Background A consistent approach to cervical spine injury (CSI) clearance for patients 65 and older remains a challenge. Clinical clearance algorithms like the National Emergency X-Radiography Utilisation Study (NEXUS) criteria have variable accuracy and the Canadian C-spine rule excludes older patients. Routine CT of the cervical spine is performed to rule out CSI but at an increased cost and low yield. Herein, we aimed to identify predictive clinical variables to selectively screen older patients for CSI.
Methods The University of Iowa’s trauma registry was interrogated to retrospectively identify all patients 65 years and older who presented with trauma from a ground-level fall from January 2012 to July 2017. The relationship between predictive variables (demographics, NEXUS criteria and distracting injuries) and presence of CSI was examined using the generalised linear modelling (GLM) framework. A training set was used to build the statistical models to identify clinical variables that can be used to predict CSI and a validation set was used to assess the reliability and consistency of the model coefficients estimated from the training set.
Results Overall, 2312 patients ≥65 admitted for ground-level falls were identified; 253 (10.9%) patients had a CSI. Using the GLM framework, the best predictive model for CSI included midline tenderness, focal neurological deficit and signs of trauma to the head/face, with midline tenderness highly predictive of CSI (OR=22.961 (15.178–34.737); p<0.001). The negative predictive value (NPV) for this model was 95.1% (93.9%–96.3%). In the absence of midline tenderness, the best model included focal neurological deficit (OR=2.601 (1.340–5.049); p=0.005) and signs of trauma to the head/face (OR=3.024 (1.898–4.815); p<0.001). The NPV was 94.3% (93.1%–95.5%).
Conclusion Midline tenderness, focal neurological deficit and signs of trauma to the head/face were significant in this older population. The absence of all three variables indicates lower likelihood of CSI for patients≥65. Future observational studies are warranted to prospectively validate this model.
- clinical assessment
- fractures and dislocations
Data availability statement
Data are available upon reasonable request. Requests for data should be send to the listed corresponding author.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
What is already known on this subject?
Cervical spine injury (CSI) clearance for patients 65 and older remains a challenge. While clinical clearance algorithms exist, the National Emergency X-Radiography Utilisation Study criteria have variable accuracy in the older population and the Canadian C-spine excludes older adults altogether. CT imaging is the current gold standard for identification of cervical spine fractures in the geriatric population.
What this study adds?
In this study, we identified three predictive clinical variables—midline tenderness, focal neurological deficit and signs of trauma to the head/face, which remain significant in the older population. The absence of all three variables indicates lower likelihood of CSI for patients 65 and older. Our study provides a platform for further research into predictive models for cervical spine imaging in older patients experiencing ground-level falls.
Each year, 2.8 million older adults (≥65 years) are treated in emergency departments (EDS) for fall injuries, which accounts for 10%–15% of all ED visits.1 2 Fall injuries, mainly head injuries and hip fractures, lead to 800 000 hospitalisations a year.3 4 Not only are these injuries serious but they are also costly. In 2015, the medical costs for falls totaled more than US$50 billion in the USA.5
For older adults admitted for fall injuries, healthcare providers must decide whether to obtain imaging studies, including those of the cervical spine. Cervical spine injuries (CSI) can occur in older patients even in lower energy mechanisms of injury such as ground-level falls. In fact, studies show that more than 60% of geriatric CSI are related to falls6 with 50% or more of these injuries associated with ground-level falls.7 8 In older adults, most injuries occur in the upper cervical spine with minor trauma.9 The existence of extensive degenerative changes and deformities makes older adults more vulnerable to injury10 and may hinder CSI diagnosis.11
CT imaging is the current gold standard for identification of cervical spine fractures.12–14 Frequent use of CT is costly and labour intensive and prolongs time in the ED for these older patients. Two widely used predictive models have been developed using clinical information to decide whether CT of the cervical spine is warranted: the Canadian C-spine rule (CCR) and the National Emergency X-Radiography Utilisation Study (NEXUS). Unfortunately, CCR excludes patients 65 and older from clinical clearance.15 16 Although, NEXUS criteria have been applied in the geriatric population,17 published studies show variable sensitivity (66%–100%) in this group of patients.18 Based on inconsistencies in the literature, the American College of Surgeons Trauma Quality Improvement Programme Best Practice Guidelines in Imaging state that neither of these two clinical decision algorithms should be used in patients of 55 and older.14 Critics of using NEXUS have emphasised a decreased sensitivity of the clinical examination in patients who may suffer from baseline dementia or delirium.19 Advocates for clinical decision-making tools point to radiology costs and increased resource utilisation.20 21 Thus, with low energy mechanism of injury and older patients, clinicians are left with using their own judgement, institutional protocols or cervical spine imaging to rule out CSI.20–23
Our aim was to determine whether the NEXUS criteria along with other clinical variables can reliably predict the presence of CSI and need for cervical spine imaging in older patients who are evaluated for a ground-level fall. Our goal was to identify a model which better predicts the necessity of CSI screening via CT for individual patients age 65 and older.
Study design and population
This retrospective cohort study was approved by the University of Iowa Institutional Review Board (IRB number 201706824). A waiver of consent was approved by the University of Iowa Institutional Review Board for all patients included in this retrospective study. The University of Iowa Trauma Registry was queried to identify all patients of 65 years and older who presented to this facility with trauma from a ground-level fall from January 2012 to July 2017. Trauma registries are databases that document acute care delivered to patients hospitalised with injuries, providing information that can be used to improve the efficiency and quality of trauma care.24 Patients were excluded if the fall was from greater than three steps or if the fall was part of another traumatic event such as a motor vehicle accident, if arrival to the institution was greater than 48 hours after the fall (those are direct inpatient transfers who were being evaluated acutely by our facility in the ED, thus not likely to have information detailing cervical spine reports or examination findings) or if there was insufficient patient data to determine whether the fall was a ground-level fall, a fall from stairs or a fall from height. This study followed the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis guidelines.25–27
Patient and public involvement
Development of the research question and outcome measures was informed by previous studies undertaken by the research team. There was no direct contact with patients required as only their medical records were reviewed. Patients were not directly involved in the design, recruitment or conduct of the study. We are unable to disseminate the findings to study participants directly.
Demographics (age, gender, race), comorbidities, mechanism of injury, physical examination findings at initial presentation, imaging studies and injuries including CSI were collected. NEXUS criteria-related data including presence of midline tenderness, focal neurological deficits, altered level of consciousness, evidence of intoxication as determined by elevated blood alcohol level greater than or equal to 80 mg/dL or positive urine drug screen for nonprescription illicit substances, any fractures, injuries resulting in large wounds, major haemorrhage, signs of trauma to the head and face, physical findings suggesting significant blunt trauma to the face, chest, torso or extremities were abstracted from the medical records. For analysis, we categorised these into distracting injuries of the head/face, chest, torso and extremity/long bone fracture categories. For patients who did not have CSI imaging performed on admission, their medical charts were extensively reviewed for any subsequent admission or visits to our institution for cervical spine symptoms or injuries from the original ground-level fall. Abstractors and data analysts were not blinded to the predictors or outcomes.
For this retrospective study, a convenience sampling method was used. Data from patients admitted between January 2016 and July 2017 were used as a training set (TS) to construct the statistical models allowing us to identify clinical variables that can be used to predict CSI. Data from patients admitted between January 2012 and December 2015 constituted the validation set (VS) used to assess the reliability and consistency of the model coefficients estimated from the TS. The generalised linear modelling (GLM) framework was used for our analyses, which allowed us to examine the relationship between a set of predictor variables (representing demographic characteristics, NEXUS criteria and distracting injuries) and each of the dichotomous outcomes, CSI and inability to use physical examination. Models were fit for all subset combinations of the predictor variables with a maximum model order of four, so that the optimal models can be identified for interpretation while avoiding overfitting. The model goodness-of-fits were compared using the Akaike information criterion (AIC)28 of which a smaller value indicates a more appropriate model. For each of our two modelling samples based on midline tenderness availability, the predictor set with the smallest model AIC was deemed optimal for that outcome. For the analysis, we coded midline tenderness as present (yes), not present (no) or unknown. We included all three levels in the modelling, but only provided the estimate for ‘yes vs no’. This was a complete case analysis.
Comparisons of the effect of the predictors across predetermined age strata (65–74, 75–84, 85+) were made using interaction terms between those predictors and the categorical age. Estimates and CIs for the OR of the outcome event were calculated along with p values. The area under the curve (AUC) was calculated for the best-fitted models as determined by AIC. All analyses were performed using SAS V.9.4.
Group comparisons were performed using χ2 tests to compare categorical variables and Mann-Whitney test for age as it was not normally distributed based on the Shapiro Wilk test. (SPSS V.25.0). P values <0.05 were considered significant. There were no missing data.
As shown in figure 1, 3340 patients were admitted for a fall between January 2012 and July 2017. Of those, 1018 patients did not meet our inclusion criteria. The remaining 2312 patients of 65 years and older admitted for ground-level falls were included in this study. Of those, 253 (10.9%) patients had a CSI. Patients who had a CSI were significantly older (83.1±8 vs 80.8±8.4, p<0.001) and were more likely to have fallen from a bed or from a chair than from standing compared with those who did not present with CSI (8.7% vs 4.5% and 7.1% vs 4.8%, respectively, p=0.004). (table 1)
A total of 837 patients (40.7%) in the non-CSI group did not undergo CSI imaging (table 1). A review of their medical records and our trauma registry showed that 49 of those patients (5.9%) were seen by the general surgery trauma service. Patients with no CSI imaging were more likely to be women (64.2% vs 59.1%, p=0.017) and older (81.6±8.2 vs 80.0±8.7, p<0.001). Mechanism of injury and GCS were similar between the patients who had no CSI imaging performed and those who were imaged. Patients who did not undergo imaging were less likely to have head and/or facial injuries (table 2). After extensive chart review of these patients, we found that none returned with cervical spine symptoms or injuries during future visits or admissions from the original ground-level fall to our institution.
CSI predictive models
Using the GLM framework, the best predictive model for CSI (table 3, sample 1) included midline tenderness, focal neurological deficit and signs of trauma to the head/face, with, as expected, midline tenderness significantly predicting CSI (OR=22.961 (15.178–34.737); p<0.001). As not all patients had midline tenderness on presentation due to delirium, sedation or lack of documentation, a second model was created to predict CSI in this population. In the absence of midline tenderness, focal neurological deficit (OR=2.601 (1.340–5.049); p=0.005) and signs of trauma to the head/face (OR=3.024 (1.898–4.815); p<0.001) were associated with CSI (table 3, sample 2). Other associated injuries were not predictive of CSI (data not shown).
Validation of the best predictive models
Our models were validated in two ways. First, we validated the models using the overall validation population. Second, we stratified our patient population based on age (65–74; 75–84; 85 and older). As shown in table 4, characteristics of patients included in the VS were similar to those of patients included in the TS. There was no significant difference in terms of gender, age, mechanism of injury, GCS injuries, midline tenderness and CSI. There were significantly more patients with signs of trauma to the face and altered level of consciousness and significantly less patients with focal neurological deficit in the VS than in the TS. Using a CSI predicted probability threshold of 0.3, in the presence of midline tenderness, our model showed a sensitivity of 61.1% (53.2%–68.9%), a specificity of 93.7% (92.3%–95.0%) and a negative predictive value (NPV) of 95.1% (93.9%–96.3%). The AUCs were good and similar for the training and TS (0.822 and 0.832, respectively). For patients in the 65 to 74 age groups, this model predicted CSI with a sensitivity of 66.7% (46.5%–86.8%), specificity of 92.3% (89.4%–95.2%) and NPV of 97.7% (96.1%–99.4%). For patients in the 75–84 age group, this model predicted CSI with a sensitivity of 61.4% (48.8%–74.0%), specificity of 93.8% (91.5%–96.0%) and NPV of 94.9% (92.8%–97.0%). In the 85+age group, this model predicted CSI with a sensitivity of 59.2% (47.7%–70.6%), specificity of 94.6% (92.5%–96.7%), and NPV of 93.5% (91.2%–95.8%).
In absence of midline tenderness and using a cut-off of 0.1, our model predicted CSI in the overall population with a low sensitivity (8.1% (2.4%–13.9%)) but with relatively good specificity (97.4% (96.5%–98.2%)) and NPV (94.3% (93.1%–95.5%)). While not as good, the AUCs for the training and TS were similar (0.6534 and 0.6558, respectively). For patients in the 65–74 age group, our model predicted CSI with a low sensitivity of 8.3% (−7.3%–24.0%], but with a specificity of 97.3% (95.6%–98.9%) and NPV of 97.0% (95.3%–98.7%). For patients in the 75–84 age group, our model predicted CSI with a sensitivity of 65.6% (49.2%–82.1%), specificity of 59.6% (55.2%–64.0%) and NPV of 96.3% (94.2%–98.5%). For patients in the 85+ age group, our model predicted CSI with a low sensitivity of 2.4% (−2.2%–7.0%) but with a specificity of 96.3% (94.7%–98.0%) and NPV of 92.0% (89.7%–94.4%).
As shown in table 5, when this model was applied to our entire study patient population (training and validation datasets), we were able to detect CSI in 219 out of 253 patients diagnosed with CSI by imaging (86.6% sensitivity, 54.5% specificity and 97.1% NPV). In comparison, NEXUS detected 227 of 253 patients diagnosed with CSI by imaging (89.7% sensitivity, 9.4% specificity and 88.1% NPV). In patients 65–74 years old, our model had a better sensitivity, specificity and NPV than NEXUS. In patients 75–84 years old, our model and NEXUS presented similar sensitivities, but our model had better specificity and NPV than NEXUS. In the 85 and older subgroup, our model was less sensitive than NEXUS but had better specificity and NPV (table 5). Using our model, 315 of 1475 patients (21%) who underwent cervical spine imaging would have been clinically cleared and would not have received cervical spine imaging.
In this retrospective study, we identified specific clinical findings associated with higher likelihood of CSI in older suffering ground-level falls, including some of the NEXUS criteria as well as additional variables (signs of trauma to the head and face). This model demonstrated the best NPV and sensitivity for detecting CSI in patients 65–74 years of age. When used on our overall population, this model presented an NPV of 97.1% with a sensitivity of 86.6%. The analysis of this cohort of patients 65 and older sustaining ground-level falls provided a predictive model that may allow clinicians to determine which patients would benefit from CT imaging to identify CSI after ground-level falls. This study emphasises the significance of good clinical examination in the era of evidence-based medicine, which may guide further imaging in this population and reduce the number of CT scans, thereby decreasing costs.
Clinical clearance has been criticised in the geriatric population for various reasons. Some point to a higher incidence of CSI in the older especially at C1–C2 level with inability to detect these injuries on physical examination.9 29 In addition, the use of the NEXUS criteria in older patients was found to have insufficient sensitivity and NPV to be clinically useful.16 18 Moreover, Tran et al 23 sought to validate modified NEXUS criteria for older patients. The modified criteria defined distracting injuries as only signs of trauma to the head and required that the patients have baseline level of mentation. This study found that the modified criteria had a 100% sensitivity and NPV when used in the older adults presenting to the ED with report related to a fall. However, only 10 patients were found to have CSI and its validity is limited to patients with baseline neurologic status as per their family member or chronic care facility staff. Patients were excluded if they were determined to have an acute change in baseline neurologic functioning as per the physician caring for the patient, including clinical intoxication. An additional factor that limits clinical clearance is that older patients may suffer baseline dementia or delirium that will affect their level of alertness and may make physical examination difficult. The NEXUS clearance tool uses the presence of a distracting injury, which is hard to standardise across providers given the subjectivity in interpretation. On the other hand, critics of routine cervical spine imaging cite increased resource utilisation and radiology costs as reasons to image selectively.20 21 This divided approach has led to variability among providers and protocols for cervical spine clearance in this population.
This variability in approach to CSI clearance was observed at our institution as 36% of the patients evaluated after ground-level falls were cleared clinically. No significant difference in mechanism of injury was observed; however, patients who were cleared using imaging were more likely to present with associated head and facial trauma. Of the population who received imaging after a ground-level fall, the incidence of CSI was 17%. When our model was applied to the study population, 21% of patients would be clinically cleared with no indication for cervical spine imaging.
Our model provides a foundation for further research into predicting which patients over age 65 need cervical spine imaging and which patients could be safely evaluated and cleared without the need for routine cervical spine imaging after a ground-level fall. Herein, we sought to identify objective criteria to distinguish the optimal clinical examination findings associated with CSI that can be easily adopted across providers and institutions when evaluating patients presenting after ground-level falls. To our knowledge, this study is one of the first to develop a cervical spine predictive tool specifically for the geriatric population who sustained a ground-level fall. Moreover, this is the first to report on the sensitivity, specificity and NPV of such a model when stratifying by age.30 When applied to our population, CSI could be clinically detected in all 39 patients 65–74 years of age who were originally diagnosed by imaging. Our data suggest that a cut-off age of 75 may be worth further study. Traditionally, an older adult is defined as having a chronological age of 65 years or more. As life expectancy has increased, medical providers are seeing patients 65 and older that represent a diverse population with a wide spectrum of functional ability and general health.30 Providers will need to adapt to this new reality and use objective data for management rather than treating all patients in this age group as the same.
Our study presents some limitations. First, the data were acquired through retrospective chart review, which introduces a risk of bias due to poor documentation. Second, this study was conducted at a single institution and, therefore, our results may not be directly applicable in other environments until validated prospectively at multiple institutions. We did not assess whether frailty and/or comorbidities affected our prediction models. Outside of patients coming from facilities, complete medical histories including medications are not always available to providers during the initial ED evaluation. Future studies are warranted to assess whether frailty and/or specific comorbidities play a role in the providers’ ability to perform clinical cervical spine clearance after ground-level falls. Additionally, this study was limited in that not all patients underwent cervical spine imaging at the time of presentation. Much consideration went into the decision to include or not include these patients. With extensive chart review of these patients, we did not find any patients who returned with cervical spine symptoms or injuries during future visits or admissions from the original ground-level fall. Exclusion of these patients would create additional selection bias as these patients are part of the population in question, though they may have been perceived as less critical at the time of triage. Our results for predictive factors have been reproduced in other analyses of trauma patients; however, populations were more heterogeneous or involved other types of trauma in addition to ground-level falls, hindering comparison. Despite these limitations, we provide evidence suggesting that not all older patients need cervical spine imaging after ground-level falls. Due to the retrospective nature of this study and because it was conducted at a single institution, our results are not directly transferable to practice. Future prospective multicentre studies are warranted to validate this model. However, our study provides a platform for further research into predictive models for cervical spine imaging in older patients experiencing ground-level falls with the goal of reducing unnecessary imaging and healthcare costs.
Our study demonstrates that midline tenderness, focal neurological deficit and signs of trauma to the head/face remain significant in the older population after ground-level falls and can be used to clinically rule out CSI. While less sensitive, in the absence of midline tenderness, focal neurological deficit and signs of trauma to the head/face remain significant and can be used to predict patients at higher risk of having CSI after ground-level fall. Future prospective, multicentre studies are warranted to validate this predictive model for imaging of the cervical spine after ground-level falls and to determine its applicability to other mechanisms of injury in the geriatric population.
Data availability statement
Data are available upon reasonable request. Requests for data should be send to the listed corresponding author.
Patient consent for publication
Handling editor Edward Carlton
JE and PZ contributed equally.
Presented at This paper was presented in part at the Region 7 American College of Surgeons Committee on Trauma Resident Paper Competition in both the 2018 (State and Regional level) and the 2019 (State Level) competitions.
Contributors All authors contributed equally to this manuscript. Study concept and design: PZ, JE, CG, DS. Acquisition of the data: ML, JJ, JE, PZ. Analysis and interpretation of the data: YZ, PT-E, PZ, JE, CG, DS. Drafting of the manuscript: JE, PZ, CG, DS. Critical revision of the manuscript for important intellectual content: JE, CG, DS. Statistical expertise: YZ, PT-E.
Funding This study was supported in part by the University of Iowa Institute for Clinical and Translational Science, which is granted with Clinical and Translational Science Award funds from the National Institutes of Health (UL1TR002537). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.