INVISTA NO SEU SUCESSO:

Network Meta-Analysis

Network Meta-analysis (NMA)

Principles and Applications

What is Network Meta-analysis (NMA)?

Network meta-analysis (NMA) is a statistical method that allows comparison of multiple treatments simultaneously, even if they have never been directly compared against each other. Unlike traditional pairwise meta-analysis, NMA integrates both:

Direct comparisons (treatment A versus treatment B directly compared in trials).
Indirect comparisons (inferring the comparison between A and C through their common comparator, e.g., B).

Why use NMA?

Medicine today faces two major challenges:

Contradictory results across multiple studies for nearly every clinical topic.
Massive amounts of data being published continuously, making interpretation challenging.

Network meta-analysis addresses these issues by:

Synthesizing evidence from multiple studies, enabling comparisons among several treatments at once.
Providing clarity when direct comparisons between treatments do not exist.
Allowing clinicians and policymakers to rank multiple treatment options according to efficacy and safety.
Facilitating informed decision-making in clinical practice and health policies.

How Does NMA Work?

Direct vs. Indirect Comparisons

Direct comparisons:
- Studies that explicitly compare two treatments (e.g., Treatment A vs Treatment B).
Indirect comparisons:
- Estimations of treatment effects between treatments that haven't been directly compared, through a common comparator.

Example:
Imagine you want to know if toothpaste is better than gel at preventing dental cavities, but no studies directly compared these two:

We have studies comparing toothpaste vs placebo and gel vs placebo separately.
By knowing the effectiveness of each treatment against placebo, we can infer how toothpaste compares to gel indirectly.

Formula for indirect comparisons:

Effect (B vs C) = Effect (A vs C) – Effect (A vs B)

Variance (B vs C) = Variance (A vs C) + Variance (A vs B)

Practical Example: Toothpaste vs Gel

Toothpaste vs placebo: Standardized Mean Difference (SMD) = -0.34
Gel vs placebo SMD = -0.19
Indirect SMD (Toothpaste vs Gel) = (-0.34) – (-0.19) = -0.15
Variance: Combined variance from both comparisons.

Combining Direct and Indirect Evidence

Combining direct and indirect evidence can produce more accurate (precise) estimates.
Example (Toothpaste vs. Gel):
- Direct SMD: 0.04
- Indirect SMD: -0.15
- Mixed effect (combined): -0.10 (more precise)

Validity and Potential Issues with NMA

Core assumption of NMA:

The common comparator (e.g., placebo) must be consistent across studies.

Example of a problem:

Comparing a placebo toothpaste (mechanical action) vs. placebo rinse (no mechanical action) might be problematic, as the placebo conditions differ significantly.

Addressing these problems:

Carefully consider the similarity of comparators at the study-design stage.
Conduct heterogeneity and inconsistency assessments using statistical methods (e.g., I² statistics, node-splitting analysis, subgroup analyses).

Validity and Criticism of NMA

Main criticism:

Although indirect comparisons respect original randomization in studies, they themselves are not randomized comparisons.
Requires careful interpretation and consideration of heterogeneity (differences between studies).

Practical Applications of NMA

1. Visual Representation of Evidence

Network plots clearly illustrate relationships between treatments tested.
Example: Comparison of antipsychotic drugs for schizophrenia (Leucht et al., Lancet 2013). Clearly displayed which drugs have been directly compared and where indirect comparisons are made.

2. Evaluating Treatments with No Direct Trials

NMA allows estimating effects for treatments without direct comparison trials.
Example: Exercise vs medications in preventing cardiovascular mortality (Naci, BMJ 2014). NMA showed comparable or superior effects of exercise compared to several medications, despite a lack of direct comparative studies.

2. Ranking Treatments

Provides a ranking of treatments from most to least effective or safe.
Example: Ranking antidepressants (Cipriani, Lancet 2011). Clearly identified antidepressants that were most effective and those associated with more side effects, guiding clinical decision-making.

Conducting NMA Using STATA (step-by-step)

STATA is commonly used software for NMA. Here’s a simplified approach:

Step-by-step Commands in STATA:

// Step 1: Prepare your data (effect sizes, standard errors, studies, and treatments)

network setup effect se, study(studyID) trt(treatment)

// Step 2: Run the NMA (assessing consistency between direct/indirect comparisons)

network meta consistency

// Step 3: Rank treatments (e.g., SUCRA, probability ranking)

network rank min

// Step 3: Generate a network plot

network plot, thickness(proportional)

// Optional: Check for inconsistency and heterogeneity

network meta inconsistency

network forestplot

What each command does:

network setup: Organizes your dataset to perform NMA.
network meta consistency: Runs the main analysis assuming consistency.
network rank: Produces a ranking of treatments.
network plot: Visualizes the entire network of evidence.
network meta inconsistency: Checks the validity of indirect vs direct comparisons.

Critical Considerations for Valid NMA

Transitivity assumption: treatments should be comparable across trials.
Consistency assumption: Direct and indirect comparisons should yield similar results.
Important to carefully assess heterogeneity and consistency through statistical tests.

Conclusion and Summary

Network Meta-analysis (NMA) is a powerful method to:

Integrate evidence across multiple treatments.
Answer clinically relevant questions even when direct evidence is lacking.
Provide a clear summary of relative treatment efficacy and safety.

NMA is increasingly influential, heavily cited, and used by policymakers, clinicians, and guideline developers.