| # OHCA Annotation Guidelines | |
| ## Overview | |
| This document provides comprehensive guidelines for manually annotating discharge notes to identify Out-of-Hospital Cardiac Arrest (OHCA) cases. | |
| ## Definition of OHCA | |
| **Out-of-Hospital Cardiac Arrest (OHCA)** is a cardiac arrest that occurs **outside** a healthcare facility (home, workplace, public spaces, etc.). | |
| ## Annotation Labels | |
| - **1 = OHCA**: Cardiac arrest that occurred outside a healthcare facility | |
| - **0 = Non-OHCA**: Everything else | |
| ## Include as OHCA (Label = 1) | |
| - Cardiac arrest at home | |
| - Cardiac arrest at work | |
| - Cardiac arrest in public spaces | |
| - Cardiac arrest witnessed by bystanders | |
| - Cardiac arrest with CPR given outside hospital | |
| ## Exclude as Non-OHCA (Label = 0) | |
| ### In-Hospital Cardiac Arrests | |
| - Cardiac arrest occurring within any healthcare facility | |
| - Code blue called on hospital ward | |
| - Arrest during surgery or medical procedures | |
| - Arrest in ICU, emergency department, or any hospital unit | |
| ### Transfer Cases | |
| - **All transfer cases are excluded**, including: | |
| - Patients transferred from outside hospitals for OHCA | |
| - Patients transferred for cardiac catheterization after OHCA | |
| ### Historical Cardiac Arrests | |
| - Patients with history of cardiac arrest who did not present with current cardiac arrest | |
| ### Non-Arrest Conditions | |
| - Chest pain without cardiac arrest | |
| - Heart attack (MI) without cardiac arrest | |
| - Shortness of breath | |
| - Syncope/fainting without arrest | |
| - Any condition where cardiac arrest did not occur | |
| ## Decision Tree | |
| ``` | |
| 1. Did a cardiac arrest occur during this admission? | |
| βββ NO β Label = 0 | |
| βββ YES β Continue to 2 | |
| 2. Did the arrest happen OUTSIDE a healthcare facility? | |
| βββ NO β Label = 0 | |
| βββ YES β Continue to 3 | |
| 3. Is this a transfer case? | |
| βββ YES β Label = 0 | |
| βββ NO β Label = 1 | |
| ``` | |
| ## Confidence Scale | |
| Rate your confidence in the annotation (1-5 scale): | |
| - **5 = Very Confident**: Clear, unambiguous case | |
| - **4 = Confident**: Strong evidence, minor uncertainty | |
| - **3 = Moderately Confident**: Some ambiguity but leaning toward decision | |
| - **2 = Uncertain**: Significant ambiguity, difficult case | |
| - **1 = Very Uncertain**: Unclear, may need expert review | |
| ## Examples with Rationale | |
| ### Example 1: OHCA at Home (Label = 1) | |
| **Text**: "Patient found down at home by spouse, immediate CPR initiated, EMS arrived and achieved ROSC." | |
| **Rationale**: | |
| - β Cardiac arrest occurred | |
| - β Outside hospital (at home) | |
| - β Not a transfer case | |
| - **Label**: 1, **Confidence**: 5 | |
| ### Example 2: In-Hospital Arrest (Label = 0) | |
| **Text**: "Patient admitted for pneumonia, developed cardiac arrest on day 3 of hospitalization." | |
| **Rationale**: | |
| - β Cardiac arrest occurred | |
| - β Inside hospital | |
| - **Label**: 0, **Confidence**: 5 | |
| ### Example 3: Transfer for OHCA (Label = 0) | |
| **Text**: "Transfer from community hospital. Patient had cardiac arrest at home, CPR by family, transferred for further care." | |
| **Rationale**: | |
| - β Cardiac arrest occurred outside hospital originally | |
| - β This is a transfer case | |
| - **Label**: 0, **Confidence**: 5 | |
| ### Example 4: Historical Arrest (Label = 0) | |
| **Text**: "Patient with history of cardiac arrest 1 year ago, now presents with chest pain." | |
| **Rationale**: | |
| - β No current cardiac arrest | |
| - β Historical arrest, not current admission | |
| - **Label**: 0, **Confidence**: 5 | |
| ### Example 5: Non-Arrest Condition (Label = 0) | |
| **Text**: "Chief complaint: Chest pain. Patient presents with acute MI, underwent emergency PCI." | |
| **Rationale**: | |
| - β No cardiac arrest occurred | |
| - **Label**: 0, **Confidence**: 5 | |
| ### Example 6: Workplace Cardiac Arrest (Label = 1) | |
| **Text**: "Patient collapsed at work, coworkers initiated CPR, EMS transported to hospital." | |
| **Rationale**: | |
| - β Cardiac arrest occurred | |
| - β Outside hospital (at work) | |
| - β Not a transfer case | |
| - **Label**: 1, **Confidence**: 5 | |
| ## Quality Control | |
| ### Before Submitting | |
| 1. **Double-check** each decision against the criteria | |
| 2. **Review** cases with confidence < 3 | |
| 3. **Add notes** for any unusual or borderline cases | |
| 4. **Ensure consistency** in similar cases | |
| ### Notes Field | |
| Use the notes field to document: | |
| - Reasoning for difficult decisions | |
| - Key phrases that influenced decision | |
| - Uncertainties or missing information | |
| - Questions for review | |
| ## Common Mistakes to Avoid | |
| 1. **Don't** include in-hospital arrests | |
| 2. **Don't** include any transfer cases (even if transferred for OHCA) | |
| 3. **Don't** include historical arrests from previous admissions | |
| 4. **Don't** include conditions where no cardiac arrest occurred | |
| ## Difficult Cases | |
| ### Transfer Cases | |
| - **All transfer cases are excluded (Label = 0)** | |
| - This includes transfers specifically for OHCA management | |
| - Focus on whether this is a transfer, not the original location of arrest | |
| ### Multiple Conditions | |
| - Focus on whether cardiac arrest actually occurred | |
| - If no arrest mentioned, label as 0 | |
| ### Incomplete Information | |
| - Base decision on available information | |
| - Use lower confidence score | |
| - Add notes about missing information | |
| ## Support | |
| If you encounter cases that don't fit these guidelines: | |
| 1. Make your best judgment | |
| 2. Use a lower confidence score | |
| 3. Document your reasoning in notes | |
| 4. Flag for expert review if needed | |
| Remember: Consistency is key for model training. When in doubt, err on the side of excluding (Label = 0) and document your uncertainty. | |
| ## Key Principles | |
| 1. **OHCA must occur outside healthcare facilities** | |
| 2. **All transfer cases are excluded** | |
| 3. **Only current cardiac arrests, not historical ones** | |
| 4. **No cardiac arrest = automatic exclusion** | |
| 5. **When uncertain, exclude and note reasons** | |