Extended two-stage designs for environmental research

BACKGROUND

The two-stage design has become a standard tool in environmental epidemiology to model multi-location data. However, its standard form is rather inflexible and poses important limitations for modelling complex risks associated with environmental factors. In this contribution, we illustrate multiple design extensions of the classical two-stage method, all implemented within a unified analytic framework.

METHODS

We extended standard two-stage meta-analytic models along the lines of linear mixed-effects models, by allowing location-specific estimates to be pooled through flexible fixed and random-effects structures. This permits the analysis of associations characterised by combinations of multivariate outcomes, hierarchical geographical structures, repeated measures, and/or longitudinal settings. The analytic framework and inferential procedures are implemented in the R package mixmeta.

RESULTS

The design extensions are illustrated in examples using multi-city time series data collected as part of the National Morbidity, Mortality and Air Pollution Study (NMMAPS). Specifically, four case studies demonstrate applications for modelling complex associations with air pollution and temperature, including non-linear exposure-response relationships, effects clustered at multiple geographical levels, differential risks by age, and effect modification by air conditioning in a longitudinal analysis.

CONCLUSIONS

The definition of several design extensions of the classical two-stage design within a unified framework, along with its implementation in freely-available software, will provide researchers with a flexible tool to address novel research questions in two-stage analyses of environmental health risks.

Extreme temperatures and mortality in Kuwait: Who is vulnerable?

BACKGROUND

Previous climate change temperature-related health studies have been performed mostly in western countries with relatively cooler temperatures than the Gulf region. Regions that are inherently hot, like Kuwait, are witnessing soaring temperatures unlike ever before. Meanwhile, Kuwait and other Gulf countries are unique in their demographic profiles due to the large number of non-national migrant workers.

OBJECTIVE

To examine the associations of hot and cold temperature extremes on the risk of mortality across gender, age groups and nationality in Kuwait.

METHODS

We investigated daily variations of all-cause non-accidental and cardiovascular mortality death counts and ambient temperatures from 2010 to 2016 in a time-series design using a negative binomial distribution. The temperature lag was modeled with distributed lag non-linear models.

RESULTS

A total of 33,472 all-cause non-accidental deaths happened during the study period. For the extreme hot temperatures and over the entire lag period, comparing the 99th percentile of temperature to the minimum mortality temperature, the risk of dying among males was 2.08 (95% CI: 1.23-3.52). Among non-Kuwaitis, males and working age group (15-64 year) had relative risks of death from extreme hot temperatures of 2.90 (1.42-5.93), and 2.59 (1.20-5.59), respectively. For extreme cold temperatures and over the entire lag period, comparing the 1st percentile of temperature to the minimum mortality temperature, the relative risk of death among Kuwaitis was 2.03 (1.05-3.93). Elderly Kuwaitis (65+ year) exposed to extreme cold temperatures had a relative risk of 2.75 (1.16-6.52).

CONCLUSIONS

Certain subpopulations in Kuwait are vulnerable to extreme temperatures with doubling to tripling risk of mortality. Nationality is an important effect modifier in temperature-related mortality studies in Kuwait and possibly the Gulf region. To the best of our knowledge, we are the first study to examine specific subpopulation vulnerabilities to temperature in this region. Our findings could carry a potential for broader insight into similar hyper-arid and hot regions.

The effects of temperature on short-term mortality risk in Kuwait: A time-series analysis

BACKGROUND

In light of climate change, health risks are expected to be exacerbated by extreme temperatures. Many studies showed that high and low ambient temperatures are associated with increased short-term mortality risk, but little is known about these risks in Kuwait and the Gulf Region.

OBJECTIVE

To examine the dose-response relationship between 24-h average ambient temperatures and daily mortality risk in Kuwait.

METHODS

We gathered mortality and meteorological data from 2010 to 2016 in Kuwait. We did a time-series analysis using a negative binomial distribution, and studied the lag effects of temperature with distributed lag non-linear models.

RESULTS

A total of 33,574 all-cause non-accidental deaths were analyzed. The relationship was shown to be non-linear. Overall relative risks of death comparing the 1st percentile (10.9 °C) and the 99th percentile (42.7 °C) to the optimum temperature were 1.67 (1.02-2.73), and 1.65 (1.09-2.48), respectively. Cold effects persisted for 9 days, while the effects of hot temperatures were the highest at lag 0 and only persisted for a week. Adjusting for PM₁₀ and ozone did not change the temperature-mortality estimates.

CONCLUSION

Our findings show evidence that there is a statistically significant positive association between temperature extremes and mortality in Kuwait. The evidence has significant implications in assessing climate vulnerability and provides insight into environmental challenges in an inherently hot and arid region.

The short-term influence of temperature on daily mortality in the temperate climate of Montreal, Canada

The purpose of this study was to determine whether short-term changes in ambient temperature were associated with daily mortality among persons who lived in Montreal, Canada, and who died in the urban area between 1984 and 2007. We made use of newly developed distributed lag non-linear Poisson models, constrained to a 30 day lag period, and we adjusted for temporal trends and nitrogen dioxide and ozone. We found a strong non-linear association with high daily maximum temperatures showing an apparent threshold at about 27°C; this association persisted until about lag 5 days. For example, we found across all lag periods that daily non-accidental mortality increased by 28.4% (95% confidence interval: 13.8-44.9%) when temperatures increased from 22.5 to 31.8°C (75-99th percentiles). This association was essentially invariant to different smoothers for time. Cold temperatures were not found to be associated with daily mortality over 30 days, although there was some evidence of a modest increased risk from 2 to 5 days. The adverse association with colder temperatures was sensitive to the smoother for time. For cardio-respiratory mortality we found increased risks for higher temperatures of a similar magnitude to that of non-accidental mortality but no effects at cold temperatures.

Bayesian bivariate meta-analysis of diagnostic test studies using integrated nested Laplace approximations

For bivariate meta-analysis of diagnostic studies, likelihood approaches are very popular. However, they often run into numerical problems with possible non-convergence. In addition, the construction of confidence intervals is controversial. Bayesian methods based on Markov chain Monte Carlo (MCMC) sampling could be used, but are often difficult to implement, and require long running times and diagnostic convergence checks. Recently, a new Bayesian deterministic inference approach for latent Gaussian models using integrated nested Laplace approximations (INLA) has been proposed. With this approach MCMC sampling becomes redundant as the posterior marginal distributions are directly and accurately approximated. By means of a real data set we investigate the influence of the prior information provided and compare the results obtained by INLA, MCMC, and the maximum likelihood procedure SAS PROC NLMIXED. Using a simulation study we further extend the comparison of INLA and SAS PROC NLMIXED by assessing their performance in terms of bias, mean-squared error, coverage probability, and convergence rate. The results indicate that INLA is more stable and gives generally better coverage probabilities for the pooled estimates and less biased estimates of variance parameters. The user-friendliness of INLA is demonstrated by documented R-code.

Particulate air pollution and health effects for cardiovascular and respiratory causes in Temuco, Chile: a wood-smoke-polluted urban area

Temuco is one of the most highly wood-smoke-polluted cities in the world. Its population in 2004 was 340,000 inhabitants with 1587 annual deaths, of which 24% were due to cardiovascular and 11% to respiratory causes. For hospital admissions, cardiovascular diseases represented 6% and respiratory diseases 13%. Emergency room visits for acute respiratory infections represented 28%. The objective of the study presented here was to determine the relationship between air pollution from particulate matter less than or equal to 10 microm in aerodynamic diameter (PM10; mostly PM2.5, or particulate matter <2.5 microm in aerodynamic diameter) and health effects measured as the daily number of deaths, hospital admissions, and emergency room visits for cardiovascular, respiratory, and acute respiratory infection (ARI) diseases. The Air Pollution Health Effects European Approach (APHEA2) protocol was followed, and a multivariate Poisson regression model was fitted, controlling for trend, seasonality, and confounders for Temuco during 1998-2006. The results show that PM10 had a significant association with daily mortality and morbidity, with the elderly (population >65 yr of age) being the group that presented the greatest risk. The relative risk for respiratory causes, with an increase of 100 microg/m3 of PM10, was 1.163 with a 95% confidence interval (CI) of 1.057-1.279 for mortality, 1.137 (CI 1.096-1.178) for hospital admissions, and 1.162 for ARI (CI 1.144-1.181). There is evidence in Temuco of positive relationships between ambient particulate levels and mortality, hospital admissions, and ARI for cardiovascular and respiratory diseases. These results are consistent with those of comparable studies in other similar cities where wood smoke is the most important air pollution problem.

A new multipollutant, no-threshold air quality health index based on short-term associations observed in daily time-series analyses

Air quality indices currently in use have been criticized because they do not capture additive effects of multiple pollutants, or reflect the apparent no-threshold concentration-response relationship between air pollution and health. We propose a new air quality health index (AQHI), constructed as the sum of excess mortality risk associated with individual pollutants from a time-series analysis of air pollution and mortality in Canadian cities, adjusted to a 0-10 scale, and calculated hourly on the basis of trailing 3-hr average pollutant concentrations. Extensive sensitivity analyses were conducted using alternative combinations of pollutants from single and multipollutant models. All formulations considered produced frequency distributions of the daily maximum AQHI that were right-skewed, with modal values of 3 or 4, and less than 10% of values at 7 or above on the 10-point scale. In the absence of a gold standard and given the uncertainty in how to best reflect the mix of pollutants, we recommend a formulation based on associations of nitrogen dioxide, ozone, and particulate matter of median aerodynamic diameter less than 2.5 microm with mortality from single-pollutant models. Further sensitivity analyses revealed good agreement of this formulation with others based on alternative sources of coefficients drawn from published studies of mortality and morbidity. These analyses provide evidence that the AQHI represents a valid approach to formulating an index with the objective of allowing people to judge the relative probability of experiencing adverse health effects from day to day. Together with health messages and a graphic display, the AQHI scale appears promising as an air quality risk communication tool.

Modelling Multivariate Outcomes in Hierarchical Data, with Application to Cluster Randomised Trials

In the cluster randomised study design, the data collected have a hierarchical structure and often include multivariate outcomes. We present a flexible modelling strategy that permits several normally distributed outcomes to be analysed simultaneously, in which intervention effects as well as individual-level and cluster-level between-outcome correlations are estimated. This is implemented in a Bayesian framework which has several advantages over a classical approach, for example in providing credible intervals for functions of model parameters and in allowing informative priors for the intracluster correlation coefficients. In order to declare such informative prior distributions, and fit models in which the between-outcome covariance matrices are constrained, priors on parameters within the covariance matrices are required. Careful specification is necessary however, in order to maintain non-negative definiteness and symmetry between the different outcomes. We propose a novel solution in the case of three multivariate outcomes, and present a modified existing approach and novel alternative for four or more outcomes. The methods are applied to an example of a cluster randomised trial in the prevention of coronary heart disease. The modelling strategy presented would also be useful in other situations involving hierarchical multivariate outcomes.

Spatial variability of fine particle mass, components, and source contributions during the regional air pollution study in St. Louis

Community time-series epidemiology typically uses either 24-hour integrated particulate matter (PM) concentrations averaged across several monitors in a city or data obtained at a central monitoring site to relate PM concentrations to human health effects. If the day-to-day variations in 24-hour integrated concentrations differ substantially across an urban area (i.e., daily measurements at monitors at different locations are not highly correlated), then there is a significant potential for exposure misclassification in community time-series epidemiology. If the annual average concentration differs across an urban area, then there is a potential for exposure misclassification in epidemiologic studies that use annual averages (or multi-year averages) as an index of exposure across different cities. The spatial variability in PM2.5 (particulate matter < or = 2.5 microm in aerodynamic diameter), its elemental components, and the contributions from each source category at 10 monitoring sites in St. Louis, Missouri were characterized using the ambient PM2.5 compositional data set of the Regional Air Pollution Study (RAPS) based on the Regional Air Monitoring System (RAMS) conducted between 1975 and 1977. Positive matrix factorization (PMF) was applied to each ambient PM2.5 compositional data set to estimate the contributions from the source categories. The spatial distributions of components and source contributions to PM2.5 at the 10 sites were characterized using Pearson correlation coefficients and coefficients of divergence. Sulfur and PM2.5 are highly correlated elements between all of the site pairs Although the secondary sulfate is the most highly correlated and shows the smallest spatial variability, there is a factor of 1.7 difference in secondary sulfate contributions between the highest and lowest site on average. Motor vehicles represent the next most highly correlated source component. However, there is a factor of 3.6 difference in motor vehicle contributions between the highest and lowest sites. The contributions from point source categories are much more variable. For example, the contributions from incinerators show a difference of a factor of 12.5 between the sites with the lowest and highest contributions. This study demonstrates that the spatial distributions of elemental components of PM2.5 and contributions from source categories can be highly heterogeneous within a given airshed and thus, there is the potential for exposure misclassification when a limited number of ambient PM monitors are used to represent population-average ambient exposures.