Risk Factors for Postoperative Complications and In-Hospital Mortality Following Surgery for Cervical Spinal Cord Injury

Acute cervical spinal cord injury (SCI) is a devastating injury that requires highly complex and sub-specialized multidisciplinary care. The estimated incidence of traumatic SCI in Australia is 21.0-32.3 per million population [1]. The rates are highest in young adult males, with the majority of injuries caused by accidental trauma [2,3]. The effects of surgery are both immediate and long-lasting. Decompression of the spinal cord may prevent the neurotoxic biochemical cascade of secondary injury mechanisms, while stabilization prevents the morbidity associated with chronic deformity [4-7].

Postoperative complications following surgery for cervical SCI can contribute to early mortality, affect neurological recovery and functional outcomes, and result in high healthcare costs [8]. Respiratory complication following cervical SCI is the main cause of in-hospital mortality, responsible for 42-87% of deaths [9-11]. Specialized respiratory care for cervical SCI patients can improve outcomes [12]. Increased risk of mortality has been associated with the development of pulmonary embolism (PE), which is most likely to occur in the immediate post-injury period [13]. Pressure ulcers (PU) are the most common complication of SCI and a major cause of chronic re-hospitalisation [14-16]. Currently, the levels of evidence about risk factors for the development of PU in the acute period remain insufficient [16].

Patient factors including age and comorbidities can influence the risk profile of SCI patients [9,10,17-19]. Complications in elderly patients are more likely to be cardiorespiratory in origin [myocardial infarction, deep vein thrombosis (DVT), PE] whereas infections (pneumonia, respiratory and urinary infection) are more frequent in younger groups of patients [10,17]. Studies have demonstrated a strong association between complete SCI and increased in-hospital mortality [9,11,19-21]. In-hospital mortality for patients aged over 65 years with complete cervical SCI is five times higher than those with incomplete injuries [19]. The severity of polytrauma has also been significantly associated with postoperative in-hospital mortality among SCI patients [18,21].

Surgical factors also influence complication rates. Existing evidence supports improved neurological recovery among cervical SCI patients undergoing early surgery [22], as well as lower in-hospital mortality [9]. A growing body of evidence supports early surgery as a means to reduce the time spent in intensive care and the overall hospital length of stay [23]. Medical complications such as PE may also be reduced by the early mobilization achieved by surgery [23]. Longer surgical duration is associated with increased postoperative complications [24]; however, anterior and posterior approaches may be equivalent in terms of outcomes in the long term [25]. Posterior approaches may result in greater blood loss, longer operative times, and longer inpatient admission duration [26].

There is a paucity of evidence regarding the association of these factors with in-hospital mortality and postoperative complications from an Australian cohort. Hence, this study aimed to investigate these associations by using a predictive model.

An overview of the study procedure is outlined in Figure 1.

Figure
1:
Flowchart depicting the study procedure

EMR: electronic medical records; ICD-10-AM: the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification

Ethical approval

The authority to access patient data through electronic medical records (EMR) was granted by the Northern Sydney Local Health District (NSLHD) Hospital Research Ethics Committee (HREC).

Study population

The Royal North Shore Hospital (RNSH) Spinal Cord Injury Service is one of two hospitals in the state of New South Wales that provides acute surgical and rehabilitation services for acute traumatic SCI patients. The RNSH caters to roughly half the population of NSW, which has a total population of 8.1 million people [27]. Adult patients over 16 years of age who sustained acute, traumatic cervical SCI from June 1, 2016, to June 1, 2021, and underwent operative management at RNSH under the orthopaedics team were eligible for inclusion in the study. The inclusion and exclusion criteria are outlined in Table 1.

Data collection

The hospital coding centre was approached with a request for data on admissions during the study period with unique billing codes. These codes were chosen from the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM) [28] and are listed in Table 2.

This process identified 88 patients who fulfilled the inclusion criteria (Figure 2). Additionally, a manual search of all emergency operations documented in the operating theatre EMR was performed using the data collection tool, Discern Analytics 2.0 (DA2) within Surginet (Oracle Cerner, 2016, Austin, TX), which yielded an additional four patients missed by hospital coding. The triage note, admission notes, operation reports, and medical imaging were reviewed and relevant data were extracted to a pre-designed Microsoft Excel spreadsheet (Microsoft Corporation, Redmond, WA).

Figure
2:
Flow diagram of sample selection

ICD-10-AM: the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification; SCI: spinal cord injury

Patient age, sex, and comorbidities including smoking, hypertension, diabetes, anticoagulation, chronic alcohol use, and coronary disease, and the American Society of Anesthesiologists (ASA) classification and American Spinal Cord Injury Association (ASIA) classification were recorded. The Injury Severity Score (ISS) was calculated manually based on the admission notes and imaging reports. Surgical factors including time to surgery, operative time, number of levels instrumented, and surgical approach were extracted from the EMR and validated against the intraoperative imaging to ensure accuracy. Postoperative adverse outcomes recorded included in-hospital mortality, any type of cerebrovascular accident (CVA), venous thromboembolic disease (DVT/PE), pressure injury, respiratory infection, urinary infection, and ventilator dependency.

Data analysis

All data analysis was performed using “R” (Ver 4.2.2 The Free Software Foundation Inc. Boston, MA). The Shapiro-Wilk test for normality was used, demonstrating that the data was not normally distributed. Continuous variables were expressed in terms of the median and interquartile range (IQR). Categorical variables were expressed as a percentage of the total sample number. Continuous variables were changed to categorical variables for the purpose of univariate analysis using the Chi-squared test.

Statistical model

A univariable selection method of logistic regression modeling was used. Models were constructed using all independent variables from the univariate analysis with p<0.20 [29,30]. Collinearity was assessed using the variance inflation factor (VIF), with VIF >10 used to define collinear variables. This threshold was chosen as it has been used in similar studies [9,31]. If collinearity was detected, the offending variable was removed from the analysis. This process was used to separately create regression models for each dependent variable. Independent variables were equally weighted within each model. The dataset was randomly split in a ratio of 70:30 for the purposes of creating training and testing subsets. These subsets were used to calculate the area under the curve (AUC) and McFadden’s R² for each model.

Study population

A total of 92 patients met the inclusion criteria. The median patient age was 54.5 years (IQR: 32.5); 77 (82.2%) of the patients were male. The median ASA score was 3 (IQR: 2), with comorbidities including smoking (n=16, 17.2%), hypertension (n=26, 28.0%), diabetes (n=12, 13.0%), anticoagulation (n=4 4.3%), chronic alcoholism (n=6, 6.5%), and coronary disease (n=10 10.7%). The median ISS was 20 (IQR: 9). The median cervical level of injury was C4, and the median ASIA classification was ASIA C. The median number of levels instrumented was 3 (IQR: 3) with greater than 5 levels instrumented in 15 (16.3%) cases. The median surgical time was 171 minutes (IQR: 130).

Postoperative complications included CVA (n=1, 1.1%), DVT/PE (n=33, 35.5%), PU (n=5, 5.4%), respiratory infection (n=28, 30.1%), urinary tract infection (n=25, 26.9%), and ventilatory dependence (n=17, 18.3%). In-hospital mortality following surgery was 6.5% (n=6). Of these patients, the primary cause of death was respiratory failure in 83.3% (n=5).

The basic characteristics of the sample are presented in Table 3.

Univariate analysis

In-hospital mortality was associated with age >65 years (Χ²=2.6, p=0.107), hypertension (Χ²=2.9, 0.091), anticoagulation (Χ²=6.6, p=0.01), chronic alcoholism (Χ²=3.6, p=0.058), coronary disease (Χ²=6.2, p=0.012), complete injury (Χ²=6.4, p=0.011), polytrauma (Χ²=2.7, p=0.099), greater than 5 levels of instrumentation (Χ²=3.02, p=0.0819), anterior instrumentation (Χ²=1.9, p=0.174), in-hospital thromboembolic disease (Χ²=2.1, p=0.145), and ventilator dependency (Χ²=13.6, p<0.001). Postoperative CVA was associated with chronic alcoholism (Χ²=3.1, p=0.076). Postoperative DVT/PE was associated with polytrauma (ISS >15) (Χ²=2.0, p=0.156). Postoperative pneumonia was associated with male gender (Χ²=3.5, p=0.06), age >65 years (Χ²=2.6, p=0.064), hypertension (Χ²=2.9, 0.085), complete injury (Χ²=2.6, 0.009), and polytrauma (Χ²=2.6, 0.002). Urinary tract infection was associated with high-level C-spine injury (C1-4) (Χ²=2.6, p=0.113). Ventilator dependence was associated with complete injuries (Χ²=19.6, p<0.001), and polytrauma (Χ²=12.1, p<0.001). These results are presented in Tables 4–10.

Multivariable analysis

The independent variables used to create a regression model for in-hospital mortality were as follows: age >65 years, hypertension, alcoholism, anticoagulation, coronary disease, complete injury, ventilator dependency, and PE/DVT. The process of collinearity assessment and model selection is outlined in Table 11. The results of the chosen regression model for in-hospital mortality are depicted in Figure 3. In this model, complete injury was significantly associated with in-hospital mortality [odds ratio (OR): 184.53, 95% confidence interval (CI): 2.41-14106.65, p=0.018]. The AUC for this model was 0.9615 and R²=0.58.

Figure
3:
Plot of logistic regression model for in-hospital mortality

*Significant at p<0.05

Formula: in-hospital mortality ~age >65 + HTN + ETOH + anticoagulation + IHD + complete injury + levels Instrumented

CI: confidence interval; ETOH: chronic alcoholism; HTN: hypertension; IHD: ischemic heart disease; ISS: Injury Severity Score; OR: odds ratio

Significant results

Our strongest finding pertaining to in-hospital mortality was its association with anticoagulation, coronary disease, complete injury, and ventilator dependency. The logistic regression model found complete injury to be significantly associated with postoperative in-hospital mortality. Studies have demonstrated a significant association between a high number of medical comorbidities and in-hospital mortality following SCI [10,19,32]. However, there is a paucity of evidence as to which specific chronic comorbidities contribute the most to postoperative complications in these patients. We have shown that in-hospital mortality risk after surgery for cervical SCI may be more strongly associated with cardiovascular comorbidities than diabetes and chronic alcohol abuse. We found that postoperative pneumonia and ventilator dependency was significantly associated with complete injury and polytrauma, as has been shown previously [33]. These findings highlight the vigilance required to prevent deterioration from respiratory decompensation in the ASIA A cohort of patients. Evidence supports providing specialized respiratory care for cervical SCI patients as a critical component of early management to avoid adverse outcomes [34].

Comparison of our findings with the literature

Table 12 provides a summary of previous studies investigating the risk factors for in-hospital mortality and complications among SCI patients using predictive models. The study population’s mean age of 51 years and its male predominance are consistent with similar studies [10,34-37]. The average level of injury was C4-5, which aligns with other studies [38-40]. However, the sample size of 92 was smaller when compared to larger studies in the literature with over 1000 patients [9,11,17,18,21]. Beck et al. investigated 706 patients with SCI in an Australian population and observed a median age of 50 years and an in-patient mortality of 15%, which is higher than our rate of 6.5%. However, their study was not limited to cervical SCI, and the inclusion criteria limited patients to those suffering from major trauma (ISS >12, ICU admission), which may account for their higher rate of in-hospital death. The majority of deaths in our study were secondary to respiratory complications, which is consistent with the literature [9-11]. Several studies have demonstrated minor differences in early complication rates between patients managed conservatively and those treated surgically [18,33], which may allow us to draw comparisons between our study cohort and nonoperatively managed patients.

Beck et al. showed that Australia is witnessing an increase in the number of older adults with traumatic SCI [41], which is consistent with other studies [36,42]. These findings exceed the rate of geriatric population growth and reflect an increasing rate of trauma from low falls in this age group [43,44]. Most studies modeling risk factors for complications following SCI report increasing age as significantly associated with in-hospital mortality [9,17-20,32,41]. Inglis et al. specifically investigated traumatic SCI in a geriatric population (n=1340) and determined that the odds of dying 50 days post-surgery are six times higher for patients aged ≥77 years versus those aged 65-76 years [19]. Our study did not find a strong association between age and postoperative complications. Some studies in the literature also demonstrate an insignificant relationship between age and postoperative mortality following cervical SCI surgery [33]. Wilson et al. investigated a group of geriatric (age >65 years) cervical SCI patients managed surgically [45]. Mortality at 90 days was significantly associated with complete injury; however, there was no significant association between mortality and age subgroups. A possible explanation for this inconsistency could be selection bias, as older patients are more likely to be conservatively managed without surgery, which avoided their inclusion in the study.

Strengths, limitations, and future research

The main strength of this study lies in the reliability of the dataset, which was rigorously extracted using multiple ICD codes and validated using multiple sources [46]. Evidence suggests significant inaccuracies inherent in hospital coding data [47-50] and underreporting of adult variants of SCI without radiological abnormality (SCIWORA) [51,52]. The data collection methodology employed in this study used multiple validation points between coding, EMR documentation, and medical imaging in an attempt to reduce information bias.

The relatively small sample size of this project likely contributed to statistical errors. A power analysis during the study planning phase would have been useful [53], as it is generally accepted that a larger sample size can improve the validity of a model by reducing the confounding errors caused by multicollinearity [54]. A larger sample size could be derived from local and state-wide trauma registries. A more detailed dataset (including time of death) would allow for Kaplan-Meier survival curves to be drawn and a Cox proportional hazard model to be constructed, as performed in similar studies [10,17]. Logistic regression models may be less precise than a Cox proportional hazard model and may have difficulty in separating the effects of inter-related, time-dependent covariates [55,56]. This model could then be validated on an external, compatible dataset to improve external validity, as has been performed in similar studies [33,57].

This study was limited to a single centre and did not include patients admitted under the neurosurgical team. Additionally, the study design was retrospective and conducted by a single author, reducing overall reliability and allowing for potential bias. As with most retrospective, non-randomised surgical studies, selection bias towards healthier, operative candidates is an inherent factor [58]. The scope of this study did not include nonoperative patients, a group more likely to be palliated secondary to old age and comorbidities, precluding surgical management. Additionally, the scope of this study was limited to patients admitted under orthopaedics. The neurosurgical team admits approximately 30% of acute SCI patients at RNSH and there is evidence to suggest that there is a lack of consensus between neurosurgeons and orthopaedic surgeons on the value of surgical intervention after SCI [59], which may have influenced the results. The grouping of venous thromboembolic disease into PE + DVT may have confounded the association of PE with mortality risk, which has been established in the literature [17]. 

The future of this type of research will involve the development of machine-learning algorithms to better understand non-linear associations between variables and possibly model mortality risk with greater precision. One recent study has suggested the superiority of a machine-learning-derived tool, the SCI Risk Score (SCIRS), which can predict in-hospital and one-year mortality using the following variables: age, neurological level and completeness of injury, fracture classification, and Abbreviated Injury Scale scores. This tool has been validated in a separate cohort and demonstrated superior AUC and sensitivity than ISS [60].

To improve surgical outcomes for patients with traumatic cervical SCI, a concerted effort must be made to prevent postoperative complications. This study investigated the patient, injury, and surgical factors contributing to in-hospital mortality and postoperative complications following surgery for traumatic SCI. Cardiovascular comorbidities including anticoagulation and coronary disease appear to present significant patient risk factors contributing to in-hospital mortality. A logistic regression analysis found complete neurological injury to be significant in predicting in-hospital mortality. Patient age appears to insignificantly influence postoperative complication rates; however, this finding may have been influenced by selection bias. Postoperative respiratory complications, especially in the setting of complete injury and ventilator dependency, are particularly devastating. The findings of this study endorse the existing literature advocating for vigilant respiratory management in these fragile patients.