4. Data sources for head injury analysis

Head injury root cause analysis is fundamentally an evidence-integration problem. No single data source can establish both the injury mechanism and the causal chain that produced it. A crash test database tells you what the dummy's head experienced under controlled, repeatable conditions; a field incident report tells you what actually happened on the road but with limited measurement fidelity; medical records establish the actual injury outcome and its severity; and sensor data — whether from event data recorders, instrumented helmets, or wearable impact monitors — provides the time-history bridge between the collision event and the occupant's or patient's response. Effective analysis depends on knowing the provenance, resolution, limitations, and failure modes of each source, and on reconciling them when they disagree.

Crash test databases: controlled boundary conditions

Crash test databases such as NHTSA's Vehicle Crash Test Database, the NHTSA Crashworthiness Data System, and Euro NCAP's published test results provide the most rigorously controlled head-impact data available. Each record typically includes the test configuration (frontal rigid barrier at 56 km/h, oblique moving deformable barrier, pole impact, side MDB, etc.), vehicle and restraint configuration, dummy type (Hybrid III 50th percentile male, THOR, or the Euro NCAP Advanced Dummy), and the full measured channels: head resultant acceleration (a_r), head injury criterion (HIC₁₅ and HIC₃₆), head angular velocity, neck forces and moments, and airbag/seatbelt deployment timing.

The analytical value of these databases lies in their repeatability and their explicit boundary conditions. When you query a database for pole-impact tests with a deploying side airbag, you know the impact velocity, the barrier geometry, the dummy seating position, and the sensor mounting location. That makes it possible to compute derived metrics consistently. For example, HIC₁₅ is computed from the resultant head acceleration a(t) in g over a 15 ms window:

HIC = max over [t1, t2] with (t2 - t1) = 15 ms of:
      (t2 - t1) * [ (1/(t2 - t1)) * ∫ a(t) dt ]^2.5

A practical caution: database records are only comparable when the dummy head CG location, the accelerometer mounting, and the filtering (typically SAE J211 CFC 1000 for head acceleration) match. Mixing unfiltered and CFC-filtered channels, or comparing a Hybrid III record against a THOR record without accounting for the different head mass and neck response, produces spurious conclusions. Always record the dummy type, filter class, and coordinate convention (SAE J211 sign convention: +X forward, +Y right, +Z down) alongside any extracted metric.

Field incident reports: real-world context, limited fidelity

Field data — police collision reports, insurance claims, NASS/CDS investigations, and manufacturer field incident reports — supply the real-world exposure that controlled tests cannot. They capture the actual collision configuration, occupant kinematics, restraint usage, and post-crash scene evidence. Their weakness is measurement resolution: head impact location is often inferred from interior contact marks, and acceleration is rarely measured at all. The analyst must therefore treat field reports as a source of causal context and configuration rather than of quantitative injury metrics.

A defensible workflow is to use field reports to define the scenario envelope — impact direction, delta-v, occupant size, restraint state — and then to select matching crash test records or run a validated simulation within that envelope. When a field report and a test record conflict (for example, the report describes a near-side impact but the closest test is a 30° oblique), the discrepancy must be documented rather than silently averaged away.

Medical records: the injury outcome ground truth

Medical records — emergency department notes, CT and MRI reports, neurosurgical operative notes, and discharge summaries — establish what injury actually occurred. For head injury analysis the critical fields are the Glasgow Coma Scale at presentation, the Marshall or Rotterdam CT classification, presence and thickness of subdural or epidural hematoma, subarachnoid hemorrhage, diffuse axonal injury findings, skull fracture pattern, and the Abbreviated Injury Scale (AIS) head score with its associated mortality risk. The AIS head score is the standard bridge between clinical outcome and engineering metric: it allows a measured HIC or angular velocity to be compared against a population-level injury risk curve.

Medical records also carry a temporal dimension that crash data lacks. The time from impact to loss of consciousness, the evolution of the CT findings on repeat imaging, and the presence of a lucid interval are diagnostically meaningful — a lucid interval followed by rapid deterioration is characteristic of epidural hematoma, whereas immediate and prolonged unconsciousness points toward diffuse injury. These clinical patterns constrain the plausible acceleration and angular-velocity regime and should be used to sanity-check any reconstructed impact.

Sensor data: the time-history bridge

Sensor data closes the gap between the event and the outcome. Sources include vehicle Event Data Recorders (which in many jurisdictions record pre-crash speed, brake status, seatbelt status, and delta-v, and in some cases occupant restraint and rollover data), instrumented helmets and mouthguards used in sports and military settings, and wearable impact monitors. These provide sampled time histories — typically 100 Hz to 1 kHz for EDR pre-crash channels and higher for dedicated head-impact sensors — from which peak acceleration, impulse, and angular velocity can be derived.

Sensor data requires careful conditioning before it can be compared with crash test channels. Check the sampling rate against the signal bandwidth of interest (head impact energy is concentrated below roughly 300 Hz, so a 1 kHz sample rate is adequate but a 100 Hz EDR channel is not), verify the sensor's mounting and its orientation relative to the head or vehicle coordinate frame, and apply the same filter class used in the reference database. A common failure is treating an EDR delta-v channel as if it were a head acceleration channel; they measure different things at different points and must not be substituted.

Integrating the four sources

The analytical discipline is to assign each source its proper evidential role and to reconcile them explicitly. The figure below shows the intended data flow: field reports and sensor data define the event and its time history; crash test databases supply the controlled reference envelope and validated injury-risk relationships; medical records supply the outcome against which the reconstruction is judged. Divergence between the reconstructed metric and the clinical outcome is itself diagnostic — it may indicate an unmodeled contact, a restraint malfunction, or a sensor calibration error, and it should trigger re-examination rather than adjustment of the numbers to fit.

In practice, the reconciliation step is where most analytical errors occur. A worked example makes the discipline concrete. Suppose a field report describes a 40 km/h frontal impact with a belted 50th-percentile male driver, the EDR records a delta-v of 38 km/h, the medical record documents a 6 mm acute subdural hematoma with GCS 13 at the scene, and the closest crash test record is a 56 km/h rigid-barrier test yielding HIC₁₅ = 480. The analyst should not simply scale the HIC by the velocity ratio. Instead, the correct procedure is to (1) confirm the EDR delta-v is a vehicle-level channel and not a head channel, (2) select or simulate a test configuration matching the 40 km/h delta-v and the actual restraint state, (3) compute HIC₁₅ under the matched configuration with the same CFC 1000 filter, and (4) compare the resulting HIC against the AIS 2+ head injury risk curve. If the matched-configuration HIC predicts a low probability of subdural hematoma but the clinical record shows one, the divergence points to an unmodeled factor — a pre-impact braking-induced head position, a seatbelt load-limiter slip, or a contact with the A-pillar — and the investigation should return to the field report and sensor data to test those hypotheses.

The four sources are complementary rather than redundant, and the analytical value comes from the tension between them. Crash test databases provide the controlled reference and the injury-risk relationships; field reports provide the real-world configuration; medical records provide the outcome ground truth; sensor data provides the time history that links event to outcome. A root cause conclusion is defensible only when it accounts for all four and explicitly explains any residual divergence.