Uni Reading List
Study guides /

Medical evidence: choose a reading route and audit a diagnostic claim

A medical result is not self-interpreting. Choose a book for the question you need to answer: why fair comparisons matter, how to appraise a paper, or how populations and study designs define the numbers. Then produce an argument someone else can check, not a list of methodological labels.

Course topic: Research Articles

Study route and exercise: University Reading List editorial project. How we curate · Sources checked:

Download blank worksheet (Markdown) ↓

Scope and responsibility

This is an educational route for reading research, not a clinical protocol or personal test/treatment recommendation. The exercises use wholly invented conditions, tests and counts; their stipulated performance is not evidence about an actual product or population. The books are specific historical editions, and their disease examples are not current care guidance. No clinical qualification, teaching trial or complete-book review is claimed.

Choose your starting point

I need a free conceptual start

Use Testing Treatments for the logic of fair comparisons and the limits of impressive claims. Its short narrative is a first route, not a diagnostic statistics manual. Keep WHO nearby only when a denominator or design needs a fuller explanation.

I have a paper and need an appraisal route

Use How to Read a Paper’s selected design-specific chapter after identifying the question and study type. Read methods before deciding what a result means; a checklist is a prompt to inspect evidence, not a score that certifies a paper.

I need population measurement and design

Use the WHO text’s measurement, design and clinical-epidemiology sections to understand who was counted and compared. Its 2006 examples are dated. Add the appraisal book when you need a more detailed paper-specific interrogation.

Compare the three books

Testing Treatments: Better Research for Better Healthcare

Imogen Evans; Hazel Thornton; Iain Chalmers; Paul Glasziou · Edition: 2 · 2011

Free explanation of why fair tests matter

Choose it when
You are new to evidence reading and want to challenge claims based on novelty, authority or a memorable story.
Use another route when
You need a detailed quantitative methods course, a current disease guideline or a full diagnostic-appraisal manual.
Selected reading
2011 second edition: chapters 6–8 for comparison, chance and evidence synthesis; chapter 4 for the distinction between a test and a beneficial screening programme.
Limits and trade-offs
The authors provide authorized full-text access. Historical stories make a methodological point but do not establish present-day treatment advice; the book is not an exercise bank with answers to every proposed activity.

Try before choosing

From chapter 6, draw two rival explanations for an invented apparent improvement: an intervention effect and a difference between the compared groups. Specify one feature that could separate them.

Check your work: Name a comparison and a source of bias, not just “use an RCT”. If you can do that but need calculations and diagnostic designs, add the relevant methods chapter rather than a second narrative primer.

Official free reading ↗

Book record and access options →

How to Read a Paper: The Basics of Evidence-Based Healthcare

Trisha M. Greenhalgh; Paul Dijkstra · Edition: 7 · 2024

Design-specific critical-appraisal guide

Choose it when
You can name the study question and want a structured route through methods, results and interpretation.
Use another route when
You expect its 2024 edition to supply current care recommendations, unrestricted online full text or a substitute for specialist statistical training.
Selected reading
Seventh edition: chapters 1, 3–5, 8 and 9; use chapter 8 for diagnostic/screening papers and chapter 9 for synthesis.
Limits and trade-offs
Wiley identifies the selected paperback as a December 2024 release; online publication and ©2025 are separate dates. The print ISBN is not an online-access entitlement. Work from the actual study, not a checklist detached from its methods.

Try before choosing

Use an available chapter-8 sample or a legitimate copy to make an index-test/reference-standard flow diagram. Leave recruitment, blinding and verification details blank if the supplied paper does not report them.

Check your work: A blank is an unresolved reporting question, not proof the method was absent. The output should explain why each missing detail changes what you can assess; locating a named checklist alone is insufficient.

Book record and access options →

Basic Epidemiology

Ruth Bonita; Robert Beaglehole; Tord Kjellström · Edition: 2 · 2006

Free population-methods bridge

Choose it when
You want definitions of populations, disease frequency and design before interpreting a research table.
Use another route when
You mainly want a quick conversational introduction or current health-policy and screening rules.
Selected reading
2006 second edition: chapter 2 on measurement, chapter 3 on designs/errors, and chapter 8 on clinical epidemiology and diagnostic tests.
Limits and trade-offs
The rights-holder repository is the free reading entry, not a licence for unrestricted reuse. The book supplies methods and study questions, while its historical population figures and policy context must remain dated.

Try before choosing

Use chapters 2–3 to label a fictional table’s target population, numerator, denominator, time window and selection mechanism. Decide whether it describes prevalence, incidence or neither.

Check your work: A one-time condition count with a defined population can describe prevalence; new events over time require a different denominator/design. If these distinctions are unclear, begin here before interpreting a positive-test percentage.

Official free reading ↗

Book record and access options →

Check your preparation

What is the denominator of sensitivity?

People with the reference condition: TP/(TP+FN). Positive predictive value uses test-positive people instead: TP/(TP+FP). Reversing these conditions changes the question.

Does an accurate test automatically make a useful screening programme?

No. Discrimination is one component. A programme-level outcome question also involves what happens after testing, benefits, harms and the comparison. This guide’s table cannot establish those outcomes.

What does an unreported methods detail establish?

That the report does not tell you enough about that detail. Do not automatically code it as well conducted or definitely biased; identify what source or clarification would resolve the uncertainty.

Six steps with a concrete output

These are your own study marks, not an assessment. They are not sent to this site.

  1. Define the question before judging the design

    Selected reading: Testing Treatments chapters 4 and 6; How to Read a Paper chapter 1.

    Make this: Write two different questions: whether an invented index test identifies a reference condition, and whether offering testing improves a specified outcome. List the population and comparison for each.

    Check your work: The first is a discrimination question; the second needs an outcome comparison. A sensitivity number cannot answer the second question by itself.

  2. Map who entered and who was measured

    Selected reading: WHO chapter 3; How to Read a Paper chapters 3–4.

    Make this: Draw recruitment → index test → reference assessment → analysis for a freely readable example or the fictional packet below. Mark exclusions and unknown information explicitly.

    Check your work: Show whether all participants receive the reference assessment and whether the analyzed group matches recruitment. Unknown fields stay unknown rather than becoming invented study facts.

  3. Make a bias-and-applicability ledger

    Selected reading: How to Read a Paper chapters 4 and 8; WHO chapter 3.

    Make this: For selection, reference assessment, blinding and the tested population, write the reported fact, the concern it raises and one reason it might matter. Keep applicability separate from internal validity.

    Check your work: A concern is not proof that bias occurred. Distinguish “the sample differs from the target” from “the study fails to measure its own sample consistently”.

  4. Build and audit the two-by-two table

    Selected reading: WHO chapter 8; How to Read a Paper chapter 8; the CDC definitions.

    Make this: Solve the original example with condition-status columns and test-result rows. Compute sensitivity, specificity and the positive-test condition proportion with named denominators.

    Check your work: Every person appears once, each column and row sums correctly, and TP+FN is not confused with TP+FP. Write the statistic’s question alongside its value.

  5. Test transport and uncertainty assumptions

    Selected reading: Testing Treatments chapter 7; How to Read a Paper chapters 5 and 8.

    Make this: Change only prevalence in the paper model, then list three reasons actual conditional performance might fail to transfer. State what sampling/measurement information a real uncertainty analysis would need.

    Check your work: Holding sensitivity and specificity fixed is an assumption, not a result about a new population. The deterministic exercise does not create a confidence interval or establish clinical utility.

  6. Write a bounded appraisal memo

    Selected reading: Testing Treatments chapter 8; How to Read a Paper chapter 9; WHO chapter 11.

    Make this: Write a short memo with your selected main book, a precise question, a source/section ledger, one checked calculation, one unresolved methods issue and one next research question.

    Check your work: The memo distinguishes source facts, invented inputs and your interpretation. Its next step concerns evidence to obtain, not a personal clinical action or an invented recommendation.

Worked exercise

Original example: the positive result changes with the base rate

Two invented cohorts each contain 10,000 people. In cohort L, 100 have condition Q; in cohort H, 1,000 do. A fictional test is stipulated to have 90% sensitivity and 98% specificity in each cohort. Build both complete tables and compare the proportion with Q among positive tests.

Assumptions

Treat reference status as known without error for everyone, with no indeterminate results or losses. Keep the same test rule and deliberately hold both conditional rates fixed. All counts and condition Q are invented; the transfer assumption is for arithmetic, not a real validation finding.

  1. Cohort L: of 100 with Q, 90 test positive and 10 negative. Of 9,900 without Q, 2% = 198 test positive and 9,702 negative. Positive tests total 288; negative tests total 9,712.
  2. Cohort H: of 1,000 with Q, 900 test positive and 100 negative. Of 9,000 without Q, 180 test positive and 8,820 negative. Positive tests total 1,080; negative tests total 8,920.
  3. Sensitivity is 90/100 = 900/1,000 = 90%; specificity is 9,702/9,900 = 8,820/9,000 = 98%. The positive predictive values are 90/288 = 31.25% and 900/1,080 ≈ 83.33%.
  4. At lower prevalence, the larger condition-absent group can contribute many false positives despite high specificity. The calculations do not show whether testing improves outcomes or whether either table applies to a particular person.

Answer and interpretation

With the stipulated equal performance, cohort L has 90 true positives among 288 positives (31.25%); cohort H has 900 among 1,080 (about 83.33%). A test-positive condition proportion is not sensitivity. Different base rates can change it without any arithmetic change in the two stipulated conditional rates.

Test the result

For each cohort, recover 10,000 from both row and column totals. Then explain why applying the second cohort’s percentage to a new population would require more than an identical test name.

Change an assumption

Follow-up: verifying positives only leaves sensitivity unidentified

Change to a new invented study and discard the earlier known prevalence and performance assumptions. Retain a total of 10,000, an error-free reference when performed and no indeterminate index results. There are 288 positives, all verified: 90 have Q and 198 do not. None of the 9,712 negatives receives the reference assessment. Which proportion is observed, and can sensitivity be determined?

  1. Among all test positives, Q is verified in 90/288 = 31.25%. The unverified negatives cannot be silently assigned to the condition-absent column.
  2. One compatible completion has no Q among negatives: FN = 0 and TN = 9,712, giving sensitivity 90/(90+0) = 100%. Another has 90 Q among negatives: FN = 90 and TN = 9,622, giving sensitivity 90/(90+90) = 50%. Both reproduce every supplied observation.

Answer and interpretation

The verified positive-test condition proportion is 31.25% for this study. Sensitivity, specificity and prevalence for all entrants are not identified by the observed data alone. The earlier exercise’s stipulated rates cannot be imported as findings of this new study.

Test the result

Check 90+198+9,712 = 10,000 and both completions’ totals. The difference is missing reference verification tied to the index result, not merely a small sample. More positive-only verification does not supply the missing negative reference statuses.

Common mistakes

Treating accuracy as programme benefit

A diagnostic table describes classifications under a reference definition. A claim about improved health outcomes requires a different question, suitable comparison and harms as well as benefits.

Borrowing performance without its population

A quoted sensitivity/specificity pair does not establish equal performance across spectrum, threshold, setting or reference procedures. Identify these conditions before transferring a result.

Turning absence of reporting into a confident verdict

Record the missing detail and its consequence. Neither an unanswered checklist item nor a prestigious source alone establishes the paper’s actual conduct.

Before moving on

  • I can state the study question and population before choosing an appraisal chapter.
  • I can reconstruct a table, name its denominators and separate stipulated model assumptions from empirical findings.
  • I can explain an unresolved bias/applicability question and what evidence could resolve it, without turning the memo into care advice.

Sources and edition checks

Sources checked:

These are proposed study activities, not official course assignments or a claim that every book has been read cover to cover.

How we curate →