class: RWH_bg_title # Building better biomarkers ### Scientific Frontiers of Medicine .RWH_footnote_title[ .RWH_footer_bold[ Rob Hunter | @renalrob ] ] .RWH_footnote_right_title[ .RWH_footer_bold[ Oct 2025 ] ] ??? BMedSci Health Sciences | Scientific Frontiers of Medicine Slides created with [xaringan](https://github.com/yihui/xaringan). --- # Learning intention - what is a biomarker? - what do we use biomarkers for? - what are the characteristics of an ideal biomarker? - strategies for developing better biomarkers <br> <!--  --> <!-- .RWH_footnote_right[.RWH_footer_style[slides at: www.kidneyfish.net/talks/]] --> ??? <!-- Notes for 2024 --> <!-- - state my credentials (miRNA in dogs and alc hep; Cr joint model; snGFR; TAKA grant...) --> <!-- - fireman study (Luft talk UCL 2024) --> <!-- - tweet on prediction (bookmarked from F Perry Wilson 14/05/24) --> <!-- - Topol article on proteomics --> <!-- Notes for 2023 Jeremy covers the new approaches to biomarkers bit (ctDNA, machine learning etc...) - so can drop this. Add back in the mis-application of biomarkers bit - Bayesian (fireworks) and PSA etc. etc. What about adding in CAST trial? Perhaps call "use and abuse of biomarkers?" or similar. Make more of SaO2 and eGFR in Black patients? Could speed up a little bit for the first slides - took around 30 mins to get to end of "what makes an ideal biomarker. --> --- class: center, middle, inverse # .white[What is a biomarker?] --- # Definition of a biomarker <br> > a medical test -- <br> <br> > a defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes, or responses to an exposure or intervention, including therapeutic interventions -- <br> <br> .red[<b>So what sort of things can be used as biomarkers?</b>] ??? May be: - physiological (e.g. BMI, ABP, height) - molecular (e.g. blood tests) - histological - imaging Definition from [FDA/NIH Biomarker working group (BEST)](https://www.ncbi.nlm.nih.gov/books/NBK326791/). --- class: center, middle, inverse # .white[What are they used for?] ---  ??? From [FDA/NIH Biomarker working group (BEST)](https://www.ncbi.nlm.nih.gov/books/NBK326791/). ---  ??? “Clinical outcomes” = how a patient feels / functions / survives “Exposures” = disease / environmental exposures / drugs… ---  <br> <table class="table" style="color: black; margin-left: auto; margin-right: auto;"> <thead> <tr> <th style="text-align:left;"> role </th> <th style="text-align:left;"> examples </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[diagnosis] </td> <td style="text-align:left;font-style: italic;"> .can-edit[HIV Ab/Ag test] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> .can-edit[...] </td> <td style="text-align:left;font-style: italic;"> .can-edit[...] </td> </tr> </tbody> </table> ??? This is an editable slide; we will fill in answers together during the session. ---  <br> <table class="table" style="color: black; margin-left: auto; margin-right: auto;"> <thead> <tr> <th style="text-align:left;"> role </th> <th style="text-align:left;"> examples </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;font-weight: bold;"> diagnosis </td> <td style="text-align:left;font-style: italic;"> sweat chloride, HIV Ab/Ag, HbA1c, eGFR </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> diagnosis (definition) </td> <td style="text-align:left;font-style: italic;"> ABP, BMI </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> risk </td> <td style="text-align:left;font-style: italic;"> LDL, BRCA1/2 </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> response (pharmacodynamic) </td> <td style="text-align:left;font-style: italic;"> ABP, HbA1c, INR, viral load </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> safety </td> <td style="text-align:left;font-style: italic;"> ALT, bilirubin, K, QTc </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> monitoring </td> <td style="text-align:left;font-style: italic;"> HbA1c, eGFR, symphisis-fundal height </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> prognostic </td> <td style="text-align:left;font-style: italic;"> CRP, Gleason score, TKV in ADPKD </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> prognostic (validated surrogate) </td> <td style="text-align:left;font-style: italic;"> ABP </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> predictive </td> <td style="text-align:left;font-style: italic;"> uACR in DKD, ER+ in breast Ca, LBBB in HFrEF </td> </tr> </tbody> </table> ??? ### Examples - diagnosis = ABP, HbA1c, eGFR, sweat chloride, HIV Ag/Ab tests, D-dimer, PSA… - risk = BRCA1/2, LDL-C… - pharmacodynamic/respose (shows a biological response has been elicited by a drug / exposure) = INR, HbA1c, ABP, viral load… - safety = ALT, bilirubin, K+, TMPT, QTc… - monitoring = eGFR, HbA1c, symphisis-fundal height, viral load, PSA… - prognostic = TKV in ADPKD, Gleason score, CRP… - predictive = uACR in DKD, squamous differentiation in NSCLC, ER+ in breast cancer (respond to hormonal Rx), LBBB in HFrEF (respond to biventricular pacing) <br> Largely academic / philosophical, but some biomarkers may act *as* biomarkers in some contexts but are better considered as bona fide physiological variables in other contexts (e.g. ABP). <br> ### Notes on prognostic and predictive biomarkers Difference between prognostic and predictive biomarkers: predictive biomarkers require comparison between treatment and control groups (so that they can predict the response to treatment). e.g. Benefit of tamoxifen in ER+ breast cancer: [EBCTCG (Lancet, 2005)](https://www.thelancet.com/journals/lancet/article/PIIS0140673605665440/fulltext). <br> Note that biomarkers that associate closely with a clinical outcome [may not necessarily function well in prediction tools](https://twitter.com/fperrywilson/status/1790389487429615655). Two potential reasons for this are: 1) if biomarker positivity is rare within the relevant patient population (so that, although the biomarker may be strongly associated with the outcome, most outcomes occur in individuals without positive biomarker); and 2) if the biomarker reports information that is already present in the risk predicion model (e.g. due to co-associations with other variables already in the model). ---  <br> <table class="table" style="color: black; margin-left: auto; margin-right: auto;"> <thead> <tr> <th style="text-align:left;"> role </th> <th style="text-align:left;"> examples </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;font-weight: bold;"> eligibility for Rx </td> <td style="text-align:left;font-style: italic;"> BMI for GLP1ra </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> eligibility for clinical trials </td> <td style="text-align:left;font-style: italic;"> uACR in nephrology trials </td> </tr> <tr> <td style="text-align:left;font-weight: bold;"> epidemiology </td> <td style="text-align:left;font-style: italic;"> Covid RNA in wastewater </td> </tr> </tbody> </table> ??? Biomarkers are also used at a population level. Again largely academic / philosophical but should we consider internet search data as a biomarker (e.g. the apocryphal Yankee candle tweets)? Does fit the definition! Analogy to Richard Dawkins extended phenotype. --- class: center, middle, inverse # .white[What makes an ideal biomarker?] --- # An ideal biomarker .pull-left[ ## accuracy - .can-edit[sensitive (low false-negatives)] - .can-edit[...] - .can-edit[...] - .can-edit[...] - .can-edit[...] - .can-edit[...] ] .pull-right[ ## practicalites - .can-edit[cheap] - .can-edit[...] - .can-edit[...] - .can-edit[...] - .can-edit[...] - .can-edit[...] ] ??? This is an editable slide; we will fill in answers together during the session. --- # An ideal biomarker .pull-left[ ## accuracy - sensitive (low false-negatives) - specific (low false-positives) - reproducible within subjects - different populations - quantitative - mechanism understood ] .pull-right[ ## practicalites - cheap - quick - easy - point-of-care - home testing - resource-poor settings ] ??? The easy criterion is really important. To illustrate how even apparently trivial layers of complexity are actually significant barriers could discuss abdominal girth *vs* BMI in obesity (correlates better with body fat and important outcomes) and low rates (25%) or uACR in CKD. See https://www.nature.com/articles/s41574-019-0310-7 and https://www.nature.com/articles/s41574-024-01012-9?fromPaywallRec=false. --- # What can go wrong? .pull-left-40[] -- .pull-right-60[] .RWH_footnote_right[.RWH_footer_style[[Sjoding et al (NEJM, 2020)](https://pubmed.ncbi.nlm.nih.gov/33326721/) | [Kellet et al (Nature, 2022)](https://pubmed.ncbi.nlm.nih.gov/36261563/)]] ??? Pulse oximeters developed in volunteers from US Airforce in 1940s. What problem might that cause? Pulse-oximetry systematically under-estimates extent of true hypoxaemia in Black individuals. --- # What can go wrong? Other examples: - PSA screening for prostate cancer - ventricular ectopics for arrhythmia risk - Covid lateral flow tests - eGFR ??? ### To talk about - PSA good at detecting cancer, but in screening does not translate into reduced mortality - ventricular ectopics plausibly bad but actually trying to treat them ended up increasing mortality (CAST) - Covid LFTs for home-testing; used because cheap and easy; however diagnostic performance was not perfect and widely misunderstood; could have increased spread if negative result taken to mean no infection - eGFR in different populations; risk of perpetuating health-inequalities in patients with African or Caribbean family origin; risk of over-medicalising older individuals with normal age-decline in GFR (not CKD) ---  <!--[:scale_c 80%]--> ??? We should challenge this assumption that increasing complexity = better accuracy (we will look at CKD273 and deep learning approach to predicting AKI). Very simple biomarkers can give powerful information. The 6-minute walk test (how far can you walk in 6 mins) predicts mortality in COPD ([Cote et al, Eur Resp J 2008](https://pubmed.ncbi.nlm.nih.gov/17989117/)). The number of press-ups predicts risk of a cardiovascular event in firefighters - better than VO₂ max ([Yang et al, JAMA 2019](https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2724778)). --- class: center, middle, inverse # .white[How can we design better biomarkers?] --- # Better biomarkers 1) more sensitive molecular assays 2) analysing tens - hundreds of tests in parallel 3) using machine learning to pick the best tests 4) analysing millions of biomarkers (also machine learning) 5) thinking outside the box --- # Sensitive molecular assays .pull-left-40[  <br>  ] .pull-right-60[ - early diagnosis - minimal residual disease - response to therapy - molecular profiling - disease resistance - clonal dynamics ] .RWH_footnote_right[.RWH_footer_style[[Corcoran et al. (NEJM, 2018)](https://www.nejm.org/doi/10.1056/NEJMra1706174) | [Dawson et al. (NEJM, 2013)](https://www.nejm.org/doi/full/10.1056/nejmoa1213261)]] ??? Approaches to detecting cancer include: CTCs (circulating tumour cells), ctDNA (circulating tumour DNA), miRNA. ctDNA can be crude quantitative (how much is there) and also mechanistic - what are the mutations? - to guide personalised chemotherapy. Concept of minimal residual disease = micro-metastases (not clinically / radiographically detectable). Dawson et al = proof-of-concept study in 30 women with metastatic breast cancer. ctDNA was more sensitive than CA 15-3. Now really very sophisticated. See [PATHFINDER trial](https://pubmed.ncbi.nlm.nih.gov/37805216/): early detection of 50 cancer types through cancer-specific DNA methylation patterns in cfDNA. Yeild 0.5%; PPV 40%; NPV 99%. --- # Hundreds of biomarkers  .RWH_footnote_right[.RWH_footer_style[[Good et al (Mol Cell Proteomics, 2010)](https://www.sciencedirect.com/science/article/pii/S1535947620345011?via%3Dihub)] & [Argiles et al (PloS One 2013)](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0062837)] ??? Concepts of "omics": proteomics, transcriptomics, metabolomics… Urinary peptidome characterised by CE-MS (capillary electrophoresis then mass spectrometry). CKD = CKD; HC = healthy control. 273 biomarker peptides identified. Subsequently validated in the diagnosis of CKD and in the ability to act as a prognostic (for CKD progression) and predictive (response to RASi) biomarker ([Pontillo & Mischak, CKJ 2017](https://pubmed.ncbi.nlm.nih.gov/28694965/); [Canadas-Garre et al., J Proteomics 2019)](https://www.sciencedirect.com/science/article/pii/S1874391918303610?via%3Dihub). --- # Hundreds of biomarkers  .RWH_footnote_right[.RWH_footer_style[[Pontillo et al (NDT, 2017)](https://academic.oup.com/ndt/article/32/9/1510/3059436)]] ??? So does CKD273 perform better than AER? Tested ability of CKD273 and AER to discriminate between progressive CKD (eGFR slope > 5 ml/min/yr) from non-progressive in a cohort of > 2000 patients with CKD (predominantly DKD). Calculated AUC in ROC-curve, stratified by eGFR. CKD273 performed better at high eGFR; AER performed better at low eGFR - but I think take-home message here is that diagnostic performance was disappointingly similar to AER. --- # Machine learning  .RWH_footnote_right[.RWH_footer_style[[Carrasco-Zanini et al (Nat Med, 2024)](https://www.nature.com/articles/s41591-024-03142-z)]] ??? UK Biobank Pharma Proteomics Project. 3000 plasma proteins. Machine learning. Compared prediction models using proteomic data with basic clinical data (+/- existing biomarkers). See also commentary by [Topol, Science 2024](https://www.science.org/doi/10.1126/science.ads5749). ---  .RWH_footnote_right[.RWH_footer_style[[Carrasco-Zanini et al (Nat Med, 2024)](https://www.nature.com/articles/s41591-024-03142-z)]] ??? Models incorporating 5 - 20 plasma proteins improved predicition for 67 diseases (median delta c-index 0.07). - DR = detection rate = TP / (FN + TP) - FPR = FP / (TN + FP) - LR = DR / FPR --- # Millions of biomarkers  .RWH_footnote_right[.RWH_footer_style[[Tomasev et al (Nature, 2019)](https://www.nature.com/articles/s41586-019-1390-1)]] -- <br> .red[<b>So how accurate do you think this was?</b>] ??? --- # Millions of biomarkers   .RWH_footnote_right[.RWH_footer_style[[Tomasev et al (Nature, 2019)](https://www.nature.com/articles/s41586-019-1390-1)]] ??? Consider: - was the test population appropriate (veterans - largely excluded women)? - was the outcome measure worth knowing about AKI (just a rise in SCr)? - is this good enough? - any better than existing models / gestalt? --- # Unknown identity  .RWH_footnote_right[.RWH_footer_style[[Srivastava et al (KI, 2019)](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6396290/)]] ??? Biological assay in FSGS. Response genes were IL1b, BMF and IGFBP3 No need to know the identity of the circulating factor or the nature of the cellular response. Could accurately identify bioactive circulating factors and therefore predict recurrent FSGS after transplant. Intriguing but totally impractical. --- <!-- # Living biosensor --> <!--  --> <!-- .RWH_footnote_right[.RWH_footer_style[[Cooper et al (Science, 2023)](https://pubmed.ncbi.nlm.nih.gov/37561843/)]] --> # When speed is important  .RWH_footnote_right[.RWH_footer_style[[Patel et al. (Nature Medicine, 2025)](https://doi.org/10.1038/s41591-025-03562-5)]] ??? The problem: - CNS tumours classically hard to phenotype (combination of morphology, single gene markers, DNA metyhlation etc.) - DNA methylation-based classification is [accurate](https://doi-org.eux.idm.oclc.org/10.1038/nature26000) but slow - takes around 28 days for a methylation array report (including sample preparation, transport etc.) <br> The solution: - adapative nanopore sampling: reject reads by reversing polarity if initial short read shows not from area of interest (although can still get some information on copy number, methylation status etc.) - therefore developed into a [very fast sequencing method](https://doi.org/10.1038/s41591-025-03562-5) that gives useful information within 30 mins - therefore good for intra-operative decision-making - required refinement of library preparation, sequencing conditions, [bioinformatic algorithm](https://www-nature-com.eux.idm.oclc.org/articles/s41587-020-00746-x), cost reduction etc. etc. --- # Learning intention - what is a biomarker? - what do we use biomarkers for? - what are the characteristics of an ideal biomarker? - strategies for developing better biomarkers .RWH_footnote_right[.RWH_footer_style[slides at: www.kidneyfish.net/talks/]] --- # Take-home points - a measure of a physiological or pathological process (or response to exposure or intervention) - the ultimate utility of a biomarker is in its ability to affect how an individual feels, functions or survives - can be used for risk-stratification, diagnosis, prognosis, personalised medicine, safety monitoring... - the value of a biomarker depends on diagnostic accuracy and practical ease of use - strategies to develop better biomarkers include highly-sensitivity molecular assays, multi-molecular biomarkers and machine learning --- class: center, middle, inverse # .white[Supplemental slides] --- class: center, middle, inverse # .white[Diagnostic performance] ??? We need to be able to measure / quantify diagnostic performance. This will depend on the context (patient population and outcome). Diagnostic accuracy is defined as: - discriminative accuracy = ability to distinguish between two states (e.g. healthy and diseased) - or predictive accuracy = ability to contribute to the posterior probability <br> This may be assessed using: - sensitivity & specificity - PPV & NPV = determined by prevalence - LR+ and LR- = independent of prevalence - AUC in ROC curve (= c-statistic) - for dichotomous outcomes - correlation analysis - for continuous outcomes <br> Remember: - accurate = close to centre of target - precise = clustered close together - sensitivity = true positives / all cases - (1 – specificity) = false positives / all negatives <br> Useful references: - [Ray et al, “Statistical Evaluation of a Biomarker” (Anesthestiology, 2010)](https://pubs.asahq.org/anesthesiology/article/112/4/1023/10681/Statistical-Evaluation-of-a-Biomarker) - [Loong (BMJ, 2003)](https://www.bmj.com/content/327/7417/716) - [Simundic, “Measures of Diagnostic Accuracy: Basic Definitions” (EJIFCC, 2009)](https://pubmed.ncbi.nlm.nih.gov/27683318/) - [Irwig et al, “Designing studies to ensure that estimates of test accuracy are transferable” (BMJ, 2002)](https://www.bmj.com/content/324/7338/669) - [CEBM on likelihood ratios](https://www.cebm.ox.ac.uk/resources/ebm-tools/likelihood-ratios#:~:text=Definition,patient%20without%20the%20target%20disorder) - [Cohrane blog](https://s4be.cochrane.org/blog/2015/03/03/ebm-for-diagnostic-tests/) - [Review on pre- and post-test probabilities in A&E](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4744617/) ---  .RWH_footnote_right[.RWH_footer_style[[Oger et al (Am J Resp Crit Care Med, 1998)](https://www.atsjournals.org/doi/10.1164/ajrccm.158.1.9710058)]] ??? ### To talk about - What could be causing the “false” positives – think about underlying biology? - False positive d-dimer in age, malignancy, pregnancy, post-surgery… - How could this be useful? …to rule OUT disease ---  <table class="table" style="color: black; margin-left: auto; margin-right: auto;"> <thead> <tr> <th style="text-align:left;"> context </th> <th style="text-align:left;"> biomarker </th> <th style="text-align:right;"> LR+ </th> <th style="text-align:right;"> LR- </th> <th style="text-align:left;"> notes </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;"> pregnant women </td> <td style="text-align:left;font-weight: bold;"> HIV test </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 640.0 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.00 </td> <td style="text-align:left;font-style: italic;"> almost perfect </td> </tr> <tr> <td style="text-align:left;"> Covid </td> <td style="text-align:left;font-weight: bold;"> LFT </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 400.0 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.60 </td> <td style="text-align:left;font-style: italic;"> rule in or out? </td> </tr> <tr> <td style="text-align:left;"> choledocholithaisis </td> <td style="text-align:left;font-weight: bold;"> elevated LFTs </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 23.0 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.60 </td> <td style="text-align:left;font-style: italic;"> </td> </tr> <tr> <td style="text-align:left;"> Covid </td> <td style="text-align:left;font-weight: bold;"> PCR </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 14.0 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.30 </td> <td style="text-align:left;font-style: italic;"> </td> </tr> <tr> <td style="text-align:left;"> choledocholithaisis </td> <td style="text-align:left;font-weight: bold;"> CBD dilatation </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 8.1 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.10 </td> <td style="text-align:left;font-style: italic;"> </td> </tr> <tr> <td style="text-align:left;"> ACS </td> <td style="text-align:left;font-weight: bold;"> ST depression </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 5.3 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.80 </td> <td style="text-align:left;font-style: italic;"> </td> </tr> <tr> <td style="text-align:left;"> malaria after travel </td> <td style="text-align:left;font-weight: bold;"> fever </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 5.1 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.12 </td> <td style="text-align:left;font-style: italic;"> </td> </tr> <tr> <td style="text-align:left;"> heart failure </td> <td style="text-align:left;font-weight: bold;"> NT-pro-BNP </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 3.8 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.10 </td> <td style="text-align:left;font-style: italic;"> c.f. ECG </td> </tr> <tr> <td style="text-align:left;"> heart failure </td> <td style="text-align:left;font-weight: bold;"> abnormal ECG </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 3.2 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.00 </td> <td style="text-align:left;font-style: italic;"> </td> </tr> <tr> <td style="text-align:left;"> GCA </td> <td style="text-align:left;font-weight: bold;"> ESR &gt; 100 | &lt; 40 </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 3.1 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.18 </td> <td style="text-align:left;font-style: italic;"> </td> </tr> <tr> <td style="text-align:left;"> VTE </td> <td style="text-align:left;font-weight: bold;"> d-dimer </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 2.4 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 0.10 </td> <td style="text-align:left;font-style: italic;"> see next slide </td> </tr> <tr> <td style="text-align:left;"> osteomyelitis in DM </td> <td style="text-align:left;font-weight: bold;"> swab culture </td> <td style="text-align:right;font-weight: bold;color: red !important;"> 1.0 </td> <td style="text-align:right;font-weight: bold;color: orange !important;"> 1.00 </td> <td style="text-align:left;font-style: italic;"> useless </td> </tr> </tbody> </table> ??? Mainly taken from [thennt.com](https://www.thennt.com/). Most biomarkers operate in a grey area (for diagnosis). Hard to think of examples with very high sensitivity and specificity – HIV test would be one; histological diagnosis of cancer another. Therefore knowing the pre-test probability is very important! Relatively poor LR- for Covid LFT meant that a negative result did NOT rule out disease if symptoms (i.e. high pre-test probability). A fancy new test (NT-pro-BNP) is not necessarily much better than a boring old one (ECG). ---   ??? Rev Thomas Bayes (1700s).Bayes theorem: the posterior probability depends on the reliability of the test AND the prior probability. Fagan’s nomogram: see [Deeks & Altman, 2004](https://pubmed.ncbi.nlm.nih.gov/15258077/). <br> ### For VTE Unselected: pre-test probability 15% (so with negative d-dimer = post-test probability ~ 3% = not good enough) = **dashed red line**. Low-risk Wells = pre-test probability 5% of DVT/PE in 3 months. Low-risk Wells and negative d-dimer = post-test probabilty 0.9% of DVT/PE in 3 months [Wells et al., NEJM (2003)](https://www.nejm.org/doi/full/10.1056/nejmoa023153) = **solid red line**. LR- d-dimer = ~0.1 ([Stein et al., Ann Int Med 2004](https://pubmed.ncbi.nlm.nih.gov/15096330/)). High-risk Wells = pre-test probabilty ~30% (Wells); LR+ = 2 (Stein). THEREFORE d-dimer can be used to rule out but not to rule in!