Absolute event probability from a baseline risk and a network odds ratio

Applies a treatment effect from a log odds ratio network to a baseline probability on the reference treatment, taken from a separate baseline natural history model. On the logit scale the effect adds, logit(p_k) = logit(p_0) + d_k, with OR_k = exp(d_k); the stated form is the same relation solved for p_k without logarithms. The ARR against the reference is then p_0 minus p_k.

Signature

p_k = p_0 * OR_k / (1 - p_0 + p_0 * OR_k)
Inputs
InputsDefinitionUnit
p_0Probability of the event with the network reference treatment in the target populationprobability from 0 to 1
OR_kOdds ratio of treatment k relative to the reference, the exponent of the network log odds ratio d_kratio, no unit
Output
p_kAbsolute probability of the event with treatment k over the period of the baseline probabilityprobability from 0 to 1

Function

Network consistency function

Expresses every pairwise relative effect in a connected network through effects relative to a common reference treatment A, on a scale where effects add, such as the log odds ratio. Here d_XY is the effect of Y relative to X, the NICE Decision Support Unit convention. Consistency is the assumption that direct and indirect evidence estimate the same d_XY.

Try this function

Implementations

  • Excel

    Treatment probability from a baseline and a log odds ratio

    With the baseline probability in BaselineProb and the network log odds ratio in LogOR, Excel adds the effect on the logit scale and converts back.

    =1/(1+EXP(-(LN(BaselineProb/(1-BaselineProb))+LogOR)))

Assumptions

  • Separate baseline model and additivity on the logit scale

    The baseline probability comes from evidence relevant to the target population and is not counted again as relative-effect evidence. The treatment effect is constant on the log odds scale across baseline risks; models linear in log risk ratios or log hazard rates use the corresponding transformation instead.

Worked examples

  • Baseline probability of 0.25 and an odds ratio of 0.8

    A probability of 0.25 on standard care and an odds ratio of 0.8 give a treatment probability of about 0.2105, reported as 0.21 in the introduction of TSD 5.

    p_0 = 0.25; OR_k = 0.8; p_k = 0.2105

Common errors

  • Applying the network odds ratio as a risk ratio

    Multiplying 0.25 by 0.8 gives 0.20 rather than about 0.2105. The error grows as the baseline probability rises and can give probabilities above 1 for odds ratios above 1.

Sources

  • Baseline natural history combined with relative effects

    Dias S, Welton NJ, Sutton AJ, Ades AE. NICE DSU Technical Support Document 5: Evidence synthesis in the baseline natural history model. Sheffield: Decision Support Unit, ScHARR, University of Sheffield; August 2011, last updated April 2012. Section 1 Introduction (the absolute probability on treatment from logit(p) = logit(0.25) + ln(0.8), giving 0.21, and separate baseline and relative effect models).

    View source →

  • Absolute effects from a logit-link network model

    Dias S, Welton NJ, Sutton AJ, Ades AE. NICE DSU Technical Support Document 2: A generalised linear modelling framework for pairwise and network meta-analysis of randomised controlled trials. Sheffield: Decision Support Unit, ScHARR, University of Sheffield; August 2011, last updated September 2016. Section 2 (logit link for binomial data) and the appendix code for absolute effects (logit(T[k]) equals the baseline log odds A plus d[k]).

    View source →

Canonical Identity