RAPEX HARM from AIS 2015 Coded Injuries

This paper presents a novel post-processing method that assigns RAPEX HARM values to AIS 2015-coded injuries by aggregating the three most severe injuries per person to holistically combine lethality (AIS-CD) and long-term consequences (pFCI), thereby providing a more comprehensive severity assessment than the current AIS-CD-only approach.

The Gist

The Big Picture: Why This Paper Matters

Imagine the European Union has a "Safety Gate" (called RAPEX) that acts like a bouncer at a club. Its job is to stop dangerous products (like a toaster that catches fire or a toy with sharp edges) from entering the market. To decide if a product is dangerous, the bouncer needs to know: "How bad could the injury be?"

For a long time, the bouncer only looked at one thing: "Could this kill you?" (This is the old way of measuring injury, called AIS-CD).

But life isn't just about dying; it's also about living with a limp, losing your hearing, or never playing sports again. This paper introduces a new way to measure injury that looks at both the risk of death and the risk of long-term disability. It's like upgrading the bouncer's checklist from "Will they die?" to "Will they die, or will they be stuck in a wheelchair forever?"

---

The Problem: The "Too Specific" Dictionary

The researchers used a massive medical dictionary called AIS 2015. Think of this dictionary as having over 2,000 specific entries for injuries.

• The Issue: It's too detailed. A doctor might write "laceration of the left index finger," but the dictionary has 10 different codes for that depending on the exact depth of the cut.
• The Result: When you try to use this dictionary to rank accidents, it gets messy. One coder might think a cut is "Level 3," and another might think it's "Level 4" just because they are being picky. It's like trying to sort a pile of sand by grain size; it takes forever and everyone disagrees.

The Solution: The authors decided to group these tiny details into bigger, more manageable buckets.

• Analogy: Instead of counting every single grain of sand, they just sort the pile into "Small Pebbles," "Medium Rocks," and "Big Boulders." This makes the data much more reliable and easier to compare.

---

The Two Scales: The "Life" Meter and the "Life" Meter

The paper uses two different rulers to measure a person's injuries:

1. The "Life or Death" Ruler (AIS-CD):
   • This measures how likely an injury is to kill you right now.
   • Analogy: It's like checking if a car crash was a "fender bender" or a "total wreck."
2. The "Future Life" Ruler (FCI-CD):
   • This measures how much your life will be ruined after you survive. Will you be blind? Will you walk with a cane?
   • Analogy: This is like checking if the car is drivable tomorrow, or if you'll need to buy a new one and learn to drive a different model.

The Catch: The "Future Life" ruler is brand new and hasn't been tested as much as the "Life or Death" one. The authors had to build a bridge to make it work with the old data.

---

The "Top 3" Rule (The Team Captain Analogy)

When a person gets hurt in a crash, they usually have more than one injury. Maybe they broke a leg, bruised a rib, and cut their head. How do you decide how "bad" the whole accident was?

• The Old Way: You just looked at the single worst injury. (If the leg was broken, the rest didn't matter).
• The New Way (The Paper's Method): They look at the three worst injuries and add them up.
  • Analogy: Imagine a sports team. If you only look at the star player, you might think the team is great. But if the other two players are terrible, the team might still lose. By looking at the "Top 3" injuries, they get a truer picture of the patient's overall condition.

They take these top three injuries, convert them into a score, and then translate that score into a HARM Level (1 to 4).

• HARM 1: A scratch. (You can go home immediately).
• HARM 2: A bruise or sprain. (You need a doctor, but you'll be fine).
• HARM 3: Serious injury. (You might be in the hospital for a while).
• HARM 4: Life-threatening or life-altering. (You might die, or you might be disabled for life).

---

What They Found (The "Aha!" Moment)

When they ran the numbers on real crash data from Germany (GIDAS), they found some surprising things:

1. Most injuries are minor: Over 70% of the injuries had almost no long-term consequences. (Most people just get a scratch or a bruise).
2. The "Future Life" ruler matters at the top end: For the most severe accidents (HARM 4), the "Life or Death" ruler was the main driver 53% of the time. But the "Future Life" ruler was the main driver 16% of the time.
   • Translation: There is a specific group of people who survive the crash but end up with a permanent, severe disability. The old system would have missed them or rated them too low because they didn't die. The new system catches them.
3. The "Double Whammy": In about 31% of the worst cases, both the risk of death and the risk of disability were high.

The Bottom Line

This paper is like upgrading the safety rating system for consumer products.

• Before: We only asked, "Will this product kill you?"
• Now: We ask, "Will this product kill you, OR will it leave you unable to walk, see, or work for the rest of your life?"

By using a smarter way to group injuries and looking at the "Top 3" worst hits, they created a system that gives a much more complete picture of danger. This helps the EU's "Safety Gate" (RAPEX) ban dangerous products before they hurt people, not just before they kill them.

Technical Summary

1. Problem Statement

The European Union's RAPEX (Rapid Alert System for non-food products) requires Hazard and Risk Assessment Methodology (HARM) evaluations that consider both injury lethality and long-term consequences (LTC). However, existing trauma databases (e.g., GIDAS, DGU, NASS CISS) typically utilize the Abbreviated Injury Scale (AIS) 2015, which provides two distinct metrics:

• AIS-CD (AIS Code): Measures threat to life (lethality).
• pFCI (predicted Functional Capacity Index): Measures long-term functional impairment (LTC).

Key Challenges:

• Lack of Integration: Current RAPEX assessments often rely solely on lethality (AIS-CD), ignoring LTC, or lack a standardized method to combine both metrics from AIS 2015 data.
• Coding Variability: There is a misalignment between specific AIS identifiers (AIS-IDs) and clinical diagnoses used by medical professionals, leading to coding inconsistencies (up-coding or down-coding).
• Validation Gap: While AIS-CD is well-validated against mortality, the pFCI metric is based on a theoretical framework pending empirical validation against real-world LTC data.
• Aggregation Complexity: Standard aggregation methods (like MAIS, which takes the single highest injury) fail to capture the cumulative severity of multiple injuries, which is critical for accurate risk assessment.

2. Methodology

The authors developed a post-processing pipeline to convert raw AIS 2015-coded trauma data into RAPEX HARM levels (1–4). The methodology involves three main stages:

A. AIS-ID Aggregation (Leveling)

To address coding variability and improve alignment with clinical reality, the authors aggregated the ~2,000 specific AIS-IDs into broader categories:

• For AIS-CD (Lethality): Three levels of detail were defined (Levels I–III) based on diagnostic specificity.
  • Level I: Organ system level (e.g., "brain injury" without specifying edema vs. swelling).
  • Level II: Differentiates subluxation vs. dislocation, named vessels, etc.
  • Level III: Differentiates contusion, laceration, avulsion, etc.
  • Skin Injury Removal: AIS-1 skin injuries were scrubbed from the dataset as they are often self-reported, non-reimbursable, and do not impact HARM ratings (FCI=1, AIS=1).
• For FCI-CD (LTC): A separate Level 1 aggregation was created focusing on easily diagnosable functional losses (motor loss, deafness, blindness) and potential future impairments (joint/cartilage damage). This reduced AIS-IDs from >2,000 to 453.
• FCI Recoding: The pFCI (where higher is worse) was inverted to FCI-CD (1 = perfect state, 5 = worst state) to align with the AIS-CD scale.

B. Person-Level Aggregation (Triple Conversion)

Instead of using the Maximum AIS (MAIS), the study adopted a New Injury Severity Score (NISS) approach, focusing on the three most severe injuries per patient.

• Transformation: An exponential transformation was applied to convert ordinal AIS-CD and FCI-CD values into a metric scale ($AIS_x$ and $FCI_x$).
• Calculation: The sum of the transformed values of the top three injuries was calculated ($NISS_x$).
• Mapping: These sums were mapped to S-Levels (0.0 to 3.0) and subsequently to HARM Levels (1 to 4).
  • HARM 1: Minor, reversible (S 0.0, 0.5).
  • HARM 2: Moderate, requires medical evaluation (S 1.0, 1.5).
  • HARM 3: Severe (S 2.0).
  • HARM 4: Life-threatening or irreversible severe impairment (S 2.5, 3.0).

C. Handling "Unknown" Codes (AIS-9)

• AIS-9 (Unknown Severity): Assigned a minimum viable AIS-CD based on the injury description. If the resulting NISS fell into the highest severity cluster, it was included.
• Non-Survivable: Injuries explicitly described as non-survivable (e.g., "torso transsection") were automatically assigned HARM 4.
• FCI-9: Due to the inability to derive a minimum FCI, a "liberal" approach was used. If FCI-9 was present but the patient's outcome was already HARM 4 based on other metrics, it was retained; otherwise, only AIS-CD was used for the final HARM determination.

3. Key Contributions

1. Novel Post-Processing Framework: A method to directly assign RAPEX HARM values to AIS 2015-coded data, bridging the gap between clinical trauma registries and product safety regulations.
2. Dual-Metric Integration: Successfully combines lethality (AIS-CD) and long-term consequences (FCI-CD) into a single severity score, moving beyond the traditional mortality-only focus.
3. Robust Aggregation Strategy: Demonstrates that aggregating the three most severe injuries (triples) provides a more accurate severity assessment than the traditional "maximum injury" (MAIS) approach.
4. Leveling Mechanism: Introduces a systematic reduction of AIS-ID detail to align with medical documentation capabilities, reducing coding bias.

4. Results

The method was validated using the GIDAS (German In-Depth Accident Study) database (1999–2024, ~1,000 crashes/year).

• LTC Prevalence: Over 70% of aggregated injuries had no or minor long-term consequences (FCI-CD 1).
• Metric Dominance by Severity:
  • Lower HARM Levels (1–2): Dominated by AIS-CD (lethality). LTC played a minimal role (only 1% of HARM 2 cases were defined by LTC).
  • Highest HARM Level (4): LTC became a significant factor.
    • AIS-CD drove the rating in 53% of HARM 4 cases.
    • FCI-CD drove the rating in 16% of HARM 4 cases.
    • Both metrics were equally severe in 31% of HARM 4 cases.
• Impact of Aggregation: The Level 1 aggregation improved the correlation between injury codes and mortality rates compared to original, highly specific AIS-IDs.
• Specific Examples: Certain injuries with low AIS-CD (e.g., knee joint dislocation, nerve lacerations) received high FCI-CD ratings, pushing the final HARM score to 4 (Life-threatening/Severe) solely due to long-term functional loss.

5. Significance and Limitations

Significance:

• Holistic Safety Assessment: Enables a more comprehensive evaluation of product safety risks by accounting for permanent disability, not just death.
• Regulatory Utility: Allows crash databases to be directly utilized for RAPEX risk clustering (e.g., identifying risks associated with specific accident types like speeding or mobile phone use).
• Standardization: Provides a reproducible method for converting complex trauma data into regulatory risk levels.

Limitations:

• FCI Validation: While AIS-CD is strongly correlated with mortality, the correlation between FCI-CD and actual long-term outcomes in real-world datasets is not yet fully established.
• Comparability: Comparing LTC across different body regions (e.g., cognitive decline vs. mobility loss) remains complex.
• Bias in HARM 4: The handling of "unknown" (AIS-9) codes introduced a bias toward the survivability metric (AIS-CD) in the highest severity category, as a reliable minimum FCI could not be derived for unknown codes.

In conclusion, this paper presents a critical advancement in injury epidemiology by operationalizing the dual-metric nature of AIS 2015 for regulatory risk assessment, proving that long-term consequences are a defining factor in the most severe injury classifications.