Non-linear relationship between ionised calcium and 28-day mortality in patients with sepsis: a retrospective cohort study from MIMIC-IV database

Abstract

Objectives

Sepsis is a life-threatening disease caused by a deregulated response to infection. This study aimed to investigate the linear and non-linear relationships between ionised calcium levels and 28-day mortality in patients with sepsis in the intensive care unit (ICU) and to generate hypotheses regarding laboratory index testing and calcium supplementation strategies.

Design

A retrospective cohort study.

Setting

Adult patients admitted to the ICU, using data extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database.

Participants

A total of 17 955 adult patients with sepsis, defined according to the sepsis-3 criteria, were included. Patients lacking ionised calcium (iCa) data were excluded.

Outcome measures

The exposure variable was iCa level measured within the first 24 hours of ICU admission, and the primary outcome was 28-day all-cause mortality. Covariates included demographics, vital signs, comorbidities, infection sites, organ support and medication use. Multivariable logistic regression models were applied with stepwise adjustment. Non-linear associations were assessed using generalised additive models (GAM) and two-piecewise linear regression. Sensitivity analyses using iCa quartiles and subgroup analyses stratified by Sequential Organ Failure Assessment (SOFA) score were also performed.

Results

The overall 28-day mortality rate was 18.3%. After full adjustment, patients in quartiles Q2–Q4 had significantly lower mortality risks than those in Q1 (P for trend <0.001). A U-shaped association was observed, with an inflection point at 1.16 mmol/L (95% CI 1.12 to 1.19). Below this threshold, each 0.1 mmol/L decrease in iCa was associated with a 13% higher risk of mortality (OR=0.87; 95% CI 0.82 to 0.92; p<0.001); whereas, above this point, mortality increased by 6% per 0.1 mmol/L increment (OR=1.06; 95% CI 1.00 to 1.10; p=0.004). Subgroup analyses stratified by SOFA score (<10 vs ≥10) confirmed the robustness of the U-shaped relationship (p>0.05).

Conclusion

The relationship between ionised calcium and 28-day mortality in patients with sepsis showed a U-shaped curve with an inflection point of 1.16 mmol/L, which is in the range of mild hypocalcaemia.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Large sample size of 17 955 sepsis patients from the well-validated Medical Information Mart for Intensive Care IV database provides robust statistical power and generalisability.

• Advanced statistical methods including generalised additive models and two-piecewise linear regression effectively identified non-linear U-shaped relationships.

• Comprehensive adjustment for multiple confounders including demographics, comorbidities, organ support and medications minimised the risk of bias and supported cautious causal interpretation within the limits of observational data.

• Retrospective observational design limits ability to establish causality between ionised calcium levels and mortality outcomes.

Introduction

Sepsis, a life-threatening organ dysfunction caused by a deregulated host response to infection,¹ accounts for approximately 19.7% of all global deaths.² Studies estimate that there are approximately 31.5 million new cases of sepsis and 5.3 million deaths worldwide each year, making it a major global public health problem.³

Despite extensive efforts to discover new treatments and understand the underlying causes of sepsis, the mortality rate associated with this condition has not decreased.^{4 5} One challenge is the lack of reliable predictors for sepsis, which hinders accurate prediction of prognosis and leads to inadequate statistical power in clinical trials for new treatments. Therefore, it is crucial to explore simple and effective biomarkers to determine the prognosis of sepsis.

Abnormal serum calcium levels may be a marker of disease severity in critical illness.⁶,⁸ Hypocalcaemia has been frequently observed in sepsis syndrome,⁹ and several studies have reported a statistically significant negative correlation between serum calcium levels and mortality rates in patients diagnosed with sepsis.¹⁰,¹² However, other studies have reported that only extreme outliers of calcium levels are associated with sepsis mortality.^{13 14}

We hypothesise that inconsistencies in the findings of the aforementioned studies may be due to the following: (1) Some of the exposure factors in the studies were total calcium and some were ionised calcium (iCa). Approximately 50% of the total calcium is iCa, which is biologically active and less susceptible to other factors (eg, albumin levels); therefore, iCa is preferred for assessment if available. (2) Small sample sizes and inadequate adjustment for confounding factors resulted in less convincing conclusions.

The relationship between ionised calcium and sepsis mortality needs to be further explored. Therefore, this study used the extensive US multicentre sepsis database, Medical Information Mart for Intensive Care (MIMIC-IV), to investigate the association between iCa and the 28-day mortality risk in sepsis.

The substantial sample size and extensive database information provided valuable statistical power, ensuring more robust and dependable findings that enhance our understanding of the relationship between iCa and the 28-day mortality risk in sepsis.

Methods

Database

This study investigated health-related data obtained from the MIMIC-IV database,¹⁵ which consists of extensive, high-quality medical records of patients at the Beth Israel Deaconess Medical Center (BIDMC), and was developed and managed by the Massachusetts Institute of Technology (MIT) Computational Physiology Laboratory. The database is available for free download on completion of an accredited course on its official website. Author (Lu Chen (RECORD 35931520)) obtained access to the database and was responsible for data extraction. This study complied with the reports of studies conducted using the observation routine collected health data (RECORD) statement.

Study population

In total, 377 207 adult patient records were found in the MIMIC-IV database. Sepsis was diagnosed according to sepsis-3 criteria.¹ Sepsis-relevant International Classification of Diseases, Ninth Revision (ICD-9) codes (99 591—99 592), or International Classification of Diseases, Tenth Revision (ICD-10) codes (R652, R6520 and R6521).^{16 17} The outcome variable was death from any cause during 28-day after intensive care unit (ICU) admission, and ionised calcium was the exposure variable (recorded as a continuous variable). We extracted iCa data at ICU admission. Patients with reported malignancy or missing exposure variable information were excluded from this study.

Data collection

All collected variables were used both for descriptive statistics of baseline characteristics and for adjustment of confounding. The selection of covariates was based on the following principles: (1) established predictors of sepsis prognosis; (2) factors previously reported in the literature to be associated with sepsis mortality; and (3) clinical expertise regarding potential confounders that may be related to both ionised calcium levels and patient outcomes. This comprehensive covariate adjustment strategy was designed to control for potential confounding biases and ensure that the estimated association between ionised calcium and 28-day mortality is as accurate and unbiased as possible. Potential variables included: demographic factors (age, gender, ethnicity), vital signs (heart rate, respiratory rate, body temperature, blood pressure), scoring system (Charlson Comorbidity Index, Sequential Organ Failure Assessment (SOFA)), life support (mechanical ventilation, dialysis), the site of infection (respiratory system, central nervous system, urinary system, liver), use of glucocorticoids (dexamethasone, methylprednisolone, cortisol), use of vasoactive drugs (dopamine, dobutamine, norepinephrine), use of antibiotics (carbapenem, cephalosporins, penicillin-based antibiotics, vancomycin) and use of intravenous gamma globulin. All variables were extracted at ICU admission.

Statistical analysis

Continuous variables are expressed as mean±SD (Gaussian distribution) or median (minimum, maximum) (skewed distribution), and categorical variables are expressed as rates. One-way analysis of variance (ANOVA) (Gaussian distribution), Kruskal-Wallis H (skewed distribution) test and χ² test (categorical variables) were used to determine any statistical differences between the means and proportions of the groups.

To investigate whether iCa was independently associated with 28-day mortality rate of sepsis patients, our statistical analysis consisted of three steps. First, we adopted univariate and multivariate logistic regression analyses to examine the association between iCa and 28-day mortality. We employed three models with sequential degrees of adjustment: an unadjusted model (Model 0, no covariates adjusted), a minimally adjusted model (Model 1, adjusted only for demographic factors: age, gender, ethnicity) and a fully adjusted model (Model 2). The fully adjusted model included all covariates presented in table 1 to control for potential confounding biases, encompassing: (1) demographic characteristics: age, gender, ethnicity; (2) vital signs: heart rate, respiratory rate, body temperature, systolic blood pressure, diastolic blood pressure; (3) scoring systems: Charlson Comorbidity Index, SOFA score; (4) life support: mechanical ventilation, dialysis; (5) infection sites: respiratory system, central nervous system, urinary system, liver; (6) medication use: glucocorticoids (dexamethasone, methylprednisolone, hydrocortisone), vasoactive drugs (norepinephrine, dopamine, dobutamine), antibiotics (carbapenems, cephalosporins, penicillin-based antibiotics, vancomycin); and (7) other treatments: intravenous gamma globulin. The selection of these covariates was based on clinical expertise and literature reports, as they are recognised as significant predictors of 28-day mortality in septic patients. Adjusting for these variables aimed to minimise confounding bias and obtain a more accurate and unbiased estimate of the association between ionised calcium levels and 28-day mortality. All models reported OR with corresponding 95% CIs. To better control potential non-linear confounding effects of continuous covariates, we performed an additional generalised additive model (GAM)-adjusted analysis. In this analysis, ionised calcium was treated as a categorical variable (quartiles), while continuous covariates (such as age, SOFA score and Charlson Comorbidity Index) were adjusted using the smoothing function s(). This method allows flexible adjustment for non-linear relationships and provides more accurate effect estimates.

Table 1. Description of baseline characteristics of patients with sepsis.

iCa (mmol/L)	Q1 (0.21–1.04）	Q2 (1.05–1.12)	Q3 (1.13–1.17)	Q4 (1.18–3.75)	P value
N, n (%)	3979 (22.16%)	4867 (27.11%)	4351 (24.23%)	4758 (26.50%)
Demographics
Age, median (minimum, maximum)	64.21 (18.13–99.16)	67.26 (18.51–100.29)	67.63 (18.16–100.01)	67.89 (18.10–97.43)	<0.001
Gender (male), n (%)	2237 (56.22%)	2939 (60.39%)	2668 (61.32%)	2813 (59.12%)	<0.001
Ethnicity (White), n (%)	2456 (61.72%)	3210 (65.95%)	2952 (67.85%)	3261 (68.54%)	<0.001
Vital signs
Systolic pressure	112.73±15.12	114.63±14.87	115.15±14.73	114.41±13.62	<0.001
Diastolic pressure	60.84±10.37	60.44±10.07	60.48±10.01	59.63±9.48	<0.001
Heart rate, mean±SD, times/minute	91.02±17.04	87.40±15.96	85.99±15.79	85.57±14.34	<0.001
Respiratory rate, mean±SD, times/minute	26.97±10.05	26.48±9.77	26.27±9.88	25.91±9.77	<0.001
Body temperature, mean±SD, degrees Celsius	36.86±0.76	36.91±0.68	36.89±0.63	36.82±0.59	<0.001
Scoring system
Charlson Comorbidity Index, mean±SD	5.45±2.86	5.59±2.79	5.49±2.64	5.61±2.53	0.020
SOFA score, mean±SD	8.59±4.41	7.26±3.82	7.09±3.74	7.06±3.68	<0.001
The site of infection
Respiratory system	1379 (34.66%)	2071 (42.55%)	1965 (45.16%)	2366 (49.73%)	<0.001
Central nervous system	723 (18.17%)	1044 (21.45%)	981 (22.55%)	1013 (21.29%)	<0.001
Urinary system	1741 (43.75%)	1794 (36.86%)	1473 (33.85%)	1589 (33.40%)	<0.001
Liver	824 (20.71%)	637 (13.09%)	478 (10.99%)	442 (9.29%)	<0.001
Life support
Mechanical ventilation, n (%)	2582 (64.89%)	3009 (61.82%)	2716 (62.42%)	3032 (63.72%)	0.015
Dialysis, n (%)	542 (13.62%)	368 (7.56%)	294 (6.76%)	318 (6.68%)	<0.001
Drugs
Glucocorticoids
Dexamethasone, n (%)	278 (6.99%)	335 (6.88%)	298 (6.85%)	256 (5.38%)	0.004
Methylprednisolone, n (%)	635 (15.96%)	746 (15.33%)	646 (14.85%)	698 (14.67%)	0.349
Hydrocortisone, n (%)	76 (1.91%)	87 (1.79%)	63 (1.45%)	67 (1.41%)	0.168
Vasoactive drugs
Norepinephrine, n (%)	1998 (50.21%)	1981 (40.70%)	1609 (36.98%)	1531 (32.18%)	<0.001
Dopamine, n (%)	384 (9.65%)	393 (8.07%)	325 (7.47%)	310 (6.52%)	<0.001
Dobutamine, n (%)	291 (7.31%)	280 (5.75%)	197 (4.53%)	183 (3.85%)	<0.001
Antibiotics
Carbapenems, n (%)	897 (22.54%)	1006 (20.67%)	821 (18.87%)	881 (18.52%)	<0.001
Cephalosporins, n (%)	314 (7.89%)	351 (7.21%)	340 (7.81%)	371 (7.80%)	0.582
Penicillin-based antibiotics, n (%)	2177 (54.71%)	2449 (50.32%)	2047 (47.05%)	1943 (40.84%)	<0.001
Vancomycin, n (%)	3377 (84.87%)	4043 (83.07%)	3539 (81.34%)	3679 (77.32%)	<0.001
Intravenous gamma globulin, n (%)	84 (2.11%)	75 (1.54%)	62 (1.42%)	66 (1.39%)	0.030
28-day mortality, n (%)	1051 (26.41%)	854 (17.55%)	671 (15.42%)	710 (14.92%)	<0.001

All variables were extracted at ICU admission.

iCa, ionised calcium; SOFA score, The Sequential Organ Failure Assessment Score.

If a non-linear association was observed, we applied a recursive algorithm based on maximum likelihood estimation to identify the optimal threshold. The procedure included: (1) coarse positioning: candidate threshold values (K) were evaluated from the 5th to 95th percentile of iCa at 5% intervals (19 points). The percentile yielding maximum likelihood was identified as P1, and the search range was narrowed to P1±4%. (2) Fine positioning: within this range, a recursive method compared likelihoods at Q1 (25th), Q2 (50th) and Q3 (75th) percentiles, selecting the optimal point and restricting the search to 25% around it. This process was repeated, excluding 50% of values each iteration, until the optimal K was found. (3) Validation: Bootstrap resampling (1000 iterations) determined the 95% CI of the threshold. Both Wald and likelihood ratio tests confirmed the threshold effect by comparing piecewise vs linear models.

Finally, to ensure result reliability, we conducted sensitivity analyses: (1) quartile analysis: patients were divided into four equal groups based on iCa quartile distribution rather than clinical thresholds, as most sepsis patients had below-normal iCa levels. Quartile grouping ensured more equal distribution of patients across groups and greater statistical power to detect mortality risk differences. (2) Subgroup analysis: given the close relationship between SOFA scores and sepsis prognosis, patients were stratified by SOFA (<10 vs ≥10).^{18 19} Smooth curve fitting examined whether the iCa-mortality association remained consistent across different disease severity levels.

All analyses were performed using the statistical packages R (http://www.R-project.org, The R Foundation) and EmpowerStats (http://www.mpowerstats.com, X&Y Solutions, Inc, Boston, MA). P values <0.05 (two-sided) were considered statistically significant.

Patient and public involvement

None.

Results

Screening process

In this study, we enrolled 377 207 patients from the MIMIC-IV database. Among these, 342 197 were non-septic patients, while 14 593 had missing information on iCa and 2462 with malignant tumour. Consequently, 17 955 cases were considered for the final analysis. The flow chart of patient selection is depicted in figure 1.

Figure 1. Flow chart of patient selection. MIMIC-IV, Medical Information Mart for Intensive Care IV.

Baseline characteristics

Patient baseline characteristics are summarised in table 1.

iCa levels in the entire population were segmented into four groups (Q1 to Q4) based on quartile divisions. Trends in the distribution of each variable across different subgroups were observed following this segmentation. Analysis revealed that, relative to the Q1 group, patients in the Q2, Q3 and Q4 groups were typically older, had a higher representation of white ethnicity, a higher Charlson Comorbidity Index, and more frequent use of dexamethasone and dopamine. These groups also exhibited lower SOFA scores, heart rate and respiratory rate. They received a lower percentage of treatments such as dexamethasone, vasoactive drug, antibiotics, intravenous gamma globulin, mechanical ventilation and dialysis. They demonstrated a lower 28-day mortality rate. The 28-day mortality rate for patients with sepsis was 18.3%.

Relationship between iCa and 28-day mortality in patients with sepsis using non-adjusted and adjusted models

Various covariate adjustment strategies were employed to elucidate the relationship between iCa and 28-day mortality in patients with sepsis. The non-adjusted and adjusted models are detailed in table 2. The GAM covariate-adjusted analysis further confirmed the robustness of the quartile results. After non-linear adjustment of continuous covariates, the trend of reduced mortality risk in Q2–Q4 compared with Q1 remained significant, indicating that our findings were not influenced by non-linear relationships of continuous variables.

Table 2. Results of univariate and multivariable logistic regression analysis (adjusted for per 0.1 mmol/L increase in iCa where applicable).

Exposure	Model 0 (n=17 955） OR (95% CI), p value	Model 1 (n=17 900） OR (95% CI), p value	Model 2 (n=17 300） OR (95% CI), p value	GAM (n=17 300） OR (95% CI), p value
iCa (mmol/L) (per 0.1 mmol/L increase)	0.81 (0.79 to 0.84) <0.0001	0.79 (0.77 to 0.82) <0.0001	0.94 (0.91 to 0.98) <0.0001	0.96 (0.92 to 0.99) 0.015
iCa quartile
Q1	Reference	Reference	Reference	Reference
Q2	0.59 (0.54 to 0.66) <0.0001	0.57 (0.51 to 0.63) <0.0001	0.80 (0.70 to 0.90) 0.0002	0.79 (0.70 to 0.90) 0.0003
Q3	0.51 (0.46 to 0.57) <0.0001	0.48 (0.43 to 0.54) <0.0001	0.73 (0.64 to 0.83) <0.0001	0.74 (0.65 to 0.84) <0.0001
Q4	0.49 (0.44 to 0.54) <0.0001	0.46 (0.41 to 0.51) <0.0001	0.75 (0.66 to 0.86) <0.0001	0.79 (0.69 to 0.90) 0.0004
P for trend	<0.0001	<0.0001	<0.0001	0.0002

Model 0: no covariates were adjusted for.

Model 1: we adjusted only for sex, age and ethnicity.

Model 2: we adjusted for all covariates presented in table 1: age, gender, ethnicity, heart rate, respiratory rate, body temperature, systolic blood pressure, diastolic blood pressure, Charlson Comorbidity Index, SOFA score, mechanical ventilation, dialysis, respiratory system, central nervous system, urinary system, liver, glucocorticoids (dexamethasone, methylprednisolone, hydrocortisone), vasoactive drugs (norepinephrine, dopamine, dobutamine), antibiotics (carbapenems, cephalosporins, penicillin-based antibiotics, vancomycin) and intravenous gamma globulin.

GAM: all variables listed in Model 2 were adjusted. However, continuous variables were adjusted as non-linearity.

GAM, generalised additive model; iCa, ionised calcium.

The unadjusted model showed that each 0.1 mmol/L increase in iCa was associated with a 19% reduction in 28-day mortality risk (OR: 0.81, 95% CI 0.79 to 0.84, p<0.0001). After adjusting for sex, age and ethnicity (Model 1), the association remained stable (OR: 0.79, 95% CI 0.77 to 0.82, p<0.0001). In the fully adjusted model (Model 2), which incorporated all covariates listed in table 1, the protective effect was attenuated but still significant, with a 6% reduction in mortality risk per 0.1 mmol/L increase in iCa (OR: 0.94, 95% CI 0.91 to 0.98, p<0.001).

To further verify robustness and provide a clinically interpretable stratification, iCa was also analysed as a categorical variable (quartiles). Compared with Q1, patients in Q2–Q4 consistently showed significantly lower mortality risk, and the overall trend remained highly significant (P for trend <0.0001). Notably, in the fully adjusted model (Model 2), the ORs for Q2–Q4 did not decrease in a strictly linear fashion with increasing iCa but instead suggested a non-linear pattern of risk reduction. This observation indicates that the association between iCa and 28-day mortality may not follow a simple linear gradient. To further address this, we applied a GAM model with non-linear adjustment of continuous covariates, and a similar non-linear trend across quartiles was also observed. This suggests that our findings are unlikely to be biased by non-linear associations of covariates.

Non-linear relationship between iCa and 28-day mortality

A generalised additive model and smoothed curve fitting were used to examine the non-linear relationship between iCa and 28-day mortality. The findings suggested a U-shaped association between iCa and 28-day mortality after adjusting for covariates (The adjustment strategy aligns with Model 2 in table 2) (refer to figure 2).

Figure 2. Correlation between iCa levels and 28-day mortality probability in sepsis patients. The solid line represents the smoothed curve, while the dashed line indicates the 95% CI. iCa, ionised calcium.

Table 3 presents the threshold effect analysis of iCa and 28-day mortality in sepsis patients, using the same covariate adjustment strategy as Model 2 in table 2. A standard logistic regression model suggested that each 0.1 mmol/L increase in iCa was associated with a 6% reduction in mortality risk (OR: 0.94, 95% CI 0.91 to 0.98, p<0.0001). However, when a two-piecewise linear regression model was applied, a clear inflection point was identified at 1.16 mmol/L (95% CI 1.12 to 1.19). Below the inflection point (iCa≤1.16 mmol/L), each 0.1 mmol/L decrease in iCa was associated with a 13% higher risk of mortality (OR: 0.87, 95% CI 0.82 to 0.92, p<0.0001). Above the inflection point (iCa>1.16 mmol/L): the association reversed, with higher iCa levels linked to a 6% increase in mortality (OR: 1.06, 95% CI 1.00 to 1.10, p=0.004). The log-likelihood ratio test was highly significant (p<0.001), supporting a non-linear association between iCa levels and 28-day mortality.

Table 3. Explanation of the non-linear association between iCa and the risk of 28-day mortality in sepsis patients (adjusted for per 0.1 mmol/L increase in iCa where applicable).

Outcome: OR (95% CI), p value	Value
Fitting model using standard logistic regression (per 0.1 mmol/L increase in iCa)	0.94 (0.91 to 0.98) <0.0001
Fitting model using two-piecewise linear model
Inflection point	1.16 (1.12 to 1.19)
≤ inflection point (per 0.1 mmol/L increase in iCa)	0.87 (0.82 to 0.92) <0.0001
> inflection point (per 0.1 mmol/L increase in iCa)	1.06 (1.00 to 1.10) 0.004
Log-likelihood ratio test	<0.001

The adjustment strategy is the same as Model 2 in table 2.

iCa, ionised calcium.

Based on the subgroup analysis presented in figure 3, a U-shaped association between basal ionised calcium levels and the risk of 28-day mortality was consistently observed in both patient groups stratified by SOFA score. This relationship remained evident regardless of disease severity, as defined by a SOFA score threshold of 10. The risk of mortality decreased with increasing ionised calcium levels up to a certain point, beyond which higher levels were associated with an elevated mortality risk.

Figure 3. Subgroup analysis of the correlation between iCa levels and 28-day mortality probability in patients with sepsis. The SOFA score of 10 was used as the cut-off to divide the patients into two groups. The solid line represents patients with the SOFA score less than 10, while the dashed line represents patients with the SOFA score of 10 or greater. iCa, ionised calcium; SOFA, the Sequential Organ Failure Assessment Score.

Discussion

In the present study, we evaluated the association between iCa levels and 28-day mortality in patients with sepsis from the MIMIC-IV database. The results indicate that iCa levels are independently associated with 28-day mortality in patients with sepsis. They remained strongly correlated even after correction for confounding risk factors. Importantly, by GAM modelling and smoothed curve fitting, a U-shaped relationship between iCa levels and 28-day mortality in patients with sepsis was found. Subgroup analysis based on SOFA score consistently revealed a U-shaped relationship between ionised calcium and 28-day mortality. This pattern was observed both in patients with lower illness severity (SOFA <10) and in those more critically ill (SOFA ≥10). The consistency of this association across severity groups suggests that ionised calcium may serve as a reliable prognostic marker in sepsis, supporting its potential utility in routine clinical assessment.

The rapid progression and high mortality of sepsis imply the necessity of identifying prognostic indicators associated with its outcome. It is recommended that scholars further explore serum markers and prognostic correlates to aid early clinical intervention.^{20 21} Calcium is an essential trace element in the human body that is responsible for physiological functions including cell signalling, neurotransmission and muscle contraction.^{22 23} Calcium disorders are uncommon in the general population, except in special cases (eg, renal failure), and are therefore often neglected.^{24 25} Nevertheless, calcium derangements in sepsis are prevalent and should be taken seriously.^{26 27} In a quiescent and healthy state, calcium levels are regulated by plasma membrane transport proteins and sequestered in cellular organelles apart from mitochondria. Nevertheless, intense external stressors, including cellular injury, oxidative stress or inflammation, may induce an overflow of cytosolic calcium ions, which are then buffered by the cell’s mitochondria.^{28 29} Excess calcium within mitochondria can harm these structures, leading to the release of mitochondrial components into the bloodstream.³⁰ This can cause uncontrolled autophagy and additional cellular damage. Energy depletion due to mitochondrial dysfunction induced by critical illnesses may result in disrupted calcium balance.

Total calcium levels are affected by a variety of factors, such as hypoalbuminaemia, and may be less accurate (in the presence of renal disease, acid-base disturbances or anionic perturbations), even after adjustment by algorithms.³¹,³³ In these cases, iCa measurement may be preferred if feasible and can improve more accurate judgement for the clinician; however, there is a paucity of high-quality studies on iCa and sepsis. In this study, we found that iCa was negatively associated with 28-day mortality from sepsis when it was less than 1.16 mmol/L. A retrospective study on elderly patients with sepsis showed that the serum calcium levels of septic patients were lower than those of the control group and that the severity of sepsis was associated with lower serum calcium levels.¹⁰ A retrospective study on paediatric sepsis also revealed that among children with sepsis, those with low blood calcium levels had a higher incidence of organ dysfunction than those with normal blood calcium levels.¹¹ These studies increase our confidence in our findings.

The normal range of iCa in adults is 1.20–1.40 mmol/L (4.8–5.6 mg/dL);⁶ nevertheless, in this study, we found that iCa was below the normal range in most septicemic patients. Furthermore, sepsis mortality was lowest at an iCa level of 1.16 mmol/L, which is slightly below the normal reference range for iCa. This suggests that mild hypocalcaemia may be protective against sepsis. A retrospective study found that mild hypocalcaemia may have a protective effect in septic patients, and calcium supplementation may have both positive and negative effects on mortality depending on the severity of the disease.³⁴ Aberegg also suggested that low iCa levels may be protective in critically ill patients; whereas, attempts to correct low levels may be harmful.⁶ Our findings are partially consistent with the perspectives of the aforementioned researchers. In a sepsis animal model, calcium supplementation was associated with decreased survival; whereas, no deaths were observed in hypocalcaemia animals.³⁵ Whether low iCa is a protective mechanism or a consequence of metabolic imbalance in sepsis requires further mechanistic studies. Because of the complex and wide-ranging effects of iCa, this result should be interpreted cautiously.

Severe hypercalcaemia is associated with an increased risk of death, which is attributable in most cases to primary hyperparathyroidism or malignancy.³⁶,³⁸ However, by exploring the non-linear relationship between iCa and sepsis, we found a significant positive correlation between elevated iCa levels and short-term mortality in patients with sepsis once iCa was greater than 1.16 mmol/L, even if severe hypercalcaemia levels were not reached (iCa<2.5 mmol/L). The mechanism, although not fully delineated, may be due to several factors, including (1) cardiovascular risk: elevated iCa can lead to increased cardiac load, causing arrhythmias, elevated blood pressure or increased myocardial contractility, which can further disrupt cardiovascular function in patients with sepsis.³⁹,⁴¹ (2) Thrombosis formation: elevated calcium levels enhance coagulation factor activity and increase blood viscosity, thereby increasing the risk of thrombosis.⁴²,⁴⁴ (3) Immune dysregulation: hypercalcaemia impairs immune cell function and disrupts immune system regulation, which increases the severity of infection, exacerbates sepsis and increases the risk of death.^{45 46}

Several strengths characterise this investigation. First, we used advanced algorithms such as GAM, smoothed curve fitting and two-segment linear models, which allow for a more accurate determination of the curvilinear and quantitative relationship between iCa and the 28-day risk of death in patients with sepsis. Second, we used the updated MIMIC-IV database, and the exposure variables were composite indicators and more abundant. We adjusted a series of covariates, including treatment, that are closely related to sepsis. Third, our study uniquely investigates the relationship between ionised calcium levels and short-term mortality, which distinguishes it from studies focusing on total serum calcium. Ionised calcium is influenced by acid-base status changes and protein binding dynamics.^{24 47} For example, acidosis increases ionised calcium, while alkalosis decreases it. Complications such as acid-base imbalance and hypoproteinaemia often occur in critically ill patients. Therefore, ionised calcium may be a more sensitive and dynamically changing indicator. Fourth, through subgroup analysis, we further verified that the relationship between ionised calcium levels and 28-day mortality remained consistent in patients with severe sepsis of different severity. In addition, we applied GAM to adjust for continuous covariates non-linearly. Compared with traditional linear adjustment, this approach better accounts for complex relationships of continuous confounders and improves the accuracy and reliability of effect estimates.

However, several limitations should be noted. First, methodological limitations: this observational study can only establish associations rather than causal relationships between iCa and sepsis mortality and relied on single iCa measurements at ICU admission rather than serial monitoring. Second, data limitations: some key variables (eg, transfusion and fluid resuscitation) are missing in the MIMIC-IV database, and these unmeasured confounding factors may affect our final results. Additionally, some sepsis patients were excluded due to missing iCa data, potentially introducing selection bias. Third, generalisability concerns: the data primarily represent US patients, limiting applicability to other populations. These limitations suggest our results should be interpreted cautiously and serve as a foundation for future prospective studies to establish causality and evaluate the clinical impact of iCa management in sepsis.

Conclusion

The relationship between iCa and 28-day mortality in patients with sepsis followed a U-shaped pattern, with an inflection point at 1.16 mmol/L, within the range of mild hypocalcaemia. Below this threshold, each 0.1 mmol/L increase in iCa was associated with a 13% reduction in mortality risk; whereas, above the threshold, each 0.1 mmol/L increase was linked to a 6% increase in risk. These findings highlight the prognostic value of iCa in sepsis and suggest that both hypocalcaemia and hypercalcaemia are associated with adverse outcomes. As this was an observational study, further prospective research is needed to determine whether interventions targeting calcium levels could improve patient outcomes.

Data availability statement

Data are available from the MIMIC-IV database (https://mimic.mit.edu/) for researchers who meet the criteria for access, including completion of a recognized training course and signing of a data use agreement.