Event
Ernest Lo, PhD, 9I制作厂免费
Tuesday, January 30, 2018 15:30to16:30
Purvis Hall
Room 24, 1020 avenue des Pins Ouest, Montreal, QC, H3A 1A2, CA
Biostatistics Research At The Quebec Public Health Institute: In Search Of An Unbiased Life Expectancy Estimator For Regional Populations.
Ernest Lo holds a PhD in physics from Princeton University and a Masters in biostatistics from 9I制作厂免费. He has worked in diverse fields including theoretical ecology, neuroimaging and bioinformatics. Ernest is currently a biostatistician and research scientist at the Quebec Public Health Institute, as well as Adjunct Professor in the department of Epidemiology, Biostatistics and Occupational Health at 9I制作厂免费. His mandates include health forecasting, estimation of social inequalities in health, and the improvement of statistical methods used in public health.
Life expectancy (LE) is a key indicator of population health whose estimated values have enormous impact for both the public and for policy makers. Although LE is routinely calculated by health agencies worldwide, little is known as to whether or not LE is in fact an unbiased estimator. Regional level estimates of life expectancy within Quebec have shown evidence of severe upward bias, leading to implausibly high values, when the standard, actuarial method is used. A geometrical argument can be used to demonstrate that this bias is produced by inaccuracy in the closure model, or the way mortality or survival is modeled over the last, open age interval. An alternative class of closure models uses extrapolation to estimate mortality over the oldest age interval; these include the Gompertz, Hsieh and Kannisto approaches. In contrast, a 鈥榬elational鈥� approach, termed the Brass method, transforms a reference survival curve to that of each population being estimated. Each of these methods is described and their performance, with respect to bias and variance, is assessed over empirical datasets and using of Monte Carlo simulation. Themes that will be addressed include: 1) strategies to evaluate bias in the absence of gold standard knowledge of the 鈥榯rue鈥� LE for a given population, 2) sensitivity of bias and variance to key parameters implicit within each LE model, 3) the relation between alternative models of LE and different approaches of 鈥榖orrowing strength鈥�. This work represents the first detailed comparison of the bias and variance of different population-level LE estimators. In addition to the statistical import of the findings, it is hoped that the results will lead to improved LE estimation by public health agencies and thus to improved public health planning and policies.
Life expectancy (LE) is a key indicator of population health whose estimated values have enormous impact for both the public and for policy makers. Although LE is routinely calculated by health agencies worldwide, little is known as to whether or not LE is in fact an unbiased estimator. Regional level estimates of life expectancy within Quebec have shown evidence of severe upward bias, leading to implausibly high values, when the standard, actuarial method is used. A geometrical argument can be used to demonstrate that this bias is produced by inaccuracy in the closure model, or the way mortality or survival is modeled over the last, open age interval. An alternative class of closure models uses extrapolation to estimate mortality over the oldest age interval; these include the Gompertz, Hsieh and Kannisto approaches. In contrast, a 鈥榬elational鈥� approach, termed the Brass method, transforms a reference survival curve to that of each population being estimated. Each of these methods is described and their performance, with respect to bias and variance, is assessed over empirical datasets and using of Monte Carlo simulation. Themes that will be addressed include: 1) strategies to evaluate bias in the absence of gold standard knowledge of the 鈥榯rue鈥� LE for a given population, 2) sensitivity of bias and variance to key parameters implicit within each LE model, 3) the relation between alternative models of LE and different approaches of 鈥榖orrowing strength鈥�. This work represents the first detailed comparison of the bias and variance of different population-level LE estimators. In addition to the statistical import of the findings, it is hoped that the results will lead to improved LE estimation by public health agencies and thus to improved public health planning and policies.