Development and Validation of a Prognostic Clinical Risk Score for Subsequent Atopic Dermatitis Risk

Atopic dermatitis (AD), is a common, complex and heterogeneous inflammatory skin disease [1]. Many patients develop AD by 5 years, directly affecting quality of life [2], and increasing risk for other comorbidities, including asthma, allergic rhinitis, and food allergies [3]. Filaggrin mutations and family history of atopy are important risk factors for AD development [1, 4]. Each provides poor predictive value when used independently to stratify risk, yet family history of atopy has been used in the research setting [5, 6] and is the current standard of care (SOC). A prognostic AD clinical risk score can empower clinicians to diagnose AD earlier on. It can also be employed by the clinical trialist to enrich birth cohorts with a priori screening for at-risk patients. We aimed to develop a clinical risk score using information available in the prenatal period, with consideration to explainability and applicability, and validate it on an external, geographically independent dataset.

The training dataset included the electronic medical records (EMR) of all infant-mother dyads captured in the Leumit Health Services (Leumit) EMR between 2010 and 2019 with a recorded AD diagnosis before the age of three (n = 7370) and a random sample of infant-mother dyads selected from the Leumit EMR whereby AD was not diagnosed in the child's EMR (n = 63,852). Leumit is one of four national health providers and its EMR captures routine data from all medical consultations, procedures, and prescriptions, along with relevant sociodemographic data. The US-based Infant Feeding Practices Study 2 (IFPS2) was used as a validation dataset [7]. The cohort included a nationally distributed panel of 1807 infants born between 2005 and 2007, of which 389 (21.5%) reported a doctor's diagnosis of AD by the age of one. Additional information about study methods and findings are available in the online repository (https://github.com/tami-myor/AD-Score/tree/main).

The scoring method is designed to generate a value from 0 to 1, representing the probable risk in probabilistic terms, and can be effectively calculated if some variables are missing. Patients were divided into high (> 0.5), medium (0.35–0.5) and low-risk groups (< 0.35) There was a significant difference in means among the risk groups in relation to the actual cases of atopic dermatitis (ANOVA, p < 0.001) for both train and validation set. When compared to the SOC, individuals in the low-risk group were less likely to develop AD than those without a parental history of atopy and individuals in the high-risk group were more likely to develop AD than those with a parental history of atopy for both train and validation set. The OR in a univariate logistic regression model for developing AD was 2.30 (CI 2.15–2.46, p < 0.001) for the medium-risk group and 4.06 (CI 3.80–4.34, p < 0.001) for the high-risk group when compared with the low-risk group in the training dataset. Separately, the OR for developing AD was 1.74 (CI 1.34–2.27, p < 0.001) for the medium-risk group and 3.36 (CI 2.51–4.49, p < 0.001) for the high-risk group when compared with the low-risk group in the validation dataset. These high-risk ORs are more than double the OR of the SOC parental history of atopy calculated in the original multivariate model (OR 1.57, CI 1.43–1.72, p < 0.001) and demonstrate the utility of this method over the current SOC.

This prognostic AD risk assessment model harnesses known environmental and familial associations into a quantifiable and applicable clinical tool. It holds the potential to inform clinicians and caregivers as to an infant's risk of developing AD thereby influencing the continuum of care, enabling earlier diagnosis, and targeting patients to adopt preventive measures for other atopic conditions. Furthermore, it can reduce cohort sizes for AD-prevention clinical trials. While developed on epidemiological data from an EMR, the validation study showed its applicability on a geographically distinct survey-based cohort, thereby opening the door to deployment in a variety of settings. An important limitation in the validation dataset is its inclusion of diagnoses until 1 year of age, as opposed to 3 years of age in the development dataset. Undoubtedly, the inclusion of genetic information would further buttress the predictive capability of the model. Future studies can further this approach by incorporating additional documented risk factors and by exploring the utility of this approach for other conditions associated with the atopic march.

Tamar Landau: data curation (supporting), formal analysis (lead), investigation (supporting), writing – review & editing (equal). Keren Gamrasni: project administration (supporting), investigation (supporting), writing – review & editing (equal). Alex Levin: project administration (supporting), investigation (supporting), writing – review & editing (equal). Yotam Barlev: project administration (supporting), investigation (supporting), writing – review & editing (equal). Oliver Sanders: resources (equal), supervision (equal), investigation (supporting), writing – review & editing (equal). Shira Benor: resources (equal), supervision (equal), investigation (supporting), writing – review & editing (equal). Michael Brandwein: conceptualization (lead), data curation (supporting), formal analysis (supporting), investigation (lead), project administration (supporting), writing – original draft preparation (lead).

This study was performed according to the approved protocols and procedures received from the Helsinki Committee of Leumit Healthcare Services, Israel (October 2022, approval number 0026-22-LEU). A written informed consent waiver was obtained by the Helsinki committee based on the nature of the study, specifically the use of deidentified and previously collected data.

Tamar Landau, Keren Gamrasni, Alex Levin, Yotam Barlev and Michael Brandwein report personal fees from MYOR Diagnostics Ltd., during the conduct of the study. MYOR Diagnostics Ltd. develops innovative, evidence-based educational tools designed to help parents and clinicians understand and manage a child's health from the earliest stages of life. The company does not have any intellectual property related to the model described in this manuscript.