Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data Analytics & AI · EBP in assessment and diagnosis

Evidence-Based Practice in Assessment and Diagnosis: Likelihood Ratios and Probability Revision

Sensitivity and specificity are properties of a test, whereas the clinical question concerns an individual patient: given this result, how probable is the diagnosis? Likelihood ratios address that question directly, and they demonstrate that a test with poor sensitivity may nonetheless be decisive; a finding with a sensitivity of only 28.4% but a specificity of 98.6% carries a positive likelihood ratio of 20.4.

Pubrica AcademyWritten by the medical writing and biostatistics teams Reviewed byPubrica biostatistics team Published9 October 2026 Last checked9 October 2026 Reading time11 minutes Sources34 peer-reviewed

Most teaching in evidence-based practice is organised around treatment effects, whereas most clinical reasoning is not. Assessment and diagnosis depend on a different arithmetic: a prior probability, a finding and a revised probability. Deeks and Altman’s account of likelihood ratios in the BMJ provides the practical foundation, and its importance lies in the fact that sensitivity and specificity address the question in reverse, describing test behaviour in patients whose diagnosis is already known. This article describes how to proceed from a test accuracy study to a decision about an individual patient.

Learning objectives

  • The reasons sensitivity and specificity do not answer the clinician’s question, and the measures that do
  • The calculation and use of positive and negative likelihood ratios
  • The thresholds that constitute strong evidence for ruling a diagnosis in or out
  • Conversion from a pre-test to a post-test probability, with a worked example
  • The definition of a clinical prediction rule, and methods for identifying one relevant to a clinical question
  • Appraisal of a diagnostic accuracy study before its results are applied

Common misconceptions

  • “A test with low sensitivity is not much use.” In Deeks and Altman’s worked example, a finding with a sensitivity of only 28.4% had a specificity of 98.6%, giving a positive likelihood ratio of 20.4; when present, the finding is close to decisive.
  • “A positive result means the patient has the condition.” It indicates that the probability has been revised upward from its initial value. From a pre-test probability of 0.1, the example finding raises the post-test probability to about 0.69, a substantial shift that nonetheless leaves approximately a one-in-three probability of no disease.
  • “Published accuracy figures transfer to my clinic.” Accuracy depends on the spectrum of patients tested. QUADAS-2 exists because patient selection is one of the four domains in which a diagnostic accuracy study may be misleading.
  • “A clinical prediction rule is just a checklist someone wrote.” The definition adopted by Ingui and Rogers requires a rule to combine three or more variables from history, physical examination or simple tests.
  • “A p-value tells me whether the test works.” Greenland and colleagues, in the European Journal of Epidemiology, catalogue the common misinterpretations; in diagnosis, the quantity required is a probability concerning the individual patient, not a test of a hypothesis about a population.

The limitations of sensitivity and specificity for clinical decisions

Sensitivity is the proportion of people with the condition who test positive. Specificity is the proportion of people without the condition who test negative. Both are conditioned on knowledge of the diagnosis: they are calculated by starting from the right-hand column of the 2x2 table and working back to the test result. The clinical question proceeds in the opposite direction: the test result is known, and the diagnosis is what must be determined.

This distinction is not merely pedantic, because the two directions diverge sharply when the condition is uncommon. A test with high sensitivity and high specificity applied in a low-prevalence setting will still generate more false positives than true positives, and no property of the test alters this; it is a consequence of the population 13. This explains why the same test performs differently in a screening clinic and a referral centre, and why accuracy figures from a tertiary case series should not be applied in primary care without examination.

What is required is a quantity that can be applied to a starting probability to produce a revised one. The likelihood ratio is such a quantity.

The clinical question and the statistic that answers it

What you want to knowThe statistic that answers itWhy the obvious choice fails
If the patient has the condition, will the test detect it?SensitivityNo failure arises here; this is the question for which sensitivity was designed, although it is seldom the clinical question
Given this positive result, how likely is the condition now?Pre-test probability combined with the positive likelihood ratioSensitivity is conditioned on already knowing the diagnosis, so it cannot be applied to a result
Given this negative result, can I stop considering the condition?Pre-test probability combined with the negative likelihood ratioHigh sensitivity alone does not rule out; the negative likelihood ratio may sit close to 1
How much does this finding change my thinking at all?Distance of the likelihood ratio from 1; above 10 or below 0.1 is strong evidenceAccuracy and the area under the curve summarise across thresholds and populations, not this result in this patient
Do these findings together justify a diagnosis?A validated clinical prediction rule combining three or more variablesMultiplying likelihood ratios for correlated findings overstates the revision

Likelihood ratio definitions and the magnitude thresholds of 10 and 0.1 from Deeks JJ and Altman DG, BMJ 2004. The three-variable definition of a clinical prediction rule is Laupacis’, as adopted by Ingui BJ and Rogers MAM, Journal of the American Medical Informatics Association 2001.

Likelihood ratios and thresholds for strong evidence

Deeks and Altman, in the BMJ, set out the two ratios plainly. The positive likelihood ratio expresses how much more likely a positive result is in a person with the condition than in a person without it; the negative likelihood ratio expresses the same for a negative result 13. A ratio of 1 indicates that the test result changes nothing. The further the ratio lies from 1, the more the result alters the probability.

Their guidance on magnitude is particularly important: likelihood ratios above 10 or below 0.1 provide strong evidence to rule a diagnosis in or out, respectively 8,13. Between those values, the result shifts the probability by an amount that may or may not cross a decision threshold, depending entirely on the starting probability.

Their worked example illustrates the asymmetry. A history of more than 40 pack-years had a sensitivity of 28.4% (42 of 148) and a specificity of 98.6% (142 of 144) 13. The positive likelihood ratio is 20.4, well above 10, so when the finding is present it is close to decisive. The negative likelihood ratio is 0.73, barely below 1, so its absence is almost uninformative. A single finding therefore has markedly different informational value depending on whether it is present or absent. This is the general pattern of most clinical findings and the reason they should be characterised by both ratios rather than by a single accuracy figure.

28.4%sensitivity of the finding (42 of 148)
98.6%specificity of the finding (142 of 144)
20.4positive likelihood ratio — strong evidence to rule in
0.73negative likelihood ratio — absence is almost uninformative

A worked clinical finding (a history of more than 40 pack-years) showing why low sensitivity does not render a finding useless. Applied to a pre-test probability of 0.1, the positive result gives a post-test probability of about 0.69. Source: Deeks JJ, Altman DG. Diagnostic tests 4: likelihood ratios. BMJ 2004;329(7458):168-169.

Conversion from pre-test to post-test probability

The likelihood ratio informs a decision only once it is applied to a starting probability. In Deeks and Altman’s example, a pre-test probability of 0.1 combined with a positive likelihood ratio of 20.4 gives a post-test probability of about 0.69 13. The calculation proceeds through odds: the pre-test probability is converted to odds, multiplied by the likelihood ratio and converted back 13. In practice, most clinicians use a nomogram or a calculator; the mechanics matter less than the habit of establishing what the starting probability was.

Two aspects of that example merit attention. A strong positive finding moved a probability of approximately one in ten to approximately seven in ten, a large and clinically consequential shift. It nonetheless left approximately three chances in ten that the condition is absent. A finding with a likelihood ratio above 10 constitutes strong evidence but not proof, and a diagnosis acted upon at a post-test probability of 0.69 is a reasoned judgement under uncertainty rather than a certainty 8,13.

The pre-test probability is where most error enters, and it is the quantity least often recorded. It derives from the prevalence in the clinical setting, from the referral pathway and from everything already known about the patient. Stating it explicitly before ordering a test is arguably the most useful habit in diagnostic reasoning, because it determines whether the test can change management.

Identification and use of clinical prediction rules

A clinical prediction rule combines several findings rather than relying on one. Ingui and Rogers, in the Journal of the American Medical Informatics Association, adopt Laupacis’ definition: a rule combining three or more variables from the history, physical examination or simple tests 21. The threshold of three distinguishes a rule from a single diagnostic finding, and it reflects the purpose of such rules: individual findings are usually too weak to move a probability across a decision threshold on their own.

Their paper also addresses a practical problem: these rules are difficult to locate in bibliographic databases because they are indexed inconsistently 21. Ingui and Rogers developed and tested search filters against a gold standard of 119 articles, work of the kind that makes a literature usable at the point of care rather than only in principle 21. When searching for a rule relevant to a specific clinical question, a tested filter will outperform free-text searching.

Before any rule is applied, the stage of evaluation it has reached should be established: derived in one dataset, validated in another, or tested for its effect on clinical decisions and outcomes. A rule that has only been derived remains a hypothesis with associated coefficients. Our PICO framework page describes how to frame the question with sufficient precision that the search returns rules for the relevant population rather than for a different one.

Appraisal of diagnostic accuracy studies

QUADAS-2, published by Whiting and colleagues in the Annals of Internal Medicine, is the instrument designed for this purpose, and its four domains correspond to the four ways in which a diagnostic accuracy study may mislead. Patient selection: were patients enrolled consecutively from a relevant population, or was a convenient case-control contrast assembled from evident cases and healthy controls? The latter inflates both sensitivity and specificity, often substantially. Index test: was it interpreted without knowledge of the reference standard, and was the threshold pre-specified rather than chosen to optimise performance in this dataset 6? Reference standard: is it likely to classify the condition correctly, and was it applied blind to the index test? Flow and timing: did all enrolled patients receive both tests, and were those excluded accounted for?

Each domain also carries an applicability judgement, which is the most important for a practitioner. A study can be at low risk of bias and nonetheless concern patients unlike those seen in the practitioner’s own setting. The inclusion criteria and the prevalence in the study sample should be read before the accuracy figures: if the prevalence differs substantially from that in the practitioner’s setting, the likelihood ratios may still transfer, but the predictive values will not 9.

Interpretation of pooled accuracy estimates

When a systematic review has pooled several accuracy studies, several points should be checked before the summary is acted upon. Sensitivity and specificity should have been pooled jointly rather than separately: Reitsma and colleagues, in the Journal of Clinical Epidemiology, set out the bivariate approach, and Harbord and colleagues, in Biostatistics, relate it to the hierarchical summary ROC formulation 10,11. Two separate univariate pools ignore the fact that the two measures vary together with the threshold, and they misstate the joint uncertainty.

Second, the summary is an average across studies that differed. Riley, Higgins and Deeks, in the BMJ, argue that a random-effects summary estimates the mean of a distribution of effects, and that where that distribution is wide, the mean is not the appropriate value to apply to an individual patient 7. Higgins and Thompson’s work in Statistics in Medicine on describing heterogeneity provides the methodological background; its implication for practice is that the range of the individual study estimates should be examined, not only the summary diamond 19.

Third, any statement about publication bias should be treated with caution. Sterne and colleagues, in the BMJ, advise that tests for funnel plot asymmetry should generally not be used when fewer than 10 studies are pooled, and that the minimum may be substantially higher under substantial heterogeneity; consequently, a review of six accuracy studies reporting a non-significant test has not demonstrated an absence of bias 8,13. McInnes and colleagues’ PRISMA-DTA statement in JAMA sets out what such a review should have reported; our pages on cardiovascular imaging AI and CNN radiology reviews examine the problems that arise when it has not 4.

Documenting diagnostic reasoning in the clinical record

The difference between a defensible diagnostic decision and an assertion is usually evident in the clinical note. A record that states the pre-test probability, the finding, its likelihood ratio and the revised probability can be audited, disputed or revisited when new information becomes available. A record that states only a conclusion cannot be revisited, because it does not indicate the basis of that conclusion.

The same assessment, recorded two ways

Cannot be revisited

Smoking history positive. Diagnosis likely. Will proceed on that basis.

No starting point. Without a pre-test probability, a positive finding cannot be converted into a post-test probability.

No magnitude. “Likely” could mean 0.3 or 0.9, and the two support different actions.

No source for the accuracy figures. There is nothing to appraise, and nothing to update if better estimates become available.

The negative case is unconsidered. A finding whose absence is uninformative is treated as though it were informative in both directions.

Nobody can disagree usefully. A colleague reviewing the note cannot identify the step that they would dispute.

Can be audited and revised

Pre-test probability estimated at around 0.1, based on presenting features and the prevalence in this referral population. The patient reports a history of more than 40 pack-years; in the published estimate this finding has sensitivity 28.4% and specificity 98.6%, giving a positive likelihood ratio of 20.4 — above the threshold of 10 conventionally taken as strong evidence to rule in. Applied to the pre-test probability, this gives a post-test probability of about 0.69, so a substantial probability of no disease remains and the finding is not treated as confirmatory. Note that had the history been negative, the negative likelihood ratio of 0.73 would have changed the estimate very little.

The pre-test probability is stated, so the revision is reproducible and the estimate itself is open to challenge.

The likelihood ratio is given with its magnitude threshold, so the strength of the evidence is explicit.

The residual probability of no disease is named rather than rounded away into a diagnosis.

The asymmetry of the finding is recorded, which prevents its absence being over-read in future.

Every step is separable, so a colleague can accept the calculation while still disputing the starting probability.

The second note is not more cautious; it reaches the same working diagnosis. It is, however, auditable: each step can be challenged independently, and the record remains valid when a better accuracy estimate becomes available or the population being seen changes.

Limitations of the framework

The framework has three principal limitations. First, likelihood ratios indicate how to revise a probability; they do not indicate what probability justifies action. That threshold depends on the consequences of treating and of not treating, and it is a judgement about values rather than a statistic. A probability of 0.69 might be well beyond the threshold for starting a safe treatment and well short of it for a procedure with serious risks 13.

Second, the evidence base for most individual clinical findings is weaker than textbooks imply, and where accuracy estimates derive from small case-control-style studies in selected populations, they are likely to be optimistic. Where the literature does not provide a defensible likelihood ratio, this should be stated rather than an estimate improvised.

Third, the calculation assumes that findings are independent, and clinical findings rarely are. Multiplying several likelihood ratios in sequence for correlated findings overstates the revision. Multivariable clinical prediction rules exist for precisely this reason, and a validated rule should be preferred to a chain of individual ratios wherever one is available 21. Our biostatistics and systematic review pages address the development and synthesis aspects of this work.

Key points

Key points

Assessment and diagnosis depend on the revision of probability rather than on test accuracy. Sensitivity and specificity are conditional on the diagnosis already being known, whereas likelihood ratios can be applied to a result. Values above 10 or below 0.1 provide strong evidence to rule a diagnosis in or out, and the two directions are usually markedly unequal: Deeks and Altman’s worked finding had a positive ratio of 20.4 and a negative ratio of 0.73. The pre-test probability should be stated before testing, because it determines whether the result can alter management: a pre-test probability of 0.1 with the finding present yields a post-test probability of about 0.69, which constitutes strong evidence but not proof.

Application to your study

Application to your study

  1. Record the pre-test probability before ordering the test. If positive and negative results would lead to the same action, the test is unnecessary.
  2. Consult both likelihood ratios rather than the overall accuracy figure. Most findings are informative in one direction and of little value in the other.
  3. Use the thresholds of 10 and 0.1 as practical reference points. Beyond these thresholds, a result constitutes strong evidence; within them, its meaning depends on the pre-test probability.
  4. Examine the study population before the results. Distinguish consecutive enrolment from a comparable population from an assembled contrast of evident cases and healthy controls; the latter inflates accuracy.
  5. Prefer a validated clinical prediction rule to a sequence of single findings. Multiplying likelihood ratios for correlated findings overstates the revision of probability; a rule combining three or more variables accounts for the dependence.
  6. Determine whether a rule has been derived, validated or evaluated for impact. A rule that has only been derived remains a hypothesis with associated coefficients.
  7. When interpreting a pooled accuracy estimate, examine the spread of the individual study results. The summary estimate is the mean of a distribution, and that distribution may be wide.
  8. Record the reasoning as well as the conclusion. Recording the pre-test probability, the finding, the likelihood ratio and the post-test probability renders the clinical note auditable.

Frequently asked questions

A likelihood ratio indicates how much a test result changes the probability of a diagnosis. The positive likelihood ratio expresses how much more likely a positive result is in a person with the condition than in a person without it; the negative likelihood ratio expresses the same for a negative result. A ratio of 1 means that the result changes nothing. It is more useful than sensitivity because sensitivity is conditioned on knowledge of the diagnosis (it answers the question “if the patient has the disease, will the test detect it?”), whereas a likelihood ratio can be applied directly to the observed result, combined with the starting probability. Deeks and Altman set this out in the BMJ.

Deeks and Altman describe likelihood ratios above 10 or below 0.1 as providing strong evidence to rule a diagnosis in or out, respectively. Between those values, whether the result matters depends entirely on the starting probability: a ratio of 3 may be decisive at a starting probability of 0.4 and irrelevant at 0.02. The two ratios for a single finding are usually markedly unequal. In their worked example, a history of more than 40 pack-years had a positive likelihood ratio of 20.4 but a negative likelihood ratio of 0.73: strong evidence when present, almost uninformative when absent.

Yes, and this is the most practically important consequence of reasoning with likelihood ratios. In Deeks and Altman’s example, the finding had a sensitivity of only 28.4% (42 of 148), so it was absent in most people with the condition. However, its specificity was 98.6% (142 of 144), which yields a positive likelihood ratio of 20.4, above the conventional threshold of 10 for strong evidence to rule in. The finding is therefore close to decisive when present and nearly uninformative when absent. Judging it by sensitivity alone would have led to a highly informative finding being discarded.

The pre-test probability is converted to odds, multiplied by the likelihood ratio, and the result converted back to a probability. In practice, most clinicians use a nomogram or a calculator, and the mechanics matter less than recognising that the result depends on the starting point. In Deeks and Altman’s worked example, a pre-test probability of 0.1 with a positive likelihood ratio of 20.4 gives a post-test probability of about 0.69. Two conclusions follow: a strong finding produced a large, consequential shift, and it nonetheless left approximately a three-in-ten probability that the condition is absent.

A rule that combines several clinical variables to estimate the probability of a diagnosis or outcome. Ingui and Rogers, in the Journal of the American Medical Informatics Association, adopt Laupacis’ definition: a rule combining three or more variables from the history, physical examination or simple tests. Such rules are difficult to locate because they are indexed inconsistently, which is why Ingui and Rogers developed and tested search filters against a gold standard of 119 articles. A tested filter should be used rather than free-text searching, and before any rule is applied, it should be established whether it has only been derived, has been validated in a separate population, or has been tested for its effect on decisions and outcomes.

Usually because of the population rather than the test. Accuracy depends on the spectrum of patients studied: a study contrasting evident cases with healthy controls will report far better sensitivity and specificity than one enrolling consecutive patients with a diagnostic question. QUADAS-2, from Whiting and colleagues in the Annals of Internal Medicine, makes patient selection one of its four domains for precisely this reason, and attaches an applicability judgement to each. Likelihood ratios transfer across settings better than predictive values do, but neither transfers from a selected case series to primary care without scrutiny of the inclusion criteria and the sample prevalence.

Only if the findings are independent, which clinical findings usually are not. Two signs that reflect the same underlying process carry overlapping information, so multiplying their likelihood ratios in sequence overstates how far the probability should move. This is the main argument for using a validated multivariable clinical prediction rule (one combining three or more variables, in Laupacis’ definition) where one exists for the clinical question, because the rule’s coefficients were fitted with the dependence present. Where no rule exists, the single strongest ratio should be applied and additional concordant findings treated as qualitative support rather than multiplied further.

Three points should be checked. First, whether the two were pooled jointly (using Reitsma and colleagues’ bivariate model or the hierarchical summary ROC approach described by Harbord and colleagues) rather than as two separate univariate estimates, which misstates the joint uncertainty. Second, whether the individual study estimates were widely spread, since Riley, Higgins and Deeks argue that a random-effects summary is the mean of a distribution and may not represent any real population. Third, whether any claim about publication bias is supportable: Sterne and colleagues advise against funnel plot asymmetry tests with fewer than 10 studies, so a non-significant test in a small review demonstrates nothing.

References

  1. Jamtvedt G, Young JM, Kristoffersen DT, Thomson O'Brien MA, Oxman AD. Audit and feedback: effects on professional practice and health care outcomes. Cochrane Database of Systematic Reviews. 2003:CD000259. read 9 October 2026. bmcmededuc.biomedcentral.com
  2. Parkes J, Hyde C, Deeks J, Milne R. Teaching critical appraisal skills in health care settings. Cochrane Database of Systematic Reviews. 2001:CD001270. read 9 October 2026. bmcmededuc.biomedcentral.com
  3. Frank RA, Salameh JP, Islam N, Yang B, Murad MH, Mustafa R, Leeflang M, Bossuyt PM, Takwoingi Y, Whiting P, Dawit H, Kang SK, Ebrahimzadeh S, Levis B, Hutton B, McInnes MDF. How to critically appraise and interpret systematic reviews and meta-analyses of diagnostic accuracy: a user guide. Radiology. 2023;307(3):e221437. doi:10.1148/RADIOL.221437 read 9 October 2026. doi.org/10.1148/RADIOL.221437
  4. McInnes MDF, Moher D, Thombs BD, McGrath TA, Bossuyt PM, et al. Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies: The PRISMA-DTA Statement. JAMA. 2018;319(4):388-396. doi:10.1001/jama.2017.19163 read 9 October 2026. doi.org/10.1001/jama.2017.19163
  5. Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, Altman DG. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology. 2016;31:337-350. doi:10.1007/s10654-016-0149-3 read 9 October 2026. doi.org/10.1007/s10654-016-0149-3
  6. Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, Lijmer JG, Moher D, Rennie D, de Vet HCW, Kressel HY, Rifai N, Golub RM, Altman DG, Hooft L, Korevaar DA, Cohen JF; STARD Group. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015;351:h5527. doi:10.1136/bmj.h5527 read 9 October 2026. doi.org/10.1136/bmj.h5527
  7. Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ. 2011;342:d549. doi:10.1136/bmj.d549 read 9 October 2026. doi.org/10.1136/bmj.d549
  8. Sterne JAC, Sutton AJ, Ioannidis JPA, Terrin N, Jones DR, Lau J, Carpenter J, Rucker G, Harbord RM, Schmid CH, Tetzlaff J, Deeks JJ, Peters J, Macaskill P, Schwarzer G, Duval S, Altman DG, Moher D, Higgins JPT. Recommendations for examining and interpreting funnel plot asymmetry in meta-analyses of randomised controlled trials. BMJ. 2011;343:d4002. doi:10.1136/bmj.d4002 read 9 October 2026. doi.org/10.1136/bmj.d4002
  9. Leeflang MMG, Deeks JJ, Gatsonis C, Bossuyt PMM; Cochrane Diagnostic Test Accuracy Working Group. Systematic reviews of diagnostic test accuracy. Annals of Internal Medicine. 2008;149(12):889-897. doi:10.7326/0003-4819-149-12-200812160-00008 read 9 October 2026. doi.org/10.7326/0003-4819-149-12-200812160-00008
  10. Harbord RM, Deeks JJ, Egger M, Whiting P, Sterne JAC. A unification of models for meta-analysis of diagnostic accuracy studies. Biostatistics. 2007;8:239-251. read 9 October 2026. arxiv.org
  11. Reitsma JB, Glas AS, Rutjes AWS, Scholten RJPM, Bossuyt PM, Zwinderman AH. Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. Journal of Clinical Epidemiology. 2005;58:982-990. read 9 October 2026. arxiv.org
  12. Dawes M, Summerskill W, Glasziou P, Cartabellotta A, Martin J, Hopayian K, Porzsolt F, Burls A, Osborne J. Sicily statement on evidence-based practice. BMC Medical Education. 2005;5:1. doi:10.1186/1472-6920-5-1 read 9 October 2026. doi.org/10.1186/1472-6920-5-1
  13. Deeks JJ, Altman DG. Diagnostic tests 4: likelihood ratios. BMJ. 2004;329(7458):168-169. doi:10.1136/bmj.329.7458.168 read 9 October 2026. doi.org/10.1136/bmj.329.7458.168
  14. Doust JA, Del Mar CB. Why do doctors use treatments that do not work? BMJ. 2004;328:474-475. read 9 October 2026. bmcmededuc.biomedcentral.com
  15. Johnston JM, Leung GM, Fielding R, Tin KYK, Ho LM. The development and validation of a knowledge, attitude and behaviour questionnaire to assess undergraduate evidence-based practice teaching and learning. Medical Education. 2003;37:992-1000. read 9 October 2026. bmcmededuc.biomedcentral.com
  16. Ramos KD, Schafer S, Tracz SM. Validation of the Fresno test of competence in evidence based medicine. BMJ. 2003;326:319-321. read 9 October 2026. bmcmededuc.biomedcentral.com
  17. Fritsche L, Greenhalgh T, Falck-Ytter Y, Neumayer HH, Kunz R. Do short courses in evidence based medicine improve knowledge and skills? Validation of Berlin questionnaire and before and after study of courses in evidence based medicine. BMJ. 2002;325:1338-1341. read 9 October 2026. bmcmededuc.biomedcentral.com
  18. Ely JW, Osheroff JA, Ebell MH, Chambliss ML, Vinson DC, Stevermer JJ, Pifer EA. Obstacles to answering doctors' questions about patient care with evidence: qualitative study. BMJ. 2002;324:710. read 9 October 2026. bmcmededuc.biomedcentral.com
  19. Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine. 2002;21(11):1539-1558. doi:10.1002/sim.1186 read 9 October 2026. doi.org/10.1002/sim.1186
  20. Haynes RB. Of studies, syntheses, synopses, and systems: the 4S evolution of services for finding current best evidence. ACP Journal Club. 2001;134:A11-A13. read 9 October 2026. bmcmededuc.biomedcentral.com
  21. Ingui BJ, Rogers MAM. Searching for clinical prediction rules in MEDLINE. Journal of the American Medical Informatics Association. 2001;8(4):391-397. doi:10.1136/jamia.2001.0080391 read 9 October 2026. doi.org/10.1136/jamia.2001.0080391
  22. Green ML. Evidence-based medicine training in graduate medical education: past, present and future. Journal of Evaluation in Clinical Practice. 2000;6:121-138. read 9 October 2026. bmcmededuc.biomedcentral.com
  23. Guyatt G, Meade MO, Jaeschke R, Cook DJ, Haynes B. Practitioners of evidence based care. BMJ. 2000;320:934-935. read 9 October 2026. bmcmededuc.biomedcentral.com
  24. Straus SE, Sackett DL. Applying evidence to the individual patient. Annals of Oncology. 1999;10:29-32. read 9 October 2026. bmcmededuc.biomedcentral.com
  25. McAlister FA, Graham I, Karr GW, Laupacis A. Evidence-based medicine and the practicing clinician. Journal of General Internal Medicine. 1999;14:236-242. read 9 October 2026. bmcmededuc.biomedcentral.com
  26. Sackett DL, Rosenberg WMC, Gray JAM, Haynes RB, Richardson WS. Evidence based medicine: what it is and what it isn't. BMJ. 1996;312(7023):71-72. doi:10.1136/bmj.312.7023.71 read 9 October 2026. doi.org/10.1136/bmj.312.7023.71
  27. Rosenberg W, Donald A. Evidence based medicine: an approach to clinical problem-solving. BMJ. 1995;310:1122-1126. read 9 October 2026. bmcmededuc.biomedcentral.com
  28. Richardson WS, Wilson MC, Nishikawa J, Hayward RS. The well-built clinical question: a key to evidence-based decisions. ACP Journal Club. 1995;123:A12-A13. read 9 October 2026. bmcmededuc.biomedcentral.com
  29. Dopson S, Mant J, Hicks N. Getting research into practice: facing the issues. Journal of Management in Medicine. 1994;8:4-12. read 9 October 2026. bmcmededuc.biomedcentral.com
  30. Guyatt GH, Rennie D. Users' guides to the medical literature. JAMA. 1993;270:2096-2097. read 9 October 2026. bmcmededuc.biomedcentral.com
  31. Oxman AD, Sackett DL, Guyatt GH. Users' guides to the medical literature. I. How to get started. JAMA. 1993;270:2093-2095. read 9 October 2026. bmcmededuc.biomedcentral.com
  32. Evidence-Based Medicine Working Group. Evidence-based medicine: a new approach to teaching the practice of medicine. JAMA. 1992;268:2420-2425. read 9 October 2026. bmcmededuc.biomedcentral.com
  33. DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986;7(3):177-188. doi:10.1016/0197-2456(86)90046-2 read 9 October 2026. doi.org/10.1016/0197-2456(86)90046-2
  34. Veroniki AA, Jackson D, Bender R, Kuss O, Langan D, Higgins JPT, Knapp G, Salanti G. Methods to calculate uncertainty in the estimated overall effect size from a random-effects meta-analysis. doi:10.1002/jrsm.1319 read 9 October 2026. doi.org/10.1002/jrsm.1319

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 9). Evidence-Based Practice in Assessment and Diagnosis: Likelihood Ratios and Probability Revision. Pubrica. https://pubrica.com/academy/meta-analysis/how-healthcare-professionals-can-use-ebp-in-assessment-diagnosis/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Diagnostic reasoning note you can copy

Four clauses, written before and after the test rather than only afterwards. Completing the note takes under a minute once the likelihood ratios are available, and it makes the decision auditable by a colleague who disagrees.

DIAGNOSTIC REASONING NOTE
Patient: ______________  Date: __________  Clinician: ______________

1. QUESTION
   Condition being considered: ______________________
   Decision this will inform: ______________________
   Probability at which I would act: ______  (and why: ____________)
   Probability at which I would stop considering it: ______

2. PRE-TEST PROBABILITY   (write this BEFORE testing)
   Estimate: ______
   Basis:  [ ] prevalence in this setting  [ ] referral pathway
           [ ] presenting features  [ ] prior results
   Note: ______________________________________________

3. FINDING OR TEST
   Finding / test: ______________________
   Result: [ ] positive  [ ] negative  [ ] indeterminate
   Positive likelihood ratio: ______   Negative likelihood ratio: ______
   Source of these figures: ______________________
   Was the source population like this patient?  [ ] yes  [ ] no
     Note: ______________________
   Strength reference: LR above 10 or below 0.1 = strong evidence to
   rule in / rule out.

4. POST-TEST PROBABILITY
   Revised estimate: ______
   Does this cross either threshold in section 1?  [ ] act  [ ] stop
     [ ] neither — further information needed, specifically: ________
   Residual probability of NO disease: ______
   Would the opposite result have changed anything?  [ ] yes  [ ] no
     (If no, reconsider whether the test was indicated)

5. IF USING A CLINICAL PREDICTION RULE INSTEAD
   Rule: ______________________
   Stage of evidence:  [ ] derived only  [ ] externally validated
                       [ ] impact on decisions/outcomes tested
   Validated in a population like this one?  [ ] yes  [ ] no
   Variables combined (3 or more by definition): ______________________

6. UNCERTAINTY TO CARRY FORWARD
   What would change this assessment: ______________________
   Review point / safety netting: ______________________

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica’s medical writing and biostatistics teams address the research aspects of this framework: appraising diagnostic accuracy studies against QUADAS-2, fitting bivariate and hierarchical models when accuracy studies are pooled, and reporting diagnostic accuracy reviews and prediction model studies according to PRISMA-DTA. Where the evidence for a finding or a rule is weaker than it appears, Pubrica states this explicitly in the manuscript.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Pubrica can review a diagnostic accuracy dataset or review protocol and advise on the conclusions it can support.

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.