American Medical Association declared obesity a “disease.”

American Medical Association declared obesity a “disease.”

"RESOLVED, That our American Medical Association recognize obesity as a disease state with 24 multiple pathophysiological aspects requiring a range of interventions to advance obesity 25 treatment and prevention. (New HOD Policy - Resolution 420)" - 06/16/2013

• New York Times (2013). A.M.A. Recognizes Obesity as a Disease
• Time.com (2013). The Best Cure for Obesity? Personal Responsibility.
• Boston.com (2013). Has obesity been mislabeled as a disease? Why doctors don’t mind.
• Bays (2013). Obesity, adiposity, and dyslipidemia: A consensus statement from the National Lipid Association.

Risk prediction and model comparison

 Risk prediction and model comparison
 
The potential approaches of prediction and comparison:

• Relative risk/hazard ratio/odds ratio, P-value.
• Sensitivity/specificity
• Area under ROC (receiver operating characteristics) curve (AUC)/Harrell's c statistic
• Somers' D (Kendall's Tau): the mathmatic conversion of Somers'D and c statistic are: [c statistic = D/2+0.5] or [D = (c - 0.5) x 2]. (SAS: PROC FREQ)
• NRI (net reclassification improvement)
• IDI (integrated discrimination improvement).
• K-fold cross-validation - Wikipedia
• ...

References:

• Pencina (2011). Extensions of net reclassification improvement calculations to measure usefulness of new biomarkers (Comment).
• Pencina (2013). Understanding increments in model performance metrics.
• Pencina (2008). Evaluating the added predictive ability of a new marker: From area under the ROC curve to reclassification and beyond.
• Measuring the Accuracy of Prediction Models - An International Symposium (2008): IDI, NRI and different c statistics (pdf)
• Hilden (2013). A note on the evaluation of novel biomarkers: do not rely on integrated discrimination improvement and net reclassification index.
• Liu (2012). Evaluating a New Risk Marker’s Predictive Contribution in Survival
• Mühlenbruch (2013). Assessing improvement in disease prediction using net reclassification improvement: impact of risk cut-offs and number of risk categories (Commentary).
• Cassell (2007). Don't be loopy: re-Sampling and simulation the SAS way (pdf).
• Kerr (2011). Evaluating the incremental value of new biomarkers with integrated discrimination improvement.

Software:

• Pepe Lab has some homemade software for biomarker/risk evaluation using Stata, SAS, R, SPSS, and even FORTRAN.
• Stata: to install 'Risk Prediction Package' (predcurve and incrisk), in a Stata session type: ".net from http://labs.fhcrc.org/pepe/stata/" and follow the instruction.To update the risk_prediction package at a later time, in Stata type: ".adoupdate risk_prediction, update".
• UCR posted SAS, Stata, and R codes for NRI & IDI on the website.
• Stata: How can I get a Somers' D after logistic regression in Stata?

 

rule of halves of diabetes

Rule of halves of diabetes
Source: DAWNStudy Diabetic Attitudes Wishes and Needs

Multicollinearity Issue

Multicollinearity Issue

Multicollinearity happens when two or more predictor/independent variables/regressors are highly correlated. I have been discussed about this issue many times by colleagues and journal reviewers. Paul Allison has a blog of some rules of thumb: When Can You Safely Ignore Multicollinearity? Wikipedia also has a article about this issue. It's true this is issue theoretically, but based on my experience in public health of chronic diseases, if the selection of predictors based on the logic/knowledge behind the model but not dump everything in one model, this should not be an issue.

Do We Really Need Zero-Inflated Models?

Do We Really Need Zero-Inflated Models?
Source: Statistical Horizon blog by Paul Allison
 
"... Of course, there are certainly situations where a zero-inflated model makes sense from the point of view of theory or common sense. For example, if the dependent variable is number of children ever born to a sample of 50-year-old women, it is reasonable to suppose that some women are biologically sterile. For these women, no variation on the predictor variables (whatever they might be) could change the expected number of children.

So next time you’re thinking about fitting a zero-inflated regression model, first consider whether a conventional negative binomial model might be good enough. Having a lot of zeros doesn’t necessarily mean that you need a zero-inflated model."

Read full text here
 
This question has haunted me for a while, thank Dr. Allison answered this question in such a layman-kind way. I like his book "Survival Analysis Using SAS: A Practical Guide" much; I don't have his book "Logistic Regression Using SAS: Theory and Application". Hope this logistic regression related book is in the same style.

More Blog on Statistical Horizon Blog