Data dictionary v1.0.1

Dataset Changes

File structure reorganization (June 2022)

1. Waveform files were moved out of ecg-metadata and into ecg-waveform.
   .../ecg-metadata/waveform* -> .../ecg-waveform/waveform*
2. Control group cohort.csv and ecg.csv were joined and concatenated between years. .../control-group/{year}/cohort.csv & .../control-group/{year}/ecg.csv -> .../control-group/ecg-cohort.csv
3. Typos were fixed in the columns of rosc-outcomes.csv. icd_discharge_datetime_offset -> icu_discharge_datetime_offset
icd_discharge_datetime_is_weekend -> icu_discharge_datetime_is_weekend

File Tree Changes

.
└── arrest-ntuh-ecg
   ├── control-group
+  │   ├── ecg-cohort.csv
   │   ├── 2015 (same changes for all years)
-  │   │   ├── cohort.csv
   │   │   ├── control (same changes for control and pre)
   │   │   │   └── ecg-metadata
   │   │   │       ├── diagnosis.csv
   │   │   │       ├── lead-data.csv
   │   │   │       ├── measurement-matrix-global.csv
   │   │   │       ├── measurement-matrix-per-lead-bitflg.csv
   │   │   │       ├── measurement-matrix-per-lead.csv
   │   │   │       ├── measurement-matrix-per-lead-definitions.csv
   │   │   │       ├── muse-info.csv
   │   │   │       ├── original-diagnosis.csv
   │   │   │       ├── original-resting-ecg-measurements.csv
   │   │   │       ├── pharma-data.csv
   │   │   │       ├── qrs-times-types.csv
   │   │   │       ├── qrs-times-types-qrs.csv
   │   │   │       ├── resting-ecg-measurements.csv
-  │   │   │       ├── waveform.csv
-  │   │   │       ├── waveform-median.npy
-  │   │   │       ├── waveform-npy.csv
-  │   │   │       └── waveform-rhythm.npy
+  │   │   │   └── ecg-waveforms
+  │   │   │       ├── waveform.csv
+  │   │   │       ├── waveform-median.npy
+  │   │   │       ├── waveform-npy.csv
+  │   │   │       └── waveform-rhythm.npy
-  │   │   ├── ecg.csv
   │   │   └── pre
   │   │       └── ... 
   │   ├── 2016
   │   │   └── ...     
   │   ├── 2017
   │   │   └── ...
   │   ├── 2018
   │   │   └── ...
   │   └── 2019
   │       └── ...
   └── study-group
       ├── cohort.csv
       ├── comorbidities.csv
       ├── ecg.csv
       ├── ecg-metadata
       │   ├── amplitude-measurements.csv
       │   ├── diagnosis.csv
       │   ├── interval-measurements.csv
       │   ├── lead-data.csv
       │   ├── measurement-matrix-global.csv
       │   ├── measurement-matrix-per-lead-bitflg.csv
       │   ├── measurement-matrix-per-lead.csv
       │   ├── measurement-matrix-per-lead-definitions.csv
       │   ├── muse-info.csv
       │   ├── original-diagnosis.csv
       │   ├── original-resting-ecg-measurements.csv
       │   ├── pharma-data.csv
       │   ├── qrs-times-types.csv
       │   ├── qrs-times-types-qrs.csv
       │   ├── resting-ecg-measurements.csv
-      │   ├── waveform.csv
-      │   ├── waveform-median.npy
-      │   ├── waveform-npy.csv
-      │   └── waveform-rhythm.npy
+      ├── ecg-waveforms
+      │   ├── waveform.csv
+      │   ├── waveform-median.npy
+      │   ├── waveform-npy.csv
+      │   └── waveform-rhythm.npy
       └── rosc-outcomes.csv

Note

The dates in this dataset have been shifted by a random number of days. Also, they have been shifted distantly into the future to make it clear that these are not the actual dates. All dates for a patient have been shifted by the same amount in order to preserve the time duration between events.

File tree

.
└── arrest-ntuh-ecg
    └── v1
        ├── control-group
        │   ├── ecg-cohort.csv
        │   ├── 2015
        │   │   ├── control
        │   │   │   ├── ecg-metadata
        │   │   │   │   └── ...
        │   │   │   └── ecg-waveforms    
        │   │   │       ├── waveform.csv
        │   │   │       ├── waveform-median.npy
        │   │   │       ├── waveform-npy.csv
        │   │   │       └── waveform-rhythm.npy
        │   │   └── pre
        │   │       ├── ecg-metadata
        │   │       │   └── ...
        │   │       └── ecg-waveforms      
        │   │           └── ...
        │   ...
        │
        └── study-group
            ├── cohort.csv
            ├── comorbidities.csv
            ├── ecg.csv
            ├── ecg-metadata
            │   └── ...
            ├── ecg-waveforms
            │   ├── waveform.csv
            │   ├── waveform-median.npy
            │   ├── waveform-npy.csv
            │   └── waveform-rhythm.npy
            └── rosc-outcomes.csv

The files are organized into two folders, study-group and control-group. The control-group is divided by year and further more into control and pre. The control folder contains the ECGs and ECG metadata for ECGs that were administered on days corresponding with ECGs from the study-group. The pre folder contains ECGs and ECG metadata for ECGs for control patients that were administered prior to that indexing date. Not all control patients have a pre ECG.

Study group

The study group contains the ECGs of the patients whose hearts restarted after cardiac arrest [return of spontaneous circulation, ROSC].

Patient 101298 is found in year groups 2015-16 and 2017. The dates have been shifted by different amounts for each of those year groups.

Control group

The control group contains the ECGs of a set of patients who meet the following parameters:

• Visited the emergency department on the same day as a case in the study group
• Had an ECG on that day
• Did not have a cardiac arrest

If the patient from this control group had an ECG prior to that day, that ECG would have the pre timetag. Only the last prior ECG is included in this dataset.

ECG waveform tables and arrays

The ECGs are stored in NumPy arrays. There is a file for the study group and each year of the control group.

ECG metadata

The ECGs were stored as XML files. The following tables contain the addition metadata that was stored in those files.