Medication administration errors in the Neonatal Intensive Care Unit: contributory factors, impact and risk prediction tool development

Abstract

Medication administration errors (MAEs) are the most common type of medication error. They are more common among neonates as compared to adults. These errors pose significant risks to patients and impose a substantial economic burden on healthcare systems. Targeting and prioritising neonates at high risk of MAEs is crucial in reducing these errors. Therefore, this research aimed to develop and internally validate a machine learning (ML)-based risk prediction model for predicting the presence of MAEs in the neonatal intensive care unit (NICU). This research consists of five studies. The first study is a systematic review and meta-analysis that critically appraises the evidence on the prevalence and contributory factors of MAEs in the NICU. The pooled prevalence of MAEs for direct observation and non-direct observation studies was 59.3% and 64.8%, respectively, with error-provoking environments being the most common cause. The second study was a cross-sectional study that employed a validated selfadministered questionnaire to determine the estimated percentage of MAE reporting and described the reasons for MAEs from nurses’ perspectives in five Malaysian public hospital NICUs. The estimated MAE reporting rate was 30.6%, with inadequate nursing staff, look-alike drugs and similar drug packaging being common reasons. The third study was a prospective direct observational study, conducted at the same study sites as the previous study to determine the prevalence of MAEs and identify factors associated with the occurrence of MAEs. The error rate recorded in this study was 68.0%, affecting 92.4% of the neonates in the NICU (92.4%). Factors significantly associated with the occurrence of MAEs were medications administered intravenously (AOR = 21.18; 95% CI = 13.35-33.61; p<0.001), unavailability of a protocol related to the preparation and administration of medications (AOR = 2.43; 95% CI = 1.54-3.84; p<0.001), number of prescribed medications (AOR = 1.11; 95% CI = 1.01-1.23; p=0.048), gestational age in weeks (AOR=0.94; 95% CI=0.91-0.97; p<0.001), non-ventilated neonates (AOR=2.03; 95% CI=1.13-3.64; p=0.018), and years of nursing experience (AOR = 1.07; 95% CI = 1.04-1.11; p<0.001). In the fourth study, an expert panel assessed each MAE detected in the direct observational study for its potential clinical and economic outcomes. A total of 1018 out of 1288 (79.1%) errors were found to be potentially moderate in severity, while only 30 (2.3%) were found to be potentially severe. The estimated cost to the Ministry of Health, Malaysia, was estimated at MYR 43 664.16 (Int$ 27 452.10). Finally, an ML model was developed and internally validated to predict the presence of MAEs in the NICU. Among ten ML algorithms assessed, adaptive boosting (AdaBoost) was the best performing model (F1 score:83.28%, accuracy: 77.63%, area under the receiver operating characteristic: 82.95%, precision: 84.72%, sensitivity: 81.88%, negative predictive value: 64.00%). The most influential features of AdaBoost were the intravenous route of administration, followed by working hours and nursing experience. Overall, the findings from this research highlighted the burden of MAEs in the NICU and the factors significantly associated with these errors. The risk prediction model could potentially prevent MAEs, thereby reducing patient harm and positively impacting the healthcare system.