Reliability and validation of an electronic penicillin allergy risk-assessment tool in a pregnant population

Abstract Background: Penicillin allergy adversely impacts patient care, yet most cases do not have true allergies. Due to high reported prevalence of penicillin allergy, clinicians require efficient, reliable clinical tools to identify low risk patients who can be safely de-labeled. Our center implemented the FIRSTLINE electronic point-of-care decision support tool to help practitioners risk stratify patients with penicillin allergy. We sought to explore the reliability and validity of this tool in relation to actual patient outcomes. We additionally compared it with two other published stratification tools: JAMA and PENFAST, to assess ability to accurately identify low risk patients appropriate for direct oral challenge. Methods: In this single-center, retrospective, observational study, 181 pregnant females with self-reported penicillin allergy between July 2019 to June 2021 at BC Women’s Hospital, Vancouver, Canada were used to assess the reliability and validity of all three tools. Physician-guided history of penicillin use and symptoms were used for scoring. Results and recommendations were compared to actual patient outcomes after clinician decision for direct oral challenge or intradermal tests. We additionally applied two other tools to the patient population and compared performance across the tools. Results: 176 (97.2%) patients were assessed and deemed not allergic. Compared to the gold standard of clinician judgement, our electronic assessment tool has a specificity 0.93(0.88, 0.97), sensitivity 0.65 (0.41, 0.85), positive predictive value (PPV) 0.54 (0.33, 0.74), negative predictive value (NPV) 0.96 (0.91, 0.98). Other clinical tools were applied to the same population with the following results: 1) PENFAST specificity 0.85 (0.79, 0.90), sensitivity 0.40 (0.19, 0.64), PPV0.25 (0.11, 0.43), and NPV 0.92 (0.86, 0.96) and 2) JAMA toolkit specificity 0.66 (0.58, 0.73), sensitivity 0.85 (0.62, 0.97), PPV 0.24 (0.14, 0.35) and NPV 0.97 (0.92, 0.99). Conclusion: We demonstrate that our point-of-care electronic algorithm is reliable at identifying low risk pregnant patients, as compared to the gold standard of allergist assessment. To our knowledge, this is the first study to provide direct comparison between multiple decision support tools using the same population, minimizing participant bias. Providing clinical algorithms to accurately identify patients with low risk penicillin allergy, can enable healthcare professionals to safely risk stratify individuals to direct penicillin oral challenges versus referral to specialists. This increases the generalizability and efficiency of penicillin allergy de-labelling.