30 years of Medical Statistics What has changed and what hasnt

1
30 years of Medical StatisticsWhat has changed
and what hasnt

• Mike Campbell
• University of Sheffield

2
Sheffield Probability and Statistics 1973
3
Paper tape circa 1971
4
Punch card - in use until 1979!
5
Computing 1976

• DIY
• Fortran, Algol
• Batch mode
• SPSS
• BMDP
• Language
• Genstat
• Interactive
• Glim, Minitab

6
Developments in computing

• Commodore Pet/Apple (1978)
• Basic
• IBM PC 1981
• BBC Computer 1981
• Specially written stats packages for BBC
  Computer
• SPSS PC 1989

7
Burwalls 1985
8
Important books and papers

• Armitage P (1971) Statistical methods in medical
  research
• Nelder JA and Wedderburn RWM (1972) Generalized
  linear models JRSSA
• Cox DR (1972) Regression models and life tables
  JRSSB
• Breslow NE (1984) Extra-Poisson variation in
  log-linear models. Appl Stat
• Bland M and Altman (1986) Method of agreement
  Lancet

9
New techniques that have had a big effect on the
practice of medical statistics

• Bootstrap (Efron 1979 JASA) (extension of
  Fishers permutation methods)
• Random effects (Laird and Ware, 1982 Biometrics)
• Ordinal models (McCullagh, 1980 JRSSB)
• Multi-level models (Goldstein , 1987)

10
Techniques no longer used

• ANOVA
• Probit analysis
• Discriminant analysis

11
Other developments

• New journals
• Statistics in Medicine (1981)
• Biostatistics (2001)
• Pharmaceutical Statistics (2002)
• New Societies
• Statisticians in the Pharmaceutical Industry
  (1977)
• International Society of Clinical Biostatistics
  (1980)

12
Career developments

• Teachers of Medical Statistics (Burwalls, 1980)
• Statistical refereeing for BMJ (1984) and Lancet
  (?1990)
• Also many other journals now recognise use of
  statistical referees
• Examinations for Royal Colleges

13
What has stayed the same

• T-tests
• Chi-squared tests
• Teaching medical statistics
• Problems with method comparisons
• Communicating statistics

14
Survival

• 1976
• Possible to know most of medical stats
• 2006
• Specialization
• For an applied statistician can learn more from
  newsletters of software packages than from
  journals