IRON-AI - IRON REPLACEMENT OPTIONS NAVIGATOR: AN R/SHINY-BASED ARTIFICIAL INTELLIGENCE TOOL TO SUPPORT DECISION-MAKING IN INTRAVENOUS IRON THERAPY
Author(s)
Francino Machado Azevedo Filho, PhD1, Layssa Andrade Oliveira Barbosa, Master2, Rosa Camilla Lucchetta, PhD2.
1Hospital Alemao Oswaldo Cruz, São Paulo, Brazil, 2Hospital Alemao Oswaldo Cruz, Haoc, Brazil.
1Hospital Alemao Oswaldo Cruz, São Paulo, Brazil, 2Hospital Alemao Oswaldo Cruz, Haoc, Brazil.
OBJECTIVES: Iron deficiency anemia (IDA) is highly prevalent and worsens clinical outcomes, quality of life, and costs. A network meta-analysis (NMA) of intravenous (IV) iron formulations revealed substantial heterogeneity in efficacy and safety, complicating interpretation. We aimed to develop IRON-AI, an interactive R/Shiny tool applying symbolic artificial intelligence to transform this evidence into clear, dynamic clinical recommendations.
METHODS: IRON-AI was built in R (4.4.0) using Shiny, with dplyr and janitor for data processing, ggplot2 for visualization, and shinydashboard for interface design. Inputs derived from a systematic review and NMA comparing six IV iron formulations. The workflow comprised: (1) normalization of key outcomes (hemoglobin change, log RR for hypophosphatemia, follow-up duration); (2) rule-based algorithms identifying statistical significance, effect direction, and GRADE certainty; (3) a symbolic inference engine applying "if-then" rules to derive clinical conclusions; and (4) interactive panels integrating temporal efficacy and efficacy-safety trade-offs.
RESULTS: The tool enables tailored comparison of IV iron formulations across subgroups, outcomes, and follow-up periods, generating automated interpretations with GRADE certainty ratings. A dynamic trade-off module integrates efficacy and hypophosphatemia risk, supporting transparent, evidence-aligned choices. The inference engine consistently flags when evidence favors an option or when uncertainty, inconsistency, or imprecision prevails. Visual panels clarify temporal effects and relative performance. Overall, IRON-AI strengthens evidence interpretation, reduces cognitive burden, and improves reproducibility. The application is freely available at https://uatshaoc.shinyapps.io/iron_ai/.
CONCLUSIONS: IRON-AI demonstrates the value of R and Shiny for HTA-oriented tools that strengthen knowledge translation and clinical decision-making. By coupling network meta-analysis with symbolic AI, it delivers dynamic, reproducible recommendations that convert complex evidence into immediate clinical insight-uniquely incorporating safety outcomes and improving the clarity, transparency, and applicability of HTA evidence.
METHODS: IRON-AI was built in R (4.4.0) using Shiny, with dplyr and janitor for data processing, ggplot2 for visualization, and shinydashboard for interface design. Inputs derived from a systematic review and NMA comparing six IV iron formulations. The workflow comprised: (1) normalization of key outcomes (hemoglobin change, log RR for hypophosphatemia, follow-up duration); (2) rule-based algorithms identifying statistical significance, effect direction, and GRADE certainty; (3) a symbolic inference engine applying "if-then" rules to derive clinical conclusions; and (4) interactive panels integrating temporal efficacy and efficacy-safety trade-offs.
RESULTS: The tool enables tailored comparison of IV iron formulations across subgroups, outcomes, and follow-up periods, generating automated interpretations with GRADE certainty ratings. A dynamic trade-off module integrates efficacy and hypophosphatemia risk, supporting transparent, evidence-aligned choices. The inference engine consistently flags when evidence favors an option or when uncertainty, inconsistency, or imprecision prevails. Visual panels clarify temporal effects and relative performance. Overall, IRON-AI strengthens evidence interpretation, reduces cognitive burden, and improves reproducibility. The application is freely available at https://uatshaoc.shinyapps.io/iron_ai/.
CONCLUSIONS: IRON-AI demonstrates the value of R and Shiny for HTA-oriented tools that strengthen knowledge translation and clinical decision-making. By coupling network meta-analysis with symbolic AI, it delivers dynamic, reproducible recommendations that convert complex evidence into immediate clinical insight-uniquely incorporating safety outcomes and improving the clarity, transparency, and applicability of HTA evidence.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HTA355
Topic
Clinical Outcomes, Health Technology Assessment, Methodological & Statistical Research
Topic Subcategory
Decision & Deliberative Processes
Disease
Personalized & Precision Medicine