J Drugs Dermatol. 2020 Dec 1;19(12):1241-1246. doi: 10.36849/JDD.2020.5006.
Drug resistance to biologics in psoriasis therapy can occur – it may be acquired during a treatment or else present itself from the beginning. To date, no biomarkers are known that may reliably guide clinicians in predicting responsiveness to biologics. Biologics may pose a substantial economic burden. Secukinumab efficiently targets IL-17 in the treatment of psoriasis.
To assess the “fast responder” patient profile, predicting it from the preliminary complete blood count (CBC) and clinical examination.
From November 2016 to May 2017 we performed a multicenter prospective open label pilot study in three Italian reference centers enrolling bio-naive plaque psoriasis patients, undergoing the initiation phase secukinumab treatment (300mg subcutaneous at week 0,1,2,3,4). We define fast responders as patients having achieved at least PASI 75 at the end of secukinumab induction phase. Clinical and CBC data at week 0 and at week 4 were analyzed with linear statistics, principal component analysis, and artificial neural networks (ANNs), also known as deep learning. Two different ANNs were employed: Auto Contractive Map (Auto-CM), an unsupervised ANNs, to study how this variables cluster and a supervised ANNs, Training with Input Selection and Testing (TWIST), to build the predictive model.
We enrolled 23 plaque psoriasis patients: 19 patients were responders and 4 were non-responders. 30 attributes were examined by Auto-CM, creating a semantic map for three main profiles: responders, non-responders and an intermediate profile. The algorithm yielded 5 of the 30 attributes to describe the 3 profiles. This allowed us to set up the predictive model. It displayed after training testing protocol an overall accuracy of 91.88% (90% for responders and 93,75% for non-responders).
The present study is possibly the first approach employing ANNs to predict drug efficacy in dermatology; a wider use of ANNs may be conducive to useful both theoretical and clinical insight. J Drugs Dermatol. 2020;19(12) doi:10.36849/JDD.2020.5006.
在银屑病治疗中，生物制剂的耐药性可能会出现——它可能是在治疗过程中获得的，也可能是从一开始就存在的。迄今为止，还没有已知的生物标志物可以可靠地指导临床医生预测对生物制剂的反应性。生物制剂可能会带来巨大的经济负担。司库奇尤单抗在治疗银屑病中有效靶向 IL-17。
从初步的全血细胞计数（CBC）和临床检查中评估“快速应答者”患者的特征。
2016 年 11 月至 2017 年 5 月，我们在意大利三个参考中心进行了一项多中心前瞻性开放标签试点研究，招募了接受生物制剂治疗的斑块状银屑病患者，接受司库奇尤单抗起始阶段治疗（300mg 皮下注射，第 0、1、2、3、4 周）。我们将快速应答者定义为在司库奇尤单抗诱导阶段结束时至少达到 PASI75 的患者。对第 0 周和第 4 周的临床和 CBC 数据进行线性统计分析、主成分分析和人工神经网络（ANNs）分析，也称为深度学习。我们使用了两种不同的 ANN：自动收缩映射（Auto-CM），一种无监督的 ANN，用于研究这些变量的聚类，以及训练与输入选择和测试（TWIST）相结合的监督 ANN，用于构建预测模型。
我们共纳入 23 例斑块状银屑病患者：19 例为应答者，4 例为无应答者。Auto-CM 检查了 30 个属性，为三个主要特征创建了一个语义图：应答者、无应答者和一个中间特征。该算法生成了 30 个属性中的 5 个来描述这 3 个特征。这使我们能够建立预测模型。经过训练测试协议，它显示出总体准确率为 91.88%（应答者为 90%，无应答者为 93.75%）。
本研究可能是首次采用人工神经网络预测皮肤科药物疗效的方法；更广泛地使用人工神经网络可能有助于理论和临床洞察力。J 皮肤病学杂志。2020;19(12)doi:10.36849/JDD.2020.5006.
