Research method
Land-Use Regression
Land-use regression (LUR) predicts a pollutant’s concentration at unmeasured locations from a short monitoring campaign plus geographic predictors — traffic, land cover, population, elevation. The output is a map used as an exposure estimate in a health model, not a personal monitor on each participant. Two campaigns in the same city can be compared to ask whether the spatial pattern, not the absolute level, is stable enough to study mortality.
Environmental epidemiologists reach for LUR when they need an address-level NO₂ or air-toxics estimate for a cohort that was never wearing a sampler. It answers 'does living in a high predicted-pollution cell track this disease?' Its main limitation is that a stable spatial ranking does not mean concentrations were unchanged, and a Cox or logistic model on predicted NO₂ cannot prove that NO₂ rather than co-pollutants, noise, or poverty caused the deaths.
Evidence
What the evidence shows
Drawn from 3 studies in this library. Each finding starts with a plain-language takeaway, then the denser detail. Supports means evidence for a finding; Challenges means evidence against a stated position; Qualifies marks scope with a short note on each study’s contribution. Challenged positions are labeled — they are not findings.
Rome’s two NO₂ LUR maps, twelve years apart, still ranked neighbourhoods similarly enough to predict deaths. Campaign measurements correlated at 0.79 (0.83 background, 0.68 traffic). Highest versus lowest predicted-NO₂ quintile had HR 1.10 on the 1995/96 map and 1.11 on the 2007 map, among 45,006 natural deaths. Stable ranking is not a claim that levels were the same.
LUR air pollutants are often one layer in a correlated traffic bundle. An administrative cohort assigned road proximity, LUR pollutants, modelled noise and NDVI greenness: 13,170 non-Alzheimer dementia, 4,201 Parkinson, 1,277 Alzheimer and 658 MS cases. Road proximity raised risks; greenness attenuated some HRs/ORs; LUR air pollution associated with NAD and PD, not AD or MS; noise was generally null.
LUR can be one of several exposure assignments in the same case-control study. Los Angeles preterm-birth analyses used monitors, LUR, and source-apportioned PM: preterm prevalence 9.2%, with ~30% odds increases for total PAHs (benzo(a)pyrene 13%) but small PAH samples, plus IQR associations for ammonium nitrate PM₂.₅ and 9–13% odds rises for benzene, diesel/EC/OC and biomass-burning PM₂.₅.
Open questions
Tensions and limits
Some items are genuine disagreements on the same question. Others mark different assays, populations, or outcomes — limits on how far one study travels — not a forced fight between papers.
These papers do not agree on which traffic-related layer carries the signal. Rome’s analysis is built on NO₂ LUR as the exposure; the neurologic cohort finds air pollution associated with some diagnoses and not others, with greenness attenuating and noise mostly null; the LA preterm paper’s PAH signal is limited by two stations and shorter sampling. Treating 'LUR showed pollution is bad' as a single result hides those splits.
- Rome NO2 maps 12 years apart still predict deaths
- Roads and air pollution track neurologic disease
- LA air toxics, especially PAHs, track preterm birth
Study Role Design N Population Outcome Rome NO2 maps 12 years apart still predict deaths Supports CohortRome adult cohort with Cox mortality models using 1995/96 and 2007 LUR NO2 maps N=684204 · 684,204 subjects; 45,006 natural-cause deaths 2001–2006 Adult residents of Rome with geocoded residential addresses Natural-cause mortality associated with residential NO2 from two LUR surfaces Roads and air pollution track neurologic disease Supports CohortMetro Vancouver administrative cohorts of adults 45–84 linking roads/pollution/noise/greenness to neurologic incidence NAD/PD cohorts ~7.3–7.4 million person-years; 13,170 NAD, 4,201 PD, 1,277 AD, 658 MS cases — exact unique-person N not a single primary figure Adults aged 45–84 in Metro Vancouver with provincial health coverage Incident non-Alzheimer dementia, Parkinson’s, Alzheimer’s, and MS LA air toxics, especially PAHs, track preterm birth Supports Case-controlPopulation-based case-control of LA preterm birth near MATES monitors with traffic toxics N=241415 · 241,415 births after exclusions from 276,891 birth records; analyses often restricted nearer monitors Births in Los Angeles County near MATES air-toxics monitors Preterm birth odds related to PAHs, traffic markers, and source-apportioned PM2.5
Common misconceptions
If two LUR maps twelve years apart both predict mortality, air quality did not change.
The Rome paper’s point is that the spatial ranking of NO₂ remained correlated (0.79). Absolute concentrations can move while the high-versus-low contrast that a Cox model uses stays similar.
An HR of 1.10 per NO₂ quintile proves that nitrogen dioxide molecules caused the extra deaths.
The Cox models use predicted NO₂ as a spatial marker. Co-pollutants, traffic, and residual confounding remain; the neurologic paper’s correlated road, noise and greenness layers are exactly that problem made explicit.
If LUR and a monitor campaign disagree, LUR is simply wrong.
LUR interpolates a campaign; PAH estimates in Los Angeles were limited by two stations and shorter periods. Disagreement can mean sparse monitoring rather than a failed land-use model, and it still limits causal claims for a single toxin.
Exam-style questions
Short-answer questions that ask you to explain or compare, not recall.
Rome used 70 school sites in 1995/96 and 78 Ogawa sites in 2007. Why can both maps give almost the same mortality HR if the campaigns are not the same network?
The health model uses the spatial contrast (highest versus lowest quintile), and measured NO₂ still correlated at 0.79 across campaigns. Different samplers and years can preserve neighbourhood ranking even if they would not reproduce the same micrograms at a given door.
How does the neurologic-disease paper use LUR differently from treating NO₂ as the only exposure?
LUR pollutants are one of several assigned layers with road proximity, noise and NDVI. Associations differ by diagnosis (NAD/PD versus AD/MS), and greenness attenuation plus null noise results are part of the evidence, not extras.
Why is a 30% PAH odds increase for preterm birth a weaker causal claim than it first appears?
PAHs were measured at only two stations for shorter periods; residual SES confounding and monitor coverage remain. LUR and source-apportioned PM were additional assignments, not a randomised reduction of PAHs.
What can LUR not replace in a study of traffic and health?
A personal exposure measurement, a design that breaks confounding among co-pollutants, and a demonstration that changing NO₂ (rather than correlated traffic) changes disease. It replaces a sparse monitor’s single city-wide value with a spatial surface.
The studies
3 studies in this library bear on Land-Use Regression, ordered by citations.
- Rome NO2 maps 12 years apart still predict deaths
Land-use regression NO2 surfaces for Rome in 1995/96 and 2007 were highly correlated, and both predicted higher natural-cause mortality in a 2001–2006 adult cohort.
- Roads and air pollution track neurologic disease
In Metro Vancouver, living near roads raised incidence of non-Alzheimer dementia, Parkinson’s, Alzheimer’s, and MS; air pollution linked to NAD and PD, noise was null.
- LA air toxics, especially PAHs, track preterm birth
In Los Angeles, entire-pregnancy PAHs associated with ~30% higher preterm-birth odds; ammonium nitrate PM2.5 and several traffic markers also raised odds.
Learn alongside