Reading a Diagnostic Test Without Fear

Every diagnostic test is really a way of changing probability.

When we perform a special test, we are not simply asking whether it is “positive” or “negative.” We are asking something more useful:

How much should this result change what I already believe about the patient in front of me?

Sensitivity, specificity, likelihood ratios and confidence intervals help answer exactly that question. The Ottawa Ankle Rules are a useful example because they show how all of these concepts work in a real clinical decision: does this patient with an acute ankle injury need an X-ray?

Start with the Ottawa Ankle Rules

The Ottawa Ankle Rules were designed to identify patients who may have an ankle or midfoot fracture after an acute injury.

An ankle X-ray is indicated when there is pain in the malleolar zone plus tenderness over the posterior distal malleolus or an inability to take four steps. A foot X-ray follows the same logic using tenderness at the navicular or base of the fifth metatarsal.

The important point is that the rule was designed to miss almost no fractures, even if that means sending some patients without fractures for imaging. That trade-off is the key to understanding sensitivity and specificity.

Think of airport security

Imagine a metal detector at an airport. A very sensitive detector is designed to catch almost every weapon. Unfortunately, it will also beep for keys, coins and belts. That is similar to the Ottawa Ankle Rules.
In the teaching example:

• Sensitivity ≈ 98%

• Specificity ≈ 35%

Of 150 patients who actually had a fracture, 147 were identified by the rule. Only three fractures were missed. But among 850 people without a fracture, more than half still tested positive and would have been sent for an X-ray.

So a negative Ottawa result is powerful.
A positive result is much less impressive.
This is the logic behind SnNout:

Sensitive test + Negative result = useful for ruling OUT.
The opposite principle is SpPin:
Specific test + Positive result = useful for ruling IN.

But these are only shortcuts. Likelihood ratios give us a more complete picture.

Likelihood ratios: how far should your probability move?

A likelihood ratio combines sensitivity and specificity into one number.
Instead of asking whether the test is simply “good” or “bad,” it asks:

How much should this result move my probability of the diagnosis?

An LR of 1 changes nothing.
As a rough guide:

For the Ottawa Ankle Rules, the positive likelihood ratio is only about 1.5.

That means a positive result barely increases the probability of fracture.
The negative likelihood ratio is about 0.06, however, which produces a much larger reduction in probability.
So the Ottawa Rules are not really trying to diagnose a fracture.
They are mainly helping us decide who probably does not need imaging.

Pre-test probability comes first

Before performing any test, we already have some idea of how likely the diagnosis is.
That is your pre-test probability.

It comes from the setting, the patient's history, mechanism, age, symptoms, ability to walk and your clinical impression.
A patient walking into a private physio clinic three days after a mild ankle inversion may have a much lower fracture probability than an older patient arriving in an emergency department after a high-energy injury.
This matters because the same test result can mean different things in different patients.

Same test.
Same sensitivity.
Same specificity.
But a very different clinical meaning depending on where you started.

Predictive values can be misleading

Patients naturally ask:

“My test is positive. What is the chance I actually have the condition?”

That is what the positive predictive value, or PPV, describes.
In the emergency-department example, a positive Ottawa result gives only about a 21% probability of fracture, while a negative result gives about a 99% probability of being fracture-free.
But PPV and NPV change with prevalence.

Move the same test from an emergency department into a lower-risk physiotherapy clinic and the meaning of a positive result changes substantially.
This is why predictive values should not simply be copied from one study and applied to every clinical population.

Confidence intervals matter too

A diagnostic paper may report an impressive sensitivity of 100%.
But how many patients was that based on?

A study detecting 15 fractures out of 15 can report 100% sensitivity, yet its confidence interval remains wide. A much larger dataset may report 98–99% sensitivity with a far narrower interval—and therefore much greater certainty.

A good rule when reading diagnostic research is:

Never read the estimate without reading its confidence interval.

A narrow interval suggests greater precision.
A wide interval tells you that the true performance could be substantially different from the headline number.

Tests do not live outside context

Sensitivity and specificity are often taught as fixed characteristics.
In reality, they may vary according to the population, setting, disease severity, examiner experience and study design.
Spectrum bias, verification bias, incorporation bias and loss to follow-up can all make a diagnostic test appear better—or worse—than it really is.
The practical question becomes:

Was this test studied in patients who look like mine?

A test studied in high-energy emergency trauma may not perform identically in a private sports clinic.

The clinical takeaway

When reading a diagnostic study, do not stop at:

“Sensitivity 90%, specificity 85%.”

Instead ask:

What is my pre-test probability?
Is this test mainly useful for ruling out or ruling in?
What are the LR+ and LR−?
How much does the result actually change probability?
How wide are the confidence intervals?
Was the test studied in patients similar to mine?

That is the real purpose of diagnostic statistics.
Not memorising formulas.
Not collecting special tests.
But understanding how much evidence a test result actually adds to your clinical reasoning.