Understanding Data Types and Effect Measures in Network Meta-Analysis (NMA)
Abstract
Network meta-analysis (NMA) allows simultaneous comparison of multiple interventions across studies by combining direct and indirect evidence. Understanding data types and their corresponding effect measures is essential to ensure correct modeling, interpretation, and comparability across networks. This article clarifies how categorical/discrete and continuous data operate within the NMA framework, including their subtypes and the logic of effect sizes.
1. Introduction
A network meta-analysis (NMA) extends conventional pairwise meta-analysis by integrating evidence across multiple interventions, even when not all have been directly compared. The validity of NMA results relies heavily on correctly identifying data types and effect measures appropriate for the included outcomes.In practice, outcomes in clinical research are broadly classified as either categorical/discrete or continuous, each demanding distinct statistical modeling strategies.
2. Categorical (Discrete) Data
Categorical data represent outcomes that occur as events or counts rather than continuous measurements. These are analyzed using relative measures of effect to compare intervention groups.
Subtypes
| Subtype | Description | Example | Common Effect Measure |
| Binary (Dichotomous) | Two possible outcomes | Death vs Alive | Odds Ratio (OR), Risk Ratio (RR), Risk Difference (RD) |
| Ordinal | Ordered categories | Mild / Moderate / Severe | Ordinal Odds Ratio, Proportional Odds Model |
| Count | Number of events per participant or time | Hospital visits per year | Rate Ratio, Incidence Rate Ratio (IRR) |
Binary and count outcomes are often modeled through logistic or Poisson regression frameworks, depending on whether events are single or recurrent. Ordinal data can be analyzed via proportional odds or cumulative logit models, depending on proportionality assumptions.
3. Continuous Data
Continuous outcomes represent numerical measurements on a scale, suitable for estimating mean differences between groups.
Subtypes
| Subtype | Description | Example | Effect Measure |
| Continuous | Quantitative measure on a fixed scale | Pain score (0–10), BP (mmHg) | Mean Difference (MD), Standardized Mean Difference (SMD) |
| Time-to-Event (Survival) | Duration until event occurs | Time to relapse, time to death | Hazard Ratio (HR) |
While classical continuous data rely on normal-distribution assumptions, time-to-event data are treated as continuous because they represent time, modeled through survival analysis (e.g., Cox proportional hazards models). The hazard ratio (HR) is the natural effect measure in this context.
4. Integrating Data Types in Network Meta-Analysis
When constructing an NMA:
- Studies within the same network should ideally use comparable outcome types and effect measures.
- Transformations (e.g., converting odds ratios to risk ratios) may be applied for harmonization, though they introduce assumptions.
- Bayesian or frequentist models can accommodate mixed outcome types, but consistency and transitivity must be verified.
Effect measures used in NMAs:
| Outcome Type | Example Outcome | Common Model | Effect Measure |
| Binary | Remission, mortality | Logit model | OR / RR |
| Count | Hospitalizations | Poisson or Negative Binomial | Rate Ratio |
| Ordinal | Severity score | Cumulative logit model | OR |
| Continuous | Blood pressure | Linear model | MD / SMD |
| Time-to-event | Survival time | Cox model / parametric | HR |
5. Conceptual Implications
Correctly categorizing the data ensures:
- Appropriate statistical model selection (e.g., binomial vs Gaussian likelihood).
- Correct weighting of studies in the network.
- Accurate interpretation of treatment ranking and SUCRA (Surface Under the Cumulative Ranking).
Furthermore:
- Time-to-event data, though event-based, belong conceptually to continuous outcomes.
- Count and ordinal data, despite numeric representation, are categorical/discrete in statistical behavior.
6. Summary Table
| Category | Subtypes | Typical Measures |
|---|---|---|
| Categorical / Discrete | Binary, Ordinal, Count | OR, RR, Rate Ratio |
| Continuous | Continuous, Time-to-event | MD, SMD, HR |
7. Key Takeaways
- NMA outcome classification is foundational for selecting the correct statistical framework.
- Binary, ordinal, and count data belong to the categorical/discrete family.
- Continuous and time-to-event outcomes belong to the continuous family.
- Effect sizes—OR, RR, MD, HR—must be interpreted according to the nature of the data and model used.
- Consistency across measures and outcome types enhances the interpretability and validity of network meta-analysis results.