Clinical workflows are information systems. Understanding where information breaks reveals why data quality issues are structural, not accidental.
During my mother’s care, I learned that the path from first symptom to clinical decision is not a pipeline—it is a series of fragile handoffs. The biggest breaks happen in three places:
Triage → Nurse documentation → Physician interpretation
These stages look simple on paper. In reality, they are complex, lossy, and shaped by human constraints.
Triage is the first point where patient experience becomes structured information. But triage nurses often juggle:
• time pressure • long patient queues • limited context • incomplete histories • language barriers
As a result, triage notes are often:
• short • broad • symptom-level only • lacking timeline • lacking nuance
For multilingual patients like my mother, this stage frequently introduced the first layer of distortion:
“Stomach pain for a few days”
In reality, her description was closer to:
“A swelling, heavy discomfort that changes location, worsens after eating, sometimes feels like pressure from inside.”
That nuance never survived triage.
Nurses gather far more detailed information than triage—but documentation is influenced by:
• EHR templates • checkbox workflows • the speed required for patient throughput • their clinical training • the limits of interpreter accuracy
Nurses may document:
• vital signs • symptom location • severity scale • brief history
But severity scales themselves often fail multilingual patients. “0–10 pain” means nothing until explained—and even then, scoring varies culturally.
My mother took years to understand the pain scale. At times, she reported her pain as “3” when she was crying from it. At other times, she said “7” when she simply felt uncomfortable.
These structured fields become “clean data,” but they are not clean—they are approximations.
Physicians often summarize, reinterpret, or compress information from both triage and nursing notes. The physician note becomes the “ground truth” that downstream systems rely on.
But physician notes are shaped by:
• cognitive load • diagnostic framing • time constraints • their interpretation of symptoms • their trust (or lack of trust) in interpreters
This can lead to:
• missing early signs • altered timelines • generalized symptoms • subtle but important misinterpretations
I saw this repeatedly:
My mother described swelling pain → interpreter simplified → nurse documented “abdominal pain” → physician summarized it as “mild chronic abdominal discomfort.”
By the time her experience entered the EHR, it had changed shape entirely.
In math and data science, we think about “signal degradation” across noisy channels. Healthcare operates the same way.
Patient → Interpreter → Triage → Nurse → Physician → EHR → Data
At each step:
• nuance disappears • semantics drift • timelines compress • emotions vanish • uncertainty is removed • meaning is reshaped
By the time data becomes “data,” it is several transformations away from the original patient.
These workflow breaks directly influence:
• EHR data quality • real-world evidence • ML model bias • diagnostic accuracy • equity for multilingual patients
This is why I am drawn to studying:
• information flow mapping • clinical documentation behavior • multilingual representation • upstream data quality
Because improving healthcare AI requires understanding not only models, but **the messy human systems that create data in the first place**.