RESUMO
As AI models continue to advance into many real-life applications, their ability to maintain reliable quality over time becomes increasingly important. The principal challenge in this task stems from the very nature of current machine learning models, dependent on the data as it was at the time of training. In this study, we present the first analysis of AI "aging": the complex, multifaceted phenomenon of AI model quality degradation as more time passes since the last model training cycle. Using datasets from four different industries (healthcare operations, transportation, finance, and weather) and four standard machine learning models, we identify and describe the main temporal degradation patterns. We also demonstrate the principal differences between temporal model degradation and related concepts that have been explored previously, such as data concept drift and continuous learning. Finally, we indicate potential causes of temporal degradation, and suggest approaches to detecting aging and reducing its impact.
Assuntos
Inteligência Artificial , Aprendizado de MáquinaRESUMO
BACKGROUND: Hypotension following orthopaedic surgery has been associated with increased morbidity and mortality. Rapid response teams (RRT) review patients on hospital wards with hypotension. AIM: To evaluate the epidemiology of hypotensive RRT activations in adult orthopaedic patients to identify contributing factors and areas for future quality improvement. METHODS: Timing of RRT activations, presumed causes of hypotension and associated treatments were assessed. RESULTS: Among 963 RRT activations in 605 patients over 3 years, the first calls of 226 of 605 patients were due to hypotension, and 213 (94.2%) of 226 had sufficient data for analysis. The median age was 79 (interquartile range 66-87) years; 58 (27.2%) were male, and comorbidities were common. Most (68%) surgery was emergent, and 75.1% received intraoperative vasopressors for hypotension. Most activations occurred within 24 h of surgery, and hypovolaemia, infection and arrhythmias were common presumed causes. Fluid boluses occurred in 173 (81.2%), and the time between surgery and RRT activation was 10 (4.0-26.5) h. in cases where fluid boluses were given, compared with 33 (15.5-61.5) h. where they were not (P < 0.001). Blood transfusion (30, 14.1%) and withholding of medications were also common. Hospital mortality was 8.5% (18), and 13.6% (29) were admitted to critical care at some stage. In-hospital death was associated with older age, functional dependence, arrhythmia and presumed infection. CONCLUSIONS: Hypotension-related RRT calls in orthopaedic patients are common. Future interventional studies might focus on perioperative fluid therapy and vaso-active medications, as well as withholding of anti-hypertensive medications preoperatively.