Kirsten Fiest, PhD June 23, 2015 1 CONDUCTING META-ANALYSES IN HEALTH RESEARCH.

Kirsten Fiest, PhD

June 23, 2015

1

CONDUCTING META-ANALYSES IN HEALTH RESEARCH

A statistical method of combining data from multiple independent sources

Powerful tool to compare the effects of interventions, determine the magnitude of association, or prevalence/incidence of disease

Often informed by the results of a systematic review

Can assess differences between subgroups that may not be possible in individual studies

2

META-ANALYSIS

SAS, STATA, SPSS, R, spreadsheets, RevMan

Graphics quality will differ

Easiest to start with data in a spreadsheet

Will need, at minimum, study identifier, effect size, and measure of error

3

STATISTICAL PROGRAMS

Fixed effect Assumes there is one true effect size to be estimated Pooled estimate is the common effect size Weighting is based entirely on the size of the study Only source of error is within studies

Random effects Allows the true effect to vary from study to study Trying to estimate the mean of a distribution of true

effects Weights assigned are more balanced Can be error within and between studies

4

ANALYTICAL METHODS

5

INTERPRETING META-ANALYTIC OUTPUT

Random-Effects Model (k = 11; tau^2 estimator: REML) 

I^2 (total heterogeneity / total variability): 99.69% 

Test for Heterogeneity: Q(df = 10) = 3990.9717, p-val < .0001

 Model Results:

 estimate se zval pval ci.lb ci.ub

-3.0934 0.9093 -3.4018 0.0007 -4.8757 -1.3111

6

INTERPRETING A FOREST PLOT

Clinical and statistical heterogeneity should be assessed Clinical heterogeneity

Factors known to influence the relationship under consideration

Eg. disease duration, age, sex

Statistical heterogeneity Measured most commonly by the I2 and Q statistics Assesses whether any observed differences may be due

to chance alone Interpret with caution (power)

7

HETEROGENEITY

8

STRATIFICATION

Potential bias for journals to publish large studies with significant results

Statistical tests to determine its presence Funnel plots Examine visually and

statistically Begg’s test is a rank

correlation method Egger’s test is a

regression-based method Trim and fill

9

PUBLICATION BIAS

Used to identify trends across an extraneous variable

Allows for the inclusion of continuous or categorical variables

Is the incidence of dementia changing over time?

Does the prevalence of epilepsy differ by geographic region?

10

META-REGRESSION

Method of comparing treatment effects

Pool data from multiple studies with one common arm

Can assess direct and indirect effects

11

NETWORK META-ANALYSES

ACT, behavioural activation; CBT, cognitive-behavioural therapy; DYN, psychodynamic therapy; IPT, interpersonal therapy; PLA, placebo; PST, problem solving therapy; SST, social skills training; SUP, supportive counselling; UC, usual care; WL, waitlist.

Barth et al., PLOS Med, 2013, 10(5)

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses)

MOOSE (Meta-Analysis of Observational Studies in Epidemiology)

Consider reporting guidelines for initial studies as well (STARD, STROBE, CONSORT)

12

REPORTING GUIDELINES

13

14Patten et al.; CJP, 2014, 59(11):60-614

15

Limited by the reporting of individual papers Definitions, estimates provided, basic study details

Quality of individuals studies may vary

Heterogeneity between estimates may weaken some conclusions

16

LIMITATIONS

Reporting guidelines

Systematic Reviews in Health Care: Meta-Analyses in Context, 2nd Edition. Egger, Smith & Altman. 2008.

Journals in your field of interest

RESOURCES

17

18