Bargir Umair Ahmed, Setia Priyanka, Desai Mukesh, S Chandrakala, Dalvi Aparna, Shinde Shweta, Gupta Maya, Jodhawat Neha, Jose Amrutha, Goriwale Mayuri, Yadav Reetika Malik, Vedpathak Disha, Temkar Lavina, Shabrish Snehal, Hule Gouri, Gowri Vijaya, Taur Prasad, Athavale Amita, Jijina Farah, Bhatia Shobna, Shukla Akash, Kalra Manas, Sivasankaran Meena, Balaji Sarath, Jain Punit, Sharma Sujata, Gangadharan Harikrishnan, Narula Gaurav, Sharma Ratna, Kini Pranoti, Mangalani Mamta, Zanwar Abhishek, Chaudhary Himanshi, Chaudhary Narendra Kumar, Khurana Ujjawal, Bavdekar Ashish, Subramaniam Girish, Raj Revathi, Saniyal Subhaprakash, Shah Nitin, Petiwala Tehsin, Kumar Prawin, Pai Venkatesh, Bhattad Sagar, Sengupta Abhinav, Soneja Manish, Upase Dayanand, Ganapule Abhijeet, Talukdar Indrani, Madkaikar Manisha
ICMR National Institute of Immunohaematology, Mumbai, India.
Birla Institute of Technology and Science, Pilani - Goa Campus, Sancoale, India.
J Clin Immunol. 2025 Aug 26;45(1):127. doi: 10.1007/s10875-025-01897-9.
Common Variable Immunodeficiency (CVID) is a heterogeneous disorder characterized by impaired antibody production and recurrent infections. In this study we investigated the clinical and immunological features of CVID in Indian patients and develops a machine learning model for predicting disease severity. We retrospectively analyzed 150 patients diagnosed with CVID over a decade at a tertiary care center in India. The median age of diagnosis was 18 years, with a male predominance (62%). The majority of patients (66.6%) had a severe phenotype, with recurrent respiratory tract infections being the most common clinical manifestation (84.2%). Gastrointestinal complications were observed in 45% of patients, while autoimmune manifestations were seen in 21%. All patients exhibited hypogammaglobulinemia. IgA levels varied, with 7.8% normal and 14.5% undetectable. IgM levels were decreased in 85.5% of patients. B-cell analysis revealed 64.4% had reduced class-switched memory B cells, with 21.7% showing very low levels. Nine adult patients presented with late-onset combined immunodeficiency. Genetic testing, performed on 52 patients, identified underlying monogenic causes in 29 pediatric and 15 adult patients. LRBA deficiency was the most common genetic defect, found in seven pediatric and three adult patients. We developed a novel machine learning-based severity prediction model for CVID patients, utilizing readily available lymphocyte subsets, class-switched memory B cell counts, and serum immunoglobulin levels to provide an accessible and robust tool for predicting disease severity using Ameratunga's clinical severity score. Random Forest outperformed other models across all metrics, achieving an accuracy of 0.853 (95% CI: 0.840-0.866). Feature importance analysis across all models identified Th-Tc ratio, CD19, and IgM levels as the most influential predictors for severity prediction. Our study highlights the diverse clinical and immunological features of CVID in Indian patients, emphasizing the need for early diagnosis and individualized management strategies. The machine learning model developed using commonly available immune parameters provide a robust tool for predicting disease severity, potentially guiding treatment strategies to improve patient outcomes.
常见变异型免疫缺陷（CVID）是一种异质性疾病，其特征为抗体产生受损和反复感染。在本研究中，我们调查了印度患者中CVID的临床和免疫学特征，并开发了一种用于预测疾病严重程度的机器学习模型。我们回顾性分析了印度一家三级医疗中心在十年间诊断为CVID的150例患者。诊断的中位年龄为18岁，男性占优势（62%）。大多数患者（66.6%）具有严重表型，反复呼吸道感染是最常见的临床表现（84.2%）。45%的患者观察到胃肠道并发症，21%的患者出现自身免疫表现。所有患者均表现为低丙种球蛋白血症。IgA水平各不相同，7.8%正常，14.5%无法检测到。85.5%的患者IgM水平降低。B细胞分析显示，64.4%的患者类别转换记忆B细胞减少，21.7%的患者水平极低。9例成年患者出现迟发性联合免疫缺陷。对52例患者进行了基因检测，在29例儿科患者和15例成年患者中确定了潜在的单基因病因。LRBA缺陷是最常见的基因缺陷，在7例儿科患者和3例成年患者中发现。我们为CVID患者开发了一种基于机器学习的新型严重程度预测模型，利用易于获得的淋巴细胞亚群、类别转换记忆B细胞计数和血清免疫球蛋白水平，使用阿梅拉图加的临床严重程度评分提供一种可及且可靠的工具来预测疾病严重程度。随机森林在所有指标上均优于其他模型，准确率达到0.853（95%CI：0.840 - 0.866）。所有模型的特征重要性分析确定Th-Tc比值、CD19和IgM水平是严重程度预测中最具影响力的预测因子。我们的研究突出了印度患者中CVID多样的临床和免疫学特征，强调了早期诊断和个体化管理策略的必要性。使用常见免疫参数开发的机器学习模型为预测疾病严重程度提供了一种强大的工具，可能指导治疗策略以改善患者预后。
