Unsupervised Analysis Applied to a Cohort of Likely Acquired Neutropenia: Is It a Unique Entity ?

INTRODUCTION. According to the recent Neutropenia Classification, a new provisional category has been added to the classical Congenital and Acquired ones named Likely Acquired Neutropenia (LAN). ¹ As described elsewhere, ² LAN is usually characterized by a mild phenotype and a peculiar immunological pattern resembling that seen in some autoimmune disorders like Sjogren syndrome and Reumathoid Arthritis.² LAN in pediatric age includes either Long Lasting (LL) (diagnosis < 3 years of age and disease lasting more than 3 years) or Late Onset (LO) (diagnosis at age ≥3 years with duration of neutropenia beyond 12 months) neutropenias (np). In a subset of patients, genetic background related to immune-dyregulations were retrieved. It is not actually known whether LAN is a homogeneous group or if the category includes different phenotypes/diseases.

AIM of the STUDY. To apply extensive Machine Learning (ML) analysis on a large group of LAN patients to delineate possible cluster by matching clinical, immunological and genetic features.

PATIENTS AND METHODS: Demographic, clinical, immunological and genetic data (NGS panel of 160 genes of Inborn Error of Immunity-IEI- and Bone Marrow Failure) for a total of “58 features” were retrieved from patients of the Italian Registry after consensus according to the Helsinki declaration. Unsupervised machine learning method, including K-prototype clustering analysis³ and three biclustering methods, such as Spectral, Cheng and Church biclustering, and Plaid⁴, were utilized to group together similar patients based on their feature values.Machine learning procedures were implemented using ClustMixType R package for clustering and the Biclust package for biclustering techniques. We used silhouette analysis to find the optimal number of clusters or biclusters. Fisher’s exact test was used to ascertain the association between the type of neutropenia and cluster. Associations with a p-value <0.05 were considered statistically significant.

RESULTS Eighty-three patients (46 females, 55%) were considered eligible for the study. Median age at onset were 11 years (IQR 2.3-15.6) and median time of follow up was 5.2 years (IQR2.7-7.9). Sixty (71%) subjects were defined as LO, and 24 (29%) were defined as LL np. Thirthy-six patients (43%) had positive antibodies against Neutrophils (AbN), while in 38 (46%) AbN were not detectable. In 9 subjects not enough samples were tested for AbN to confirm or to exclude their presence. In 9/70 (13%) subjects variants of IEI were found. The immunological pattern were characterized by a reduction of B memory (44% of the cohort), B switched memory cells (53%) and T regulatory cells (72%) whereas T γ/δ (64%), HLADR+ T cells (44%) and Double Negative B cells (56%) were increased.

No significant differences regarding clinical, immunological and genetic features were seen between the subgroups of patients with positive vs negative AbN and with LO vs LL. Most importantly, the powerful biclustering analysis, did not reveal any additional clusters related to combination of features thus suggesting that all subgroups of LAN were not likely to be different diseases.

CONCLUSIONS

By applying sophisticated ML tools it looks that AbN positive or negative and LL or LO neutropenia patients are not likely to be different disease and that these phenotypes can all be gathered within the new provisional LAN category.

Implementation of this cohort and analyses focused on immunology and genetics will be warranted to better characterize LAN

REFERENCES

1.Fioredda F, Skokowa J, Tamary H, et al . The European Guidelines on Diagnosis and Management of Neutropenia in Adults and Children: A Consensus Between the European Hematology Association and the EuNet-INNOCHRON COST Action. Hemasphere 2023;7:e872.

2.Fioredda F, Beccaria A, Casartelli P, et al . Late-onset and long-lasting neutropenias in the young: A new entity anticipating immune-dysregulation disorders. Am J Hematol 2024;99:534-542

3.Aschenbruck R, Szepannek G. Cluster validation for mixed-type data. Archives of Data Science,2020 Series A, 6(1), p.02.

4. Castanho EN, Aidos H, Madeira SC. Biclustering data analysis: a comprehensive survey. Briefings in Bioinformatics 2024;25 :bbae34225(4).