Manning Feinleib

Methodological issues in screening for dementia: The problem of education adjustment

The methodological problems of developing efficient and unbiased screening methods for population-based studies of dementia have received scant attention. The potential advantages of education-adjusted screening methods are discussed. The implications for adjustment techniques of a negative correlation of educational attainment with age are emphasized. Two education adjustment methods, a stratified regression method and a nonparametric method, which take the age-education correlation into account are described, compared, and illustrated.

The Framingham disability study: Physical disability among community-dwelling survivors of stroke

The relationship between stroke and physical disability was examined in a cohort of adult, Framingham, Massachusetts, residents who, between 1948 and 1951, were assembled for a longitudinal examination of cardiovascular disease. Multivariate analyses examined the amount of residual disability attributable to stroke among 2540 community-dwelling survivors, 27 years after their initial examination, after controlling for age, cardiovascular risk factors, other cardiovascular diseases, and eight general health conditions related to physical disability. Among men living in the community, a history of stroke explained 12% of the variance in physical disability. Suffering a stroke, however, was not as strongly related to physical disability among women living in the community, accounting for only 3% of the variance. Results suggest that although older men and women die from the same major causes, they may not be disabled by the same conditions.

On the relation between blood pressure change and initial value

Abstract The methods of Feinleib, Halperin and Garrison (FHG) [1,4] and maximum likelihood (ML) [5] were used to analyze the relation between initial value and change over time of serial blood pressure measurements among participants in the Framingham Heart Study [3]. Both methods assume a linear trend in expected blood pressure over time, with variation in intercept and slope over individuals. If individual values of the slope and intercept have a joint normal distribution in the population, the coefficient of regression of slope on initial value over individuals will also be linear. It is shown here that the FHG method tends to underestimate that coefficient of regression. Significantly positive regressions were found under the ML approach while the FHG approach failed to claim significance in most of the age-sex groups analyzed. A direct comparison of the efficacies of the two significance tests shows that the asymptotic relative efficiency (ARE) of the FHG and ML procedures is always less than one indicating that the ML test has larger power, i.e. is more likely to detect an association between change and initial value when one exists. These results establish the ML procedure as the preferred method for this problem. Since significantly positive regressions were found in most age groups for both sexes under the ML approach, we infer that there exists some degree of positive association between blood pressure change and initial value. Although the expected increase in the rate of change of systolic blood pressure (SBP) per mmHg increase in initial value is not very great, a younger woman or a middle-aged man with an SBP initial value 20 mmHg higher than the mean initial level of their age-sex group has an expected rate of increase almost twice the mean rate of increase in their groups.

The magnitude and nature of the decrease in coronary heart disease mortality rate

The age-adjusted cardiovascular mortality rate has fallen by 40% in the United States over the last 30 years, primarily since the late 1960s. Although cardiovascular disease is still the leading cause of death in this country, the rate of deaths from cardiovascular disease have dropped to below 50%, and the rate continues to decrease. In actual numbers, it has been estimated that as many as 300,000 people between the ages of 35 and 65 years would have died from coronary heart disease between 1968 and 1978 if the mortality rate had remained unchanged. The extent of the decline varies with different areas of the country and most dramatically affects the black population. Between 1970 and 1980, life expectancy increased by 2.7 years for white men, 2.5 years for white women, 3.7 years for black men, and 4.0 years for black women. Attempts to analyze the decline in mortality must take 2 features of mortality into consideration: incidence and case fatality, which reflect changes in primary prevention factors and in treatment methods, respectively. However, data on cardiovascular disease conflict and are difficult to interpret. Many methods of estimating the potential effect of altering or eliminating potential risk factors such as smoking, hypertension and cholesterol have been explored, with encouraging results.

Regression analysis of data with correlated errors: An example from the NHLBI twin study

Epidemiologic studies often involve genetically related individuals, spouses, or repeat observations on the same individual. When regression analysis is required in such studies, significant correlation of the residuals may affect the estimates of the standard errors of the regression coefficients. Ordinary least squares may not provide the best (minimum variance) estimates of the regression coefficients. Generalized least squares (weighted least squares) is more appropriate when the covariance matrix of the errors is known or can be estimated with some degree of confidence. Data from a twin study of pulmonary function were analyzed by three different regression techniques and comparisons of the coefficients and standard errors are made to illustrate the potential effects of correlated errors.

Data needed for improving the health of minorities

Identified needs for minority health data, obstacles in obtaining the data, and potential solutions are reviewed. Vital statistics for whites and blacks have been available by states for many years. Recent revisions provide data on Hispanics, and new resolutions will provide data on Asian and Pacific Islander subgroups. But limitations persist in providing accurate statistics for minority subgroups. A major obstacle is the inadequacy of census denominator estimates, due to differential undercounts, paucity of postcensal estimates for states and localities, and the validity of the race and ethnicity data. Important issues revolve around quality, comparability, and intraperson variability of self-identification in determining race and ethnicity, versus external assessment. National survey data have oversampled for black and Hispanic minorities, but not others. The Disadvantaged Minority Health Improvement Act of 1990 provides some solutions, including an extramural grants program to strengthen minority statistics, which the National Center for Health Statistics has implemented to improve minority health assessment at all levels.