Grauer JN

Venous thromboembolic events (VTEs), encompassing both deep vein thrombosis (DVT) and pulmonary embolism (PE), are potentially fatal events that can occur after orthopedic surgery.¹ In patients who do not receive prophylaxis, VTE incidence can be as high as 70% for total hip arthroplasty,² 26% for hip fracture,³ and 5% for ankle fracture.⁴ Based on the relatively low incidence of VTE after ankle fractures and insufficient evidence for VTE prophylaxis in this population, the American Orthopaedic Foot and Ankle Society and the American College of Chest Physicians do not recommend routine screening or prophylaxis for VTE in patients with ankle fractures.^1,5 Nevertheless, certain patients may be at increased risk for VTE after open reduction and internal fixation (ORIF) of an ankle fracture. In such cases, further consideration for prophylaxis may be warranted.

Other studies of VTEs have identified general risk factors of increased age, obesity, prior thromboembolic disease, oral contraceptive use, multitrauma, varicose veins, and prolonged immobilization, among others.^1,6,7 In orthopedics, most of this research comes from total joint arthroplasty and hip fracture studies. However, there is relatively limited data for ankle fracture. The best studies directly addressing VTE after ORIF of ankle fractures have had important limitations, including missing patient data and suboptimal capture of VTE occurrences,^8-10 possibly leading to underestimates of the incidence of VTEs.

Given the limited data available, we conducted a retrospective national-cohort study to determine the incidence of and independent risk factors for VTEs after ankle fracture ORIF. If patients who are at higher risk for VTE can be identified, they can and should be carefully monitored and be considered for VTE prophylaxis. This information is needed for patient counseling and clinical decision-making.

Materials and Methods

This retrospective study used the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database, which captures data from more than 370 participating US hospitals.¹¹ In ACS-NSQIP, 150 patient variables are collected from operative reports, medical records, and patient interviews by trained clinical reviewers.^11,12 Patients are identified prospectively and randomly sampled at participating hospitals. Routine auditing is performed to ensure high-quality data. Clinical data are collected for the entire 30-day postoperative period, regardless of discharge status during this time.

Patients who underwent ankle fracture ORIF between 2005 and 2012 were identified in the ACS-NSQIP database. They were initially selected by the postoperative diagnosis of ankle fracture (International Classification of Diseases, Ninth Revision codes 824.0-824.9). Of these patients, only those with primary Current Procedural Terminology codes 27766 (ORIF of medial malleolus fracture), 27769 (ORIF of posterior malleolus fracture), 27792 (ORIF of lateral malleolus fracture), 27814 (ORIF of bimalleollar fracture), and 27822/27823 (ORIF of trimalleollar fracture) were included in the analysis. Patients with incomplete perioperative data were excluded, leaving 4412 patients (out of the initial 4785) for analysis.

Patient characteristics, including sex, age, height, weight, and history of smoking, were collected from the ACS-NSQIP database. Body mass index (BMI) was calculated from each patient’s height and weight. Age was divided into approximately 20-year increments, beginning with age 18 years, in order to compare younger, middle-aged, and elderly groups of patients with ankle fractures. BMI was divided into categories based on the World Health Organization definitions of obesity: under 25 kg/m² (normal weight), 25 to 30 kg/m² (overweight), 30 to 35 kg/m² (class I obesity), and 35 kg/m² or over (class II and class III obesity).¹³

Information about medical comorbidities is also available in the ACS-NSQIP database. History of pulmonary disease was defined as a history of dyspnea, severe chronic obstructive pulmonary disease, ventilator-assisted respiration within 48 hours before surgery, or current pneumonia. History of heart disease was defined as a history of congestive heart failure (CHF) or angina within 1 month before admission, myocardial infarction within 6 months before admission, cardiac surgery, or percutaneous coronary intervention. American Society of Anesthesiologists (ASA) classes 3 and above signify severe systemic disease. Steroid use was defined as requiring regular administration of corticosteroid medications within 1 month before surgery. Disseminated cancer was defined as a malignancy that has spread to 1 or more sites besides the primary site.

Functional status was defined as the ability to perform activities of daily living (ADLs) within 30 days before surgery. Best functional status during this period was recorded. ACS-NSQIP defines ADLs as the “activities usually performed in the course of a normal day in a person’s life,” including bathing, feeding, dressing, toileting, and mobility. An independent patient does not require assistance for any ADLs; a partially dependent patient requires assistance for some ADLs; and a totally dependent patient requires assistance in all ADLs. Partially and totally dependent patients were grouped for analysis. Anesthesia type was separated into general and nongeneral, which includes monitored anesthesia care, spinal anesthesia, and regional anesthesia.

ACS-NSQIP also records the occurrence of multiple events up to 30 days after surgery. For our study, VTE was defined as the occurrence of a DVT or a PE during this period. ACS-NSQIP defines DVT as a new blood clot or thrombus identified within a vein—with confirmation by duplex ultrasonography, venogram, or computed tomography (CT)—that required therapy (anticoagulation, placement of vena cava filter, and/or clipping of vena cava). PE is recorded if ventilation/perfusion (VQ) scan, CT examination, transesophageal echocardiogram, pulmonary arteriogram, CT angiogram, or any other definitive modality is positive.

Statistical analyses were performed with Stata Version 11.2 (StataCorp). Demographic and comorbidity variables were tested for association with occurrence of VTE using bivariate and multivariate logistic regression.

Final multivariate models were constructed with a backward stepwise process that initially included all potential variables and sequentially excluded variables with the highest P value until only those with P < .200 remained. Variables with .050 < P < .200 were left in the model to control for potential confounding but are not considered significantly associated with the outcome. Statistical significance was established at a 2-sided α of 0.050 (P < .050). The fitness of the final logistic regression model was assessed with the C statistic and the Hosmer-Lemeshow goodness-of-fit test.

Results

For the 4412 ankle fracture patients who met the inclusion criteria, mean (SD) age was 50.9 (18.2) years, and mean (SD) BMI was 30.4 (7.6) kg/m². The cohort was 40.4% male. Surgery was performed on 235 patients (5.3%) with medial malleolus fracture, 1143 patients (25.9%) with lateral malleolus fracture, 1705 patients (38.6%) with bimalleollar fracture, and 1329 patients (30.1%) with trimalleollar fracture. Table 1 summarizes the patient characteristics.

Of the 33 patients (0.8%) with a VTE recorded within the first 30 postoperative days, 16 (0.4% of all patients) had a DVT recorded, 14 (0.3% of all patients) had a PE recorded, and 3 (0.1% of all patients) had both a DVT and a PE recorded. In 13 (39.4%) of the 33 patients with a VTE, the event occurred after discharge. VTEs were reported a mean (SD) of 11.5 (9.6) days after surgery. No patient in this study died of VTE.

Bivariate logistic regressions were performed to test the association of each patient variable with the occurrence of a VTE. Results are listed in Table 2. The bivariate analyses revealed significant associations between VTE after ankle fracture ORIF and the patient variables of age 60 years or older (odds ratio [OR], 2.40; 95% confidence interval [CI], 1.01-5.72), class I obesity (BMI, 30-35 kg/m²: OR, 5.15, 95% CI, 1.14-23.28), class II and class III obesity (BMI, ≥35 kg/m²: OR, 6.33, 95% CI, 1.41-28.38), ASA classes 3 and 4 (OR, 3.05; 95% CI, 1.53-6.08), history of heart disease (OR, 5.10; 95% CI, 2.08-12.49), history of hypertension (OR, 2.81; 95% CI, 1.39-5.66), and dependent functional status (OR, 3.39; 95% CI, 1.52-7.56).

Multivariate logistic regression was used to control for potential confounding variables and determine which factors were independently associated with VTEs. Results of this analysis are listed in Table 2 as well. The multivariate analysis revealed that the patient variables of class I obesity (BMI, 30-35 kg/m²: OR, 4.77; 95% CI, 1.05-21.72; P = .044), class II and class III obesity (BMI, ≥35 kg/m²: OR, 4.71; 95% CI, 1.03-21.68; P = .046), history of heart disease (OR, 3.28; 95% CI, 1.20-8.97; P = .020), and dependent functional status (OR, 2.59; 95% CI, 1.11-6.04; P = .028) were independently associated with an increased rate of VTEs. Of note, anesthesia type was not significantly associated with occurrence of VTE on bivariate or multivariate analysis.

The C statistic of the final multivariate model was 0.76, indicating very good distinguishing ability. The Hosmer-Lemeshow goodness-of-fit test showed no evidence of lack of fit.

Discussion

Citing the lack of conclusive evidence and the low incidence of VTE after ankle fracture surgery, current recommendations are to avoid routine VTE prophylaxis in the postoperative management of patients who undergo this surgery.^1,5 However, it is important to identify patients who are at increased risk, as some may benefit from VTE prophylaxis. In the present study, we used the large, high-quality ACS-NSQIP database collecting information from multiple US hospitals to examine risk factors for VTE after ankle fracture ORIF. We identified 4412 patients who underwent ankle fracture ORIF between 2005 and 2012, and found an overall VTE incidence of 0.8%. Multivariate analysis identified obesity, history of heart disease, and dependent functional status as independent risk factors for VTE after ankle fracture ORIF.

This study’s 0.8% incidence of VTE after ankle fracture ORIF is consistent with the range (0.29%-5%) reported in other ankle fracture studies.^4,8-10,14-18 We found that VTEs occurred a mean of about 11 days after surgery, and no patient died of VTE.

Obesity (BMI, ≥30 kg/m²) had the strongest association with VTEs in this study. Obesity, which is a growing public health concern, can make postoperative care and mobilization more difficult.¹⁹ Obesity has previously been associated with VTEs after ankle fractures, and BMI of over 25 kg/m² is one of the Caprini criteria for thrombosis risk factor assessment.^6,10 In our study, however, BMI of 25 to 30 kg/m² was not associated with an increased VTE rate, indicating that moderately overweight patients may not be at significantly higher risk for VTE (compared with patients with normal BMI) and may not need VTE prophylaxis. VTE prophylaxis after ankle fracture surgery may be considered in patients with BMI over 30 kg/m².

History of heart disease was also associated with VTEs in this study. Patients with a history of heart disease were at 3 times the risk for VTE within 30 days of ankle fracture surgery. This association is also consistent with the Caprini criteria, which include acute myocardial infarction and CHF as risk factors for venous thrombosis.⁶ Other studies have found associations between CHF and VTE and between cardiovascular risk factors and VTE.^7,20 The association between cardiovascular disease and VTE may derive from the decreased venous flow rate associated with CHF or an overall vascular disease state. These patients may benefit from heightened surveillance and postoperative prophylaxis for VTE.

Dependent functional status was the final risk factor found to be associated with VTE after ankle fracture ORIF. This association likely derives from an inability to mobilize independently, leading to increased venous stasis. Immobilization has been previously associated with increased risk for VTE after ankle surgery.^7,14,16,20 Caretakers should be aware of this increased risk during the postoperative period and diligently monitor these patients for signs and symptoms of VTE. Prophylaxis may also be considered in this patient population.

Several risk factors that were significant on bivariate analysis (increased age; increased ASA class; history of diabetes, pulmonary disease, hypertension) were not significant in the final multivariate model. This finding suggests covariance between these factors and those that were significant in the final multivariate model. In particular, age and increased overall comorbidity (represented by increased ASA class) were not significant in our multivariate model—contrary to findings of other studies.^8-10 It is possible that history of heart disease alone was responsible for the association between overall comorbidity and VTE in those studies. In the present study, separating and controlling for individual comorbidities could have allowed this association to be more precisely characterized.

The characteristics of the ACS-NSQIP database limited our study in several ways. First, although ACS-NSQIP makes significant efforts to collect as many patient variables as possible, some information is not captured. Data about additional factors that may affect VTE risk (eg, history of previous VTE, hypercoagulable state, history of malignancy other than disseminated cancer, tourniquet time, patient position in operating room) were not available. Second, data are collected only on those postoperative adverse events that occur within 30 days after surgery; data on VTEs that occur later are not captured. However, it has been shown that the majority of VTEs occur within the first 30 days after lower extremity trauma and surgery,^21,22 so this follow-up interval was deemed adequate for capture of VTE data. Third, the database does not include information on the prophylactic regimens used for these patients—which may have weakened the associations between predictor variables and VTE risk and led to an underestimated effect size. VTE incidence, as well as the odds of developing a VTE with one of the identified risk factors, may actually be higher than reported in this study.

Conclusion

VTEs are serious complications that can occur after ORIF of ankle fractures. In this study, the overall incidence of VTE after ankle fracture ORIF was 0.8%. Although the American Orthopaedic Foot and Ankle Society and the American College of Chest Physicians do not recommend routine screening or prophylaxis for VTE in patients with ankle fractures,^1,5 the results of this study showed there may be a benefit in emphasizing VTE prophylaxis after ankle fracture ORIF in patients with obesity, history of heart disease, or dependent functional status. At minimum, these patients should be more carefully monitored for development of VTEs.

Venous thromboembolic events (VTEs), encompassing both deep vein thrombosis (DVT) and pulmonary embolism (PE), are potentially fatal events that can occur after orthopedic surgery.¹ In patients who do not receive prophylaxis, VTE incidence can be as high as 70% for total hip arthroplasty,² 26% for hip fracture,³ and 5% for ankle fracture.⁴ Based on the relatively low incidence of VTE after ankle fractures and insufficient evidence for VTE prophylaxis in this population, the American Orthopaedic Foot and Ankle Society and the American College of Chest Physicians do not recommend routine screening or prophylaxis for VTE in patients with ankle fractures.^1,5 Nevertheless, certain patients may be at increased risk for VTE after open reduction and internal fixation (ORIF) of an ankle fracture. In such cases, further consideration for prophylaxis may be warranted.

Other studies of VTEs have identified general risk factors of increased age, obesity, prior thromboembolic disease, oral contraceptive use, multitrauma, varicose veins, and prolonged immobilization, among others.^1,6,7 In orthopedics, most of this research comes from total joint arthroplasty and hip fracture studies. However, there is relatively limited data for ankle fracture. The best studies directly addressing VTE after ORIF of ankle fractures have had important limitations, including missing patient data and suboptimal capture of VTE occurrences,^8-10 possibly leading to underestimates of the incidence of VTEs.

Given the limited data available, we conducted a retrospective national-cohort study to determine the incidence of and independent risk factors for VTEs after ankle fracture ORIF. If patients who are at higher risk for VTE can be identified, they can and should be carefully monitored and be considered for VTE prophylaxis. This information is needed for patient counseling and clinical decision-making.

Materials and Methods

This retrospective study used the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database, which captures data from more than 370 participating US hospitals.¹¹ In ACS-NSQIP, 150 patient variables are collected from operative reports, medical records, and patient interviews by trained clinical reviewers.^11,12 Patients are identified prospectively and randomly sampled at participating hospitals. Routine auditing is performed to ensure high-quality data. Clinical data are collected for the entire 30-day postoperative period, regardless of discharge status during this time.

Patients who underwent ankle fracture ORIF between 2005 and 2012 were identified in the ACS-NSQIP database. They were initially selected by the postoperative diagnosis of ankle fracture (International Classification of Diseases, Ninth Revision codes 824.0-824.9). Of these patients, only those with primary Current Procedural Terminology codes 27766 (ORIF of medial malleolus fracture), 27769 (ORIF of posterior malleolus fracture), 27792 (ORIF of lateral malleolus fracture), 27814 (ORIF of bimalleollar fracture), and 27822/27823 (ORIF of trimalleollar fracture) were included in the analysis. Patients with incomplete perioperative data were excluded, leaving 4412 patients (out of the initial 4785) for analysis.

Patient characteristics, including sex, age, height, weight, and history of smoking, were collected from the ACS-NSQIP database. Body mass index (BMI) was calculated from each patient’s height and weight. Age was divided into approximately 20-year increments, beginning with age 18 years, in order to compare younger, middle-aged, and elderly groups of patients with ankle fractures. BMI was divided into categories based on the World Health Organization definitions of obesity: under 25 kg/m² (normal weight), 25 to 30 kg/m² (overweight), 30 to 35 kg/m² (class I obesity), and 35 kg/m² or over (class II and class III obesity).¹³

Information about medical comorbidities is also available in the ACS-NSQIP database. History of pulmonary disease was defined as a history of dyspnea, severe chronic obstructive pulmonary disease, ventilator-assisted respiration within 48 hours before surgery, or current pneumonia. History of heart disease was defined as a history of congestive heart failure (CHF) or angina within 1 month before admission, myocardial infarction within 6 months before admission, cardiac surgery, or percutaneous coronary intervention. American Society of Anesthesiologists (ASA) classes 3 and above signify severe systemic disease. Steroid use was defined as requiring regular administration of corticosteroid medications within 1 month before surgery. Disseminated cancer was defined as a malignancy that has spread to 1 or more sites besides the primary site.

Functional status was defined as the ability to perform activities of daily living (ADLs) within 30 days before surgery. Best functional status during this period was recorded. ACS-NSQIP defines ADLs as the “activities usually performed in the course of a normal day in a person’s life,” including bathing, feeding, dressing, toileting, and mobility. An independent patient does not require assistance for any ADLs; a partially dependent patient requires assistance for some ADLs; and a totally dependent patient requires assistance in all ADLs. Partially and totally dependent patients were grouped for analysis. Anesthesia type was separated into general and nongeneral, which includes monitored anesthesia care, spinal anesthesia, and regional anesthesia.

ACS-NSQIP also records the occurrence of multiple events up to 30 days after surgery. For our study, VTE was defined as the occurrence of a DVT or a PE during this period. ACS-NSQIP defines DVT as a new blood clot or thrombus identified within a vein—with confirmation by duplex ultrasonography, venogram, or computed tomography (CT)—that required therapy (anticoagulation, placement of vena cava filter, and/or clipping of vena cava). PE is recorded if ventilation/perfusion (VQ) scan, CT examination, transesophageal echocardiogram, pulmonary arteriogram, CT angiogram, or any other definitive modality is positive.

Statistical analyses were performed with Stata Version 11.2 (StataCorp). Demographic and comorbidity variables were tested for association with occurrence of VTE using bivariate and multivariate logistic regression.

Final multivariate models were constructed with a backward stepwise process that initially included all potential variables and sequentially excluded variables with the highest P value until only those with P < .200 remained. Variables with .050 < P < .200 were left in the model to control for potential confounding but are not considered significantly associated with the outcome. Statistical significance was established at a 2-sided α of 0.050 (P < .050). The fitness of the final logistic regression model was assessed with the C statistic and the Hosmer-Lemeshow goodness-of-fit test.

Results

For the 4412 ankle fracture patients who met the inclusion criteria, mean (SD) age was 50.9 (18.2) years, and mean (SD) BMI was 30.4 (7.6) kg/m². The cohort was 40.4% male. Surgery was performed on 235 patients (5.3%) with medial malleolus fracture, 1143 patients (25.9%) with lateral malleolus fracture, 1705 patients (38.6%) with bimalleollar fracture, and 1329 patients (30.1%) with trimalleollar fracture. Table 1 summarizes the patient characteristics.

Of the 33 patients (0.8%) with a VTE recorded within the first 30 postoperative days, 16 (0.4% of all patients) had a DVT recorded, 14 (0.3% of all patients) had a PE recorded, and 3 (0.1% of all patients) had both a DVT and a PE recorded. In 13 (39.4%) of the 33 patients with a VTE, the event occurred after discharge. VTEs were reported a mean (SD) of 11.5 (9.6) days after surgery. No patient in this study died of VTE.

Bivariate logistic regressions were performed to test the association of each patient variable with the occurrence of a VTE. Results are listed in Table 2. The bivariate analyses revealed significant associations between VTE after ankle fracture ORIF and the patient variables of age 60 years or older (odds ratio [OR], 2.40; 95% confidence interval [CI], 1.01-5.72), class I obesity (BMI, 30-35 kg/m²: OR, 5.15, 95% CI, 1.14-23.28), class II and class III obesity (BMI, ≥35 kg/m²: OR, 6.33, 95% CI, 1.41-28.38), ASA classes 3 and 4 (OR, 3.05; 95% CI, 1.53-6.08), history of heart disease (OR, 5.10; 95% CI, 2.08-12.49), history of hypertension (OR, 2.81; 95% CI, 1.39-5.66), and dependent functional status (OR, 3.39; 95% CI, 1.52-7.56).

Multivariate logistic regression was used to control for potential confounding variables and determine which factors were independently associated with VTEs. Results of this analysis are listed in Table 2 as well. The multivariate analysis revealed that the patient variables of class I obesity (BMI, 30-35 kg/m²: OR, 4.77; 95% CI, 1.05-21.72; P = .044), class II and class III obesity (BMI, ≥35 kg/m²: OR, 4.71; 95% CI, 1.03-21.68; P = .046), history of heart disease (OR, 3.28; 95% CI, 1.20-8.97; P = .020), and dependent functional status (OR, 2.59; 95% CI, 1.11-6.04; P = .028) were independently associated with an increased rate of VTEs. Of note, anesthesia type was not significantly associated with occurrence of VTE on bivariate or multivariate analysis.

The C statistic of the final multivariate model was 0.76, indicating very good distinguishing ability. The Hosmer-Lemeshow goodness-of-fit test showed no evidence of lack of fit.

Discussion

Citing the lack of conclusive evidence and the low incidence of VTE after ankle fracture surgery, current recommendations are to avoid routine VTE prophylaxis in the postoperative management of patients who undergo this surgery.^1,5 However, it is important to identify patients who are at increased risk, as some may benefit from VTE prophylaxis. In the present study, we used the large, high-quality ACS-NSQIP database collecting information from multiple US hospitals to examine risk factors for VTE after ankle fracture ORIF. We identified 4412 patients who underwent ankle fracture ORIF between 2005 and 2012, and found an overall VTE incidence of 0.8%. Multivariate analysis identified obesity, history of heart disease, and dependent functional status as independent risk factors for VTE after ankle fracture ORIF.

This study’s 0.8% incidence of VTE after ankle fracture ORIF is consistent with the range (0.29%-5%) reported in other ankle fracture studies.^4,8-10,14-18 We found that VTEs occurred a mean of about 11 days after surgery, and no patient died of VTE.

Obesity (BMI, ≥30 kg/m²) had the strongest association with VTEs in this study. Obesity, which is a growing public health concern, can make postoperative care and mobilization more difficult.¹⁹ Obesity has previously been associated with VTEs after ankle fractures, and BMI of over 25 kg/m² is one of the Caprini criteria for thrombosis risk factor assessment.^6,10 In our study, however, BMI of 25 to 30 kg/m² was not associated with an increased VTE rate, indicating that moderately overweight patients may not be at significantly higher risk for VTE (compared with patients with normal BMI) and may not need VTE prophylaxis. VTE prophylaxis after ankle fracture surgery may be considered in patients with BMI over 30 kg/m².

History of heart disease was also associated with VTEs in this study. Patients with a history of heart disease were at 3 times the risk for VTE within 30 days of ankle fracture surgery. This association is also consistent with the Caprini criteria, which include acute myocardial infarction and CHF as risk factors for venous thrombosis.⁶ Other studies have found associations between CHF and VTE and between cardiovascular risk factors and VTE.^7,20 The association between cardiovascular disease and VTE may derive from the decreased venous flow rate associated with CHF or an overall vascular disease state. These patients may benefit from heightened surveillance and postoperative prophylaxis for VTE.

Dependent functional status was the final risk factor found to be associated with VTE after ankle fracture ORIF. This association likely derives from an inability to mobilize independently, leading to increased venous stasis. Immobilization has been previously associated with increased risk for VTE after ankle surgery.^7,14,16,20 Caretakers should be aware of this increased risk during the postoperative period and diligently monitor these patients for signs and symptoms of VTE. Prophylaxis may also be considered in this patient population.

Several risk factors that were significant on bivariate analysis (increased age; increased ASA class; history of diabetes, pulmonary disease, hypertension) were not significant in the final multivariate model. This finding suggests covariance between these factors and those that were significant in the final multivariate model. In particular, age and increased overall comorbidity (represented by increased ASA class) were not significant in our multivariate model—contrary to findings of other studies.^8-10 It is possible that history of heart disease alone was responsible for the association between overall comorbidity and VTE in those studies. In the present study, separating and controlling for individual comorbidities could have allowed this association to be more precisely characterized.

The characteristics of the ACS-NSQIP database limited our study in several ways. First, although ACS-NSQIP makes significant efforts to collect as many patient variables as possible, some information is not captured. Data about additional factors that may affect VTE risk (eg, history of previous VTE, hypercoagulable state, history of malignancy other than disseminated cancer, tourniquet time, patient position in operating room) were not available. Second, data are collected only on those postoperative adverse events that occur within 30 days after surgery; data on VTEs that occur later are not captured. However, it has been shown that the majority of VTEs occur within the first 30 days after lower extremity trauma and surgery,^21,22 so this follow-up interval was deemed adequate for capture of VTE data. Third, the database does not include information on the prophylactic regimens used for these patients—which may have weakened the associations between predictor variables and VTE risk and led to an underestimated effect size. VTE incidence, as well as the odds of developing a VTE with one of the identified risk factors, may actually be higher than reported in this study.

Conclusion

VTEs are serious complications that can occur after ORIF of ankle fractures. In this study, the overall incidence of VTE after ankle fracture ORIF was 0.8%. Although the American Orthopaedic Foot and Ankle Society and the American College of Chest Physicians do not recommend routine screening or prophylaxis for VTE in patients with ankle fractures,^1,5 the results of this study showed there may be a benefit in emphasizing VTE prophylaxis after ankle fracture ORIF in patients with obesity, history of heart disease, or dependent functional status. At minimum, these patients should be more carefully monitored for development of VTEs.

Venous thromboembolic events (VTEs), encompassing both deep vein thrombosis (DVT) and pulmonary embolism (PE), are potentially fatal events that can occur after orthopedic surgery.¹ In patients who do not receive prophylaxis, VTE incidence can be as high as 70% for total hip arthroplasty,² 26% for hip fracture,³ and 5% for ankle fracture.⁴ Based on the relatively low incidence of VTE after ankle fractures and insufficient evidence for VTE prophylaxis in this population, the American Orthopaedic Foot and Ankle Society and the American College of Chest Physicians do not recommend routine screening or prophylaxis for VTE in patients with ankle fractures.^1,5 Nevertheless, certain patients may be at increased risk for VTE after open reduction and internal fixation (ORIF) of an ankle fracture. In such cases, further consideration for prophylaxis may be warranted.

Other studies of VTEs have identified general risk factors of increased age, obesity, prior thromboembolic disease, oral contraceptive use, multitrauma, varicose veins, and prolonged immobilization, among others.^1,6,7 In orthopedics, most of this research comes from total joint arthroplasty and hip fracture studies. However, there is relatively limited data for ankle fracture. The best studies directly addressing VTE after ORIF of ankle fractures have had important limitations, including missing patient data and suboptimal capture of VTE occurrences,^8-10 possibly leading to underestimates of the incidence of VTEs.

Given the limited data available, we conducted a retrospective national-cohort study to determine the incidence of and independent risk factors for VTEs after ankle fracture ORIF. If patients who are at higher risk for VTE can be identified, they can and should be carefully monitored and be considered for VTE prophylaxis. This information is needed for patient counseling and clinical decision-making.

Materials and Methods

This retrospective study used the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database, which captures data from more than 370 participating US hospitals.¹¹ In ACS-NSQIP, 150 patient variables are collected from operative reports, medical records, and patient interviews by trained clinical reviewers.^11,12 Patients are identified prospectively and randomly sampled at participating hospitals. Routine auditing is performed to ensure high-quality data. Clinical data are collected for the entire 30-day postoperative period, regardless of discharge status during this time.

Patients who underwent ankle fracture ORIF between 2005 and 2012 were identified in the ACS-NSQIP database. They were initially selected by the postoperative diagnosis of ankle fracture (International Classification of Diseases, Ninth Revision codes 824.0-824.9). Of these patients, only those with primary Current Procedural Terminology codes 27766 (ORIF of medial malleolus fracture), 27769 (ORIF of posterior malleolus fracture), 27792 (ORIF of lateral malleolus fracture), 27814 (ORIF of bimalleollar fracture), and 27822/27823 (ORIF of trimalleollar fracture) were included in the analysis. Patients with incomplete perioperative data were excluded, leaving 4412 patients (out of the initial 4785) for analysis.

Patient characteristics, including sex, age, height, weight, and history of smoking, were collected from the ACS-NSQIP database. Body mass index (BMI) was calculated from each patient’s height and weight. Age was divided into approximately 20-year increments, beginning with age 18 years, in order to compare younger, middle-aged, and elderly groups of patients with ankle fractures. BMI was divided into categories based on the World Health Organization definitions of obesity: under 25 kg/m² (normal weight), 25 to 30 kg/m² (overweight), 30 to 35 kg/m² (class I obesity), and 35 kg/m² or over (class II and class III obesity).¹³

Information about medical comorbidities is also available in the ACS-NSQIP database. History of pulmonary disease was defined as a history of dyspnea, severe chronic obstructive pulmonary disease, ventilator-assisted respiration within 48 hours before surgery, or current pneumonia. History of heart disease was defined as a history of congestive heart failure (CHF) or angina within 1 month before admission, myocardial infarction within 6 months before admission, cardiac surgery, or percutaneous coronary intervention. American Society of Anesthesiologists (ASA) classes 3 and above signify severe systemic disease. Steroid use was defined as requiring regular administration of corticosteroid medications within 1 month before surgery. Disseminated cancer was defined as a malignancy that has spread to 1 or more sites besides the primary site.

Functional status was defined as the ability to perform activities of daily living (ADLs) within 30 days before surgery. Best functional status during this period was recorded. ACS-NSQIP defines ADLs as the “activities usually performed in the course of a normal day in a person’s life,” including bathing, feeding, dressing, toileting, and mobility. An independent patient does not require assistance for any ADLs; a partially dependent patient requires assistance for some ADLs; and a totally dependent patient requires assistance in all ADLs. Partially and totally dependent patients were grouped for analysis. Anesthesia type was separated into general and nongeneral, which includes monitored anesthesia care, spinal anesthesia, and regional anesthesia.

ACS-NSQIP also records the occurrence of multiple events up to 30 days after surgery. For our study, VTE was defined as the occurrence of a DVT or a PE during this period. ACS-NSQIP defines DVT as a new blood clot or thrombus identified within a vein—with confirmation by duplex ultrasonography, venogram, or computed tomography (CT)—that required therapy (anticoagulation, placement of vena cava filter, and/or clipping of vena cava). PE is recorded if ventilation/perfusion (VQ) scan, CT examination, transesophageal echocardiogram, pulmonary arteriogram, CT angiogram, or any other definitive modality is positive.

Statistical analyses were performed with Stata Version 11.2 (StataCorp). Demographic and comorbidity variables were tested for association with occurrence of VTE using bivariate and multivariate logistic regression.

Final multivariate models were constructed with a backward stepwise process that initially included all potential variables and sequentially excluded variables with the highest P value until only those with P < .200 remained. Variables with .050 < P < .200 were left in the model to control for potential confounding but are not considered significantly associated with the outcome. Statistical significance was established at a 2-sided α of 0.050 (P < .050). The fitness of the final logistic regression model was assessed with the C statistic and the Hosmer-Lemeshow goodness-of-fit test.

Results

For the 4412 ankle fracture patients who met the inclusion criteria, mean (SD) age was 50.9 (18.2) years, and mean (SD) BMI was 30.4 (7.6) kg/m². The cohort was 40.4% male. Surgery was performed on 235 patients (5.3%) with medial malleolus fracture, 1143 patients (25.9%) with lateral malleolus fracture, 1705 patients (38.6%) with bimalleollar fracture, and 1329 patients (30.1%) with trimalleollar fracture. Table 1 summarizes the patient characteristics.

Of the 33 patients (0.8%) with a VTE recorded within the first 30 postoperative days, 16 (0.4% of all patients) had a DVT recorded, 14 (0.3% of all patients) had a PE recorded, and 3 (0.1% of all patients) had both a DVT and a PE recorded. In 13 (39.4%) of the 33 patients with a VTE, the event occurred after discharge. VTEs were reported a mean (SD) of 11.5 (9.6) days after surgery. No patient in this study died of VTE.

Bivariate logistic regressions were performed to test the association of each patient variable with the occurrence of a VTE. Results are listed in Table 2. The bivariate analyses revealed significant associations between VTE after ankle fracture ORIF and the patient variables of age 60 years or older (odds ratio [OR], 2.40; 95% confidence interval [CI], 1.01-5.72), class I obesity (BMI, 30-35 kg/m²: OR, 5.15, 95% CI, 1.14-23.28), class II and class III obesity (BMI, ≥35 kg/m²: OR, 6.33, 95% CI, 1.41-28.38), ASA classes 3 and 4 (OR, 3.05; 95% CI, 1.53-6.08), history of heart disease (OR, 5.10; 95% CI, 2.08-12.49), history of hypertension (OR, 2.81; 95% CI, 1.39-5.66), and dependent functional status (OR, 3.39; 95% CI, 1.52-7.56).

Multivariate logistic regression was used to control for potential confounding variables and determine which factors were independently associated with VTEs. Results of this analysis are listed in Table 2 as well. The multivariate analysis revealed that the patient variables of class I obesity (BMI, 30-35 kg/m²: OR, 4.77; 95% CI, 1.05-21.72; P = .044), class II and class III obesity (BMI, ≥35 kg/m²: OR, 4.71; 95% CI, 1.03-21.68; P = .046), history of heart disease (OR, 3.28; 95% CI, 1.20-8.97; P = .020), and dependent functional status (OR, 2.59; 95% CI, 1.11-6.04; P = .028) were independently associated with an increased rate of VTEs. Of note, anesthesia type was not significantly associated with occurrence of VTE on bivariate or multivariate analysis.

The C statistic of the final multivariate model was 0.76, indicating very good distinguishing ability. The Hosmer-Lemeshow goodness-of-fit test showed no evidence of lack of fit.

Discussion

Citing the lack of conclusive evidence and the low incidence of VTE after ankle fracture surgery, current recommendations are to avoid routine VTE prophylaxis in the postoperative management of patients who undergo this surgery.^1,5 However, it is important to identify patients who are at increased risk, as some may benefit from VTE prophylaxis. In the present study, we used the large, high-quality ACS-NSQIP database collecting information from multiple US hospitals to examine risk factors for VTE after ankle fracture ORIF. We identified 4412 patients who underwent ankle fracture ORIF between 2005 and 2012, and found an overall VTE incidence of 0.8%. Multivariate analysis identified obesity, history of heart disease, and dependent functional status as independent risk factors for VTE after ankle fracture ORIF.

This study’s 0.8% incidence of VTE after ankle fracture ORIF is consistent with the range (0.29%-5%) reported in other ankle fracture studies.^4,8-10,14-18 We found that VTEs occurred a mean of about 11 days after surgery, and no patient died of VTE.

Obesity (BMI, ≥30 kg/m²) had the strongest association with VTEs in this study. Obesity, which is a growing public health concern, can make postoperative care and mobilization more difficult.¹⁹ Obesity has previously been associated with VTEs after ankle fractures, and BMI of over 25 kg/m² is one of the Caprini criteria for thrombosis risk factor assessment.^6,10 In our study, however, BMI of 25 to 30 kg/m² was not associated with an increased VTE rate, indicating that moderately overweight patients may not be at significantly higher risk for VTE (compared with patients with normal BMI) and may not need VTE prophylaxis. VTE prophylaxis after ankle fracture surgery may be considered in patients with BMI over 30 kg/m².

History of heart disease was also associated with VTEs in this study. Patients with a history of heart disease were at 3 times the risk for VTE within 30 days of ankle fracture surgery. This association is also consistent with the Caprini criteria, which include acute myocardial infarction and CHF as risk factors for venous thrombosis.⁶ Other studies have found associations between CHF and VTE and between cardiovascular risk factors and VTE.^7,20 The association between cardiovascular disease and VTE may derive from the decreased venous flow rate associated with CHF or an overall vascular disease state. These patients may benefit from heightened surveillance and postoperative prophylaxis for VTE.

Dependent functional status was the final risk factor found to be associated with VTE after ankle fracture ORIF. This association likely derives from an inability to mobilize independently, leading to increased venous stasis. Immobilization has been previously associated with increased risk for VTE after ankle surgery.^7,14,16,20 Caretakers should be aware of this increased risk during the postoperative period and diligently monitor these patients for signs and symptoms of VTE. Prophylaxis may also be considered in this patient population.

Several risk factors that were significant on bivariate analysis (increased age; increased ASA class; history of diabetes, pulmonary disease, hypertension) were not significant in the final multivariate model. This finding suggests covariance between these factors and those that were significant in the final multivariate model. In particular, age and increased overall comorbidity (represented by increased ASA class) were not significant in our multivariate model—contrary to findings of other studies.^8-10 It is possible that history of heart disease alone was responsible for the association between overall comorbidity and VTE in those studies. In the present study, separating and controlling for individual comorbidities could have allowed this association to be more precisely characterized.

The characteristics of the ACS-NSQIP database limited our study in several ways. First, although ACS-NSQIP makes significant efforts to collect as many patient variables as possible, some information is not captured. Data about additional factors that may affect VTE risk (eg, history of previous VTE, hypercoagulable state, history of malignancy other than disseminated cancer, tourniquet time, patient position in operating room) were not available. Second, data are collected only on those postoperative adverse events that occur within 30 days after surgery; data on VTEs that occur later are not captured. However, it has been shown that the majority of VTEs occur within the first 30 days after lower extremity trauma and surgery,^21,22 so this follow-up interval was deemed adequate for capture of VTE data. Third, the database does not include information on the prophylactic regimens used for these patients—which may have weakened the associations between predictor variables and VTE risk and led to an underestimated effect size. VTE incidence, as well as the odds of developing a VTE with one of the identified risk factors, may actually be higher than reported in this study.

Conclusion

VTEs are serious complications that can occur after ORIF of ankle fractures. In this study, the overall incidence of VTE after ankle fracture ORIF was 0.8%. Although the American Orthopaedic Foot and Ankle Society and the American College of Chest Physicians do not recommend routine screening or prophylaxis for VTE in patients with ankle fractures,^1,5 the results of this study showed there may be a benefit in emphasizing VTE prophylaxis after ankle fracture ORIF in patients with obesity, history of heart disease, or dependent functional status. At minimum, these patients should be more carefully monitored for development of VTEs.

Mortality rate is an important indicator of the severity of traumatic injuries, and these values have been described for different orthopedic injuries and fractures. Studies have identified 3 distinct trends in patient survival when compared with the age- and sex-matched uninjured population: 

1.  Hip fractures bring about a transient increase in mortality relative to age-matched controls that normalizes after a few months to 1 year.^1-10

2.  Thoracic and lumbar compression fractures are associated with an ongoing, lifelong increase in mortality rate relative to age-matched controls without an initial marked upswing.^11-15

3.  Certain injuries such as isolated rib or wrist fractures do not adversely affect survival relative to age-matched controls.^12,16-18

Understanding the mortality patterns after these injuries can help guide management and even facilitate the development of appropriate treatment algorithms.^19-21 While studies have examined mortality in specific odontoid fracture types,²² such mortality trends have not been broadly established in persons with cervical spine fractures. 

Cervical spine fractures are common: 60% of spine fractures localize to this region,^23-26 and this equates to 2% to 3% of all blunt-trauma patients.^27,28 These injuries can lead to devastating consequences, including neurologic compromise, permanent disability, and death.^29-31

Studies have estimated that up to 20% of cervical fractures involve the odontoid process.^23-26 These injuries are more common among the elderly population because of their greater prevalence of osteoporosis and likelihood of falling.³² Because of demographic similarities to those of the hip fracture population, a survival analysis of all odontoid fractures is particularly interesting. Published odontoid mortality rates vary significantly, with reports ranging from 13% to 44%.^22,33-35 Unfortunately, these studies largely evaluated survival rates specific to an individual treatment modality, such as nonoperative compared with operative, or specific to certain odontoid fracture types (eg, type II). Additionally, studies have generally only considered survivorship during initial hospitalization, have been specific to a constrained age group, or have been based solely on inpatient records that do not permit the longer-term follow-up critical to determining the effect of odontoid fractures on overall mortality.^36-39

Likewise, mortality rates after fractures of the subaxial spine (ie, the motion segments between C3 and C7) have yet to be established. In 1 study, the mortality risk of a cohort of elderly patients with cervical fractures appeared to be elevated for the first 6 to 12 months after the traumatic event.⁴⁰ However, the sample size was too small to examine mortality beyond 1 year.

In this context, the purpose of the current study was to determine the mortality rates at several time points (3 months, 1 year, and 2 years) of patients 50 years or older (start of the second mode of the bimodal age distribution of odontoid fractures^41-44) with fractures of the odontoid and subaxial cervical spine. A secondary purpose of this study was to compare survival rates of these 2 cohorts relative to each other and to the general population.

Materials and Methods

Identification of Cervical Fractures and Collection of Demographic Information

This protocol was approved by the human investigation committee of our institution. Every computed tomography (CT) scan of the cervical spine performed in the emergency department (ED) of an academic hospital between November 27, 1997, and December 31, 2006, was identified. Since the threshold for obtaining a CT scan of a patient with suspected cervical spine trauma is relatively low, it was assumed that virtually all acute cervical spine fractures during this time period would be successfully identified through this approach.

Radiology reports for all identified CT scans were reviewed for any findings consistent with acute fractures and/or dislocations of the cervical spine (Figure 1). Every study noted to be positive or equivocal for cervical trauma was directly visualized, including those that did not specifically mention the presence or absence of an injury. Scans with no signs of acute trauma or that showed fractures caused by a pathologic process or penetrating mechanism (eg, metastatic lesions or gunshot wounds) were omitted from this series. Finally, relevant demographic information, such as the medical record number, age, gender, and date of study, was recorded for every subject in this group.

Fracture Classification

Next, the level and the type of cervical injury were documented for each patient. Fractures were segregated according to their involvement with the odontoid or the subaxial vertebrae. 

Odontoid fractures were categorized into type I (limited to the tip), type II (across the base of the process) and type III (through the base with extension into the C2 vertebral body).^45,46 Since many systems for classifying subaxial cervical spine trauma require a subjective inference of the injury mechanism, which is difficult to ascertain from imaging studies alone, all of these fractures were pooled together.

A preliminary survey of the data indicated that the odontoid fractures appeared to exhibit a bimodal age distribution, with the beginning of the second cluster occurring around age 50 years (Figure 2). As noted above, this has been shown in previous studies.^41-44 As a consequence, the mortalities of those older than 50 years became the focus of this study. To control for comorbid conditions, mechanism of injury, and to allow for more direct comparison with the odontoid fractures in this study, the same age demarcation was used for subaxial cervical fractures.

Mortality Data

The mortality status of every patient diagnosed with an acute cervical injury at our institution between November 27, 1997, and December 31, 2006, was determined by referencing the National Death Index (NDI). The NDI is a computerized database of death records maintained by the National Center for Health Statistics (NCHS). The time window for the current study was selected because we had access to NDI information only through 2007 at the time of this study. Social Security numbers (SSNs), which were available for approximately half of the subjects, were used to search the NDI catalog. For individuals whose SSNs were unavailable, patient names and birthdates were considered to be sufficient to confirm a true match. Our center’s medical records of this cohort were also examined to verify whether any had died during their initial hospitalizations and to substantiate the NDI data. Finally, patient deaths were categorized as trauma (eg, motor vehicle accident, fall from a height) or medical comorbidity (eg, diabetes mellitus, cancer, congestive heart failure), based on information in the NDI listing.

Age- and Sex-Matched Controls

Age- and sex-matched controls were determined from the Wide-ranging Online Data for Epidemiologic Research (WONDER) application distributed by the Centers for Disease Control and Prevention (http://wonder.cdc.gov). Composite mortality data from the state in which the study was performed was obtained for the years between 1999 and 2007, and this information was further stratified according to gender and age to estimate the mortality rates and construct survival curves for each group. Controls were used to establish a standardized mortality ratio (SMR) for subjects 50 years and older, a value that compares the number of observed deaths with the figure expected for matched populations from the general population.

Statistical Methods

Statistical analyses were performed by using both SAS 9.2 (SAS Institute Inc., Cary, North Carolina) and R (version 2.9; www.r-project.org, Auckland, New Zealand). Relevant comparisons were planned, and all tests were 2-sided. The Wilcoxon rank sum test was applied to compare the survival times of patients with odontoid fractures with different documented causes of death, and Pearson χ² test was used to compare the age distributions of odontoid and subaxial fractures. Survival rates at 3 months, 1 year, and 2 years were estimated from Kaplan-Meier curves. The relative survival of these cohorts was compared by completing a 2-sample log-rank test. In addition, a 1-sample log-rank test was implemented to compare the mortality from either odontoid or subaxial cervical spine fractures with that of the age- and gender-matched general population. Statistical significance was defined as a 2-sided α error of less than 0.05 (P < .05).

Results

Fifty-nine patients were diagnosed with odontoid fractures (28 men, 31 women), and 233 patients were diagnosed with subaxial cervical spine fractures (168 men, 65 women).

Odontoid Fracture Patients

Odontoid fracture patients exhibited a distinct bimodal age distribution (Figure 2). In the younger population, there were 14 subjects, 3 of whom died within days of the injury (mean, 12 days; 78.6% survival). At 2-year follow-up, there were no further deaths. The fractures that caused death were high-energy injuries, and only early deaths occurred in these cases. 

Because of the significant bimodal age distribution, it was believed these cohorts could not be directly compared. As a result, the remaining analysis focused on the older age group. In the older population mode (50 years and older) were 45 patients with odontoid fractures. Of the 12 subjects who died after odontoid fracture, 5 were assigned a trauma code as the cause of death, while a medical comorbidity code was assigned for the remaining 7. Mean survival time of those who died secondary to trauma was significantly shorter than the medical comorbidity group (P = .025). 

In the cohort of subjects older than 50 years, 3-month, 1-year, and 2-year survival rates were 84.4%, 82.2%, and 72.9%, respectively. Figure 2 shows the 1- and 2-year follow-up data by age group.

Analysis was performed relative to gender. Of male patients (n = 22), the 3-month, 1-year, and 2-year survival rates were 72.7%, 72.7%, and 62.7%, respectively. Among women (n = 23), the 3-month, 1-year, and 2-year survival rates were 95.7%, 91.3%, and 82.6%, respectively.

Figure 3 shows the Kaplan-Meier survival curves of the older patients with odontoid fractures. A comparison of the curves for each gender showed no significant disparities between the male and female survival (Figure 3A, P = .124). Compared with age-matched male counterparts, the survival of male subjects with odontoid fractures was significantly worse (Figure 3B, P < .001). Men experienced an initial acute decline in survival, with the remainder of the survival curve matching that of the general male population. In contrast, odontoid fractures did not adversely affect female survival compared with the matched population (Figure 3C, P = .568). 

The 2-year SMR of 2.98 for men showed that odontoid fractures led to greater mortality compared with a sex- and age-matched population. This means that men older than 50 years who sustained an odontoid fracture had nearly 3 times the mortality rate after 2 years compared with a normal, matched population; this increase is attributed to the 3-month time point that subsequently normalized. The female rate was 1.33 times that of a matched population, a difference that is not statistically significant.

Subaxial Fracture Patients

Of the 91 patients older than 50 years with subaxial fractures, 3-month, 1-year, and 2-year survival rates were 87.9%, 85.7%, and 85.7%, respectively. Figure 4 shows the 1- and 2-year follow-up data by age group.

Gender-specific analysis was performed. For men (n = 58), the 3-month, 1-year, and 2-year survival rates were 87.9%, 84.5%, and 84.5%, respectively. Among women (n = 33), 4 deaths were recorded at all time points (87.9% survival). 

Figure 5 shows Kaplan-Meier survival curves for the older population with subaxial fractures. A comparison of the curves between genders again showed no significant differences between male and female survival (P = .683, Figure 5A). Compared with age- and gender-matched counterparts, men showed decreased relative survival (P < .0001, Figure 5B), whereas subaxial fractures did not decrease female survival (P = .554, Figure 5C).

The 2-year SMR of 2.90 for men showed higher mortality rates relative to sex- and age-matched controls. Men who were both 50 years old and sustained a subaxial fracture were 2.9 times as likely to die within 2 years of follow-up compared with their counterparts. Similar to odontoid fractures, this increase occurred by the 3-month time point and subsequently normalized. The female rate, which was 1.34 times that of the uninjured population, was not statistically significant.

Comparison of Odontoid and Subaxial Fracture Patients

The survival of subaxial injuries was not significantly different from that of odontoid fractures (P = .113, Figure 6A). When analyzed by gender and controlled for age, the rates in both male (P = .347, Figure 6B) and female (P = .643, Figure 6C) patients did not differ between fracture types.

Discussion

The US population is aging rapidly, with the demographic older than 65 years predicted to more than double in size between 2010 and 2050.⁴⁷ As our elderly population grows, the incidence of age-related injuries will rise accordingly. An understanding of mortality risks associated with different fractures will not only assist practitioners in advising patients regarding prognosis but may also lead to improvements in clinical care.^19,48-50 While we know cervical spine trauma is associated with significant morbidity,^29-31 little is known about associated moderate-term mortality rates that can be compared with other known injury patterns, such as hip fractures or osteoporotic compression fractures. 

An interesting finding of the present study is the bimodal age distribution of the 59 odontoid fractures (Figure 2). The 14 patients younger than 50 years included 3 individuals who died, all within days of their presentation from severe multisystem trauma. This is consistent with the determination that high-energy forces are required to fracture the odontoid process in younger individuals.^{38,45,46,51,52} Given the severity of their nonspinal injuries, the cause of death was likely not primarily related to their odontoid fractures. Also in line with previous studies, the majority (76%) of odontoid fractures were documented in subjects older than 50 years.^32,53,54 Within our cohort older than 50 years, the deaths appear to be spread evenly across age groups and do not seem to be skewed by the oldest portion of the population (Figure 2).

Our gender-specific analyses revealed that older men with odontoid injuries exhibited higher mortality compared with an age-matched male cohort, with 6 of the 8 deaths occurring within 3 months. However, after this exaggerated decline in survival, the rate normalized towards general population mortality rates (Figure 3B). As in the younger cohort, these earlier deaths were largely attributable to multisystem trauma, whereas medical comorbidities were implicated in those who died later. In contrast, the Kaplan-Meier curve of older women with odontoid fractures closely approximates that of age-matched women at every time point (Figure 3C), indicating that these injuries do not decrease survival as they do in their male counterparts.

When comparing the survival of older patients with subaxial cervical spine fractures with that of gender- and age-matched controls, the mortality rates of women were, once again, essentially equivalent. However, the survival of older men was significantly compromised by these injuries. In men, 7 of the 9 deaths were within 3 months, with the remaining 2 deaths occurring within 7 months. Nevertheless, beyond this initial period of elevated mortality, the survival curve again stabilized and paralleled that of the general population. As with odontoid fractures, there was no sustained increase in the mortality of male patients who lived at least 3 months after injury.

The mortality rates of odontoid and subaxial fractures were also compared in the older population. When controlled for age, there was no difference in mortality rates between these 2 groups. When individually analyzed in both men and women, the mortality rates of both fracture types matched those of the general population at all time points.

It is useful to contextualize our findings alongside the mortality of older individuals with other fracture types. Based on our results, we believe that the survival curves of geriatric men with odontoid or subaxial cervical spine fractures most closely resemble the characteristic pattern seen in hip fractures. Hip fractures have shown an early spike in mortality by as much as 8% to 49% in the first 6 to 12 months that returns to baseline after 1 year.^1-10 This presumably reflects the natural history of these injuries in response to appropriate therapeutic interventions. Interestingly, the male mortality rates for both odontoid and subaxial cervical spine fractures in this study are largely analogous to those reported by various hip fracture surveys.^1,5,55-58 In contrast, similar to prior studies of rib or wrist fractures, older women with these cervical spine fractures did not show a survival decrease after their injuries.^12,16-18

While the reasons underlying the differential effects of cervical fractures on the mortality of men and women have not been established, one explanation is that the female geriatric population is relatively more osteoporotic; thus, cervical injuries may occur after lower-energy forces, leading to less severe associated trauma that could otherwise decrease survival. Another explanation is that men are more likely to be involved in high-energy accidents,^59,60 thus decreasing their overall survival after injury.

This investigation is not without limitations. Our primary concern is the determination of survival. The NDI maintained by the NCHS is an extremely reliable tool regularly employed by epidemiologists to collect mortality data. However, it is possible that deaths may have been missed. We believe this number would be small, because the NDI database provided multiple probable matches that were carefully compared with supplemental personal information. It is also possible that deaths that were not appropriately registered with the NDI are not represented in this series. Another limitation lies in the determination of controls. As with any case–control study, the patients sustaining these odontoid fractures may differ in some significant way from the average population. A final limitation is that a small portion of patients in the study have only 1-year follow-up, because patient data was collected through 2006, although access to NDI data ended in 2007.

Conclusion

Our results indicate that the survival of older men with either odontoid or subaxial cervical spine fractures shares many of the same mortality characteristics as hip fractures, with diminished survival in the first 3 months that normalizes to the survival rate of the age-matched population. Interestingly, and perhaps because of disparate rates of osteoporosis and traumatic forces, the mortality rates in the female cohort were similar to that of the age-matched general population at all time points. These trends were nearly identical for both odontoid and subaxial cervical fractures.

Mortality rate is an important indicator of the severity of traumatic injuries, and these values have been described for different orthopedic injuries and fractures. Studies have identified 3 distinct trends in patient survival when compared with the age- and sex-matched uninjured population: 

1.  Hip fractures bring about a transient increase in mortality relative to age-matched controls that normalizes after a few months to 1 year.^1-10

2.  Thoracic and lumbar compression fractures are associated with an ongoing, lifelong increase in mortality rate relative to age-matched controls without an initial marked upswing.^11-15

3.  Certain injuries such as isolated rib or wrist fractures do not adversely affect survival relative to age-matched controls.^12,16-18

Understanding the mortality patterns after these injuries can help guide management and even facilitate the development of appropriate treatment algorithms.^19-21 While studies have examined mortality in specific odontoid fracture types,²² such mortality trends have not been broadly established in persons with cervical spine fractures. 

Cervical spine fractures are common: 60% of spine fractures localize to this region,^23-26 and this equates to 2% to 3% of all blunt-trauma patients.^27,28 These injuries can lead to devastating consequences, including neurologic compromise, permanent disability, and death.^29-31

Studies have estimated that up to 20% of cervical fractures involve the odontoid process.^23-26 These injuries are more common among the elderly population because of their greater prevalence of osteoporosis and likelihood of falling.³² Because of demographic similarities to those of the hip fracture population, a survival analysis of all odontoid fractures is particularly interesting. Published odontoid mortality rates vary significantly, with reports ranging from 13% to 44%.^22,33-35 Unfortunately, these studies largely evaluated survival rates specific to an individual treatment modality, such as nonoperative compared with operative, or specific to certain odontoid fracture types (eg, type II). Additionally, studies have generally only considered survivorship during initial hospitalization, have been specific to a constrained age group, or have been based solely on inpatient records that do not permit the longer-term follow-up critical to determining the effect of odontoid fractures on overall mortality.^36-39

Likewise, mortality rates after fractures of the subaxial spine (ie, the motion segments between C3 and C7) have yet to be established. In 1 study, the mortality risk of a cohort of elderly patients with cervical fractures appeared to be elevated for the first 6 to 12 months after the traumatic event.⁴⁰ However, the sample size was too small to examine mortality beyond 1 year.

In this context, the purpose of the current study was to determine the mortality rates at several time points (3 months, 1 year, and 2 years) of patients 50 years or older (start of the second mode of the bimodal age distribution of odontoid fractures^41-44) with fractures of the odontoid and subaxial cervical spine. A secondary purpose of this study was to compare survival rates of these 2 cohorts relative to each other and to the general population.

Materials and Methods

Identification of Cervical Fractures and Collection of Demographic Information

This protocol was approved by the human investigation committee of our institution. Every computed tomography (CT) scan of the cervical spine performed in the emergency department (ED) of an academic hospital between November 27, 1997, and December 31, 2006, was identified. Since the threshold for obtaining a CT scan of a patient with suspected cervical spine trauma is relatively low, it was assumed that virtually all acute cervical spine fractures during this time period would be successfully identified through this approach.

Radiology reports for all identified CT scans were reviewed for any findings consistent with acute fractures and/or dislocations of the cervical spine (Figure 1). Every study noted to be positive or equivocal for cervical trauma was directly visualized, including those that did not specifically mention the presence or absence of an injury. Scans with no signs of acute trauma or that showed fractures caused by a pathologic process or penetrating mechanism (eg, metastatic lesions or gunshot wounds) were omitted from this series. Finally, relevant demographic information, such as the medical record number, age, gender, and date of study, was recorded for every subject in this group.

Fracture Classification

Next, the level and the type of cervical injury were documented for each patient. Fractures were segregated according to their involvement with the odontoid or the subaxial vertebrae. 

Odontoid fractures were categorized into type I (limited to the tip), type II (across the base of the process) and type III (through the base with extension into the C2 vertebral body).^45,46 Since many systems for classifying subaxial cervical spine trauma require a subjective inference of the injury mechanism, which is difficult to ascertain from imaging studies alone, all of these fractures were pooled together.

A preliminary survey of the data indicated that the odontoid fractures appeared to exhibit a bimodal age distribution, with the beginning of the second cluster occurring around age 50 years (Figure 2). As noted above, this has been shown in previous studies.^41-44 As a consequence, the mortalities of those older than 50 years became the focus of this study. To control for comorbid conditions, mechanism of injury, and to allow for more direct comparison with the odontoid fractures in this study, the same age demarcation was used for subaxial cervical fractures.

Mortality Data

The mortality status of every patient diagnosed with an acute cervical injury at our institution between November 27, 1997, and December 31, 2006, was determined by referencing the National Death Index (NDI). The NDI is a computerized database of death records maintained by the National Center for Health Statistics (NCHS). The time window for the current study was selected because we had access to NDI information only through 2007 at the time of this study. Social Security numbers (SSNs), which were available for approximately half of the subjects, were used to search the NDI catalog. For individuals whose SSNs were unavailable, patient names and birthdates were considered to be sufficient to confirm a true match. Our center’s medical records of this cohort were also examined to verify whether any had died during their initial hospitalizations and to substantiate the NDI data. Finally, patient deaths were categorized as trauma (eg, motor vehicle accident, fall from a height) or medical comorbidity (eg, diabetes mellitus, cancer, congestive heart failure), based on information in the NDI listing.

Age- and Sex-Matched Controls

Age- and sex-matched controls were determined from the Wide-ranging Online Data for Epidemiologic Research (WONDER) application distributed by the Centers for Disease Control and Prevention (http://wonder.cdc.gov). Composite mortality data from the state in which the study was performed was obtained for the years between 1999 and 2007, and this information was further stratified according to gender and age to estimate the mortality rates and construct survival curves for each group. Controls were used to establish a standardized mortality ratio (SMR) for subjects 50 years and older, a value that compares the number of observed deaths with the figure expected for matched populations from the general population.

Statistical Methods

Statistical analyses were performed by using both SAS 9.2 (SAS Institute Inc., Cary, North Carolina) and R (version 2.9; www.r-project.org, Auckland, New Zealand). Relevant comparisons were planned, and all tests were 2-sided. The Wilcoxon rank sum test was applied to compare the survival times of patients with odontoid fractures with different documented causes of death, and Pearson χ² test was used to compare the age distributions of odontoid and subaxial fractures. Survival rates at 3 months, 1 year, and 2 years were estimated from Kaplan-Meier curves. The relative survival of these cohorts was compared by completing a 2-sample log-rank test. In addition, a 1-sample log-rank test was implemented to compare the mortality from either odontoid or subaxial cervical spine fractures with that of the age- and gender-matched general population. Statistical significance was defined as a 2-sided α error of less than 0.05 (P < .05).

Results

Fifty-nine patients were diagnosed with odontoid fractures (28 men, 31 women), and 233 patients were diagnosed with subaxial cervical spine fractures (168 men, 65 women).

Odontoid Fracture Patients

Odontoid fracture patients exhibited a distinct bimodal age distribution (Figure 2). In the younger population, there were 14 subjects, 3 of whom died within days of the injury (mean, 12 days; 78.6% survival). At 2-year follow-up, there were no further deaths. The fractures that caused death were high-energy injuries, and only early deaths occurred in these cases. 

Because of the significant bimodal age distribution, it was believed these cohorts could not be directly compared. As a result, the remaining analysis focused on the older age group. In the older population mode (50 years and older) were 45 patients with odontoid fractures. Of the 12 subjects who died after odontoid fracture, 5 were assigned a trauma code as the cause of death, while a medical comorbidity code was assigned for the remaining 7. Mean survival time of those who died secondary to trauma was significantly shorter than the medical comorbidity group (P = .025). 

In the cohort of subjects older than 50 years, 3-month, 1-year, and 2-year survival rates were 84.4%, 82.2%, and 72.9%, respectively. Figure 2 shows the 1- and 2-year follow-up data by age group.

Analysis was performed relative to gender. Of male patients (n = 22), the 3-month, 1-year, and 2-year survival rates were 72.7%, 72.7%, and 62.7%, respectively. Among women (n = 23), the 3-month, 1-year, and 2-year survival rates were 95.7%, 91.3%, and 82.6%, respectively.

Figure 3 shows the Kaplan-Meier survival curves of the older patients with odontoid fractures. A comparison of the curves for each gender showed no significant disparities between the male and female survival (Figure 3A, P = .124). Compared with age-matched male counterparts, the survival of male subjects with odontoid fractures was significantly worse (Figure 3B, P < .001). Men experienced an initial acute decline in survival, with the remainder of the survival curve matching that of the general male population. In contrast, odontoid fractures did not adversely affect female survival compared with the matched population (Figure 3C, P = .568). 

The 2-year SMR of 2.98 for men showed that odontoid fractures led to greater mortality compared with a sex- and age-matched population. This means that men older than 50 years who sustained an odontoid fracture had nearly 3 times the mortality rate after 2 years compared with a normal, matched population; this increase is attributed to the 3-month time point that subsequently normalized. The female rate was 1.33 times that of a matched population, a difference that is not statistically significant.

Subaxial Fracture Patients

Of the 91 patients older than 50 years with subaxial fractures, 3-month, 1-year, and 2-year survival rates were 87.9%, 85.7%, and 85.7%, respectively. Figure 4 shows the 1- and 2-year follow-up data by age group.

Gender-specific analysis was performed. For men (n = 58), the 3-month, 1-year, and 2-year survival rates were 87.9%, 84.5%, and 84.5%, respectively. Among women (n = 33), 4 deaths were recorded at all time points (87.9% survival). 

Figure 5 shows Kaplan-Meier survival curves for the older population with subaxial fractures. A comparison of the curves between genders again showed no significant differences between male and female survival (P = .683, Figure 5A). Compared with age- and gender-matched counterparts, men showed decreased relative survival (P < .0001, Figure 5B), whereas subaxial fractures did not decrease female survival (P = .554, Figure 5C).

The 2-year SMR of 2.90 for men showed higher mortality rates relative to sex- and age-matched controls. Men who were both 50 years old and sustained a subaxial fracture were 2.9 times as likely to die within 2 years of follow-up compared with their counterparts. Similar to odontoid fractures, this increase occurred by the 3-month time point and subsequently normalized. The female rate, which was 1.34 times that of the uninjured population, was not statistically significant.

Comparison of Odontoid and Subaxial Fracture Patients

The survival of subaxial injuries was not significantly different from that of odontoid fractures (P = .113, Figure 6A). When analyzed by gender and controlled for age, the rates in both male (P = .347, Figure 6B) and female (P = .643, Figure 6C) patients did not differ between fracture types.

Discussion

The US population is aging rapidly, with the demographic older than 65 years predicted to more than double in size between 2010 and 2050.⁴⁷ As our elderly population grows, the incidence of age-related injuries will rise accordingly. An understanding of mortality risks associated with different fractures will not only assist practitioners in advising patients regarding prognosis but may also lead to improvements in clinical care.^19,48-50 While we know cervical spine trauma is associated with significant morbidity,^29-31 little is known about associated moderate-term mortality rates that can be compared with other known injury patterns, such as hip fractures or osteoporotic compression fractures. 

An interesting finding of the present study is the bimodal age distribution of the 59 odontoid fractures (Figure 2). The 14 patients younger than 50 years included 3 individuals who died, all within days of their presentation from severe multisystem trauma. This is consistent with the determination that high-energy forces are required to fracture the odontoid process in younger individuals.^{38,45,46,51,52} Given the severity of their nonspinal injuries, the cause of death was likely not primarily related to their odontoid fractures. Also in line with previous studies, the majority (76%) of odontoid fractures were documented in subjects older than 50 years.^32,53,54 Within our cohort older than 50 years, the deaths appear to be spread evenly across age groups and do not seem to be skewed by the oldest portion of the population (Figure 2).

Our gender-specific analyses revealed that older men with odontoid injuries exhibited higher mortality compared with an age-matched male cohort, with 6 of the 8 deaths occurring within 3 months. However, after this exaggerated decline in survival, the rate normalized towards general population mortality rates (Figure 3B). As in the younger cohort, these earlier deaths were largely attributable to multisystem trauma, whereas medical comorbidities were implicated in those who died later. In contrast, the Kaplan-Meier curve of older women with odontoid fractures closely approximates that of age-matched women at every time point (Figure 3C), indicating that these injuries do not decrease survival as they do in their male counterparts.

When comparing the survival of older patients with subaxial cervical spine fractures with that of gender- and age-matched controls, the mortality rates of women were, once again, essentially equivalent. However, the survival of older men was significantly compromised by these injuries. In men, 7 of the 9 deaths were within 3 months, with the remaining 2 deaths occurring within 7 months. Nevertheless, beyond this initial period of elevated mortality, the survival curve again stabilized and paralleled that of the general population. As with odontoid fractures, there was no sustained increase in the mortality of male patients who lived at least 3 months after injury.

The mortality rates of odontoid and subaxial fractures were also compared in the older population. When controlled for age, there was no difference in mortality rates between these 2 groups. When individually analyzed in both men and women, the mortality rates of both fracture types matched those of the general population at all time points.

It is useful to contextualize our findings alongside the mortality of older individuals with other fracture types. Based on our results, we believe that the survival curves of geriatric men with odontoid or subaxial cervical spine fractures most closely resemble the characteristic pattern seen in hip fractures. Hip fractures have shown an early spike in mortality by as much as 8% to 49% in the first 6 to 12 months that returns to baseline after 1 year.^1-10 This presumably reflects the natural history of these injuries in response to appropriate therapeutic interventions. Interestingly, the male mortality rates for both odontoid and subaxial cervical spine fractures in this study are largely analogous to those reported by various hip fracture surveys.^1,5,55-58 In contrast, similar to prior studies of rib or wrist fractures, older women with these cervical spine fractures did not show a survival decrease after their injuries.^12,16-18

While the reasons underlying the differential effects of cervical fractures on the mortality of men and women have not been established, one explanation is that the female geriatric population is relatively more osteoporotic; thus, cervical injuries may occur after lower-energy forces, leading to less severe associated trauma that could otherwise decrease survival. Another explanation is that men are more likely to be involved in high-energy accidents,^59,60 thus decreasing their overall survival after injury.

This investigation is not without limitations. Our primary concern is the determination of survival. The NDI maintained by the NCHS is an extremely reliable tool regularly employed by epidemiologists to collect mortality data. However, it is possible that deaths may have been missed. We believe this number would be small, because the NDI database provided multiple probable matches that were carefully compared with supplemental personal information. It is also possible that deaths that were not appropriately registered with the NDI are not represented in this series. Another limitation lies in the determination of controls. As with any case–control study, the patients sustaining these odontoid fractures may differ in some significant way from the average population. A final limitation is that a small portion of patients in the study have only 1-year follow-up, because patient data was collected through 2006, although access to NDI data ended in 2007.

Conclusion

Our results indicate that the survival of older men with either odontoid or subaxial cervical spine fractures shares many of the same mortality characteristics as hip fractures, with diminished survival in the first 3 months that normalizes to the survival rate of the age-matched population. Interestingly, and perhaps because of disparate rates of osteoporosis and traumatic forces, the mortality rates in the female cohort were similar to that of the age-matched general population at all time points. These trends were nearly identical for both odontoid and subaxial cervical fractures.

Mortality rate is an important indicator of the severity of traumatic injuries, and these values have been described for different orthopedic injuries and fractures. Studies have identified 3 distinct trends in patient survival when compared with the age- and sex-matched uninjured population: 

1.  Hip fractures bring about a transient increase in mortality relative to age-matched controls that normalizes after a few months to 1 year.^1-10

2.  Thoracic and lumbar compression fractures are associated with an ongoing, lifelong increase in mortality rate relative to age-matched controls without an initial marked upswing.^11-15

3.  Certain injuries such as isolated rib or wrist fractures do not adversely affect survival relative to age-matched controls.^12,16-18

Understanding the mortality patterns after these injuries can help guide management and even facilitate the development of appropriate treatment algorithms.^19-21 While studies have examined mortality in specific odontoid fracture types,²² such mortality trends have not been broadly established in persons with cervical spine fractures. 

Cervical spine fractures are common: 60% of spine fractures localize to this region,^23-26 and this equates to 2% to 3% of all blunt-trauma patients.^27,28 These injuries can lead to devastating consequences, including neurologic compromise, permanent disability, and death.^29-31

Studies have estimated that up to 20% of cervical fractures involve the odontoid process.^23-26 These injuries are more common among the elderly population because of their greater prevalence of osteoporosis and likelihood of falling.³² Because of demographic similarities to those of the hip fracture population, a survival analysis of all odontoid fractures is particularly interesting. Published odontoid mortality rates vary significantly, with reports ranging from 13% to 44%.^22,33-35 Unfortunately, these studies largely evaluated survival rates specific to an individual treatment modality, such as nonoperative compared with operative, or specific to certain odontoid fracture types (eg, type II). Additionally, studies have generally only considered survivorship during initial hospitalization, have been specific to a constrained age group, or have been based solely on inpatient records that do not permit the longer-term follow-up critical to determining the effect of odontoid fractures on overall mortality.^36-39

Likewise, mortality rates after fractures of the subaxial spine (ie, the motion segments between C3 and C7) have yet to be established. In 1 study, the mortality risk of a cohort of elderly patients with cervical fractures appeared to be elevated for the first 6 to 12 months after the traumatic event.⁴⁰ However, the sample size was too small to examine mortality beyond 1 year.

In this context, the purpose of the current study was to determine the mortality rates at several time points (3 months, 1 year, and 2 years) of patients 50 years or older (start of the second mode of the bimodal age distribution of odontoid fractures^41-44) with fractures of the odontoid and subaxial cervical spine. A secondary purpose of this study was to compare survival rates of these 2 cohorts relative to each other and to the general population.

Materials and Methods

Identification of Cervical Fractures and Collection of Demographic Information

This protocol was approved by the human investigation committee of our institution. Every computed tomography (CT) scan of the cervical spine performed in the emergency department (ED) of an academic hospital between November 27, 1997, and December 31, 2006, was identified. Since the threshold for obtaining a CT scan of a patient with suspected cervical spine trauma is relatively low, it was assumed that virtually all acute cervical spine fractures during this time period would be successfully identified through this approach.

Radiology reports for all identified CT scans were reviewed for any findings consistent with acute fractures and/or dislocations of the cervical spine (Figure 1). Every study noted to be positive or equivocal for cervical trauma was directly visualized, including those that did not specifically mention the presence or absence of an injury. Scans with no signs of acute trauma or that showed fractures caused by a pathologic process or penetrating mechanism (eg, metastatic lesions or gunshot wounds) were omitted from this series. Finally, relevant demographic information, such as the medical record number, age, gender, and date of study, was recorded for every subject in this group.

Fracture Classification

Next, the level and the type of cervical injury were documented for each patient. Fractures were segregated according to their involvement with the odontoid or the subaxial vertebrae. 

Odontoid fractures were categorized into type I (limited to the tip), type II (across the base of the process) and type III (through the base with extension into the C2 vertebral body).^45,46 Since many systems for classifying subaxial cervical spine trauma require a subjective inference of the injury mechanism, which is difficult to ascertain from imaging studies alone, all of these fractures were pooled together.

A preliminary survey of the data indicated that the odontoid fractures appeared to exhibit a bimodal age distribution, with the beginning of the second cluster occurring around age 50 years (Figure 2). As noted above, this has been shown in previous studies.^41-44 As a consequence, the mortalities of those older than 50 years became the focus of this study. To control for comorbid conditions, mechanism of injury, and to allow for more direct comparison with the odontoid fractures in this study, the same age demarcation was used for subaxial cervical fractures.

Mortality Data

The mortality status of every patient diagnosed with an acute cervical injury at our institution between November 27, 1997, and December 31, 2006, was determined by referencing the National Death Index (NDI). The NDI is a computerized database of death records maintained by the National Center for Health Statistics (NCHS). The time window for the current study was selected because we had access to NDI information only through 2007 at the time of this study. Social Security numbers (SSNs), which were available for approximately half of the subjects, were used to search the NDI catalog. For individuals whose SSNs were unavailable, patient names and birthdates were considered to be sufficient to confirm a true match. Our center’s medical records of this cohort were also examined to verify whether any had died during their initial hospitalizations and to substantiate the NDI data. Finally, patient deaths were categorized as trauma (eg, motor vehicle accident, fall from a height) or medical comorbidity (eg, diabetes mellitus, cancer, congestive heart failure), based on information in the NDI listing.

Age- and Sex-Matched Controls

Age- and sex-matched controls were determined from the Wide-ranging Online Data for Epidemiologic Research (WONDER) application distributed by the Centers for Disease Control and Prevention (http://wonder.cdc.gov). Composite mortality data from the state in which the study was performed was obtained for the years between 1999 and 2007, and this information was further stratified according to gender and age to estimate the mortality rates and construct survival curves for each group. Controls were used to establish a standardized mortality ratio (SMR) for subjects 50 years and older, a value that compares the number of observed deaths with the figure expected for matched populations from the general population.

Statistical Methods

Statistical analyses were performed by using both SAS 9.2 (SAS Institute Inc., Cary, North Carolina) and R (version 2.9; www.r-project.org, Auckland, New Zealand). Relevant comparisons were planned, and all tests were 2-sided. The Wilcoxon rank sum test was applied to compare the survival times of patients with odontoid fractures with different documented causes of death, and Pearson χ² test was used to compare the age distributions of odontoid and subaxial fractures. Survival rates at 3 months, 1 year, and 2 years were estimated from Kaplan-Meier curves. The relative survival of these cohorts was compared by completing a 2-sample log-rank test. In addition, a 1-sample log-rank test was implemented to compare the mortality from either odontoid or subaxial cervical spine fractures with that of the age- and gender-matched general population. Statistical significance was defined as a 2-sided α error of less than 0.05 (P < .05).

Results

Fifty-nine patients were diagnosed with odontoid fractures (28 men, 31 women), and 233 patients were diagnosed with subaxial cervical spine fractures (168 men, 65 women).

Odontoid Fracture Patients

Odontoid fracture patients exhibited a distinct bimodal age distribution (Figure 2). In the younger population, there were 14 subjects, 3 of whom died within days of the injury (mean, 12 days; 78.6% survival). At 2-year follow-up, there were no further deaths. The fractures that caused death were high-energy injuries, and only early deaths occurred in these cases. 

Because of the significant bimodal age distribution, it was believed these cohorts could not be directly compared. As a result, the remaining analysis focused on the older age group. In the older population mode (50 years and older) were 45 patients with odontoid fractures. Of the 12 subjects who died after odontoid fracture, 5 were assigned a trauma code as the cause of death, while a medical comorbidity code was assigned for the remaining 7. Mean survival time of those who died secondary to trauma was significantly shorter than the medical comorbidity group (P = .025). 

In the cohort of subjects older than 50 years, 3-month, 1-year, and 2-year survival rates were 84.4%, 82.2%, and 72.9%, respectively. Figure 2 shows the 1- and 2-year follow-up data by age group.

Analysis was performed relative to gender. Of male patients (n = 22), the 3-month, 1-year, and 2-year survival rates were 72.7%, 72.7%, and 62.7%, respectively. Among women (n = 23), the 3-month, 1-year, and 2-year survival rates were 95.7%, 91.3%, and 82.6%, respectively.

Figure 3 shows the Kaplan-Meier survival curves of the older patients with odontoid fractures. A comparison of the curves for each gender showed no significant disparities between the male and female survival (Figure 3A, P = .124). Compared with age-matched male counterparts, the survival of male subjects with odontoid fractures was significantly worse (Figure 3B, P < .001). Men experienced an initial acute decline in survival, with the remainder of the survival curve matching that of the general male population. In contrast, odontoid fractures did not adversely affect female survival compared with the matched population (Figure 3C, P = .568). 

The 2-year SMR of 2.98 for men showed that odontoid fractures led to greater mortality compared with a sex- and age-matched population. This means that men older than 50 years who sustained an odontoid fracture had nearly 3 times the mortality rate after 2 years compared with a normal, matched population; this increase is attributed to the 3-month time point that subsequently normalized. The female rate was 1.33 times that of a matched population, a difference that is not statistically significant.

Subaxial Fracture Patients

Of the 91 patients older than 50 years with subaxial fractures, 3-month, 1-year, and 2-year survival rates were 87.9%, 85.7%, and 85.7%, respectively. Figure 4 shows the 1- and 2-year follow-up data by age group.

Gender-specific analysis was performed. For men (n = 58), the 3-month, 1-year, and 2-year survival rates were 87.9%, 84.5%, and 84.5%, respectively. Among women (n = 33), 4 deaths were recorded at all time points (87.9% survival). 

Figure 5 shows Kaplan-Meier survival curves for the older population with subaxial fractures. A comparison of the curves between genders again showed no significant differences between male and female survival (P = .683, Figure 5A). Compared with age- and gender-matched counterparts, men showed decreased relative survival (P < .0001, Figure 5B), whereas subaxial fractures did not decrease female survival (P = .554, Figure 5C).

The 2-year SMR of 2.90 for men showed higher mortality rates relative to sex- and age-matched controls. Men who were both 50 years old and sustained a subaxial fracture were 2.9 times as likely to die within 2 years of follow-up compared with their counterparts. Similar to odontoid fractures, this increase occurred by the 3-month time point and subsequently normalized. The female rate, which was 1.34 times that of the uninjured population, was not statistically significant.

Comparison of Odontoid and Subaxial Fracture Patients

The survival of subaxial injuries was not significantly different from that of odontoid fractures (P = .113, Figure 6A). When analyzed by gender and controlled for age, the rates in both male (P = .347, Figure 6B) and female (P = .643, Figure 6C) patients did not differ between fracture types.

Discussion

The US population is aging rapidly, with the demographic older than 65 years predicted to more than double in size between 2010 and 2050.⁴⁷ As our elderly population grows, the incidence of age-related injuries will rise accordingly. An understanding of mortality risks associated with different fractures will not only assist practitioners in advising patients regarding prognosis but may also lead to improvements in clinical care.^19,48-50 While we know cervical spine trauma is associated with significant morbidity,^29-31 little is known about associated moderate-term mortality rates that can be compared with other known injury patterns, such as hip fractures or osteoporotic compression fractures. 

An interesting finding of the present study is the bimodal age distribution of the 59 odontoid fractures (Figure 2). The 14 patients younger than 50 years included 3 individuals who died, all within days of their presentation from severe multisystem trauma. This is consistent with the determination that high-energy forces are required to fracture the odontoid process in younger individuals.^{38,45,46,51,52} Given the severity of their nonspinal injuries, the cause of death was likely not primarily related to their odontoid fractures. Also in line with previous studies, the majority (76%) of odontoid fractures were documented in subjects older than 50 years.^32,53,54 Within our cohort older than 50 years, the deaths appear to be spread evenly across age groups and do not seem to be skewed by the oldest portion of the population (Figure 2).

Our gender-specific analyses revealed that older men with odontoid injuries exhibited higher mortality compared with an age-matched male cohort, with 6 of the 8 deaths occurring within 3 months. However, after this exaggerated decline in survival, the rate normalized towards general population mortality rates (Figure 3B). As in the younger cohort, these earlier deaths were largely attributable to multisystem trauma, whereas medical comorbidities were implicated in those who died later. In contrast, the Kaplan-Meier curve of older women with odontoid fractures closely approximates that of age-matched women at every time point (Figure 3C), indicating that these injuries do not decrease survival as they do in their male counterparts.

When comparing the survival of older patients with subaxial cervical spine fractures with that of gender- and age-matched controls, the mortality rates of women were, once again, essentially equivalent. However, the survival of older men was significantly compromised by these injuries. In men, 7 of the 9 deaths were within 3 months, with the remaining 2 deaths occurring within 7 months. Nevertheless, beyond this initial period of elevated mortality, the survival curve again stabilized and paralleled that of the general population. As with odontoid fractures, there was no sustained increase in the mortality of male patients who lived at least 3 months after injury.

The mortality rates of odontoid and subaxial fractures were also compared in the older population. When controlled for age, there was no difference in mortality rates between these 2 groups. When individually analyzed in both men and women, the mortality rates of both fracture types matched those of the general population at all time points.

It is useful to contextualize our findings alongside the mortality of older individuals with other fracture types. Based on our results, we believe that the survival curves of geriatric men with odontoid or subaxial cervical spine fractures most closely resemble the characteristic pattern seen in hip fractures. Hip fractures have shown an early spike in mortality by as much as 8% to 49% in the first 6 to 12 months that returns to baseline after 1 year.^1-10 This presumably reflects the natural history of these injuries in response to appropriate therapeutic interventions. Interestingly, the male mortality rates for both odontoid and subaxial cervical spine fractures in this study are largely analogous to those reported by various hip fracture surveys.^1,5,55-58 In contrast, similar to prior studies of rib or wrist fractures, older women with these cervical spine fractures did not show a survival decrease after their injuries.^12,16-18

While the reasons underlying the differential effects of cervical fractures on the mortality of men and women have not been established, one explanation is that the female geriatric population is relatively more osteoporotic; thus, cervical injuries may occur after lower-energy forces, leading to less severe associated trauma that could otherwise decrease survival. Another explanation is that men are more likely to be involved in high-energy accidents,^59,60 thus decreasing their overall survival after injury.

This investigation is not without limitations. Our primary concern is the determination of survival. The NDI maintained by the NCHS is an extremely reliable tool regularly employed by epidemiologists to collect mortality data. However, it is possible that deaths may have been missed. We believe this number would be small, because the NDI database provided multiple probable matches that were carefully compared with supplemental personal information. It is also possible that deaths that were not appropriately registered with the NDI are not represented in this series. Another limitation lies in the determination of controls. As with any case–control study, the patients sustaining these odontoid fractures may differ in some significant way from the average population. A final limitation is that a small portion of patients in the study have only 1-year follow-up, because patient data was collected through 2006, although access to NDI data ended in 2007.

Conclusion

Our results indicate that the survival of older men with either odontoid or subaxial cervical spine fractures shares many of the same mortality characteristics as hip fractures, with diminished survival in the first 3 months that normalizes to the survival rate of the age-matched population. Interestingly, and perhaps because of disparate rates of osteoporosis and traumatic forces, the mortality rates in the female cohort were similar to that of the age-matched general population at all time points. These trends were nearly identical for both odontoid and subaxial cervical fractures.

Assessment of cervical spine range of motion (ROM) is an integral aspect of the physical examination for cervical conditions,^1-3 surgical outcomes,⁴ and functional impairment.¹ In fact, the emphasis being placed on such functional measures before and after treatments is increasing.^4,5

Cervical spine range of motion is routinely used as an outcome measure in clinical studies.^6-8 Underscoring the importance of defining cervical spine ROM, studies have found it to be a preoperative predictor of outcomes of anterior cervical surgery,⁹ and other studies have suggested it is a determinant of athletes’ return to play.¹⁰

Spinal ROM measurements can be used to determine the degree of disability experienced by a patient with a spinal condition as defined in the Guides to the Evaluation of Permanent Impairment by the American Medical Association (AMA).¹ In the medicolegal realm, ROM measurements made by clinicians can influence the dollar amounts of awards in legal claims, and, according to the AMA guides, the difference in cervical spine ROM between normality and disability or impairment can be as little as 5°.

Although cervical spine ROM is routinely assessed and documented in clinical practice, no universal protocol exists for its evaluation.^11,12 In fact, considerable inter-examiner variation in visual estimates of ROM has been found,^13-16 and significant inaccuracies have been reported.^17,18

Goniometers have been shown to be reliable and highly accurate, with low inter-examiner and intra-examiner variability.^5,19-21 Nevertheless, logistics²² and costs²¹ generally limit their being accepted in routine clinical practice. Among many methods available for assessing ROM, visual estimation is the least reliable or accurate,²³ but it is the quickest and least expensive and is recommended in textbooks that describe the spinal-specific physical examination.²⁴ Despite the superiority of goniometers in measuring ROM, these significant barriers have limited their use in clinical practice. When assessing cervical spine ROM, most clinicians prefer visual estimates over goniometers.

We conducted a study to determine whether training could improve the accuracy of visual estimates. We compared the accuracy of visual estimates of cervical spine ROM with that of a radiographically validated electrogoniometer and then investigated whether accuracy and reliability of visual estimates could be improved with a session of instruction and demonstration. Assessments of accuracy were made immediately after and 1 month after this training session.

Materials and Methods

Assessments Made Before Training

This study was approved by our institution’s human investigation committee and was conducted in accordance with the ethical standards of that committee.

Cervical spine ROM was assessed by 8 examiners (2 attending spine surgeons, 4 orthopedic residents, 2 medical students). They were informed they would be participating in a study evaluating visual estimates of motion but were given no other information prior to the study.

Four healthy volunteer subjects (examiners who rotated through the role) were assessed. No subject reported any ongoing neck or spine discomfort or had had any previous spinal surgery. One at a time, subjects were fitted with a cervical harness electrogoniometer capable of measuring angulation of the cervical spine to the nearest degree (modified electrogoniometer, torsiometer, and display from Biometrics, Gwent, UK; Figures 1A, 1B). This electrogoniometer has been shown to have a mean (SD) error of 2.3° (2.6°) relative to radiographic assessments.⁸

With the electrogoniometer fitted, each subject was instructed to sit upright in a chair with his back to the backrest and his head neutrally positioned. The electrogoniometer was then zeroed, and the subject proceeded with 5 series of flexion-extension, left and right lateral bending, and left and right rotation movements. The subject was instructed to make 1 movement in full motion in each direction and the other 4 movements in less than full motion to yield a variety of excursions for assessment. Each subject was instructed to pause at the apex of each motion. During these pauses, the examiners recorded their visual estimates of movement in each direction while the investigator recorded degrees of motion (displayed by the electrogoniometer) in flexion-extension, lateral bending, and rotation (Figures 2A–2D). The electrogoniometer display was not visible to subjects or examiners.

A total of 840 independent visual estimates of 120 distinct movements were recorded.

Training, and Assessments Made Immediately Thereafter

After the first round of visual estimates, the 8 examiners were verbally instructed in cervical spine ROM assessment and were asked to observe 1 subject, fitted with the electrogoniometer, demonstrating partial and full cervical motions while the investigator announced the electrogoniometric measurements. The motions demonstrated included 15°, 30°, and the extremes of cervical spine ROM in each of 6 directions from neutral.

After this training session, each of the 4 subjects from the first round of assessments was again fitted with the harness electrogoniometer and instructed to repeat the movements in turn while examiners visually estimated cervical spine ROM and independently recorded their estimates. Meanwhile, the investigator recorded the degree of motion during each movement (as measured by the electrogoniometer). Again, a total of 840 independent visual estimates of 120 distinct movements were recorded.

Assessments Made 1 Month After Training

One month after the training session, the examiners and the investigator reconvened to assess the same 4 subjects using a procedure for simultaneous visual estimation and electrogoniometric measurement identical to that used 1 month earlier. No additional training was given. Again, 840 independent visual estimates of 120 distinct movements were recorded.

Data Analysis

The reliabilities of visual estimates were analyzed by calculating the intraclass coefficients (ICCs) using random-effect 1-way analyses of variance. By convention, ICCs of < 0.2, 0.2 to 0.39, 0.4 to 0.59, 0.6 to 0.8, and > 0.8 correspond to poor, fair, moderate, substantial, and perfect reliability, respectively.²⁵

We compared the visual estimates and electrogoniometric measurements made for 3 planes of motion (flexion-extension, lateral bending, axial rotation) before, immediately after, and 1 month after training and drew trend lines generated by linear regression relative to a line of perfect correlation.

Mean errors in examiners’ visual estimates (relative to elec­trogoniometric measurements) made before, immediately after, and 1 month after training were calculated. Paired Student t tests were then used to compare the mean errors before training with the mean errors immediately after and 1 month after training.

All analyses were performed with SPSS for Windows 16.0 (SPSS, Chicago, Illinois).

Results

Inter-examiner reliability of the visual estimates in all planes of motion ranged from 0.51 to 0.79 (suggestive of moderate to substantial reliability). For reference, standard goniometers measuring knee ROM have inter-examiner ICCs of 0.89 to 0.98²⁶ (suggestive of perfect reliability). The ICCs before, immediately after, and 1 month after training were not significantly different. 

As expected, there were significant errors in visual estimates of cervical spine ROM in all planes. Initial errors in visual estimates (relative to electrogoniometric measurements) were 23.9° (flexion-extension), 15.5° (lateral bending), and 19.3° (axial rotation) (Table, Figure 3).

Immediately after training, mean errors in visual estimates decreased to 12.0° (flexion-extension), 11.7° (lateral bending), and 16.4° (axial rotation) (Table, Figure 3). In all 3 planes of cervical motion, these improvements were statistically significant.

One month after training, mean errors in visual estimates were 14.4° (flexion-extension), 13.9° (lateral bending), and 16.2° (axial rotation) (Table, Figure 3). Only the improvement in the estimate of flexion-extension (the direction of the largest error initially) remained statistically significant—a 39.7% decrease in error.

We also considered how errors varied with degree of motion observed. In flexion-extension, the tendency to overestimate at larger degrees of motion was not apparent after training, and 1 month after training we found a tendency to underestimate at smaller degrees of motion (Figure 4A). The tendency to overestimate lateral bending before training did not persist immediately after or 1 month after training (Figure 4B). Estimates of axial rotation correlated well with goniometer measurements before training and were also well correlated immediately after and 1 month after training (Figure 4C).

Discussion

Visual estimation of spinal motion is unreliable and inaccurate, but its widespread use in clinical practice continues. Goniometers are far more accurate and reliable but are seldom used. We investigated whether a training session featuring verbal instruction and demonstration with an electrogoniometer could improve visual estimates and whether potential improvement in visual estimates would remain 1 month after training.

Widely variable ICCs (0.42-0.90) have been reported for visual estimates of cervical spine ROM.^17,18,22 Our findings on the reliability of these estimates are consistent with the literature.

We recorded the greatest initial error in estimates of motion in flexion-extension. Previous studies have also found the greatest error and least reliability in visual estimates in this plane.^14,15,18 Visual estimation may be more difficult in flexion-extension because the shoulders cannot be used as landmarks, whereas they serve as approximate 90° reference points during estimation of lateral bending and axial rotation. Demonstration of 15°, 30° and the extremes of ROM during the training session may have provided alternative reference points during visual estimation after training—decreasing the error to within the range found in other planes of motion.

Initial errors in visual estimates were 23.9° (flexion-extension), 15.5° (lateral bending), and 19.3° (axial rotation). Based on normative cervical spine ROM in a healthy population— 126° ± 12° for flexion-extension, 86° ± 5° for lateral bending, 151° ± 23° for axial rotation²²—the errors we identified are 18.9% of the normal range of flexion-extension, 18.0% of lateral bending, and 12.8% of axial rotation.

Training clearly improved the accuracy of visual estimates of cervical spine ROM. Estimates were statistically improved for all planes immediately after training and remained significantly improved for flexion-extension (the plane of largest error initially) 1 month after training. Before training, mean errors varied across planes. Training normalized mean errors to about 15°, and this effect lasted in flexion-extension, lateral bending, and axial rotation (Figures 4A–4C). Of note, before training these percentage errors increased with increased motion from neutral in the flexion-extension and lateral bending planes. At full ROM, percentage errors in estimates were greater. After training, percentage errors did not increase appreciably with increasing motion.

Readers will naturally reflect on the clinical significance of the motion assessment improvements demonstrated after the training session described in this study. We must be aware that functional assessments are increasingly being emphasized in the clinical arena—with respect to clinical conditions, surgical outcomes, and functional impairments. We highlight a point made earlier: A difference of only 5° can affect impairment ratings in the medicolegal realm.¹ In estimating flexion-extension motion, lasting improvements of almost 10° were demonstrated and maintained 1 month after the training session described in this study.

Nevertheless, mean errors in visual estimation remained at about 15° in all planes of motion, despite our modest improvements. This finding raises the question of whether visually estimated ROM should be pertinent to assessments of impairment and disability. Although visual estimates of ROM may have more utility as a screening test for impairment and disability, fine differences in ROM simply cannot be reliably assessed by visual estimation.

This study has limitations. First, it was conducted at a single institution where the evaluators received most of their training. Their skill in visually estimating cervical spine ROM may not be generalizable to a larger population of spine specialists who are practicing at other institutions and may have different training backgrounds.

Second, only healthy subjects were assessed. Some studies of cervical spine ROM have shown better reliability in symptomatic subjects relative to asymptomatic subjects.^13,14 To attempt to overcome this limitation, we assessed many different excursions of motion that were often not to the extremes of motion.

Third, the “gold standard” we used for motion assessment was an electrogoniometer, which has some inherent error (previously validated mean [SD] error of 2.3° [2.6°] relative to radiographs⁸). Although obtaining radiographs of each movement would have more closely resembled the gold standard, the radiation dose associated with such a study is prohibitive.

Last, the assessors included medical students. The medical students’ estimates, however, tended to be more accurate than the residents’ or attending surgeons’ (though the difference was not statistically significant). This tendency may reflect the medical students’ closer attention to detail.  Clearly, including medical students in the study did not negatively affect the accuracy of the estimates or the validity of our findings.

Conclusion

Despite its limitations, visual assessment of cervical spine motion remains the gold standard in clinical practice and is routinely recorded and reported. Mean errors ranged from 15.5° to 23.9°, depending on plane of motion being assessed, but these improved after a training session.

Visual estimates of motion in flexion-extension were most improved by training, as the initial errors in this plane were the largest. Statistically significant improvement of about 10° remained for flexion-extension motion estimates 1 month after training.

During a time when we are increasingly emphasizing functional outcomes, such a degree of improvement could be of clinical significance. Our study results support a call for more formalized training of ROM assessment, but clinicians should also be aware of the limitations of visual estimates of cervical spine ROM, and our study results support scrutiny of visual assessment of ROM as a criterion for diagnosing permanent impairment or disability.

Assessment of cervical spine range of motion (ROM) is an integral aspect of the physical examination for cervical conditions,^1-3 surgical outcomes,⁴ and functional impairment.¹ In fact, the emphasis being placed on such functional measures before and after treatments is increasing.^4,5

Cervical spine range of motion is routinely used as an outcome measure in clinical studies.^6-8 Underscoring the importance of defining cervical spine ROM, studies have found it to be a preoperative predictor of outcomes of anterior cervical surgery,⁹ and other studies have suggested it is a determinant of athletes’ return to play.¹⁰

Spinal ROM measurements can be used to determine the degree of disability experienced by a patient with a spinal condition as defined in the Guides to the Evaluation of Permanent Impairment by the American Medical Association (AMA).¹ In the medicolegal realm, ROM measurements made by clinicians can influence the dollar amounts of awards in legal claims, and, according to the AMA guides, the difference in cervical spine ROM between normality and disability or impairment can be as little as 5°.

Although cervical spine ROM is routinely assessed and documented in clinical practice, no universal protocol exists for its evaluation.^11,12 In fact, considerable inter-examiner variation in visual estimates of ROM has been found,^13-16 and significant inaccuracies have been reported.^17,18

Goniometers have been shown to be reliable and highly accurate, with low inter-examiner and intra-examiner variability.^5,19-21 Nevertheless, logistics²² and costs²¹ generally limit their being accepted in routine clinical practice. Among many methods available for assessing ROM, visual estimation is the least reliable or accurate,²³ but it is the quickest and least expensive and is recommended in textbooks that describe the spinal-specific physical examination.²⁴ Despite the superiority of goniometers in measuring ROM, these significant barriers have limited their use in clinical practice. When assessing cervical spine ROM, most clinicians prefer visual estimates over goniometers.

We conducted a study to determine whether training could improve the accuracy of visual estimates. We compared the accuracy of visual estimates of cervical spine ROM with that of a radiographically validated electrogoniometer and then investigated whether accuracy and reliability of visual estimates could be improved with a session of instruction and demonstration. Assessments of accuracy were made immediately after and 1 month after this training session.

Materials and Methods

Assessments Made Before Training

This study was approved by our institution’s human investigation committee and was conducted in accordance with the ethical standards of that committee.

Cervical spine ROM was assessed by 8 examiners (2 attending spine surgeons, 4 orthopedic residents, 2 medical students). They were informed they would be participating in a study evaluating visual estimates of motion but were given no other information prior to the study.

Four healthy volunteer subjects (examiners who rotated through the role) were assessed. No subject reported any ongoing neck or spine discomfort or had had any previous spinal surgery. One at a time, subjects were fitted with a cervical harness electrogoniometer capable of measuring angulation of the cervical spine to the nearest degree (modified electrogoniometer, torsiometer, and display from Biometrics, Gwent, UK; Figures 1A, 1B). This electrogoniometer has been shown to have a mean (SD) error of 2.3° (2.6°) relative to radiographic assessments.⁸

With the electrogoniometer fitted, each subject was instructed to sit upright in a chair with his back to the backrest and his head neutrally positioned. The electrogoniometer was then zeroed, and the subject proceeded with 5 series of flexion-extension, left and right lateral bending, and left and right rotation movements. The subject was instructed to make 1 movement in full motion in each direction and the other 4 movements in less than full motion to yield a variety of excursions for assessment. Each subject was instructed to pause at the apex of each motion. During these pauses, the examiners recorded their visual estimates of movement in each direction while the investigator recorded degrees of motion (displayed by the electrogoniometer) in flexion-extension, lateral bending, and rotation (Figures 2A–2D). The electrogoniometer display was not visible to subjects or examiners.

A total of 840 independent visual estimates of 120 distinct movements were recorded.

Training, and Assessments Made Immediately Thereafter

After the first round of visual estimates, the 8 examiners were verbally instructed in cervical spine ROM assessment and were asked to observe 1 subject, fitted with the electrogoniometer, demonstrating partial and full cervical motions while the investigator announced the electrogoniometric measurements. The motions demonstrated included 15°, 30°, and the extremes of cervical spine ROM in each of 6 directions from neutral.

After this training session, each of the 4 subjects from the first round of assessments was again fitted with the harness electrogoniometer and instructed to repeat the movements in turn while examiners visually estimated cervical spine ROM and independently recorded their estimates. Meanwhile, the investigator recorded the degree of motion during each movement (as measured by the electrogoniometer). Again, a total of 840 independent visual estimates of 120 distinct movements were recorded.

Assessments Made 1 Month After Training

One month after the training session, the examiners and the investigator reconvened to assess the same 4 subjects using a procedure for simultaneous visual estimation and electrogoniometric measurement identical to that used 1 month earlier. No additional training was given. Again, 840 independent visual estimates of 120 distinct movements were recorded.

Data Analysis

The reliabilities of visual estimates were analyzed by calculating the intraclass coefficients (ICCs) using random-effect 1-way analyses of variance. By convention, ICCs of < 0.2, 0.2 to 0.39, 0.4 to 0.59, 0.6 to 0.8, and > 0.8 correspond to poor, fair, moderate, substantial, and perfect reliability, respectively.²⁵

We compared the visual estimates and electrogoniometric measurements made for 3 planes of motion (flexion-extension, lateral bending, axial rotation) before, immediately after, and 1 month after training and drew trend lines generated by linear regression relative to a line of perfect correlation.

Mean errors in examiners’ visual estimates (relative to elec­trogoniometric measurements) made before, immediately after, and 1 month after training were calculated. Paired Student t tests were then used to compare the mean errors before training with the mean errors immediately after and 1 month after training.

All analyses were performed with SPSS for Windows 16.0 (SPSS, Chicago, Illinois).

Results

Inter-examiner reliability of the visual estimates in all planes of motion ranged from 0.51 to 0.79 (suggestive of moderate to substantial reliability). For reference, standard goniometers measuring knee ROM have inter-examiner ICCs of 0.89 to 0.98²⁶ (suggestive of perfect reliability). The ICCs before, immediately after, and 1 month after training were not significantly different. 

As expected, there were significant errors in visual estimates of cervical spine ROM in all planes. Initial errors in visual estimates (relative to electrogoniometric measurements) were 23.9° (flexion-extension), 15.5° (lateral bending), and 19.3° (axial rotation) (Table, Figure 3).

Immediately after training, mean errors in visual estimates decreased to 12.0° (flexion-extension), 11.7° (lateral bending), and 16.4° (axial rotation) (Table, Figure 3). In all 3 planes of cervical motion, these improvements were statistically significant.

One month after training, mean errors in visual estimates were 14.4° (flexion-extension), 13.9° (lateral bending), and 16.2° (axial rotation) (Table, Figure 3). Only the improvement in the estimate of flexion-extension (the direction of the largest error initially) remained statistically significant—a 39.7% decrease in error.

We also considered how errors varied with degree of motion observed. In flexion-extension, the tendency to overestimate at larger degrees of motion was not apparent after training, and 1 month after training we found a tendency to underestimate at smaller degrees of motion (Figure 4A). The tendency to overestimate lateral bending before training did not persist immediately after or 1 month after training (Figure 4B). Estimates of axial rotation correlated well with goniometer measurements before training and were also well correlated immediately after and 1 month after training (Figure 4C).

Discussion

Visual estimation of spinal motion is unreliable and inaccurate, but its widespread use in clinical practice continues. Goniometers are far more accurate and reliable but are seldom used. We investigated whether a training session featuring verbal instruction and demonstration with an electrogoniometer could improve visual estimates and whether potential improvement in visual estimates would remain 1 month after training.

Widely variable ICCs (0.42-0.90) have been reported for visual estimates of cervical spine ROM.^17,18,22 Our findings on the reliability of these estimates are consistent with the literature.

We recorded the greatest initial error in estimates of motion in flexion-extension. Previous studies have also found the greatest error and least reliability in visual estimates in this plane.^14,15,18 Visual estimation may be more difficult in flexion-extension because the shoulders cannot be used as landmarks, whereas they serve as approximate 90° reference points during estimation of lateral bending and axial rotation. Demonstration of 15°, 30° and the extremes of ROM during the training session may have provided alternative reference points during visual estimation after training—decreasing the error to within the range found in other planes of motion.

Initial errors in visual estimates were 23.9° (flexion-extension), 15.5° (lateral bending), and 19.3° (axial rotation). Based on normative cervical spine ROM in a healthy population— 126° ± 12° for flexion-extension, 86° ± 5° for lateral bending, 151° ± 23° for axial rotation²²—the errors we identified are 18.9% of the normal range of flexion-extension, 18.0% of lateral bending, and 12.8% of axial rotation.

Training clearly improved the accuracy of visual estimates of cervical spine ROM. Estimates were statistically improved for all planes immediately after training and remained significantly improved for flexion-extension (the plane of largest error initially) 1 month after training. Before training, mean errors varied across planes. Training normalized mean errors to about 15°, and this effect lasted in flexion-extension, lateral bending, and axial rotation (Figures 4A–4C). Of note, before training these percentage errors increased with increased motion from neutral in the flexion-extension and lateral bending planes. At full ROM, percentage errors in estimates were greater. After training, percentage errors did not increase appreciably with increasing motion.

Readers will naturally reflect on the clinical significance of the motion assessment improvements demonstrated after the training session described in this study. We must be aware that functional assessments are increasingly being emphasized in the clinical arena—with respect to clinical conditions, surgical outcomes, and functional impairments. We highlight a point made earlier: A difference of only 5° can affect impairment ratings in the medicolegal realm.¹ In estimating flexion-extension motion, lasting improvements of almost 10° were demonstrated and maintained 1 month after the training session described in this study.

Nevertheless, mean errors in visual estimation remained at about 15° in all planes of motion, despite our modest improvements. This finding raises the question of whether visually estimated ROM should be pertinent to assessments of impairment and disability. Although visual estimates of ROM may have more utility as a screening test for impairment and disability, fine differences in ROM simply cannot be reliably assessed by visual estimation.

This study has limitations. First, it was conducted at a single institution where the evaluators received most of their training. Their skill in visually estimating cervical spine ROM may not be generalizable to a larger population of spine specialists who are practicing at other institutions and may have different training backgrounds.

Second, only healthy subjects were assessed. Some studies of cervical spine ROM have shown better reliability in symptomatic subjects relative to asymptomatic subjects.^13,14 To attempt to overcome this limitation, we assessed many different excursions of motion that were often not to the extremes of motion.

Third, the “gold standard” we used for motion assessment was an electrogoniometer, which has some inherent error (previously validated mean [SD] error of 2.3° [2.6°] relative to radiographs⁸). Although obtaining radiographs of each movement would have more closely resembled the gold standard, the radiation dose associated with such a study is prohibitive.

Last, the assessors included medical students. The medical students’ estimates, however, tended to be more accurate than the residents’ or attending surgeons’ (though the difference was not statistically significant). This tendency may reflect the medical students’ closer attention to detail.  Clearly, including medical students in the study did not negatively affect the accuracy of the estimates or the validity of our findings.

Conclusion

Despite its limitations, visual assessment of cervical spine motion remains the gold standard in clinical practice and is routinely recorded and reported. Mean errors ranged from 15.5° to 23.9°, depending on plane of motion being assessed, but these improved after a training session.

Visual estimates of motion in flexion-extension were most improved by training, as the initial errors in this plane were the largest. Statistically significant improvement of about 10° remained for flexion-extension motion estimates 1 month after training.

During a time when we are increasingly emphasizing functional outcomes, such a degree of improvement could be of clinical significance. Our study results support a call for more formalized training of ROM assessment, but clinicians should also be aware of the limitations of visual estimates of cervical spine ROM, and our study results support scrutiny of visual assessment of ROM as a criterion for diagnosing permanent impairment or disability.

Assessment of cervical spine range of motion (ROM) is an integral aspect of the physical examination for cervical conditions,^1-3 surgical outcomes,⁴ and functional impairment.¹ In fact, the emphasis being placed on such functional measures before and after treatments is increasing.^4,5

Cervical spine range of motion is routinely used as an outcome measure in clinical studies.^6-8 Underscoring the importance of defining cervical spine ROM, studies have found it to be a preoperative predictor of outcomes of anterior cervical surgery,⁹ and other studies have suggested it is a determinant of athletes’ return to play.¹⁰

Spinal ROM measurements can be used to determine the degree of disability experienced by a patient with a spinal condition as defined in the Guides to the Evaluation of Permanent Impairment by the American Medical Association (AMA).¹ In the medicolegal realm, ROM measurements made by clinicians can influence the dollar amounts of awards in legal claims, and, according to the AMA guides, the difference in cervical spine ROM between normality and disability or impairment can be as little as 5°.

Although cervical spine ROM is routinely assessed and documented in clinical practice, no universal protocol exists for its evaluation.^11,12 In fact, considerable inter-examiner variation in visual estimates of ROM has been found,^13-16 and significant inaccuracies have been reported.^17,18

Goniometers have been shown to be reliable and highly accurate, with low inter-examiner and intra-examiner variability.^5,19-21 Nevertheless, logistics²² and costs²¹ generally limit their being accepted in routine clinical practice. Among many methods available for assessing ROM, visual estimation is the least reliable or accurate,²³ but it is the quickest and least expensive and is recommended in textbooks that describe the spinal-specific physical examination.²⁴ Despite the superiority of goniometers in measuring ROM, these significant barriers have limited their use in clinical practice. When assessing cervical spine ROM, most clinicians prefer visual estimates over goniometers.

We conducted a study to determine whether training could improve the accuracy of visual estimates. We compared the accuracy of visual estimates of cervical spine ROM with that of a radiographically validated electrogoniometer and then investigated whether accuracy and reliability of visual estimates could be improved with a session of instruction and demonstration. Assessments of accuracy were made immediately after and 1 month after this training session.

Materials and Methods

Assessments Made Before Training

This study was approved by our institution’s human investigation committee and was conducted in accordance with the ethical standards of that committee.

Cervical spine ROM was assessed by 8 examiners (2 attending spine surgeons, 4 orthopedic residents, 2 medical students). They were informed they would be participating in a study evaluating visual estimates of motion but were given no other information prior to the study.

Four healthy volunteer subjects (examiners who rotated through the role) were assessed. No subject reported any ongoing neck or spine discomfort or had had any previous spinal surgery. One at a time, subjects were fitted with a cervical harness electrogoniometer capable of measuring angulation of the cervical spine to the nearest degree (modified electrogoniometer, torsiometer, and display from Biometrics, Gwent, UK; Figures 1A, 1B). This electrogoniometer has been shown to have a mean (SD) error of 2.3° (2.6°) relative to radiographic assessments.⁸

With the electrogoniometer fitted, each subject was instructed to sit upright in a chair with his back to the backrest and his head neutrally positioned. The electrogoniometer was then zeroed, and the subject proceeded with 5 series of flexion-extension, left and right lateral bending, and left and right rotation movements. The subject was instructed to make 1 movement in full motion in each direction and the other 4 movements in less than full motion to yield a variety of excursions for assessment. Each subject was instructed to pause at the apex of each motion. During these pauses, the examiners recorded their visual estimates of movement in each direction while the investigator recorded degrees of motion (displayed by the electrogoniometer) in flexion-extension, lateral bending, and rotation (Figures 2A–2D). The electrogoniometer display was not visible to subjects or examiners.

A total of 840 independent visual estimates of 120 distinct movements were recorded.

Training, and Assessments Made Immediately Thereafter

After the first round of visual estimates, the 8 examiners were verbally instructed in cervical spine ROM assessment and were asked to observe 1 subject, fitted with the electrogoniometer, demonstrating partial and full cervical motions while the investigator announced the electrogoniometric measurements. The motions demonstrated included 15°, 30°, and the extremes of cervical spine ROM in each of 6 directions from neutral.

After this training session, each of the 4 subjects from the first round of assessments was again fitted with the harness electrogoniometer and instructed to repeat the movements in turn while examiners visually estimated cervical spine ROM and independently recorded their estimates. Meanwhile, the investigator recorded the degree of motion during each movement (as measured by the electrogoniometer). Again, a total of 840 independent visual estimates of 120 distinct movements were recorded.

Assessments Made 1 Month After Training

One month after the training session, the examiners and the investigator reconvened to assess the same 4 subjects using a procedure for simultaneous visual estimation and electrogoniometric measurement identical to that used 1 month earlier. No additional training was given. Again, 840 independent visual estimates of 120 distinct movements were recorded.

Data Analysis

The reliabilities of visual estimates were analyzed by calculating the intraclass coefficients (ICCs) using random-effect 1-way analyses of variance. By convention, ICCs of < 0.2, 0.2 to 0.39, 0.4 to 0.59, 0.6 to 0.8, and > 0.8 correspond to poor, fair, moderate, substantial, and perfect reliability, respectively.²⁵

We compared the visual estimates and electrogoniometric measurements made for 3 planes of motion (flexion-extension, lateral bending, axial rotation) before, immediately after, and 1 month after training and drew trend lines generated by linear regression relative to a line of perfect correlation.

Mean errors in examiners’ visual estimates (relative to elec­trogoniometric measurements) made before, immediately after, and 1 month after training were calculated. Paired Student t tests were then used to compare the mean errors before training with the mean errors immediately after and 1 month after training.

All analyses were performed with SPSS for Windows 16.0 (SPSS, Chicago, Illinois).

Results

Inter-examiner reliability of the visual estimates in all planes of motion ranged from 0.51 to 0.79 (suggestive of moderate to substantial reliability). For reference, standard goniometers measuring knee ROM have inter-examiner ICCs of 0.89 to 0.98²⁶ (suggestive of perfect reliability). The ICCs before, immediately after, and 1 month after training were not significantly different. 

As expected, there were significant errors in visual estimates of cervical spine ROM in all planes. Initial errors in visual estimates (relative to electrogoniometric measurements) were 23.9° (flexion-extension), 15.5° (lateral bending), and 19.3° (axial rotation) (Table, Figure 3).

Immediately after training, mean errors in visual estimates decreased to 12.0° (flexion-extension), 11.7° (lateral bending), and 16.4° (axial rotation) (Table, Figure 3). In all 3 planes of cervical motion, these improvements were statistically significant.

One month after training, mean errors in visual estimates were 14.4° (flexion-extension), 13.9° (lateral bending), and 16.2° (axial rotation) (Table, Figure 3). Only the improvement in the estimate of flexion-extension (the direction of the largest error initially) remained statistically significant—a 39.7% decrease in error.

We also considered how errors varied with degree of motion observed. In flexion-extension, the tendency to overestimate at larger degrees of motion was not apparent after training, and 1 month after training we found a tendency to underestimate at smaller degrees of motion (Figure 4A). The tendency to overestimate lateral bending before training did not persist immediately after or 1 month after training (Figure 4B). Estimates of axial rotation correlated well with goniometer measurements before training and were also well correlated immediately after and 1 month after training (Figure 4C).

Discussion

Visual estimation of spinal motion is unreliable and inaccurate, but its widespread use in clinical practice continues. Goniometers are far more accurate and reliable but are seldom used. We investigated whether a training session featuring verbal instruction and demonstration with an electrogoniometer could improve visual estimates and whether potential improvement in visual estimates would remain 1 month after training.

Widely variable ICCs (0.42-0.90) have been reported for visual estimates of cervical spine ROM.^17,18,22 Our findings on the reliability of these estimates are consistent with the literature.

We recorded the greatest initial error in estimates of motion in flexion-extension. Previous studies have also found the greatest error and least reliability in visual estimates in this plane.^14,15,18 Visual estimation may be more difficult in flexion-extension because the shoulders cannot be used as landmarks, whereas they serve as approximate 90° reference points during estimation of lateral bending and axial rotation. Demonstration of 15°, 30° and the extremes of ROM during the training session may have provided alternative reference points during visual estimation after training—decreasing the error to within the range found in other planes of motion.

Initial errors in visual estimates were 23.9° (flexion-extension), 15.5° (lateral bending), and 19.3° (axial rotation). Based on normative cervical spine ROM in a healthy population— 126° ± 12° for flexion-extension, 86° ± 5° for lateral bending, 151° ± 23° for axial rotation²²—the errors we identified are 18.9% of the normal range of flexion-extension, 18.0% of lateral bending, and 12.8% of axial rotation.

Training clearly improved the accuracy of visual estimates of cervical spine ROM. Estimates were statistically improved for all planes immediately after training and remained significantly improved for flexion-extension (the plane of largest error initially) 1 month after training. Before training, mean errors varied across planes. Training normalized mean errors to about 15°, and this effect lasted in flexion-extension, lateral bending, and axial rotation (Figures 4A–4C). Of note, before training these percentage errors increased with increased motion from neutral in the flexion-extension and lateral bending planes. At full ROM, percentage errors in estimates were greater. After training, percentage errors did not increase appreciably with increasing motion.

Readers will naturally reflect on the clinical significance of the motion assessment improvements demonstrated after the training session described in this study. We must be aware that functional assessments are increasingly being emphasized in the clinical arena—with respect to clinical conditions, surgical outcomes, and functional impairments. We highlight a point made earlier: A difference of only 5° can affect impairment ratings in the medicolegal realm.¹ In estimating flexion-extension motion, lasting improvements of almost 10° were demonstrated and maintained 1 month after the training session described in this study.

Nevertheless, mean errors in visual estimation remained at about 15° in all planes of motion, despite our modest improvements. This finding raises the question of whether visually estimated ROM should be pertinent to assessments of impairment and disability. Although visual estimates of ROM may have more utility as a screening test for impairment and disability, fine differences in ROM simply cannot be reliably assessed by visual estimation.

This study has limitations. First, it was conducted at a single institution where the evaluators received most of their training. Their skill in visually estimating cervical spine ROM may not be generalizable to a larger population of spine specialists who are practicing at other institutions and may have different training backgrounds.

Second, only healthy subjects were assessed. Some studies of cervical spine ROM have shown better reliability in symptomatic subjects relative to asymptomatic subjects.^13,14 To attempt to overcome this limitation, we assessed many different excursions of motion that were often not to the extremes of motion.

Third, the “gold standard” we used for motion assessment was an electrogoniometer, which has some inherent error (previously validated mean [SD] error of 2.3° [2.6°] relative to radiographs⁸). Although obtaining radiographs of each movement would have more closely resembled the gold standard, the radiation dose associated with such a study is prohibitive.

Last, the assessors included medical students. The medical students’ estimates, however, tended to be more accurate than the residents’ or attending surgeons’ (though the difference was not statistically significant). This tendency may reflect the medical students’ closer attention to detail.  Clearly, including medical students in the study did not negatively affect the accuracy of the estimates or the validity of our findings.

Conclusion

Despite its limitations, visual assessment of cervical spine motion remains the gold standard in clinical practice and is routinely recorded and reported. Mean errors ranged from 15.5° to 23.9°, depending on plane of motion being assessed, but these improved after a training session.

Visual estimates of motion in flexion-extension were most improved by training, as the initial errors in this plane were the largest. Statistically significant improvement of about 10° remained for flexion-extension motion estimates 1 month after training.

During a time when we are increasingly emphasizing functional outcomes, such a degree of improvement could be of clinical significance. Our study results support a call for more formalized training of ROM assessment, but clinicians should also be aware of the limitations of visual estimates of cervical spine ROM, and our study results support scrutiny of visual assessment of ROM as a criterion for diagnosing permanent impairment or disability.