Diagnostic performance of package insert-based screening for potential drug–drug interactions in ICU patients receiving injectable medications: a retrospective diagnostic accuracy study using Lexi-Interact as the comparator in Japan

BACKGROUND: No established gold standard exists for identifying potential drug-drug interactions (pDDIs). In Japan, package inserts are commonly used for pDDI screening in clinical practice, whereas internationally curated databases such as Lexi-Interact serve widely as clinical decision support tools. However, the performance of package insert-based pDDI identification in intensive care unit (ICU) settings has not been fully evaluated. Therefore, in this study, we aimed to evaluate the concordance of package insert-based screening for detecting pDDIs in ICU patients receiving injectable medications, using Lexi-Interact as the comparator. METHODS: This single-center retrospective cohort study included adult ICU admissions involving patients aged ≥ 18 years at two ICUs of a tertiary hospital. Clinically significant pDDIs among injectable drugs were assessed at three predefined timepoints (day 2, day 5, and the day before ICU discharge). Package insert-based pDDI identification was compared with Lexi-Interact as the comparator. The primary outcome was the negative predictive value (NPV) of package insert-based screening for clinically significant pDDIs. Diagnostic performance was evaluated using sensitivity, specificity, positive predictive value (PPV), and weighted kappa in relation to Lexi-Interact as a pragmatic comparator. Robustness was assessed via ICU admission-level cluster bootstrap analysis with 10,000 resamples. RESULTS: Among 345 eligible ICU admissions, package insert-based screening yielded an NPV of 0.98, a sensitivity of 0.80, a specificity of 0.91, and a PPV of 0.51 compared with Lexi-Interact. Agreement between the two sources was moderate (weighted kappa = 0.57). Bootstrap analysis confirmed stable estimates of diagnostic performance. Interactions identified exclusively by Lexi-Interact were predominantly category C, with only one category D interaction and no category X interactions. CONCLUSIONS: In ICU settings, package insert-based identification of pDDIs involving injectable drugs showed high NPV, sensitivity, and specificity relative to Lexi-Interact. These findings suggest that package insert-based screening may serve as a practical initial screening approach with substantial concordance with Lexi-Interact classifications in ICU pharmacotherapy. Nevertheless, discrepancies in interaction severity classification highlight the potential benefit of complementary use of internationally established databases to enhance clinical interpretation and medication safety in critical care.