Calculating confidence intervals for a net-effect estimate




Poster session 2 Thursday: Evidence synthesis - methods / improving conduct and reporting


Thursday 14 September 2017 - 12:30 to 14:00


All authors in correct order:

Alper BS1, Thabane L2, Kunnamo I3, Iorio A2
1 EBSCO Health DynaMed Plus, USA
2 McMaster University, Canada
3 Duodecim, Finland
Presenting author and contact person

Presenting author:

Brian Alper

Contact person:

Abstract text
Background: Evidence-based medicine (EBM) is the integration of clinical expertise, patient values and current best evidence into the decision-making process for patient care. Practicing EBM involves determining the balance of benefits and harms for an individual patient, including consideration of all important outcomes and their relative importance. A decision analysis has previously seemed too complex for routine use in practicing EBM.

Objectives: We developed a simplified decision-analysis model to combine effect estimates for important outcomes and report a net effect estimate with a 95% confidence interval.

Methods: Statistical formulas for combinations of effect estimates were evaluated to understand the assumptions necessary for their use. Practical methods were devised to represent these concepts for clinical decisions and the process was demonstrated with three clinical examples.

Results: Assumptions for use of the model are that effect estimates: 1) have data that conform to a normal distribution; 2) are independent and not correlated with each other; and, 3) are expressed using the same units of measure. Each effect estimate is assigned a multiplier which conveys relative importance and converts all effect estimates to reference unit of measure. Statistical formulas to derive standard deviations of the mean from confidence interval widths for each effect estimate, then produce a 95% confidence interval for a net effect estimate, do not require statistical software.

Conclusions: For relatively simple decisions (choosing between options without dependencies on a series of decisions) with understanding and acceptance of three initial assumptions, a simple decision analysis can be used to generate a net effect estimate with a 95% confidence interval to determine the likelihood of net benefit or net harm. This model could be used for individual decision making with the individual’s preferences defining the relative importance multipliers. For population-level recommendations as occurs in guidelines, the range of plausible relative importance multipliers can be used for a sensitivity analysis to identify preference-sensitive decisions.