clinicians.build · interactive · august 26, 2026

The Escalation Gap

A hospital's sepsis software fires and somebody starts the bundle 79.7% of the time. Somebody finishes it 62.9% of the time. The 16.8 points in between are the escalation path — and they are the only part of it anyone measures.

Story: Alexandra Byrne, “AI software to detect hospital drug diversion has a problem: Humans,” STAT News, Aug 25 2026
Data: CMS Provider Data Catalog — Timely and Effective Care – Hospital (measures SEP_1, SEV_SEP_3HR, SEP_SH_3HR, OP_18b, OP_22, Safe Use of Opioids), reporting period Jan 1 2024 – Jun 30 2025

In late September 2024, patients at Adventist Health in Bakersfield noticed a travel nurse behaving strangely. She was taking their IV opioids from a secured cabinet and charting them as given. The hospital was running machine-learning software built to catch exactly that pattern. It fired. Hospital managers ignored the alerts, auditors found.

There is no dataset of ignored diversion alerts. Hospitals are not required to disclose that they run this software, or to report when it fails, so no denominator exists. But there is one place where CMS does publish, hospital by hospital, what happened after a clinical trigger fired — and it tells the same story with 3,084 data points.

What the sepsis bundle measures are, precisely

SEV_SEP_3HR — of patients with severe sepsis, the share who got the whole 3‑hour package: lactate drawn, blood cultures before antibiotics, broad‑spectrum antibiotics, fluids where indicated. The first move.

SEP_1 — the composite. Every 3‑hour element plus the 6‑hour work: repeat lactate, and for septic shock, vasopressors and a documented reassessment. The path completed and closed out. These are separate CMS measures with overlapping but not identical denominators — SEP_1 covers severe sepsis and septic shock together. SEP_1 requires the 3‑hour elements, so it is close to nested inside SEV_SEP_3HR, but not strictly so. Treat the vertical distance as a strong indicator, not an exact conditional dropout rate.

Every reporting hospital, first move against finished path

Each dot is one hospital. Horizontal: how often the first move landed. Vertical: how often the whole thing got finished. The dashed navy line is y = x — where a hospital would sit if starting and finishing were the same act. Hover any dot. Then drag the sample‑size floor.

Brush the axes
n ≥ 11
All owners Nonprofit For-profit Government
Highlight critical-access (333) Reset all
Hospitals shown
3,084
of 3,084 plotted
Mean gap
16.8
points lost start → finish
Below the line
98.6%
finish worse than they start
Spread (SD)
16.2
of the vertical measure
Correlation r
0.71
x against y
Hospital (dot size = sepsis cases reported) Critical-access, when highlighted y = x — starting and finishing at the same rate

Drag the floor to 100 and watch a third of the spread evaporate

The dots that look most alarming — the hospital at 0%, the hospital at 100% — are almost all small ones. Below 30 reported sepsis cases the standard deviation of completion is 22.8 points. At 300 cases or more it is 14.0. Nothing about care quality changed between those two groups. The arithmetic got quieter.

DenominatorHospitalsMean SEP_1SDScoring 0 or 100
Under 30 cases27558.622.83.3%
30–9986763.316.50.5%
100–2991,61362.914.80.1%
300 or more32965.614.00.0%

The gap itself, though, does not evaporate. It narrows from 19.0 points to 15.4 and stops. Every subgroup you can build out of this file — nonprofit, for-profit, government, tiny, enormous — finishes worse than it starts, by between fifteen and nineteen points. That is not a small-hospital problem. That is the shape of the thing.

What this means if you are shipping an alert

The detector is the easy half. Sepsis surveillance is a solved detection problem — the trigger logic is twenty years old and it fires. Diversion surveillance is a solved detection problem too; at Bakersfield it fired for weeks. What both stories have in common is that everything downstream of the firing was unowned, uninstrumented, and invisible to the vendor.

If your product can be fully ignored and never tell you, you don't have telemetry. You have a dashboard.Builder's Briefing, August 26, 2026

The instrumentation nobody ships: who received it, what their queue looked like when it landed, what the false‑positive rate did to their trust by week six, whether closing it required a reason, and whether anyone ever audited the closures. None of that is in a model card. All of it decides whether the model does anything at all.

⚠︎ AI-generated · not reviewed by a human · verify against the linked sources before relying on it. Hospital-level values are transcribed from the CMS Timely and Effective Care file; the gap framing, the small-sample analysis and the diversion analogy are this page's interpretation, not CMS's.