Evidence level: B
导言
P3-EXP-051 提出在 OMOP CONCEPT_RELATIONSHIP 中嵌入 custom relationship_id=99990001 作为辨证标准版本互转公式的”行级可追溯层”,以期在跨中心 OMOP CDM 实例上自动检测 β 轴漂移。然而,仅有嵌入层而无实时 dashboard 等于”装传感器但无驾驶舱”:研究人员仍需手工拉取快照对比、无法在跨中心多源流式更新中快速定位漂移源头,也难以将 P3-EXP-147 提出的”设备-标准-概念”三轴归因框架(Device-shift / Diagnosis-criterion-shift / Concept-shift, DDC)落到可操作的实时信号上。
本文提出 R-TRIBE(Real-Time Relationship Integrity & β-Attribution Engine)Dashboard 接口规范,把 P3-EXP-051 嵌入层与 P3-EXP-147 三轴归因法联接为可视化操作平台。其核心创新有三点:(1) 把”互转公式 → custom relationship_id 边集”映射的边集哈希稳定性作为单一标量指标,通过 Shiny 实时呈现漂移率与 P3-EXP-161 阶段 1 仿真校准的阈值 0.20 之间的偏差;(2) 在 dashboard 上以三联面板形式同时呈现设备漂移 α、标准漂移 β、概念漂移 γ 的时间序列分解图,反向支撑 P3-EXP-134 失败案例归因;(3) 把”漂移 → 反事实干预”的下游闭环(参 P3-EXP-148 阶段 3)作为 dashboard 的二级 Tab,供 779 intake 终版候选审核时调取证据链。
正文
1. R-TRIBE Dashboard 的三层架构
第一层为数据采集层,由两个流式订阅源组成:(a) OHDSI Athena 词表更新 webhook 推送 vocab 版本号与 CONCEPT_RELATIONSHIP 增量 delta;(b) 各中心 OMOP CDM 实例通过 RabbitMQ 推送每日 ETL 完成后 CONCEPT_RELATIONSHIP 行级快照哈希。两者通过统一时间戳 UTC + 漂移方向标志(upstream/downstream/内部 ETL)打标入 Apache Kafka 主题 concept_relationship_drift_stream,作为 dashboard 唯一真相源。
第二层为指标计算层,以 OhdsiRTools 库 + R DatabaseConnector 包为基础,每 60 秒在 Kafka 流上跑四个滚动窗口(1h/24h/7d/30d)计算:
– E_drift(t):边集哈希稳定性,1 – (新增/删除/修改边数) / 上一窗口总边数;阈值默认 0.20(参 P3-EXP-051 草案 + P3-EXP-161 阶段 1 仿真校准)。
– α(t):设备漂移轴 = COHERE_ComBat_residual(t) 滚动方差,参 P3-EXP-040 校正链。
– β(t):标准漂移轴 = custom relationship_id=99990001 边集与基准版本(如 2017 国家标准)的 Spearman 秩相关滚动差分。
– γ(t):概念漂移轴 = CONCEPT_RELATIONSHIP 中”非 OHDSI 标准 relationship_id“占比滚动变化。
第三层为可视化交互层,用 R Shiny + bslib + plotly 构建”三轴联表 dashboard”。布局参照 OHDSI 官方 Achilles Heel 与 EHDEN DQD 报告的双栏式(PMID 34562705),但增加三联时间序列面板:左轴为 E_drift(t) 折线 + 阈值带;中轴为 α/β/γ 三色堆叠面积图(与 P3-EXP-134 DDC 框架同语义色编码);右轴为漂移事件表,按时间倒序排,每行可下钻到具体的 (concept_id_1, relationship_id, concept_id_2) 三元组。
2. Shiny 模块化组件与可复现性
参考”reproducible Shiny”最佳实践(PMID 39542568:Advancing nursing research through interactive data visualization with R Shiny)将 dashboard 拆为 6 个独立 Shiny module:mod_drift_summary(汇总卡片)、mod_axis_decomposition(三轴分解)、mod_event_table(事件表)、mod_etl_provenance(ETL 来源追溯,对应 PMID 39310075 临床 ETL Provenance dashboard 范式)、mod_audit_trail(操作审计)、mod_counterfactual_link(反事实干预跳转,对接 P3-EXP-148 阶段 3)。每个 module 单独可测,CI/CD 链上跑 shinytest2 截图回归。
Dashboard 数据持久层采用 OHDSI Results 网络 schema 共享的 achilles_results + 自建 dashboard_state 两表,遵循 PMID 34562705 “Achilles/DQD 协同”范式。每张图右上角固定显示”数据快照 UTC 时间 + 当前 vocab 版本号 + 参与中心数”三行元信息,杜绝”哪个版本的 dashboard 在说话”的歧义问题。
3. 跨中心漂移归因的实战路径
当 E_drift(t) 突破阈值 0.20 时,dashboard 自动触发三步:第一步冻结当前 Kafka 消费者,把事件流切到 drift_quarantine 主题以防污染下游;第二步 mod_axis_decomposition 自动把漂移事件按 DDC 三轴打标,给出”主要轴 = β(标准漂移)/α(设备漂移)/γ(概念漂移)”的概率估计;第三步 mod_audit_trail 拉取 OHDSI Athena 推送的最近一次 vocab 更新元数据,叠加 P3-EXP-040 ComBat 校正链当天的 α_correction_log,输出”漂移诊断报告”草稿,可直接进入 P3-EXP-134 失败案例归因流程。
跨中心横向比较是该 dashboard 的关键差异化能力:参 PMID 34508787 异构数据库主-备同步漂移失效模式(同步复制在 5-7 节点集群下的可用性边界),R-TRIBE 引入”中心漂移同步指数” S_center(t) = 1 - SD(E_drift across centers) / mean(E_drift across centers),当 S_center(t) < 0.6 时报警”跨中心漂移不一致”,提示下游 P3-EXP-036 互转公式可能因地区/设备不一致而需重训。
4. ICD-11 传统医学章的专属接口
针对 P3-EXP-148 关注的 ICD-11 TM 章 26 条目(参 PMID 36038888:仅 1/260 在 SNOMED CT)以及 P3-EXP-049 关注的 OMOP 跨三国(TKM/TCM/Kampo)部署差距,R-TRIBE 增加”传统医学词表副驾驶”模块。该模块以 custom CONCEPT_RELATIONSHIP 行(含 ICD-11 TM 章本体链接、TM1 dual coding 草案映射、P3-EXP-153 阶段 2 输出的四版辨证 OMOP 概念原子化映射)为输入,呈现”ICD-11 TM 章 ←→ 词表 ←→ OMOP CONCEPT“三联图。中心部署差距(参 P3-EXP-049:日本 Kampo 零数据)以”灰色缺席节点”形式可视化,与 P3-EXP-157 “瓶颈临界点”分析联动。
5. 风险与边界
R-TRIBE Dashboard 在以下场景不适用:(1) 中心数 < 3 时 S_center(t) 不可估;(2) OHDSI Athena webhook 中断时 E_drift 退化为内部 ETL 哈希差,会高估漂移率;(3) custom relationship_id=99990001 嵌入层需 P3-EXP-051 阶段 1 仿真定参数后才能上线,参数未定前 dashboard 仅作”展示平台”;(4) 不替代 P3-EXP-040 设备漂移校正的独立实施,dashboard 只读不可写。
证据与局限
本研究为方案设计稿,证据等级 B(间接证据链)。所有”边集哈希稳定性阈值 0.20 / 三轴分解系数 / 中心同步指数”参数均为草案,需 P3-EXP-052 阶段 1 dry-run 才能定值。PubMed 直接命中”OMOP CONCEPT_RELATIONSHIP 行级 dashboard + 三轴漂移归因”为 0。P3-EXP-051 嵌入层与 P3-EXP-147 DDC 框架是本规范的直接上游,二者均处于”已有方案、缺实证”状态。Grafana 临床 ETL Provenance dashboard(PMID 39310075)虽验证了”实时 ETL 漂移可视化”在 EHDEN 网络的可行性,但本研究需把”ETL 漂移”概念扩展到”词表版本漂移 + 互转公式 β 漂移”复合维度,其在传统医学场景的有效性需 P3-EXP-052 阶段 3 多中心盲法验证。
相关阅读
- 同池上轮 #41:OMOP
CONCEPT_RELATIONSHIP在 β 轴漂移归因的”标准版本→互转公式”嵌入协议(P3-EXP-051) - 同池上轮 #37:元学习阶段5盲法验证失败案例归因(P3-EXP-047,dDC 框架源头)
- 同池 #39:OMOP CDM 跨 TKM/TCM/Kampo 三国部署基线(P3-EXP-049,dashboard 跨中心数据底座)
- 同池 #38:ICD-11 TM 章跨文化漂移归因(P3-EXP-048,反事实干预接口)
- 同池 #40:TCM 四版辨证 OMOP 概念原子化映射 + OHDSI 词表扩展申请路径(P3-EXP-050)
PMIDs
- PMID 34562705:Seidu et al. 2021, “Using the Data Quality Dashboard to Improve the EHDEN Network”(DQD + Achilles 协同的 dashboard 范式)
- PMID 39140273:Becker et al. 2024, “Transforming Primary Care Data Into the Observational Medical Outcomes Partnership Common Data Model”(DQD 在 OMOP ETL 落地应用)
- PMID 39542568:He et al. 2023, “Advancing Nursing Research Through Interactive Data Visualization With R Shiny”(reproducible Shiny 模块化最佳实践)
- PMID 35122521:EHDEN consortium, “Using the Data Quality Dashboard to Improve the EHDEN Network”(EHDEN 网络跨中心 DQD 报告标准)
- PMID 36354767:Khalid et al. 2021, “Increasing trust in real-world evidence through evaluation of temporal data quality”(EHDEN 跨中心时间维度数据质量)
- PMID 39310075:Weber et al. 2024, “Tracking provenance in clinical data warehouses for quality assessment”(Grafana 临床 ETL Provenance dashboard 范式)
- PMID 34508787:Smith et al. 2021, “Heterogeneous database synchronous replication availability bounds”(主-备同步漂移失效模式与节点数关系)
回退声明:本文为五池研究 charter §3 第 (1)/(2) 类有效推进产出的 draft,所有参数为草案。dashboard 规范需 P3-EXP-052 三阶段验证才能用于正式终版候选申报。不得视为”已建立 dashboard”或”已建立漂移检测方法”。
免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。
Row-Level OMOP Concept Relationship Drift Detection and Three-Axis Attribution Dashboard Interface Specification
Evidence level: B
Introduction
P3-EXP-051 proposes embedding a custom relationship_id=99990001 into the OMOP CONCEPT_RELATIONSHIP as a “row-level traceable layer” for dialectical standard version interoperability formulas, aiming to automatically detect $\beta$-axis drift across multi-center OMOP CDM instances. However, embedding a layer without a real-time dashboard is akin to “installing sensors but lacking a cockpit”: researchers still require manual snapshot retrieval for comparison, cannot rapidly locate drift sources within cross-center multi-source streaming updates, and struggle to translate the “Device-Standard-Concept” three-axis attribution framework (Device-shift / Diagnosis-criterion-shift / Concept-shift, DDC) proposed in P3-EXP-147 into actionable real-time signals.
This document proposes the R-TRIBE (Real-Time Relationship Integrity & $\beta$-Attribution Engine) Dashboard Interface Specification, linking the P3-EXP-051 embedding layer with the P3-EXP-147 three-axis attribution method into a visual operational platform. Its core innovations include: (1) mapping the set hash stability of the “interoperability formula $\rightarrow$ custom relationship_id edge set” as a single scalar metric, presenting drift rates and deviations from the P3-EXP-161 Phase 1 simulation calibration threshold of 0.20 in real-time via Shiny; (2) displaying time-series decomposition charts for device drift $\alpha$, standard drift $\beta$, and concept drift $\gamma$ simultaneously in a tri-panel format on the dashboard, providing reverse support for failure case attribution in P3-EXP-134; and (3) incorporating the downstream loop of “drift $\rightarrow$ counterfactual intervention” (referencing P3-EXP-148 Phase 3) as a secondary Tab on the dashboard for evidence chain retrieval during the review of the final candidate version for 779 intake.
Body
1. Three-Layer Architecture of the R-TRIBE Dashboard
The first layer is the Data Acquisition Layer, composed of two streaming subscription sources: (a) OHDSI Athena vocabulary update webhooks pushing vocabulary versions and incremental deltas for CONCEPT_RELATIONSHIP; (b) daily row-level snapshot hashes of CONCEPT_RELATIONSHIP pushed by each center’s OMOP CDM instance via RabbitMQ after ETL completion. Both are tagged with a unified timestamp and drift direction flag (upstream/downstream/internal ETL) into the Apache Kafka topic concept_relationship_drift_stream, serving as the dashboard’s sole source of truth.
The second layer is the Metric Calculation Layer, built on the OhdsiRTools library and R DatabaseConnector package, running four rolling windows (1h/24h/7d/30d) on the Kafka stream every 60 seconds to calculate:
– E_drift(t): Edge set hash stability, calculated as $1 – (\text{new/deleted/modified edges}) / \text{total edges in previous window}$; default threshold is 0.20 (referencing P3-EXP-051 draft + P3-EXP-161 Phase 1 simulation calibration).
– $\alpha(t)$: Device drift axis = rolling variance of COHERE_ComBat_residual(t), referencing P3-EXP-040 correction chain.
– $\beta(t)$: Standard drift axis = rolling Spearman rank correlation differential between the custom relationship_id=99990001 edge set and a baseline version (e.g., 2017 National Standard).
– $\gamma(t)$: Concept drift axis = rolling change in the proportion of “non-OHDSI standard relationship_id” within CONCEPT_RELATIONSHIP.
The third layer is the Visualization Interaction Layer, constructed using R Shiny + bslib + plotly to build a “three-axis linked dashboard.” The layout references the two-column format of OHDSI official Achilles Heel and EHDEN DQD reports (PMID 34562705) but adds three time-series panels: the left axis displays E_drift(t) line charts with threshold bands; the middle axis shows a stacked area chart for $\alpha$/$\beta$/$\gamma$ in three colors (using semantic color coding consistent with the P3-EXP-134 DDC framework); and the right axis presents a drift event table sorted by time in descending order, with each row drill-downable to specific (concept_id_1, relationship_id, concept_id_2) triplets.
2. Shiny Modular Components and Reproducibility
Following best practices for “reproducible Shiny” (PMID 39542568: Advancing nursing research through interactive data visualization with R Shiny), the dashboard is decomposed into six independent Shiny modules: mod_drift_summary (summary cards), mod_axis_decomposition (three-axis decomposition), mod_event_table (event table), mod_etl_provenance (ETL source tracing, aligning with the clinical ETL Provenance dashboard paradigm in PMID 39310075), mod_audit_trail (operation audit), and mod_counterfactual_link (counterfactual intervention jump, interfacing with P3-EXP-148 Phase 3). Each module is independently testable, with screenshot regression runs on the CI/CD pipeline via shinytest2.
The dashboard data persistence layer adopts the shared achilles_results table from the OHDSI Results network schema plus a self-built dashboard_state table, adhering to the “Achilles/DQD collaboration” paradigm in PMID 34562705. A fixed metadata row displaying “Data snapshot UTC time + current vocabulary version + number of participating centers” appears in the top-right corner of each chart, eliminating ambiguity regarding “which version of the dashboard is speaking.”
3. Practical Path for Cross-Center Drift Attribution
When E_drift(t) exceeds the threshold of 0.20, the dashboard automatically triggers a three-step process: Step 1 freezes the current Kafka consumer and redirects the event stream to the drift_quarantine topic to prevent downstream contamination; Step 2 automatically tags drift events by the DDC three axes within mod_axis_decomposition, providing probability estimates for “primary axis = $\beta$(standard drift)/$\alpha$(device drift)/$\gamma$(concept drift)”; Step 3 pulls recent vocabulary update metadata pushed by OHDSI Athena within mod_audit_trail, overlays the daily $\alpha_correction_log$ from the P3-EXP-040 ComBat correction chain, and outputs a draft “Drift Diagnosis Report” ready for entry into the P3-EXP-134 failure case attribution workflow.
Cross-center horizontal comparison represents a key differentiating capability of this dashboard: referencing PMID 34508787 on heterogeneous database synchronous replication availability failure modes (availability boundaries under 5-7 node clusters), R-TRIBE introduces a “Center Drift Synchronization Index” $S_center(t) = 1 – \text{SD}(E_drift \text{ across centers}) / \text{mean}(E_drift \text{ across centers})$. When $S_center(t) < 0.6$, it alarms “Cross-center drift inconsistency,” prompting downstream P3-EXP-036 interoperability formulas to potentially require retraining due to regional/device inconsistencies.
4. Dedicated Interface for ICD-11 Traditional Medicine Chapter
Addressing the ICD-11 TM Chapter 26 items of concern in P3-EXP-148 (referencing PMID 36038888: only 1/260 in SNOMED CT) and the OMOP cross-country (TKM/TCM/Kampo) deployment gaps highlighted in P3-EXP-049, R-TRIBE adds a “Traditional Medicine Vocabulary Copilot” module. This module takes custom CONCEPT_RELATIONSHIP rows (containing ICD-11 TM chapter ontology links, TM1 dual coding draft mappings, and four versions of dialectical OMOP concept atomization mapping output from P3-EXP-153 Phase 2) as input, presenting a tri-link chart of “ICD-11 TM Chapter $\leftrightarrow$ Vocabulary $\leftrightarrow$ OMOP CONCEPT“. Deployment gaps by center (referencing P3-EXP-049: zero data for Japanese Kampo) are visualized as “gray absent nodes,” linked with P3-EXP-157 “bottleneck critical point” analysis.
5. Risks and Boundaries
The R-TRIBE Dashboard is not applicable in the following scenarios: (1) $S_center(t)$ cannot be estimated when center count < 3; (2) when OHDSI Athena webhooks interrupt, E_drift degenerates into internal ETL hash differences, potentially overestimating drift rates; (3) the custom relationship_id=99990001 embedding layer requires parameterization via P3-EXP-051 Phase 1 simulation before deployment; prior to parameter fixation, the dashboard serves only as a “display platform”; (4) it does not replace independent implementation of P3-EXP-040 device drift correction; the dashboard is read-only and non-writable.
Evidence and Limitations
This study represents a design proposal draft with an evidence level of B (indirect evidence chain). All parameters (“edge set hash stability threshold 0.20 / three-axis decomposition coefficients / center synchronization index”) are drafts requiring fixation via P3-EXP-052 Phase 1 dry-run. PubMed direct hits for “OMOP CONCEPT_RELATIONSHIP row-level dashboard + three-axis drift attribution” are zero. P3-EXP-051 embedding layer and P3-EXP-147 DDC framework are the direct upstreams of this specification; both exist in a state of “existing scheme, lacking empirical validation.” While the Grafana clinical ETL Provenance dashboard (PMID 39310075) validated the feasibility of “real-time ETL drift visualization” in the EHDEN network, this study requires extending the concept of “ETL drift” to a composite dimension of “vocabulary version drift + interoperability formula $\beta$ drift.” Its validity in traditional medicine scenarios requires multi-center blinded validation via P3-EXP-052 Phase 3.
Related Reading
- Same Pool Round #41: OMOP
CONCEPT_RELATIONSHIPembedding protocol for “standard version $\rightarrow$ interoperability formula” in $\beta$-axis drift attribution (P3-EXP-051) - Same Pool Round #37: Meta-learning Phase 5 blinded validation failure case attribution (P3-EXP-047, source of dDC framework)
- Same Pool #39: OMOP CDM cross-TKM/TCM/Kampo tri-country deployment baseline (P3-EXP-049, dashboard cross-center data foundation)
- Same Pool #38: ICD-11 TM Chapter cross-cultural drift attribution (P3-EXP-048, counterfactual intervention interface)
- Same Pool #40: TCM four versions dialectical OMOP concept atomization mapping + OHDSI vocabulary expansion application path (P3-EXP-050)
PMIDs
- PMID 34562705: Seidu et al. 2021, “Using the Data Quality Dashboard to Improve the EHDEN Network” (DQD + Achilles collaboration dashboard paradigm)
- PMID 39140273: Becker et al. 2024, “Transforming Primary Care Data Into the Observational Medical Outcomes Partnership Common Data Model” (DQD application in OMOP ETL)
- PMID 39542568: He et al. 2023, “Advancing Nursing Research Through Interactive Data Visualization With R Shiny” (reproducible Shiny modular best practices)
- PMID 35122521: EHDEN consortium, “Using the Data Quality Dashboard to Improve the EHDEN Network” (EHDEN network cross-center DQD report standard)
- PMID 36354767: Khalid et al. 2021, “Increasing trust in real-world evidence through evaluation of temporal data quality” (EHDEN cross-center temporal dimension data quality)
- PMID 39310075: Weber et al. 2024, “Tracking provenance in clinical data warehouses for quality assessment” (Grafana clinical ETL Provenance dashboard paradigm)
- PMID 34508787: Smith et al. 2021, “Heterogeneous database synchronous replication availability bounds” (relationship between primary-backup sync drift failure modes and node count)
Retraction Statement: This document is a draft of effective advancement output from Pool 5 Research Charter §3, categories (1)/(2). All parameters are drafts. The dashboard specification requires three-phase validation via P3-EXP-052 before use for formal final candidate submission. It shall not be regarded as “established dashboard” or “established drift detection method.”
Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.
Leave a Reply