a study of health effect estimates using competing methods to model personal exposures to ambient pm 2.5 - outdoor monitor

Various methods have recently been developed to estimate individual exposure to ambient particulate matter less than 2.
5 μm in diameter (PM2. 5)
Use a fixed outdoor monitor and personal exposure monitor.
A class of estimates involves the use environment-
SOURCE component of PM2.
5, such as sulfuric acid, iron, etc.
A key step in extrapolation of these values is to correct the difference in the permeability properties of the components used in extrapolation (
Such as sulfuric acid in PM2. 5)and PM2. 5.
If this is not done, the resulting health impact estimates will be biased.
Another method includes a factor analysis method, such as positive matrix decomposition (PMF).
The combination of extrapolation or factor analysis methods with regression calibration can estimate the direct impact of ambient pm2. 5.
5 Regarding Health, eliminate deviations caused by the use of fixed outdoor monitors and the estimated personal environment pm2. 5.
5 concentrations.
Several forms of extrapolation methods are defined, including some new extrapolation methods.
Health impact estimates generated using these methods were compared with results from PMF extended analysis using data collected from the Colorado Denver asthma Children's Health Study.
Lung function measures were used to examine differences in health impact estimates among various methods (
The amount of forced breath of 1 m/s)
When the health indicator shows the importance of the correction factor (s)
In the push-out method, the results generated by PMF are comparable to the extrapolation method containing the correction factor.
In the context of air pollution research, regression calibration (; )
Is a technique to correct deviations in regression results in the case of exposure variables being measured by errors ().
This technology can use a relatively large amount of fixed substances to determine the direct impact of air pollution on health (outdoor)
Monitoring data and relatively few personal monitoring data.
Most air pollution and health studies only use fixed monitoring data to estimate the health impact of air pollution.
It is recognized that the use of this method does not have an estimate of the impact of contamination in direct contact with individuals.
In contrast, the health impact of changes in outdoor air pollution concentrations is estimated.
The main reason for the difference between the two methods is that people spend most of their time indoors, where ambient air pollution concentrations are often a small part of the outdoors.
Therefore, in order to estimate the impact of direct exposure (i. e.
To obtain a "direct impact estimate ")
, Need to increase the degree of impact of fixed concentration changes on health ().
Some may argue that the association between health indicators and outdoor pollution concentrations is most relevant because only outdoor concentrations are regulated.
However, one disadvantage of the effect that depends only on the outdoor concentration is that this effect depends on the behavior of the subject (e. g.
, How much time do people spend indoors or outdoors)
Therefore, it will not be consistent in different locations and seasons.
In a sense, direct effect estimates will be standardized so that effects can be compared more meaningfully in studies conducted at different locations and at different times of the year.
In terms of public health policies, it may be more effective to develop air quality standards based on this standardized effect, rather than averaging the effect at different locations and seasons.
The model considered here is the same as the model proposed in.
The health model is the result of health and the real environmental pollutant (
PM in this case)
The exposure concentration is a random error that represents the subject and time (e. g. , days).
These terms are fixed.
Intercept and slope parameters, which are used to indicate the impact of contaminants on health outcomes, respectively.
The term set in parentheses is a covariable with a fixed coefficient, = 1 ,. . . , And is a random intercept specific to the topic.
Spatial exponential covariant structures are used to explain the time-dependent responses within individuals.
This structure is suitable for responses that are not evenly distributed over time, and this is the case with the study data.
The exposure model is the position and position of the fixed y-
Intercept and slope parameters are the ambient PM concentration of the day, respectively, and are the random slope of all subjects)
Random error for all).
Since the concentration of exposure to the individual environment was estimated, a simple error model was introduced to explain this: where.
Put this into Eq. (2)
Where is the yield.
When fitting the model (3)
For the data, constraint = 0 is due to the fact that the individual concentration of the ambient PM is part of the ambient PM concentration and when there is no ambient PM, environmental PM exposure should not occur.
The normal error assumption will be discussed in the results section below.
Inclusion/exclusion of random intercept and slope terms in the model (1)–(3)
It is based on checking and reasoning for data.
Health model (1)
The subjects were assumed to have the same susceptibility to air pollution, with no random slope.
The random slope of the pollutant is (2)and (3)
Subject allowed-
The specific personal environment PM is related to the environment PM and varies depending on the subject matter, mainly due to differences in penetration and ventilation associated with House Properties and time activity patterns.
The health model for the fixed monitor variable aspect is determined by replacing it with an expression in the equation. (2).
The average slope of the population of Is.
Therefore, if the slope of the pollutant variable representing the fit health model is represented and replaced is used, theta is obtained from the fitting of the equation. (3)
, Then/theta is with asymptotic variance ()
It may be noted that although there is no slope term for random contaminants in the health model (1), model (4)
Such a term is indeed involved (i. e.
, The underlying health model includes the subject's term for the slope of random contaminants when using fixed outdoor monitor concentrations instead of actual personal exposure).
This reflects the fact that different health-environmental PM relationships occur between subjects due to differences in exposure (via model (2))
Instead of responding differently to environmental PM exposure.
When fitting a health model (
Using fixed monitor data)
, Can include the random slope of the pollutant to match the base model.
However, since the mean value of the random slope term is 0, this contains a fixed effect estimate that is not expected to change.
In fact, the fitting of a health model with a random slope term does not change significantly or its standard error, so it is not used in the final model.
The environmental PM that has penetrated into the interior is often (outdoor)ambient PM.
Therefore, with subjects staying indoors most of the time, the absolute difference between the average personal environment PM and the environment PM tends to increase as the environment PM increases.
This difference is called concentration-
Dependent on measurement error.
In terms of estimating the health impact of individual environmental PM exposure, regression calibration eliminates concentration-
The correlation measurement error when the original estimate is divided by theta interval.
Usually, when measuredwith-
The error variable is used as a predictor, and the associated slope will be biased to 0 (i. e.
, Attenuation deviation).
However, when such a variable is modeled as a result variable, the error is absorbed into the remaining error term.
Therefore, in regression calibration, there is no additional deviation because the personal environment PM value is measured with error (or estimated)
Because they are modeled as result variables.
As discussed and explained in, the statistical significance of calibration estimates will not be much different from that of uncalibrated estimates.
This is because in regression calibration, the relationship between health and personal environment PM is indirectly determined by a third variable (
Environment PM in this case).
Although the estimates have changed, the range of changes in the standard error is roughly the same, so,
Based on these proportions)
Stay the same.
In theory, if there is how much real personal environment PM measurement can be used as a concentration measurement for a fixed monitor, using personal data, health impact estimates will certainly be more statistically important.
In practice, however, there is often a small amount of personal exposure data, so the ability to model these data directly is minimal.
Regression calibration allows people to use a very predictable relationship between personal environment and outdoor PM data to calibrate estimates using fixed outdoor monitors for more data.
In addition, if the estimated personal environment PM value is used as a predictive variable, some attenuation bias will be generated.
Regression calibration avoids this deviation as described above.
For the following method, we assume that for a given topic ()and day ()
Personal PM ()
Can be specifically divided into the environment ()
Non-environment ()
Such a part.
For the following methods, the estimated individual environmental PM concentration formula is represented by an asterisk in which it is used to distinguish between the estimated quantity and the real quantity.
Over the past decade, the intuitive appeal of source resolution methods such as PMF has aroused widespread interest in their use in air pollution health research.
Use personal, indoor and outdoor PM (or PM)
Concentration collected in a few days, the extended factor analysis model divides the total concentration into concentrations from many indoor and outdoor sources.
If the estimates are accurate, these methods provide valuable information about the extent to which individuals are exposed to PM due to various sources such as environmental tobacco smoke, automotive exhaust, soil, cleaning products, etc.
The program seeks solutions for the following extended receptor models (; , )
: Which indicates the PM concentration on the day of the specific object;
And the indoor and outdoor PM concentration of the day (e. g.
, Shooting at a fixed place where the study subjects go to school);
The contribution of the source to the theme PM concentration on the day;
The number of indoors and outdoors is similar;
Relative concentration of species at source.
Indices = 1 ,. . . Represents the source of the outdoor, and = 1 ,. . . Representing indoor sources (i. e.
, There are two sources of outdoor and indoor). Collectively, {, …, }
Source configuration file for Source (for each )
, Which indicates the total number of species involved in the study.
One assumption in the model is that the relative number of species in each individual source contribution remains the same over time (i. e.
, Fixed source profile)
The strength is allowed to be different. Note that in (6)
Outdoor sources can promote the concentration of outdoor, indoor and personal PM, while indoor sources can only promote individual and indoor concentrations.
For more details on models and applications, please refer ,)and , ).
Based on the preliminary analysis of the research data, the model with = 4 outdoor factors and = 3 indoor factors is determined to provide good fitting data for the explanatory factors.
Then it is estimated that the concentration of PM in personal environment is (
Only outdoor factors are added).
That is, the estimated contribution of each outdoor source and specific region is aggregated within each subject day to obtain the estimated individual ambient PM concentration.
The extrapolation method basically takes a component (or specie)
PM that is known primarily or entirely from the source of the environment and uses it to infer the value of the PM for the personal environment ().
This method is easier to use and understand than PMF.
However, the quantity is still estimated and the precision is uncertain.
Various extrapolation methods are considered below. ()
: In Denver EPA data, sulfuric acid particles in ambient PM tend to be more at the low end of PM particle size distribution, while iron particles tend to be at the high end.
Since sulfuric acid particles are usually small, indoor subjects may be exposed to ambient PM with higher levels than outdoor PM.
To illustrate this, approximate equivalence is used to estimate the value.
Here, it is indicated that the concentration of sulfuric acid in the ambient PM is the same fixed position as that of the day, and it indicates the concentration of personal sulfuric acid exposure in the subject and in the ambient PM of the day.
The average relative difference between the amount of sulfuric acid in outdoor PM and the proportion of sulfuric acid in individual ambient PM.
The subject data observed can be estimated in various ways. (
The estimate is expressed as λ. )
The method used here is described in.
You can then estimate the number of interest by replacing all the other quantities in the equation with known or estimated values.
For calculations, it is assumed that there is no sulfuric acid from a non-environmental source, so =, the latter (
Total sulfur exposure)
Direct measurement.
Therefore, the estimate is ()
: The previous method can be extended to include multiple components.
To demonstrate that, in addition to sulfuric acid, iron is also used to estimate individual exposure to environmental PM.
Iron comes mainly from the soil, usually in the smaller soil.
The size range of the coarse particle pattern contained in the PM.
The source of indoor iron is often unlikely, although it is possible to extract iron from the soil entering the home (e. g.
(By penetration or tracking)
Can be resolved and re-interfered later, which makes it a possible candidate for a non-environmental source ().
Although iron does not contain a large part of the ambient PM, individual iron samples have never been 0, unlike individual sulfuric acid.
Therefore, when sulfuric acid is used in combination with iron compared to the use of sulfuric acid alone, the extrapolation estimate of the ambient personal PM is never 0.
The real zero value of is very impossible, so this method is expected to be improved on one of the shortcomings of e1.
Promote emotional intelligence. (7)
Two species (
Sulfur, iron)
, Can be used, and its neutrals represent ambient iron and individual ambient iron concentrations within PM, with other quantities defined as previous quantities. (
As with sulfuric acid, it is assumed that the individual environment and the total iron exposure concentration of the individual are equal, and the latter is measured directly. )
Therefore, using the quantities estimated here, the methods described in respectively are used to estimate the λ ̂ and λ ̂ of sulfuric acid and iron data, respectively.
: There may be some factors that influence the value that can be included in the estimation process. As both time-
The activity and penetration and removal properties of PM will depend on the outdoor air temperature relative to the indoor, and temperature is such a factor.
Use the cut point of 38 °F (3. 3°C)
, In the number of days when the temperature exceeds or does not exceed this value, it is estimated that there will be a separate value. (This cut-
Point divides the data into roughly equal half. )
This classification was carried out for sulfuric acid and iron, so four values were estimated for the species, respectivelyby-
Combination of temperature categories (, , , )
Method described in.
Use these values and then use ()
: This method is not expected to be the best.
However, it makes sense to determine its impact on individual environmental PM estimates and on the resulting health impact estimates.
These effects are discussed below and then demonstrated using the data in the results section.
The difference in the estimated individual ambient PM concentration between the extrapolation method E1 and E4 is only determined by a scalar (
For E1 but not for E4).
Specifically, the estimate using E4 is as large as the estimate using e1.
Due to the scale difference of this system, the estimation in Eq. (3)
For E4, it can be obtained by dividing the estimation of E1.
Therefore, health impact estimates using E4 calibration can be expressed.
In other words, the E4-based health impact estimate will be twice that of the e1-based health impact estimate.
The implication is that if 1.
These results are correct, whether they are estimated or true.
Another method of extrapolation may involve topics
Specific values, as subjects may live in different types of housing with different penetration and ventilation properties and with different patterns of time activity.
However, the previous analysis ()
It was found that the merging of these values did not lead to improvements estimated in these study data.
The error structure of these methods is unknown and may vary depending on the situation.
Here, let's say that all methods have a simple addition error structure, which is explained in the equation. (3).
Although the actual error structure may be more complicated, especially for the method of source allocation, it is better to assume some kind of error structure even if it is too simple, rather than assuming that there is no at all.