Development and validation of a risk nomogram to estimate risk of hyponatremia after spinal cord injury: A retrospective single-center study

Abstract

Objective

This study aimed to establish a nomogram-based assessment for predicting the risk of hyponatremia after spinal cord injury (SCI).

Design

The study is a retrospective single-center study.

Participants

SCI patients hospitalized in the First Affiliated Hospital of Guangxi Medical University.

Setting

The First Affiliated Hospital of Guangxi Medical University, Nanning, Guangxi, China.

Methods

We performed a retrospective clinical study to collect SCI patients hospitalized in the First Affiliated Hospital of Guangxi Medical University from 2016 to 2020. Based on their clinical scores, the SCI patients were grouped as either hyponatremic or non-hyponatremic, SCI patients in 2016–2019 were identified as the training set, and patients in 2020 were identified as the test set. A nomogram was generated, the calibration curve, receiver operating characteristic (ROC) curve, and decision curve analysis (DCA) were used to validate the model.

Results

A total of 895 SCI patients were retrieved. After excluding patients with incomplete data, 883 patients were finally included in this study and used to construct the nomograms. The indicators used in the nomogram included sex, completeness of SCI, pneumonia, urinary tract infection, fever, constipation, white blood cell (WBC), albumin and serum Ca²⁺. These indices were determined by the least absolute shrinkage and selection operator (LASSO) regression analysis. The C-index of the model was 0.81, the area under the curve (AUC) of the training set was 0.82(Cl:0.79–0.85), and the validation set was 0.79(Cl:0.73–0.85).

Conclusions

Nomogram has good predictive ability, sex, completeness of SCI, pneumonia, urinary tract infection, fever, constipation, WBC, albumin and serum Ca²⁺ were predictors of hyponatremia after SCI.

Keywords:

• Nomogram
• Spinal cord injury
• Hyponatremia

Introduction

Spinal cord injury (SCI) is one of the important public health problems in the world, and the incidence rate of SCI varies greatly from region to region. There are many complications of SCI, among which hyponatremia is a common complication, and hyponatremia is associated with adverse consequences (1). In many studies, it has been proven that the mortality rate of SCI inpatients with combined hyponatremia is higher.² If not corrected in a timely manner, acute severe hyponatremia may lead to death. Hyponatremia, a condition when serum sodium levels fall to <135 mmol/L, is the second-most common complication associated with lung infection in patients with SCI (1). Hyponatremia is associated with increased hospital stay and mortality (2). In clinical practice, due to lack of awareness, patients with hyponatremia often cannot be treated in time, which reduces the rehabilitation effect of patients and even endangers life. The symptoms associated with acute and severe hyponatremia include seizures, acute psychosis, and permanent brain injury, leading to coma and death, the symptoms of chronic hyponatremia, on the other hand, include headache, nausea, vomiting, gait changes, muscle spasms, irritability, and disorientation (3). With the non-availability of proper prediction models, treatment for hyponatremia in patients with SCI often gets delayed, which ultimately affects their rehabilitation and even endangers their life. Therefore, developing prediction models for identifying high-risk factors for hyponatremia have practical significance. These models can be used for early diagnosis and treatment and therefore improve patient recovery.

Nomogram is a prediction model that can predict the probability of a specific outcome by adding the scores of each risk factors. In clinical settings, nomograms can be applied to facilitate timely clinical decisions and promoting personalized treatments, which can ultimately improve patient prognosis. In comparison to other statistical models, nomogram prediction models can easily be applied to clinical decision-making (4). Accordingly, nomogram is often used to forecast the clinical outcomes of cancer patients (5,6). Furthermore, nomograms have recently been used to predict the outcome of nervous system diseases (7,8), including SCI (9).

Although nomogram prediction models are frequently used in cancer studies, their usage in the analyses of hyponatremia post SCI is scarce. In this study, we constructed a nomogram to predict the probability of hyponatremia after SCI. We, therefore, have proposed a tool for the early identification of high-risk patients for hyponatremia post SCI. This prediction model can provide a means for improving patient rehabilitation, shortening the time of hospital stay, and reducing the mortality of patients with SCI.

Materials and methods

Subjects

The study is a retrospective cohort study. Based on the screening of the electronic medical record system of the First Affiliated Hospital of Guangxi Medical University between January 1, 2016 and December 31, 2020, a total of 895 SCI patients were retrieved. After excluding patients with incomplete data, 883 patients were finally included in this study, which included 630 men and 253 women of a mean age 50.14 years. The patient inclusion criteria were as follows: (1) the patients who met the diagnostic criteria for SCI based on the imaging and clinical data, (2) the patients whose medical records were completely available. The patient exclusion criteria were as follows: (1) nervous system diseases that may cause autonomic dysfunction such as Parkinson's disease and Shy-Drager syndrome, and (2) incomplete medical records (10).

This study was approved by the Medical Ethics Committee of First Affiliated Hospital of Guangxi Medical University on November 28, 2022, with approval number 2022-E414-01, and study title is analysis of risk factors related to SCI complicated with hyponatremia and establishment of Nomogram prediction model and verification. This study is a retrospective clinical study and does not require the patient to sign an informed consent form, followed the ethical standards of the responsible committee on human experimentation (institutional or regional) and with the Helsinki Declaration of 1975.

Data collection

Sample size for data collection: This study used univariate and multivariate regression models to analyze the correlation between various research indicators and the incidence of hyponatremia after spinal cord injury, with a total of 12 input variables included. According to the principle of logistic regression analysis, the sample size should be 10–15 times the number of input variables (11). In this study, based on 15 folds estimation, the minimum sample size for SCI cases is 180. This study included 883 SCI patients, meeting the sample size requirements.

To determine the most important features, SCI patients in 2016–2019 were identified as the training set, and patients in 2020 were identified as the test set. Analyze the development set to determine the clinical characteristics for predicting each outcome. Then, using the most important variables identified from the development data, create the final predictive model and nomogram, and evaluate the predictive performance of the final model in the validation set (12).

We conducted demographic data analysis (such as age and sex) on the included patients, analyzed their post SCI medical history and related comorbidities, evaluated and analyzed the severity of post SCI injuries (such as complete or incomplete injuries), and evaluated the level of nerve damage. We record whether there are complications after SCI, such as pneumonia, fever, degree of injury, urinary tract infection, intestinal obstruction, constipation, etc. The selection of these related variables is based on previous relevant research (13,14).

We have collected dynamic changes in blood cytological indicators such as blood routine, hypersensitive C-reactive protein, electrolytes, and liver and kidney function of patients monitored by the laboratory department of our hospital. Based on clinical observations and previous literature reports, select relevant factors that may affect hyponatremia after SCI for quantification and assignment.

The included cases were assigned to either hyponatremia or non-hyponatremia group. The clinical data collected were as follows: (1) sex, age, completeness of SCI, pneumonia, urinary tract infection, fever, constipation, diabetes, diarrhea, (2) laboratory tests: white blood cells (WBC), albumin (ALB), potassium (K), and calcium (Ca) in serum (15).

Statistical analysis

SPSS 26.0 statistical software was used for statistical analysis. The measurement data were expressed as mean ± standard. Independent sample t-test was used for comparison between groups. The count data were expressed as a percentage, using chi-square test, using SPSS software. Among the included subjects, the patients in 2016–2019 were the training set, and in 2020 were the test set. We used the R software to run LASSO (least absolute shrinkage and selection operator) regression analysis based on glmnet package. And then we used the LASSO to analyze the risk factors of SCI combined with hyponatremia. The LASSO regression analysis was used to further screen the model parameters. The variables with zero regression coefficient were excluded from the model, and the variables with the strongest correlation were selected to construct the prediction model and displayed in the nomogram. The model was verified by the internal validation method of the training set (including ROC, DCA, C-index). The discrimination (ROC curve), calibration curve and clinical decision curve (DCA curve) were drawn, and the consistency index (C-index) was calculated.

Results

Overall, 883 patients who met the inclusion and exclusion criteria were included in this study. The patients were assigned to two test groups: non-hyponatremia (n = 576) and hyponatremia (n = 307), respectively. Table 1 represents the clinical and laboratory features of the two test groups. The incidence of hyponatremia was 34.76%.

Table 1 Analysis of general clinical data of spinal cord injury with hyponatremia.

Clinical indicators	Hyponatremia group (n = 307)	Non hyponatremia group (n = 576)	Total (n = 883)
Sex
Male	240 (21.8%)	390 (32.3%)	630 (71.35%)
Female	67 (78.2%)	186 (67.7%)	253 (28.65%)
Age
≥60y	91 (29.6%)	162 (28.1%)	253 (28.56%)
<60y	216 (70.4%)	414 (71.9%)	630 (71.35%)
completeness of SCI
Yes	188 (61.2%)	100 (17.4%)	288 (32.62%)
No	119 (38.8%)	476 (82.6%)	595 (67.38%)
Pneumonia
Yes	189 (61.6%)	140 (24.3%)	329 (37.26%)
No	118 (38.4%)	436 (75.7%)	554 (62.74%)
Urinary tract infection
Yes	78 (25.4%)	89 (15.5%)	167 (18.91%)
No	229 (74.6%)	487 (84.5%)	716 (81.09%)
Fever
Yes	228 (74.3%)	184 (31.9%)	412 (46.66%)
No	79 (25.7%)	392 (68.1%)	471 (53.34%)
Constipation
Yes	94 (30.6%)	105 (18.2%)	199 (22.54%)
No	213 (69.4%)	471 (81.8%)	684 (77.46%)
DM
Yes	32 (10.4%)	24 (4.2%)	56 (6.34%)
No	275 (89.6%)	552 (95.8%)	827 (93.66%)
WBC (10^9/L)
<3.5 or >9.5	244 (79.5%)	273 (47.4%)	517 (58.55%)
3.5–9.5	63 (20.5%)	303 (52.6%)	366 (41.45%)
Albumin (g/L)
<40 or >55	262 (85.3%)	392 (68.1%)	654 (74.07%)
40–55	45 (14.7%)	184 (31.9%)	229 (25.93%)
Serum K (mmol/L)
<3.5 or >5.3	25 (8.1%)	34 (5.9%)	59 (6.68%)
3.5–5.3	282 (91.9%)	542 (94.1%)	824 (93.32%)
Serum Ca	(mmol/L)
<2.08 or >2.6	171 (55.7%)	132 (22.9%)	303 (34.31%)
2.08–2.6	136 (44.3%)	444 (77.1%)	580 (65.69%)
Diarrhea
Yes	22 (7.2%)	17 (3%)	39 (4.42%)
No	285 (92.8%)	559 (97%)	844 (95.58%)

WBC: white blood cell; DM: diabetes.

The typical conditions of SCI are depicted in Table 1.

The flowchart of the article is shown in Figure 1.

Figure 1 The flow chart.

All available indicators of non-hyponatremia and hyponatremia groups were analyzed by LASSO regression. LASSO regression was applied for drawing binomial deviance (Fig. 2(A)) and coefficient (Fig. 2(B)). We constructed a risk prediction model using the R language equation method based on the Xiantao Academic Online Website (16). There is a significant correlation between sex, completeness of SCI, pneumonia, urinary tract infection, fever, constipation, WBC, albumin, and serum Ca²⁺. Finally, a nomogram was constructed to predict the hyponatremia risk in patients with SCI (Fig. 3).

Figure 2 All perioperative parameters were calculated in LASSO analysis. A. Binomial deviance was plotted using the LASSO binary logistic regression model, and 9 parameters were statistically significant. B. Coefcient profles of the 9 features were plotted using the LASSO binary logistic regression model.

Figure 3 A novel nomogram was constructed to predict hyponatremia after SCI risk by calculating the total score of 9 parameters.

The AUC (ROC curve) of the training set was 0.82 (Fig. 4(A)), and test set was 0.79 (Fig. 4(B)). Calibration curves of the predictive hyponatremia after SCI risk nomogram were shown in Fig. 5, the diagonal dotted line represents a perfect prediction by an ideal model, the solid line represents the performance of the training set (Fig. 5(A)) and test set (Fig. 5(B)), and the results indicating that a closer fit to the diagonal dotted line represents a better prediction.

Figure 4 Receiver operating characteristic curve (ROC) validation of the hyponatremia after SCI. The y-axis represents the true positive rate of the risk prediction, the x-axis represents the false positive rate of the risk prediction. The thick blue line represents the performance of the nomogram in the training set (A) and test set (B).

Figure 5 Calibration curves of the predictive hyponatremia after SCI risk nomogram. The y-axis represents actual diagnosed cases of hyponatremia after SCI, the x-axis represents the predicted risk of hyponatremia after SCI. The diagonal dotted line represents a perfect prediction by an ideal model, the solid line represents the performance of the training set (A) and test set (B), and the results indicating that a closer fit to the diagonal dotted line represents a better prediction.

The DCA from the training set is shown in Fig. 6A, and from the test set shown in Fig. 6B. The C-index of the nomogram was 0.81, Together, the data suggested that the model exhibits better predictive ability.

Figure 6 The decision curve analysis. The horizontal axis represents the risk probability threshold, the vertical axis represents the net benefit, and each curve represents the change of the net benefit of each variable (each model) with the risk threshold. All: represents the intervention of all populations, none: represents no intervention. (A) From the training set, (B) from the test set.

One typical case with SCI is illustrated in Fig. 7.

Figure 7 Radiographic examination of the patient with SCI. (A) Lateral radiographs, the arrow points to the 4/5 cervical dislocation. (B) Magnetic resonance imaging (MRI) in the sagittal position, the arrow points to injured cervical spinal cord. (C) Computed tomography (CT) in the sagittal position, the arrow points to the 4/5 cervical dislocation.

Discussion

SCI-induced hyponatremia is an under-recognized disease and has not been mentioned in the European Guidelines for Hyponatremia (17). Hyponatremia is one of the risk factors for early death in patients with acute severe traumatic cervical SCI (13).

There are several reasons for SCI combined with hyponatremia. The existing research methods are mostly based on retrospective analysis of past cases. Various studies have reported that severe hyponatremia is associated with traumatic brain injury, higher American Spinal Cord Injury Association injury score, bradycardia, vasopressors, and nosocomial pneumonia (18). However, some studies suggested traumatic cervical SCI as the only factor significantly associated with hyponatremia.¹⁴ Furthermore, Song et al. reported that MRI scan representing bleeding changes are closely related to the occurrence of hyponatremia, while hypotension is also a reasonable predictor of hyponatremia (19). Although these findings are consistent with our observations, they cannot accurately determine the specific influence of different risk factors on hyponatremia and cannot accurately predict the probability of occurrence of hyponatremia. Therefore, making it difficult to propose better clinical decisions.

Hyponatremia can result in serious consequences in patients with SCI. Therefore, it is imperative to develop suitable and effective prediction tools for determining the probability of its incidence in such patients. Nomogram is one such prediction model that facilitates making better clinical decisions. In the present study, independent risk factors were identified by LASSO, and a nomogram was constructed to quantify and visualize the impact of different risk factors on the outcomes. The risk factors were expressed in line segments with scales, which could accurately predict the probability of occurrence of the outcomes. The present study provides a new method for predicting the risk of hyponatremia after SCI. Our data implies that sex, complete SCI, pneumonia, urinary tract infection, fever, constipation, WBC, ALB and Ca²⁺ are important risk factors for hyponatremia after SCI. These parameters can be used by physicians for the better management of SCI patients susceptible to hyponatremia and therefore facilitate timely and effective intervention to avoid delays or interruptions in the treatment.

In this study, the incidence of hyponatremia was 38.10% (240/630) in males and 26.48% (67/253) in females, suggesting that sex was one of the risk factors for hyponatremia. Previous studies on patients with cervical SCI have shown that sex is not risk factor for hyponatremia (19), which is inconsistent with this study. It is considered to be related to the small sample size of previous studies and only cervical SCI. Hyponatremia is associated with longer hospital stay, but is not associated with increased mortality (20). Studies suggest that complete SCI is the only factor significantly associated with hyponatremia,¹⁴ this is similar to our findings.

Hyponatremia is one of the most common electrolyte disorders in hospitalized patients with pneumonia and is closely related to pneumonia (21). In the current guidelines for community-acquired pneumonia (CAP), hyponatremia is one of the criteria for the severity of pneumonia, while hypernatremia is not specified (21). Our study found that hyponatremia is also closely related to pneumonia in the SCI population. Respiratory failure, pneumonia and pleural effusion are common serious complications in SCI patients (22), we should be more alert to the occurrence of hyponatremia in patients with these complications (23). Fever and intestinal constipation can lead to fluid loss and increase the risk of hyponatremia. SCI is related to fever, and literature reports suggest that high fever is an independent factor in the occurrence of hyponatremia in cervical SCI. After cervical SCI, the regulatory function of the temperature regulation center is disrupted, which easily leads to central high fever. After high fever, there is obvious dehydration, and patients are prone to water and electrolyte metabolism disorders. Hyponatremia is prone to occur, which is consistent with this study. Patients with concurrent urinary tract infections may experience fever, which can easily lead to electrolyte imbalance and hyponatremia. A single center clinical study in Japan showed that in-hospital pneumonia and severity of injury are one of the risk factors for hyponatremia in SCI patients (17). The incidence rate of hyponatremia in patients with elevated WBC count in this study is 79.5% (Table 1). The elevated WBC count is often associated with inflammation and infection. The WBC count in patients with pneumonia and urinary tract infection is high. Therefore, the WBC count is also a related factor to predict hyponatremia. This study found that 85.3% of patients with abnormal albumin had hyponatremia, while 14.7% of patients with normal albumin had hyponatremia. Bajaj et al. (24) conducted a study on hyponatremia and albumin, and the results showed that the cure rate of hyponatremia in the albumin infusion group was significantly higher than that in the control group, on the contrary, low albumin could lead to water and sodium retention aggravating hyponatremia, which was basically consistent with the conclusions of this study. And we also discovered that, inappropriate ion concentration (hyponatremia) will perturb homeostasis and have significant and profound impact. These changes also impact osmotic pressure and the concentration of other metal ions, such as the serum Ca²⁺. We will pay more attention to the relationship between sodium and calcium ions in electrolyte disorders in our future research (25).

This study is based on single-center clinical data. We plan to conduct multi-center clinical research in the future.

Conclusions

The nomogram constructed in this study will be the first of its kind that could help clinicians assess the risk of hyponatremia in SCI patients. Through this nomogram, clinicians can analyze individual risk factors and adopt necessary medical interventions in a targeted manner.

Disclaimer statements

Contributors Qian Wei and Xuefeng Lu: Original draft, formal analysis, validation, data curation. Haifeng Bu, Yikai Chen, Sijing Tuo, Xiaoxia Ye, Laoyi Geer, Xiuwei Tan, Jiling Wang, Yanlan Wu: Investigation. Jichong Zhu, Jie Jiang, and zihong Yang: Software. Yiji Su and Song Fangming: Conceptualization, methodology, writing - review & editing, supervision, project administration. Su Yiji is the corresponding author, and Song Fangming is the co -corresponding author.

Conflicts of interest The authors declare no competing interests.

Additional information

Funding

This study was funded by National Natural Science Foundation of China (NSFDA grant: No.81960773), Open project of Guangxi Key Laboratory of regenerative medicine (GUI Zai Zhong Kai: 201705), and Guangxi Health and Family Planning Commission 2017 self-financing research projects (No. Z20170565), Key Laboratory of Basic Research of Traditional Chinese Medicine, Guangxi University of Traditional Chinese Medicine (16-380-58-03).