Infusion Informatics to Improve Intravenous Medication Safety

Kang-Yu Hsu, Purdue University


Patient safety has been one priority in healthcare since the publishing of the Institute of Medicine report “To Err is Human: Building a Safer Health System” in 2000. Multiple studies have since demonstrated concerns about adverse drug events specifically caused by intravenous (IV) infusion errors. One approach of reducing adverse events and safeguarding patients from IV medication harm is the adoption of modern smart infusion pumps which are equipped with Dose Error Reduction Systems (DERS) and drug library (DL). However, while innovative technologies have been introduced to a complex system as healthcare for reducing the likelihood of adverse events, they might trigger latent errors which can be found only when the technology is practically embedded into the working environment as a whole. Therefore, in order to identify such errors from adopting smart pumps into IV medication delivery system, we need to investigate the records of infusion administration and also have a systemic view considering the task, the individual, the team, the work environment and the organization. In this research, we focused on the three topics in the area of smart pump informatics, corresponding to the key limitations, persisting issues and the most notable potential improvements summarized in multiple systematic reviews: (1) delayed DL updates issue, (2) contributing factors of using outdated DLs, and (3) impact from standardizing IV medication concentration. We aimed to extend the current understanding of the issues, mitigate the present limitation of using smart infusion devices, and improve the outcomes of IV medication use safety during administration. The analysis was conducted using infusion records and smart infusion pump device logs from multiple healthcare systems, ranging from January 2014 to December 2016. First, we identified the consequences and potential patient harm when administering an infusion using a smart pump with an outdated. More than 20%, 16.2% and 19.4% of infusion events were found to be administered using pumps with outdated DLs in the three hospitals. In the meanwhile, workarounds, latent use errors, missed alerts, alert fatigue, and dosage/rate miscalculation were identified in this study. Second, by integrating infusion records and smart pump machine logs, two distinctive steps of an infusion pump library update were defined, pending a new library and completing the library installation, on the pump of interest. Both steps contribute equally to the overall delay of the drug library update. Our results validated the hypotheses that the general pump utilization has a significant impact on the completion of drug library updates on individual pumps. Lastly, two prediction models were developed to identify unintended programmed infusions after standardizing IV medication concentrations. When the old smart pump programming behavior can still be implemented after the new concentration is applied, it raises concern of unintended overdose or underdose infusions, which are hardly found from infusion records. The presented models, basic model and copula-based model, enabled to predict unintended infusions with (mean sensitivity rate, mean specificity rate, mean AUC scores) as (91.29%, 66.21%, 0.903) and (83.62%, 78.99%, 0.895) respectively.




Yih, Purdue University.

Subject Area

Health care management|Information Technology|Artificial intelligence

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server