A contingency table is a grid that compares test results (positive/negative) against the true disease status(present/absent).
It organizes outcomes into true positives, false positives, false negatives, and true negatives, allowing calculation of sensitivity, specificity, predictive values, and likelihood ratios.

Also known as:

• Diagnostic 2×2 Table

• Diagnostic Test Performance Table

• Confusion Matrix

• Sensitivity – How often the test correctly identifies people with the condition
  (True positive rate)

• Specificity – How often the test correctly identifies people without the condition
  (True negative rate)

• Positive Likelihood Ratio (LR+) – How much more likely a positive test is in someone with the condition than without it
  (Higher values strengthen “rule-in” confidence)

• Negative Likelihood Ratio (LR−) – How much less likely a negative test is in someone with the condition than without it
  (Lower values strengthen “rule-out” confidence)

• Pre-test probability (prevalence) – Baseline probability of disease before testing

• Post-test probability – Probability of disease after accounting for test results
  (Derived using likelihood ratios)

Using PICO to search for evidence

PICO can also be used in reverse—to formulate a focused clinical question that guides literature searching. Clearly defining each PICO element improves search precision, reduces irrelevant results, and helps identify studies that directly address a clinical uncertainty.

What it is:
The specific group of patients or clinical condition being studied.

Why it matters:
Determines whether the study applies to your patient or population.

Where to find it in an article:

• Title

• Abstract

• Methods → Participants / Study Population

• Inclusion / Exclusion Criteria

Common keywords/phrases:

• “patients with…”

• “adults aged…”

• “diagnosed with…”

• “inclusion criteria”

• “baseline characteristics”

What it is:
The treatment, test, exposure, or action being evaluated.

Why it matters:
Defines what is being done differently from usual care or placebo.

Where to find it in an article:

• Abstract

• Methods → Intervention / Treatment Protocol

• Study Design section

Common keywords/phrases:

• “treated with…”

• “received…”

• “intervention group”

• “administered”

• “exposure”

A Fagan nomogram is a visual tool that uses Bayes’ theorem to convert a pre-test probability and a likelihood ratiointo a post-test probability.
It shows how diagnostic test results change the estimated probability of disease, helping clinicians translate test performance into clinical decision-making.

• Absolute Risk Reduction (ARR) – Difference in outcome risk between treatment and control groups

• Absolute Risk Increase (ARI) – Increase in outcome risk caused by an intervention

• Number Needed to Treat (NNT) – Number of patients who need treatment to prevent one adverse outcome

• Number Needed to Harm (NNH) – Number of patients who need treatment for one to experience harm

Accuracy

• Accuracy – Overall proportion of correct test results (true positives + true negatives)
  (Can be misleading when prevalence is very high or very low)

Let’s use the JUPITER statin article to practice using the contingency and nomogram tools.

STEP 1 — Define the clinical question

Population: Apparently healthy adults with normal LDL but elevated hs-CRP
Intervention: Rosuvastatin 20 mg daily
Comparison: Placebo
Outcome: Major cardiovascular events

STEP 2 — Extract absolute event counts (Results section)

From the NEJM paper:

• Rosuvastatin group: 142 events / 8901

• Placebo group: 251 events / 8901

These raw counts matter more than hazard ratios for this exercise.

STEP 3 — Populate the contingency table

Treat statin use as the “test” and cardiovascular event as the outcome.

A: Statin Event (+) n=142

B: Statin Event (-) n=8759

C: Placebo Event (+) n=251

D: Placebo Event (-) n=8650

This allows calculation of:

• Absolute risk reduction

• Relative risk reduction

• Number needed to treat (NNT)

STEP 4 — Output

From these inputs, your calculator will show:

• ARR ≈ 1.2% over ~2 years

• NNT ≈ 95 over ~2 years

• Much larger relative reductions than absolute reductions

STEP 5 — Use the Fagan nomogram conceptually

Although JUPITER is a treatment trial (not a diagnostic test), the nomogram analogy helps interpretation:

• Pre-test probability: Low baseline event risk (healthy population)

• Likelihood of benefit: Statin modestly shifts probability

• Post-test probability: Still low absolute event risk

Visually, even a strong “line” does not cross into large absolute benefit territory.

INTERPRETATION:

1. Relative risk reduction exaggerates perceived benefit

• 44% relative risk reduction sounds dramatic

• Absolute risk reduction was ~0.6% per year

2. NNT is large in low-risk populations

• ~169 treated for 1 year to prevent one composite event

• ~500 treated for 1 year to prevent one MI

• This is easy to miss without a 2×2 table

3. Early trial termination may overestimate benefit

• Trial stopped at 1.9 years instead of planned 4

• Long-term harms and durability of benefit remain uncertain

4. Harms matter when baseline risk is low

• Increased incidence of diabetes observed

• When NNT is high, even small harms can shift net benefit

Using PICO to read a journal article

The PICO framework provides a structured way to extract what matters most from a journal article. By identifying the patient population, intervention, comparison, and outcomes, readers can quickly determine whether a study is relevant, how it was designed, and whether its findings apply to their clinical context.

What it is:
The alternative to the intervention (placebo, standard care, another treatment, or no intervention).

Why it matters:
Provides the reference point needed to interpret effect size and clinical relevance.

Where to find it in an article:

• Abstract

• Methods → Control / Comparator

• Randomization section

Common keywords/phrases:

• “compared with…”

• “control group”

• “placebo”

• “standard of care”

• “usual care”

Note: Some studies (especially observational ones) may not have a formal comparator.

What it is:
The clinical result(s) measured to determine effectiveness or harm.

Why it matters:
Ensures the study measures outcomes that are meaningful for patients or decision-making.

Where to find it in an article:

• Abstract

• Methods → Outcomes

• Results

• Tables and Figures

Common keywords/phrases:

• “primary outcome”

• “secondary outcome”

• “measured by…”

• “incidence of…”

• “mortality, symptom score, risk reduction”

Big picture: what this worksheet is asking

This worksheet helps you answer two core questions:

1. Are the study methods valid?

2. If valid, are the results useful for clinical decision-making?

You are not being asked to judge whether the test is “good” in absolute terms — only whether the study design allows the results to be trusted and interpreted.

This worksheet helps you decide whether a diagnostic accuracy study is trustworthy and clinically meaningful. The first section focuses on study validity—whether the right patients were studied, whether a proper reference standard was used, and whether bias was minimized. The second section focuses on interpreting the results using 2×2 tables, sensitivity, specificity, and likelihood ratios to understand how test results change disease probability.

General rule

👉 Almost all answers come from the Methods
If you can’t find it in Methods → often “Unclear” is the correct answer.

R — Representative spectrum of patients

(Page 1 of worksheet)

What this is really asking

“Do the patients in the study look like the patients you’d actually use this test on?”

Where to look in the article

• Methods → Participants / Study population

• Inclusion / exclusion criteria

• Setting (ED, outpatient, ICU, screening population)

KEYWORDS:

• “consecutive patients”

• “random sample”

• “suspected [condition]”

• “referred for evaluation”

• “inclusion criteria”

• “exclusion criteria”

How to decide

• Yes → broad, real-world population; consecutive or random enrollment

• No → highly selected, extreme cases only, obvious disease vs obvious non-disease

• Unclear → patient selection not described

📌 Tip: If the study only includes very sick patients or clear positives, sensitivity/specificity may be inflated.

A — Reference standard applied regardless of index test

(Page 1 of worksheet)

What this is really asking

“Did everyone get the gold standard diagnosis — or only people with certain test results?”

Where to look

• Methods → Reference standard

• Flow diagram (if present)

Keywords

• “all patients underwent…”

• “reference standard performed on…”

• “follow-up used as reference”

• “verification bias”

How to decide

• Yes → everyone had the same reference standard (or adequate follow-up)

• No → only test-positive patients got the reference standard

• Unclear → not stated

📌 Why this matters: If negatives aren’t verified, false negatives can be missed.

Mbo — Masked (blinded) comparison

(Page 1 of worksheet)

What this is really asking

“Were the test results interpreted independently?”

Where to look

• Methods → Blinding

• Radiology / pathology interpretation description

Keywords

• “blinded”

• “masked”

• “interpreted independently”

• “read without knowledge of…”

How to decide

• Yes → index test and reference standard read without knowledge of each other

• No → interpreters knew the other result

• Unclear → blinding not mentioned

📌 Tip: If blinding isn’t stated, “Unclear” is acceptable.

Replication question (Page 3)

“Are methods described in enough detail to repeat the test?”

Look for

• Test protocol details

• Cutoffs / thresholds

• Timing of tests

If the test is vaguely described → No or Unclear

Are test characteristics presented?

(Page 2 of worksheet)

What this is asking

“Can I build or see a 2×2 table?”

LOOK FOR:

• Sensitivity / specificity

• Likelihood ratios

• Predictive values

• Raw numbers (TP, FP, FN, TN)

If raw counts aren’t given but can be derived → that still counts.

How to approach the 2×2 table (don’t panic)

The worksheet walks you through a worked example using dementia prevalence and explicitly shows how to complete the table step-by-step Diagnostic Article Critical App….

KEY MINDSET

You are not expected to memorize formulas.
You are expected to:

• understand what the numbers represent

• know how they relate to clinical decisions

Interpreting likelihood ratios (clinically)

From the worksheet language Diagnostic Article Critical App…:

• LR+ > 1 → increases probability of disease (rule-in)

• LR− < 1 → decreases probability of disease (rule-out)

Rough intuition:

• LR+ ≥ 10 → strong rule-in

• LR− ≤ 0.1 → strong rule-out

How to answer “Yes / No / Unclear” without stress

Use this rule:

• Yes → explicitly stated

• No → explicitly violated

• Unclear → not described

“Unclear” is not a wrong answer — it often reflects poor reporting.