Advanced Medical Statistics - 2013

week	date (for you to complete)	material (link to pdf)	datasets
my YouTube videos theoldorganplayer: link to videos
1		Assignment details Survival analysis Chan Biostatistics survival analysis article additional information for Chan article (including R Commander equivalent analysis) Example Course handbook (students will be provided with individually edited version of this)	Description of all datasets survival1.sav survival1a.sav survival_collett_p290.sav lung_cancer_prentice1973.sav breast_cancer_survival.sav
2		continue survival analysis handout
3		Logistic regression	andy_f_eel.sav andy_f_burnout.sav miles_shevlin_p152_data6_2.sav
4		Multiple regression	miles_shevlin_p169_data7_1.sav campbell_mlr_p12.sav airpoll.sav daniel_random_p514_ed5_multreg_chd_risk.sav
5		Continue multiple regression handout above
6		Mixed models approaches: Getting familiar with the Linear Mixed Models (LMM) options in SPSS	t_test_paired_long_format.sav twisk_longitudinal_wide.sav twisk_longitudinal_long.sav
7		Repeated measures 2 - Four repeated measures in a single group	same as above
8		Repeated measures 3 - Five repeated measures in two treatment groups	bandb_wide.sav The following links provide the data at the various stages described in the chapter - you should not need them: bandb_long_start.sav bandb_trad_treat_transpossed.sav bandb_btheb_treat_transpossed.sav bandb_long_completed.sav bandb_long_month.sav
9	Each candidate can select either the factor analysis or binary analysis chapter to work through	Factor analysis R script for correlation matrix input	med_factor.sav psy_factor.sav grnt_fem.sav (two factor model) t_b_factor.sav computer_s_2009.sav grnt_fem.dat
10		Analysing a set of binary questions	somtot.dat (somtat = mus scores; 30 vars. n=161) somtot.txt somtot.sav sexualat.dat sexualat.sav (sexual attitude data from Dmitris Rizopoulos itm r package; 10 vars n=1077)

Welcome. This is the second course in medical statistics (you can see the first course here.) This course has been designed specifically for medics and those concerned directly with healthcare provision, while students from other disciplines are welcome to use the material it is advised that they search out examples of the various techniques discussed within their own discipline to compliment the materials presented here.

The course assumes that you have completed the introductory course before undertaking this one - if you have not done this it is unlikely you will understand the material above and specifically you will probably find using R for the various exercises extremely difficult.

This is a 10 week (around 100 hours of work) course which introduces you to survival analysis, logistic regression, and various other varieties of multiple regression, however the focus of the course is on the modern analysis of repeated measures, the so called Linear Mixed Models (LMM) approach. There has been much research and development of specific software for this type of analysis over the last decade and statistical practices are changing rapidly. Linked closely with this approach are what are called multilevel models.

Both the LMM and multilevel approach have much in common as they both allow the modelling of non independent data. For example you may have a cluster randomised trial where patients are seem by particular GPs, within various practices or investigating the effects of a training program across many subjects (containing several trainers) across many locations. Considering the GP evample, regardless of how standardised the GPs try to be you need to be able to take into account the GP effect upon the particular cluster of patients along with the practice effect at a higher level. A similar argument goes for the training program example. Often repeated measures for a particular patient/client/subject mean it is also essential to take into account including the individual baseline measure and any possible correlation between the repeated measures for themsleves and the independent nature of the data across different patients.

Finally we look briefly at a data reduction techniques - Factor analysis (Principal component analysis), this much older technique, which was developed in the early 20th century allows an analyst to take a large number of variables, and given certian circumstances, reduce them down to a core set of one or more factors each of which consists of a subset of the original variables. The classic example of this is where a whole battery of tests concerning Intelligence is reduced down to a few factors, such as motor skills, language, etc. It is also a very common technique in psychiatry where a questionnaire has a large number of items to provide a smaller set of indicators concerning certain traits. Unfortunatley this techique does not work well for a set of binary variables, a common situation in questionnaire design so we consider this seperately.

An aspect of this course that you should find familiar, as it reflects the approach taken in the Essential Medical statistics course, is the introduction of a variety of software packages, for specific situations, the two core software applications here being SPSS (for a short time called PASW) and R, including a free add on called R commander.

R, being a free open source application, has seen a meteoric rise in use in the last few years and is being increasingly used for introductory statistics courses. I will also be introducing you to MLwIN, a free application specifically for carrying out multilevel modeling

This course requires you to carry out a large number of practical exercises yourselves and to facilitate this I have included screenshots in the pdf handouts and also over 30 HD youtube videos. It is essential that from the very first week you carry out the analyses yourselves and do not just read the notes and watch the videos, as the assignment requires you to carry out a similar analysis.

Required resources

To complete this course you will need:

A copy of Norman G R, Streiner D L. 2008 (3rd ed) Biostatistics: The bare Essentials. [there are two versions of this book, you need the cheaper one without the SPSS disk] isbn: 978 155009 3476 Peoples medical publishing house http://www.pmph-usa.com/products/biostatistics-the-bare-essentials If you are a student of Edinburgh university/Rcsed you can access an electronic copy at: http://lib.myilibrary.com.ezproxy.webfeat.lib.ed.ac.uk/Open.aspx?id=227415
Statistics at square two by Mike Campbell et al BMJ publications, an excellent little book, available as a ebook at Edinburgh University
A copy of SPSS (formally PASW) any version above 17 will be sufficient.
A copy of 'R' available free from http://www.r-project.org/

Additional none essential texts

Understanding clinical papers 2nd ed. by David Bowers, Allan House, David Owens.
If you are a student of Edinburgh university/Rcsed you can access an electronic copy at (edin uni / rcsed students): http://www.dawsonera.com.ezproxy.webfeat.lib.ed.ac.uk/depp/reader/protected/external/AbstractView/S9780470091319
statistics workbook for evidence based healthcare by Peat, Burton and Elliott at (edin uni / rcsed students): http://www.dawsonera.com.ezproxy.webfeat.lib.ed.ac.uk/depp/reader/protected/external/AbstractView/S9781444300505