Search

Background Local policy makers increasingly need information on health-related indicators at smaller geographic levels like districts or neighbourhoods. Although more large data sources have become available, direct estimates of the prevalence of a health-related indicator cannot be produced for neighbourhoods for which only small samples or no samples are available. Small area estimation provides a solution, but unit-level models for binary-valued outcomes that can handle both non-linear effects of the predictors and spatially correlated random effects in a unified framework are rarely encountered. Methods We used data on 26 binary-valued health-related indicators collected on 387,195 persons in the Netherlands. We associated the health-related indicators at the individual level with a set of 12 predictors obtained from national registry data. We formulated a structured additive regression model for small area estimation. The model captured potential non-linear relations between the predictors and the outcome through additive terms in a functional form using penalized splines and included a term that accounted for spatially correlated heterogeneity between neighbourhoods. The registry data were used to predict individual outcomes which in turn are aggregated into higher geographical levels, i.e. neighbourhoods. We validated our method by comparing the estimated prevalences with observed prevalences at the individual level and by comparing the estimated prevalences with direct estimates obtained by weighting methods at municipality level. Results We estimated the prevalence of the 26 health-related indicators for 415 municipalities, 2599 districts and 11,432 neighbourhoods in the Netherlands. We illustrate our method on overweight data and show that there are distinct geographic patterns in the overweight prevalence. Calibration plots show that the estimated prevalences agree very well with observed prevalences at the individual level. The estimated prevalences agree reasonably well with the direct estimates at the municipal level. Conclusions Structured additive regression is a useful tool to provide small area estimates in a unified framework. We are able to produce valid nationwide small area estimates of 26 health-related indicators at neighbourhood level in the Netherlands. The results can be used for local policy makers to make appropriate health policy decisions.

We studied the spatial distribution of cancer incidence rates around a large steel plant and its association with historical exposure. The study population was close to 600,000. The incidence data was collected for 1995–2006. From historical emission data the air pollution concentrations for polycyclic aromatic hydrocarbons (PAH) and metals were modelled. Data were analyzed using Bayesian hierarchical Poisson regression models. The standardized incidence ratio (SIR) for lung cancer was up to 40% higher than average in postcodes located in two municipalities adjacent to the industrial area. Increased incidence rates could partly be explained by differences in socioeconomic status (SES). In the highest exposure category (approximately 45,000 inhabitants) a statistically significant increased relative risk (RR) of 1.21 (1.01–1.43) was found after adjustment for SES. The elevated RRs were similar for men and women. Additional analyses in a subsample of the population with personal smoking data from a recent survey suggested that the observed association between lung cancer and plant emission, after adjustment for SES, could still be caused by residual confounding. Therefore, we cannot indisputably conclude that past emissions from the steel plant have contributed to the increased risk of lung cancer.

In the HYENA study (HYpertension and Exposure to Noise near Airports) noise annoyances due to aircraft and road traffic noise were assessed in subjects that lived in the vicinity of 6 major European airports using the 11-point ICBEN scale (International Commission on Biological Effects of Noise). A distinction was made between the annoyance during the day and during the night. Lden and Lnight were considered as indicators of noise exposure. Pooled data analyses showed clear exposure–response relationships between the noise level and the noise annoyance for both exposures. The exposure–response curves for road noise were congruent with the EU standard curves used for predicting the number of highly noise annoyed subjects in European communities. Annoyance ratings due to aircraft noise, however, were higher than predicted by the EU standard curves. The data supports other findings suggesting that the people's attitude towards aircraft noise has changed over the years, and that the EU standard curve for aircraft noise should be modified. Keywords: Aircraft noise, Road traffic noise, Annoyance, Exposure–response curve, Change of annoyance

As a part of the EXPOLIS study, a stochastic exposure-modeling framework was developed. The framework is useful to compare exposure distributions of different (sub-) populations or different scenarios, and to gain insight into population exposure distributions and exposure determinants. It was implemented in an MS-Excel workbook using @Risk add-on software. Basic concept of the framework is that time-weighted average exposure is a sum of partial exposures in the visited microenvironments. Partial exposure is determined by the concentration and the time spent in the microenvironment. In the absence of data, indoor concentrations are derived as a function of ambient concentrations, effective penetration rates and contribution of indoor sources. Framework input parameters are described by probability distributions. A lognormal distribution is assumed for the microenvironment concentrations and for the contribution of indoor sources, and a beta distribution for the time spent in a microenvironment and for the penetration factor. Mean and standard deviation values parameterize the distributions. In this paper, Latin Hypercube sampling is used for the input distributions. The outcome of the framework is an estimate of the population exposure distribution for the selected air pollutant. The framework is best suited for averaging times from 24 h upwards. Sensitivity analyses can be performed to determine the most influential factors of exposure. The application of the framework is illustrated in two examples. The EXPOLIS PM(2.5) example uses microenvironment measurement and time-activity data from the EXPOLIS study to model PM(2.5) population exposure distributions in four European cities. The results are compared to the observed personal exposure distributions from the same study. The Dutch PM(10) example uses input data from several (Dutch) databases and from literature, and shows a more complex application of the framework for comparison of scenarios and subpopulations.

Objectives The effect modifying impact of annoyance due to aircraft noise and road traffic noise on the relationships between the aircraft noise level and road traffic noise level on the prevalence of hypertension was investigated in 4861 subjects of the HYENA study (HYpertension and Exposure to Noise near Airports). Methods Different models were investigated either including the noise level and noise annoyance variables separately, or simultaneously, or together with an interaction term referring to the same noise source for the noise level and the noise annoyance. Results Significant effect modification was found with respect to the association between aircraft noise and hypertension. The association was stronger in more annoyed subjects. No clear interaction was found with respect to road traffic noise. The comparison of the magnitude of the main effects (per standard deviation or inter-quartile range) of noise level and noise annoyance variables revealed stronger associations with hypertension for the noise levels. Conclusion There is some indication that the noise level has a stronger predictive meaning for the relationship between noise exposure and hypertension than the reported noise annoyance (main effects). The results from the Hyena study support the hypothesis that noise annoyance acts as an effect modifier of the relationship between the noise level and hypertension.

A group of male amateur cyclists was studied in the summer of 1991 in the east of the Netherlands. Lung function was measured before and after training sessions or competitive races on a number of occasions. Continuous heart rate monitors were employed to document exercise levels. Heart rate averaged 161 beats/min during training, and 176 beats/min during races. Exercise duration averaged 75 min. Ozone concentrations during exercise were obtained from the nearest stations of the National Monitoring Network. The difference between pre- and postexercise lung function values was related to these ozone concentrations. Ozone concentrations were low on most occasions with an average of 87 micrograms/m3 and a maximum of 195 micrograms/m3. The difference between pre- and postexercise lung function was found to be negatively related to the ozone concentration during exercise. When all observations obtained at ozone concentrations higher than 120 micrograms/m3 were removed from the analysis, the relationship with ozone was still significant. The data also suggested that effects of ozone on lung function were stronger in midsummer than in the late summer. The difference between pre- and postexercise acute symptoms was positively related to ozone, for shortness of breath, chest tightness, and wheeze. Cough and eye irritation were not related to ozone. These results indicate that in healthy young men vigorously exercising outdoors, ozone is related to lung function changes and acute respiratory symptom changes at low levels of exposure.