Our aim was to determine the prevalence of VDD and analyse its association with morbidity, mortality and length of PICU stay. Specifically, we sought to describe the clinical and sociodemographic characteristics of these patients, to analyse the association of 25(OH)D levels with the PRISM III score, and the independent association of VDD with sociodemographic and clinical characteristics, laboratory parameters for systemic inflammatory response syndrome (SIRS), morbidity and mortality.

We conducted a prospective cohort study. Patients were recruited consecutively as they were admitted to the PICU and classified as VD-deficient (<20ng/mL) or not deficient. Once all patients had been recruited, the second phase consisted of a prevalence study.

Children aged 6 months to 17 years admitted to a medical-surgical PICU in a tertiary level university hospital. The exclusion criteria were: chronic renal disease, parathyroid disorder, malabsorption disorder and vitamin D supplementation.

We collected data for age (months), weight (kg), body mass index (BMI [kg/m2]), race (white, black, Asian), underlying disease, scheduled/emergency admission; date of admission and parental educational attainment. We analysed complete blood count values, levels of 25(OH)D, parathyroid hormone (PTH), C-reactive protein (CRP), procalcitonin (PCT), total and ionised calcium, magnesium and blood gases within 48h from admission. In patients that underwent surgery under extracorporeal membrane circulation (ECMO) or with large-volume fluid resuscitation, blood samples were collected at least 12h after to prevent potential dilution effect of these interventions. We measured levels of 25(OH)D by electrochemiluminescence binding assay (Cobas® e-602, Roche, Switzerland). We recorded the length of stay in the PICU (in days), the need for MV for more than 24h, the use of vasoactive drugs (VADs), the need for prolonged antibiotherapy (>7 days), parenteral nutrition (PN) and continuous renal replacement therapy (CRRT), as well as the duration of all these treatments.

We defined VDD as a level of less than 20ng/mL.2,13,14 We used an operational definition of morbidity, which included requiring MV for more than 24h, use of VADs, CRRT, antibiotherapy of more than 7 days’ duration or PN during the PICU stay.

We conducted the analysis in collaboration with an expert of the IBIMA-FIMABIS AMEC unit using the software R. We calculated the prevalence with a 95% confidence interval (CI). We have described continuous variables as mean and standard deviation or as median and interquartile range based on the normality of the data, and categorical variables as percentages with 95% CIs. We assessed the association between VD, baseline characteristics and outcome measures using the χ2 or Fisher's exact test for categorical variables, and Student's t test, ANOVA, Tukey's HSD test, the Mann–Whitney U test or the Kruskal–Wallis test as applicable. We used multivariate logistic regression to assess the association between VDD and patient characteristics and between VDD and morbidity and mortality. To determine whether 25(OH)D levels were independently associated with length of stay, morbidity and mortality, we included variables previously associated with these outcomes in the reviewed literature in the linear model, and those with P-values of less than 0.01 in the univariate analysis in the logistic regression model. We considered P-values of 0.05 statistically significant.

We calculated the sample size using the described prevalence of VDD in healthy Spanish children of 51%.15 To estimate the VDD proportion with a 5.6% precision and a 95% CI, we needed a minimum of 307 patients for the second phase of the study.

For the first phase (cohort study), we calculated the sample size based on a relative risk (RR) of mortality at 90 days of 1.27. To achieve a precision of 0.20 units in the estimation of the logarithm of the RR with an asymptotic 95% CI for a two-sided test, assuming the expected RR to be 1.73 [CI (RR): 1.73*exp(−0.20); 1.73*exp(+0.20)] and that the proportion of individuals in the reference group relative to the total to be 50%, we needed to include 100 patients in the reference group and 100 patients in the experimental group, for a total of 200 patients in the study.

The study protocol was approved by the competent regional research ethics committee. We provided an informational leaflet to the families and obtained their informed consent in every instance before including patients in the study. Data were recorded on an anonymous basis in an encrypted electronic database. The project adhered to the principles of the Declaration of Helsinki and standards for good clinical practice.

During the period under study, 615 patients were admitted, of who 466 met the inclusion criteria. Out of the latter, we did not obtain a signed informed consent for 126, so the final sample consisted of 340 patients. Table 1 shows the sociodemographic and baseline characteristics of the patients.

The mean 25(OH)D level for the overall sample was 22.28±10.55ng/mL (95% CI, 21.15–23.41). The prevalence of VD deficiency was 43.8% (95% CI, 38.53%–49.07%). One hundred and twenty-eight patients (37.6%) had VD insufficiency (20–30ng/mL), 41 patients (12.1%) had levels between 30 and 40ng/mL, and only 22 (6.5%) had optimal levels (>40ng/mL).

The bivariate analysis (Tables 1 and 2) found that patients with VDD were older, were admitted more frequently in the spring, had higher PRISM III scores and stayed longer in the PICU. In spring, the mean 25(OH)D level was 20.25±9.7ng/mL, while in autumn it was 26.9±11.7ng/mL. The highest level of 25(OH)D corresponded to August (32.42ng/mL) and the lowest to April (17.1ng/mL). We found a statistically significant association between parental educational attainment and VDD. The mean 25(OH)D levels for children of parents with no education, an elementary education, a secondary education and a higher education were 24.9, 22.7, 22.6 and 19.6ng/mL, respectively. Using bivariate logistic regression, we found that the risk of having VDD was greater in patients whose parents had a higher education compared to patients whose parents had an elementary education (OR, 0.498; 95% CI, 0.26–0.94) or a secondary education (OR, 0.37; 95% CI, 0.2–0.7).

The multivariate analysis (Table 3) found an increased risk of VDD in patients admitted during winter compared to autumn (OR, 2.93; 95% CI, 1.1–8.7; P=.04) and in patients that scored higher in the PRISM III scale (OR, 1.07; 95% CI, 1.02–1.13; P=0.005).

We also found an association between 25(OH)D levels and several laboratory parameters (Table 4). Patients with VDD had higher white blood cell and neutrophil counts, lower platelet counts, lower levels of albumin, ionised calcium, lower pH, and higher serum phosphate and PCT levels. The mean PTH level was 56±49pg/mL in patients with VDD and 414±34.6pg/mL in patients without VDD (P=0.004). Forty-one patients (27.5%) in the VDD group had hyperparathyroidism (PTH>60pg/mL), compared to 22 (12.6%) in the group without VDD (P=0.001). We found a negative linear correlation of PTH levels with both 25(OH)D levels (Pearson's r, −0.22; R2, 0.047; P<0.001) and ionised calcium levels (Spearman's ρ, −0.33; R2, 0.18; P<0.001).

The bivariate analysis (Table 1) showed that children with VDD stayed longer in the PICU. However, in the multivariate analysis adjusted for age, PRISM III score, morbidity and reason for admission we did not find a statistically significant association between VDD and PICU length of stay.

The multivariate analysis found an increased morbidity in patients with VDD (Table 1). The percentage of patients requiring more than 24 hours of MV was higher in the VDD group, as was the total duration of MV. Patients with VDD required VADs more frequently and for longer periods (median, 3 vs 2 days; mean, 7.5 vs 3.05 days, P=0.24), although the difference was not statistically significant. The need for CRRT was more frequent in patients with VDD, as were prolonged antibiotherapy and the use of PN.

In the VDD group, patients with secondary hyperparathyroidism had increased morbidity compared to patients with PTH levels of less than 60pg/mL (65.9% vs 59.3%; P<0.001).

In the multivariate logistic regression analysis, the adjusted OR for VDD was 5.55 (95% CI, 2.64–11.69). C-reactive protein levels, the PRISM III score and the PICU length of stay were also associated with an increased risk of morbidity, while ionised calcium levels and age had a protective effect (Table 5).

While mortality was higher in patients with VDD in the bivariate analysis, the difference was not statistically significant. Levels of 25(OH)D were lower in patients that died compared to survivors at discharge from PICU (14±8.81ng/mL vs 22.53±10.53ng/mL; P=0.012). In the multivariate regression analysis, higher levels of 25(OH)D acted as protective factors against mortality (OR, 0.84; 95% CI, 0.70–0.99; P=0.04). This effect was independent of hypocalcaemia, as shown by the second logistic regression model (Table 6). In this model, PTH levels were a risk factor for mortality (OR, 1.021; 95% CI, 1.003–1.04). However, when we included hypocalcaemia in model 2, the association of PTH with mortality was no longer significant, which suggests that calcium level is a confounding factor and potential source of bias (Table 6). The PRISM III score and emergency admission to the PICU were also associated with increased risk of mortality in the multivariate analysis (OR, 1.32 and 25.95, respectively).