Researchers Find Flaws in Algorithm Used To Identify Atypical Medication Orders (venturebeat.com) 9
Can algorithms identify unusual medication orders or profiles more accurately than humans? Not necessarily. From a report: A study coauthored by researchers at the Universite Laval and CHU Sainte-Justine in Montreal found that one model physicians used to screen patients performed poorly on some orders. The study offers a reminder that unvetted AI and machine learning may negatively impact outcomes in medicine. Pharmacists review lists of active medications -- i.e., pharmacological profiles -- for inpatients under their care. This process aims to identify medications that could be abused, but most medication orders don't show drug-related problems. Publications from over a decade ago illustrate technology's potential to help pharmacists streamline workflows by taking on tasks like reviewing orders. But while more recent research has investigated AI's potential in pharmacology, few studies have demonstrated its efficacy. The coauthors of this latest work looked at a model deployed in a tertiary-care mother-and-child academic hospital between April 2020 and August 2020. The model was trained on a dataset of 2,846,502 medication orders from 2005 to 2018. These had been extracted from a pharmacy database and preprocessed into 1,063,173 profiles. Prior to data collection, the model was retrained every month with 10 years of the most recent data from the database in order to minimize drift, which occurs when a model loses its predictive power.
Re: This is not how I envisioned the Internet in 2 (Score:2)
Your annoying rants have not gotten any better with time. Please stop. But I will point out that bitching about other people's porch cams and security cams being on the internet is dumb. First, no lack of expectation of privacy is needed to record you visually. The expectation of privacy thing only comes up if you're recording in an explicitly private place like a bedroom or bathroom or you're recording audio. If there's no audio and it's not a bedroom or bathroom, your non-expectation of privacy is not req
Anonymous whiner (Score:1)
Neural nets are stupid in this application (Score:2)
Using symbolic rule-based systems (actually, medical code-based systems) to check prescription histories would give more reliable answers. You might use a genetic algorithm to build the rules, but in-reality, a hand-tuned check algorithm would probably do better.
Re: Neural nets are stupid in this application (Score:2)
Finding atypical prescriptions is dumb anyway. What we need is to automatically detect harmful drug combinations. This is something the folks who publish the PDR should be doing.
Re: (Score:2)
I don't necessarily agree with "more reliable", but it's probably more trace-able, which may be more important. We like to know why things go wrong, especially in a litigious society.
One of the big drawbacks of neural nets (NN) is difficulty of trace-ability. That's why I propose factor tables (link in sig). It's generally less automatic than NN, but can be more regimented in return, so each result & layer is trace-able. It's to run AI
Ask who judged the quality? (Score:2)
The only people with skin in the game besides the patients are pharmacists.
25 pharmacist from the same location as the data came were asked to rate the algorithm. One could suggest a two-fold conflict of interest:
a) pharmacist fears that an AI could replace this aspect of their jobs
b) pharmacists are potentially seeing their own orders and really don't want to upset the status quo.
Why not have an independent review by people with no vested interest in the result?