A systematic review of predictive models for hospital-acquired pressure injury using machine learning.
You Zhou, Xiaoxi Yang, Shuli Ma and 2 others
PMID 36310417WHAT IT FOUND
Machine learning models predicted hospital-acquired pressure injuries with areas under the curve from 0.68 to 0.99, but none underwent external validation.
High risk of bias affected 15 of 23 studies. Current models are not ready for clinical use.
Key findings
01The 23 included studies reported areas under the receiver-operating characteristic curve ranging from 0.68 to 0.99 for their best machine learning models.
02No study performed external validation, and 10 studies did not report any validation method, limiting confidence that these models work in different hospitals.
03Fifteen of the 23 included studies were judged to have a high risk of bias, with the main problems in the analysis domain.
STILL TO COME
How it was doneWhat they foundWhat it means for RNs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
No study performed external validation, so it is unknown whether any model would work in a different hospital or patient population. Fifteen of the 23 studies had a high risk of bias, and the overall quality of evidence was rated as poor. The review included only English-language publications since 2010, which may have missed relevant studies. A meta-analysis was not possible because the included studies used different performance indicators. The prevalence of pressure injuries varied widely across studies, from 0.5% to 55.6%, making pooled interpretation difficult.
Declared interests
Not reported in the supplied text.
The easy way to misread this
Do not treat the reported areas under the curve as proof that these models are clinically ready. None underwent external validation, and 15 of the 23 studies had a high risk of bias, so the reported performance may not hold in a real-world hospital.