Data source
Inpatient data were extracted from the Japanese Diagnosis Procedure Combination database, a national database containing administrative claims and discharge data9. All academic hospitals are obliged to participate in the database, and more than 1000 community hospitals voluntarily contribute to the database. Overall, the database provides data for approximately 50% of all acute-care inpatients in Japan. The database contains the following information: encrypted unique identifiers; age and sex; body weight and height; admission and discharge dates; diagnoses coded according to the International Classification of Diseases (ICD), 10th revision; surgical and nonsurgical procedures coded according to Japanese original codes (K codes); drugs prescribed; and discharge status. A previous study showed that the validity of diagnoses and procedure records in the database was high (sensitivity and specificity of primary diagnoses: 78.9% and 93.2%, respectively)10. The database clearly differentiates between comorbidities that were already present at admission and complications that occurred after admission, and many studies using the database have been reported elsewhere4,11,12.
This study was approved by the Institutional Review Board of The University of Tokyo [approval number: 3501-(3) (December 25th, 2017)]. The requirement for informed consent was waived by the Ethics committee of The University of Tokyo because of the anonymous nature of the data. All study were performed in accordance with relevant guidelines and regulations.
Patient selection
From July 2010 to March 2017, we screened all patients who were admitted with C2 fracture (ICD-10 code: S12.1) and further identified odontoid fracture using Japanese disease codes. The inclusion criteria were age of ≥ 65 years and admission for treatment of odontoid fracture by at least one of three procedures (halo-vest immobilization (K1444), ASF(K142-1), or PSF (K142-2)) during hospitalization. We excluded patients with multiple fractures (any fractures other than odontoid fractures), with severe consciousness disturbance at admission, who underwent combined surgery (both ASF and PSF), or who died within 2 days of admission. The patients who were treated with halo-vest before or after ASF or PSF were included in the surgery group.
Covariates and outcomes
We compared the three procedures (halo-vest immobilization, ASF, and PSF) using the following covariates at admission: age; sex; body mass index (BMI) (kg/m2); smoking status; academic hospitals; emergency admission; ambulance use; primarily admitted to intensive care unit; oxygenation, hemodialysis, or renal catheter use on admission; pre-existing comorbidities such as diabetes mellitus (E10–E14), hypertension (I10–I15), or chronic lung disease (J40–J47); history of cerebrovascular disease (I60–I69), cardiac disease (I20–I25, I30–I52), hepatic cirrhosis (K74), or dementia (F00–F03); Japan Coma Scale score on admission, which is correlated with the Glasgow Coma Scale score13; Charlson comorbidity index (CCI)14; and Barthel index15. Use of navigation (K9391) was identified among the surgical groups. We categorized eligible patients into two age groups: 65 to 79 years and ≥ 80 years. BMI was categorized into underweight (< 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25.0–29.9 kg/m2), obesity (≥ 30.0 kg/m2), and missing according to the World Health Organization definition. Smoking status was categorized into nonsmoking, smoking, and missing.
The primary endpoint was overall in-hospital mortality. The secondary endpoints were at least one complication after admission, post-treatment length of stay (PLOS), and total hospitalization cost in US dollars (USD). We identified complications after admission from the diagnoses recorded after admission using the following ICD-10 codes and defined at least one complication as at least one of the following complications during hospitalization: sepsis (A40–A41), pulmonary embolism (I26), respiratory complications [pneumonia (J12–J18, J69), respiratory failure (J96), respiratory disorders (J95)], acute coronary syndrome (I21–I24), heart failure (I50), stroke (I60–I64), urinary tract infection (N30, N34, N36–N37, N39), and renal failure (N17–N19). PLOS was defined as the length of stay from the day treated with halo-vest, ASF, or PSF to discharge (or death). Total hospitalization cost includes item-by-item prices for surgical, pharmaceutical, laboratory, nursing care, and other inpatient services, that are offered by universal health care in Japan. The currency exchange rate was set at 100 Japanese yen per USD to account for the average rate of the study period.
Statistical analysis
We used a propensity score-based method to account for differences in observed factors that might affect either the treatment assignment or outcome16. The propensity score was defined as the probability of a patient undergoing halo-vest immobilization, ASF, or PSF based on the patient’s baseline covariates. Covariate selection was prespecified by using both potential confounding factors and variables that can serve as proxies for unknown or unmeasured confounding variables. The propensity score was estimated using a multinomial logistic model with the procedure received as the dependent variable and the following baseline factors as independent variables17: age; sex; BMI category; smoking status; ambulance use; emergency admission; admission to intensive care unit before treatment; oxygenation therapy before treatment; use of urinary catheter; pre-existence of diabetes mellitus, hypertension, or chronic lung disease; history of cerebrovascular disease, cardiac disease, hepatic disease, dementia, or osteoporosis; at least one comorbidity; Japan Coma Scale score category; Barthel index; and CCI category on admission.
To balance the patients’ baseline characteristics among the three procedures, a matching weight approach was applied18. Matching weights is recommended for comparing outcomes across multiple treatment groups when the covariates’ overlaps are relatively limited, outcomes are rare, or exposure distributions are unequal19. Each patient was weighted by the inverse probability with the lower propensity score of the three procedures as the numerator19. The patients would receive each of the treatments among halo-vest immobilization, ASF, or PSF, allowing average treatment effects to be estimated. Baseline covariate balance was checked after weighting, using a p value of > 0.05 calculated by analysis of variance or the chi-squared test among the three treatments.
We compared the following outcomes among the three groups (halo-vest immobilization, ASF, and PSF) using analysis of variance and the chi-square test in the matching weighted cohort: overall in-hospital death, complications after admission, PLOS, total hospitalization costs, and Barthel index at discharge. We further conducted logistic regression analyses to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) for overall in-hospital death and at least one complication after admission. We also conducted a linear regression analysis to estimate the regression coefficient and 95% CI for the PLOS. Moreover, we conducted a multivariable logistic regression analysis with adjustment for age, sex, BMI category, smoking status, and CCI category in the non-weighted and weighted cohorts to identify risk factors for in-hospital death. The following sensitivity analyses were undertaken to assess the robustness of the results. We combined the ASF and PSF groups as the surgery group and compared halo-vest immobilization with the surgery group using propensity score-matching analysis and matching weight analysis to balance the baseline variables.
Statistical analyses were performed using Stata/MP version 15 software (StataCorp, College Station, TX, USA). A two-tailed significance level of p < 0.05 and 95% CIs were used in the analyses.
Ethical approval
The study design was approved by the Institutional Review Board of The University of Tokyo.
Consent to participate
The requirement for informed consent was waived because of the anonymous nature of the data.