Assigning Secondary Progressive Multiple Sclerosis (SPMS) course consistently is challenging as it is based on a gradual worsening in neurological disability independent of relapses. Clinical SPMS assignment may therefore vary between registries depending on clinical practice. Consequently, a comparison of SPMS between registries would benefit from an objective definition of SPMS.
To validate three different methods for classifying patients into Relapsing Remitting Multiple Sclerosis (RRMS) or SPMS, compared to the clinical assignment, in five European Multiple Sclerosis (MS) registries.
Data from MS registries in Czech Republic (11,336 patients), Denmark (10,255 patients), Germany (23,185 patients), Sweden (11,247 patients), and the United Kingdom (UK) (5,086 patients) were used. Patients with either RRMS or SPMS, age ≥ 18 years at index date (date with the latest Expanded Disability Status Scale (EDSS) observation) were included. Index period was 01/2017 - 12/2019. Three EDSS centric classification methods were applied; method 1: a modified real world EXPAND criteria (Kappos, L. et al., 2018. The Lancet 391(10127), 2018), method 2: the data-derived definition from Melbourne University but without pyramidal Functional Score (Lorscheider, J. et al., 2016. Brain 139(9)), method 3: the decision tree classifier from Karolinska Institutet (Ramanujam, R. et al., 2020. medRxiv, 2020.07.09.20149674). The classifications were compared to the clinical assignment, where sensitivity (SPMS as true positive), specificity (RRMS as true negative) and accuracy were calculated as similarity measurements.
The overall classification performance (sensitivity, specificity, accuracy) among classifiable patients were; method 1: (0.47, 0.85, 0.79), method 2: (0.77, 0.87, 0.85), method 3: (0.84, 0.83, 0.84). The proportions of unclassifiable patients with each method were; method 1: 20.0%, method 2: 32.2%, method 3: 0%. Methods 2 & 3 provided a high sensitivity, specificity and accuracy, while method 1 provided high specificity but low sensitivity. Method 3 was the only method having no unclassifiable patients.
Our findings suggest that these methods can be used to objectively assign SPMS with a fairly high performance in different registries. The method of choice depends on the research question and to what degree unclassifiable patients are tolerable.