GoalA model can only respond to what it was given. Find out what this one can’t see.
1Change the features
2Send an observation
3Read the takeaway
Step 1 · Change the featuresThese four inputs are all the model was trained on.
Age 76+6
Medications 6+15
Prior fall history+0
Mobility status+0
→
Fall risk model
Takes in 4 features
AgeMedsFallsMobility
Learned from past patients
FellDid not fall
Gives a score for how much this patient resembles those who fell.
→
Output · in the patient chart
OKAFOR, M76 F · 4B-12
Fall Risk Score
22Low risk
Auto-calculatedToday 07:00
Move a slider or pick an option to see the score respond.
Step 2 · What you noticed this shiftSelect an observation to see whether the model can use it.
✕
Outside the model’s view.The model can’t take in new information. It wasn’t trained on this, so the score stays at 22 no matter what you add.Only you can weigh it. That’s why the score informs your clinical judgment and doesn’t replace it.
TakeawayThe score moved every time you changed a feature, and not at all for what you saw at the bedside. The model doesn’t adapt to new context: it reflects only its features and the label it learned to predict. Every decision still needs your clinical judgment.