Developing a Decision Instrument to Guide Abdominal-pelvic CT Imaging of Blunt Trauma Patients

RESEARCH STRATEGY

Significance

Injuries to the abdomen and pelvis can lead to significant morbidity and mortality in adult blunt trauma patients. Clinicians, concerned about abdominal and pelvic injuries, have increasingly utilized abdominal-pelvic (A/P) CT imaging of blunt trauma patients.

CT imaging is compelling because of its documented high sensitivity in detecting injuries, and has lead some to recommend "pan-scan" imaging, which includes head, neck, chest and A/P CT imaging, of all blunt trauma patients. However, well-designed comparative trials have repeatedly failed to demonstrate improved patient outcomes thorough the use of "pan-scanning". It is hard to justify routine imaging if there is no benefit, particularly since A/P CT imaging is expensive and exposes patients to ionizing radiation and risk of lethal malignant transformation. As a consequence, recent recommendations emphasize the use of selective imaging in assessing most trauma patients.

At the present time, there are no definite recommendations to guide selective A/P CT imaging of blunt trauma patients. In current practice, imaging decisions are typically based on clinical judgment, which suffers from poor specificity and substantial variability, making it an imprecise and unreliable tool. A rigorously develop decision instrument could provide substantial benefit in guiding selective A/P CT imaging.

A decrease in CT utilization by as little as 10% would translate to substantial, direct, long-term benefits. There were approximately 1.5 million blunt trauma activation visits at the 1,632 adult trauma centers in the United States in 2015. Nearly half of these patients underwent abdominal-pelvic CT imaging, implying that nation-wide approximately 750,000 blunt trauma patients undergo A/P CT imaging each year. If a decision instrument could allow clinicians to safely forego at least 10% of these CTs, up to 75,000 CTs could be eliminated annually in the U.S. This reduction would provide cost and resource savings, reduced radiation exposure and iatrogenic cancers in the young vulnerable trauma population, and facilitate more efficient trauma evaluations, translating to shorter ED length of stay.

Approach

Investigators will employ the classic design strategy described by Stiell in developing the decision instrument. The development process includes separate derivation and validation phases. The clinical criteria identified in the derivation phase will form the basis of the instrument, with "high-risk" classification assigned to patients exhibiting one or more of the criteria, and "low-risk" classification assigned to patients who exhibit none of the criteria. Investigators will then conduct a separate validation assessment to determine whether the instrument retains optimal performance among a new cohort of patients. The validation phase will focus on ensuring that the final instrument has the high sensitivity and negative predictive value (NPV) needed to ensure its safe application, while retaining the highest possible specificity (and potential to decrease imaging).

The primary goal of the study will be to develop an instrument that reliably identifies patients who have injuries of major clinical significance that are evident on A/P CT imaging. The injuries of interest in this goal include injuries that require intervention, as well as any injuries to the aorta (including those not requiring intervention), and select injuries to the spine.

The secondary goal of the study will be to develop a tool that reliably identifies patients who have injuries of major or minor clinical significance. The injuries of interest related to this goal include injuries that require intervention, injuries that merit clinical observation, any injuries to the aorta, and select injuries to the spine.

To be clinically reliable, the DI must exhibit a sensitivity and negative predictive value of 98% or greater. At this level of precision, the risk of unfavorable outcomes due to missed injuries is equivalent to the risk of lethal malignant transformation from the radiation exposure associated with increased imaging. Validating a 95% lower confidence bound of 98.0% for a measured NPV of 100.0% requires evaluations on 183 patients assigned "low risk" status by the instrument. Similar validation of the sensitivity requires evaluations on 183 patients who have sustained injuries of major clinical significance. Because "low risk" patients are more prevalent than those sustaining major injuries, the ultimate sample estimate is driven by the need to enroll 183 patients who have sustained major A/P injuries.

Protocol

Phase I - Derivation

Clinicians will assess the presence or absence of each of the candidate criteria for each imaged patient during their initial emergency department evaluation. These assessments will be recorded prior to initiating CT imaging.

The injury status of each patient will be based on final radiologic interpretation of the initial A/P CT imaging obtained in the ED. Investigators require that relevant injuries be evident on CT imaging because the risk classification assigned by the instrument is immaterial if an injury is not visible on CT imaging. Injuries will be further classified as being of major, minor, or insignificant concern based on review of the patients care during their index visit.

The three level injury classification reflects the fact that many clinicians are comfortable in missing minor injuries provided that all injuries requiring intervention are identified, while other clinicians are uncomfortable using tools that occasionally miss minor injuries that require no intervention, and seek to identify all injuries regardless of their clinical significance.

To accommodate these perspectives, investigators will employ identical techniques to develop a two separate decision instruments. The first, and primary goal of the study, is to develop an instrument that identifies all injuries requiring intervention, while a second goal is to create an instrument that identifies all significant injuries, whether they require intervention or not.

Investigators will employ binary recursive partitioning to derive the decision instruments. The primary instrument must exhibit a sensitivity and negative predictive value greater than 98.0% in detecting and excluding major injuries. The second instrument must exhibit similar sensitivity and NPV in detecting and excluding major and minor injuries.

Phase II - Validation

In the second phase of the study, investigators will apply the instruments to a new cohort of blunt trauma patients to assess their performance and determine whether they retain sufficient sensitivity to support their use as A/P CT decision tools.

Clinicians will again document the presence or absence of criteria during their initial emergency department evaluations, and record their impressions prior to imaging. Investigators will base the presence or absence of injury on the final radiologic interpretation of the initial A/P CT imaging obtained in the ED. Injuries will be further classified as being of major, minor, or insignificant concern based on review of the patients care during their index visit.

The primary outcome will be the sensitivity of the instrument in classifying as "high-risk" all patients harboring injuries of major importance. Investigators will also calculate the NPV, specificity, and count the number of "low-risk" patients for whom imaging might have been safely omitted had the decision instrument been implemented.

The secondary outcome will be to perform a similar analysis focused on identifying patients with injuries major or minor clinical significance, and will measure the sensitivity, NPV, specificity, and potential to reduce imaging using this additional version of the instrument.

Data Collection

Criteria assessments.

Physicians will assess the status of the individual criteria during their initial emergency department evaluations and prior to ordering CT imaging. Investigators will record demographic information for each patient, along with the clinician assessments of the presence or absence of potential predictor variables. The potential predictor variables include abdominal pain or tenderness, flank pain or tenderness, pelvic pain or tenderness, hip or iliac pain or tenderness, midline lumbar spine or sacral pain, abnormal alertness, evidence of intoxication, distracting painful injury, and hypotension. Investigators will also record historical information and emergency department vital signs measurements.

Data collection during the validation phase will follow a similar process, but will be limited to demographic information and the criteria that compose the decision instruments.

Patient outcomes.

The presence of significant major or minor injuries will be based on final radiologic interpretation of the initial CT imaging obtained in the ED. The classification into major or minor status will be based on review of the patients care during their index visit to determine whether the patient underwent any interventional procedures.

All injury assessments will be completed by trained investigators blinded to criteria assessments and final classification by the decision instrument. Investigators require that relevant injuries for the study must be evident on CT imaging because risk classification assigned by the instrument is immaterial if an injury is not visible on CT imaging.

Injuries of major clinical significance consist of all abdominal and pelvic injuries requiring intervention, as well as any injury to the aorta, and any injury to the spine involving instability or neurological compromise. Injuries that require only observation, but no intervention, will be considered clinically minor, provided they do not involve the aorta or spine, while injuries that required neither intervention nor observation, and that do not involve the aorta or spine will be considered insignificant.

Formulation of the optimal A/P CT decision instrument.

Investigators will treat each individual criterion as a dichotomous variable. For analytic purposes in the derivation phase, each variable will be considered negative (normal) unless it can be assessed and found to be abnormal. Thus, variables that cannot be measured in an individual patient (such as coagulopathy in an unconscious individual) will be documented as "unable to assess" and treated statistically as negative. This will ensure that the derivation process is based on observed assessments and is as robust as possible. Investigators will use the individual criteria to construct the nodal points of the recursive partitions.

Investigators will employ binary recursive partitioning to identify a combination of candidate criteria that predicts significant abdominal-pelvic injuries with > 98.0% sensitivity, excludes significant injuries with > 98.0% NPV, and retains the greatest specificity. Sensitivity will be used as the outcome measure for constructing the partitions. Investigators will terminate the partitioning under the following circumstances: 1) partitioning produces an instrument exhibiting a sensitivity of 100.0%, 2) partitioning exhausts all patients without developing a sufficiently sensitive rule, 3) partitioning exhausts all variables (criteria) without identifying a sufficiently sensitive rule. Circumstance I will confirm the hypothesis driving the first specific aim. Circumstance 2 and 3 may confirm the first hypothesis, provided the lower confidence limit of for the sensitivity of the resultant rule exceeds 98.0%, otherwise these circumstances will be interpreted to refute the first hypothesis.

Investigators will create two distinct decision instruments. The first, and primary instrument will focus on identifying only those patients having injuries of major clinical significance. Investigators will also develop a second instrument that identifies patients having injuries of either major or minor significance.

Validation of the A/P CT decision instrument.

The validation phase will again treat each individual criterion as a dichotomous variable. Each variable will be considered negative (normal) only if it is assessed and found to be normal. Variables that cannot be measured in an individual patient (i.e. coagulopathy in an unconscious individual) will be documented as "un-assessable" and treated statistically as positive. This ensures that patients will be assigned low-risk status only if all criteria are adequately assessed and found to be negative. Patients will be excluded from low-risk classification (and assigned high-risk status) if they have one or more positive or un-assessed criteria. This requirement ensures patients are not assigned low-risk status on the basis of an inadequate assessment.

Using the final risk assignments and patient outcomes, investigators will assign each case to one of the following four categories: 1 - true-positive status (injured patients classified as "high-risk" by the DI); 2 - false-negative status (injured patients classified as "low-risk" by the DI); 3 - false-positive status (uninjured patients classified as "high-risk" by the DI); and 4 - true-negative status (uninjured patients classified as "low-risk" by the DI). Investigators will employ these values to calculate point measures and 95 percent confidence intervals for sensitivity, NPV and specificity of the decision instrument. Investigators will consider the instrument to be validated if the lower 95% confidence interval for its measured sensitivity exceeds 98.0%.

Assessing the potential to decrease A/P CT imaging.

Investigators must assess whether application of the tool to blunt trauma patients actually has the potential to decrease A/P CT imaging. Because of the tool's inherent sensitivity, "low risk" patients will be free of injury and receive no benefit from imaging. They are ideal candidates to exclude from imaging. Thus, measuring the proportion of "low-risk" patients will provide the investigators with an estimate of potential imaging reductions. Reductions in charges and radiation exposure will be determined by respectively summing radiographic charges and life-time decreases in radiation morbidity and mortality for all "no risk" cases.