Simulation study links hybrid machine learning to ring-foundation capacity
A mobile app using artificial intelligence (AI), developed by a researcher, was linked to better knowledge of drug calculations, higher scores on medication-related clinical decision-making tests, and greater confidence in their abilities among 56 nurses. The study used both quantitative and qualitative methods and randomly assigned nurses to groups at one non-governmental hospital in Egypt. The findings show short-term training results. They do not prove that the app reduces medication errors or improves patient care.
The study mainly focused on a randomized controlled trial with two groups. It also held focus groups after the training with nurses who used the app to explore whether the training was practical and acceptable. Nurses who provided direct patient care were recruited through the hospital's staff training program. After their starting skills were assessed, they were assigned to either the app-based training group or the traditional teaching group, with 28 nurses in each group.
The researchers used random assignment in blocks of four and made sure that nurses from different clinical areas were included in the assignment process. They used opaque, sealed envelopes with numbers in order to keep group assignments hidden until they were needed. The people who assessed nurses after training did not know which group each nurse belonged to. Both groups took part in three training sessions over two weeks. Each session lasted about 60 to 90 minutes, and both groups received the same learning content and about the same amount of teaching time.
The app used interactive cases about calculating oral medication doses, adjusting intravenous infusions, calculating doses for children, handling high-alert medications, and finding errors. Its rule-based system changed how difficult each case was and gave feedback based on set rules, including whether answers were correct, whether mistakes were repeated, and how quickly nurses responded. Nurses could work with virtual patient charts, infusion pumps, and patient reactions. The app also produced reports showing their performance.
Nurses in the control group received in-person lectures, guided discussions, and paper-based problem-solving exercises. The researchers measured outcomes just before training and two weeks after the final session. A 16-question multiple-choice test measured knowledge of oral drug calculations, intravenous infusion rates, and doses based on body weight. The test had a KR-20 reliability score of 0.82 and a two-week test-retest reliability score of r = 0.86. An eight-question survey developed by the researchers measured clinical judgment related to medications, safety checks, evaluation of information, and the use of calculation results. It had a Cronbach's α of 0.87 and a test-retest reliability score of r = 0.84. A 10-question self-efficacy scale adapted for the study had a Cronbach's α of 0.90 and a test-retest reliability score of r = 0.88.
After training, 42.9% of nurses in the app group reached the study's high-knowledge category, while 50% of nurses in the control group remained in the low-knowledge category. An independent t-test found a statistically significant difference between the groups, with a large effect size (Cohen's d = 0.89). An analysis that adjusted for starting scores also found a significant difference (F = 25.32, p < 0.001). In the app group, knowledge scores rose significantly from before to after training (t = 7.985, p < 0.001). At the start, 64.3% of these nurses were in the low-knowledge category, and none were in the high-knowledge category.
Scores for clinical decision-making related to medications were also significantly different between the groups, with Cohen's d = 0.90 and a significant result in the analysis adjusted for starting scores (F = 12.35, p = 0.001). After training, 42.9% of nurses in the app group reached the high decision-making category, compared with 14.3% in the control group. Scores in the app group also rose significantly from before to after training (t = 5.665, p < 0.001). These findings relate to scores on the study's questionnaire. They do not show that medication errors or patient outcomes actually changed.
Self-efficacy scores, which reflect people's belief in their ability to do tasks, were also significantly different between the groups. The effect size was Cohen's d = 0.81, and the analysis adjusted for starting scores found a significant difference (F = 16.54, p < 0.001). Scores in the app group increased from before to after training (t = 5.315, p < 0.001). After training, 25% of nurses in the app group were placed in the high self-efficacy category. In the control group, 64.3% were reported as having moderate self-efficacy and 35.7% as having low self-efficacy; no nurses in this group were reported in the high category. Because the researchers adapted a general self-efficacy scale, these results reflect nurses' reported confidence in the study setting, not directly observed clinical performance.
All 28 nurses in the app group took part in four focus groups, with seven nurses in each group. The researchers identified six main themes: perceived benefits for clinical work; how realistic the scenarios were and how well they connected theory with practice; confidence and ability; ease of use and engagement; training efficiency and resource savings; and challenges and suggested improvements. Nurses said the scenarios felt realistic, the app was engaging and easy to follow, and the training helped them apply calculations to clinical cases. They also said they were initially unsure about using the new technology and wanted more guidance and more detailed content. Only nurses who used the app took part in these discussions, so the study did not compare their experiences with those of nurses in the control group.
The trial has several important limits. It included only 56 nurses from one hospital, which makes the estimates less precise and limits how well the findings may apply to other settings. The follow-up period was short, so the researchers could not determine whether the improvements lasted over time. The outcome tools were created by the researchers or adapted from existing tools, and the study did not measure direct outcomes such as medication-error rates on hospital wards. The researchers could not fully rule out the possibility that nurses in different groups influenced one another or that their awareness of being studied affected their performance. One of the study's authors also developed the app. Larger studies across several hospitals, with longer follow-up periods, are needed to find out whether the results can be repeated, whether the app affects real clinical outcomes, and whether it can be used successfully in other settings.
Subject of Research: The short-term effects of adaptive AI mobile simulation on drug-calculation training outcomes among practicing nurses
Article Title: Adaptive AI-based mobile simulation for drug calculation competency in nurses: a mixed-methods RCT
Article References: Salem, M. F. A., Sharaf, A. K., & abdelhafeez abdelmonseif Younis, A. (2026). Adaptive AI-based mobile simulation for drug calculation competency in nurses: a mixed-methods RCT. BMC Nursing, 25(1), Article 828. https://doi.org/10.1186/s12912-026-05313-4
Image Credits: AI Generated
DOI: 10.1186/s12912-026-05313-4
Keywords: nursing education, artificial intelligence, mobile simulation, drug calculation, clinical decision-making, self-efficacy, randomized controlled trial, Egypt, adaptive, AI-based, mobile, simulation