The Positive and Negative Effects of Calcium Supplementation on Mortality in Septic ICU Patients Depend on Disease Severity: A Retrospective Study from the MIMIC-III

Background Calcium administration in septic patients with hypocalcemia is a controversial issue. The present study preliminarily investigated the effects of calcium supplementation on the length of hospitalization and mortality in septic ICU patients with different severities of hypocalcemia and disease. Method A total of 5761 eligible septic patients, including 2689 who received calcium supplementation and 3072 who did not receive calcium supplementation, were extracted from the Medical Information Mart for Intensive Care III (MIMIC-III) database. The cofounding covariates between the calcium supplement and nonsupplement groups were balanced using the propensity score matching model. We compared the length of stay (LOS) in the ICU and hospital with 28-day and hospital mortality and stratified the analysis according to the sequential organ failure assessment (SOFA) score and ionized calcium (iCa) at the first ICU admission in the matched groups. Results The results showed that iCa at the first ICU admission was associated with mortality in sepsis patients (HR: 0.421; 95% CI: 0.211∼0.837), but the lowest mortality rate was observed in patients with mild hypocalcemia. A total of 993 paired patients were included in the analysis after propensity score matching. Regardless of the SOFA score or presence of iCa, the LOS in the ICU was higher in the calcium supplement group than in the nonsupplement group. The survival analysis was stratified by the SOFA score and showed that calcium supplementation reduced mortality when the patient's SOFA score was ≥8 (p=0.002), and it worsened the outcome when the patient's SOFA score was ≤4 (p=0.010). It had no significant effect on patients with SOFA scores ranging from 5 to 7 (p=0.911). Conclusion Our results showed that mild hypocalcemia may be protective in septic patients, and calcium supplementation may have positive and negative effects on mortality depending on disease severity. The SOFA score may be a valuable clinical index for decisions regarding calcium administration.


Introduction
Hypocalcemia is a widely recognized biochemical abnormality in critically ill patients [1,2], and septic patients are at particular risk for hypocalcemia [3,4]. e etiology of hypocalcemia has been extensively researched in septic patients [5,6], and many factors, such as decreased parathyroid hormone (PTH) release, tissue calcium accumulation, ionized calcium (iCa) release into ascites fluid, and hypomagnesemia, are involved in causing hypocalcemia during sepsis. However, the major factors that cause hypocalcemia in sepsis are unclear, and more than half of patients in the intensive care unit (ICU) have no identifiable etiology for hypocalcemia.
iCa is a ubiquitous intracellular messenger and coenzyme throughout the body. Without adequate calcium regulation, the body ceases to function properly and many different clinical signs manifest [7]. Hypocalcemia is associated with poor outcomes in septic patients [8][9][10]. Patients with severe hypocalcemia who fail to recover a normal iCa level during the early stage have significantly higher mortality [2]. Calcium should be replaced to prevent lifethreatening complications, such as laryngospasm, tetany, seizures, and cardiac abnormalities [11,12]. However, the treatment of hypocalcemia in critically ill patients is a controversial issue [13]. We lack evidence-based guidance because no randomized controlled trial has evaluated the effects of calcium administration on prognosis in septic patients. Studies in critically ill patients also yielded conflicting results regarding calcium supplementation. One large retrospective study showed that calcium supplementation improved the 28-day and 90-day survival of ICU patients [14]. However, several clinical observation studies showed that calcium supplementation in critically ill patients provided no benefit or worsened outcomes [15,16]. Experiments using an animal model of sepsis showed that calcium administration increased mortality [17].
iCa is more reliable and sensitive than total calcium for hypocalcemia evaluation. However, iCa is not a routine measurement in the ICU setting, and some treatments, such as a high dose of albumin or fast transfusion, can easily affect iCa. e identification of a reference index to substitute for iCa or total calcium would be valuable in the decisionmaking process for clinical calcium administration. Previous studies on the effects of calcium administration on the prognosis of septic patients did not stratify patients according to disease severity. Whether the effects of calcium administration varied in patients with different SOFA scores were not known. erefore, we performed a retrospective study by extracting the data of sepsis patients from the Medical Information Mart for Intensive Care III (MIMIC-III) database to preliminarily evaluate whether calcium supplementation should be administered during the ICU stay and to identify suitable conditions for calcium supplementation in septic ICU patients. A preprint of the results was previously published [18].

Design.
e present study preliminarily investigated the effects of calcium supplementation on the prognosis of septic patients in the ICU. Factors including age, sex, iCa, and lactate at the first ICU admission and other complications were examined to determine whether there were associations with hospital mortality in septic patients. e prognostic differences between the calcium-supplemented and nonsupplemented groups were stratified and analyzed according to iCa at the first ICU admission and SOFA scores. e ranges of iCa and SOFA scores that worsened or benefited the outcome were examined using survival analysis.

Database.
is study was a single-center, retrospective, observational study. We used the MIMIC-III (latest version 1.4) [19]. Queries were stored on a public repository GitHub (https://github.com/alistairewj/sepsis3mimic/tree/master/query). e explicit sepsis codes were introduced at the BIDMC in 2004, and the group of admissions between 2008 and 2012 was easily identifiable in the database. erefore, only ICU admissions from 2008 to 2012 (n � 23,620) were enrolled in the present study. e admissions included 3 nonadult patients, 7,536 patients with secondary (or greater) admissions (to avoid repeated measures), 2,298 patients who underwent cardiothoracic surgery (their postoperative physiological derangements did not translate to the same mortality risk as to the other ICU patients), 1,974 patients with suspected infections that lasted longer than 24 hours before and after ICU admission (we focused on the majority of patients who were admitted to the ICU with sepsis to ensure independence between data points because the MIMIC-III only contained ICU data), and 6,030 patients with SOFA scores lower than 2 and were considered noninfectious; 18 patients with missing data were excluded. e final cohort contained 5,761 patients (Figure 1), and 2,689 of these patients received calcium supplementation during their ICU stay.

Statistical Analysis.
e normality of the distribution of continuous variables was tested using the Kolmogorov-Smirnov test. Data with a normal distribution are expressed as the mean ± SD and were compared using Student's t-test. Otherwise, data are expressed as the median with quartiles and were compared using the Mann-Whitney U test. Categorical variables are expressed as percentages and were compared using the chi-squared test or Fisher's exact test as appropriate. Variables between survivors and nonsurvivors were compared using univariate analysis to screen for potential risk factors associated with hospital mortality. Covariates with p < 0.1 in the single-factor analysis were included in Cox regression analysis to investigate the significant risk factors associated with hospital mortality. e proportional hazard assumption and the linearity relationship between the covariates and outcome were confirmed by the Schoenfeld residual test, and the linearity relationship between the covariates and outcome was confirmed by the Martale residual test.
To balance the confounding covariates between the calcium-supplemented and nonsupplemented groups, factors that were previously associated with mortality from sepsis, including generic patient characteristics, such as age [20] and sex [21], disease severity indicators, including the SOFA score [22], lactate level on first ICU admission [23], experience of septic shock [24] or ventilation [25], and complications associated with mortality in sepsis from the Cox regression analysis, were selected into the propensity score matching model, which was processed using fuzzy matching to create a 1 : 1 matching with 0.0028 toleration. e blood iCa stage was defined according to the clinical regulations of our department: normal iCa range (1.15∼1.30 mmol/L), mild hypocalcemia (1.10∼1.15 mmol/L), moderate hypocalcemia (1.00∼1.10 mmol/L), and severe hypocalcemia (<1.00 mmol/L), and hypercalcemia (>1.30 mmol/L). e length of stay (LOS) in the ICU and hospital between the calcium-supplemented and nonsupplemented groups was compared using multiple comparisons. To investigate which particular situations were suitable for calcium supplementation, each interval in the stratified analysis was evaluated according to the intersections of the mortality curves of the two matching groups. e SOFA scores were categorized into the following intervals: ≤4, 5∼7, and ≥8. iCa was categorized into the following intervals: ≤1.00, 1.01∼1.20, and >1.20 mmol/L. e survival analysis for the calcium-supplemented and nonsupplemented cohorts in each interval was performed using landmark survival analysis to determine the effect of calcium supplementation on hospital mortality. e Kaplan-Meier survival curves are depicted.
Statistical analyses were performed using IBM SPSS software (version 25). Two-tailed p < 0.05 indicated a statistically significant difference.  (Table 3). ese results suggest that hypocalcemia is common in sepsis, and the decrease in iCa at the first ICU admission poses an increased risk of mortality.

Longer Hospitalization in Septic Patients with Calcium Supplementation.
ere were 2,689 (46.68%) patients who were admitted and received calcium supplementation and 3,072 (53.32%) patients who were admitted and did not receive calcium supplementation during the ICU stay. After propensity score matching according to age, sex, the SOFA score, lactate at the first ICU admission, and other complications associated with mortality in sepsis patients, 993 pairs of patients were ultimately included in the analysis.  Overall, the ICU LOS was significantly longer in the calcium-supplemented group than in the nonsupplemented group ( Table 4). We stratified patients according to the SOFA score and iCa at the first ICU admission to examine whether the effect of calcium supplementation on hospitalization varied according to disease severity. e results showed that regardless of the SOFA score and iCa, most ICU (Figure 2(a) and Figure 3(a)) and hospital (Figures 2(b) and 3(b)) stays were longer in the calcium-supplemented group than in the nonsupplemented group, but some intervals were not significantly different between the matched groups.

Calcium Supplementation Exerts Positive and Negative
Effects on the Mortality of Septic Patients. We compared the  mortality curves between the calcium supplement group and the nonsupplement group. e mortality of the matching groups in each iCa interval is shown in Table 5. Overall, the 28-day and hospital mortality in the calcium-supplemented group were higher than those in the nonsupplemented group (17.03% vs. 15.96% and 18.32% vs. 16.25%, respectively). e mortality curve for the 28-day (Figure 2(c)) and hospital (Figure 2(d)) patients showed a U-shaped curve, which suggests that an iCa level that is too high or too low increases the mortality risk for septic patients. However, the minimum of this curve was located in the mild hypocalcemia interval and not in the normal iCa range, and administration of calcium supplementation during this interval would increase mortality. is hypothesis was confirmed in the landmark analysis, which showed that administration of calcium supplementation to patients with an iCa level within 1.01∼1.20 mmol/L yielded a lower survival rate at the later stage of disease (p � 0.040) (Figure 4). ese results suggest that mild hypocalcemia is protective. In contrast, a lower mortality rate was observed when iCa was <1.01 or >1.20 mmol/L. e 28-day (Figure 3(c)) and hospital (Figure 3(d)) mortality progressively increased with increasing SOFA scores in the matching groups that were stratified by the SOFA score. e 28-day and hospital mortality rates between the matching groups were not significantly different after propensity score matching (13.49% vs. 14.30%, p � 0.818, 14.5% vs. 14.6%, p � 0.949, respectively)  Critical Care Research and Practice (Table 4). However, the results of the survival analysis showed that calcium supplementation reduced mortality when the patient's SOFA score was ≥8 (p � 0.002), and higher mortality was observed with calcium supplementation when the SOFA score was ≤4 (p � 0.010). ere was no significant difference when the SOFA score was within the range of 5∼7 (p � 0.911) ( Figure 5), which suggested that calcium supplementation has the opposite effect on mortality in patients with different disease severities.

Discussion
Treatment of hypocalcemia in septic patients remains controversial. e international guidelines developed by the Surviving Sepsis Campaign provided no recommendations for calcium administration as a therapeutic measure [26]. Evidence-based guidance is lacking because no randomized controlled trials have investigated whether calcium supplementation should be administered to septic patients [13]. Unlike previous studies, our investigation assessed the prognosis of septic patients with or without calcium supplementation using stratified analyses of iCa and SOFA scores. We found that calcium supplementation exerted positive and negative effects on mortality for different SOFA score intervals, which may be clinically valuable because the SOFA score could be a substitute for iCa and could help clinicians make decisions regarding calcium supplementation when iCa measurements are unavailable. e mortality curve in the nonsupplemented patients stratified by the iCa interval showed a U-shaped curve, which suggests that abnormal iCa poses a risk to septic patients. Because hypocalcemia is more prevalent in septic patients than hypercalcemia (in our case, 70.12% vs. 3.30%) and patients with severe hypocalcemia require critical care   e LOS in the ICU (a) and hospital (b) in the calcium supplement group was longer than that in the nonsupplemented group. e data for each interval are expressed as the median with the interquartile range and connected using the LOWESS smoothing technique (Mann-Whitney U test, * p < 0.05; * * p < 0.01; * * * p < 0.001; * * * * p < 0.0001, and n.s. not significant); higher 28-day (c) and hospital (d) mortality were found in the calcium supplement group when iCa was within 1.04∼1.17 mmol/L, and lower 28-day (c) and hospital (d) mortality were found in the calcium supplement group when iCa was <1.04 or >1.17. e data for each interval are expressed as percentages and connected using the LOWESS smoothing technique.
for a longer period [2], hypocalcemia may be another prognostic marker of sepsis. However, we found that the minimum point of this mortality curve was located at the mild hypocalcemia interval and not in the normal iCa range, which suggests that patients maintained under mild hypocalcemia may receive a benefit in terms of outcomes. Our recent experiment using a septic model also showed that mice pretreated with EDTA-2Na before cecal ligation and  Figure 3: LOS and mortality in septic patients with or without calcium supplementation stratified by the SOFA score. e LOS in the ICU (a) and hospital (b) in the calcium-supplemented group was higher than that in the nonsupplemented group. e data for each interval are expressed as the median with the interquartile range (Mann-Whitney U test, * p < 0.05, * * p < 0.01, * * * p < 0.001, * * * * p < 0.0001, and n.s. not significant). A higher 28-day (c) and hospital (d) mortality were found in the calcium supplement group when the patient's SOFA score was ≥8, and a lower 28-day (c) and hospital (d) mortality were found in the calcium supplement group when the patient's SOFA score was ≤4.
e data for each interval are expressed as percentages (chi-squared test, * p < 0.05, and * * p < 0.01). Critical Care Research and Practice 7 puncture (CLP) had 20∼50% lower mortality than those in the non-EDTA-2Na treatment group (unpublished data). Despite these findings, whether changes in blood iCa are a protective mechanism or simply a consequence of metabolic dysregulation when the body undergoes a critical illness must be established. If it is protective, then calcium supplementation may increase the burden on the body, which would downregulate iCa and worsen the outcome. Overall, the ICU and hospital LOS in septic patients with calcium supplementation were longer than those in patients who did not receive supplementation. To examine whether the LOS varied across different subranges of iCa and disease severities, we stratified the LOS according to iCa and SOFA scores. However, the LOS tended to be higher in the calcium treatment groups for most subranges, particularly the ICU LOS, which suggests that calcium supplementation contributes to longer hospital stays. A similar phenomenon was seen in critically ill patients receiving calcium supplementation [14]. However, patients who stayed longer in the ICU may have a greater chance of receiving calcium supplementation and a higher likelihood of experiencing hypocalcemia. We could not exclude this possibility because causality could not be determined in our case. A large randomized controlled trial exploring the use of calcium supplementation in septic or ICU patients with different degrees of hypocalcemia is required to exclude this bias.
Unlike hospitalization, the two mortality curves crossed, which suggest that calcium supplementation exhibits the opposite effect on mortality on a case-by-case basis. It would be valuable for clinical decision-making to determine which situation is suitable for calcium supplementation. erefore, the intersection of the mortality curves was calculated, and  Critical Care Research and Practice we found that patients receiving calcium supplementation tended to have higher hospital mortality when their iCa level at the first ICU admission was approximately 1.01∼1.20 mmol/L, which suggests that the attempted correction of blood calcium at this interval may be harmful.
Most iCa levels at the first ICU admission for septic patients were located within this iCa interval, which may explain why some studies revealed that calcium supplementation had an adverse effect on the prognosis of septic patients [16,27] and in septic models [17,28]. However, there was a trend showing that calcium supplementation decreased mortality when iCa at the first ICU admission was <1.01 mmol/L or >1.20 mmol/L. However, we could not confirm this finding in the survival analysis, which may be due to the small number of patients in these subgroups. e double effect of calcium supplementation was significant when the patients were stratified according to the SOFA score. Because iCa is fast-changing and not routinely measured in clinical practice, using the SOFA score as a substitute reference index for clinical calcium supplement decision-making may be valuable, particularly when iCa is not available. e incidence of severe hypocalcemia tends to be more pervasive in severe septic cases, and a negative correlation between the iCa concentration and mortality was observed in our study and previous studies [9,10]. Moritoki et al. reported that the iCa concentration was not independently associated with the mortality, and only extreme iCa abnormalities were independent predictors of mortality [10]. Consistent with these results, calcium supplementation may benefit the outcome when the SOFA score is ≥8 and when patients experience severe hypocalcemia. e underlying mechanism of the positive and negative effects of calcium supplementation in septic patients has not been well established. e administration of calcium supplementation to patients with severe hypocalcemia may prevent life-threatening complications, such as cardiac arrest or seizures, which may explain why patients with iCa levels <1.00 mmol/L had the highest mortality compared to patients with other iCa ranges. iCa also plays an important role in maintaining hemodynamics [29], and a direct relationship between iCa and arterial pressure was found in critically ill patients [30]. Calcium supplementation may improve hemodynamics by increasing the mean arterial pressure (MAP), the left ventricular stroke work index, and CO in critically ill patients [31,32] and may improve heart function [33]. However, calcium supplementation may be deleterious at the cellular level because iCa may be shuttled into cells [34][35][36].
erefore, the administration of additional parenteral calcium may aggregate the accumulation of cytosolic iCa concentrations, which would overactivate certain pathways and generate reactive oxygen species that trigger cell death [37]. Studies using calcium blockers as a treatment for sepsis showed an improvement in patient outcome [38][39][40], which supports this hypothesis. erefore, the advantages and disadvantages should be weighed before calcium supplementation is administered to the patient. Notably, our study showed that calcium supplementation for patients with an iCa level at the first ICU admission >1.20 mmol/L tended to decrease mortality because excessive iCa levels are also a risk factor for increased mortality in septic patients. Notably, although hypocalcemia is very common in these patients, some do not present with hypocalcemia despite relatively high SOFA scores. ese patients may not have an iCa influx problem. erefore, it would be valuable to examine the contribution of calcium supplementation to hemodynamics in these individuals.
ere are several limitations in this study. First, it was retrospective and purely observational. Some patients who may have received calcium supplementation or treatments such as fast transfusion that would affect blood iCa levels prior to admission to the ICU were not included in the analysis because the MIMIC-III is an ICU database. Second, the formula, dose, timing, and duration of calcium administration are important factors that influence the therapeutic effect of calcium replacement. Our approach did not investigate these factors because there were 5 types of calcium replacement agents in use in the MIMIC-III database, and hypocalcemia may have occurred at any time during their ICU stay, which led to the variation in the administration, time of administration and duration of administration from case to case. erefore, our study only preliminarily analyzed whether calcium supplementation should be administered and under which conditions it is suitable for ICU septic patients. ird, an immortal time bias exists because patients who stayed longer in the ICU had a greater chance of experiencing hypocalcemia and receiving calcium supplementation. Although a longer hospitalization was noted in the calcium supplementation group, the causality between longer hospitalization and calcium supplementation could not be determined in our study. Fourth, calcium agents can be used for hypocalcemia therapy and other purposes and we did not identify the calcium agent used for hypocalcemia therapy or other purposes in the MIMIC-III database. Fifth, the dynamics of iCa may be fastchanging and hypocalcemia and hypercalcemia may occur in the same patient during their ICU stay [10]. Our study focused on iCa at the first ICU admission, which may be inadequate. However, we found that it was significantly correlated with mortality in septic patients. Finally, our study presented only a rough reference range of iCa and SOFA scores for calcium supplementation, and the sample size of patients with iCa <1.01 or >1.20 mmol/L was small. erefore, the study was underpowered to produce convincing conclusions for this subgroup. Larger amounts of data are needed, and controlled intervention studies should be performed to assess the range in which this intervention should be applied for clinical decision-making.

Conclusions
Septic patients are at particular risk for hypocalcemia, and the effect of calcium administration on septic patients with different disease severities is largely unknown. We extracted the data from septic patients admitted to the ICU from the MIMIC-III database and compared the prognosis between calcium-supplemented and nonsupplemented patients. Our findings suggest that mild hypocalcemia is protective in septic patients. ere were positive and negative effects of calcium supplementation on mortality in septic patients with different SOFA score intervals. erefore, the SOFA score may be a valuable clinical index for decisions regarding calcium administration, particularly when iCa is not available. Further controlled intervention studies should be performed to verify the advantages and disadvantages of this intervention and apply them to clinical decision-making.
Data Availability e calcium supplementation and noncalcium supplementation patient datasets in this study are available after requirement since the Excel files cannot be uploaded in this submission platform.

Disclosure
is manuscript has been uploaded as a preprint on the preprint platform named Research Square (https://www. researchsquare.com/article/rs-16486/v1).