OBGYN MSc Stat Module (2016-2018) :
Contents_4. Meta-analysis
Please note : the data presented in all course material for the statistical module are generated by computers to demonstrate the methodologies, and should not be confused with actual clinical information
Introduction Heterogeneity Publication Bias Combine Data
Since the mid-1980s, there is an increasing tendency for clinicians to base their managements, not from any single set of research results, but from a comprehensive search and collation of publications concerning a clinical problem. Such a practice is generally termed Evidence Based Practice, and the process of collecting, collating, and evaluating available research is term the Systematic Review.

Systematic Review is a large subject, and covered elsewhere in the MSc. Course. The statistical module will cover the statistical procedures used, meta-analysis.

Meta-analysis itself is a large subject, consisting of many methods and procedures to handle data obtained from multiple sources. In this module, only the most commonly used procedures covering parametric data from one or two group projects are covered. These are

  • Heterogeneity : to determine whether the studies are sufficiently similar to be pooled
  • Publication Bias : to determine whether there is a bias in the selection of reports to be included
  • Combining data : to produce a summary conclusion
Data input

The data is a two column table.

  • Each row consists of data from a study to be included in the meta-analysis
  • Column 1 is the Effect Size
  • Column 2 is the Standard Error of the Effect Size
Type of Effect that can be included in Meta-analysis

Any statistical affect that can be presumed to be Normally distributed, and accompanied by its Standard Error, can be subjected to meta-analysis. Examples are as follows

  • Single Group Effects
    • Mean and the Standard Error of the Mean
    • Proportion and the Standard Error of proportion
    • Fisher's Z Transformed Pearson's Correlation and its Standard Error
  • Two Group analysis
    • Difference between two means and the Standard Error of the difference
    • Risk Difference and its Standard Error
    • Log(Risk Ratio) and its Standard Error
    • Log(Odds Ratio) and its Standard Error

Example data

Group 1
Oxytocics + Ergometrine
Group 2
oxytocin
Difference
(mean1 - mean2)
SEdiff
n1mean1SD1n2mean2SD2
100750250120790275-4035.7
360850175355875200-2514.0
5060027555700300-10056.3
8090022578950250-5037.8
5012503308075028550054.6

Discussions in subsequent panels on his page will use the default example data from StatPgm 4. Meta-Analysis. The data is computer generated and does not represent real clinical information.

EffectSE
-4035.7
-2514.0
-10056.3
-5037.8
50054.6
The data is from 5 imagined controlled trials comparing the use of oxytocin + ergometrine against using oxytocin only in the management of the third stage of labour, the outcome measurement being the volume of blood loss. Data from these trials are presented in the table to the left.

The last 2 columns, the effect (mean1 - mean2) and its Standard Error (SE) are tabulated as to the right, and used for demonstrating meta-analysis on this page.