Processing speed accounts for the discriminative power of a brief smartphone Go/No-Go task among individuals with multiple sclerosis

Background Identifying and longitudinally measuring cognitive decline remains challenging because traditional assessments rely largely on episodic, in-person administration of analog scales by trained examiners. Smartphone technology offers a scalable alternative for extracting distinct digital biomarkers from continuous sensor and event streams. Following the integration of several cognitive modules into the self-administered Neurological Functions Test Suite (NeuFun-TS), this study evaluates the clinical utility and incremental validity of a final candidate module: a brief (30 s per hand) equiprobable (50/50) Go/No-Go (50/50GNG) choice-reaction task. Methods We analyzed 3,270 quality-controlled 30-s runs from 303 participants, with primary analyses focused on the first run per hand (488 observations; 18,650 stimuli) across 34 healthy donors (HDs) and 213 multiple sclerosis (MS) participants. From the raw event stream, we screened 16 candidate metrics and retained 10 based on their psychometric properties. Using these 10 metrics, we evaluated diagnostic group separation, test–retest reliability, correlations with clinical and magnetic resonance imaging (MRI) benchmarks, and incremental validity over existing digital modules: the randomized Symbol Digit Modalities Test (rSDMT) and the Motor Sequencing Test (MST), using paired observations. Results Although nine of the 10 outcomes significantly separated HD from MS and seven separated relapsing–remitting from progressive MS subtypes (all adjusted p < 0.05), discriminative power was driven almost entirely by response latency rather than by commission errors or inhibitory metrics. Response latency accounted for the majority of its full-run separation within the first 12 s. Multi-predictor composite scores offered no diagnostic advantage over reaction time alone. Furthermore, the Go/No-Go task showed poor-to-moderate longitudinal test–retest reliability and did not provide incremental clinical value over existing modules. Conclusion In this brief implementation, the smartphone Go/No-Go task served as a choice-reaction-time measure reflecting processing-speed impairment in MS; however, its discriminative capability was redundant with reaction-time metrics already implemented on the platform, so the module was excluded from NeuFun-TS. These findings do not generalize to longer continuous performance tests, whose error-based measures require many more stimuli than a 30-s run provides.