Genotype-based severity scoring system in Wolfram Syndrome: correlation with onset of cardinal symptoms and

Background: Wolfram syndrome is a rare genetic disorder characterized by antibody-negative early-onset atypical diabetes mellitus, optic nerve atrophy, sensorineural hearing loss, central diabetes insipidus (arginine vasopressin deficiency), and progressive neurodegeneration, with significant variability in disease severity. We assessed the accuracy of a genotype-based severity scoring system for predicting the onset of cardinal symptoms in Wolfram syndrome. Methods: The scoring system is based on the type of WFS1 variants (in-frame or out-of-frame) and their location relative to transmembrane domains, defining six severity classes. Severity scores were assigned to 324 patients with documented onset ages for diabetes mellitus, optic atrophy, hearing loss, and central diabetes insipidus. A computational rule-based framework and genotype-only gradient-boosting machine-learning models were developed to independently evaluate the scoring system and to estimate age of onset. Results: The severity score showed a clear association with earlier onset of diabetes mellitus and, less consistently, optic atrophy, but no correlation with hearing loss or central diabetes insipidus. Patients with in-frame variants outside transmembrane domains showed milder disease, especially the WFS1 c.1672C>T (p.Arg558Cys) variant, which is associated with a mild form of Wolfram syndrome prevalent in the Ashkenazi Jewish population, whereas out-of-frame variants showed the earliest onset. Severity scores 3 and 4 deviated from the expected progression, suggesting that transmembrane-domain involvement in both alleles increases severity. Consolidation into three tiers (Mild, Moderate, Severe) yielded median diabetes mellitus onset ages of 9.0, 6.5, and 4.6 years. The computational framework reproduced the six-class annotation with 94.4% accuracy, and the engineered genotype model outperformed the original registry score (diabetes mellitus exact-sextile accuracy 29.4% vs. 19.3%). Conclusion: The genotype-based severity score is most informative for diabetes mellitus, modestly informative for optic atrophy, and not currently useful for hearing loss or central diabetes insipidus. Because the analysis relies on observed events without time-to-event censoring and the three-tier consolidation is a post hoc summary of the same dataset, results should be regarded as exploratory and hypothesis-generating rather than as a validated clinical prediction tool, while still offering useful insight into genotype-related progression of Wolfram syndrome.