3. SWIFT Accurately Predicts Lichen Sclerosus among Premenarchal Girls

Abstract

<h3>Background</h3> Prevalence of lichen sclerosus (LS) is 1 in 900 in premenarchal girls. Clinical presentation is elusive, and diagnosis can be challenging. Early identification of LS increases the likelihood of timely and effective treatment. We therefore sought to create a risk-stratification tool to help guide clinical diagnosis in cases of suspected LS in premenarchal girls presenting with vulvar complaints. <h3>Methods</h3> This study utilized retrospective chart review of patients seen at the pediatric and adolescent gynecology clinic in a tertiary hospital between July 2019 and September 2020. Patients were included if they were ≤ 13 years old, were premenarchal, and had vulvar complaints. History, physical exam, and genital culture results were documented across 26 variables. LS diagnosis was made based on history, physical exam and vulvar biopsy in few cases. A comprehensive logistic regression was constructed comprising all variables with LS diagnosis as a dichotomous outcome. Following a step-wise regression procedure on complete-records observations, surviving predictor variables were tuned so as to maximize accuracy in predicting LS status. Pursuant to this model discovery, the data were randomly partitioned into a separate training set and testing set (1:1 allocation). This prediction was iterated 500 times. Statistical analysis was completed using R (v.4.0.0). Study was approved by the institutional review board. <h3>Results</h3> In total, 81 patient visits met inclusion criteria with 69 unique patients. Among the 69 patients, 19 were diagnosed with LS and 50 were diagnosed with other diseases including but not limited to straddle injury, acute vulvovaginitis, sexual abuse, and vulvar irritation. Average age of patients was 6 years old (0-13 years old; Table 1). The optimal regression model included five predictors: Soreness/Pain (S), Whitening (W), Urinary Incontinence (I), Fissures (F), and Clitoral hood thickening (T). The most accurate model using predictive training-testing approach is logOdds(LS) = -71 + 47•S + 83•W + 48•I + 41•F + 30•T with a model accuracy of 96.6% (Figure 1). <h3>Conclusions</h3> Using statistical modeling, we identified five predictors for LS including soreness/pain (S), whitening of the skin (W), urinary incontinence (I), fissures (F), clitoral hood thickening (T; SWIFT). We developed a model for predicting LS with 96.6% accuracy among premenarchal patients presenting with vulvar complains. If future studies involving diverse population show replicability of this model, clinicians should consider using this model to stratify patients for risk of LS.