Both low and high serum ferritin levels predict mortality risk in hemodialysis patients without inflammation

Serum ferritin concentration is clinically used as a marker of iron store, and it is also an independent predictor of mortality and other adverse outcomes in hemodialysis patients. Kalantar-Zadeh et al. [1] previously showed that serum ferritin correlated positively with hospitalization days and frequency, and that the increase in it predicted death in 12 months in 101 hemodialysis patients. They performed another historical cohort study in 58,058 maintenance hemodialysis [2] showing a higher risk for all-cause and cardiovascular mortality only in serum ferritin ranges >1200 ng/mL as compared with the reference range (100–199 ng/mL) in models adjusted for case-mix variables. When further adjustment for 9 variables for the malnutrition–inflammation–cachexia syndrome, the association between high serum ferritin and mortality was much less significant. Recently, Karaboyas et al. [3] reported a trend of increase in serum ferritin in the US after the implementation of the new bundled payment system in the US, with the median level of serum ferritin of 601 ng/mL in 2009 and 887 ng/mL in 2012.

In contrast to the above studies in the US, Hasuike et al. [4] found in a cohort of 90 Japanese hemodialysis patients that serum ferritin >100 ng/mL was associated with higher mortality risk than those with lower ferritin levels. Recently, Kuragano et al. [5] performed a prospective cohort study of 1086 Japanese maintenance hemodialysis patients with repeated measurements of serum ferritin, showing that the risk of death and/or adverse events was higher in those whose ferritin levels were kept high and in those showing large fluctuation in ferritin. In their study, high ferritin was defined as >100 ng/mL based on the mean and median serum ferritin levels (127 and 79 ng/mL, respectively) measured during the study period. Thus, there are considerable international differences in serum ferritin concentrations and in the association between serum ferritin concentration and adverse outcomes, which may affect the upper limit of serum ferritin concentration in the management of anemia in hemodialysis patients. However, the reason is unknown why such difference exists in the ferritin–mortality relationship among countries.

Based on the facts that serum ferritin is lower in Japanese hemodialysis patients than those in other countries [6], that serum ferritin is increased in the presence of inflammation [7], and that the Japanese dialysis patients are generally less inflamed as shown by a lower C-reactive protein (CRP) level than those in the Western countries [8, 9], we hypothesized that the association between serum ferritin and the adverse outcomes could be modified by the presence of inflammation.

In this study, we addressed a question whether the associations between serum ferritin and adverse clinical outcomes are modified by the presence of inflammation in hemodialysis patients using the dataset of J-DOPPS III and IV. In the stratum with low CRP levels, serum ferritin showed a U-shaped association with all-cause mortality after adjustment for relevant confounding factors. However, such relationship was not significant in the high CRP counterparts. Similar results were obtained when the total cohort was stratified by serum albumin. In contrast, we found a linear and positive association between serum ferritin and CVD hospitalization regardless of the levels of the serum markers of inflammation. These results support the view that the association between serum ferritin and all-cause mortality is modified by inflammation, whereas the presence of inflammation does not apparently modify the relationship between serum ferritin and CVD hospitalization.

We report here that the association between serum ferritin and mortality risk was different in the presence and the absence of inflammation. Kalantar-Zadeh et al. [2] were the first that reported the inter-relationship among serum ferritin, inflammation, and mortality risk of hemodialysis patients. They showed in the US hemodialysis patients that mortality risk was higher only in serum ferritin ranges >1200 ng/mL as compared with the reference range between 100 and 199 ng/mL, but that the hazard ratios were close to 1 (no association) in models further adjusted for variables related to malnutrition and inflammation. The mean serum ferritin concentration was 556 ng/mL in those who received IV iron at baseline and 718 ng/mL in those who did not receive it. They discussed that the increased mortality risk in those with high serum ferritin concentration was due mostly to the confounding effects by systemic inflammation and/or malnutrition. In contrast, a recent study by Park et al. [7] showed that the groups with high serum ferritin level (>142.6 ng/mL) showed a higher all-cause mortality risk than those with lower serum ferritin levels, and that the association between serum ferritin and mortality risk was independent of systemic inflammation and nutritional status in incident hemodialysis patients by including CRP as one of the covariates in the Cox model. The median serum ferritin in the Korean hemodialysis cohort was 213 ng/mL, much lower than that in the US patients, but slightly higher than that of Japanese hemodialysis patients [5, 16]. The study by Park et al. [7] may be of clinical importance because the results suggest the possible direct effects of ferritin independent of inflammation. However, their study does not answer the question whether or not systemic inflammation affects the association between serum ferritin and mortality risk.

Although we do not know precise mechanisms for the observed difference in the serum ferritin–mortality association between those having different degrees of inflammation, some explanations are possible. As discussed by Park et al. [7], ferritin appears to reflect some systemic conditions, other than inflammation, that could result in increased mortality and other adverse outcomes. Serum ferritin was reported to be increased in the presence of renal cell carcinoma [17] and other malignancies [18]. Also, increased tissue ferritin levels in the general [19] and in hemodialysis populations [20] are believed to indicate oxidative stress and tissue damage induced by iron released from ferritin. Furthermore, ferritin can act as a suppressor of immune system [21] which is involved in not only infectious disease but also cardiovascular disease. Therefore, even in the absence or at a very low grade of inflammation, an elevated level of serum ferritin could predict poor clinical outcomes independent of the markers of inflammation, as shown in the present study. On the other hand, it is known that there is considerable overlapping in the roles of inflammation, oxidative stress and PEW in prognosis of patients with CKD [14]. In addition, serum ferritin and CRP concentrations correlate with each other to a moderate extent [7]. Therefore, among subjects with signs of inflammation or at a higher grade of inflammation, it may be difficult to separate the roles of ferritin and inflammation in epidemiologic studies, as we report in this study.

In addition to the serum ferritin–mortality relationship, we examined the possible modification by the presence of inflammation of the association between serum ferritin and CVD hospitalization. Unexpectedly, we observed a linear and positive association between serum ferritin and CVD hospitalization in the total cohort as well as in any strata by CRP or serum albumin. These results indicate that the presence of inflammation does not modify the relationship between serum ferritin and CVD hospitalization. We have no clear explanation for the discrepancy between all-cause mortality and CVD hospitalization as endpoints. However, these observations raise a possibility that the effect modification by inflammation occurs in deaths not from CVD, but from infectious disease and cancer. Anyway, it is an important finding that higher serum ferritin predicted higher risk of hospitalization due to CVD in hemodialysis patients regardless of inflammatory status.

We noticed the death risk was higher in those with serum ferritin <50 ng/mL as compared with those with serum ferritin between 50 and 99.9 ng/mL in both strata of CRP <0.3 mg/dL and serum albumin ≥3.8 g/dL. On the other hand, such an increase in mortality risk in the lowest ferritin category was not found in strata of CRP ≥0.3 mg/dL or serum albumin <3.8 g/dL. These results indicate that both low and high serum ferritin levels are predictive of mortality risk in hemodialysis patients without signs of inflammation, and such relationship is hardly seen in the presence of inflammation. Although we have no clear explanations why serum ferritin <50 ng/mL was associated with higher mortality risk in hemodialysis patients without signs of inflammation, at least two explanations are possible. First, since such relationship was not found for CVD hospitalization, the higher risk of death in such subgroup of patients may have derived from non-CVD causes, such as gastrointestinal disease, causing chronic blood loss and iron deficiency. This speculation is supported by the cause of death in the four categories of serum ferritin stratified by CRP. Second, iron deficiency itself may be hazardous. In the general population, the harm of iron deficiency is reported even in those without anemia. Iron supplementation to non-anemic women decreased the degree of fatigue [22, 23] and sleep disturbances [23], both of which are known predictors of worse long-term outcomes in hemodialysis patients [24, 25]. To our knowledge, no previous cohort study reported the relationship of iron deficiency itself with mortality in the general population and other populations. Thus, this study is the first that suggests the potential harm of iron deficiency in the less inflamed subpopulation of hemodialysis patients.

The U-shaped association between serum ferritin and mortality in hemodialysis patients without signs of inflammation in this study appear somewhat different from the report by Kalantar-Zadeh et al. [2], showing that the mortality risk was lower in the lowest range of serum ferritin than in serum ferritin between 100 and 199 ng/mL. We speculate that the discrepancy between these studies can be explained by the different statistical models used in these studies. The US study [2] used a time-dependent model in which serum ferritin and the markers of inflammation and malnutrition were included as time-varying variables. In contrast, we used the traditional baseline model and the inflammation markers were handled as stratification factors based on our research question. It should be noted that baseline models give the associations in a long run, whereas time-dependent models indicate the associations (not always causality) in a short term [26]. We interpret the results by Kalantar-Zadeh et al. [2] to indicated that hemodialysis patients with a low ferritin (<50 ng/mL) are less likely to die in a short term as compared with those with serum ferritin of 100–199 ng/mL regardless of the presence of inflammation. In contrast, our results can be interpreted to indicate that hemodialysis patients with serum ferritin <50 ng/mL without inflammation have an increased risk of death in a long term, whereas such association between low ferritin and mortality was not seen in the presence of inflammation. We performed additional analyses with time-dependent models in which serum ferritin, ESA dose, and IV iron dose were included as time-varying variables which were updated quarterly. In such models, we could not detect the increased risk of death in the lowest ferritin category (<50 ng/mL) in the stratum without inflammation, but the mortality risk of the highest category of serum ferritin was significantly higher than that of the reference ferritin category in stratum with signs of inflammation (data not shown). These results with time-dependent models are similar to the report by Kalantar-Zadeh et al. [2], supporting the above interpretation of the results. Thus, our finding is novel and cannot be made without stratification of the cohort.

It is unknown whether or not a change in serum ferritin due to management of anemia brings on increased risk of the adverse events. Increased serum ferritin concentrations may result from higher dosing of IV iron or lower dosing of ESA [3]. In the US, serum ferritin level is reported to be higher than that before the implementation of the new bundle payment system [3], but no significant increase was found in mortality rate in the US hemodialysis population [27]. The Japanese Society for Dialysis Therapy (JSDT) has published the revised 2015 version of JSDT guideline [28]. In patients having hemoglobin lower than the target range, the new JSDT guideline: (1) proposes iron replacement prior to treatment with ESA in patients not treated with ESA or iron, if serum ferritin is <50 ng/mL; (2) suggests iron replacement therapy in patients treated with ESA, if serum ferritin is <100 ng/mL and transferrin saturation is <20 %, and (3) suggests iron replacement therapy when no condition is identified that could reduce iron utilization rate, if serum ferritin is <100 ng/mL or transferrin saturation is <20 %. As compared with the 2008 version of the JSDT guideline [12], which recommended iron replacement be initiated only in patients with serum ferritin ≤100 ng/mL and transferrin saturation ≤20 %, these revised statements may result in more cases being treated with iron and possibly higher serum ferritin levels than before. On the other hand, the new JSDT guideline does not recommend iron replacement therapy by which serum ferritin goes up to 300 ng/mL or higher. We do not know what happens in Japanese hemodialysis patients, who are less inflamed than those in the US, if the practice pattern is changed and serum ferritin level becomes higher. We need careful follow-up of the nationwide changes in serum ferritin concentration, mortality rate, and their relationship in patients with and without signs of inflammation.

This study has some limitations. First, because of the observational design, the results of this study do not necessarily imply causality. Second, we used serum ferritin concentrations by a single measurement at baseline. Also, we did not take the possible changes during the follow-up period into consideration. Therefore, our results could have underestimated or overestimated the true association. Third, because of the limited number (85) of deaths in the stratum with serum albumin ≥3.8 g/dL, the 95 %CI was large.

In conclusion, we revealed different patterns of association of serum ferritin and mortality in hemodialysis patients with versus without systemic inflammation. Combined with the known international difference in serum CRP and other inflammation markers, this finding may explain the apparent discrepancy in the association between serum ferritin concentration and mortality among countries. Further studies are needed to determine whether the results of this observational study can be translated into the management of anemia for hemodialysis patients.