Skip to content
Submit incident
AII-1002-62161
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.

University of Pennsylvania AI algorithm for end-of-life care declined during pandemic

An algorithm at the University of Pennsylvania Health System designed to prompt oncologists to discuss end-of-life care with cancer patients degraded during the COVID-19 pandemic, according to a 2022 study. The tool failed hundreds of times to generate the needed prompts, potentially resulting in unnecessary chemotherapy for some patients who would have benefited from advance planning discussions.

18 January 2025 Β· University of Pennsylvania Health System Β· Reliability and hallucination
18 January 2025
AII-1002-62161View source β†—
Quick facts
Category
Reliability and hallucination
Country
United States
Source
Daily News
Date
18 January 2025
Drafted by
Raidu Incident Detection

What happened

A routine tech checkup revealed an algorithm at the University of Pennsylvania Health System designed to predict patient mortality and prompt oncologists to discuss end-of-life preferences decayed during the COVID-19 pandemic, performing 7 percentage points worse at predicting death.

The tool failed hundreds of times to prompt doctors to initiate important conversations about treatment and end-of-life preferences with patients who needed them.

Reported impact

The prompt failures may have resulted in unnecessary chemotherapy for patients who would have benefited from advance planning discussions with their doctors.

Lead researcher Ravi Parikh believes several algorithms designed to enhance medical care weakened during the pandemic, not just the one at Penn Medicine, suggesting a broader problem across institutions that do not routinely monitor algorithm performance.

Index analysis

FDA Commissioner Robert Califf stated that no single U.S. health system is capable of validating an AI algorithm deployed in clinical care, highlighting systemic capacity gaps.

Researchers note that artificial intelligence systems require consistent monitoring and staffing to function properly, and hospitals may lack the resources needed to validate and maintain these tools.

Governance controls

Many institutions are not routinely monitoring the performance of their AI products, according to researchers involved in the study.

The study was published in 2022 and documented the algorithm's degradation during the pandemic period.

Classification

Category
Reliability and hallucination
Industry
Healthcare, drugs and biotechnology
Country
United States
Harm status
Potential harm
Intent
Unintentional
Timing
Post-deployment
Harmed
Consumers
AI systems
End-of-life care prediction algorithm
Regulations
FCRA (cited)

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

Daily News
Health care AI, intended to save money, turns out to require a lot of expensive humans
18 January 2025