Takaisin Tulosta

Calculation of relative mean difference

Lisätietoa aiheesta
Aleksi Raudasoja
3.3.2026

Background

To provide comparable and applicable estimates on the benefits of different treatments, we primarily use relative treatment effects for both binary and continuous variables. For binary variables, this means using relative risks, and for continuous variables, relative mean difference (RMD) or ratio of means (ROM). These relative methods are less used for continuous outcomes; however, it may give a more accurate comparison between different treatments. This is because the absolute treatment effects for continuous outcomes are also often correlated with baseline symptoms ( «Furukawa TA, Levine SZ, Tanaka S, ym. Initial seve...»1, «Stone MB, Yaseen ZS, Miller BJ, ym. Response to ac...»2). Therefore, if treatments are tested on patients with different baseline severity of disease or risk profiles, absolute effects may not represent true differences in treatment effect. Relative effect indicates the average impact expected from treatment versus control (e.g., not starting a treatment/placebo).

However, if empirical evidence exists for the absolute treatment effect not being correlated with the baseline severity, then the mean difference may be a better option to present the treatment benefit.

Data conversion methods

Relative risk is usually reported for binary variables, so they rarely require conversion.

For continuous variables, RMD or ROM are rarely reported. Below, we describe the methods used in calculation:

  1. Ratio of means (ROM)
    1. a. (1 – ROM) = RMD
  2. Mean difference (MD)
    1. a. MD / weighted control mean
    2. b. Where the control mean is estimated with the same weighting as the meta-analysis
  3. Standardized mean difference (SMD)
    1. a. Since SMD = MD / SDpooled
    2. b. RMD = SMD x SDpooled / weighted control mean
    3. c. weighted average SD may be used to convert meta-analysis results to MD
  4. 4. Where none of these data is available or the data is of poor quality (e.g., mistakes or fixed effects meta-analysis used where a random effects model should have been used)
    1. a. Calculate ROM by extracting the data from individual studies
  5. Conversion to percentage scale
    1. a. (1-ROM) x 100 %
    2. b. RMD x 100 %

References

  1. Furukawa TA, Levine SZ, Tanaka S, ym. Initial severity of schizophrenia and efficacy of antipsychotics: participant-level meta-analysis of 6 placebo-controlled studies. JAMA Psychiatry 2015;72(1):14-21 «PMID: 25372935»PubMed
  2. Stone MB, Yaseen ZS, Miller BJ, ym. Response to acute monotherapy for major depressive disorder in randomized, placebo controlled trials submitted to the US Food and Drug Administration: individual participant data analysis. BMJ 2022;378():e067606 «PMID: 35918097»PubMed