{ "protocol_version": "1.0.0", "study_id": "pm25-health-v2", "title": "A dual-track methodological study of ambient particulate pollution and population health", "question": "Which parts of the observed relationship between long-term particulate exposure and population health are reproducible associations, and which can be supported by an explicit causal identification design?", "claim_target": "causal", "estimands": [ { "id": "A-country-association", "population": "Eligible countries or economies with repeated annual measurements", "contrast": "A 10 microgram per cubic metre difference in population-weighted annual PM2.5 exposure", "outcome": "Change in period life expectancy at birth; disease-specific age-standardized mortality is the preferred future outcome", "time_horizon": "Current, 1-year, 3-year, 5-year and 10-year exposure windows", "weighting": "Equal-country primary estimand; population-weighted result required as a distinct secondary estimand when population data are added" }, { "id": "B-policy-att", "population": "Locations affected by a documented pollution policy or natural experiment and eligible comparison locations", "contrast": "Policy-induced change in particulate exposure versus the counterfactual exposure path without treatment", "outcome": "Life expectancy or age-standardized all-cause and cardiorespiratory mortality", "time_horizon": "Dynamic event-time effects before and after intervention", "weighting": "Average treatment effect on treated locations, aggregated by treatment-cohort size" } ], "population": { "unit": "Country-year for Track A; policy-defined subnational location-year is preferred for Track B", "time": "Annual observations", "eligibility": [ "Geographic unit has a stable identifier", "Outcome and exposure timing can be aligned without forward filling", "At least three pre-periods and three post-periods for a Track B event study", "Data provenance and measurement type are recorded" ], "exclusions": [ "World Bank regional and income aggregates", "Observations with impossible units or unresolved geography", "Track B interventions without a defensible comparison group or assignment mechanism" ] }, "exposure": { "id": "pm25", "label": "Population-weighted mean annual PM2.5 exposure", "unit": "micrograms per cubic metre", "timing": "Annual and pre-specified moving-average or distributed-lag windows", "primary": true }, "outcomes": [ { "id": "life", "label": "Period life expectancy at birth", "unit": "years", "timing": "Annual", "primary": false }, { "id": "asdr_cardiorespiratory", "label": "Age-standardized cardiorespiratory mortality", "unit": "deaths per 100,000", "timing": "Annual", "primary": true } ], "variable_roles": { "confounders": [ { "id": "gdp_ppp_lag1", "rationale": "Prior economic conditions can affect pollution, infrastructure and mortality", "timing": "At least one year before the outcome" }, { "id": "urban_pct_lag1", "rationale": "Prior urban form can affect exposure, service access and health", "timing": "At least one year before the outcome" }, { "id": "smoking_prevalence_lagged", "rationale": "Smoking affects cardiorespiratory mortality and can covary with development", "timing": "Prior or baseline value; pending data" }, { "id": "temperature_and_climate", "rationale": "Climate affects particulate formation, transport and mortality", "timing": "Contemporaneous exogenous weather summaries or lagged climate normals; pending data" }, { "id": "conflict_and_epidemic_shocks", "rationale": "Major shocks affect mortality and economic activity", "timing": "Pre-specified indicators or exclusions; pending data" } ], "mediators": [ { "id": "health_ppp", "rationale": "Pollution burden and policy can alter health spending; development may affect health through spending", "timing": "Do not include in the total-effect primary model; use only in a labelled mechanism analysis" } ], "colliders_or_descendants": [ { "id": "age65_pct", "rationale": "Longer survival changes the share aged 65 and older, making it partly downstream of the outcome process", "timing": "Exclude from the primary causal adjustment set" } ], "descriptive_only": [ { "id": "region", "rationale": "Useful for sample description, stratification and region-by-year shocks", "timing": "Time invariant" }, { "id": "income_group", "rationale": "Useful for heterogeneity; current classification must not be treated as a historical confounder", "timing": "Version-specific classification" } ] }, "data_sources": [ { "id": "wb-pm25-gbd2023", "provider": "World Bank World Development Indicators; underlying GBD 2023 exposure estimates", "version_or_access_date": "API snapshot last updated 2026-07-13", "url": "https://data.worldbank.org/indicator/EN.ATM.PM25.MC.M3", "licence": "CC BY 4.0 as stated by World Bank", "measurement_type": "modelled", "uncertainty_available": false }, { "id": "wb-life-expectancy", "provider": "World Bank WDI; UN World Population Prospects and national statistical sources", "version_or_access_date": "API snapshot last updated 2026-07-13", "url": "https://data.worldbank.org/indicator/SP.DYN.LE00.IN", "licence": "CC BY 4.0 as stated by World Bank", "measurement_type": "hybrid", "uncertainty_available": false }, { "id": "who-ghe2021-causes", "provider": "World Health Organization Global Health Estimates 2021", "version_or_access_date": "2024 release; evaluated 2026-07-20", "url": "https://www.who.int/data/gho/data/themes/mortality-and-global-health-estimates/ghe-leading-causes-of-death", "licence": "WHO terms must be recorded for each downloaded asset", "measurement_type": "hybrid", "uncertainty_available": true } ], "track_a": { "enabled": true, "design": "Longitudinal ecological panel with pre-specified alternative time windows and falsification tests", "primary_model": "Country and year fixed effects with lagged GDP and urbanization; country-clustered uncertainty; life expectancy is a methodological demonstration outcome", "diagnostics": [ "complete-case and missingness flow", "within-unit exposure variation", "country-specific trend sensitivity", "one-year and five-year change models", "future-exposure placebo", "3-year, 5-year and 10-year exposure windows", "nonlinear spline contrast", "country-clustered and Driscoll-Kraay uncertainty", "COVID-period exclusion", "leave-one-region-out" ], "status": "data-ready", "blocking_input": null }, "track_b": { "enabled": true, "design": "Group-time difference-in-differences for staggered policy adoption, with regression discontinuity or instrumental variables used when assignment is threshold- or instrument-based", "primary_model": "Cohort-specific ATT using not-yet-treated or never-treated controls, aggregated by treatment-cohort size", "diagnostics": [ "treatment timing and reversals", "comparison-group overlap", "event-time pre-trends", "anticipation window", "negative-control health outcome", "exposure first stage", "spillover mapping", "alternative event windows", "unit-cluster bootstrap" ], "status": "blocked", "blocking_input": "A versioned location-year intervention dataset with policy start, geographic scope, assignment mechanism and plausible comparison locations" }, "decision_rules": { "primary_specification": "Track A primary estimate is the lagged-confounder two-way fixed-effects model; Track B primary estimate is the pre-registered design-specific ATT or local effect.", "robustness_rule": "An association is called stable only if its direction and material magnitude survive trend, change, exposure-window and uncertainty checks; null compatibility must be reported rather than converted to no effect.", "falsification_rule": "A material future-exposure association, failed Track B pre-trend, failed exposure first stage or effect on a pre-specified negative-control outcome prevents an evidence upgrade.", "causal_language_rule": "Track A never licenses causal language. Track B licenses a bounded causal interpretation only after the assignment mechanism, overlap, timing, falsification and interference assumptions are documented and pass their pre-specified checks." } }