Data on 32,551 adult patients were included in our analysis. There were 28,062 survivors
and 4489 non-survivors (28-day mortality rate: 13.8 %). As shown in Table 1, variables including age (62.6 ± 17.9 vs. 71.3 ± 16.0, p  0.001), sex (male percentage:
56.70 % vs. 53.68 %, p  0.001), SAPS-1 (13.1 ± 5.2 vs. 17.6 ± 5.8, p  0.001), SOFA
(5.1 ± 3.8 vs. 8.1 ± 4.6, p  0.001), Asian population (3.54 vs. 4.15 %, p = 0.044),
hypertension (31.79 vs. 29.18 %, p = 0.001), congestive heart failure (19.88 vs. 31.98 %
p  0.001), chronic pulmonary disease (16.89 % vs. 19.11 %, p  0.001), renal failure
(6.13 vs. 9.71 %, p  0.001), liver disease (5.06 vs. 7.19 %, p  0.001), alcohol
abuse (5.41 vs. 3.80 %, p  0.001), serum iCa on ICU entry (1.133 ± 0.103 vs. 1.113 ± 0.146 mmol/L,
p  0.001), serum creatinine (1.32 ± 1.48 vs. 1.76 ± 1.52 mg/dL, p  0.001) and calcium
supplementation (33.56 vs. 31.03 %, p = 0.001) were significantly different between
survivors and non-survivors. Variables with p  0.1 were entered into Cox regression
model (Table 2), which showed that calcium supplementation was associated with reduced risk of death
(hazard ratio: 0.51; 95 % CI 0.47–0.56).
Table 1. Differences of clinical characteristics between survivors and non-survivors (28-day
mortality)
Table 2. Cox regression model showing variables associated with 28-day mortality
Variables were compared between calcium and non-calcium groups (Table 3). Mild forms of acute renal failure as reflected by mild creatinine elevation was
not significantly different between calcium and non-calcium groups (1.36 ± 1.53 vs.
1.38 ± 1.47 mg/dL, p = 0.26). Calcium supplementation was used as the dependent variable.
Covariates including sex (59.44 vs. 54.70 %, p  0.001), SAPS-1 (16.0 ± 5.1 vs. 12.2 ± 5.2,
p  0.001), SOFA (7.5 ± 3.9 vs. 4.4 ± 3.6, p  0.001), hypertension (30.7 vs. 31.9 %,
p = 0.033), congestive heart failure (20.10 vs. 22.28 %, p  0.001), chronic pulmonary
disease (15.92 vs. 17.91 %, p  0.001), renal failure (6.17 vs. 6.86 %, p = 0.023),
AIDS (0.58 vs. 0.78 %, p = 0.047), iCa on ICU entry (1.127 ± 0.104 vs. 1.135 ± 0.121,
p  0.001) were associated with the choice of calcium supplementation. Because SAPSI-1
and SOFA both measured the same clinical characteristics, we used SAPSI-1 to calculate
propensity score. After radius matching, there were 5238 in non-calcium group and
8719 in the calcium group. The balance of covariates is shown in Fig. 1, which demonstrates that these variables are well balanced after matching. Figures 2 and 3 are Kaplan–Meier survival curves showing the 28- and 90-day mortality by calcium
groups. The results showed that calcium supplementation was associated with improved
28- and 90-day mortality (p  0.05 for both Log-rank test).
Table 3. Differences of clinical characteristics between calcium and non-calcium groups
Fig. 1. Standardized bias (%) across covariates before and after propensity score matching.
The result showed that candidate covariates were well matched
Fig. 2. Kaplan–Meier survival curves showing that 28-day mortality was reduced with calcium
supplementation
Fig. 3. Kaplan–Meier survival curves showing that 90-day mortality was reduced with calcium
supplementation
Multivariable dose–response relationship
In patients received calcium supplementation, we investigated the dose–response relationship.
The median calcium intake was 13.9 mmol (interquartile range: 4.6–111.9 mmol) in patients
received calcium supplementation during their ICU stay. Distribution of total calcium
intake stratified by different serum iCa is displayed in Fig. 4. The multivariable model included variables age, sex, SAPS-1, SOFA, Asian population,
hypertension, congestive heart failure, chronic pulmonary disease, renal failure,
liver disease, alcohol abuse, serum iCa on ICU entry and serum creatinine (Table 4). The result showed that the dose of calcium intake was significantly associated
with 90-day mortality (hazard ratio: 1.0004, p  0.001).
Fig. 4. Distribution of total calcium intake stratified by serum iCa. The overall median calcium
intake was 13.9 mmol (interquartile range: 4.6–111.9 mmol)
Table 4. Cox proportional hazard model investigating the association of calcium supplementation
with 90-day mortality
Subgroup analysis
Patients were grouped into subsets by their minimum calcium levels of 1.0 (11,404),
0.9–0.99 (2388), 0.8–0.89 (1326), 0.7–0.79 (343) and 0.7 (316). Multivariable regression
model was fitted to adjust for confounding variables to investigate the effect of
calcium supplementation on mortality risk. The results showed that in subgroups with
iCa 1, calcium supplementation was associated with reduced risk of death (OR: 0.41,
95 % CI 0.37–0.45). Results of other subgroups are shown in Fig. 5.
Fig. 5. Subgroup analysis by dividing patients into subsets according to their minimum iCa.
Within each subgroup, multivariable regression model was used to control for confounding
factors including creatinine, sex, age, SAPS-I, SOFA, Asian population, hypertension,
congestive heart failure, chronic pulmonary disease, renal failure, liver disease
and alcohol abuse