Reading the whole story at a glance

One picture, the whole analysis

A forest plot looks busy, but it's built from just a handful of elements, and none of them is decoration. Every square, every line, every position is carrying a specific piece of information.

Here's the whole vocabulary, in one sentence: each row is a study, drawn as a square at its effect (left or right) with whiskers showing its confidence interval; the square's size is its weight; and at the bottom, a diamond shows the pooled result — its position the combined effect, its width the combined uncertainty. A vertical line marks "no effect."

That's it. Six things. Learn to see each one and you can read any forest plot ever printed. So let's take one apart — literally.

Take it apart

Below is a forest plot from a real-style meta-analysis: five trials of a treatment, measured as a risk ratio — a value below 1 means the treatment reduces risk. Tap each part to find out what it's telling you — work through all five.

Risk ratio scale. Null line (no effect) at RR = 1. Squares: study effect and weight. Whiskers: 95% CI. Diamond: pooled result.

All five parts explored — you can continue.

The trap: size is not effect

One misreading catches almost everyone, so let's kill it now.

A bigger square does not mean a bigger effect. The square's size is its weight — precision, driven by sample size. The effect is the square's horizontal position, not its area. These are two completely separate dimensions of the same square: where it sits (effect) and how big it is (weight).

Look at the plot you just took apart: Study A has the biggest square, but Study C sits furthest from the null line. A is the heaviest, but C found the biggest effect. Size and position are telling you two different things about two different studies — read them separately.

So the most misread pattern is a large square sitting right on the null line: a big, precise, heavily-weighted study showing no effect at all. Position is what. Size is how much it counts. Keep them apart.

Where's "no effect"?

One more thing to check before you trust your reading of any forest plot: where is the null line? Because it isn't always at zero.

It depends on what the effect is measured as — and you know both kinds from Module 3:

• For a difference (a risk difference, or a mean difference), "no effect" is a difference of 0. The null line sits at 0.
• For a ratio (a relative risk, odds ratio, or hazard ratio), "no effect" is a ratio of 1 — the two arms are equal. The null line sits at 1, and the axis is usually on a log scale (which is why this plot's axis is spaced the way it is).

Get this wrong and you'll misread every significance call on the chart. A diamond sitting at 0.73 on a risk-ratio plot is a 27% reduction and — since its whole interval stays left of 1 — significant; but only if you know the line to beat is 1, not 0. Always find the null line first, then read everything relative to it.

Now read one

Using this plot (risk ratio, null line at 1), answer four questions by reading the chart.

The scatter tells a story

That last question was bigger than it looked. How spread out the squares are is heterogeneity, read by eye.

When the squares cluster tightly around a shared value — like the plot you just read — the trials broadly agree, and the fixed-effect world from the last lesson looks credible. When they're flung across the axis, some far left, some far right, the trials genuinely disagree — you're likely in the random-effects world, and a single pooled number hides real conflict.

So before you even reach for a statistic, a forest plot tells you which world you're in — just from the shape of the scatter. You can feel the heterogeneity before you measure it. And measuring it — turning that visual scatter into a single number, I², so the judgement isn't just eyeballing — is exactly where the next lesson goes.

Why this matters for HTA

Every meta-analysis in every dossier arrives as a forest plot. Reading it fluently — and sceptically — is a core appraisal skill, and most of what you need is on the chart if you know where to look.

• Read weight before you read effect. Find the largest squares first — they're driving the pooled result. If one study dominates the weight, the meta-analysis is largely that study; check whether it's one you trust (and whose it is).
• Read the diamond against the right null line. Confirm whether the scale is a difference (null 0) or a ratio (null 1), then check whether the diamond crosses it. A diamond that just barely clears the line is a fragile "significant" — one that could vanish under an honest random-effects model or with one study removed.
• Read the scatter for hidden disagreement. Widely spread squares under a confident-looking narrow diamond is a red flag: it usually means a fixed-effect model was used on heterogeneous trials, manufacturing false precision. The picture shows you the conflict the summary number hides.

"The forest plot doesn't just report the answer — it shows you how much to trust it. Learn to read the whole picture, and the summary line can never fool you."

Reading a forest plot, in one breath

• A forest plot compresses an entire meta-analysis into one picture where every mark means something — nothing is decoration.
• Each row is a study: its position is the effect, its square size is the weight (precision), its whiskers are its confidence interval.
• Size is weight, not effect — the most common misreading. A big square near the null line is a precise study finding nothing.
• The diamond is the pooled result: position = combined effect, width = combined uncertainty. Whether it crosses the null line (0 for differences, 1 for ratios) decides significance.
• The scatter of the squares is heterogeneity by eye — tight means agreement, spread means real disagreement between trials.

"Fluent readers don't study a forest plot — they glance, and the whole story falls out: who counts, whether they agree, and how far to trust the answer."

That completes the picture this module has been building. You can find the studies, count them, search for them, judge each one's risk of bias, combine them by weight, choose the right model, and now read the result at a glance. One thing still hangs over the forest plot, though: that scatter. We've been eyeballing whether studies "agree" — but eyeballing isn't a method. The next lesson makes it one, with a number for heterogeneity (I²) and a way to spot when the studies you've pooled shouldn't have been pooled at all.