Transcript 11 Years` Experience of Teaching Med Stats to Maths

Eleven Years Experience of Teaching
Medical Statistics to Mathematics
Undergraduates
Chris Roberts
Biostatistics Group
School of Health Sciences
University of Manchester
Why Teach Medical Statistics to Mathematics Students
Motivation (Biostats Group)
• Recruit PhD students.
• Promote medical statistics as a career option for maths students.
• Develop links between the Biostatistics Group and the Statistics Group in the
School of Mathematics.
Motivation (School of Maths)
• Make the BSc in Maths/Maths and Statistics more attractive for applicants
with an interest in statistics.
• Provide an additional 3rd year options.
Statistics Course Units in the Mathematics BSc
Year 1
Semester 1
Semester 2
Probability 1
Introduction to Statistics
Year 2
(Choice of 6 from 14)
Probability 2
Statistical Methods
Practical Statistics
Year 3
(Choice of 6 from 22)
(Choice of 6 from 23)
Statistical Inference
Generalized Linear Models
Linear Models
Time Series Analysis
Multivariate Statistics
Design and Analysis of Experiments
Medical Statistics
Social Statistics (New in 2011/12)
Statistical Computing
Module Curriculum (2001)
•
•
•
•
Randomised controlled trials: Introduction to RCT, Bias concealment
Randomisation; Statistical and ethical issues concerning randomised
experimentation on patients; Sample size and Power; Treatment allocation
methods; Equivalence and Non-inferiority Trials; Intention-to-treat; Subgroups
analyses; Cross-over Trials; Meta-analysis. (18 hrs) Roberts
Epidemiological studies: Causal inference concerning risks from
observational studies; Confounding; Cohort and case-control designs; Methods
of analysis; Incident rate ratios; Relative risks and odds-ratios; Stratification;
Application of generalised linear models. (9 hrs McNamee)
Measurement Error: Problems arising from measurement errors in diagnostic
testing, screening, and epidemiological studies; Statistical methods related to
measurement error. (6 hrs Dunn)
Complex survey sampling: Sampling frames and sampling fractions; Methods
of random sampling (simple, systematic, stratified and clustered); Implications of
design based methods of statistical analysis. (3 hrs Pickles)
Module Curriculum (2002)
Randomised controlled trials:
Introduction to RCT, Bias concealment Randomisation; Statistical and ethical
issues concerning randomised experimentation on patients; Sample size and
Power; Treatment allocation methods; Adjustment for Baseline; Equivalence
and Non-inferiority Trials; Intention-to-treat and CACE Estimation; Subgroups
analyses; Cross-over Trials; Meta-analysis. (24hrs ) Roberts
Epidemiological studies:
Causal inference concerning risks from observational studies; Confounding;
Cohort and case-control designs; Methods of analysis; Incident rate ratios;
Relative risks and odds-ratios; Stratification; Application of generalised linear
models. (12hrs) McNamee
Biostatistics MSc / MSc in Statistics Biostatistics
Pathway
Three or 8 modules on the Biostatistics MSc / MSc in Statistics (Biostatistics
Pathway) taught by Biostatistics Group
• Introduction to Clinical Trials
• Epidemiology
• Advanced Topics in biostatistics
Medical Statistics lecture are now part of the Introduction to Clinical Trials
Current Syllabus (2011)
•
•
•
•
•
•
•
•
•
•
Introduction to RCT, Bias concealment Randomisation; Statistical and ethical
issues concerning randomised experimentation on patients;
Basic methods of analysis including revision of statistical inference.
Sample size and Power.
Treatment allocation methods.
Adjustment for Baseline.
Equivalence and Non-inferiority Trials.
Intention-to-treat and CACE Estimation.
Subgroups analyses.
Cross-over Trials.
Meta-analysis.
(22 hours lectures – 11 hours example class)
30
20
10
0
Number of Students
40
50
Number of Students
2001
2006
Academic Year
2011
Features of the Medical Statistics Modules
•
•
•
•
•
Interpretation.
Discussion of practical aspect of trials.
No survival analysis.
Discussion of inappropriate methods.
Intention-to-treat introduced by considering CACE estimation.
Inappropriate Methods Covered
Analysis with baseline data
• Within treatment group tests of change.
• Test to check randomisation.
Equivalence and non-inferiority
• Use of test of the null hypothesis of no difference.
Sub-group analyses
• Separate test of sub-groups.
• Multiple testing.
Crossover trials
• Use of paired t-test.
• Tests of carry-over effect.
Treatment protocol violations
• Per-protocol and As Treated Estimates
ITT and CACE Estimation
CONTROL
INTERVENTION
Always Treatment
T + 
Always Treatment
T+ 
As Randomised
R
As Randomised
R + 
Always Control
Always Control
C
C
= CACE effect / Average Treatment Effect of the Treated
The Students
Mathematic students very different to medical students and medical researcher
• Non-communicative.
• Unwilling to ask or answer question.
Statistic teaching focused on the mathematics of statistics.
• Limited understanding of statistical inference.
• Lack confidence regarding interpretation of result.
• Limited exposure to design issues.
Students anxious about essay style writing.
Low level of computing skills.
• Lack experience of using statistical software.
• Course work almost always hand-written.
Teaching Format
Most courses on the Maths BSc follow a standard format
• Each course consisted of 2 lectures per week + Practical class.
• Chalk or Handouts.
• Assessment
• 80% end of course exam
• 20% course work or in-course test
To make the course attractive to the typical Maths student we chose to follow
this format.
Compromises
Course expected to have a minimum of 15 students it was therefore
important to make the course attractive to the typical mathematics
student.
• Include some algebraic deviation in course.
• Only limited amount of critical appraisal as some students
uncomfortable with prose style writing.
• Examination and course work includes some algebraic work and
well as hand-calculator calculations.
• No critical appraisal in the examination – limited amount in course
work.
• No computing practical classes.
• Use edited statistical output in handouts, exercises and
examination.
Course Work Assessment
Article from BMJ
• Low medical technical content.
• Simple statistical methods.
Typical Tasks
• Manual check calculations in paper (e.g. t-test / test of proportions/ sample
size)
• Comment of results - critical appraisal.
• Derivation of methods
 Maximum sample size for the difference of two proportions of a given
magnitude.
 Power/sample size for unequal allocation.
 Fisher’s exact test.
 Confidence interval for the rate ratio using the delta method.
• Missing data issues.
Course work papers
•
•
•
•
•
Quinn. Suturing versus conservative management of lacerations of the hand:
randomised controlled trial BMJ. 2002 August 10; 325(7359): 299.
Quist-Paulsen. Randomised controlled trial of smoking cessation intervention
after admission for coronary heart disease BMJ. 2003 doi: 10.1136.
Heal et al. Does single application of topical chloramphenicol to high risk
sutured wounds reduce incidence of wound infection after minor surgery?
Prospective randomised placebo controlled double blind trial BMJ. 2009; doi:
10.1136
Hickson et al. Use of probiotic Lactobacillus preparation to prevent diarrhoea
associated with antibiotics: randomised double blind placebo controlled trial
BMJ. 2007; doi: 10.1136
Melchart et al. Acupuncture in patients with tension-type headache:
randomised controlled trial BMJ. 2005; doi: 10.1136
Reading List
Course Text
Matthews JNS (2000) An Introduction to Randomised Controlled Trials. Arnold
London ISBN 0-340-76143-1.
Background Reading
Campbell MJ & Machin D (1999) Medical Statistics: A commonsense approach.
John Wiley London
Why Teach Medical Statistics to Mathematics
Students
Motivation (Biostats Group)
• Recruit PhD students.
• Promote medical statistics as a career option for maths students.
• Develop links between the Biostatistics Group and the Statistics Group in the
School of Mathematics.
Motivation (School of Maths)
• Make the BSc in Maths/Maths and Statistics more attractive for applicants with
an interest in statistics.
• Provide an additional 3rd year options.