Mining NHTSA CISS Data for Crash Reconstruction

A $190,000 Demand and No Comparable Data

You're reviewing a broadside collision at a four-way stop. A 2021 Hyundai Tucson struck the driver's side of a 2019 Toyota Camry at roughly 25 mph. The plaintiff's attorney has attached an orthopedic report listing a displaced rib fracture, a grade 2 liver laceration, and a lumbar disc protrusion. The demand is $190,000.

Your reconstructionist's report estimates Delta-V at 14 to 18 mph for the Camry. That feels about right. But the demand package is anchored to anecdote and narrative, not population-level injury data. You need a baseline. You need to know how often a 15 mph side-impact actually produces AIS 3 abdominal injuries in restrained occupants of midsize sedans.

That baseline exists. It lives inside NHTSA's Crash Investigation Sampling System, or CISS, and most adjusters and attorneys have never queried it directly.

What CISS Actually Contains

CISS replaced the legacy NASS-CDS program in 2017. NHTSA redesigned the sampling frame to reflect the modern vehicle fleet, including the explosion of SUVs, crossovers, and advanced driver-assistance systems that NASS-CDS couldn't adequately capture. Each year, CISS investigators examine around 5,000 crashes across 32 primary sampling units nationwide. Every case includes a full scene diagram, vehicle inspection with exterior and interior damage documentation, EDR imaging when available, and coded occupant injuries using the AIS 2015 revision.

The depth per case is what matters. A single CISS record can include crush depth measurements at six points across the damage profile, the principal direction of force (PDOF) coded to 10-degree increments, barrier equivalent speed, and whether the seatbelt pretensioner or load limiter fired. Injury records are coded by body region, severity (AIS 1 through 6), and source (steering column contact, door intrusion, seatbelt loading).

For crash reconstruction, this is primary-source gold. It connects physical evidence to occupant outcomes at a resolution no hospital billing dataset can match.

Why Reconstructionists Should Care About CISS

Crash reconstruction has always relied on reference datasets. For decades, that meant NASS-CDS, which ran from 1979 to 2015. Thousands of published biomechanical studies use NASS-CDS as their foundation. But the vehicle fleet has changed so dramatically that applying a 2008 NASS case to a 2022 collision introduces real methodological risk.

Consider side-impact protection alone. FMVSS 214, the federal side-impact standard, was updated in 2012 to include an oblique pole test. Vehicles manufactured after 2015 are structurally different in the B-pillar, door trim, and curtain airbag coverage. CISS captures these vehicles. NASS-CDS largely doesn't.

If you're estimating Delta-V for a late-model crossover and benchmarking against NASS cases from 2009, your confidence interval is wider than you think. CISS narrows it. You can filter by vehicle class, model year range, impact configuration, and restraint status to pull a cohort of comparable real-world crashes. That cohort gives you the distribution of Delta-V values, not just a point estimate from your own analysis.

"A single-case reconstruction tells you what probably happened. A population benchmark tells you how unusual that outcome is. You need both."
Senior biomechanical engineer, SAE-certified reconstructionist

Querying CISS for Injury Causation

Injury causation is where NHTSA CISS data becomes indispensable for claims and litigation. The dataset lets you answer a precise question: given this crash configuration and this Delta-V range, what is the observed frequency of a specific injury at a specific severity level?

Take the Camry broadside from earlier. You filter CISS for near-side impacts to midsize passenger cars, Delta-V between 12 and 20 mph, belted drivers, model years 2018 and later. You pull 47 weighted cases. In that cohort, AIS 3+ abdominal injuries appear in roughly 8% of occupants. Rib fractures at AIS 2 or higher show up in about 22%. Lumbar disc injuries above AIS 1 are under 5%.

Now your evaluation has teeth. The rib fracture is consistent with the crash. The liver laceration is uncommon but not implausible at the upper end of the Delta-V range. The lumbar disc protrusion, at AIS 2, is a statistical outlier for this configuration. That doesn't make it fabricated. But it tells you exactly where to focus your IME request and what questions to put to the treating physician.

This is the kind of analysis where Silent Witness's Delta-V estimation and AIS injury probability scoring give you a defensible anchor. The platform cross-references crash-photo-derived force data against population-level injury distributions, producing the same comparison you'd build manually from CISS, but in about five minutes.

How to Access and Filter CISS Records

NHTSA publishes CISS data through two channels. The first is the NHTSA research and testing databases portal, where you can download SAS or CSV files organized by case year. The second is the Case Viewer, a web interface that lets you browse individual cases with photos, scene diagrams, and coded variables.

For reconstruction benchmarking, the flat files are more useful. You'll work primarily with these tables:

• GV (General Vehicle): impact speed, Delta-V, PDOF, deformation location, vehicle make/model/year, and CDC (Collision Deformation Classification) codes.
• OA (Occupant Assessment): restraint use, airbag deployment status, seating position, ejection, and overall injury severity (MAIS).
• OI (Occupant Injury): individual injury records coded by AIS body region, severity, and injury source. This is where you connect specific injuries to specific contact points.
• DI (Distraction/Impairment): driver condition variables relevant to comparative fault arguments.

Joining GV to OA to OI on case ID and vehicle number gives you the full chain from crash forces to occupant outcomes. The sampling weights (RATEFIN in most years) are essential for producing nationally representative estimates. Unweighted counts alone can mislead you because CISS oversamples serious crashes by design.

Common Pitfalls When Using CISS

The dataset is powerful but not self-explanatory. Three mistakes show up repeatedly in reports that cite CISS data.

Ignoring the weighting scheme. CISS uses a stratified, probability-proportional-to-size sampling design. Every case has an analysis weight that accounts for the sampling probability. Reporting raw case counts as if they represent national frequencies will get you challenged in deposition. A Daubert motion targeting unweighted CISS analysis is an easy win for opposing counsel.

Treating CDC codes as precise measurements. The Collision Deformation Classification system is a shorthand. A code like 12-FDEN-3 tells you the damage is frontal, distributed, engaging narrow structures, with a severity rating of 3 on a 1-to-7 scale. It is not a crush measurement. It's not a substitute for photogrammetric damage analysis. Use CDC codes for filtering cohorts, not as standalone evidence.

Cherry-picking comparison cases. If you filter CISS until you have four cases that support your position, you don't have a benchmark. You have an anecdote with a government logo. Defensible use means specifying your filter criteria before you look at injury outcomes, documenting the weighted sample size, and reporting the full distribution of results. Even when the distribution includes outcomes that cut against your argument.

CISS and the Shift Toward Photo-Based Reconstruction

One of the structural limitations of CISS is response time. Investigators typically arrive at the scene or inspect vehicles within days, sometimes weeks. The scene has been cleared. Vehicles may have been partially repaired. That's a different evidence quality than what an adjuster or attorney captures in the first 48 hours through phone photos and dashcam footage.

The emerging approach pairs rapid, photo-based reconstruction tools with CISS as the validation layer. You run an initial damage severity and Delta-V estimate from claim photos. Then you benchmark that estimate against the CISS cohort for the same vehicle class, impact type, and damage profile. If your photo-based estimate of 16 mph Delta-V falls inside the interquartile range for comparable CISS cases, your confidence is high. If it's an outlier, you know to order a full inspection or an EDR download before moving forward.

This workflow is how we built the Silent Witness validation pipeline. Our 96% accuracy rate against known Delta-V values is benchmarked partly against EDR reads and partly against CISS population data. When the platform scores a crash, it's not generating a number from a black box. It's placing your crash within a distribution that has been validated against the same federal data a retained expert would use.

Practical Applications by Role

For adjusters and SIU investigators, CISS data answers the question that drives every BI evaluation: is this injury profile consistent with this crash? When a $75,000 demand on a parking-lot fender bender includes an AIS 3 cervical injury, and the CISS cohort for sub-5-mph rear impacts shows AIS 3+ cervical injuries in less than 0.3% of restrained occupants, you have a documented basis for your counter-offer or your referral to SIU. That's not claim denial. That's triage grounded in federal data.

For plaintiff attorneys, the same data works in reverse. If your client's AIS 2 rotator cuff tear appears in 18% of the CISS cohort for comparable T-bone collisions, you have a population-level argument that the injury is not just plausible but expected. Defense experts who testify that the crash couldn't have caused the injury now have to explain away the federal dataset. That's a harder deposition.

For defense attorneys, CISS benchmarking exposes inflated demands. When the biomechanical evidence shows a crash severity in the bottom quartile of the CISS cohort, and the plaintiff's claimed injuries map to the top decile of outcomes, you have a clean exhibit for mediation or trial. Pair it with a platform-generated damage-versus-injury mismatch score, and the story tells itself without needing three days of expert testimony.

Where CISS Falls Short

No dataset answers every question. CISS doesn't cover commercial vehicles over 10,000 pounds GVWR, so heavy-truck and tractor-trailer crashes require NHTSA's Large Truck Crash Causation Study or FMCSA data instead. Motorcycle and pedestrian crashes are underrepresented. And because CISS focuses on tow-away crashes, the lowest-severity impacts (the MIST cases that generate a disproportionate share of BI claims) are systematically excluded from the sample.

For minor-impact soft-tissue claims, you'll still need to rely on published biomechanical literature, volunteer sled-test studies, and platform tools that model occupant kinematics at low Delta-V values. CISS gives you the moderate-to-severe backbone. The low end requires other sources.

This content is for informational purposes and does not constitute legal or medical advice.

If you want to see how a specific crash benchmarks against population data, the free Delta-V calculator takes three photos and about two minutes.