本稿第1部は、労災保険における身体部分のコーディングに関するものです。
保険金請求の約15%は、複数の身体部分として、具体的に傷害を受けた身体部分を特定しないあいまいな記述でコーディングされています。保険金請求の複雑性およびリスク・プロファイルは、手や指の傷害と首や頭の傷害とでは大きく異なるものの、いずれも複数の身体部分としてコーディングされます。複数の身体部分に係る保険金請求は、テキストマイニング・アルゴリズムを用いて非構造化データ（鑑定人のノートなど）を抽出することで具体的な身体部分の組み合わせに変換することが出来ます。
Figure 1: Body part coding accuracy - single vs multiple body parts
Consider the following:
- 33% of claims coded as multiple body parts are incorrect with a single body part referenced in the unstructured data.
- 40% of claims coded as a single body part are incorrect with multiple body parts referenced in the unstructured data.
- 45% of claims reference more than one injured body part in the adjuster notes and should be coded as multiple body parts—3 times the reported rate of 15%.
Over 40% of body part coding is inaccurate. Hover over the graphic above to explore coding accuracy of body parts.
Claims that reference multiple body parts in the unstructured data cost significantly more on average. Hover over the graphic below to explore the average severity of multiple body part claims.
Figure 2: Average severity by number of body parts
Text mining identifies the specific body parts underlying a multiple body part claim. Hover over the graphic below to explore the body parts most frequently referenced in multiple body part claims.
Figure 3: Most frequently referenced body parts for multiple body part claims
It is crucial to understand the individual body parts underlying a multiple body part claim. Hover over the graphic below to explore the average severity of the most frequent combinations of body parts in a multiple body part claim.
Figure 4: Average severity by body part combinations
Significant value is lost in the data by coding as multiple body parts. Text mining helps translate body part coding into meaningful data points for adjuster interpretation as well as for predictive analytics.
* All values above are representative of a workers' compensation claim population.