This was a prospective observational cohort study conducted in Sacramento County, California, involving 5 EMS agencies and 11 hospitals (one level I trauma center, 3 level II trauma centers, and 7 non-trauma centers). We obtained Institutional Review Board approval at all sites.

Inclusion and exclusion criteria regarding participants for this study has been outlined previously (19, 20). Briefly, our participants were patients 55 years and older who experienced head trauma and were transported by EMS between August 1, 2015 and September 30, 2016. Participants were excluded if they experienced penetrating head trauma, inter-facility transport, or did not undergo cranial CT imaging at their index ED visit. In addition, patients who did not consent to a follow-up telephone call or those who did not have a reliable means for follow-up were also excluded. EMS providers were given standardized data collection forms that included information pertaining to demographics, anticoagulant usage, and other clinical variables.

The data collection method has been described previously (19, 20). In summary, a standardized prehospital patient care report (PCR) form was completed for each patient by participating EMS agencies. Information collected included the patient’s demographics, vital signs, medical history, and history of present illness (e.g., vomiting, headache, loss of consciousness, amnesia, seizure after head injury), mechanism of injury, initial GCS score, and EMS provider judgment for tICH. EMS providers were asked “What is your suspicion for the patient having intracranial hemorrhage (bleeding in the brain)?” Responses were recorded as ordinal categories (<1%, 1–5%, >5–10%, >10–50%, or >50%).

Each patient’s EMS record was then linked to hospital electronic medical records (EMR) using patient name, date of birth, and date of EMS transport. A trained research coordinator then summarized the data from the PCR and EMR. This included patient name, date of birth, mechanism of injury, detail regarding EMS transport, initial GCS score, history of vomiting, headache, loss of consciousness, amnesia, or seizure after head injury, EMS suspicion for tICH, antiplatelet and anticoagulant use, and initial cranial CT results. All tests, including cranial CT imaging, were ordered at the discretion of the treating physician.

The primary outcome measured was the presence of tICH on initial cranial CT imaging. tICH included subarachnoid hemorrhage, subdural hemorrhage, epidural hemorrhage, intraparenchymal hemorrhage/contusion, or intraventricular hemorrhage.

We formatted and de-identified the data and recoded the variables using STATA 14.0 statistical software (STATA Corp, College Station, TX). Descriptive statistics were used to characterize the study population. Non-normal interval data were reported with medians and quartiles 1 (Q1) and 3 (Q3). The incidence of tICH was stratified by EMS provider judgment ordinal categories (<1%, 1–5%, >5–10%, >10–50%, or >50%).

We compared the sensitivity and specificity of EMS provider judgment to identify tICH to other sets of triage criteria, including current field triage criteria, current field triage criteria plus additional risk factors identified using a multivariate logistical regression model, and actual transport. EMS provider judgment was stratified as <1% judgment for tICH vs. 1% or higher judgment for tICH. Current field triage criteria included physiologic criteria (Step 1), anatomic criteria (Step 2), and mechanism of injury criteria (Step 3) (7).

Multivariate logistical regression risk factors were identified using a regression model that included a parsimonious set of covariates, comprised of age 85 years or older (ideal cut-point based on receiver operating curve), loss of consciousness or amnesia, history of vomiting, abnormal initial EMS GCS score (dichotomized where abnormal equals GCS score less than 15), evidence of trauma above the clavicles, mechanism of injury other than a fall from standing height or less, any anticoagulant or antiplatelet use, and the presence of any Step 1 to 3 field triage criteria. Anticoagulant or antiplatelet use included the use of warfarin, direct oral anticoagulants (dabigatran, rivaroxaban, apixaban, or edoxaban), aspirin, and other antiplatelet medications (clopidogrel, ticlopidine, prasugrel, dipyridamole, cilostazol, or ticagrelor). To identify the ideal cut-point for age, we explored various age cut-points using a receiver operating curve for nonparametric data and chose the cut-point that maximized sensitivity and specificity (Youden Index) (21). Significance was defined as a p value <0.05 and the results of the multivariate logistic regression model were presented as adjusted OR with 95% CIs. We used complete-case analysis to handle missing data due to the infrequency of missing data (<1%) and model fit was evaluated using the c-statistic (22, 23).

To evaluate for potential selection bias, we compared age, male sex, EMS GCS score, Injury Severity Score (ISS), and incidence of tICH between those who did and did not have an EMS provider judgment recoded. Because a higher risk threshold for EMS provider judgment might be tolerated, we also evaluated the sensitivity and specificity of EMS provider judgment stratified as 5% or lower judgment for tICH vs. >5% judgment for tICH.

Our initial cohort included 1,304 patients, of which 673 patients (51.6%) had an EMS provider judgment for tICH documented. These 673 patients had a median age of 75 years (interquartile range 64–85) and 319 (47.0%) were male. Most (72.8%) of these patients had a GCS score of 15 and the majority(73.8%) had experienced a fall from standing height or less. Following emergency department evaluation, most patients (57.9%) were later discharged home. The complete description of patients included in this study is documented in Table 1. There were no differences in age, male sex, GCS score, ISS, and the presence of tICH in patients with and without an EMS provider judgment recorded (Table 2).

Overall, 76 (11.3%) patients enrolled in the study had a tICH on initial cranial CT imaging. EMS providers most commonly had a low suspicion for tICH and rated 265 patients (39.4%) as <1% risk and 201 patients (29.9%) as 1–5% risk. In general, as EMS provider judgment for tICH increased, the incidence of tICH also increased (Table 3).

Adjusted analysis demonstrated a history of vomiting (OR 4.70, 95% CI 1.58–13.97), evidence of trauma above the clavicle (OR 2.41, 95% CI1.09–5.32), an abnormal initial EMS GCS score (OR2.23, 95% CI 1.26–3.96), Step 1 to 3 field triage criteria (OR 2.04, 95% CI 1.05 to 3.98), and loss of consciousness or amnesia (OR 2.02, 95% CI 1.14 to 3.59) as predictive of the incidence of tICH on initial cranial CT (Table 4).

Using a threshold of 1% or higher suspicion for tICH, EMS provider judgment had a sensitivity of77.6% (95% CI 67.1–85.5%) and a specificity of 41.5% (95% CI 37.7–45.5%). Using a threshold of >5% suspicion for tICH, EMS provider judgment had a sensitivity of 53.9% (95% CI 42.8–64.7%) and a specificity of 72.2% (95% CI 68.5–75.6%).

Steps 1 to 3 of the field triage criteria was poorly sensitive (26.3%, 95% CI 17.7–37.2%) in identifying tICH and Steps 1 to 3 of the field triage criteria plus multivariate logistical regression risk factors was sensitive (97.4%, 95% CI 90.9–99.3%) but poorly specific (12.9%, 95% CI 10.4–15.8%). Actual transport had similar accuracy as EMS provider judgment (sensitivity 71.1%, 95% CI 60.0–80.0%; specificity35.3%, 95% CI 31.6–38.3%) (Table 5). The c-statistic for the logistic regression model was 0.73, which indicates an overall good model fit.