EVENT DOSSIER
The researchers proposed the DeMMO framework, aimed at addressing practical challenges in modeling longitudinal digital mobility outcomes across diseases. This framework achieves a mechanism for selective information sharing between mutually exclusive queues without the need for paired participants by directly inferring symbolic relationships from learned longitudinal mappings. The research team evaluated the framework on the Mobilise-D dataset, which included longitudinal observations, various clinical outcomes, and four mutually exclusive queues. Results showed that DeMMO achieved the best overall predictive performance in eight strong benchmark models and outperformed baseline models in most individual outcomes. Additionally, the framework is interpretable; it identifies reliable longitudinal digital mobility patterns that can guide subsequent clinical validation, while meeting the needs of modeling time evolution within a single disease and joint modeling across multiple diseases.
DeMMO 框架提出并解决了跨疾病纵向数字移动性结果建模问题。该研究首次定义并探讨了这一实践难题，旨在同时满足单病种内时间演变与多病种联合建模的需求。DeMMO 是一个可解释的框架，其核心技术贡献是通过从学习到的纵向映射中直接推断符号关系，实现无需配对参与者即可跨队列选择性信息共享的机制。研究人员在 Mobilise-D 数据集上进行了评估，该数据集包含纵向观察、多种临床结果及四个互斥队列。与八种强基准模型相比，DeMMO 取得了最佳整体预测性能，并在大多数个体结果上优于基线。此外，稳定性选择进一步识别了可指导后续临床验证的可靠纵向数字移动性模式。