11 posts · 6,226 views
Evidence-Based Public Health
10 posts
Upon*the.People
1 post
Sort by Latest Post, Most Popular
View by Condensed, Full
- January 7, 2010
- 05:40 PM
- 746 views
Achtung, Baby: Hazard Ratios
by Ryan in Evidence-Based Public Health
The hazard ratio is the statistic of choice for nearly all medical research involving time. And by far the most common method of deriving hazard ratios from data is via the Cox Proportional Hazards model. In a great little editorial in this month's Epidemiology, Miguel Hernán lays out what we lose and what we can gain with a more subtle approach.... Read more »
Hernán, M. (2010) The Hazards of Hazard Ratios. Epidemiology, 21(1), 13-15. DOI: 10.1097/EDE.0b013e3181c1ea43
- November 19, 2009
- 09:11 PM
- 711 views
Multilevel (Quantile) Regression - A Question for Gelman
by Ryan in Evidence-Based Public Health
Multilevel (or hierarchical) regression modeling is very popular in the social sciences. So what I want to do is a hierarchical quantile regression of the 75% quantile of time spent in jail. And that was my question for Andrew Gelman.... Read more »
Carmichael SL, Witte JS, & Shaw GM. (2009) Nutrient pathways and neural tube defects: a semi-Bayesian hierarchical analysis. Epidemiology (Cambridge, Mass.), 20(1), 67-73. PMID: 19234400
- November 10, 2009
- 11:33 PM
- 686 views
'Cause I said so... The Sufficient-Component Cause model and what it can tell us about cancer screening: Part I
by Ryan in Evidence-Based Public Health
... Read more »
Flanders WD. (2006) On the relationship of sufficient component cause models with potential outcome (counterfactual) models. European journal of epidemiology, 21(12), 847-53. PMID: 17048084
- February 8, 2010
- 01:49 PM
- 601 views
Methods Monday: Maximum Likelihood in SAS using PROC NLP
by Ryan in Evidence-Based Public Health
I've been working on fitting some excess relative risk (ERR) models to case-control data on occupational exposures lately. ERR models are of the form:RR=1+β*XIn SAS, unfortunately, we don't have unlimited freedom in defining the form of the model we want to fit, but a recent paper by Langholz and Richardson [behind firewall] describes a way that we can solve for parameters once we specify the likelihood function. (For those interested, the likelihood function can be thought of as the function that would be most likely to give rise to the data. We define it with some variables, and then try to solve for the variable(s) that maximize the likelihood function. This falls into the class of methods called maximum likelihood estimation.)The general conditional logistic likelihood is pretty simple (phi represents the odds or rate ratio function) :The best way to conceptualize this equation is as: divide the data you observed by all possible permutations of the data.This function is then maximized with respect to beta (for the mat-inclined, an iterative process minimizes the derivative of the log of the function to look for the global maximum).The method described by Langholz and Richardson makes use of a nifty little SAS procedure called PROC NLP (the NLP stands for non-linear programming). It basically does exactly what I just described: you can specify a function and host of parameters, and it will iteratively search for a maximum value of the function, and spit out the parameters that yield the maximum.A cool extension of this is that you can define complex "mixture models" that contain two distinct models that are each exponentiated: one to alpha, one to 1-alpha. You then multiply the two exponentiated models together. If you then maximize the likelihood, including the parameter alpha, you get a neat little value that tells you the relative importance of each of the two models in the full mixture model. For example:RR=[(βX)^α]*[(exp(βX))^(1-α)]PROC NLP lets you specify this model form and get an estimate of which model (linear or exponential) fits better, depending on whether alpha is closer to zero or one.Langholz, B., & Richardson, D. (2009). Fitting General Relative Risk Models for Survival Time and Matched Case-Control Analysis American Journal of Epidemiology, 171 (3), 377-383 DOI: 10.1093/aje/kwp403... Read more »
Langholz, B., & Richardson, D. (2009) Fitting General Relative Risk Models for Survival Time and Matched Case-Control Analysis. American Journal of Epidemiology, 171(3), 377-383. DOI: 10.1093/aje/kwp403
- January 9, 2010
- 02:55 PM
- 599 views
Crime and Public Health: Plumbum Causa
by Ryan in Evidence-Based Public Health
There is a distressingly myopic tendency among our existing social programs. Health departments ignore crime, and miss a valuable opportunity to improve social well-being.... Read more »
Wright, J., Dietrich, K., Ris, M., Hornung, R., Wessel, S., Lanphear, B., Ho, M., & Rae, M. (2008) Association of Prenatal and Childhood Blood Lead Concentrations with Criminal Arrests in Early Adulthood. PLoS Medicine, 5(5). DOI: 10.1371/journal.pmed.0050101
- November 13, 2009
- 09:50 PM
- 531 views
Methods Blogging: The Prostate Cancer Prevention Trial
by Ryan in Evidence-Based Public Health
It's somewhat defeating to acknowledge, but a large part of the strength (and beauty) of this study lies in its simplicity. Placebo-controlled, double-blinded, randomized; these are the things biostatisticians dream of. Luckily, I'm not one; but I can still appreciate it.... Read more »
Thompson, I. (2003) The Influence of Finasteride on the Development of Prostate Cancer. New England Journal of Medicine, 349(3), 215-224. DOI: 10.1056/NEJMoa030660
Gerald L. Andriole, M.D., E. David Crawford, M.D., Robert L. Grubb, III, M.D., Saundra S. Buys, M.D., David Chia, Ph.D., Timothy R. Church, Ph.D., Mona N. Fouad, M.D., Edward P. Gelmann, M.D., Paul A. Kvale, M.D., Douglas J. Reding, M.D., Joel L. Weissfel. (2009) Mortality Results from a Randomized Prostate-Cancer Screening Trial. New England Journal of Medicine, 360(17), 1797-1797. DOI: 10.1056/NEJMx090012
- May 28, 2010
- 09:22 AM
- 494 views
Falling Child Mortality - Where we are on Millennium Development Goal 4
by Ryan in Upon*the.People
The release several days ago of revised estimates for global child mortality showing that mortality has fallen faster than we previously expected was a cause for celebration. As one of the eight targets of the Millennium Development Goals, child mortality is among the better indicators we have for the health status of a given population, and is, in the words of Michael Marmot, "the health outcome most sensitive to the effects of absolute material deprivation."[Children in Burma; The Irrawaddy]So where are we? The authors report that 7.7 million children will die in 2010, compared to 11.9 million in 1990. The reaction to this, of course, can be nothing but mixed - things are getting better, and at a faster rate than we previously supposed, but the number of deaths each year is still numbingly high. Note that the previous UN estimate (here, as of 5/27/10) was that, in 2008, 8.77 million children died, while the current study estimates the number at 7.95 million. It's hard to think of 800,000 children as a calculating error, but so it is.These new numbers are a 35% reduction since 1990, which is the reference year for the MDGs. The MDG for child mortality aims for a two-thirds reduction by 2015. So we've come about half of the way in 80% of the time... not stunning progress, but as the fall in mortality seems to be accelerating, the goal (which would comprise, according to the new study, 3.9 million child deaths in 2015) is not beyond reach.I want to comment here on the methodology used to estimate the number of deaths, as nearly every headline on the topic has highlighted the fact that these estimates are substantially lower than previous estimates (in the MSM's defense, the authors highlight the same issue in the discussion, as they should).It takes only a moment's consideration to realize the mind-boggling difficulty of getting an accurate count of the millions of children who die in a year. Anyone familiar with global health will know that there is a wide range of quality in the estimates put out by nations and international organizations. As a start towards classifying the quality of data, the authors grouped data sources into 10 types:Demographic and Health Surveys, World Fertility Surveys, Multiple Indicator Cluster Surveys, Pan Arab Project for Family Health Surveys, CDC Reproductive Health Surveys, census birth histories, complete vital regristration systems, incomplete vital and sample registration systems in low- and middle-income countries, complete vital and sample registration systems in low- and middle-income countries, and a category for all other sources.The basic approach is this. You assume that the function describing mortality in a country between 1990 and 2010 has a certain shape, but that we don't know what that shape is. Instead of assuming a particular function, such as linear or logistic, that describes the relationship, one can define a "function of functions" that describes the probability of a given function giving rise to the observed data (here, the various estimates of child mortality in a given country). This probability is also known as the likelihood (as in, the likelihood that this function would give rise to this data).The authors of the current paper used a technique called Gaussian Process Regression (GPR, or Kriging) to estimate the best function that would describe the child mortality curve for each country. Because GPR is an interpolation technique, it "fills in" the points between actual data points, which in this case are the individual estimates of child mortality. The term 'Gaussian Process' refers to what is more commonly known as a random walk. By assuming that the shape of the curve is a random walk between the observed points, the curve is allowed to obtain nearly any shape. The 'Regression' part comes in when the error is minimized - namely, the best Gaussian Process is the one described by the curve which would require the fewest random steps to fit the observed data points. The figure below shows a typical GPR; the green lines define the uncertainty, which increases as you get farther from an observed data point.[A Gaussian Process Regression; Wikipedia]This is all a gross simplification. The point is that GPR is a flexible interpolation technique that can be used when you don't know the "shape" your data should take, and you want to account for unknown deviations from a particular form (the authors give as a known example the HIV/AIDS epidemic in Sub-Saharan Africa, which caused an increase in child mortality).But back to child mortality. The authors employed this GPR technique to obtain country-by-country estimates, and then summed to derive their final predictions. This allows us to compare nations and regions on their child mortality statistics.Unfortunately, two regions may have had no reduction in child mortality between 1990 and 2010 - souther Sub-Saharan Africa and Oceania. Quite stunningly, the authors estimate that child mortality in Swaziland increased from 25.8 million to 38.3 million between 1990 and 2010.But the overall picture of global child mortality is, in the end, a bit brighter this week. We might meet MDG 4 after all. Global health infrastructure and social systems may be stronger than we previously supposed. Most importantly, fewer children are dying today than at any date in recent history.Reference:Rajaratnam, J., Marcus, J., Flaxman, A., Wang, H., Levin-Rector, A., Dwyer, L., Costa, M., Lopez, A., & Murray, C. (2010). Neonatal, postneonatal, childhood, and under-5 mortality for 187 countries, 1970–2010: a systematic analysis of progress towards Millennium Development Goal 4 The Lancet DOI: 10.1016/S0140-6736(10)60703-9... Read more »
Rajaratnam, J., Marcus, J., Flaxman, A., Wang, H., Levin-Rector, A., Dwyer, L., Costa, M., Lopez, A., & Murray, C. (2010) Neonatal, postneonatal, childhood, and under-5 mortality for 187 countries, 1970–2010: a systematic analysis of progress towards Millennium Development Goal 4. The Lancet. DOI: 10.1016/S0140-6736(10)60703-9
- November 16, 2009
- 10:50 PM
- 493 views
Prevalence in Place of Plausibility: NCCAM Call for Comments
by Ryan in Evidence-Based Public Health
NCCAM has funded, to the tune of half a million dollars, of study of magnets and carpal tunnel syndrome.... Read more »
Colbert, A., Wahbeh, H., Harling, N., Connelly, E., Schiffke, H., Forsten, C., Gregory, W., Markov, M., Souder, J., Elmer, P.... (2007) Static Magnetic Field Therapy: A Critical Review of Treatment Parameters. Evidence-based Complementary and Alternative Medicine, 6(2), 133-139. DOI: 10.1093/ecam/nem131
- January 6, 2010
- 03:30 PM
- 491 views
Innovation in Health: Socialism and Innovation
by Ryan in Evidence-Based Public Health
What's the motivation for innovation in healthcare, and does any degree of socialization at any level have an impact?... Read more »
Conrad, D., & Perry, L. (2009) Quality-Based Financial Incentives in Health Care: Can We Improve Quality by Paying for It?. Annual Review of Public Health, 30(1), 357-371. DOI: 10.1146/annurev.publhealth.031308.100243
- November 12, 2009
- 02:58 PM
- 472 views
Cell Phones and Cancer?
by Ryan in Evidence-Based Public Health
CNN has a story about the link between cell phone usage and tumors. Unfortunately, the article tends towards the sensational, and doesn't cite the many recent studies that have failed to find a link between cell phone use and cancer.... Read more »
Ahlbom A, Feychting M, Green A, Kheifets L, Savitz DA, Swerdlow AJ, & ICNIRP (International Commission for Non-Ionizing Radiation Protection) Standing Committee on Epidemiology. (2009) Epidemiologic evidence on mobile phones and tumor risk: a review. Epidemiology (Cambridge, Mass.), 20(5), 639-52. PMID: 19593153
- February 2, 2010
- 09:41 PM
- 402 views
More on abstinence
by Ryan in Evidence-Based Public Health
A recent paper describes a significant benefit from abstinence-only education. The story is more complicated...... Read more »
Jemmott, J., Jemmott, L., & Fong, G. (2010) Efficacy of a Theory-Based Abstinence-Only Intervention Over 24 Months: A Randomized Controlled Trial With Young Adolescents. Archives of Pediatrics and Adolescent Medicine, 164(2), 152-159. DOI: 10.1001/archpediatrics.2009.267
topics
join us!
Do you write about peer-reviewed research in your blog? Use ResearchBlogging.org to make it easy for your readers — and others from around the world — to find your serious posts about academic research.
If you don't have a blog, you can still use our site to learn about fascinating developments in cutting-edge research from around the world.
