Scalable exposure assessment approaches that capture personal exposure to particles for purposes of epidemiology are currently limited, but valuable, particularly in low-/middle-income countries where sources of personal exposure are often distinct from those of ambient concentrations. We measured 2 × 24-h integrated personal exposure to PM2.5 and black carbon in two seasons in 402 participants living in peri-urban South India. Means (sd) of PM2.5 personal exposure were 55.1(82.8) µg/m3 for men and ...
Scalable exposure assessment approaches that capture personal exposure to particles for purposes of epidemiology are currently limited, but valuable, particularly in low-/middle-income countries where sources of personal exposure are often distinct from those of ambient concentrations. We measured 2 × 24-h integrated personal exposure to PM2.5 and black carbon in two seasons in 402 participants living in peri-urban South India. Means (sd) of PM2.5 personal exposure were 55.1(82.8) µg/m3 for men and 58.5(58.8) µg/m3 for women; corresponding figures for black carbon were 4.6(7.0) µg/m3 and 6.1(9.6) µg/m3. Most variability in personal exposure was within participant (intra-class correlation ~20%). Personal exposure measurements were not correlated (Rspearman < 0.2) with annual ambient concentration at residence modeled by land-use regression; no subgroup with moderate or good agreement could be identified (weighted kappa ≤ 0.3 in all subgroups). We developed models to predict personal exposure in men and women separately, based on time-invariant characteristics collected at baseline (individual, household, and general time-activity) using forward stepwise model building with mixed models. Models for women included cooking activities and household socio-economic position, while models for men included smoking and occupation. Models performed moderately in terms of between-participant variance explained (38-53%) and correlations between predictions and measurements (Rspearman: 0.30-0.50). More detailed, time-varying time-activity data did not substantially improve the performance of the models. Our results demonstrate the feasibility of predicting personal exposure in support of epidemiological studies investigating long-term particulate matter exposure in settings characterized by solid fuel use and high occupational exposure to particles.
+
