Department of Medical Laboratory, First Affiliated Hospital, Sun Yat-sen University, 58 Zhongshan Er Road, Guangzhou, Guangdong Province 510080, China.
Department of Medical Laboratory, Second Affiliated Hospital of Guangzhou Medical University, 250 Changgang Zhong Road, 510260, China.
Clin Chim Acta. 2023 May 1;545:117368. doi: 10.1016/j.cca.2023.117368. Epub 2023 Apr 29.
Iron deficiency anemia (IDA) and thalassemia trait (TT) are the most common causes of microcytic hypochromic anemia (MHA) and are endemic in lower resource settings and rural areas with poor medical infrastructure. Accurate discrimination between IDA and TT is an essential issue for MHA patients. Although various discriminant formulas have been reported, distinguishing between IDA and TT is still a challenging problem due to the diversity of anemic populations.
We retrospectively collected laboratory data from 798 MHA patients. High proportions of α-TT (43.33 %) and TT concomitant with IDA (TT&IDA) patients (14.04 %) were found among TT patients. Five machine learning (ML) approaches, including Liner SVC (L-SVC), support vector machine learning (SVM), Extreme gradient boosting (XGB), Logistic Regression (LR), and Random Forest (RF), were applied to develop a discriminant model. Performance was assessed and compared with six existing discriminant formulas.
The RF model was chosen as the discriminant algorithm, namely TT@MHA. TT@MHA was tested in an interlaboratory cohort with a sensitivity, specificity, accuracy, and AUC of 91.91 %, 91.00 %, 91.53 %, and 0.942, respectively. A webpage tool of TT@MHA (https://dxonline.deepwise.com/prediction/index.html?baseUrl=%2Fapi%2F&id=26408&topicName=undefined&from=share&platformType=wisdom) was developed to facilitate the healthcare providers in rural areas.
The ML-based TT@MHA algorithm, with high sensitivity and specificity, could help discriminate TT patients from MHA patients, especially in populations with high proportions of α-TT patients and TT&IDA patients. Moreover, a user-friendly webpage tool for TT@MHA could facilitate healthcare providers in rural areas where advanced technologies are not accessible.
缺铁性贫血（IDA）和地中海贫血（TT）是小细胞低色素性贫血（MHA）最常见的原因，在资源匮乏的环境和医疗基础设施较差的农村地区流行。准确区分 IDA 和 TT 是 MHA 患者的关键问题。尽管已经报道了多种判别公式，但由于贫血人群的多样性，区分 IDA 和 TT 仍然是一个具有挑战性的问题。
我们回顾性收集了 798 例 MHA 患者的实验室数据。在 TT 患者中，α-TT 的比例较高（43.33%），TT 合并 IDA（TT&IDA）患者的比例也较高（14.04%）。我们应用了 5 种机器学习（ML）方法，包括线性支持向量机（L-SVC）、支持向量机学习（SVM）、极端梯度提升（XGB）、逻辑回归（LR）和随机森林（RF），来开发一个判别模型。并评估了其性能，并与 6 种现有判别公式进行了比较。
选择 RF 模型作为判别算法，即 TT@MHA。TT@MHA 在一个实验室间队列中进行了测试，其敏感性、特异性、准确性和 AUC 分别为 91.91%、91.00%、91.53%和 0.942。我们开发了一个 TT@MHA 的网页工具（https://dxonline.deepwise.com/prediction/index.html?baseUrl=%2Fapi%2F&id=26408&topicName=undefined&from=share&platformType=wisdom），以方便农村地区的医疗保健提供者使用。
基于 ML 的 TT@MHA 算法具有较高的敏感性和特异性，可帮助区分 TT 患者和 MHA 患者，特别是在 α-TT 患者和 TT&IDA 患者比例较高的人群中。此外，TT@MHA 的网页工具易于使用，可方便农村地区无法获得先进技术的医疗保健提供者使用。
