Karen Kafadar

Geographical and environmental epidemiology: methods for small-area studies

Part 1 Introduction: geographihcal epidemiology and ecological studies small-area studies - purpose and methods health and the environment - the significance of chemicals and radiation. Part 2 Data, computational methods and mapping: mortality data cancer incidence data for adults cancer incidence data for children congenital anomalies specialized registers population counts in small areas use of routine data in studies of point sources of environmental pollution socio-economic confounding use of record linkage in small-area studies confidentiality practical approaches to disease mapping estimating environmental exposures mapping environmental exposure. Part 3 Statistical methods: statistical methods for geographical correlation studies Bayesian methods for mapping disease risk statistical methods for analyzing point-source exposures some comments on methods for investigating disease risk around a point source methods for the assessment of disease clusters. Part 4 Studies of health and the environment: environmental epidemiology - a historical perspective guidelines for the investigation of clusters of adverse health events studies of diseas clustering - problems of interpretation. Part 5 Case studies: childhood leukaemia around the Sellafield nuclear plant the epidemic of respiratory cancer associated with erionite fibres in the Cappadocian region of Turkey soya bean as a risk factor of epidemic asthma the Seveso accident cancer of the larynx and lung near incinerators of waste solvents and oils in Britain a study of geographical correlations in China.

A Biweight Approach to the One-Sample Problem

Abstract A “t”-like statistic, replacing the classical mean by a biweight location estimator in the numerator and the sample variance by a corresponding variance term in the denominator, is proposed as a modification to that used by Gross (1976) and is evaluated for its efficiency in constructing confidence intervals in symmetric, stretched-tailed situations. The one-sample biweight “t” is shown, via Monte Carlo simulations, to be efficient for samples of moderate sizes (in terms of expected length of the confidence intervals). For smaller samples (size five), the sum of the biweight weights is useful in rescaling biweight “t”. For several samples of common population width, a root mean square of the variances affords greater stability when the underlying distribution is not extremely stretched-tailed.

SMOOTHING GEOGRAPHICAL DATA, PARTICULARLY RATES OF DISEASE

This paper proposes a linear smoother for geographically-defined data that consist of standardized rates (for example, adjusted for age). The smoother is viewed as a special case of one that applies to data in the form of ratios, and situations are described under which such a smoother can and cannot be useful. Its application to mortality rates due to prostate cancer in both whites and non-whites demonstrates its potential to highlight features in the data that might otherwise remain obscure. Some open questions concerning inference of apparent trends are discussed.

Get smart Colorado: impact of a mass media campaign to improve community antibiotic use.

Context: Large-scale strategies are needed to reduce overuse of antibiotics in US communities. Objectives: To evaluate the impact of a mass media campaign—“Get Smart Colorado”—on public exposure to campaign, antibiotic use, and office visit rates. Design: Nonrandomized controlled trial. Setting: Two metropolitan communities in Colorado, United States. Subjects: The general public, managed care enrollees, and physicians residing in the mass media (2.2 million persons) and comparison (0.53 million persons) communities. Intervention: The campaign consisting of paid outdoor advertising, earned media and physician advocacy ran between November 2002 and February 2003. Principal Measures: Antibiotics dispensed per 1000 persons or managed care enrollees, and the proportion of office visits receiving antibiotics measured during 10 to 12 months before and after the campaign. Results: After the mass media campaign, there was a 3.8% net decrease in retail pharmacy antibiotic dispenses per 1000 persons (P = 0.30) and an 8.8% net decrease in managed care-associated antibiotic dispenses per 1000 members (P = 0.03) in the mass media community. Most of the decline occurred among pediatric members, and corresponded with a decline in pediatric office visit rates. There was no change in the office visit prescription rates among pediatric or adult managed care members, nor in visit rates for complications of acute respiratory tract infections. Conclusions: A low-cost mass media campaign was associated with a reduction in antibiotic use in the community, and seems to be mediated through decreases in office visits rates among children. The campaign seems to be cost-saving.

Choosing among two-dimensional smoothers in practice

Abstract Seven smoothers representing two general classes of smoothers are evaluated in terms of their ability to capture patterns and ignore random fluctuations in two-dimensional structures. Two patterns are generated as responses at locations representing the coordinates of 86 counties within 500 miles of San Francisco: a null pattern, having a constant level surface, and a structured pattern, having a ridge, peak, and a depression in the surface. Three levels of Gaussian noise are assessed as typical of the sizes of variation that arise in connection with measurement error models or with disease mortality rates. Two criteria used to evaluate the smoothers suggest that weighted averages (weights inversely proportional to the square of the distance) and a nonlinear smoother proposed by Tukey (headbanging) are likely to be the most successful in practice.

Competing and Using Moral versus Urban Measures in Statistical Applications

Abstract We discuss various measures of urban-versus-rural character, particularly when used for adjustment of epidemiologic or environmental data. For example, because urban-versus-rural differences may conceal more subtle effects, such a measure may be used to adjust for urban-versus-rural differences in county mortality rates, so that underlying geographic or demographic patterns may be revealed. Urban-versus-rural measures frequently have been defined as categorical variables given by ranges of (i) total population; (ii) population density; and (iii) percent urban population. When regions are counties in the United States, such measures are easily obtained, but their relation to ones concept of urban-versus-rural character is often weak, as illustrated with several examples. A class of alternative measures is proposed. These measures are functions of the individual populations of places (usually minor civil divisions) in the county: (1) population of the largest place; (2) root sum of squared populat...

Geographic trends in prostate cancer mortality: An application of spatial smoothers and the need for adjustment

Prostate cancer mortality among whites and nonwhites in U.S. counties are analyzed for geographic effects. To better visualize geographical effects, the data are smoothed with a bivariate smoother using age-specific rates. Among nonwhites, an important explanatory variable is the proportion of African Americans. A relationship between the mortality rate and this variable is derived, and the data are adjusted for this variable using this relationship. When the rates are adjusted for age only, among whites there is a north-south gradient: rates are higher in the north, lower in the south. Among nonwhites, the gradient runs east to west: higher in the east, lower in the west. The latter gradient disappears when the rates are further adjusted for African Americans. The study reveals the importance of both smoothing the data to visualize patterns in geography and adjusting the data for an important variable to identify underlying patterns. The additional adjustment permits the identification of other areas of the country with elevated or depressed rates.

Simultaneous smoothing and adjusting mortality rates in U.S. counties: melanoma in White females and White males

Detecting patterns in health-related data for geographic areas is facilitated with the use of exploratory methods, especially smoothing. In addition, these data often must be adjusted for known prognostic factors such as age and gender. The analysis in this paper focuses on mortality rates due to malignant melanoma in White males and White females; these data are adjusted for both age and latitude, separately for males and females, and then smoothed using (a) a non-linear smoother known as weighted head-banging, and (b) a new method that incorporates the adjustment and the smoothing simultaneously. Maps of the continental United States show regions of high rates, even after having adjusted for age and latitude, and suggest the possibility of other variables that may influence the rates.

Gaussian white-noise generation for digital signal synthesis

An algorithm for simulating a random Gaussian noise signal on a computer is proposed. The power spectrum of the generated signal is constant, and the ordinates of the phase spectrum are independent and uniformly distributed. A computer program has been developed based on this algorithm and used with a signal generator to test electronic systems.

Understanding medical research : a practitioner's guide

Part 1 Principles of estimation: describing population characteristics screening tests. Part 2 Principles of hypothesis testing: approach to the study of a risk factor and an outcome study design spurious association causal association more about statistical tests. Part 3. Part 4 Application to conducting a study: a systematic approach gathering the data - designing a data form entering and analyzing the data writing the paper. Appendices: a short guide to statistical tests and measures sample protocol sample data forms determining sample size randomization.