Introduction
An increasingly large and frail aging population are at risk for fragility fractures of the hip, which is associated with a high morbidity and 1-year mortality [
1]. There are a number of scores predicting the postoperative mortality that rely on accurate definition of comorbidity and formulaic calculations. These scoring systems are invented to identify patients at risk for early mortality and may provide a possibility to optimize patients pre- and post-operatively. The most commonly used are POSSUM [
2,
3], the Charlson Comorbidity Index [
4] and Nottingham hip fracture score [
5]. Previously, the Sernbo score has been found to be a useful predictor of 1-year mortality in a small cohort of patients with femoral neck fracture [
6]. The Sernbo score is a simple four-component score (age, habitat, walking aids and mental state), initially developed as a tool for decision making for treatment with either a total- or hemiarthroplasty for femoral neck fractures [
7,
8]. The aim of this study was to validate Sernbo score on a national level. The Sernbo score can easily be calculated using information obtained during routine orthopaedic patient assessment.
Source of data and terminology
In the Swedish Hip Fracture Register (SHFR), patients with hip fractures treated in Sweden have been registered since 1988. In 2016, the completeness of SHFR was 84% [
13]. Baseline data on all patients include age, sex, ASA category (1–2 or 3–5), pre-fracture walking ability, habitat, cognitive status and type of fracture. We classified walking ability as either with or without any walking aid. Habitat, as either living independent or at a sheltered home. In the current study, cognitive status was classified as lucid or cognitive impairment and the type of fracture as femoral neck or trochanteric/subtrochanteric.
Date of death was obtained through record linkage with the National Death Register. In Sweden, The National Register of Causes of Death is cross-checked continuously with the National Death Register. All data were linked to the patients using the unique personal identification number assigned to all Swedish citizens. The Sernbo score was graded according to the total number of points and divided into low risk (17–20 points), intermediate risk (14 points) and high risk (8–11 points), as described previously Mellner et al. [
6] (Table
1).
Table 1
The Sernbo score was graded according to the total number of points and divided into low risk (17–20 points), intermediate risk (14 points) and high risk (8–11 points)
Age |
< 80 years | 5 | 4.6% |
≥ 80 years | 2 | 22.8% |
Social situation |
Own home | 5 | 13.6% |
Sheltered home | 2 | 13.8% |
Walking aids |
None or one stick | 5 | 6.8% |
Two sticks or walking frame | 2 | 20.5% |
Mental status |
Alert | 5 | 12.2% |
Slight confusion | 2 | 15.1% |
Statistical analysis
Descriptive data were presented with means and standard deviation (SD), range and percentages. A univariate and multivariate logistic regression analysis was performed to predict mortality. Associations were quantified using odds ratio (OR). ROC (Receiver-operating characteristic) curves were calculated to validate the mortality thresholds given by the Sernbo score and to determine their sensitivity and specificity. C-statistics (area under the curve, AUC) with 95% confidence interval (95% CI) was used to assess discrimination of the outcomes. The Hosmer–Lemeshow test were used to assess calibration of goodness-of-fit. Nagelkerke’s R2 were used for testing the predictive value. Significance level was set at 0.05 and all tests were two-tailed. Kaplan–Meier survival curves were used to compare the different groups with log-rank test. The statistical analysis was performed using SPSS Statistics software 24.0 for Mac (SPSS Inc., Chicago, IL) and using the MedCalc (Medcalc Software, Ostend, Belgium) for the ROC analysis.
Ethics
The study was conducted in accordance with the ethical principles of the Helsinki Declaration and was approved by Regional Ethical Review Board at the Karolinska Institute (DNR: 2017/1088-31).
Discussion
In this nationwide, retrospective cohort study based on the registry data, the Sernbo accurately identified patients with high risk of death during the first postoperative year. This easy, clinically applicable scoring system could be used more systematically to tailor pre- and post-operative care and might guide in the choice of surgical treatment (i.e. hemi- or total-hip arthroplasty) for patients with an acute hip fracture.
Our results in this large cohort of hip fracture patients corroborates the findings of Dawe et al., as well as a previous study from our department, both showing acceptable predictive abilities with an AUC of 0.69 and 0.79, respectively [
6,
15]. The results suggested that the mortality in this group of patients is dependent on several additional unknown factors not included in these models. However, as the complexity of the scoring system increases, it might be at the expense of its clinical applicability. In this registry setting resulted in lower discriminative power compared to the more selected cohort at a single centre in Sweden [
6]. Moreover, the latter study focused on a subpopulation of patients with displaced femoral neck fractures fit for hip arthroplasty surgery [
6]. The Sernbo score was originally developed as a tool for decision making for treatment with either a THA or HA for displaced femoral neck fractures [
7]. This might explain lower discriminative power when using Sernbo score on the whole hip fracture population. However, concurrent with our results, Söderqvist et al. [
15] found no difference in mortality at 4 months and 24 months between the different types of hip fractures [
15].
Karres et al., showed in a review of six scoring models for predicting 30-day mortality that none of the included models yielded neither good nor excellent discriminative power [
16]. Trevisan et al. compared the survival rates between 2000 and 2015 and found that Charlson comorbidity index had the best predictive ability for mortality at 30 days and 1 year [
17]. The Sernbo score seemed to be at par with most other predictive instruments in the literature [
18].
The Charlson comorbidity index, Orthopaedic-POSSUM, Estimation of Physiological Ability and Surgical Stress and the Nottingham Hip Fracture Score, have all been evaluated as predictors of mortality in patients with hip fractures. In contrast to these scoring systems, the Sernbo score is far less complex which increases its daily clinical applicability [
17]. In a recent publication, Jonsson et al., investigated the discrimination and accuracy of the POSSUM score, Portsmouth-POSSUM (P-POSSUM) score and the Nottingham Hip Fracture Score for prediction of mortality [
19]. The authors found a moderate calibration and poor discriminative ability and concluded that mortality and morbidity in hip fracture patients are largely dependent on unknown factors that are not included in these scores [
19]. The Sernbo score seems to perform on par with the above-mentioned scoring systems for mortality.
In the present study, we did not have data on co-morbidities, laboratory tests, such as haemoglobin or albumin on which other scoring system are based on i.e. O-POSSUM and NHFC [
16]. In a publication by Karres et al. [
16], presented an AUC of 0.78 in a fairly complex predicting model for 30-day mortality in hip fracture patients which included above 85 years of age, in-hospital fracture, signs of malnutrition, previous myocardial infarction, congestive heart failure, current pneumonia, chronic renal failure, malignancy and elevated serum urea.
Several other factors affecting the postoperative mortality in hip fracture patients have been reported, including the components in Sernbo score and fracture type, high ASA grade, high Charlson comorbidity score on admission, an abnormal ECG, increased C-reactive protein level, low haemoglobin level and hypoalbuminaemia [
20‐
22]. Timing until surgery is associated with an increased mortality [
23,
24]. A shorter length of hospital stay after hip fracture has been proposed to be associated with an increased 30-day mortality [
25]. Previous studies have described a higher short-term mortality in males with a hip fracture [
15,
27].
The major strength of the present study is the analysis of data from a large nationwide cohort of patients with hip fracture with a high degree of external validity by including patients with all types of hip fractures [
28]. The unique personal identify number minimizes lost to follow-up and enables linkage to accurate mortality data at the National Death Register.
This registry-based, retrospective study design has inherited limitations. A limitation to the study is that it is not reproducible. We have not performed any test for reproducibility, such as interobserver and intraobserver reliability of each item. The exclusion of 30% of the registered patients due to missing data, mainly concerning cognitive function, which is an optional question in the registry. The assessment of calibration indicated a lower goodness of fit in our statistical models, we had a large sample size which might overestimate this issue. Poor calibration does not necessarily suggest a clinically relevant differences between prediction by the model and the observed outcome, we observed a fairly small, but still significant difference measured with the Hosmer–Lemeshow test [
29,
30].