pyAKI - An Open Source Solution to Automated KDIGO classification

This data is the result of the experiment used in the validation experiment of pyAKI (https://github.com/aidh-ms/pyAKI). PyAKI is an open source software used for determining Acute Kidney Injury (AKI) stages accroding to the Kidney Injury Improving Global Outcomes (KDIGO) criteria. The data is extracted from the MIMIC Demo dataset and was then labelled by physicians and our software alike and used for comparison between both labels. The MIMIC demo dataset is a subset of the full MIMIC dataset which can be shared under the Open Data Commons Open Database License. The full MIMIC dataset is not open for publication and requires special training and identification to be worked with.

For further details on the methodology we kindly refer to our publication which can be found in the GitHub repository, mentioned above. The columns contain the following data:
- stay_id: int, unique identificator for each patient
- charttime: datetime, arbitrary date that identifies each point in time during the ICU stay in an hourly frequence
- urineoutput: float, the hourly urine output in ml
- human_urineoutput_stage: int, the AKI stage according to urine output, determined by a physician
- machine_urineoutput_stage: int, the AKI stage according to urine output, determined by pyAKI
- urineoutput_stage_vote: int, in case of discrepancy or incorrectness of both labels, the decision which AKI stage according to urine output was correct, or an alternative determined by experts
- urineoutput_stage_comment: string, free text comment explaining the reasoning behind a stage correction
- creat: float, the creatinine values in mg/dl
- human_abs_creat_stage: int, the AKI stage according to absolute creatinine criterion, determined by a physician
- machine_abs_creat_stage: int, the AKI stage according to absolute creatinine criterion, determined by pyAKI
- abs_creat_stage_vote: int, in case of discrepancy or incorrectness of both labels, the decision which AKI stage according to absolute creatinine criterion was correct, or an alternative determined by experts
- abs_creat_stage_comment: string, free text comment explaining the reasoning behind a stage correction
- human_rel_creat_stage: int, the AKI stage according to relative creatinine criterion, determined by a physician
- machine_rel_creat_stage: int, the AKI stage according to relative creatinine criterion, determined by pyAKI
- rel_creat_stage_vote: int, in case of discrepancy or incorrectness of both labels, the decision which AKI stage according to relative creatinine criterion was correct, or an alternative determined by experts
- rel_creat_stage_comment: string, free text comment explaining the reasoning behind a stage correction
- weight: float, the weight of each patient
- rrt_status: bool, the status of renal replacementtherapy (RRT), 0 or NaN indicating no RRT ist present, 1 indicating RRT is present
- human_rrt_stage: int, AKI stage according to dialysis criterion, determined by physicians
- machine_rrt_stage: int, AKI stage according to dialysis criterion, determined by pyAKI
- rrt_stage_vote: int, in case of discrepancy or incorrectness of both labels, the decision which AKI stage according to dialysis criterion was correct, or an alternative determined by experts
- rrt_stage_comment: string, free text comment explaining the reasoning behind a stage correction
- human_stage: the overall AKI stage, determined by physicians
- machine_stage: the overall AKI stage, determined by physicians
- stage_vote: int, in case of discrepancy or incorrectness of both labels, the decision which AKI overall stage was correct, or an alternative determined by experts
- stage_comment: string, free text comment explaining the reasoning behind a stage correction

本数据集为用于验证pyAKI（https://github.com/aidh-ms/pyAKI）的实验结果。PyAKI是一款开源软件，可依据全球改善肾脏病预后委员会（Kidney Injury Improving Global Outcomes, KDIGO）标准判定急性肾损伤（Acute Kidney Injury, AKI）分期。本数据集提取自MIMIC Demo数据集，随后分别由临床医师与本软件进行标注，用于对比两组标注结果。MIMIC Demo数据集是完整MIMIC数据集的子集，可在开放数据共用开放数据库许可（Open Data Commons Open Database License）下共享；完整MIMIC数据集未对外开放，使用前需接受专项培训并完成身份认证。

若需了解方法学的更多细节，请参阅上述GitHub仓库中收录的相关论文。

数据集各字段含义如下：
- stay_id：整数型，每位患者的唯一标识符
- charttime：日期时间型，以小时为频率标识重症监护病房（Intensive Care Unit, ICU）住院期间各时间点的时间戳
- urineoutput：浮点型，每小时尿量，单位为毫升（ml）
- human_urineoutput_stage：整数型，由医师依据尿量判定的AKI分期
- machine_urineoutput_stage：整数型，由pyAKI依据尿量判定的AKI分期
- urineoutput_stage_vote：整数型，当两组标注存在分歧或存在标注错误时，由专家判定的正确尿量相关AKI分期，或经专家调整后的替代分期
- urineoutput_stage_comment：字符串型，用于解释分期调整依据的自由文本注释
- creat：浮点型，肌酐值，单位为毫克每分升（mg/dl）
- human_abs_creat_stage：整数型，由医师依据血清肌酐绝对值标准判定的AKI分期
- machine_abs_creat_stage：整数型，由pyAKI依据血清肌酐绝对值标准判定的AKI分期
- abs_creat_stage_vote：整数型，当两组标注存在分歧或存在标注错误时，由专家判定的正确血清肌酐绝对值相关AKI分期，或经专家调整后的替代分期
- abs_creat_stage_comment：字符串型，用于解释分期调整依据的自由文本注释
- human_rel_creat_stage：整数型，由医师依据血清肌酐相对变化标准判定的AKI分期
- machine_rel_creat_stage：整数型，由pyAKI依据血清肌酐相对变化标准判定的AKI分期
- rel_creat_stage_vote：整数型，当两组标注存在分歧或存在标注错误时，由专家判定的正确血清肌酐相对变化相关AKI分期，或经专家调整后的替代分期
- rel_creat_stage_comment：字符串型，用于解释分期调整依据的自由文本注释
- weight：浮点型，患者体重
- rrt_status：布尔型，肾脏替代治疗（Renal Replacement Therapy, RRT）状态：0或空值（NaN）代表未实施肾脏替代治疗，1代表已实施肾脏替代治疗
- human_rrt_stage：整数型，由医师依据透析标准判定的AKI分期
- machine_rrt_stage：整数型，由pyAKI依据透析标准判定的AKI分期
- rrt_stage_vote：整数型，当两组标注存在分歧或存在标注错误时，由专家判定的正确透析相关AKI分期，或经专家调整后的替代分期
- rrt_stage_comment：字符串型，用于解释分期调整依据的自由文本注释
- human_stage：整数型，由医师判定的整体AKI分期
- machine_stage：整数型，由医师判定的整体AKI分期
- stage_vote：整数型，当两组标注存在分歧或存在标注错误时，由专家判定的正确整体AKI分期，或经专家调整后的替代分期
- stage_comment：字符串型，用于解释分期调整依据的自由文本注释