Exposure measurement error in PM2.5 health effects studies: A pooled analysis of eight personal exposure validation studies

Background Exposure measurement error is a concern in long-term PM2.5 health studies using ambient concentrations as exposures. We assessed error magnitude by estimating calibration coefficients as the association between personal PM2.5 exposures from validation studies and typically available surrogate exposures. Methods Daily personal and ambient PM2.5, and when available sulfate, measurements were compiled from nine cities, over 2 to 12 days. True exposure was defined as personal exposure to PM2.5 of ambient origin. Since PM2.5 of ambient origin could only be determined for five cities, personal exposure to total PM2.5 was also considered. Surrogate exposures were estimated as ambient PM2.5 at the nearest monitor or predicted outside subjects’ homes. We estimated calibration coefficients by regressing true on surrogate exposures in random effects models. Results When monthly-averaged personal PM2.5 of ambient origin was used as the true exposure, calibration coefficients equaled 0.31 (95% CI:0.14, 0.47) for nearest monitor and 0.54 (95% CI:0.42, 0.65) for outdoor home predictions. Between-city heterogeneity was not found for outdoor home PM2.5 for either true exposure. Heterogeneity was significant for nearest monitor PM2.5, for both true exposures, but not after adjusting for city-average motor vehicle number for total personal PM2.5. Conclusions Calibration coefficients were <1, consistent with previously reported chronic health risks using nearest monitor exposures being under-estimated when ambient concentrations are the exposure of interest. Calibration coefficients were closer to 1 for outdoor home predictions, likely reflecting less spatial error. Further research is needed to determine how our findings can be incorporated in future health studies.

The observed city-specific calibration coefficients are on the x-axis, while the predicted city-specific calibration coefficients from the leave one city out procedure are on the y-axis. 9 S3 Relationship between the city-average number of cars per housing unit and the population density (left) and the percentage of detached homes in the study area (right). . . . . . . . . 11
Briefly, all subjects were recruited at senior or community centers, through doctors referrals or advertisements. All subjects were non-smokers, living in non-smoking homes. Latitude and longitude for the addresses of the participants, zip-codes or census blocks were available. Because the goals of the validation studies were different, some studies sampled specific population groups, such as the the elderly, patients with MI, COPD or CHF, children, and adults. All subjects younger than 18 years were excluded from our analyses, since long-term air pollution health studies are often focused on adult mortality.
For all studies, personal PM 2.5 exposures were measured with Personal Environmental Monitors (PEMs), which included small inertial impactors for PM 2.5 collection on Teflon filters. All personal samplers were attached to pumps placed in carrying bags. Subjects had to keep the monitors within their breathing zones at all times, except when showering or using the restroom. During prolonged times of inactivity (such as sleeping or watching TV), subjects were allowed to remove the carrying bag from their body, but were instructed to keep the monitors as close as possible to their breathing zones. Personal SO 2− 4 measurements were collected for four cities (Baltimore, MD, Boston, MA, Steubenville, OH and Atlanta, GA). For Baltimore and Boston the PM 2.5 filters were analyzed for SO 2− 4 with ion chromatography, while for Steubenville and Atlanta, SO 2− 4 were collected on fluoropore filters that were then analyzed also via ion chromatography.

S2. Sensitivity Analyses
In the main analysis for personal PM 2.5 of ambient origin we included 35 subjects with COPD and 12 subjects with MI, while in the analysis for total personal PM 2.5 we included 88 subjects with COPD, 12 subjects with MI and 21 subjects with CHD. Senior subjects accounted for the 54.4% of the total number of subjects in the 9 cities and 63.6% in the five cities with available personal PM 2.5 of ambient origin exposures. We conducted sensitivity analyses to assess whether subpopulation status modifies the estimated calibration coefficients.
We found no effect modification of the association between any true and surrogate exposures by COPD (p-value range: 0.254 -0.939) or CHD (p-value range: 0.214 -0.663) status. For individuals with MI, we found a significantly lower calibration coefficient for personal PM 2.5 of ambient origin in relation to the nearest ambient monitor concentrations (p-value < 0.001), but not for total personal PM 2.5 (p-value = 0.258) nor for outdoor home predictions (p-values: 0.217 and 0.489 for PM 2.5 of ambient origin and total personal PM 2.5 respectively).
We also found significant effect modification by age, with subjects younger than 65 years of age having lower calibration coefficients than their older counterparts. Less bias, therefore, would be expected in the health effect estimates from use of surrogate exposures in the health models for older subjects than younger. This effect modification was statistically significant for all combinations, except for that between personal PM 2.5 of ambient origin and outdoor home model predictions (p-value = 0.06). When the interaction between age and PM 2.5 concentrations at the nearest ambient monitor was included in Model 1, we were no longer able to detect any between-city heterogeneity for the calibration coefficient for personal PM 2.5 of ambient origin (p-value = 0.50).The stratified calibration coefficients are presented in Table  A3.
Further, we found that estimated calibration coefficients were similar irrespective of the method used to calculate monthly ambient concentrations. Correlations between the two estimation methods were moderate (Spearman r s = 0.61). When all days within the month were used in the calculation, the calibration coefficient for personal PM 2.5 of ambient origin was 0.31 (95%CI: 0.14, 0.47), vs. 0.35 (95%CI: 0.26, 0.43) when monthly ambient concentrations were calculated using only those days with personal monitoring. This sensitivity analysis could only be performed for the nearest ambient monitor concentrations, as the outdoor home model predictions were calculated at the monthly level only.
Finally, in a restricted analysis using data only from the five cities for which PM 2.5 of ambient origin were also available, the calibration coefficients for total personal PM 2.5 were somewhat higher, although comparable and with overlapping confidence intervals. For the nearest monitor the calibration coefficient was 0.59 (95%CI: 0.35, 0.83), while for the outdoor home model predictions it was 1.05 (95%CI: 0.75, 1.34).    and nearest monitor concentrations. The gray line represents perfect prediction. The observed city-specific calibration coefficients are on the x-axis, while the predicted city-specific calibration coefficients from the leave one city out procedure are on the y-axis.