Statistical tools to understand and undertake anticancer research

1
Statistical tools to understand and undertake
anti-cancer research
Valter TORRI, MD Department of Oncology

MARIO NEGRI INSTITUTE, MILAN
2
General principle

• Statistic is a tool for describing and explaining
  variability of observations
• Several different types of investigation
• observational studies
• exploratory experiments
• confirmatory studies.
• Principles of experimental design apply equally
  to experiments in the life sciences involving
  cell cultures, animals and humans

3
Requirements for a well-designed experiment

• Avoidation of bias
• Randomisation
• Masking
• Analysis on appropriate sample (ITT)
• Control of chance effect
• Power probability of detecting a difference
  between treatment groups, assuming that a
  difference exists.

4
Sample size determination
5
Sample size for comparative trials
Power 90 a 0.05
6
Power and relevance of results

• Power depends on events, which are function of
  observation time
• same power with
• many patients followed for a short time
• relatively few patients followed for longer time
• quality of information different
• early effects
• late effects

7
Statistical vs. clinical significance

• p value tells us the probability that the
  difference between two treatments was due to
  chance
• Statistical significance, requirement for
  determining clinical significance,
  is not enough to signify a clinical difference
• Confidence intervals help us to understand how
  close our estimate is to the "truth

8
Relevance of summary measures

• Baseline OS Abs. Diff. NNT
• HR 0.80 0.67 0.50
  0.80 0.67 0.50
• 0.8 3.7 6.1 9.4
  27 16 11
• 0.5 7.4 12.9 20.7
  14 8 5
• 0.2 7.6 14.0 24.7
  13 7 4
• 0.1 5.8 11.4 21.6
  17 9 5

Relative measures tells part, but not all of the
story Absolute measures do it better