MRS CLEAN: A Synthetic Tabular Dataset of Stroke Patients based on the MR CLEAN Trials

MRS CLEAN This repository contains a synthetic tabular dataset of stroke patients based on the MR CLEAN Registry and MR CLEAN-LATE, MR CLEAN-MED and MR CLEAN-NOIV trials. We evaluated multiple synthetic data generation methods and found that TabDDPM works best, generating data that is statistically similar to the real data. On a functional outcome prediction task, prediction models trained on synthetic data show similar performance to models trained on real data when evaluating on a (real) held out test set. For more information on the development of the TabDDPM model, we refer to our preprint. Variable Codebook: Variable codename Human‑readable name Datatype Possible values Interpretation Distribution StudySubjectID Study subject identifier string S0000–S14999 Unique patient identifier n=15,000, all unique r_age Age (years) integer 25–93 Age at stroke onset 72 (64–81) r_sex Sex binary 0, 1 0 = male, 1 = female 0: 48.7%, 1: 51.3% bl_NIHSS_sumscore Baseline NIHSS score integer 0–42 National Institutes of Health Stroke Scale; stroke severity (higher = worse) 15 (11–20) bl_phy_rrsyst Baseline systolic blood pressure (mmHg) integer ~90–250 Systolic BP at baseline 150 (134–167) bl_phy_rrdiast Baseline diastolic blood pressure (mmHg) integer ~45–185 Diastolic BP at baseline 82 (73–91) bl_lab_glucose Baseline blood glucose (mmol/L) integer 1–100 Fasting or random glucose 7 (6–8) bl_hist_is Prior ischemic stroke binary 0, 1 0 = no, 1 = yes 0: 84.2%, 1: 15.8% bl_hist_ht Hypertension binary 0, 1 0 = no, 1 = yes 0: 48.5%, 1: 51.5% bl_hist_af Atrial fibrillation binary 0, 1 0 = no, 1 = yes 0: 78.6%, 1: 21.4% bl_hist_dm Diabetes mellitus binary 0, 1 0 = no, 1 = yes 0: 85.5%, 1: 14.5% bl_hist_mi Prior myocardial infarction binary 0, 1 0 = no, 1 = yes 0: 88.0%, 1: 12.0% bl_hist_pad Peripheral artery disease binary 0, 1 0 = no, 1 = yes 0: 93.5%, 1: 6.5% bl_hist_hc Hypercholesterolemia binary 0, 1 0 = no, 1 = yes 0: 72.4%, 1: 27.6% bl_hist_smoking Smoking status ordinal 0, 1, 2 0 = never, 1 = former, 2 = current 0: 62.1%, 1: 21.1%, 2: 16.8% bl_hist_premrs Pre-stroke modified Rankin Scale ordinal 0–5 Functional status before stroke; 0 = no symptoms, 5 = severe disability 0: 67.4%, 1: 16.0%, 2: 7.6%, 3: 5.5%, 4: 3.1%, 5: 0.4% bl_hist_apt Antiplatelet therapy at baseline binary 0, 1 0 = no, 1 = yes 0: 70.8%, 1: 29.2% bl_hist_hep Heparin at baseline binary 0, 1 0 = no, 1 = yes 0: 98.8%, 1: 1.2% bl_hist_ahtd Antihypertensive drugs at baseline binary 0, 1 0 = no, 1 = yes 0: 45.7%, 1: 54.3% bl_hist_stat Statins at baseline binary 0, 1 0 = no, 1 = yes 0: 65.8%, 1: 34.2% occlside_LeftRight Occlusion side categorical 1, 2, 3, 4 1 = left, 2 = right, 3 = both, 4 = unknown 1: 52.3%, 2: 47.4%, 3: 0.2%, 4: 0.03% bl_cta_prox_occl_loc_CatSAP Proximal occlusion location (CatSAP) categorical 0, 1, 2, 3, 4 0 = ICA, 1 = ICA‑T, 2 = M1 proximal, 3 = M1 distal, 4 = M2 0: 0.3%, 1: 4.2%, 2: 18.0%, 3: 58.4%, 4: 19.1% bl_ASPECTS Alberta Stroke Program Early CT Score ordinal 0–10 Early ischemic change on NCCT; higher = less damage; 10 = normal 10: 39.3%, 8: 19.4%, 9: 18.7%, 7: 11.2%, 6: 5.0%, 0–5: 6.4% collateral_grade Collateral grade on CTA ordinal 0, 1, 2, 3 0 = absent, 1 = poor, 2 = moderate, 3 = good 0: 4.3%, 1: 35.0%, 2: 40.7%, 3: 20.0% postetici_c Post-procedural eTICI score ordinal 0–5 Extended Thrombolysis in Cerebral Infarction; reperfusion grade 0: 13.1%, 1: 5.8%, 2: 17.6%, 3: 21.6%, 4: 10.1%, 5: 31.8% r_ivt Intravenous thrombolysis binary 0, 1, 9 0 = no, 1 = yes, 9 = missing/unknown 0: 35.5%, 1: 64.4%, 9: 0.1% TI.onsetlsw_to_groin Onset to groin puncture (minutes) integer ~38–1190 Time from last seen well to groin puncture 190 (131–249) deepest_anesthmanag Deepest anesthesia during procedure categorical 0, 1, 2, 3, 4 0 = conscious sedation, 1 = general anesthesia, 2–4 = other 0: 80.8%, 1: 19.0%, 2–4: 0.3% TI.groin_to_endproc Groin puncture to end of procedure (minutes) integer ~10–280 Procedure duration 55 (36–74) number_iat_attempts Number of intra-arterial thrombectomy attempts integer 0–17 Number of device passes 1 (0–2) balloon_used Balloon guide catheter used binary 0, 1 0 = no, 1 = yes 0: 44.6%, 1: 55.4% fu24h_NIHSS_sumscore NIHSS at 24 hours integer 0–42 Stroke severity at 24h follow-up 9 (3–15) sICH_HBC Symptomatic intracranial hemorrhage (Heidelberg) binary 0, 1 0 = no, 1 = yes; per Heidelberg Bleeding Classification 0: 95.3%, 1: 4.7% mrs_90d Modified Rankin Scale at 90 days ordinal 0–6 Primary outcome; 0 = no symptoms, 6 = death; mRS ≤2 = functional independence 0: 7.3%, 1: 15.2%, 2: 20.7%, 3: 13.9%, 4: 12.5%, 5: 6.6%, 6: 23.7% SAE_any Any serious adverse event binary 0, 1 0 = no, 1 = yes 0: 63.5%, 1: 36.5% SAE_strokeprog Stroke progression (SAE) binary 0, 1 0 = no, 1 = yes; neurological deterioration with imaging confirmation 0: 92.2%, 1: 7.8% SAE_pneu Pneumonia (SAE) binary 0, 1 0 = no, 1 = yes 0: 91.3%, 1: 8.7% Abbreviations NIHSS: National Institutes of Health Stroke Scale mRS: Modified Rankin Scale ASPECTS: Alberta Stroke Program Early CT Score CTA: Computed tomography angiography eTICI: Extended Thrombolysis in Cerebral Infarction sICH: Symptomatic intracranial hemorrhage SAE: Serious adverse event ICA: Internal carotid artery ICA‑T: Internal carotid artery terminus M1/M2: Middle cerebral artery segments Access Conditions Access to this dataset is granted on an individual basis. By requesting and receiving access, you agree to the following conditions: 1. No RedistributionYou may not share, distribute, upload, publish, sublicense, or otherwise provide the dataset—whether in full, in part, or in modified form—to any third party.This prohibition includes (but is not limited to) sharing via email, cloud services, repositories, supplementary materials, or public or private archives.' 2. Personal, Authorized Use OnlyAccess is granted only to the individual whose request is approved.You may not transfer your access or allow others to use your copy of the dataset. 3. Attribution RequirementAny publication or public output that uses or is derived from this dataset must include proper attribution (please use the citation at the right side of the page). References Pirson FAV, et al. MR CLEAN-LATE: a multicenter randomized clinical trial of endovascular treatment of acute ischemic stroke. Trials 2021;22:160. PMC7903604 von Kummer R, et al. The Heidelberg Bleeding Classification. Stroke 2015;46:2981–2986. Berkhemer OA, et al. A randomized trial of intraarterial treatment for acute ischemic stroke (MR CLEAN). N Engl J Med 2015;372:11–20.MRS CLEAN 本仓库包含基于MR CLEAN注册库及MR CLEAN-LATE、MR CLEAN-MED、MR CLEAN-NOIV三项临床试验构建的脑卒中患者合成表格型（tabular）数据集。研究团队评估了多种合成数据生成方法，最终发现TabDDPM模型效果最优，其生成的数据与真实数据在统计学层面具有高度相似性。在功能结局预测任务中，基于合成数据训练的预测模型在真实留存测试集上的表现，与基于真实数据训练的模型相当。 如需了解TabDDPM模型的开发细节，请参阅我们的预印本文章。 ### 变量编码手册 | 变量代号 | 可读名称 | 数据类型 | 可能取值 | 含义 | 分布 | | --- | --- | --- | --- | --- | --- | | StudySubjectID | 研究对象标识符 | 字符串 | S0000–S14999 | 唯一患者标识符，共15000条且全部不重复 | n=15,000，全部唯一 | | r_age | 年龄（岁） | 整数型 | 25–93 | 脑卒中发病时年龄 | 中位数72（四分位距64–81） | | r_sex | 性别 | 二分类 | 0, 1 | 0=男性，1=女性 | 男性占48.7%，女性占51.3% | | bl_NIHSS_sumscore | 基线NIHSS评分（National Institutes of Health Stroke Scale，美国国立卫生研究院卒中量表） | 整数型 | 0–42 | 卒中严重程度，分值越高病情越重 | 中位数15（四分位距11–20） | | bl_phy_rrsyst | 基线收缩压 | 整数型 | ~90–250 | 基线收缩压（mmHg） | 中位数150（四分位距134–167） | | bl_phy_rrdiast | 基线舒张压 | 整数型 | ~45–185 | 基线舒张压（mmHg） | 中位数82（四分位距73–91） | | bl_lab_glucose | 基线血糖 | 整数型 | 1–100 | 空腹或随机血糖（mmol/L） | 中位数7（四分位距6–8） | | bl_hist_is | 既往缺血性脑卒中 | 二分类 | 0, 1 | 0=否，1=是 | 0占84.2%，1占15.8% | | bl_hist_ht | 高血压 | 二分类 | 0, 1 | 0=否，1=是 | 0占48.5%，1占51.5% | | bl_hist_af | 心房颤动 | 二分类 | 0, 1 | 0=否，1=是 | 0占78.6%，1占21.4% | | bl_hist_dm | 糖尿病 | 二分类 | 0, 1 | 0=否，1=是 | 0占85.5%，1占14.5% | | bl_hist_mi | 既往心肌梗死 | 二分类 | 0, 1 | 0=否，1=是 | 0占88.0%，1占12.0% | | bl_hist_pad | 外周动脉疾病 | 二分类 | 0, 1 | 0=否，1=是 | 0占93.5%，1占6.5% | | bl_hist_hc | 高胆固醇血症 | 二分类 | 0, 1 | 0=否，1=是 | 0占72.4%，1占27.6% | | bl_hist_smoking | 吸烟状态 | 有序分类 | 0, 1, 2 | 0=从不吸烟，1=既往吸烟，2=当前吸烟 | 0占62.1%，1占21.1%，2占16.8% | | bl_hist_premrs | 卒中前改良Rankin量表（mRS，Modified Rankin Scale）评分 | 有序分类 | 0–5 | 卒中前功能状态，0=无症状，5=重度残疾 | 0: 67.4%, 1: 16.0%, 2: 7.6%, 3: 5.5%, 4: 3.1%, 5: 0.4% | | bl_hist_apt | 基线抗血小板治疗 | 二分类 | 0, 1 | 0=否，1=是 | 0占70.8%，1占29.2% | | bl_hist_hep | 基线肝素使用 | 二分类 | 0, 1 | 0=否，1=是 | 0占98.8%，1占1.2% | | bl_hist_ahtd | 基线抗高血压药物使用 | 二分类 | 0, 1 | 0=否，1=是 | 0占45.7%，1占54.3% | | bl_hist_stat | 基线他汀类药物使用 | 二分类 | 0, 1 | 0=否，1=是 | 0占65.8%，1占34.2% | | occlside_LeftRight | 闭塞侧别 | 分类变量 | 1, 2, 3, 4 | 1=左侧，2=右侧，3=双侧，4=未知 | 1占52.3%，2占47.4%，3占0.2%，4占0.03% | | bl_cta_prox_occl_loc_CatSAP | 近端闭塞部位（CatSAP分类） | 分类变量 | 0, 1, 2, 3, 4 | 0=颈内动脉（ICA，Internal carotid artery），1=颈内动脉末端（ICA-T，Internal carotid artery terminus），2=M1段近端，3=M1段远端，4=M2段 | 0占0.3%，1占4.2%，2占18.0%，3占58.4%，4占19.1% | | bl_ASPECTS | Alberta卒中项目早期CT评分（ASPECTS，Alberta Stroke Program Early CT Score） | 有序分类 | 0–10 | 非增强CT早期缺血改变，分值越高损伤越小，10=正常 | 10: 39.3%, 8: 19.4%, 9: 18.7%, 7: 11.2%, 6: 5.0%, 0–5: 6.4% | | collateral_grade | CTA侧支循环分级 | 有序分类 | 0, 1, 2, 3 | 0=无侧支循环，1=差，2=中等，3=良好 | 0占4.3%，1占35.0%，2占40.7%，3占20.0% | | postetici_c | 术后扩展脑梗死溶栓分级（eTICI，Extended Thrombolysis in Cerebral Infarction）评分 | 有序分类 | 0–5 | 再灌注分级 | 0: 13.1%, 1: 5.8%, 2: 17.6%, 3: 21.6%, 4: 10.1%, 5: 31.8% | | r_ivt | 静脉溶栓 | 二分类 | 0, 1, 9 | 0=否，1=是，9=缺失/未知 | 0占35.5%，1占64.4%，9占0.1% | | TI.onsetlsw_to_groin | 发病至股动脉穿刺时间 | 整数型 | ~38–1190 | 最后一次被观察到正常至股动脉穿刺的时间（分钟） | 中位数190（四分位距131–249） | | deepest_anesthmanag | 术中最深麻醉方式 | 分类变量 | 0, 1, 2, 3, 4 | 0=清醒镇静，1=全身麻醉，2–4=其他 | 0占80.8%，1占19.0%，2–4占0.3% | | TI.groin_to_endproc | 股动脉穿刺至手术结束时间 | 整数型 | ~10–280 | 手术时长（分钟） | 中位数55（四分位距36–74） | | number_iat_attempts | 血管内取栓尝试次数 | 整数型 | 0–17 | 器械通过次数 | 中位数1（四分位距0–2） | | balloon_used | 球囊导引导管使用情况 | 二分类 | 0, 1 | 0=否，1=是 | 0占44.6%，1占55.4% | | fu24h_NIHSS_sumscore | 24小时NIHSS评分 | 整数型 | 0–42 | 24小时随访卒中严重程度 | 中位数9（四分位距3–15） | | sICH_HBC | 症状性颅内出血（sICH，Symptomatic intracranial hemorrhage，海德堡分类） | 二分类 | 0, 1 | 0=否，1=是，采用海德堡出血分类标准 | 0占95.3%，1占4.7% | | mrs_90d | 90天改良Rankin量表（mRS，Modified Rankin Scale）评分 | 有序分类 | 0–6 | 主要结局指标，0=无症状，6=死亡；mRS≤2提示功能独立 | 0: 7.3%, 1: 15.2%, 2: 20.7%, 3: 13.9%, 4: 12.5%, 5: 6.6%, 6: 23.7% | | SAE_any | 任意严重不良事件（SAE，Serious adverse event） | 二分类 | 0, 1 | 0=否，1=是 | 0占63.5%，1占36.5% | | SAE_strokeprog | 卒中进展（严重不良事件） | 二分类 | 0, 1 | 0=否，1=是，经影像学确认的神经功能恶化 | 0占92.2%，1占7.8% | | SAE_pneu | 肺炎（严重不良事件） | 二分类 | 0, 1 | 0=否，1=是 | 0占91.3%，1占8.7% | ### 缩写说明 - NIHSS：美国国立卫生研究院卒中量表（National Institutes of Health Stroke Scale） - mRS：改良Rankin量表（Modified Rankin Scale） - ASPECTS：Alberta卒中项目早期CT评分（Alberta Stroke Program Early CT Score） - CTA：计算机断层血管造影（Computed tomography angiography） - eTICI：扩展脑梗死溶栓分级（Extended Thrombolysis in Cerebral Infarction） - sICH：症状性颅内出血（Symptomatic intracranial hemorrhage） - SAE：严重不良事件（Serious adverse event） - ICA：颈内动脉（Internal carotid artery） - ICA-T：颈内动脉末端（Internal carotid artery terminus） - M1/M2：大脑中动脉分段（Middle cerebral artery segments） ### 使用权限 本数据集需经个体申请获批后方可获取。申请并获得访问权限后，您需遵守以下条款： 1. **禁止再分发**：您不得将数据集（无论全部、部分或修改形式）共享、分发、上传、发布、转授权或许可给任何第三方。本禁令包括但不限于通过电子邮件、云服务、代码仓库、补充材料或公开/私有归档渠道进行共享。 2. **仅限个人授权使用**：访问权限仅授予申请获批的个人。您不得转让访问权限，或允许他人使用您获取的数据集副本。 3. **署名要求**：任何使用或基于本数据集产生的出版物或公开成果，均需注明恰当的来源（请使用页面右侧的引用格式）。 ### 参考文献 1. Pirson FAV, et al. MR CLEAN-LATE: a multicenter randomized clinical trial of endovascular treatment of acute ischemic stroke. *Trials* 2021;22:160. PMC7903604 2. von Kummer R, et al. The Heidelberg Bleeding Classification. *Stroke* 2015;46:2981–2986. 3. Berkhemer OA, et al. A randomized trial of intraarterial treatment for acute ischemic stroke (MR CLEAN). *N Engl J Med* 2015;372:11–20.