What is a network meta-analysis?
🔗 What is Network Meta-Analysis (NMA)?
Network meta-analysis (NMA) is a statistical method that allows researchers to compare several treatments at the same time, even if some of those treatments have never been directly compared in head-to-head clinical trials.
Traditional meta-analysis usually compares two treatments at a time (for example, Drug A vs Drug B). This is called pairwise meta-analysis.
Network meta-analysis is an extension of this approach. It combines results from many randomized controlled trials (RCTs) to create a network of treatment comparisons. Using this network, researchers can make indirect comparisons between treatments that were not directly tested against each other.
In simple terms:
> Network meta-analysis allows researchers to compare multiple treatments using both direct evidence (head-to-head trials) and indirect evidence (through a common comparator).
This approach is also sometimes called multiple treatment comparison (MTC).
🔎 How indirect comparisons work
Imagine we have studies that compare:
- Drug A vs placebo
- Drug B vs placebo
Even if Drug A and Drug B were never compared directly, researchers can still estimate how they compare indirectly through placebo.
This creates a network of evidence linking all treatments.
🏥 Clinical Example
Researchers want to determine which antidepressant works best for major depressive disorder. Some trials compare fluoxetine vs placebo, others compare sertraline vs placebo, and others compare venlafaxine vs fluoxetine. Network meta-analysis combines all these studies to estimate how all antidepressants compare with each other, even when some drugs were never directly compared.
🧒 Explain it to a 10-year-old
Imagine three runners who never raced each other.
If Runner A beat Runner C, and Runner B also beat Runner C, you can guess how Runner A and Runner B might compare, even though they never raced directly.
🧠 Memory Hook
“Network meta-analysis = building a web of trials so treatments can compete even if they never fought directly.” 🕸️
Network meta-analyses are useful for 3 main reasons.
1. They allow visualisation of a large amount of evidence.
2. They allow estimation of the relative effectiveness of multiple treatments
3. They allow ranking of treatments
Explain point 1.
🔗 What are Network Meta-Analyses (NMA) Useful For?
Network meta-analyses are useful because they allow researchers to compare many different treatments at the same time using evidence from multiple studies. This helps researchers understand how different interventions perform relative to each other.
Below are the main reasons they are useful.
1️⃣ They allow visualisation of a large amount of evidence
Network meta-analysis often produces a network diagram that shows how different treatments are connected through various clinical trials.
Each treatment is represented as a node, and lines between them represent studies that compared those treatments.
This visual network helps researchers see how much evidence exists and how treatments are linked.
🏥 Clinical example
Researchers studying treatments for schizophrenia create a network diagram linking antipsychotics such as risperidone, olanzapine, quetiapine, and placebo. The diagram shows which drugs have been directly compared in clinical trials.
🧒 Explain it to a 10-year-old
Imagine a map showing which friends have played games against each other. The map helps you see who has competed with whom.
🧠 Memory Hook
“NMA draws the spider web of studies so you can see how treatments are connected.” 🕸️
Network meta-analyses are useful for 3 main reasons.
1. They allow visualisation of a large amount of evidence.
2. They allow estimation of the relative effectiveness of multiple treatments
3. They allow ranking of treatments
Explain point 2.
🔗 What are Network Meta-Analyses (NMA) Useful For?
Network meta-analyses are useful because they allow researchers to compare many different treatments at the same time using evidence from multiple studies. This helps researchers understand how different interventions perform relative to each other.
Below are the main reasons they are useful.
2️⃣ They allow estimation of the relative effectiveness of multiple treatments
By combining both direct comparisons (head-to-head trials) and indirect comparisons, network meta-analysis can estimate how effective all treatments are relative to each other.
This allows researchers to answer questions like which treatment works better overall.
🏥 Clinical example
A network meta-analysis comparing different antidepressants can estimate which medications are more effective in treating major depressive disorder, even when some drugs have never been directly compared in the same trial.
🧒 Explain it to a 10-year-old
If two kids each played against the same opponent, you can guess who might be better, even if they never played each other.
🧠 Memory Hook
“NMA lets treatments compete even if they never fought in the same study.” 🥊
Network meta-analyses are useful for 3 main reasons.
1. They allow visualisation of a large amount of evidence.
2. They allow estimation of the relative effectiveness of multiple treatments
3. They allow ranking of treatments
Explain point 3.
🔗 What are Network Meta-Analyses (NMA) Useful For?
Network meta-analyses are useful because they allow researchers to compare many different treatments at the same time using evidence from multiple studies. This helps researchers understand how different interventions perform relative to each other.
3️⃣ They allow ranking of treatments
Network meta-analysis can produce a rank order of treatments, showing which interventions are likely to be most effective and which are least effective.
This helps clinicians and policymakers make better treatment decisions.
🏥 Clinical example
A network meta-analysis of treatments for generalised anxiety disorder may rank therapies such as CBT, SSRIs, and benzodiazepines according to their effectiveness and tolerability.
🧒 Explain it to a 10-year-old
It’s like creating a leaderboard in a video game showing which players are the best.
🧠 Memory Hook
“NMA turns research into a treatment leaderboard.” 🏆
What is the adjusted indirect treatment comparison (ITC) method, aka the ‘Anchored ITC’?
🔗 What is the Adjusted Indirect Treatment Comparison (ITC) or Anchored ITC?
The adjusted indirect treatment comparison (ITC) method is a statistical approach used to compare two treatments even when they have never been directly compared in the same clinical trial.
This method was first described by Bucher et al. (1997). It allows researchers to estimate the effect of Treatment A vs Treatment C by using results from studies that compared:
- A vs B, and
- B vs C
In this situation, Treatment B acts as the “anchor” or common comparator linking the two studies. Because both A and C were compared to B, researchers can indirectly estimate how A and C compare with each other.
The method was originally developed using odds ratios (ORs) as the measure of treatment effect.
In simple terms:
> If two treatments were each compared with the same third treatment, we can use that shared comparison to estimate how the two treatments compare with each other.
⚠️ Limitation
A key limitation is that this method only works in simple situations involving three treatments and two trials.
In other words:
- One study compares A vs B
- Another study compares B vs C
- Researchers then estimate A vs C indirectly
Because it relies on this simple structure, it cannot easily handle large networks of many treatments, which is why network meta-analysis is often used instead.
🏥 Clinical Example
One trial compares Drug A (sertraline) with placebo (B) for depression, and another trial compares Drug C (venlafaxine) with placebo (B).
Using the anchored ITC method, researchers can estimate how sertraline compares with venlafaxine, even though the two drugs were never directly compared in the same trial.
🧒 Explain it to a 10-year-old
Imagine two kids each raced against the same runner.
If Kid A beat that runner by a lot and Kid C only beat them by a little, you can guess that Kid A might be faster than Kid C, even though they never raced each other.
🧠 Memory Hook
“Anchored ITC = compare two treatments by tying them to the same anchor.” ⚓
What is the Lumley method of network meta-analysis?
🔗 What is Network Meta-Analysis (Lumley)?
Network meta-analysis (NMA) is a statistical method used to compare many treatments at the same time by combining both direct and indirect evidence from clinical trials.
It is also known as multiple treatment comparison (MTC). This approach allows researchers to compare treatments even if some of them have never been directly compared in the same clinical trial.
The method described by Lumley allows comparisons between treatments when more than one common comparator or linking treatment connects them in a network of studies.
In simple terms:
> Network meta-analysis builds a network of studies linking treatments together, allowing researchers to estimate how all treatments compare with each other.
🕸️ Understanding the Network Diagram
In a network meta-analysis diagram:
- Each node (circle) represents a treatment or intervention.
- Lines connecting nodes represent randomised controlled trials (RCTs) where those treatments were directly compared.
- Thicker lines indicate that more studies exist comparing those treatments.
🔁 What are Closed Loops?
Sometimes the network forms closed loops, meaning several treatments are linked through both direct and indirect comparisons.
For example:
- If studies compare B vs C, C vs E, and B vs E, these treatments form a closed loop.
Closed loops allow researchers to cross-check evidence and improve the reliability of comparisons.
🔄 How indirect comparisons work
Indirect comparisons occur when treatments are compared through a common comparator.
For example:
- If studies compare A vs B and B vs D, researchers can estimate A vs D using B as the linking treatment.
Similarly:
- If E vs F and F vs G are studied, researchers can estimate E vs G using F as the comparator.
🏥 Clinical Example
A network meta-analysis compares several antipsychotic medications for schizophrenia. Some trials compare risperidone vs placebo, others compare olanzapine vs placebo, and others compare quetiapine vs risperidone. By linking these trials together, the analysis estimates how all antipsychotics compare with each other, even when some drugs were never directly compared.
🧒 Explain it to a 10-year-old
Imagine a group of runners where some have raced each other and some haven’t.
By looking at all the race results together, you can figure out who is probably faster even if two runners never raced directly.
🧠 Memory Hook
“Network meta-analysis = a spider web of studies where every treatment can compete through the network.” 🕸️
“Incoherence” or “inconsistency” is a type of bias that can arise in a network meta-analysis. Explain what it is.
⚠️ What is Incoherence (or Inconsistency) in Network Meta-Analysis?
Incoherence (also called inconsistency) occurs when the direct evidence and indirect evidence in a network meta-analysis do not agree with each other.
In network meta-analysis, researchers combine:
- Direct comparisons (head-to-head trials, e.g., A vs B)
- Indirect comparisons (e.g., estimating A vs C through B)
Ideally, these estimates should point to the same conclusion.
However, inconsistency occurs when these results conflict, suggesting that the network model may not be reliable.
In simple terms:
> Incoherence means the different paths of evidence in the network give different answers about which treatment works better.
Because of this, researchers must investigate the cause of the inconsistency to ensure the analysis is robust.
🔎 Why inconsistency can occur
Inconsistency often happens because the studies being compared are not truly similar. Possible reasons include:
- Different patient populations in different studies
- Different versions of a treatment (e.g., different doses or regimens)
- Studies conducted in different time periods, settings, or healthcare contexts
For example, patients in trials comparing A vs B may differ from those in B vs C trials.
🏥 Clinical Example
Suppose trials show that Drug A is better than Drug B, and Drug B is better than Drug C. Based on indirect comparison, we would expect Drug A to be better than Drug C.
However, if a direct A vs C study shows Drug C is actually better than Drug A, the network contains inconsistency, suggesting that differences in patient populations or treatment protocols may be influencing the results.
🧒 Explain it to a 10-year-old
Imagine three runners.
If Runner A beat Runner B and Runner B beat Runner C, you would expect Runner A to beat Runner C. But if Runner C suddenly beats Runner A in another race, something doesn’t add up.
🧠 Memory Hook
“Inconsistency = when the research triangle doesn’t add up.” 🔺
Network meta-analyses follow several assumptions to reduce bias.
These include:
1. Similarity assumption
2. Homogeneity assumption
3. Consistency and transivitivity assumption
What is the “similarity assumption”?
🔎 What is the Similarity Assumption in Network Meta-Analysis?
The similarity assumption is an important rule used when designing a network meta-analysis (NMA). It means that the studies included in the analysis should be sufficiently similar to each other so that their results can be fairly compared.
Researchers should only combine studies if they are comparable in important aspects such as:
- Patient characteristics (e.g., age, disease severity)
- Study design
- Outcomes being measured
- Follow-up duration
- Treatment conditions (dose, regimen, etc.)
These factors are called effect modifiers, because they can influence how well a treatment works.
If studies differ too much in these factors, the comparison becomes unfair or biased, which can lead to heterogeneity and inconsistency in the network meta-analysis.
In simple terms:
> The similarity assumption means that studies included in a network meta-analysis should be similar enough that differences in results are due to treatments, not differences between the studies themselves.
If the studies are sufficiently similar, researchers assume they are measuring the same underlying treatment effects, and any differences between study results are mostly due to chance rather than systematic differences.
🏥 Clinical Example
A network meta-analysis compares several antidepressants for major depressive disorder. For the similarity assumption to hold, the studies should include similar patient populations (e.g., similar depression severity), similar treatment durations, and comparable outcome measures. If some studies include only mild depression while others include severe treatment-resistant depression, the comparisons may become unreliable.
🧒 Explain it to a 10-year-old
Imagine comparing how fast different kids run in races.
To make it fair, the races should have kids of similar ages running the same distance, not adults racing toddlers.
🧠 Memory Hook
“Similarity assumption = compare apples with apples, not apples with astronauts.” 🍎🚀
Network meta-analyses follow several assumptions to reduce bias.
These include:
1. Similarity assumption
2. Homogeneity assumption
3. Consistency and transivitivity assumption
What is the “homogeneity assumption”?
🔎 What is the Homogeneity Assumption in Network Meta-Analysis?
The homogeneity assumption means that the results of studies comparing the same treatments should be reasonably consistent with each other.
In pairwise comparisons (for example, several trials comparing Treatment A vs Treatment B), the studies should produce similar estimates of treatment effect.
If the results vary greatly between studies, this is called heterogeneity, and it suggests that something important differs between the studies, such as:
- different patient populations
- different treatment doses or regimens
- different study designs or follow-up periods
For the homogeneity assumption to hold, there should be no major or meaningful differences in the results across studies comparing the same treatments.
In simple terms:
> The homogeneity assumption means that studies comparing the same treatments should produce similar results, so they can be reliably combined in a meta-analysis.
If this assumption is violated, the pooled estimate from the meta-analysis may not be trustworthy.
🏥 Clinical Example
Several trials compare sertraline vs placebo for major depressive disorder. If all the trials show a similar level of improvement with sertraline, the results are considered homogeneous. However, if some studies show very strong benefit while others show almost none, heterogeneity exists, suggesting the studies may differ in important ways.
🧒 Explain it to a 10-year-old
Imagine several classes doing the same math test.
If most classes get similar scores, the results are consistent. But if some classes get very high scores and others very low, something different must be happening.
🧠 Memory Hook
“Homogeneity = the studies should sing the same tune.” 🎵
Network meta-analyses follow several assumptions to reduce bias.
These include:
1. Similarity assumption
2. Homogeneity assumption
3. Consistency and transivitivity assumption
What is the “consistency and transivity assumption”?
🔗 What are the Consistency and Transitivity Assumptions in Network Meta-Analysis?
The consistency and transitivity assumptions are important rules that help ensure the results of a network meta-analysis (NMA) are reliable.
1️⃣ Consistency
Consistency means that direct evidence and indirect evidence should agree with each other.
- Direct evidence comes from studies that compare two treatments directly (e.g., A vs B).
- Indirect evidence comes from comparing treatments through a shared comparator (e.g., A vs C estimated using A vs B and B vs C studies).
For the network meta-analysis to be valid, these two types of evidence should produce similar results.
If the direct and indirect estimates are very different, this is called inconsistency, and it suggests there may be bias or important differences between the studies.
2️⃣ Transitivity
Transitivity is the underlying assumption that makes indirect comparisons possible.
It means that if treatment A is compared with B, and B is compared with C, we can indirectly compare A with C, as long as the studies are comparable.
In other words, if:
- A is better than B
- B is better than C
then we would expect:
- A to be better than C
This assumption works only if the studies are similar in important characteristics, such as patient populations, disease severity, and treatment conditions.
🔄 How this works in a network
In closed-loop networks, there may be:
- direct comparisons (A vs C)
- indirect comparisons (A vs B vs C)
If the analysis is valid, both paths should give similar results.
🏥 Clinical Example
Suppose clinical trials show that Drug A is better than Drug B for depression, and Drug B is better than Drug C. Using transitivity, researchers can estimate that Drug A should be better than Drug C. If a direct A vs C study later shows a completely different result, the network would show inconsistency.
🧒 Explain it to a 10-year-old
Imagine three runners in races.
- Runner A beats Runner B
- Runner B beats Runner C
You would expect Runner A to beat Runner C.
If Runner C suddenly beats Runner A, something doesn’t make sense.
🧠 Memory Hook
“Consistency means the triangle of treatments must add up.” 🔺
Statistical analyses in a network meta-analysis can be done either with a 1) frequentist or 2) Bayesian approach.
Explain what the “frequentist” approach is.
📊 What is the Frequentist Approach in Network Meta-Analysis?
The frequentist approach is a traditional statistical method used in network meta-analysis to estimate how effective treatments are based on the observed data from studies.
In this approach, probability is interpreted as the long-run frequency of outcomes if the same study were repeated many times. In other words, the method focuses on what the data show, rather than incorporating prior beliefs or previous knowledge.
Researchers use the study data to calculate:
- a point estimate of the treatment effect (the best estimate of how well a treatment works), such as
- Odds Ratio (OR)
- Risk Ratio (RR)
- Mean Difference
- a 95% confidence interval (CI) around that estimate, which shows the range in which the true effect is likely to lie.
This is similar to the way results are reported in standard pairwise meta-analyses.
In simple terms:
> The frequentist approach estimates treatment effects using only the study data and shows the result with a confidence interval indicating uncertainty.
🏥 Clinical Example
A network meta-analysis compares several antidepressants for major depressive disorder. Using the frequentist approach, the analysis calculates the odds ratio for treatment response for each drug compared with placebo and provides 95% confidence intervals to show how precise those estimates are.
🧒 Explain it to a 10-year-old
Imagine flipping a coin many times to see how often it lands on heads.
The frequentist approach uses the actual results of the flips to estimate how likely heads is.
🧠 Memory Hook
“Frequentist = trust the data, not your gut.” 📊
Statistical analyses in a network meta-analysis can be done either with a 1) frequentist or 2) Bayesian approach.
Explain what the “Bayesian” approach is.
📊 What is the Bayesian Approach in Network Meta-Analysis?
The Bayesian approach is a statistical method that estimates treatment effects by combining the observed data from studies with prior information or beliefs about the treatments.
In this approach, researchers start with a prior belief about the possible values of the treatment effect. This prior information may come from previous studies, expert knowledge, or existing evidence.
When new study data are analysed, the prior belief is updated using the observed data, producing a posterior distribution (an updated estimate of the treatment effect).
This approach models the full probability distribution of the treatment effect, meaning it explicitly accounts for uncertainty in the estimates.
Because of this, Bayesian analyses can make direct probability statements, such as:
- the probability that Treatment A is better than Treatment B
- the probability that one treatment ranks as the best option
In simple terms:
> The Bayesian approach combines previous knowledge with new data to estimate how likely different treatment effects are.
🏥 Clinical Example
A network meta-analysis compares several antipsychotic medications for schizophrenia. Using the Bayesian approach, researchers combine prior evidence about drug effects with the results of current trials and estimate the probability that each medication is the most effective treatment.
🧒 Explain it to a 10-year-old
Imagine guessing how many candies are in a jar.
You might first make a guess based on what you’ve seen before, and then update your guess after counting some candies.
🧠 Memory Hook
“Bayesian = start with a belief, then let the data update your mind.” 🧠📊
Look at this network plot with 4 interventions and a placebo. Can you interpret what it means?
📊 Interpreting the Network Meta-Analysis Plot (4 Interventions + Placebo)
This figure shows the results of a network meta-analysis comparing four treatments (A, B, C, D) and placebo.
The diagram combines two parts:
1️⃣ A network plot (showing how treatments are connected by studies).
2️⃣ A results table (showing how the treatments compare with each other).
1️⃣ What the Network Plot Shows
The network plot on the right shows the treatments as circles (called nodes):
- A
- B
- C
- D
- Placebo
Lines between them represent clinical trials where those treatments were directly compared.
The diagram shows that:
- D and placebo act as common comparators linking the different treatments.
- This means some treatments were not directly compared with each other but can still be compared indirectly through placebo or treatment D.
This network allows researchers to estimate how all treatments compare, even if some were never tested head-to-head.
2️⃣ How to Read the Results Table
The table on the left shows comparisons between treatments.
Important rules for reading it:
- Drugs are listed alphabetically (A, B, C, D, Placebo).
- Comparisons should be read from left to right (Treatment 1 vs Treatment 2).
- Each cell shows an effect estimate (here an odds ratio – OR) with a 95% confidence interval (CI).
Example format:
1.72 (1.05–2.91)
This means:
- 1.72 = estimated treatment effect (OR)
- 1.05–2.91 = range of uncertainty (95% CI)
3️⃣ How to Interpret the Odds Ratio (OR)
- OR > 1 → favours the treatment being compared (treatment 1)
- OR < 1 → favours the other treatment (treatment 2)
If the confidence interval crosses 1, the result is not statistically significant.
In the table:
- Significant results are shown in bold and underlined.
4️⃣ Examples from the Table
A vs B
Result:
1.12 (0.31–3.44)
Because the confidence interval includes 1, there is no statistically significant difference between A and B.
A vs D
Result:
1.72 (1.05–2.91)
This result does not include 1, so it is statistically significant.
➡️ This means Treatment A appears more effective than Treatment D.
A vs Placebo
2.88 (1.27–3.04)
This is also significant, showing Treatment A is better than placebo.
5️⃣ Overall Interpretation
From this network meta-analysis:
- Treatment A appears to perform better than D and placebo
- Some comparisons (like A vs B) show no clear difference
- The network structure allows these comparisons even when direct trials are missing
Because placebo and treatment D connect the network, they allow indirect comparisons between treatments.
✅ Key Idea
Network meta-analysis combines direct evidence (head-to-head trials) and indirect evidence (through shared comparators) to compare multiple treatments in a single analysis.
What is the “fixed effect approach” in the context of a network meta-analysis?
📊 What is the Fixed Effect Approach in Network Meta-Analysis?
The fixed effect approach is a statistical method used in network meta-analysis that assumes all studies are estimating the same true treatment effect.
In this model, researchers assume that the actual effect of the treatment is identical in every study. If the results of the studies are slightly different, those differences are assumed to be caused only by random sampling error (chance variation within each study), not by real differences between studies.
This approach is similar to the fixed effect model used in traditional pairwise meta-analysis.
In simple terms:
> The fixed effect model assumes that all studies are trying to measure one single true treatment effect, and any differences between study results happen just because of random variation.
🔎 What this assumption means
Under the fixed effect approach:
- There is one true effect size for the treatment.
- All studies are considered estimating that same effect.
- Differences between study results are assumed to occur due to sampling error (within-study variation).
This model works best when the studies are very similar in design, population, and treatment conditions.
🏥 Clinical Example
Suppose several clinical trials compare sertraline vs placebo for depression. In a fixed effect model, researchers assume that the true effectiveness of sertraline is the same in all studies, and any small differences between trial results occur because of random variation in the samples.
🧒 Explain it to a 10-year-old
Imagine several thermometers measuring the same temperature.
If they show slightly different readings, you assume the temperature is actually the same, and the small differences happen because the thermometers are not perfectly precise.
🧠 Memory Hook
“Fixed effect = one true answer, many slightly noisy measurements.” 🎯
What is the “random effects approach” in the context of a network meta-analysis?
📊 What is the Random Effects Approach in Network Meta-Analysis?
The random effects approach is a statistical model used in network meta-analysis that assumes the true treatment effect may vary between different studies.
Unlike the fixed effect model (which assumes one single true effect), the random effects model assumes that each study may estimate a slightly different true effect. These differences can occur because studies often involve different patient groups, study conditions, or treatment protocols.
Therefore, the variation in study results comes from two sources:
1️⃣ Sampling error (random variation within each study).
2️⃣ Real differences between studies, known as heterogeneity.
Heterogeneity can occur because studies differ in factors such as:
- Disease severity
- Age of patients
- Drug dose
- Length of follow-up
- Other clinical or methodological differences.
In network meta-analysis, this concept is extended further: effect estimates may vary not only across studies but also across different treatment comparisons, including both direct and indirect comparisons.
In simple terms:
> The random effects model assumes that the true treatment effect is not exactly the same in every study, because real differences exist between the studies.
🏥 Clinical Example
Several studies compare different antidepressants for major depressive disorder. Some studies involve younger patients with mild depression, while others include older patients with severe depression. Because these populations differ, the true treatment effects may vary across studies, so a random effects model is used.
🧒 Explain it to a 10-year-old
Imagine measuring how fast kids run in different schools.
Some schools might have younger kids, different tracks, or different weather, so the results may naturally vary a bit.
🧠 Memory Hook
“Random effects = different studies, different realities.” 🎲
What is a “treatment ranking probability” or “rankogram” or “rank order” in the context of network meta-analyses?
🏆 What is a Treatment Ranking Probability (Rankogram / Rank Order) in Network Meta-Analysis?
In network meta-analysis (NMA), researchers often want to know which treatment is most effective overall. One way to show this is by calculating treatment ranking probabilities.
A treatment ranking probability estimates the chance that each treatment will rank at a particular position (best, second best, third best, etc.) when all treatments are compared.
These probabilities are often displayed in a graph called a rankogram. A rankogram shows the probability that each treatment will occupy each rank in the ordering of treatments from best to worst.
This method is commonly used in Bayesian network meta-analysis, because Bayesian models can estimate the probability of each possible ranking position.
In simple terms:
> A rankogram shows how likely each treatment is to be the best, second best, third best, and so on.
Researchers usually report these rankings alongside the actual treatment effect estimates (such as odds ratios or risk ratios), so readers can see both the size of the treatment effect and the ranking probabilities.
🏥 Clinical Example
A network meta-analysis compares several antidepressants for major depressive disorder. The analysis estimates the probability that each drug is the most effective, second most effective, and so on. For example, one medication might have a 60% probability of being the best treatment, while another has only a 10% probability.
🧒 Explain it to a 10-year-old
Imagine several runners in a race.
Instead of just saying who won once, you calculate how likely each runner is to finish 1st, 2nd, 3rd, or 4th based on lots of race results.
🧠 Memory Hook
“Rankogram = the treatment leaderboard with probabilities.” 🏆
Look at the data below. Interpret it and provide a ranking for which drug is the best to worst.
📊 Interpreting the Treatment Rankings (Network Meta-Analysis)
This figure shows rank probabilities for five treatments: A, B, C, D, and placebo.
The rankings estimate how likely each treatment is to be 1st (best), 2nd, 3rd, 4th, or 5th (worst) based on the network meta-analysis.
The table shows the probability of each treatment occupying each rank. For example:
- A has a 0.46 (46%) chance of being ranked 1st.
- C has a 0.48 (48%) chance of being ranked 2nd.
- B has a 0.49 (49%) chance of being ranked 4th.
- Placebo has an 0.85 (85%) chance of being ranked last (5th).
The bar chart and line graph simply visualize these probabilities.
🏆 Overall Ranking of Treatments (Best → Worst)
Based on the pattern of probabilities across all ranks, the treatments are ordered approximately as:
1️⃣ A – Best treatment
2️⃣ C
3️⃣ D
4️⃣ B
5️⃣ Placebo – Worst
🔎 Why this is the ranking
🥇 A (Best)
- 46% probability of being ranked 1st
- 42% probability of being ranked 2nd
- Very low probability of being among the worst ranks.
➡️ This means A is most often near the top.
🥈 C (Second best)
- Moderate probability of being 1st (20%)
- Highest probability of being 2nd (48%)
➡️ C is usually near the top but slightly behind A.
🥉 D (Third)
- Some chance of being 1st (18%)
- Often ranked 3rd or 4th
➡️ D is moderately effective but not among the very best.
4️⃣ B
- Low probability of being 1st (15%)
- 49% probability of being ranked 4th
➡️ B is usually in the lower rankings.
5️⃣ Placebo (Worst)
- 85% probability of being ranked 5th (last)
➡️ Placebo is almost always the least effective treatment, which is expected.
🧠 Simple Way to Understand the Graphs
The graphs show how often each treatment lands in each position if the analysis were repeated many times.
Think of it like running many races between the treatments and recording how often each one finishes in each position.
✅ Final ranking
** ⚡ The 5-Second Trick to Interpret Rankograms in Network Meta-Analysis **
When you see rank probabilities or rankograms, you don’t need to analyse every number.
You can quickly identify the best and worst treatments using this simple rule.
🧠 Step 1 — Look at the probability of Rank 1 (best)
The treatment with the highest probability of being ranked 1st is usually the best overall treatment.
From the table:
| Treatment | Probability of Rank 1 |
| ——— | ——————— |
| A | 0.46 |
| B | 0.15 |
| C | 0.20 |
| D | 0.18 |
| Placebo | 0.01 |
✅ A has the highest probability of being ranked first (46%) → best treatment.
🧠 Step 2 — Look at the probability of the worst rank
The treatment with the highest probability of the lowest rank is the worst treatment.
| Treatment | Probability of Rank 5 |
| ——— | ——————— |
| A | 0.01 |
| B | 0.04 |
| C | 0.03 |
| D | 0.07 |
| Placebo | 0.85 |
✅ Placebo has an 85% chance of being last → worst treatment.
🧠 Step 3 — Look at the middle ranks
For the remaining drugs, check where the largest probability lies.
| Treatment | Highest probability rank |
| ——— | ———————— |
| C | Rank 2 (0.48) |
| D | Rank 3 (0.44) |
| B | Rank 4 (0.49) |
So the order becomes:
A → C → D → B → Placebo
🎯 The Exam Shortcut
When interpreting rankograms:
1️⃣ Highest probability of Rank 1 → Best drug
2️⃣ Highest probability of Worst Rank → Worst drug
3️⃣ Use the biggest middle probabilities to fill the rest
🧠 Ultra-short Memory Trick
“Top-rank, bottom-rank, fill the middle.” 🥇➡️🥉➡️🥈
**
Rank | Treatment |
| —— | ———– |
| 🥇 1st | A |
| 🥈 2nd | C |
| 🥉 3rd | D |
| 4th | B |
| 5th | Placebo |
The steps of a network meta-analysis include:
1. Define the review question and inclusion criteria
2. Search and select studies
3. Perform titles/ abstract and full-test reading
4. Risk of bias assessment
5. Extraction of data, network building and statistical analysis
6. Synthesis of results
7. Interpretation of results and conclusions
Explain step 1.
Step 1 of a Network Meta-Analysis: Define the Review Question and Inclusion Criteria
The first step in a network meta-analysis (NMA) is to clearly define what question the study is trying to answer and which studies will be included.
Just like in a traditional pairwise meta-analysis, researchers must carefully define the research question. This includes clearly identifying:
- The patient population (e.g., adults with depression)
- The treatments being compared (these treatments become the nodes in the network)
- The outcomes being measured (e.g., symptom improvement)
All the treatments included in the analysis should be precisely defined so that they can be compared accurately. Whenever possible, researchers try to include all relevant treatments or drugs so the network comparison is as complete as possible.
Researchers also define inclusion criteria, which specify which studies will be allowed into the analysis. For example, they might include only:
- Randomized controlled trials (RCTs)
- Studies involving adults
- Studies measuring a specific outcome
To ensure the review is conducted properly and transparently, researchers usually follow established guidelines such as the PRISMA-NMA guidelines (an extension of PRISMA used specifically for network meta-analyses, recommended by groups like the Cochrane Collaboration).
In simple terms:
> Step 1 is about clearly deciding what question the review will answer and which studies are allowed to be included.
🏥 Clinical Example
Researchers want to compare all available antidepressants for major depressive disorder. They define their question as: Which antidepressant is most effective for improving depressive symptoms in adults? They decide to include randomized controlled trials comparing any antidepressant with another antidepressant or placebo.
🧒 Explain it to a 10-year-old
Imagine you want to find out which video game is the best.
First, you decide which games will be included in the competition and what counts as winning.
🧠 Memory Hook
“Step 1: Ask the question and choose the players before starting the tournament.” 🏆
The steps of a network meta-analysis include:
1. Define the review question and inclusion criteria
2. Search and select studies
3. Perform titles/ abstract and full-test reading
4. Risk of bias assessment
5. Extraction of data, network building and statistical analysis
6. Synthesis of results
7. Interpretation of results and conclusions
Explain step 2.
Step 2 of a Network Meta-Analysis: Search and Select Studies
⭐ Key Idea (1 sentence)
Researchers must perform a broad and systematic search to find all relevant studies and select the ones that meet the inclusion criteria.
📚 Explanation (High-school level)
After deciding the research question and inclusion criteria in Step 1, the next step is to search for all the studies that could answer that question.
Researchers perform a systematic search of scientific databases (such as PubMed, Embase, or Cochrane Library) using carefully designed search terms. The goal is to make sure the search is broad enough to capture every relevant study so that important evidence is not missed.
Once all potential studies are found, researchers select the studies that meet the inclusion criteria defined in Step 1. This usually involves removing:
- studies that are not relevant
- studies that are not the right type (e.g., not randomized trials)
- duplicate studies
The key idea is that the search must be comprehensive and unbiased, ensuring all studies of interest are included.
In simple terms:
> Step 2 is about finding every relevant study and choosing the ones that fit the research question.
🏥 Clinical Example
Researchers conducting a network meta-analysis of antidepressants for major depressive disorder search multiple databases for randomized controlled trials comparing antidepressants or placebo. They identify hundreds of studies, then select only those that meet their criteria, such as trials involving adult patients with depression.
🧒 Explain it to a 10-year-old
Imagine you want to find the best pizza in your city.
First you look everywhere for pizza shops, then you pick the ones that actually sell pizza and are open.
🧠 Memory Hook
“Step 2: Search everywhere before choosing your contestants.” 🔎🍕
The steps of a network meta-analysis include:
1. Define the review question and inclusion criteria
2. Search and select studies
3. Perform titles/ abstract and full-test reading
4. Risk of bias assessment
5. Extraction of data, network building and statistical analysis
6. Synthesis of results
7. Interpretation of results and conclusions
Explain step 3.
Step 3 of a Network Meta-Analysis: Screen Titles/Abstracts and Read Full Text
⭐ Key Idea (1 sentence)
Researchers systematically review the titles, abstracts, and full articles of studies to confirm which ones truly meet the inclusion criteria and contain the necessary information.
📚 Explanation (High-school level)
After finding many potential studies in Step 2, researchers must carefully screen them to determine which ones should actually be included in the analysis.
This process happens in two stages:
1️⃣ Title and Abstract Screening
Researchers first read the title and abstract of each study. This allows them to quickly remove studies that are clearly irrelevant (for example, wrong population, wrong treatment, or wrong study design).
2️⃣ Full-Text Review
For studies that seem relevant, researchers then read the entire article to confirm whether it truly meets the inclusion criteria.
These steps must be done systematically and carefully because important information about effect modifiers may be found in the full paper. Effect modifiers are characteristics (such as patient age, disease severity, or treatment dose) that can influence treatment effects.
If studies differ too much in these factors, it may violate the transitivity assumption of network meta-analysis, which requires that studies be comparable.
In simple terms:
> Step 3 ensures that only appropriate and comparable studies are included in the network meta-analysis.
🏥 Clinical Example
Researchers studying antipsychotic treatments for schizophrenia initially identify 500 studies. After reading titles and abstracts, they remove studies involving children or observational designs. They then read the full text of the remaining studies to ensure they involve randomized trials comparing antipsychotics or placebo in adults with schizophrenia.
🧒 Explain it to a 10-year-old
Imagine you are choosing players for a soccer team.
First you quickly look at everyone who wants to join, then you check their skills carefully to see if they really qualify.
🧠 Memory Hook
“Step 3: Skim the cover, then read the book before letting it join the team.” 📚⚽
The steps of a network meta-analysis include:
1. Define the review question and inclusion criteria
2. Search and select studies
3. Perform titles/ abstract and full-test reading
4. Risk of bias assessment
5. Extraction of data, network building and statistical analysis
6. Synthesis of results
7. Interpretation of results and conclusions
Explain step 4.
Step 4 of a Network Meta-Analysis: Risk of Bias Assessment
⭐ Key Idea (1 sentence)
Researchers must evaluate the quality of each study and check for possible bias to ensure the results of the network meta-analysis are reliable and comparable.
📚 Explanation (High-school level)
After selecting the studies, researchers must examine how well each study was conducted. This step is called risk of bias assessment.
Bias refers to systematic errors in how a study was designed, conducted, or reported, which could lead to misleading results.
Researchers evaluate each trial using methodological quality criteria, such as:
- whether participants were randomly assigned to treatments
- whether blinding was used
- whether all participants were accounted for at the end of the study
- whether the study reported all outcomes honestly
This step is especially important in network meta-analysis, because poor-quality studies can distort the network results and threaten key assumptions such as similarity and consistency between studies.
In simple terms:
> Step 4 checks whether the studies are trustworthy enough to be included in the analysis.
🏥 Clinical Example
In a network meta-analysis comparing antidepressants, researchers assess whether each trial used proper randomization and blinding. If a study did not randomize patients properly or had many dropouts, it may be classified as high risk of bias and its findings interpreted cautiously.
🧒 Explain it to a 10-year-old
Imagine you are judging a race.
Before trusting the results, you check whether everyone followed the same rules and nobody cheated.
🧠 Memory Hook
“Step 4: Check the players for cheating before counting the score.” 🏁
The steps of a network meta-analysis include:
1. Define the review question and inclusion criteria
2. Search and select studies
3. Perform titles/ abstract and full-test reading
4. Risk of bias assessment
5. Extraction of data, network building and statistical analysis
6. Synthesis of results
7. Interpretation of results and conclusions
Explain step 5.
Step 5 of a Network Meta-Analysis: Extraction of Data, Network Building, and Statistical Analysis
⭐ Key Idea (1 sentence)
Researchers collect the important information from each study, build the treatment network, and run statistical analyses to compare all treatments in the network.
📚 Explanation (High-school level)
Once the relevant studies have been selected and assessed for bias, researchers move on to extracting the data from those studies.
1️⃣ Data Extraction
Researchers collect both quantitative data (numbers such as effect sizes, odds ratios, and sample sizes) and qualitative data (information about study design, patient characteristics, treatment doses, etc.).
2️⃣ Network Building
Using the treatments studied in the trials, researchers draw a network diagram.
Each treatment becomes a node, and lines between nodes represent direct comparisons from clinical trials.
Researchers then examine the network geometry, meaning they check how the treatments are connected and whether the network is suitable for analysis.
3️⃣ Statistical Analysis
Next, statistical models are used to compare treatments across the network. This includes:
- Performing pairwise meta-analyses for treatments directly compared in trials
- Running network meta-analysis models to combine direct and indirect evidence
- Checking for inconsistency between different sources of evidence
- Evaluating model fit and convergence (to ensure the statistical model works properly)
Researchers may also perform treatment ranking analyses to estimate which treatments are most effective overall.
In simple terms:
> Step 5 turns the collected studies into a network of treatments and uses statistical models to compare them.
🏥 Clinical Example
In a network meta-analysis comparing antipsychotic medications for schizophrenia, researchers extract data such as response rates and sample sizes from each trial. They then build a network connecting drugs like risperidone, olanzapine, quetiapine, and placebo, and perform statistical analyses to estimate which drug is most effective.
🧒 Explain it to a 10-year-old
Imagine you collect results from many soccer games.
You write down the scores, draw a chart showing which teams played each other, and then use the results to figure out which team is strongest overall.
🧠 Memory Hook
“Step 5: Gather the numbers, draw the network, and let the math decide the winner.” 📊🏆
The steps of a network meta-analysis include:
1. Define the review question and inclusion criteria
2. Search and select studies
3. Perform titles/ abstract and full-test reading
4. Risk of bias assessment
5. Extraction of data, network building and statistical analysis
6. Synthesis of results
7. Interpretation of results and conclusions
Explain step 6.
Step 6 of a Network Meta-Analysis: Synthesis of Results
⭐ Key Idea (1 sentence)
Researchers summarize and present the results of the analysis in clear formats such as tables, diagrams, and ranking graphs so the findings are easy to understand.
📚 Explanation (High-school level)
After performing the statistical analyses in Step 5, the next step is to combine and present the results in a clear and organized way. This step is called synthesis of results.
Researchers summarize the findings using visual and numerical tools such as:
- Tables – showing treatment comparisons and effect sizes (e.g., odds ratios or risk ratios).
- Network diagrams – illustrating how treatments are connected and compared.
- Rankograms or ranking plots – showing the probability that each treatment is the best, second best, third best, and so on.
These summaries help readers understand how the treatments compare overall, which ones appear most effective, and how confident we are in the results.
In simple terms:
> Step 6 takes the complicated statistical results and turns them into clear summaries that people can easily interpret.
🏥 Clinical Example
In a network meta-analysis comparing antidepressants, researchers present a table showing the odds ratios for each drug compared with placebo and with other antidepressants. They also include a rankogram showing the probability that each medication is the most effective treatment.
🧒 Explain it to a 10-year-old
Imagine you collected results from many races between different runners.
Now you make charts and tables showing who won the most races and who usually finishes near the top.
🧠 Memory Hook
“Step 6: Turn the math into pictures so that people can understand it.” 📊📈
The steps of a network meta-analysis include:
1. Define the review question and inclusion criteria
2. Search and select studies
3. Perform titles/ abstract and full-test reading
4. Risk of bias assessment
5. Extraction of data, network building and statistical analysis
6. Synthesis of results
7. Interpretation of results and conclusions
Explain step 7.
Step 7 of a Network Meta-Analysis: Interpretation of Results and Conclusions
⭐ Key Idea (1 sentence)
Researchers interpret what the results actually mean for patients and clinical practice, and evaluate how reliable the evidence is before drawing conclusions.
📚 Explanation (High-school level)
After summarizing the results in Step 6, researchers must interpret what those results mean in the real world. This means looking at the findings in the context of the disease and the available treatments.
For example, they consider:
- Which treatment appears most effective
- Whether the differences between treatments are clinically meaningful
- Whether the results are consistent and reliable
Researchers must also carefully interpret ranking figures (such as rankograms) because a treatment ranked first does not always mean it is dramatically better than others.
To assess how trustworthy the evidence is, researchers often use formal evaluation tools such as:
- GRADE (Grading of Recommendations, Assessment, Development and Evaluation) – used to judge the strength and quality of evidence.
- R-AMSTAR (Revised Assessment of Multiple Systematic Reviews) – used to evaluate the methodological quality of systematic reviews, including those that use network meta-analysis.
Using these tools helps determine how confident clinicians should be in the results and whether the findings should influence clinical decisions.
In simple terms:
> Step 7 turns the statistical results into real-world conclusions about which treatments work best and how reliable the evidence is.
🏥 Clinical Example
A network meta-analysis finds that Drug A ranks highest for treating major depressive disorder, but the differences between drugs are small and the evidence quality is rated moderate using GRADE. Researchers conclude that while Drug A may be slightly better, several antidepressants are similarly effective, so treatment choice may depend on side effects and patient preference.
🧒 Explain it to a 10-year-old
Imagine you finished a tournament between different teams.
Now you decide which team really played the best and whether the results were fair and trustworthy.
🧠 Memory Hook
“Step 7: Turn the numbers into real-world medical decisions.” 🩺📊
-
CAP 1: Intro to Evidence-Based Practice19
-
CAP 2: Intro to Data and Statistics18
-
CAP 2.2: Statistical Test Types (REVISE)23
-
CAP 2.3: Reliability & Validity (REVISE)21
-
CAP 2.4: P-values, Confidence Intervals3
-
CAP 2.5: Type 1 and 2 Errors, Power, Causation7
-
CAP 3: Bias19
-
CAP 4: Randomised Controlled Trials15
-
CAP 5: Case Control Studies15
-
CAP 6: Cohort Studies26
-
CAP 7: Diagnostic Studies25
-
CAP 8: Systematic Reviews & Meta-Analyses36
-
CAP 9: Economic Analyses17
-
CAP 10: Qualitative Research25
-
CAP 11: Network Meta-Analysis24
