Pediatric Blunt Abdominal Trauma Clinical Prediction Model

Overview

The submitted proposal is designed to reduce morbidity and mortality to injured children. Significant variability in the initial trauma assessment exists among institutions. The proposed project is a prospective, observational, multi-institutional study of children following blunt abdominal trauma. The specific goals of the project are to: 1) Document history, physical exam findings, imaging, and laboratory values, which are available to physicians during the initial trauma resuscitation prior to a decision on whether to order an abdominal computed tomography (CT) to evaluate for potential intra-abdominal injury; and 2) Derive and validate a multi-variable clinical prediction rule based on data variables readily available during the pediatric trauma resuscitation to identify patients at low risk for intra-abdominal injury, in which unnecessary CT might safely be avoided. Information from this study could be used to develop a more standardized approach to the evaluation for intra-abdominal injury following blunt trauma in children. This information could lead to significant improvement in the early recognition of injury and to improved resource utilization.

The proposed research project is a prospective, observational study. Data will be collected from each institution and entered into a secure REDCap database. Two phases of data collection will occur: 1) patient demographics and specific clinical variables which may be associated with IAI which are available and recorded during the initial trauma evaluation 2) laboratory and imaging results available after the initial assessment, identified injuries, patient disposition, interventions and outcomes. Primary Outcome Variables: 1) Intra-abdominal injury (IAI) (presence of solid or hollow organ injury (spleen, liver, kidney, GI tract, adrenal, pancreas, intra-abdominal vascular structure, bladder, ureter, gallbladder, abdominal wall fascia)) 2) IAI requiring intervention (abdominal operation, angio-embolization, blood transfusion) Time Course: The initial data collection period will extend for one year. Data will be recorded in a password protected redcap database which is readily available online to all participating institutions. Study setting: Eleven Level One Pediatric Trauma Centers Data collection points: demographics (age, mechanism, alert level), physical exam, labs (AST, HCT, amylase, UA, base deficit), FAST, imaging results, injuries, outcomes (admission, ICU admit, need for intervention, missed injuries, ISS). Data Analysis: De-identified data will be used during the data analysis phase to minimize the risk of loss of confidentiality to the patients. Data analysis with development of a clinical prediction rule (CPM) will be performed as follows. A logistic regression will be used to fit a predictive model for both IAI and IAI requiring interventions. SAS 9.3 will be used for the statistical analysis. Validation (including sensitivity and negative predictive value) of the derived CPM would then be performed in a subsequent study using a second population of patients. Internal validation of the prediction model was assessed by creating a split sample using a random selection process; half of the sample was used as the initial cohort to develop the prediction model for estimates of all covariates, and the remaining half was used as the validation cohort to compare the true to the predicted outcomes. A receiver operating characteristic (ROC) curve is created by plotting sensitivity against (1- specificity) for different cut-off points of the predicted outcome. A bootstrap study of 1000 replications was performed on the level of sensitivity, specificity and TP/FP/FN/TN validation. Validation (including sensitivity and negative predictive value) of the derived CPM would then be performed in a subsequent study using a second population of patients. Quality assurance plan: Data quality will be evaluated bi-monthly by a team consisting of the PI, an expert in clinical prediction models and a statistician. This panel will be able to review de-identified data from all institutions but will not have access to make any changes in the data entered in the centralized database. Data checks to compare data entered into the registry against predefined rules for range or consistency: Data checks to compare entered registry data against predefined rules for range and consistency will be performed bi-monthly. Source data verification: Source data collection is the responsibility of each individual institution. A data analysis team consisting of a statistician and an expert in clinical prediction models will evaluate the data bi-monthly to look for wide outliers (beyond pre-defined range) and impossible data values (not clinically possible) and request clarification from the individual institutions when necessary. Data dictionary that contains detailed descriptions of each variable: The vast majority of the data collection points are discrete variables with no opportunity for "free text" data entry. Many of these variables are specifically described on the data collection tools. Data definitions and question from the individual sites are addressed in a frequently updated Frequently Asked Questions (FAQ) document which is sent out to the sites bi-monthly following data verification review. Standard Operating Procedures to address registry operations and analysis activities: The database is established through a secure Vanderbilt Redcap web-based site. Individual site data collection will be analyzed and reviewed bi-monthly with feedback to each of the sites if data inconsistencies exist. Plan for missing data to address situations where variables are reported as missing, unavailable, "non-reported," cannot be interpreted, or considered missing because of data inconsistency or out-of-range results: Critical data which are missing, unavailable or not reported will not be utilized for the development of the blunt abdominal trauma clinical prediction model (BAT CPM). The investigators will attempt clarification for out of range results prior to data analysis. Variables which are available and reported in less than 50% of the patients will likely be excluded from the CPM because they will not form the basis of a practicable risk stratification model. The developed CPM will require validation in a second un-related population prior to widespread application of the CPM.