Introduction

Spinal cord injury (SCI) is a severe trauma to the central nervous system, often resulting in sensory and motor dysfunction or even lifelong disability in patients, imposing heavy medical and economic burdens on individuals, families, and society1. Epidemiological data show that the annual incidence of SCI is approximately (0.7–1.2) per 900,000 population. In recent years, with the increase in accidental injuries such as traffic accidents and falls from heights, its incidence has been on an annual upward trend2. The prognosis of SCI patients is affected by multiple factors, including injury level, injury severity, timing of treatment, and management of complications. Mortality and survival outcomes of SCI vary by injury characteristics. Studies have found that mortality is extremely high in the first year after injury, which depends on many factors such as the severity of trauma, management of early complications, quality of rescue and medical assistance in the acute phase, and admission to specialized centers3. SCI mortality is also closely associated with the American Spinal Injury Association (ASIA) grade, level of neurological injury, mechanical ventilation, advanced age, and sex4. Thietje et al.5 found that the average survival time of patients with high-level injuries was shorter than that of tetraplegic patients with low-level injuries, and there was a negative correlation between survival rate and higher neurological injury level or complete injury. Charlifue et al.6 identified the need for mechanical ventilation as another important factor affecting the survival rate of SCI patients. Varma et al.7 found through research that advanced age was positively correlated with mortality in SCI patients—the older the patient, the higher the mortality rate. Additionally, studies have shown that the 1-year survival rate of SCI patients ranges from 79 to 100%, the 5-year survival rate from 85 to 96%, and the 10-year survival rate from 81 to 93%. Studies comparing survival rates between males and females usually show that females have a higher survival rate8. Furthermore, several other factors should be considered as potential influences on survival prognosis, including low educational attainment, maladaptive behaviors (substance abuse, suicide attempts), and poor quality of care9.

Among these factors, metabolic disorders—one of the common complications—have gradually become a key factor affecting patients’ rehabilitation progress and long-term prognosis10,11. As one of the most important minerals in the human body, calcium is not only a major component of bones and teeth, but also plays an irreplaceable role in physiological processes such as nerve conduction, muscle contraction, cellular signal transduction, and regulation of enzyme activity12. Under normal circumstances, the human body maintains the serum calcium concentration within a stable range through the synergistic effect of multiple hormones, including parathyroid hormone, vitamin D, and calcitonin13. After the occurrence of SCI, patients often develop calcium metabolism imbalance due to factors such as prolonged bed rest, neuroregulatory dysfunction, gastrointestinal disorders, and drug interventions14. Srichuachom et al.15 found that the overall prevalence of hypocalcemia in trauma patients upon admission to the emergency department was 56%, and the mortality rate of patients with hypocalcemia was significantly higher than that of patients with normocalcemia. Hypocalcemia is relatively common in adult trauma patients, and the mortality rate in this population is notably higher. Serum calcium levels play a crucial role in SCI repair: calcium ions are involved in nerve injury repair, and long-term hypocalcemia can reduce the calcium ion concentration in the extracellular environment, thereby impairing neuronal function16. Zhang et al.17 observed that the intracellular calcium ion concentration increases significantly after SCI, and there is a significant correlation between changes in intracellular calcium ion concentration and motor function. Intracellular calcium ion overload may play an important role in the pathogenesis of SCI. In addition, long-term hypocalcemia can also affect the bone mineral density (BMD) of SCI patients. It not only leads to reduced bone mass, but also impairs the microstructure of bones, resulting in poor prognosis associated with severe osteoporosis18. Gifre et al.19 followed 35 SCI patients for 12 months and found that 52% of them developed osteoporosis. In recent years, dynamic monitoring of the trajectory of physiological indicators has gradually become an important method for evaluating disease progression and prognosis. As a quantitative indicator reflecting the dynamic changes of calcium metabolism, calcium trajectory can more comprehensively capture the fluctuation characteristics of calcium levels during the course of the disease, and is of greater clinical significance compared with calcium concentration detection at a single time point20,21. In critically ill patients, persistent changes or severe fluctuations in calcium levels often indicate a poor prognosis, and early identification and intervention of abnormal calcium trajectories may improve patients’ clinical outcomes. However, there are currently no relevant studies on the dynamic change pattern of calcium trajectories in SCI patients and its association with prognosis, and systematic observation of the changing trend of calcium metabolism is lacking.

The MIMIC-IV (Medical Information Mart for Intensive Care IV) database is one of the most widely used public databases for intensive care medicine internationally, containing detailed clinical information of a large number of critically ill patients, including demographic data, laboratory test results, medication records, imaging data, and prognostic outcomes22. Therefore, based on the MIMIC-IV database, this study intends to extract serum calcium test data from patients with spinal cord injury, construct a calcium trajectory model, and analyze the correlation between different calcium trajectory types and patient prognosis, aiming to provide new ideas for early clinical identification of high-risk patients and improvement of patients’ clinical outcomes.

Methods

Data source

The data of this study were derived from version 3.0 of the MIMIC-IV (Medical Information Mart for Intensive Care-IV) database, a large-scale public clinical database developed through collaboration between the Laboratory for Computational Physiology at the Massachusetts Institute of Technology and the Beth Israel Deaconess Medical Center. It contains detailed clinical information of patients admitted to the intensive care unit (ICU) of the Beth Israel Deaconess Medical Center from 2008 to 2019, covering multi-dimensional data such as demographic information, diagnosis and treatment records, laboratory test results, vital sign monitoring data, medication information, and prognostic outcomes. The MIMIC-IV database strictly protects patient privacy through de-identification processing; thus, this study did not require approval from the Institutional Review Board (IRB) or informed consent.

Our research team obtained access permissions to the MIMIC-IV database (project number: 68561984), and the researchers have completed and obtained certification for the human research protection training provided by the National Institutes of Health (NIH). During data extraction, Structured Query Language (SQL) was used to link and screen relevant tables in the database, including core tables such as patient basic information (patients), admission records (admissions), ICU stay records (icustays), diagnostic coding (diagnoses_icd), laboratory test results (labevents), and prognostic outcomes (outcomes), ensuring the completeness and accuracy of the extracted data.

Study population

The study population consisted of patients with SCI selected from the MIMIC-IV database. The inclusion criteria were as follows: (1) adult patients aged ≥ 18 years; (2) definite diagnosis of SCI based on the 10th Revision of the International Classification of Diseases (ICD-10) codes (code range: S14.0–S34.9, T09.3, T10.3, T11.3, T12.3, S04.1); (3) ICU stay duration ≥ 24 h to ensure sufficient clinical data records; (4) availability of at least 2 serum calcium test results in laboratory examinations for constructing calcium trajectories; (5) complete clinical data, including demographic information, history of underlying diseases, laboratory test data, and prognostic outcome indicators required for this study. The exclusion criteria were: (1) patients with concurrent severe hepatic or renal failure, end-stage malignant tumors, or other end-stage diseases; (2) patients with congenital calcium metabolism disorders or long-term use of drugs affecting calcium metabolism with unclear medication duration; (3) patients with significant missing clinical data or logical errors that could not be supplemented or corrected after manual verification; (4) patients with repeated ICU admissions, where only the first ICU stay record meeting the inclusion criteria was retained.

Variable acquisition

Clinical data were retrieved using SQL and PostgreSQL (Version 9.6). The included clinical variables were as follows: gender, age, hospital length of stay (los_hospital_day), ICU length of stay (los_icu_day), ICU survival time (time_icu), in-ICU death status (death_icu_yn), in-hospital survival time (time_hosp), in-hospital death status (death_hosp_yn), presence of diabetes mellitus (Diabetes), presence of hypertension (Hypertension), heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean arterial pressure (MAP), body temperature (Tem), oxygen saturation (sPO₂), red blood cell count (RBC), white blood cell count (WBC), serum creatinine (Serum_Creatinine), blood glucose (Glu), serum potassium (Potassium), partial pressure of oxygen (PO₂), partial pressure of carbon dioxide (PCO₂), blood pH (pH), serum calcium, lactate, Sequential Organ Failure Assessment score (SOFA), Acute Physiology Score III (APSIII), Simplified Acute Physiology Score Ⅱ (SAPSⅡ), Oxford Acute Severity of Illness Score (OASIS), Glasgow Coma Scale score (GCS), and Charlson Comorbidity Index (CCI).

Outcome indicators

The primary outcomes were all-cause mortality within 30 days and 60 days of ICU admission. Secondary outcomes included ICU length of stay, hospital length of stay, and functional indicators.

Statistical analysis

Data were standardized, and samples with a missing value proportion exceeding 20% were excluded. The Group-Based Trajectory Model (GBTM) was used to identify the unique longitudinal change characteristics of serum calcium levels (Note: “Pmean” in the original text is presumed to be a typo for the core variable "serum calcium levels" based on the study context). To accurately fit the dynamic change trend of serum calcium levels over time, this study first determined the optimal order by comparing the model fitting effects of different polynomial orders (linear, quadratic, and cubic). By comparing the Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), and residual sum of squares of models with different orders, the quadratic polynomial was finally selected as the trajectory fitting function. Combined with the clinical practical scenario of serum calcium testing in the MIMIC-IV database and data integrity, the time scale unit for GBTM modeling was set to “days”, with the time of ICU admission as the baseline time point (T = 0). Subsequent testing times were standardized as "Day 1, Day 2… Day n after admission". In clinical practice, serum calcium testing time points are affected by factors such as the severity of patients’ conditions, adjustments to treatment plans, and doctors’ diagnosis and treatment decisions, resulting in significant imbalance. Specifically, some patients undergo intensive testing in the first 3 days after admission with extended intervals between subsequent tests, while others have fewer testing points due to early discharge after condition improvement. To address this issue, a two-step approach was adopted in this study: The first step was testing point screening and time alignment. First, all patients’ serum calcium test data were sorted by "time after admission", and abnormal data with ambiguous test time records (e.g., only marked as “during hospitalization” without a specific time) or exceeding the ICU length of stay were excluded. Then, the "time window alignment method" was used to map each patient’s test time to preset standard time nodes. If there was no corresponding test value for a standard node, linear interpolation was used only when the interval between adjacent tests was ≤ 3 days to fill in the value; if the interval between adjacent tests was > 3 days, the node was marked as “missing” to avoid bias caused by excessive interpolation range. The second step was weight assignment and model robustness verification. In GBTM modeling, weights were assigned to each patient’s test data based on "testing point density": a weight of 1.0 was set for patients with ≥ 5 testing points, 0.8 for those with 3–4 testing points, and 0.6 for those with 2 testing points (no samples with 1 testing point existed, as the inclusion criteria required at least 2 tests). The logic behind this weight assignment was that patients with denser testing points had more reliable calcium level change trends and thus should be given higher contribution weight, while patients with fewer testing points had their interference on the overall trajectory classification reduced by lowering their weights. Meanwhile, sensitivity analysis was conducted to verify the robustness of the processing method: trajectory classification was performed based on the “unweighted model” and “weighted model” respectively, and the number of trajectories, sample proportion of each trajectory group, and entropy value (classification accuracy) of the two models were compared. The results showed that the entropy value of the weighted model (0.89) was higher than that of the unweighted model (0.82), and the consistency of trajectory classification results between the two models reached 91.3% (Kappa = 0.85, P < 0.001), indicating that the method for addressing the imbalance of testing time points in this study was reliable and did not introduce significant bias.

A two-stage iterative model fitting method was used for analysis: first, a polynomial function was applied to the trajectories to determine the number of groups, with the number of groups in the model ranging from 1 (corresponding to trajectories with no obvious characteristics) to 6; after selecting the appropriate number of groups, the model further determined the specific structure of each trajectory. The model fitting effect was evaluated using BIC, entropy value (> 0.7), and the minimum group size criterion (> 5%). For missing covariate data, multiple imputation was used to fill in missing values for continuous variables with a small amount of missing data, while the most frequent category was used for categorical variables. Additionally, the consistency between the predicted group probabilities and the established population proportions was calibrated. Furthermore, this study also balanced the operational convenience of the model and its intuitiveness in clinical scenarios.

Based on the different classifications of calcium trajectories, the Kruskal–Wallis test or analysis of variance (ANOVA) was used to determine the basic characteristics of continuous variables, while the chi-square test or Fisher’s exact test was applied to examine the basic characteristics of categorical variables. The Kaplan–Meier method was used to plot survival curves for different groups, and the log-rank test was employed to assess differences in survival rates between groups. A Cox proportional hazards regression model was constructed to further analyze the independent effect of calcium trajectories on the prognosis of patients with SCI after adjusting for potential confounding factors such as demographic characteristics, clinical indicators, and comorbidities. Additionally, subgroup analysis was conducted to evaluate the robustness of the results. To verify the incremental predictive value of serum calcium trajectories as prognostic biomarkers, the C-statistic and Net Reclassification Improvement (NRI) were further used to assess the optimization effect of combining calcium trajectories with conventional prognostic scoring systems on the predictive model for ICU mortality risk in SCI patients. All statistical analyses were performed using SPSS 26.0 or R 4.2.0 software, with a two-tailed test significance level (α) set at 0.05.

Results

Results of trajectory analysis

Model fitting indices for different numbers of categories (ranging from 1 to 6), including Log Likelihood, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Scaled Bayesian Information Criterion (SABIC), are presented in Fig. 1A. The log likelihood value increases with the increasing number of categories. However, in the two-category model, AIC and BIC reach the lowest levels, indicating an optimal balance between model fit and complexity. Additionally, an entropy value close to 1 enhances the classification accuracy, guiding us to select the two-category model as the optimal solution (Fig. 1B,C).

Fig. 1
figure 1

(A) presents the model selection based on scaled and ranked values of log-likelihood, AIC, BIC, and SA-BIC; (B) shows the changes in predicted mean values over time for the two categories along with their corresponding confidence intervals; (C) depicts the predicted probability density distributions of the two categories under different time and value conditions.

Full size image

Baseline characteristics of patients

A total of 699 patients diagnosed with SCI were included in this study. Based on the pattern of serum calcium ion levels, they were divided into two groups: Class 1 (n = 244) and Class 2 (n = 455). Class 1 was characterized by a "steep decline in serum calcium ion levels", while Class 2 was characterized by a "slow but persistent decline in serum calcium ion levels". Differences between the two groups were analyzed and compared. In terms of clinical indicators, there were no statistically significant differences between the two groups in sex, age, in-ICU death status (death_icu_yn), in-hospital death status (death_hosp_yn), presence of diabetes mellitus (Diabetes), presence of hypertension (Hypertension), heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean arterial pressure (MAP), body temperature (Tem), age group (age_group), SOFA group (sofa_group), Sequential Organ Failure Assessment (SOFA) score, Acute Physiology and Chronic Health Evaluation III (APS III) score, Simplified Acute Physiology Score II (SAPS II), Oxford Acute Severity of Illness Score (OASIS), Charlson Comorbidity Index (CCI), injury level, ASIA classification, or timing of surgery (all P > 0.05). However, Class 1 had significantly longer hospital length of stay (los_hospital_day), ICU length of stay (los_icu_day), ICU survival time (time_icu), and in-hospital survival time (time_hosp) compared with Class 2 (all P < 0.05). Regarding blood biochemical indicators, a significant difference was observed in oxygen saturation (sPO₂) between the two groups (P < 0.05), while no significant differences were found in red blood cell count (RBC), white blood cell count (WBC), serum creatinine (Serum_Creatinine), blood glucose (Glu), serum potassium (potassium), partial pressure of oxygen (PO₂), partial pressure of carbon dioxide (PCO₂), blood pH value, serum calcium (calcium), or lactate between the two groups (all P > 0.05). (Table 1).

Table 1 Baseline characteristics of patients.
Full size table

Results of survival analysis

Kaplan–Meier survival curves were used to analyze the 30-day and 60-day in-ICU mortality rates between the two groups of patients. The results showed significant differences in mortality rates between the two groups. Regarding the 30-day in-ICU mortality, the survival probability of Class 1 within the first 30 days was significantly higher than that of Class 2 (P < 0.05), and the survival curve indicated a survival benefit in Class 1 (Fig. 2A). For the 60-day in-ICU mortality, the 60-day survival probability of Class 1 was significantly higher than that of Class 2 (P < 0.05), with significant differences in survival curves between the two groups at all stages (P < 0.05) (Fig. 2B). Throughout the curve, the survival rate of patients in Class 1 remained stable overall, while that of patients in Class 2 decreased significantly, suggesting that the risk of in-ICU death in Class 2 was higher than in Class 1. This phenomenon may be related to the disease status, clinical diagnosis and treatment, and treatment response of the two groups.

Fig. 2
figure 2

Kaplan–Meier survival curves illustrating the in-ICU mortality rates over time for different serum calcium ion trajectory categories (Class 1 and Class 2). (A) Kaplan–Meier survival curve for 30-day in-ICU all-cause mortality; (B) Kaplan–Meier survival curve for 60-day in-ICU all-cause mortality.

Full size image

Cox regression analysis

Univariate and multivariate Cox regression analyses were performed in this study to explore the relationship between different serum calcium ion trajectory categories and 30-day and 60-day in-ICU mortality. Compared with serum calcium ion trajectory Class 1, trajectory Class 2 was significantly associated with an increased risk of in-ICU death at both time points (P < 0.05). After adjusting for multiple confounding factors such as age, sex, diabetes mellitus, hypertension, and various physiological parameters, the hazard ratio (HR) for Class 2 ranged from 1.969 to 2.122, with all P-values < 0.05 (Table 2). These findings suggest that Class 1 may present more favorable clinical manifestations and is associated with improved clinical prognosis in patients with spinal cord injury. The consistently lower risk of death associated with trajectory Class 1 highlights the application value of serum calcium ion trajectory analysis in the stratified management and personalized treatment of spinal cord injury.

Table 2 Univariate and multivariate cox regression analyses of the two calcium ion patterns.
Full size table

Subgroup analysis

Different subgroups stratified by sex, age, hypertension, and diabetes mellitus were correlated with trajectory categories. In males, Class 2 showed a non-significant trend toward an increased risk of death (HR = 2.090, P = 0.064), while no significant difference was observed in females (HR = 1.772, P = 0.305), with no interaction between sexes (P > 0.05). Older patients exhibited a non-significant trend of elevated death risk, whereas no significance was noted in younger patients (HR = 2.011, P = 0.320), and no interaction was found between different age groups (P > 0.05). Hypertensive patients faced an increased risk of mortality (HR = 3.282, P = 0.027), but no significance was observed in non-hypertensive patients (HR = 1.449, P = 0.362), with no significant interaction related to hypertension (P > 0.05). Neither diabetic patients (HR = 2.525, P = 0.251) nor non-diabetic patients (HR = 1.734, P = 0.119) showed significant differences, and no significant interaction was found regarding diabetes mellitus (P > 0.05). (Table 3).

Table 3 Subgroup analysis of the association between serum calcium ion trajectories and mortality in spinal cord injury.
Full size table

Results of predictive performance

The C-statistic ranges from 0.5 to 1.0, with values closer to 1.0 indicating a stronger ability of the model to distinguish between “death” and “survival” outcomes. The C-statistic of Model 4 (fully adjusted combined model) was 0.782 (95% CI: 0.721–0.843), which represented an increase of 0.059 compared with Model 1 (P = 0.018) and a slight improvement compared with Model 3 (ΔC = 0.017, P = 0.049). This suggests that combining calcium trajectories with conventional indicators can significantly enhance the discriminatory ability of the model.

The NRI is used to evaluate the degree of optimization in the prognostic risk stratification of patients by the model after the inclusion of a new indicator (calcium trajectories). It is divided into Overall NRI, Event NRI (for patients with the death outcome), and Non-event NRI (for patients who survived). An NRI > 0 indicates that the new indicator can improve the accuracy of risk classification. In this study, "high risk (mortality probability ≥ 20%)" and "low risk (mortality probability < 20%)" were used as stratification thresholds to calculate the NRI of Model 3 and Model 4 relative to Model 1. The Overall NRI of Model 3 compared with Model 1 was 0.236 (95% CI: 0.089–0.383, P = 0.002), among which the Event NRI was 0.185 (95% CI: 0.052–0.318, P = 0.006) and the Non-event NRI was 0.051 (95% CI: − 0.012–0.114, P = 0.108). This indicates that after adding calcium trajectories, 18.5% of patients who died were correctly reclassified as high-risk, and 5.1% of patients who survived were correctly reclassified as low-risk, resulting in a significant improvement in the overall accuracy of risk classification. The Overall NRI of Model 4 compared with Model 1 was 0.274 (95% CI: 0.118–0.430, P < 0.001), with an Event NRI of 0.213 (95% CI: 0.075–0.351, P = 0.002) and a Non-event NRI of 0.061 (95% CI: − 0.008–0.130, P = 0.082). The NRI was further improved after full adjustment, and the statistical significance of the Event NRI was stronger, suggesting that the incremental predictive value of calcium trajectories is more stable after controlling for confounding factors. (Table 4).

Table 4 Comparison of predictive performance of different predictive models for 30-day all-cause mortality in the ICU among patients with SCI.
Full size table

Discussion

As one of the most critical electrolytes in the human body, serum calcium plays a vital role in maintaining physiological functions such as nerve conduction, muscle contraction, cell signal transduction, and bone metabolism through its concentration homeostasis23. SCI, as a severe central nervous system trauma, is often accompanied by systemic stress responses and multisystem metabolic disorders, among which calcium metabolism abnormalities are a typical manifestation24. Studies have shown that SCI patients frequently experience fluctuations in serum calcium levels during the acute phase of injury, which may be associated with the following mechanisms: on the one hand, sympathetic nerve dysfunction caused by trauma can inhibit osteoblast activity and enhance osteoclast function, leading to increased release of bone calcium, which may result in transient hypercalcemia in the early stage; on the other hand, disuse osteoporosis due to long-term bed rest after SCI, increased renal calcium excretion, and reduced intestinal calcium absorption may lead to subsequent hypocalcemia25,26,27. In addition, calcium ions exert a dual role in the nerve repair process after SCI: an appropriate influx of calcium ions is necessary for nerve regeneration, but excessive influx can activate apoptotic pathways and exacerbate nerve cell damage28,29. Although previous studies have focused on calcium metabolism abnormalities in critically ill patients, there is currently a lack of reports on the dynamic trajectory of serum calcium levels in SCI patients and its relationship with prognosis30. Based on large-sample data from the MIMIC-IV database, this study is the first to reveal the longitudinal change characteristics of serum calcium in SCI patients through trajectory model analysis, providing a new perspective for understanding the relationship between calcium metabolism and SCI prognosis.

In this study, GBTM was used to classify the serum calcium trajectories of 699 SCI patients into two categories: Class 1, characterized by "a steep decline in serum calcium ion levels," and Class 2, characterized by "a slow but persistent decline in serum calcium ion levels." Baseline characteristic analysis showed that the duration of ICU stay, total hospital stay, ICU survival time, and in-hospital survival time in Class 1 patients were significantly longer than those in Class 2 (P < 0.05). Survival analysis and Cox regression results revealed that the 30-day and 60-day all-cause mortality rates in the ICU were significantly lower in Class 1 patients than in Class 2 (P < 0.05). After adjusting for confounding factors such as age, sex, diabetes mellitus, and hypertension, Class 2 remained an independent risk factor for increased mortality (30-day HR = 2.064, 60-day HR = 2.122). These results suggest that the dynamic change pattern of serum calcium may better reflect the severity and prognosis of SCI patients than a single calcium level measurement. Mechanistically, the rapid decline in calcium levels in Class 1 patients may be related to the “compensatory regulation” of the acute stress response. Traumatic stress in the acute phase of SCI can activate the hypothalamic–pituitary–adrenal axis, prompting the temporary release of bone calcium to meet the body’s needs. The subsequent rapid decline may reflect the body’s rapid restoration of calcium homeostasis through pathways such as renal excretion and bone redeposition, indicating that these patients have stronger metabolic regulation capabilities and a more timely stress response to injury. In contrast, the persistent slow decline in calcium levels in Class 2 patients may reflect long-term metabolic disorders: on the one hand, long-term suppression of sympathetic nerve function after SCI can lead to sustained enhancement of osteoclast activity, with bone calcium loss exceeding the body’s compensatory capacity; on the other hand, persistent inflammatory responses can damage renal tubular function, affecting calcium reabsorption and resulting in a gradual decline in serum calcium levels31,32. This long-term hypocalcemic state may further trigger a cascade of reactions, such as weakened myocardial contractility, increased risk of arrhythmias, and suppressed immune function, ultimately exacerbating multiple organ dysfunction and increasing the risk of death33. Furthermore, the NRI analysis confirmed that calcium trajectories can enable 23.6%-27.4% of patients to obtain more accurate risk stratification, and in particular, significantly improve the ability to identify high-risk patients among those with the death outcome (Event NRI = 0.185–0.213). Serum calcium trajectories are not only a potential biomarker for the prognosis of patients with SCI but also can provide significant incremental predictive value for conventional prognostic scoring systems.

Furthermore, subgroup analysis revealed that the association between Class 2 trajectory and increased mortality risk was more pronounced in hypertensive patients (HR = 3.282, P = 0.027), whereas no significant difference was observed in non-hypertensive patients (HR = 1.449, P = 0.362). This may suggest that serum calcium trajectories could also reflect systemic stress and vascular risk. Similar abnormalities in calcium metabolism have been shown to be associated with prognosis in non-SCI ICU populations. The systemic stress response activates the hypothalamic–pituitary–adrenal axis, triggering the release of bone calcium and subsequent compensatory regulation of calcium homeostasis. Persistent inflammatory responses impair renal tubular function, affecting calcium reabsorption. Additionally, vascular endothelial dysfunction caused by underlying conditions such as hypertension further exacerbates abnormal calcium channel regulation. These processes are unrelated to SCI but represent common pathological changes in critically ill patients30,33,34,35. The subgroup results of hypertensive patients in this study may precisely reflect this shared mechanism. Hypertensive patients already have pre-existing calcium signaling disorders in vascular endothelial cells. Calcium metabolism abnormalities induced by SCI overlap with underlying vascular lesions, forming a vicious cycle of "calcium metabolism disorder—vascular contractile dysfunction—tissue hypoxia," which ultimately amplifies the risk of death34,35. In this case, changes in serum calcium trajectories are more likely to reflect systemic vascular/metabolic vulnerability rather than mechanisms solely related to SCI. This finding suggests that monitoring calcium trajectories may have greater prognostic value for SCI patients with comorbid hypertension.

This study has several limitations. First, the data were derived from the MIMIC-IV database, which only includes patients admitted to the ICU between 2008 and 2019. This may have led to the SCI patients with milder conditions who did not require ICU admission, resulting in selection bias. Second, the timing of serum calcium measurements was determined by clinical practice, with no standardized testing frequency, which may have affected the accuracy of the trajectory model in capturing dynamic changes in calcium levels. Third, the sample size of some subgroups in the subgroup analysis was relatively small, potentially leading to insufficient statistical power. Fourth, as a retrospective observational study, it cannot establish a causal relationship between calcium trajectories and mortality risk; prospective studies will be needed to validate this relationship in the future. Finally, due to limitations in data sources, this study did not include a non-SCI critically ill control group, making it impossible to verify the specificity of calcium trajectories through direct comparison.

Conclusion

This study confirms that the dynamic change trajectory of serum calcium in patients with SCI can serve as a potential biomarker for prognostic assessment. Among these trajectories, the "slow but continuous decline" type (Class 2) indicates a higher risk of ICU mortality. Furthermore, calcium trajectories can provide significant incremental predictive value for conventional prognostic scoring systems. However, the results of this study are only applicable to critically ill SCI patients admitted to the ICU. Clinicians can identify high-risk populations by dynamically monitoring the serum calcium levels of SCI patients and take targeted intervention measures to improve prognosis. Future studies may further combine the specific characteristics of spinal cord injury to explore the association between calcium trajectories and neural repair mechanisms, thereby providing a theoretical basis for the precise treatment of SCI.