What rankings estimate
Rankings summarize the ordering implied by network estimates. A P-score can be understood as a frequentist ranking measure based on the extent of certainty that one treatment is better than competing treatments.
A simplified interpretation is P_i = average over j of P(treatment i is better than treatment j), with direction determined by whether lower or higher outcome values are beneficial. Exact computation is handled by the netmeta ranking functions.
Rankograms
A rankogram displays the probability or relative support for each treatment occupying each rank. In Bayesian NMA this is based on posterior rank probabilities. In frequentist netmeta workflows, rankogram support depends on available package functionality and may not be available for all exact binary methods.
Coventra reports P-scores where estimable and returns a status message when rankogram plots are not implemented for a selected exact binary network method.
Changeable settings
- Ranking direction is user-changeable: lower values can be better for harms or higher values can be better for benefits.
- Common or random model output can determine which ranking values are used where both are available.
- Rankings are optional NMA modules, so users can omit them to reduce latency and avoid overemphasis.
Limitations
- A first-ranked treatment can still have wide intervals and low certainty.
- Rankings can be unstable in sparse networks.
- Rankings are sensitive to outcome direction, treatment grouping, and transitivity violations.
- Rankings do not include harms, feasibility, cost, or certainty unless the review team interprets them alongside those domains.
Reporting checklist
- Report ranking direction.
- Report rankings next to effect estimates and uncertainty intervals.
- State whether rankogram output was available for the selected NMA method.
- Avoid recommending treatment order from rankings alone.
- Discuss certainty of evidence and direct evidence support.