PREG-ADR: desenvolvimento e validação de um instrumento preditivo para reações adversas a medicamentos em gestantes hospitalizadas

Abstract

Aim: To develop and validate a predictive tool to estimate the risk of adverse drug reactions (ADRs) in hospitalized pregnant women (PREG-ADR). Methods: A case-control study nested within a cohort of high-risk pregnant women was conducted at a university hospital in Natal, Brazil, between September 2021 and July 2023. Eligible participants were hospitalized for > 24 hours and administered at least one medication. Cases were defined as patients with ADRs classified as "probable" or "definite" according to the Naranjo algorithm. Independent predictors were identified using multivariate logistic regression. Model calibration was assessed by the Hosmer-Lemeshow test and Brier score. Discrimination was evaluated via the area under the ROC curve (AUC), with bias-corrected c-statistics derived from bootstrapping. Results: The study included 330 pregnant women (110 cases and 220 controls). Eight independent risk factors for ADRs were identified: gestational hypertensive syndrome, self-reported drug allergy, aPTT ≥ 27.9 s, hospitalization duration ≥ 7 days, blurred vision, sedation, nausea, and polypharmacy (≥ 5 medications). The PREG-ADR tool yielded a total score (0–100), with a cut-off ≥ 34, AUC of 0.77 (95% CI: 0.71–0.82), sensitivity of 58.93% (95% CI: 53.62–64.24), and specificity of 79.82% (95% CI: 75.49–84.15). The model showed good calibration (p = 0.528) and acceptable predictive performance (Brier score = 0.27). Conclusion: PREG-ADR demonstrated robust discriminative ability, leveraging eight routinely monitored independent risk factors in hospitalized pregnant women.