Table 2.

Error analyses on the validation dataset.

Clinical text examples and the causes of Natural Language Processing (NLP) errors
Error analysis on injury onset
	1	“She has chronic pain—neck, low back, B/La shoulders. She has fibromyalgia and also fell a few weeks ago which worsened her back pain.” NLP incorrectly associated the event (“fall”) that occurred “a few weeks ago” with the shoulder problem when performing a cross-sentence search.
	2	Prior condition reported on day 0 visit: “My left shoulder pain never went away despite still doing physical therapy and living on NSAIDsb. Now it is constant and much worse today.” NLP incorrectly captured “today” as the shoulder pain onset date when performing a cross-sentence search.
Error analysis on injury duration
	3	On day 136, “States in past pain would travel to left shoulder causing numbness to left arm and lasting a few days but today denies any numbness.” NLP incorrectly identified the injury duration based on a resolved shoulder symptom.
Error analysis on injury cause
	4	“...with 1 day of pain in the left arm and shoulder. Denies any injury. Did some lifting yesterday.” NLP identified the cause as unknown, failing to identify the possible cause (daily activity).
	5	“She has been working on the computer a lot. Overhead movement exacerbates the pain... No injury or trauma.” NLP identified the cause as unknown, failing to identify the possible cause (daily activity).
	6	“...who complains of left shoulder pain that started 3 weeks ago after vacuuming.” NLP identified the cause as unknown, failing to identify the possible cause (daily activity).
	7	“...likely subdeltoid bursitis and supraspinatus tendinopathy in the setting of DMc likely from acute movement with pain when getting IVd placed.” NLP identified the cause as unknown, failing to identify the possible cause (accident).
	8	“Patient reports left shoulder pain with movement; no trauma. Patient worked for years caring for young children and had to carry and lift them.” NLP identified the cause as unknown, failing to identify the possible cause (daily activity).

^aB/L: bilateral.

^bNSAIDs: Nonsteroidal anti-inflammatory drugs.

^cDM: diabetes mellitus.

^dIV: intravenous.