Background
Elderly patients are at high risk for postoperative complications and increased mortality after hip fracture (HF) surgery due to frailty and co-morbidities. According to available studies, one-year mortality after HF surgery ranges from 12 to 37% [
1]. The prediction of postoperative outcome could help with clinical decision making. In previous years, several postoperative mortality scores have been developed for this specific purpose in the setting of elective surgeries. However, these scores often misinterpret orthogeriatric patients’ outcomes and risk of death [
2‐
4].
The PICO model was used to clearly define a clinical question. The population is represented by elderly patients with hip fractures. The intervention aims to identify prognostic factors which improve the accuracy of postoperative mortality prediction in the specific sub-population. Four potential prognostic factors were examined, namely: CRP D
0, NLR D
0, age and gender. C-reactive protein (CRP) and the neutrophil-to-lymphocyte ratio (NLR) reflect the inflammatory status of patients during the peri-operative period. Furthermore, the NLR at admission (D
0) and at the fifth day (D
5) have been shown to be associated with postoperative complications [
3,
5‐
7]. We compared predicted mortality to observed mortality. The outcome of the study is to improve mortality prediction by analysing if the addition of these variables to the POSPOM score, one of the best validated prognostic tools in the peri-operative period, may increase its discriminative capacity.
Methods
Study design
Retrospective analysis of a single-centre cohort.
Settings
Included patients were admitted to the university hospital of St. Luc in Belgium (Cliniques universitaires Saint-Luc) between 2010 and 2016. Study cases were directly recruited after diagnosis of hip fracture in the emergency department or surgical ward. Date of admission, date of discharge, date of intra-hospital death, age, sex, co-morbidities, NLR D0, CRP D0 values of patients were recorded. Follow-up was terminated upon hospital discharge or intra-hospital death. Data registration and management were performed in agreement with Belgian law and the Helsinki declaration.
Participants
After receiving ethical committee approval (Commission d’Ethique Biomédicale Hospitalo-Facultaire - CEBHF) of the Catholic University of Louvain (Chairperson: Prof J-M Maloteaux, n°2010/23DEC/406), the authors were granted a waiver for written informed consent due to the retrospective nature of the study and analysis of anonymised data. The database included a total of 782 patients with a diagnosis of HF. Patients lacking personal data concerning co-morbidities were excluded from the study. All patients included in the study were treated following the same early surgical care protocol (i.e. 81% operated within the first 24 h). This methodology consisted of obtaining medical clearance from the emergency department as soon after the diagnosis as possible. Patients were then wait-listed and commonly operated the same day. In cases of treatment with anti-vitamin K medication, coagulation was restored with vitamin K. Patients under anti-platelets treatment were operated on without additional delay, in agreement with the surgeon, under general anaesthesia. When possible, regional anaesthesia/analgesia was proposed and performed, including a fascia iliaca block (single shot). Particular attention was paid to haemodynamic control, with the use of invasive blood pressure monitoring and dynamic variables when indicated and applicable. Postoperative follow-up was performed by an inter-disciplinary medical team consisting of orthopaedic surgeons, anaesthetists, geriatricians, an internal medicine specialist dedicated to peri-operative medicine, and physiotherapists.
Variables
In previous studies, advanced age and male gender were identified as risk factors in patients with HF [
3,
5]. Thus, age and gender were chosen as variables with potential discriminative capacity.
Age and gender were registered during pre-operative evaluation. Data on NLR D0 and CRP D0 were taken from the first blood sample obtained from the patient at admission and before surgery. In our clinical practice, blood testing is only realised in presence of an anamnestic or clinical problem, in order to not delay surgery. All blood analyses were performed on venous blood samples and were processed in a blood analyser (Sysmex; TAO Medical Electronics, Kobe, Japan) for full blood count and differential count of leukocytes. The NLR value was obtained by calculating the ratio between registered neutrophils and lymphocytes counts. The CRP value was determined based on a serum sample by a turbidimetry process (UniCel® DxC 800; Beckman Coulter, Pasadena, California, USA) and is expressed in mg l− 1.
The POSPOM score of each patient was calculated as the sum of the points assigned to each item (age, co-morbidities and type of surgery). With regards to the first variable (age), older patients received higher points. The second variable (co-morbidities) consisted of the total number of points assigned to each of the 17 validated co-morbidities. The third variable (type of surgery) was identical for all patients (“orthopaedic trauma”) and therefore each patient received 14 points. Depending on the total number of points, a percentage predicted risk of in-hospital mortality was assigned to each patient [
8].
Data collection
Data collection (Date of admission, date of discharge, date of intra-hospital death, age, sex, co-morbidities, NLR D0, CRP D0 values) was performed using systematic, standardised and computerised medical charts issued by the institutional software (Medical Explorer v9, Saint-Luc university Hospital, Brussels, 2009).
Control of potential biases was performed using a prospective listing and a standardisation of the data collection process. The increased weight of the variable “age” was intentional.
The population was divided into two groups: patients discharged from the hospital and patients who died in hospital. Quantitative variables such as the NLR, CRP and age were analysed in the descriptive analysis.
Statistical methods
Pearson’s correlation coefficient was computed to identify potential linear association between log(NLR) and log(CRP).
In order to determine the performance of each individual variable (original POSPOM setting, age, gender, NLR and CRP) in the prediction of postoperative mortality, we computed the AUC with DeLong confidence intervals. A logistic regression model was then used to combine POSPOM score with the four other variables to determine whether the addition of one of these variables could improve the predictive value of the POSPOM score.
All analyses were performed using R 3.3.2 (R Foundation for Statistical Computing, 2016, Vienna, Austria) and the ggplot2 and pROC packages.
Discussion
The age, gender, NLR D0 and CRP D0 did not show discriminative capacity in predicting in-hospital mortality after HF. Furthermore, the addition of these variables to the POSPOM did not improve its performance.
Forget and colleagues previously identified age as a risk factor in elderly patients after surgery for HF [
3]. Although age has already been integrated in the original POSPOM, we tested the variable independently in order to give it a higher importance in our scoring system.
The NLR as an inflammatory marker has proven its association with complications and long-term outcome in the setting of gastro-intestinal pathologies treated medically or surgically [
6,
9‐
11]. The NLR has also been associated with complications after major cardiac events or in a variety of cancers [
12,
13]. Forget and colleagues developed a four-item score which included the NLR D
5 after surgery, to predict one-year mortality after surgery for HF [
3]. Fisher and colleagues concluded in a recent paper that a high NLR at admission is an independent indicator of fracture risk in orthogeriatric patients and a significant risk factor and moderate predictor for intra-hospital mortality [
7].
In our study, all CRP and NLR values taken at admission were included for analysis, simulating real life conditions. The score initially proposed by Forget and colleagues uses the NLR from a blood sample on D
5, reflecting an inflammatory state [
3]. The disadvantage of such a score integrating the NLR at D
5 is the unavailability of the result on admission.
The calculated POSPOM score of our cohort predicts a mortality of 13.6%. This was higher than the 4.5% observed in our cohort. This difference may be explained by a multitude of causes.
First, our patients were followed-up using a multidisciplinary approach, the “co-care”-concept. This approach has been shown to reduce mortality compared with standard care [
14]. In a meta-analysis and systematic review, Grigoryan and colleagues found that orthogeriatric collaboration was associated with a significant reduction in intra-hospital and long-term mortality, with a relative risk (RR) of 0.60 [95% CI = 0.43–0.84] and 0.83 [95% CI = 0.74–0.94] respectively. Hospital length of stay was reduced in the co-care model, with a standardised mean difference (SMD) of − 0.61 [95% CI = − 0.95, − 0.28] [
15].
Second, in the validation paper published by Le Manach and colleagues, the percentage of major orthopaedic surgery (such as HF) was only 1.76% of total surgery [
8]. Minor surgery was the most represented type of orthopaedic surgery, comprising 19.69%. Consequently, our study population of elderly HF patients was likely under-represented. Third, the optimal timing of surgery for HF in elderly patients still remains unclear [
16‐
18]. Time to surgery was short in our study (81% of patients were operated within 24 h of admission). As detailed data of the POSPOM population was unavailable, we were unable to determine the significance of this variable.
Recently, several articles have been published showing a lack of accuracy of existing scores. For example, Boddaert and colleagues compared pre-operative surgical scores (ASA classification, POSPOM, Nottingham Hip Fracture score) and geriatric scores (Cumulative Illness rating Scale, Charlson comorbidity index) on a dataset of patients hospitalised in a peri-operative geriatric ward: They concluded no superiority in discriminative capacity of specific or geriatric scores in terms of short and long-term postoperative mortality prediction [
2]. This result can be explained by the hypothesis that all of the scores are not designed to detect diminished physiological capacity in this frail population and suggest that more attention should be paid to frailty assessment rather than pre-operative characteristics, even in emergency conditions. A rapid, multidisciplinary clinical action plan in a shared ward for this patient sub-set, particularly designed to maintaining autonomy and reduce fall risk, could decrease postoperative mortality.
The intended heterogeneity in this cohort reflects a typical geriatric population and common clinical practice but can be considered as a potential source of bias. Furthermore, the relatively small number of events (32 deaths of 782 patients) limited the power of this study. With regards to the biological markers, the unavailability of the parameters in many patients limits the interpretation of these analyses. Specifically, no exclusion criteria were applied to obtained NLR values despite the fact that patients taking steroid therapy and smokers can show higher neutrophil counts. Patients with malignancies were also not excluded, thereby improving the generalisability of these results, whilst remaining a potential confounding factor. Furthermore, an ongoing infection is often a cause of confusion and fall in geriatric patients, resulting in high CRP D
0 and NLR D
0 values. Also, a malnourished state is frequently observed in elderly patients and is typically associated with lymphopaenia. Finally, the age-related impairment of the immune system may play a role in anomalies in this population [
19].