Methods and analysis
Study design
This is a multicentre prospective observational study of paediatric patients who visit the ED with a traumatic elbow injury. The data will be collected in four different hospitals; one academic hospital, two large teaching hospitals and one general hospital.
The first part of this research is the development of the decision rule with use of clinical parameters obtained from baseline patient characteristics, patient interview and physical examination. An expert panel of orthopaedic elbow surgeons, paediatric orthopaedic surgeons and trauma surgeons will determine which clinical parameters provide a possible predictive value for elbow fractures. We will collect all clinical parameters (originating from the expert panel) via the patient history and physical examination from all patients. All patients will get radiographs of the affected elbows and will be provided emergency care according to local hospital protocols. These patient data will be processed in a multivariable logistic regression analysis to determine the clinical parameters that predict the presence or absence of an elbow fracture. Only the clinical parameters which significantly predict a fracture, within this prediction model, will be used to formulate the new decision rule.
In the second part of this research, the newly developed decision rule will undergo internal validation using separate data identically gathered in a prospective fashion. At the same time, we will determine the primary outcome measurements: the potential absolute reduction in the number of X-ray examinations, a calculation detailing how much costs have been saved by taking more selective X-rays and a calculation on time saved during an ED visit.
The final study will include a completed version of the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) checklist. With this checklist, we hope to improve the transparency of our research by identifying the important factors in the prognostic prediction model, according to the TRIPOD statement.
Study population
The study population is defined as all consecutive children/adolescents aged 2–17 years who visit the ED of one of the participating hospitals with pain following elbow trauma. The anatomical region of the elbow is defined as the bony and articular surfaces of the distal humerus, the proximal ulna and the proximal radius.11 A traumatic injury is defined as any direct or indirect low-energetic or high-energetic trauma involving the elbow. A full list describing the inclusion and exclusion criteria is given in box 1.
Box 1Inclusion and exclusion criteria used during the selection of paediatric patients in this study
Inclusion criteria
Patients aged between 2 and 17 years.
Traumatic injury of the elbow (maximum 72 hours prior to presentation on the emergency department).
Pain in the anatomical region of the elbow joint.
Exclusion criteria
Pre-existent neurological pathology, genetic disorders and/or bone disorders in the affected limb.
Current ipsilateral fracture of wrist or shoulder.
Previous fracture of the ipsilateral upper extremity (from clavicle to distal phalanges) <3 months.
Patients referred from another hospital where X-rays of the elbow were performed.
A multitrauma patient (Injury Severity Score >16).
Children with an intellectual disability.
Unable to communicate in Dutch or English.
Sample size
A traditional sample size calculation is not recommended due to the multivariable character of this study. A sample size calculation through its ability to accurately estimate effect size is chosen, therefore, a modified convenience sample will be used. A logistic regression analysis is used to determine the potential variables for the final decision rule. The variance between outcomes per variable within a regression analysis dictates the sample size per variable. The variance between outcomes for the potential variables in the decision rule is estimated to be very small (predominantly yes/no answers). Jenkins and Quintana-Ascencio, Riley et al and Steyerberg12–14 described a detailed calculation for an adequate sample size in clinical decision/prediction models. Based on their recommendations, we aim to include 400 patients for our study. To summarise: on average 8–12 participants are needed per variable to ensure a valid prediction can be made concerning the variables’ discriminative value. To increase the accuracy of the prediction model, we will focus on 10 predictive potential variables with a high a priori chance of underlying traumatic injury to the bone. Therefore, we will need to include a minimum of 100 patients (10 variables × 10 patients). To ensure an accurate internal validation for our updated clinical decision rule a minimum of 300 patients must be included.14 Based on these estimates, we have chosen to include 400 patients for our research; 100 patients (25%) for the development of the decision rule and 300 patients (75%) for the internal validation of the decision rule.
Statistical analysis
The data from standardised electronic case report forms (CRFs) will be used to develop the prediction model, by using a multistep logistic multivariable analysis in a shrinkage model. The shrinkage model used will be a ridge regression, because of its ability to analyse data suffering from multicollinearity (multiple independent variables are correlated). During the first step, a univariate logistic regression analysis will be used to estimate the regression coefficients and analyse the correlation between a variable and the presence or absence of a fracture. The regression coefficients will be processed, in the second step, through a multivariate shrinkage model to establish significant regression coefficients and generate a relative risk score per variable. Sensitivity, specificity, positive predictive value and negative predictive values will be gathered. The accuracy of the model will be estimated by a goodness-of-fit test with a graphical calibration curve and a receiver operating characteristics curve with a discriminative curve. Overfitting will be controlled by calculating the optimism estimation of the C-statistic. Internal validation will be performed through bootstrapping to estimate overfitting and adjust the model accordingly. The final decision rule will be presented as a simplified risk score for easy use by emergency care physicians.
Missing data
We will use three strategies to avoid or adequately substitute potential missing data: (1) optimising the study design and implementation methods to avoid missing data, such as; training doctors, creating simplified CRFs and adhering to normal treatment protocol, (2) sending regular updates to all participating hospitals and (3) investigating patterns of missing data to allow analyses to explore potential reasons for missing data and impute missing values by chained equations to avoid bias.
Study procedures
Data collection will take place starting May 2023 and will be completed after including 400 patients, preferably within a 2-year period. All paediatric patients presenting to the ED following a traumatic injury of the elbow will receive care as usual according to hospital protocol. To develop the decision rule, we will collect patient characteristics in a standardised fashion during the interview and physical examination. A standardised electronic CRF will be generated to collect the data during the participant’s visit. The CRF will contain basis information on patient characteristics such as age, gender, injured arm, it will also include a physical examination, the results of the X-ray and the possible predictive clinical parameters. The attending (orthopaedic/surgical) physicians collecting the data will receive instructions and training before recruiting participants to the study. Possible predictive clinical parameters are patient age and gender, point tenderness at lateral or medial distal humerus, radial head, olecranon, limited range of motion for supination/pronation/flexion and extension, hypoesthesia of the lower arm, increased capillary refill test, visible haematoma and trauma injury mechanism. All participants will receive plain elbow radiographs, according to Dutch guidelines; anterior–posterior view with the hand in anatomical position and a lateral view with the thumb in upwards position.15 Additional imaging for associated injuries or to confirm suspected diagnosis will be performed at the discretion of the treating orthopaedic or trauma physician.
Primary outcome parameters
Our primary outcome measurement is the existence of a fracture on the conventional X-ray diagnosed by a musculoskeletal radiologist. A fracture is defined as a partial or complete disruption of one or more of the cortices in the ulna, radius or humerus within the elbow region and all epiphysial growth plate injuries visible on AP or lateral view. Avulsions or displacement of apophyseal growth plates are also defined as a fracture. All additional imaging (radiography, MRI or CT) performed by the on-call physician and radiologist will be taken into account when diagnosing the fracture.
Our primary outcome, a fracture of the elbow, will be measured after inclusion of all 400 patients. All participants will receive a conventional radiography. After inclusion has ended all conventional radiographs will be gathered for final inspection. This will be done by two musculoskeletal radiologist in consensual agreement and blinded to the clinical parameters and medical history of the patient. The two musculoskeletal radiologist will provide a detailed diagnostic report, after reaching consensus, for every radiography performed on our patients. This final report will dictate the presence or absence of a fracture of the elbow after traumatic injury to the elbow in the paediatric patient.
Withdrawal of individual subjects
Participants can leave the study at any time for any reason if they wish to do so without any consequences. The principal investigator or treating physician can decide to withdraw a subject from the study for urgent medical reasons.
Regulation statement
The study will be conducted according to the principles of the Declaration of Helsinki (64th World Medical Association General Assembly, Fortaleza, Brazil 2013) and in accordance with the Medical Research Involving Human Subjects Act (WMO, valid since 1 July 2021).
Recruitment and consent
Potential participants and/or the parents or legal guardians of the participants will be asked to join our study by the physician on call in the ED prior to regular diagnosis and treatment. Verbal informed consent will be given. Participants have no obligation to participate and will receive diagnosis and treatment as normal. Participants who are willing to join will receive similar treatment, the only difference is that clinical parameters recorded during patient interview and physical examination will be more extensive and will be recorded in a CRF.
Patient and public involvement
No patients were involved during the creation of this study protocol.
Administrative aspects
Handling and storage of data and documents
All acquired patient-related data will be anonymously coded with a referencing legend for safe used by members of the research team. Research data will be stored in a database (SPSS V.25 and Castor EDC and SMS) and can be traced to individual persons only by authorised personnel. The personnel authorised to view the database include the members of the research team, members of the healthcare inspection and members of the medical ethics committee. Review of the data may be necessary to ensure the reliability and quality of the research. The handling of personal data is in compliance with the Dutch act on Implementation of the General Data protection Regulation (in Dutch: ‘Wet Algemene Verordening Gegevensbescherming persoonsgegevens’), the EU General Data Protection Regulation and the privacy regulation of all involved hospitals.
