The effect of modifiable risk factors on geographic mortality differentials : a modelling study

Background Australian mortality rates are higher in regional and remote areas than in major cities. The degree to which this is driven by variation in modifiable risk factors is unknown.

Methods We applied a risk prediction equation incorporating smoking, cholesterol and blood pressure to a national, population based survey to project all-causes mortality risk by geographic region. We then modelled life expectancies at different levels of mortality risk by geographic region using a risk percentiles model. Finally we set high values of each risk factor to a target level and modelled the subsequent shift in the population to lower levels of mortality risk and longer life expectancy.

Results Survival is poorer in both Inner Regional and Outer Regional/Remote areas compared to Major Cities for men and women at both high and low levels of predicted mortality risk. For men smoking, high cholesterol and high systolic blood pressure were each associated with the mortality difference between Major Cities and Outer Regional/Remote areas--accounting for 21.4%, 20.3% and 7.7% of the difference respectively. For women smoking and high cholesterol accounted for 29.4% and 24.0% of the difference respectively but high blood pressure did not contribute to the observed mortality differences. The three risk factors taken together accounted for 45.4% (men) and 35.6% (women) of the mortality difference. The contribution of risk factors to the corresponding differences for inner regional areas was smaller, with only high cholesterol and smoking contributing to the difference in men-- accounting for 8.8% and 6.3% respectively-- and only smoking contributing to the difference in women--accounting for 12.3%.

Conclusions These results suggest that health intervention programs aimed at smoking, blood pressure and total cholesterol could have a substantial impact on mortality inequities for Outer Regional/Remote areas. Background: Australian mortality rates are higher in regional and remote areas than in major cities. The degree to which this is driven by variation in modifiable risk factors is unknown. Methods. We applied a risk prediction equation incorporating smoking, cholesterol and blood pressure to a national, population based survey to project all-causes mortality risk by geographic region. We then modelled life expectancies at different levels of mortality risk by geographic region using a risk percentiles model. Finally we set high values of each risk factor to a target level and modelled the subsequent shift in the population to lower levels of mortality risk and longer life expectancy. Results: Survival is poorer in both Inner Regional and Outer Regional/Remote areas compared to Major Cities for men and women at both high and low levels of predicted mortality risk. For men smoking, high cholesterol and high systolic blood pressure were each associated with the mortality difference between Major Cities and Outer Regional/Remote areas - accounting for 21.4%, 20.3% and 7.7% of the difference respectively. For women smoking and high cholesterol accounted for 29.4% and 24.0% of the difference respectively but high blood pressure did not contribute to the observed mortality differences. The three risk factors taken together accounted for 45.4% (men) and 35.6% (women) of the mortality difference. The contribution of risk factors to the corresponding differences for inner regional areas was smaller, with only high cholesterol and smoking contributing to the difference in men - accounting for 8.8% and 6.3% respectively - and only smoking contributing to the difference in women - accounting for 12.3%. Conclusions: These results suggest that health intervention programs aimed at smoking, blood pressure and total cholesterol could have a substantial impact on mortality inequities for Outer Regional/Remote areas.

Computing of high breakdown regression estimators without sorting on graphics processing units

We present an approach to computing high-breakdown regression estimators in parallel on graphics processing units (GPU).We show that sorting the residuals is not necessary, and it can be substituted by calculating the median. We present and compare various methods to calculate the median and order statistics on GPUs. We introduce an alternative method based on the optimization of a convex function, and showits numerical superiority when calculating the order statistics of very large arrays on GPUs.

Does the choice of neighbourhood supermarket access measure influence associations with individual-level fruit and vegetable consumption? A case study from Glasgow

Background
Previous studies have provided mixed evidence with regards to associations between food store access and dietary outcomes. This study examines the most commonly applied measures of locational access to assess whether associations between supermarket access and fruit and vegetable consumption are affected by the choice of access measure and scale.

Method
Supermarket location data from Glasgow, UK (n = 119), and fruit and vegetable intake data from the 'Health and Well-Being' Survey (n = 1041) were used to compare various measures of locational access. These exposure variables included proximity estimates (with different points-of-origin used to vary levels of aggregation) and density measures using three approaches (Euclidean and road network buffers and Kernel density estimation) at distances ranging from 0.4 km to 5 km. Further analysis was conducted to assess the impact of using smaller buffer sizes for individuals who did not own a car. Associations between these multiple access measures and fruit and vegetable consumption were estimated using linear regression models.

Results
Levels of spatial aggregation did not impact on the proximity estimates. Counts of supermarkets within Euclidean buffers were associated with fruit and vegetable consumption at 1 km, 2 km and 3 km, and for our road network buffers at 2 km, 3 km, and 4 km. Kernel density estimates provided the strongest associations and were significant at a distance of 2 km, 3 km, 4 km and 5 km. Presence of a supermarket within 0.4 km of road network distance from where people lived was positively associated with fruit consumption amongst those without a car (coef. 0.657; s.e. 0.247; p0.008).

Conclusions
The associations between locational access to supermarkets and individual-level dietary behaviour are sensitive to the method by which the food environment variable is captured. Care needs to be taken to ensure robust and conceptually appropriate measures of access are used and these should be grounded in a clear a priori reasoning.

The epidemiology of cryptosporidiosis in New Zealand, 1997-2006

Aims New Zealand has a high incidence of cryptosporidiosis compared to other developed countries. This study aimed to describe the epidemiology of this disease in detail and to identify potential risk factors.

Methods We analysed anonymous cryptosporidiosis notification (1997–2006) and hospitalisation data (1996–2006). Cases were designated as “urban” or “rural” and assigned a deprivation level based on their home address. Association between disease rates and animal density was studied using a simple linear regression model, at the territorial authority level.

Results Over the 10-year period 1997–2006, the average annual rate of notified cryptosporidiosis was 22.0 cases per 100,000 population. The number of hospitalisations was equivalent to 3.6% of the notified cases. There was only 1 reported fatality. The annual incidence of infection appeared fairly stable, but showed marked seasonality with a peak rate in spring (September–November in New Zealand). The highest rates were among Europeans, children 0–9 years of age, and those living in low deprivation areas. Notification rates showed large geographic variations, with rates in rural areas 2.8 times higher than in urban areas, and with rural areas also experiencing the most pronounced spring peak. At the territorial authority (TA) level, rates were also correlated with farm animal density.

Conclusions Most transmission of Cryptosporidium in New Zealand appears to be zoonotic: from farm animals to humans. Prevention should focus on reducing transmission in rural setting, though more research is needed to identify which strategies are likely to be most effective in that environment.

The epidemiology of giardiasis in New Zealand, 1997-2006

Aims New Zealand has a higher incidence rate of giardiasis than other developed countries. This study aimed to describe the epidemiology of this disease in detail and to identify potential risk factors.

Methods We analysed anonymous giardiasis notification (1997–2006) and hospitalisation data (1990–2006). Cases were designated as urban or rural and assigned a deprivation level based on their home address. Association between disease rates and animal density was studied using a simple linear regression model, at the territorial authority (TA) level.

Results Over the 10-year period 1997–2006 the average annual rate of notified giardiasis was 44.1 cases per 100,000 population. The number of hospitalisations was equivalent to 1.7% of the notified cases. There were 2 reported fatalities. The annual incidence of notified cases declined over this period whereas hospitalisations remained fairly constant. Giardiasis showed little seasonality. The highest rates were among children 0–9 years old, those 30–39 years old, Europeans, and those living in low deprivation areas. Notification rates were slightly higher in rural areas. The correlation between giardiasis and farm animal density was not significant at the TA level.

Conclusions The public health importance of giardiasis to New Zealand mainly comes from its relatively high rates in this country. The distribution of cases is consistent with largely anthroponotic (human) reservoirs, with a relatively small contribution from zoonotic sources in rural environments and a modest contribution from overseas travel. Prevention efforts could include continuing efforts to improve hand washing, nappy handling, and other hygiene measures and travel health advice relating to enteric infections.

A hybrid neural network model for noisy data regression

A hybrid neural network model, based on the fusion of fuzzy adaptive resonance theory (FA ART) and the general regression neural network (GRNN), is proposed in this paper. Both FA and the GRNN are incremental learning systems and are very fast in network training. The proposed hybrid model, denoted as GRNNFA, is able to retain these advantages and, at the same time, to reduce the computational requirements in calculating and storing information of the kernels. A clustering version of the GRNN is designed with data compression by FA for noise removal. An adaptive gradient-based kernel width optimization algorithm has also been devised. Convergence of the gradient descent algorithm can be accelerated by the geometric incremental growth of the updating factor. A series of experiments with four benchmark datasets have been conducted to assess and compare effectiveness of GRNNFA with other approaches. The GRNNFA model is also employed in a novel application task for predicting the evacuation time of patrons at typical karaoke centers in Hong Kong in the event of fire. The results positively demonstrate the applicability of GRNNFA in noisy data regression problems.

Fusion of GRNN and FA for online noisy data regression

A new online neural-network-based regression model for noisy data is proposed in this paper. It is a hybrid system combining the Fuzzy ART (FA) and General Regression Neural Network (GRNN) models. Both the FA and GRNN models are fast incremental learning systems. The proposed hybrid model, denoted as GRNNFA-online, retains the online learning properties of both models. The kernel centers of the GRNN are obtained by compressing the training samples using the FA model. The width of each kernel is then estimated by the K-nearest-neighbors (kNN) method. A heuristic is proposed to tune the value of Kof the kNN dynamically based on the concept of gradient-descent. The performance of the GRNNFA-online model was evaluated using two benchmark datasets, i.e., OZONE and Friedman#1. The experimental results demonstrated the convergence of the prediction errors. Bootstrapping was employed to assess the performance statistically. The final prediction errors are analyzed and compared with those from other systems.Bootstrapping was employed to assess the performance statistically. The final prediction errors are analyzed and compared with those from other systems.

Does parkland influence walking? The relationship between area of parkland and walking trips in Melbourne, Australia

Background : Using two different measures of park area, at three buffer distances, we sought to investigate the ways in which park area and proximity to parks, are related to the frequency of walking (for all purposes) in Australian adults. Little previous research has been conducted in this area, and results of existing research have been mixed.

Methods : Residents of 50 urban areas in metropolitan Melbourne, Australia completed a physical activity survey (n = 2305). Respondents reported how often they walked for >=10 minutes in the previous month. Walking frequency was dichotomised to 'less than weekly' (less than 1/week) and 'at least weekly' (1/week or more). Using Geographic Information Systems, Euclidean buffers were created around each respondent's home at three distances: 400metres (m), 800 m and 1200 m. Total area of parkland in each person's buffer was calculated for the three buffers. Additionally, total area of 'larger parks', (park space >= park with Australian Rules Football oval (17,862 m2)), was calculated for each set of buffers. Area of park was categorised into tertiles for area of all parks, and area of larger parks (the lowest tertile was used as the reference category). Multilevel logistic regression, with individuals nested within areas, was used to estimate the effect of area of parkland on walking frequency.

Results : No statistically significant associations were found between walking frequency and park area (total and large parks) within 400 m of respondent's homes. For total park area within 800 m, the odds of walking at least weekly were lower for those in the mid (OR 0.65, 95% CI 0.46-0.91) and highest (OR 0.65, 95% CI 0.44-0.95) tertile of park area compared to those living in areas with the least amount of park area. Similar results were observed for total park area in the 1200 m buffers. When only larger parks were investigated, again more frequent walking was less likely when respondents had access to a greater amount of park area.

Conclusions : In this study we found that more park area in residential environments reduced the odds of walking more frequently. Other area characteristics such as street connectivity and destinations may underlay these associations by negatively correlating with park area.

Global non-smooth optimization in robust multivariate regression

Robust regression in statistics leads to challenging optimization problems. Here, we study one such problem, in which the objective is non-smooth, non-convex and expensive to calculate. We study the numerical performance of several derivative-free optimization algorithms with the aim of computing robust multivariate estimators. Our experiences demonstrate that the existing algorithms often fail to deliver optimal solutions. We introduce three new methods that use Powell's derivative-free algorithm. The proposed methods are reliable and can be used when processing very large data sets containing outliers.

Modeling of a stereo vision system using a genetic algorithm based fuzzy linear regression

In this paper a fuzzy linear regression (FLR) model integrated with a genetic algorithm (GA) is proposed. The proposed GA-FLR model is applied to modeling of a stereo vision system. A set of empirical data from stereo vision object measurement is collected based on the full factorial design technique. Three regression models, namely ordinary least-squares regression (OLS), FLR, and GA-FLR, are developed, and with their performances compared. The results show that the proposed GA-FLR model performs better than OLS and FLR in modeling of a stereo vision system.