κ/λ) showed a trend towards higher AUCs in the expanded models; however, the differences did not reach statistical significance (Figure 2). Improvements in the predictive ability of the models when the FLC κ/λ ratio was added to existing heart failure biomarkers (i.e. BNP and NT-proBNP) are presented in Table 2. In the BNP-based model (Table 2), although the AUC increased from 0.688 in the BNP-only model to 0.795 in the expanded model with the FLC κ/λ ratio, the difference was not statistically significant (ΔAUC=0.107, p=0.065). However, marked improvements were seen in the NRI and IDI. The overall NRI was 0.962 (95% CI [0.562–1.306]; p<0.001), showing a particularly marked improvement in the event (death) group (NRI for events: 0.600, p<0.001). Furthermore, the IDI was 0.137 (95% CI [0.073–0.209]; p<0.001), indicating a highly significant clinical improvement.
A similar trend was observed for the NT-proBNP-based model (Table 2). Adding the FLC κ/λ ratio to NT-proBNP improved the AUC from 0.708 to 0.804 (p=0.102). Regarding the reclassification ability of the expanded model, the overall NRI reached 0.677 (95% CI [0.253–1.077]; p<0.001), showing significant improvement in both the event (death) group (NRI 0.440) and the surviving group (NRI 0.237). The IDI was 0.126 (95% CI [0.061–0.196]; p<0.001), demonstrating a strong additive predictive effect over the baseline model.
To assess the added predictive value of the FLC κ/λ ratio when combined with natriuretic peptides, we evaluated model performance using IDI metrics (Figure 3). In the case of the BNP model, the addition of the FLC κ/λ ratio to BNP resulted in an IDI of 0.1367 (95% CI [0.0734–0.2088]; p<0.001), indicating a significant improvement in discrimination. Similarly, in the NT-proBNP model, the addition of the FLC κ/λ ratio yielded an IDI of 0.1259 (95% CI [0.0609–0.1956]; p<0.001), demonstrating a statistically significant improvement in discrimination. While there is no formal consensus on absolute thresholds, an IDI >0.01 is often considered to reflect a meaningful improvement, and an IDI >0.02 is regarded as a substantial improvement in clinical risk prediction.15 Our results far exceed these empirical benchmarks, suggesting a high magnitude of added usefulness offered by the new marker.
The IDI analysis suggested that the addition of the FLC κ/λ ratio to natriuretic peptides may provide incremental prognostic information beyond natriuretic peptides alone in this dataset. However, given the small sample size and limited number of events, these results should be interpreted as exploratory.
In Figure 3, the grey-shaded area between the dashed (base model) and solid (expanded model) lines represents the total improvement in model performance. The red lines in Figure 3 show the sensitivity in the event (death) group, whereas the black lines show the false-positive rate in the non-event (survivors) group.
Discussion
This study evaluated the potential prognostic relevance of circulating biomarkers reflecting cardiac stress, inflammation and immune activation in patients with advanced heart failure. In this study cohort, BNP, NT-proBNP, IL-6 and the FLC κ/λ ratio were associated with long-term outcomes. In addition, analyses of reclassification metrics suggested that the FLC κ/λ ratio may provide incremental prognostic information when considered with natriuretic peptides.
Consistent with previous observations in heart failure populations, patients in our cohort exhibited a relative reduction in FLC κ levels and an elevation in FLC λ levels compared with reference populations, resulting in a lower κ/λ ratio.11 Importantly, the FLC κ/λ ratio emerged as an inverse predictor of mortality risk, suggesting that dysregulated immunoglobulin light chain balance, potentially reflecting chronic inflammation, immune activation or impaired clearance, contributes to adverse outcomes in advanced heart failure.
Incremental Predictive Value Beyond Natriuretic Peptides
When the FLC κ/λ ratio was added to BNP- or NT-proBNP-based models, modest increases in AUC were observed, but these changes did not reach statistical significance. This finding is not unexpected, because AUC-based comparisons are known to be relatively insensitive to incremental improvements when baseline models already demonstrate moderate-to-good discrimination.13 In such contexts, reliance on AUC alone may underestimate the true clinical utility of newly added biomarkers.
In contrast, reclassification-based metrics revealed a notable improvement in predictive performance. The addition of the FLC κ/λ ratio resulted in relatively high NRI values for both the BNP (0.962) and NT-proBNP (0.677) models, indicating an improvement in correctly classifying patients according to mortality risk. Notably, the improvement appeared to be greater in the event (death) group, suggesting that the addition of the FLC κ/λ ratio may help identify higher-risk individuals who may not be fully captured by natriuretic peptides alone.
At the same time, additional improvements were observed in the non-event group, suggesting improved identification of lower-risk patients.
The IDI analysis further supported the consistency of these findings. IDI values of 0.137 for the BNP-based model and 0.126 for the NT-proBNP-based model were higher than commonly used thresholds for potentially meaningful (≥0.01) and substantial (≥0.02) improvement. Because IDI directly reflects changes in the discrimination slope (i.e. the average separation of predicted risks between events and non-events), it sensitively captures improvements in risk prediction that may be obscured in AUC-based analyses. The consistency of IDI results across both natriuretic peptide models suggests consistency of the additional effect conferred by the FLC κ/λ ratio.
This study demonstrates that several biomarkers are significantly associated with long-term outcomes following ventriculoplasty in patients with advanced heart failure. BNP and NT-proBNP, well-established indicators of cardiac stress, were associated with mortality, consistent with previous literature on heart failure prognosis.16,17
IL-6, a pro-inflammatory cytokine, exhibited a moderate association with mortality, underscoring the contribution of systemic inflammation to disease progression.
Notably, the FLC κ/λ ratio was observed as a significant inverse predictor of mortality risk. This finding may reflect underlying immune dysregulation, although the mechanistic basis of the effect warrants further investigation. Although individual FLC κ and λ levels approached significance, only their ratio achieved strong statistical significance as a predictor of mortality. CRP and TNF-α did not show significant associations with mortality, which may be due to biological variability or the limited statistical power of this study. However, an observed association of the FLC κ/λ ratio with mortality, especially when combined with natriuretic peptides, suggest its potential utility as a novel biomarker for risk stratification.
Our findings are consistent with evidence from previous studies. In an encephalomyocarditis virus-induced myocarditis model in mice, FLC expression increased in conjunction with the progression to heart failure.18 In another study, marked differences in FLC concentrations were observed between healthy controls and patients with heart failure with reduced ejection fraction (HFrEF) due to ischaemic heart disease or dilated cardiomyopathy.19 Compared with healthy individuals, patients with heart failure had lower FLC κ levels, elevated FLC λ levels and a reduced κ/λ ratio. Linear regression analysis revealed an inverse association between NT-proBNP and the κ/λ ratio, as well as negative correlations between the κ/λ ratio and both end-diastolic and end-systolic left ventricular dimensions.19 In addition, a positive correlation was found between the κ/λ ratio and ejection fraction.19 Similar trends were observed in patients with heart failure and preserved ejection fraction (HFpEF), particularly those with hypertrophic cardiomyopathy, namely higher FLC λ levels and lower κ/λ ratios compared with healthy controls.19 These findings indicate that changes in FLC λ levels and the κ/λ ratio are evident in both HFrEF and HFpEF.
Jackson et al. analysed 628 patients recently hospitalised for decompensated heart failure and found that combined FLC (cFLC) levels, encompassing both κ and λ isotypes, were present in 43% of patients.20 Mortality risk increased progressively across cFLC quartiles, with those in the highest quartile having more than double the unadjusted risk compared with those in the lowest quartile.20 After multivariable adjustment, cFLC remained an independent predictor of mortality, with those in the highest quartile having approximately a 50% higher adjusted risk.20 Elevated cFLCs have also been reported in acute myocardial infarction, and their levels are correlated with ejection fraction; high FLC κ or λ concentrations are commonly observed in patients with impaired systolic function.21
Overproduction of FLCs has been associated with an increased risk of mortality. In a cohort of over 15,000 individuals aged ≥50 years, excluding those with multiple myeloma, mortality and its causes were tracked following FLC measurement.22 After adjusting for age, sex and renal function, elevated combined FLC levels were linked to an HR for death of 2.07.22 Another study reported a markedly increased risk of mortality within the first 100 days after enrolment into the study (HR 7.1) among individuals with elevated FLCs, with cardiovascular causes accounting for 41% of these deaths.23
In our recent study, we analysed data from 1,105 residents who participated in a community health screening in 2007.24 Using propensity score matching for age and sex, we selected 400 participants (200 men, 200 women) for further analysis. Logistic regression analysis, applied to 11-year follow-up survival data, identified BNP, FLC λ and HbA1c as significant predictors of mortality.24 Elevated levels of these biomarkers were associated with increased mortality, underscoring their prognostic significance.24
In certain diseases, variations in circulating FLC κ and λ levels suggest that the κ/λ ratio may fluctuate even in cases of polyclonal activation.25–27 In our studies, patients with heart failure had a reduced κ/λ ratio.10,11 Although the regulatory mechanisms driving differential B cell and plasma cell responses remain uncertain, it is plausible that FLC κ and λ are regulated by distinct pathways, potentially mediated by nuclear factor-κB. The differing responses may also result from antigens with restricted epitopes, which promote selective expression of light chains. Antibodies featuring λ light chains exhibit specificities that differ from and complement those of κ-containing antibodies.28,29
Importantly, reclassification analysis in the present study demonstrated that incorporating the FLC κ/λ ratio in models based on BNP or NT-proBNP was associated with significantly enhanced predictive accuracy. The observed NRI and IDI gains suggest potential clinical relevance of integrating FLC measurements into standard biomarker panels.
Overall, the present study suggests the potential value of integrating the FLC κ/λ ratio with traditional cardiac biomarkers to improve prognostic models. The findings also suggest that FLCs may reflect immune and cardiovascular stress, and could potentially contribute to clinical risk stratification and treatment monitoring. Because the data in this study were collected immediately before surgery in patients with advanced heart failure, the results suggest the potential clinical relevance of these biomarkers in predicting postoperative outcomes and supporting patient selection for surgical intervention.
Clinical Implications
Together, the findings of the present study suggest that the FLC κ/λ ratio offers complementary prognostic information when used alongside natriuretic peptides. By improving risk reclassification, particularly among patients who ultimately died, the combined biomarker approach may contribute to supporting clinical risk assessments in patients with advanced heart failure undergoing ventriculoplasty. The observed improvement across the NRI and IDI metrics suggests the potential utility of incorporating immune-related biomarkers into conventional heart failure prognostic models.
Limitations
This study has some limitations. First, the study used a retrospective historical cohort and several baseline clinical characteristics were not systematically recorded, which limited the ability to adjust for established prognostic factors. Second, the sample size was relatively small and the number of observed events was limited. As a result, the statistical estimates should be interpreted with caution, particularly given the wide CIs.
Conclusion
The FLC κ/λ ratio was associated with long-term outcomes in patients with advanced heart failure undergoing ventriculoplasty. When combined with established biomarkers, namely BNP and NT-proBNP, the FLC κ/λ ratio may improve prognostic assessment. Future prospective studies are warranted to validate these findings.