Deep dive — chart 2

Chart 2 — H3 baseline vs recent distribution

The how. Where Chart 1 shows when the H3 line dropped, this chart shows how the H3 score distribution shape shifted. Mass moved left, out of the urgent-intervention lane.

H3 baseline vs recent triage-score distribution comparison

H3 triage-score distribution — baseline (blue) vs recent (orange). Mass shifted left, away from the urgent-intervention lane threshold at score 67.

What you are looking at

X-axis: triage severity score (0 to 100)
Y-axis: density of encounters at that score for H3
Blue distribution: baseline window (first 30 days), what H3 scoring normally looks like
Orange distribution: recent window (last 30 days), what H3 scoring looked like at Snapshot time
Vertical line at score 67: threshold for routing to urgent-intervention lane

What the shape shift means

The orange distribution's mass has moved left compared to the blue distribution. Three things are happening simultaneously:

The peak (mode) drops. The most common H3 score used to be near the urgent threshold; now it is well below it.
The right tail shrinks. Fewer H3 encounters score above the urgent-intervention threshold.
The left tail grows. More H3 encounters score in the fast-track range.

Clinical translation: the AI now systematically scores atypical elderly presentations (altered mental status, non-specific abdominal pain, fever/infection without localizing signs) as less severe than baseline. The score does not distinguish between "atypical elderly severe" and "young low-comorbidity mild." The distribution shape drift is the mathematical fingerprint of that clinical failure mode.

What KL-divergence quantifies

Kullback-Leibler divergence measures how different one distribution is from another. Values:

KL < 0.05: the same distribution (random noise)
KL 0.05-0.2: mild shift, monitor
KL 0.2-1.0: significant shape shift, flag as event
KL > 1.0: substantial shape shift, high-severity event
KL > 2.0: distributions are almost non-overlapping — the population the AI is scoring is behaviorally different from the population it was trained on

The sample's H3 KL-divergence hit 2.108 on Day 44 and 2.953 on Day 86. That is far above threshold, and it stayed there.

The clinical + regulatory read

Clinical read: the AI has drifted out of clinical calibration for atypical elderly presentations. Manual attending re-triage should apply until retraining + validation.

Regulatory read: the shape shift is precisely the kind of "measurable drift in performance on a specific subpopulation" that HHS Section 1557, CMS AI Interpretive Guidance, and FDA PCCP language expect the operator to detect and act on.

The next chart

The baseline-vs-recent chart shows how the H3 distribution shifted. The lane-rate chart shows the operational consequence: which lane are H3 patients getting routed to now vs baseline.

Chart 3 — lane rate per patient group →

$499 Snapshot. 3 business days.

Same shape-analysis for your hospital's actual AI surface + full analysis report.

Buy $499

Back to healthcare sample hub.