GlucoPRISM: Protocol-Supervised Pretraining for Cross-Cohort Glucose Transfer

Computer Science > Machine Learning arXiv:2610.12528 (cs) [Submitted on 8 Oct 2026] Title: GlucoPRISM: Protocol-Supervised Pretraining for Cross-Cohort Glucose Transfer Authors: Adinath Madhavrao Dukre, Abdullah Mohammad, Yifan Lu, Eran Segal, Ahmed Metwally, Imran Razzak.

计算机科学 > 机器学习 arXiv:2610.12528 (cs) [提交于 2026 年 10 月 8 日] 标题：GlucoPRISM：用于跨队列葡萄糖迁移的协议监督预训练 作者：Adinath Madhavrao Dukre, Abdullah Mohammad, Yifan Lu, Eran Segal, Ahmed Metwally, Imran Razzak。

Abstract: Continuous glucose monitoring (CGM) foundation models encode each day as one vector that mixes who the person is, what happened that day and how the sensor measured it, so measurement differences travel with the representation into new cohorts. We ask whether pretraining can place the measurement component that harms transfer in known coordinates, so that deployment drops them with a fixed projection and no target data.

摘要：连续葡萄糖监测（CGM）基础模型将每一天编码为一个向量，其中混合了个人身份、当日发生事件以及传感器测量方式，因此测量差异会随表示一同进入新的队列。我们探讨预训练是否能将损害迁移的测量成分置于已知坐标中，从而在部署时通过固定投影将其剔除，且无需目标数据。

GlucoPRISM supervises a GlucoFM backbone with response contracts, rules for how each part of the representation should change when only the day or only the sensor changes: a 64-dimensional Trait band should change with neither, a 48-dimensional State band only across days, and a 16-dimensional Sensor band should track the device. Synthetic sensor views calibrated on paired Dexcom and Libre recordings, different-day pairs and glucose-summary targets supply this supervision without clinical labels.

GlucoPRISM 通过响应契约监督 GlucoFM 主干，这些契约规定了当仅日期或仅传感器发生变化时，表示的各部分应如何改变：64 维的特征带（Trait band）应保持不变，48 维的状态带（State band）仅随日期变化，而 16 维的传感器带（Sensor band）应追踪设备。在配对的 Dexcom 和 Libre 记录、不同日期对以及葡萄糖汇总目标上校准的合成传感器视图，在无需临床标签的情况下提供了这种监督。

Across twelve cross-cohort directions, dropping the Sensor band from the same frozen encoder raises mean transfer AUROC from 68.75 to 73.57, against 69.3 for GlucoFM. The gain has two parts. Into CGMacros, the one target cohort recorded with two sensor types, filtering adds 11 to 19 points, and this holds when the generator is recalibrated on half of the participants and scored on the other half.

在十二个跨队列方向上，从相同的冻结编码器中剔除传感器带，使平均迁移 AUROC 从 68.75 提高到 73.57，而 GlucoFM 为 69.3。这种增益包含两部分。在 CGMacros（唯一记录了两种传感器类型的目标队列）中，过滤增加了 11 到 19 个点，当生成器在半数参与者上重新校准并在另一半上评分时，这一结果依然成立。

On the other eight directions filtering is neutral and the encoder itself exceeds GlucoFM by 3.5 points. The device remains linearly decodable after filtering, so the filter removes a device-related component that harms transfer rather than all measurement information. The retained bands improve window AUROC on thirteen of fourteen public task—cohort cells (Holm-adjusted p=0.031) and correlation on fourteen of fifteen clinical endpoints in an external cohort of 9,303 participants.

在其他八个方向上，过滤效果中性，且编码器本身比 GlucoFM 高出 3.5 个点。过滤后设备仍然是线性可解码的，因此过滤器移除的是损害迁移的设备相关成分，而非所有测量信息。保留的频带在十四个公共任务-队列单元中的十三个上提高了窗口 AUROC（Holm 校正 p=0.031），并在一个包含 9,303 名参与者的外部队列中，在十五个临床终点中的十四个上提高了相关性。