Hepius Community

Hepius Community A health - tech community OUR VISION
A world in which no one is harmed while giving or receiving health care.

OUR MISSION
Drive forward policies, strategies and actions, based on science, data,
patient experience, system design and partnerships, to eliminate
all sources of avoidable risk and harm to patients and health workers. GOAL
Achieve the maximum possible reduction in avoidable harm due
to unsafe health care in developing countries and globally.

When someone with cancer gets a fever in hospital, doctors start strong antibiotics and check again two to three days la...
01/11/2026

When someone with cancer gets a fever in hospital, doctors start strong antibiotics and check again two to three days later. A new model predicted, at the 48-hour mark, whose fever would still be there, using data hospitals already collect.

Why it matters: the temperature chart did most of the work. Tested on 8,396 hospital stays in Germany and the US, the model scored 0.78 to 0.82, where 0.5 is a coin toss and 1.0 is perfect.

The catch: it looked back at old records, and the authors say it must never be used on its own to stop antibiotics or skip a scan.

New peer-reviewed study (npj Digital Medicine, 31 Oct 2026), explained in plain English:
https://hepius.co/ai-persistent-fever-cancer-antibiotics-study/

A model using routine hospital data predicted which cancer patients' fevers would last on antibiotics, at hospitals in Germany and the US. Explained.

Sometimes a doctor writes that you need a specialist, and the referral is never sent. An AI that reads every note each w...
01/11/2026

Sometimes a doctor writes that you need a specialist, and the referral is never sent. An AI that reads every note each week found about 7 times more of these missed referrals in seven Singapore clinics.

Why it matters: it almost never raised a false alarm (99.9% specificity), and in a one-week test about 6 in 10 of its flags were real, with far less staff time than doctors reviewing by hand.

The catch: it still caught only about half of missed referrals, and the authors say clinics also need to change how they work so referrals are not missed in the first place.

New peer-reviewed study (npj Digital Medicine, 30 Oct 2026), explained in plain English:
https://hepius.co/ai-missed-referrals-primary-care-study/

An AI that reads doctors' notes found about 7 times more missed specialist referrals a week in 7 Singapore clinics, but caught only half. Explained.

When one large US health system rolled out an AI note-taker, 7 in 10 clinicians in community pediatrics used it. Fewer t...
29/10/2026

When one large US health system rolled out an AI note-taker, 7 in 10 clinicians in community pediatrics used it. Fewer than 1 in 5 in academic general surgery did.

Why it matters: in some groups the tool was used in roughly 80% to 90% of eligible visits by the end of the six-month study, while others tried it and stopped. The authors say support should fit each clinic rather than be the same everywhere.

The catch: it is one US health system, and the study tracked who used the tool, not whether it improved care.

New peer-reviewed study (npj Digital Medicine, 26 Oct 2026), explained in plain English:
https://hepius.co/ai-scribe-adoption-study/

A study of 2,901 clinicians found AI scribe use ranged from 69% in community pediatrics to 16% in academic general surgery. The new study, explained.

When medical software gives the wrong answer, does the doctor catch it? Not as often as we assume.Why it matters: across...
29/10/2026

When medical software gives the wrong answer, does the doctor catch it? Not as often as we assume.

Why it matters: across 46 study reports, clinicians' odds of getting it right were about 10 times higher when the computer's advice was correct than when it was wrong. In real-world heart tracing (ECG) studies, about 1 in 4 wrong computer readings went uncorrected.

The catch: the main figures rest on a small number of studies that varied a lot, and few measured harm to patients.

New peer-reviewed review (npj Digital Medicine, 23 Oct 2026), explained in plain English:
https://hepius.co/clinical-ai-wrong-advice-human-oversight-review/

A review of 46 reports finds wrong computer advice sharply cuts clinicians' accuracy, and about 1 in 4 wrong ECG readings went uncorrected. Explained.

An AI tool missed about 1 in 4 broken bones on its own. The hospital's X-ray staff missed about 1 in 12.Here's the twist...
28/10/2026

An AI tool missed about 1 in 4 broken bones on its own. The hospital's X-ray staff missed about 1 in 12.

Here's the twist: when the staff used it as a second pair of eyes across 2,496 emergency X-rays, they made 10 fewer mistakes.

The catch: the AI gave no answer for almost 9 in every 100 scans, and staff rarely changed their minds, even when the AI was right.

New peer-reviewed study (npj Digital Medicine, 27 Oct 2026), explained in plain English:
https://hepius.co/ai-fracture-detection-emergency-department-study/

In 2,496 emergency X-rays, AI alone missed about 1 in 4 fractures, yet as a second check it cut radiographers' mistakes. The new study, explained.

Address

Addis Ababa

Telephone

+251965717111

Alerts

Be the first to know and let us send you an email when Hepius Community posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Contact The Business

Send a message to Hepius Community:

Shortcuts

Share