Attenuation of exposure-response rate ratios at higher exposures: a simulation study focusing on frailty and measurement error

Long-term air pollution exposure and diabetes risk in American older adults: A national secondary data-based cohort study

Type 2 diabetes is a major public health concern. Several studies have found an increased diabetes risk associated with long-term air pollution exposure. However, most current studies are limited in their generalizability, exposure assessment, or the ability to differentiate incidence and prevalence cases. We assessed the association between air pollution and first documented diabetes occurrence in a national U.S. cohort of older adults to estimate diabetes risk. We included all Medicare enrollees 65 years and older in the fee-for-service program, part A and part B, in the contiguous United States (2000-2016). Participants were followed annually until the first recorded diabetes diagnosis, end of enrollment, or death (264, 869, 458 person-years). We obtained annual estimates of fine particulate matter (PM2.5), nitrogen dioxide (NO2), and warm-months ozone (O3) exposures from highly spatiotemporally resolved prediction models. We assessed the simultaneous effects of the pollutants on diabetes risk using survival analyses. We repeated the models in cohorts restricted to ZIP codes with air pollution levels not exceeding the national ambient air quality standards (NAAQS) during the study period. We identified 10, 024, 879 diabetes cases of 41, 780, 637 people (3.8% of person-years). The hazard ratio (HR) for first diabetes occurrence was 1.074 (95% CI 1.058; 1.089) for 5 μg/m3 increase in PM2.5, 1.055 (95% CI 1.050; 1.060) for 5 ppb increase in NO2, and 0.999 (95% CI 0.993; 1.004) for 5 ppb increase in O3. Both for NO2 and PM2.5 there was evidence of non-linear exposure-response curves with stronger associations at lower levels (NO2 ≤ 36 ppb, PM2.5 ≤ 8.2 μg/m3). Furthermore, associations remained in the restricted low-level cohorts. The O3-diabetes exposure-response relationship differed greatly between models and require further investigation. In conclusion, exposures to PM2.5 and NO2 are associated with increased diabetes risk, even when restricting the exposure to levels below the NAAQS set by the U.S. EPA.

Long-term air pollution exposure and diabetes risk in American older adults: A national secondary data-based cohort study

Type 2 diabetes is a major public health concern. Several studies have found an increased diabetes risk associated with long-term air pollution exposure. However, most current studies are limited in their generalizability, exposure assessment, or the ability to differentiate incidence and prevalence cases. We assessed the association between air pollution and first documented diabetes occurrence in a national U.S. cohort of older adults to estimate diabetes risk. We included all Medicare enrollees 65 years and older in the fee-for-service program, part A and part B, in the contiguous United States (2000-2016). Participants were followed annually until the first recorded diabetes diagnosis, end of enrollment, or death (264, 869, 458 person-years). We obtained annual estimates of fine particulate matter (PM_2.5), nitrogen dioxide (NO₂), and warm-months ozone (O₃) exposures from highly spatiotemporally resolved prediction models. We assessed the simultaneous effects of the pollutants on diabetes risk using survival analyses. We repeated the models in cohorts restricted to ZIP codes with air pollution levels not exceeding the national ambient air quality standards (NAAQS) during the study period. We identified 10, 024, 879 diabetes cases of 41, 780, 637 people (3.8% of person-years). The hazard ratio (HR) for first diabetes occurrence was 1.074 (95% CI 1.058; 1.089) for 5 μg/m³ increase in PM_2.5, 1.055 (95% CI 1.050; 1.060) for 5 ppb increase in NO₂, and 0.999 (95% CI 0.993; 1.004) for 5 ppb increase in O₃. Both for NO₂ and PM_2.5 there was evidence of non-linear exposure-response curves with stronger associations at lower levels (NO₂ ≤ 36 ppb, PM_2.5 ≤ 8.2 μg/m³). Furthermore, associations remained in the restricted low-level cohorts. The O₃-diabetes exposure-response relationship differed greatly between models and require further investigation. In conclusion, exposures to PM_2.5 and NO₂ are associated with increased diabetes risk, even when restricting the exposure to levels below the NAAQS set by the U.S. EPA.

Low-Concentration Air Pollution and Mortality in American Older Adults: A National Cohort Analysis (2001-2017)

Mounting epidemiological evidence has documented the associations between long-term exposure to multiple air pollutants and increased mortality. There is a pressing need to determine whether risks persist at low concentrations including below current national standards. Air pollution levels have decreased in the United States, and better understanding of the health effects of low-level air pollution is essential for the amendment of National Ambient Air Quality Standards (NAAQS). A nationwide, population-based, open cohort study was conducted to estimate the association between long-term exposure to low-level PM_2.5, NO₂, O₃, and all-cause mortality. The study population included all Medicare enrollees (ages 65 years or older) in the contiguous U.S. from 2001 to 2017. We further defined three low-exposure subcohorts comprised of Medicare enrollees who were always exposed to low-level PM_2.5 (annual mean ≤12-μg/m³), NO₂ (annual mean ≤53-ppb), and O₃ (warm-season mean ≤50-ppb), respectively, over the study period. Of the 68.7-million Medicare enrollees, 33.1% (22.8-million, mean age 75.9 years), 93.8% (64.5-million, mean age 76.2 years), and 65.0% (44.7-million, mean age 75.6 years) were always exposed to low-level annual PM_2.5, annual NO₂, and warm-season O₃ over the study period, respectively. Among the low-exposure cohorts, a 10-μg/m³ increase in PM_2.5, 10-ppb increase in NO₂, and 10-ppb increase in warm-season O₃, were, respectively, associated with an increase in mortality rate ranging between 10 and 13%, 2 and 4%, and 12 and 14% in single-pollutant models, and between 6 and 8%, 1 and 3%, and 9 and 11% in tripollutant models, using three statistical approaches. There was strong evidence of linearity in concentration-response relationships for PM_2.5 and NO₂ at levels below the current NAAQS, suggesting that no safe threshold exists for health-harmful pollution levels. For O₃, the concentration-response relationship shows an increasingly positive association at levels above 40-ppb. In conclusion, exposure to low levels of PM_2.5, NO₂, and warm-season O₃ was associated with an increased risk of all-cause mortality.

Nonlinear effect of air pollution on adult pneumonia hospital visits in the coastal city of Qingdao, China: A time-series analysis

Many studies have illustrated adverse effects of short-term exposure to air pollution on human health, which usually assumes a linear exposure-response (E-R) function in the delineation of health effects due to air pollution. However, nonlinearity may exist in the association between air pollutant concentrations and health outcomes such as adult pneumonia hospital visits, and there is a research gap in understanding the nonlinearity. Here, we utilized both the distributed lag model (DLM) and nonlinear model (DLNM) to compare the linear and nonlinear impacts of air pollution on adult pneumonia hospital visits in the coastal city of Qingdao, China. While both models show adverse effects of air pollutants on adult pneumonia hospital visits, the DLNM shows an attenuation of E-R curves at high concentrations. Moreover, the DLNM may reveal delayed health effects that may be missed in the DLM, e.g., ozone exposure and pneumonia hospital visits. With the stratified analysis of air pollutants on adult pneumonia hospital visits, both models consistently reveal that the influence of air pollutants is higher during the cold season than during the warm season. Nevertheless, they may behave differently in terms of other subgroups, such as age, gender and visit types. For instance, while no significant impact due to PM_2.5 in any of the subgroups abovementioned emerges based on DLM, the results from DLNM indicate statistically significant impacts for the subgroups of elderly, female and emergency department (ED) visits. With respect to adjustment by two-pollutants, PM₁₀ effect estimates for pneumonia hospital visits were the most robust in both DLM and DLNM, followed by NO₂ and SO₂ based on the DLNM. Considering the estimated health effects of air pollution relying on the assumed E-R functions, our results demonstrate that the traditional linear association assumptions may overlook some potential health risks.

Benzene exposure and risk of benzene poisoning in Chinese workers

OBJECTIVES

Benzene is a known haematoxin and leukemogen that can cause benzene poisoning (BP), that is, a persistent reduction in white cell counts that is strongly associated with increased risk of lymphohaematopoietic malignancies. Data are needed on the exposure-response, particularly at low doses and susceptible populations for clinical and regulatory purposes.

METHODS

In a case-cohort study among 110 631 Chinese workers first employed 1949-1987 and followed up during 1972-1999, we evaluated BP risk according to benzene exposure level and investigated risk modification by subject (sex, attained age) and exposure-related factors (latency, exposure windows, age at first benzene exposure, coexposure to toluene) using excess relative risk and excess absolute risk models.

RESULTS

There were 538 BP cases and 909 benzene-exposed referents. The exposure metric with best model fit was cumulative benzene exposure during a 5-year risk window, followed by a 9-month lag period before BP diagnosis. Estimated excess absolute risk of BP at age 60 increased from 0.5% for subjects in the lowest benzene exposure category (>0 to 10 ppm-years) to 5.0% for those in the highest category (>100 ppm-years) compared with unexposed subjects. Increased risks were apparent at low cumulative exposure levels and for workers who were first exposed at <30 years of age.

CONCLUSIONS

Our data show a clear association between benzene exposure and BP, beginning at low cumulative benzene exposure levels with no threshold, and with higher risks for workers exposed at younger ages. These findings are important because BP has been linked to a strongly increased development of lymphohaematopoietic malignancies.

Urinary arsenic and heart disease mortality in NHANES 2003-2014

BACKGROUND

Evidence evaluating the prospective association between low-to moderate-inorganic arsenic (iAs) exposure and cardiovascular disease in the general US population is limited. We evaluated the association between urinary arsenic concentrations in National Health and Nutrition Examination Survey (NHANES) 2003-2014 and heart disease mortality linked from the National Death Index through 2015.

METHODS

We modeled iAs exposure as urinary total arsenic and dimethylarsinate among participants with low seafood intake, based on low arsenobetaine levels (N = 4990). We estimated multivariable adjusted hazard ratios (HRs) for heart disease mortality per interquartile range (IQR) increase in urinary arsenic levels using survey-weighted, Cox proportional hazards models, and evaluated flexible dose-response analyses using restricted quadratic spline models. We updated a previously published relative risk of coronary heart disease mortality from a dose-response meta-analysis per a doubling of water iAs (e.g., from 10 to 20 μg/L) with our results from NHANES 2003-2014, assuming all iAs exposure came from drinking water.

RESULTS

A total of 77 fatal heart disease events occurred (median follow-up time 75 months). The adjusted HRs (95% CI) of heart disease mortality for an increase in urinary total arsenic and DMA corresponding to the interquartile range were 1.20 (0.83, 1.74) and 1.18 (0.68, 2.05), respectively. Restricted quadratic splines indicate a significant association between increasing urinary total arsenic and the HR of fatal heart disease for all participants at the lowest exposure levels <4.5 μg/L. The updated pooled relative risk of coronary heart disease mortality per doubling of water iAs (μg/L) was 1.16 (95% CI 1.07, 1.25).

CONCLUSIONS

Despite a small number of events, relatively short follow-up time, and high analytical limits of detection for urinary arsenic species, iAs exposure at low-to moderate-levels is consistent with increased heart disease mortality in NHANES 2003-2014 although the associations were only significant in flexible dose-response models.

Consequences of Depletion of Susceptibles for Hazard Ratio Estimators Based on Propensity Scores

Supplemental Digital Content is available in the text.

We use simulated data to examine the consequences of depletion of susceptibles for hazard ratio (HR) estimators based on a propensity score (PS). First, we show that the depletion of susceptibles attenuates marginal HRs toward the null by amounts that increase with the incidence of the outcome, the variance of susceptibility, and the impact of susceptibility on the outcome. If susceptibility is binary then the Bross bias multiplier, originally intended to quantify bias in a risk ratio from a binary confounder, also quantifies the ratio of the instantaneous marginal HR to the conditional HR as susceptibles are depleted differentially. Second, we show how HR estimates that are conditioned on a PS tend to be between the true conditional and marginal HRs, closer to the conditional HR if treatment status is strongly associated with susceptibility and closer to the marginal HR if treatment status is weakly associated with susceptibility. We show that associations of susceptibility with the PS matter to the marginal HR in the treated (ATT) though not to the marginal HR in the entire cohort (ATE). Third, we show how the PS can be updated periodically to reduce depletion-of-susceptibles bias in conditional estimators. Although marginal estimators can hit their ATE or ATT targets consistently without updating the PS, we show how their targets themselves can be misleading as they are attenuated toward the null. Finally, we discuss implications for the interpretation of HRs and their relevance to underlying scientific and clinical questions. See video Abstract: http://links.lww.com/EDE/B727.

A Bayesian hierarchical approach to account for left-censored and missing radiation doses prone to classical measurement error when analyzing lung cancer mortality due to γ-ray exposure in the French cohort of uranium miners

Epidemiological data on cohorts of occupationally exposed uranium miners are currently used to assess health risks associated with chronic exposure to low doses of ionizing radiation. Nevertheless, exposure uncertainty is ubiquitous and questions the validity of statistical inference in these cohorts. This paper highlights the flexibility and relevance of the Bayesian hierarchical approach to account for both missing and left-censored (i.e. only known to be lower than a fixed detection limit) radiation doses that are prone to measurement error, when estimating radiation-related risks. Up to the authors' knowledge, this is the first time these three sources of uncertainty are dealt with simultaneously in radiation epidemiology. To illustrate the issue, this paper focuses on the specific problem of accounting for these three sources of uncertainty when estimating the association between occupational exposure to low levels of γ-radiation and lung cancer mortality in the post-55 sub-cohort of French uranium miners. The impact of these three sources of dose uncertainty is of marginal importance when estimating the risk of death by lung cancer among French uranium miners. The corrected excess hazard ratio (EHR) is 0.81 per 100 mSv (95% credible interval: [0.28; 1.75]). Interestingly, even if the 95% credible interval of the corrected EHR is wider than the uncorrected one, a statistically significant positive association remains between γ-ray exposure and the risk of death by lung cancer, after accounting for dose uncertainty. Sensitivity analyses show that the results obtained are robust to different assumptions. Because of its flexible and modular nature, the Bayesian hierarchical models proposed in this work could be easily extended to account for high proportions of missing and left-censored dose values or exposure data, prone to more complex patterns of measurement error.

Exposure measurement error in air pollution studies: A framework for assessing shared, multiplicative measurement error in ensemble learning estimates of nitrogen oxides

BACKGROUND

Increasingly ensemble learning-based spatiotemporal models are being used to estimate residential air pollution exposures in epidemiological studies. While these machine learning models typically have improved performance, they suffer from exposure measurement error that is inherent in all models. Our objective is to develop a framework to formally assess shared, multiplicative measurement error (SMME) in our previously published three-stage, ensemble learning-based nitrogen oxides (NO_x) model to identify its spatial and temporal patterns and predictors.

METHODS

By treating the ensembles as an external dosimetry system, we quantified shared and unshared, multiplicative and additive (SUMA) measurement error components in our exposure model. We used generalized additive models (GAMs) with a smooth term for location to identify geographic locations with significantly elevated SMME and explain their spatial and temporal determinants.

RESULTS

We found evidence of significant shared and unshared multiplicative error (p < 0.0001) in our ensemble-learning based spatiotemporal NO_x model predictions. Unshared multiplicative error was 26 times larger than SMME. We observed significant geographic (p < 0.0001) and temporal variation in SMME with the majority (43%) of predictions with elevated SMME occurring in the earliest time-period (1992-2000). Densely populated urban prediction regions with complex air pollution sources generally exhibited highest odds of elevated SMME.

CONCLUSIONS

We developed a novel statistical framework to formally evaluate the magnitude and drivers of SMME in ensemble learning-based exposure models. Our framework can be used to inform building future improved exposure models.

Reanalysis of Reported Associations of Beryllium and Lung Cancer in a Large Occupational Cohort

OBJECTIVE

The aim of this study was to verify and extend reported associations of beryllium exposure and lung cancer by reanalyzing data from a large occupational cohort at three beryllium processing plants.

METHODS

We used standardization and Poisson regression to evaluate the effect of cumulative and maximum exposure, unlagged, and lagged 10 years, adjusting for plant, employment tenure, and date of hire. Exposure was modeled either categorically or continuously using splines.

RESULTS

Categorical analyses displayed previously reported effect patterns, but not the spline analysis, which provided a more consistent picture of risk across all analyzed groups.

CONCLUSIONS

We found modestly but monotonically increasing risk in the full cohort, by duration of tenure, and within most subgroups defined by plant and date of hire. Regression-based point-wise confidence bands, however, did not clearly separate risk for low versus high exposure groups.

Shared and unshared exposure measurement error in occupational cohort studies and their effects on statistical inference in proportional hazards models

Exposure measurement error represents one of the most important sources of uncertainty in epidemiology. When exposure uncertainty is not or only poorly accounted for, it can lead to biased risk estimates and a distortion of the shape of the exposure-response relationship. In occupational cohort studies, the time-dependent nature of exposure and changes in the method of exposure assessment may create complex error structures. When a method of group-level exposure assessment is used, individual worker practices and the imprecision of the instrument used to measure the average exposure for a group of workers may give rise to errors that are shared between workers, within workers or both. In contrast to unshared measurement error, the effects of shared errors remain largely unknown. Moreover, exposure uncertainty and magnitude of exposure are typically highest for the earliest years of exposure. We conduct a simulation study based on exposure data of the French cohort of uranium miners to compare the effects of shared and unshared exposure uncertainty on risk estimation and on the shape of the exposure-response curve in proportional hazards models. Our results indicate that uncertainty components shared within workers cause more bias in risk estimation and a more severe attenuation of the exposure-response relationship than unshared exposure uncertainty or exposure uncertainty shared between individuals. These findings underline the importance of careful characterisation and modeling of exposure uncertainty in observational studies.

Disease fatality and bias in survival cohorts

OBJECTIVES

Simulate how the effect of exposure on disease occurrence and fatality influences the presence and magnitude of bias in survivor cohorts, motivated by an actual survivor cohort under study.

METHODS

We simulated a cohort of 50,000 subjects exposed to a disease-causing exposure over time and followed forty years, where disease incidence was the outcome of interest. We simulated this 'inception' cohort under different assumptions about the effect of exposure on disease occurrence and fatality after disease occurrence. We then created a corresponding 'survivor' (or 'cross-sectional') cohort, where cohort enrollment took place at a specific date after exposure began in the inception cohort; subjects dying prior to that enrollment date were excluded. The disease of interest caused all deaths in our simulations, but was not always fatal. In the survivor cohort, person-time at risk began before enrollment for all subjects who did not die prior to enrollment. We compared exposure-disease associations in each inception cohort to those in corresponding survivor cohorts to determine how different assumptions impacted bias in the survivor cohorts. All subjects in both inception and survivor cohorts were considered equally susceptible to the effect of exposure in causing disease. We used Cox proportional hazards regression to calculate effect measures.

RESULTS

There was no bias in survivor cohort estimates when case fatality among diseased subjects was independent of exposure. This was true even when the disease was highly fatal and more highly exposed subjects were more likely to develop disease and die. Assuming a positive exposure-response in the inception cohort, survivor cohort rate ratios were biased downwards when case fatality was greater with higher exposure.

CONCLUSIONS

Survivor cohort effect estimates for fatal outcomes are not always biased, although precision can decrease.