By name: Study Design description: "Environmental epidemiology study design patterns. Use when designing time series studies, case-crossover analyses, or planning exposure assessment strategies." metadata: labels: [epidemiology, study-design, time-series, case-crossover] triggers: files: ['**/*.R'] keywords: [case-crossover, time-series, exposure, outcome, study design, ecological]
Study Design
Priority: P1 (OPERATIONAL)
📋 Common Designs
Time Series (Most Common for DLNM)
- Unit: Day (or week/month).
- Outcome: Daily counts (deaths, hospital admissions) for a defined population.
- Exposure: Daily average (or max) from monitoring data.
- Strengths: Population-level inference, controls for individual confounders by design.
- Weakness: Ecological fallacy — population-level association ≠ individual risk.
Case-Crossover
- Unit: Individual event.
- Design: Each case serves as its own control. Compare exposure on event day vs. control days.
- Control selection: Time-stratified (same DOW in same month/year).
- R package:
survival::clogit()withstrata(). - Use when: Individual-level exposure data available; want within-person comparison.
📊 Outcome Data
| Source | Typical Variables | Format |
|---|---|---|
| Mortality registers | Date, age, sex, cause (ICD-10) | Daily counts by stratum |
| Hospital admissions | Date, diagnosis, age, sex | Daily counts by cause |
| Emergency visits | Date, diagnosis, triage category | Daily counts |
- ICD-10 codes: Define outcome clearly (e.g., cardiovascular: I00-I99, respiratory: J00-J99).
- Age stratification: Common groups: 0-64, 65-74, 75+.
🌡 Exposure Assessment
- Fixed monitors: Daily PM2.5, O3, temperature from regulatory networks.
- Spatial interpolation: Kriging, IDW to assign exposure to populations between monitors.
- Satellite data: AOD-derived PM2.5 estimates (e.g., van Donkelaar datasets).
- Lag alignment: Ensure exposure date aligns with outcome date (same day = lag 0).
⚠️ Key Considerations
- Minimum series length: ≥ 3 years recommended for seasonal control.
- Population stability: Assume fixed population over study period.
- Harvesting debate: Short-term displacement vs. true excess mortality.
- Multiple comparisons: Pre-specify primary analyses; label exploratory analyses.