“EBHC Statistical Toolkit” David M. Thompson Dept. of Biostatistics and Epidemiology College of Public Health, OUHSC Learning to Practice and Teach Evidence-Based Health Care Fifth Annual Workshop September 24-25, 2010 1 5th Annual EBHC Workshop 9-24-2010
21
Embed
“EBHC Statistical Toolkit” David M. Thompson Dept. of Biostatistics and Epidemiology College of Public Health, OUHSC Learning to Practice and Teach Evidence-Based.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
“EBHC Statistical Toolkit”
David M. ThompsonDept. of Biostatistics and Epidemiology
College of Public Health, OUHSC
Learning to Practice and Teach Evidence-Based Health Care
Fifth Annual WorkshopSeptember 24-25, 2010
5th Annual EBHC Workshop 9-24-2010
2
Statistical tools answer questions
by testing hypotheses
and generating p-values
by estimating parameters
and generating confidence intervals
on those estimates
5th Annual EBHC Workshop 9-24-2010
3
Glossaries and online calculators
• 5th Annual Workshop - Learning to Practice and Teach EBHC
• OUHSC Bird Library - Evidence Based Healthcare
• Duke - UNC Chapel Hill Intro to EBP
• EBM calculators at Can. Inst. of Health Research
Absolute |EER - CER| Harmful or beneficial events per person
“Number neededto …”
1/ |EER - CER|
Persons per harmful or beneficial event
NNT NNH NNT
9
Outcomes measured in other ways require other statistical tools
5th Annual EBHC Workshop 9-24-2010
10
Boilerplate
“Continuous variables were analyzed using t-tests or, when appropriate, their nonparametric analogs. Associations between categorical variables were assessed using Chi-square tests or, when expected values were small, Fisher’s exact tests.”
5th Annual EBHC Workshop 9-24-2010
11
Statistical tools fit the features of the question
• P Population• I Intervention, prognostic factor, or
exposure• C Comparison group• O Primary outcome• (Study design)
5th Annual EBHC Workshop 9-24-2010
12
Statistical tools fit the features of the question
5th Annual EBHC Workshop 9-24-2010
OutcomeComparison group defined by Intervention or Exposure
Population CovariatesAge, SexDisease SeverityComorbid conditions
13
Features of statistical model
• Statistical interaction or
“effect modification”
• Correlated observations of the outcome
• Multiple comparisons
5th Annual EBHC Workshop 9-24-2010
14
Interaction between marital status and C1 enrollment regarding incidence of infant death
5th Annual EBHC Workshop 9-24-2010
15
Certain study designs obtain(and take advantage of) nonindependent (or correlated ) observations of the outcome.
Observations can be correlated• temporally• spatially• hierarchically
5th Annual EBHC Workshop 9-24-2010
16
Statistical tools that appropriatelyhandle correlated observations
• Repeated measures analysis of variance
• Linear mixed models– for numeric outcomes
• Generalized linear models– for outcomes that are binary, categorical,
ordinal, or counts– conditional and marginal models
5th Annual EBHC Workshop 9-24-2010
17
Multiple comparisons
The probability of detecting and reporting differences that don’t truly exist accumulates in a study that examines several hypothesis tests.
5th Annual EBHC Workshop 9-24-2010
185th Annual EBHC Workshop 9-24-2010
19
The right statistical tool for the question.
“Between-group differences in HbA1c were assessed using a mixed regression model that accounted for the study’s repeated and, therefore, correlated measurements on each subject. …”
5th Annual EBHC Workshop 9-24-2010
20
“… Hypothesis testing focused on the model’s estimate of group*time interaction to assess whether change in HbA1c over time differed between the treatment groups. …”
5th Annual EBHC Workshop 9-24-2010
21
“…The model also produced stratum-specific estimates of the change in HbA1c levels over time (in mg/dL/year) along with 95% confidence intervals.”