Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "Ariel Linden, DrPH" <ariel.linden@gmail.com> |
To | <statalist@hsphsun2.harvard.edu> |
Subject | re: Re: st: Alternatives to clogit using generalised additive models (GAM) |
Date | Wed, 21 Dec 2011 10:26:20 -0500 |
I would add regression splines to Steve's suggestion as well. In fact, the approach used in -gam- (written by Patrick Royston and Gareth Ambler), is more akin to splines than polynomials. Patrick has another program you should consider for this problem - mvrs- which is the multiple variable alternative to -uvrs- (the univariate model). See: Lambert P, Royston P (2009) Further development of flexible parametric models for survival analysis. Stata Journal 9: 265-290. Royston P, Sauerbrei W (2008) Multivariable model-building. A pragmatic approach to regression analysis based on fractional polynomials for modelling continuous variables. Wiley-Blackwell, Chichester. Royston P, Sauerbrei W (2007) Multivariable modeling with cubic regression splines: a principled approach. Stata Journal 7: 45-70. Ariel Date: Tue, 20 Dec 2011 18:22:27 -0500 From: Steve Samuels <sjsamuels@gmail.com> Subject: Re: st: Alternatives to clogit using generalised additive models (GAM) See -fracpoly- and -mfp- which will fit very flexible models in any regression command, including -clogit-. "search fractional polynomials, all" will turn up several contributed commands that enhance -fracpoly- amd -mfp-. Steve sjsamuels@gmail.com On Dec 20, 2011, at 8:51 AM, Lucas Salas wrote: Hi, I am performing a matched case-control analysis in which my outcome is binomial (Cancer yes/no) and my exposure is continuous non-normal. I have already fitted some conditional logistic regression to account for matching using quartiles as boundaries to overcome the non-normality. However, I am interested in explore the dose-response curves and the fitting of my model using a GAM approach. The actual GAM module for Stata is the Fortran app developed by Hastie and Tibshirani, which do not allow any fixed effects adjustment (conditional or unconditional). I have seen these potential alternatives in different webpages: 1. Introducing a match dummy variable in my GAM model. As you could expect my strata are rather sparse so this is not helpful. 2. Fitting a Cox model using exact partial likelihood and using the match as the strata. The logic behind the conditional logistic regression is ok, but I cannot figure out how this applies to a GAM model, and how can I extrapolate this to graph the dose response. 3. Using a vectorial GAM to account for the matching. I have not found any Stata alternative to this procedure. I'd be very grateful for any potential ideas, or if you know any kind of user written ado that might be useful for this data. Thanks, Lucas * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/