Details

Autor(en) / Beteiligte

• Waterson, James
• Kia, Arash
• Rolt, Stuart
• Bargary, Norma
• Garcia, Samuel
• Robertson, Jeremy
• Benavoli, Alessio
• Bergström, David
• Burke, Kevin

Titel

Predictive models for intravenous infusion longevity via infusion pump event log analysis

Ist Teil von

• Journal of critical care, 2024-06, Vol.81, p.154577, Article 154577

Ort / Verlag

Philadelphia: Elsevier Inc

Erscheinungsjahr

2024

Link zum Volltext

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• Infusion failure may have severe consequences for critical care patients, particularly if the medications are critical short half-life infusions. Interruptions to infusions, and infusions not running ‘to time’ can also lead to subtherapeutic management of patients receiving medications that require therapeutic monitoring. We aimed to identify determinants of longevity of infusions in critical care delivered by syringe driver and by volume pump, and to create, and test models for predicting the likelihood of successful infusion longevity. Data from 2,341,456 syringe pump and 3,498,279 large volume pump infusions were examined with a dependent variable of Infusion Longevity with explanatory variables of Medication Type, Concentration, Infusion Rate, and type of Critical Care Unit. Explanatory variables were added and removed from the core explanatory variables with each addition of explanatory ability being assessed against change in the adjusted goodness of fit (F-Score) within five statistical models. The study showed that ‘Infusion-Rate’ had the highest importance among other predictors of interruption. Eliminating this variable caused a 4.13% decrease in the F-Score. Following that, the ‘Care-Unit’ (3.45%), ‘Medication’ (2.75%), and ‘Concentration/Dose’ (1.41%). In 86% of the infusions, Random Forest was the most successful prediction model; in the remaining 14%, XGBoost was a better predictor. ANOVA confirms the statistical significance of F-Score reductions. t-tests were performed to check the significance of the F-Scores before and after the elimination of each predictor. The model identified and ranked predictive factors of longevity for infusions. This information can inform critical care clinicians of which types of infusion warrant more intense observation or proactive intravenous access management, in which kinds of care unit infusion failure is more likely, and what is the average length of uninterrupted infusion that can be expected in these care areas. We found that final medication concentration and medication type were of less significance to infusion longevity than rate and care unit. Optimizing rate settings to improve infusion longevity for continuous infusions through compounding to create customized concentrations for individual patients may be possible in the light of the results. Explanatory variables of added interest included pump occlusion pressure settings, and deployed syringe size and brand. A ‘sweet spot’ of infusion delivery rate for low flow infusions giving acceptable occlusion detection and shortened time-to-alarm by the pump along with extended infusion longevity was detected at 0.5–0.8 ml/h.