{
  "id": "liquid-biopsy",
  "title": "Liquid Biopsy",
  "lang": "en",
  "source_lang": "en",
  "summary": "Liquid biopsy analyzes cells or molecular material in body fluids to characterize disease, guide selected cancer treatments, and investigate residual disease or early detection.",
  "category": "health",
  "subcategory": "medicine",
  "type": "concept",
  "aliases": [
    "Fluid biopsy"
  ],
  "generated_at": "2026-11-04",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/liquid-biopsy/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/liquid-biopsy/",
      "json": "https://aiwiki.page/zh/liquid-biopsy.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Liquid biopsy is the analysis of disease-associated cells, nucleic acids, or other biological material obtained from blood or another body fluid. It is most closely associated with cancer, where it can reveal molecular features of a tumor without removing a tissue specimen. Its principal targets include circulating tumor DNA, circulating tumor cells, and molecular cargo carried by extracellular vesicles. The term describes a family of sampling and analytical approaches, not a single test.\n\nIts importance lies in the possibility of obtaining molecular information repeatedly through relatively accessible specimens. In defined clinical settings, plasma testing can identify alterations that determine eligibility for particular cancer drugs. Other applications include investigating treatment response, detecting molecular residual disease after treatment, and screening for previously undiagnosed cancer. These uses have different evidentiary requirements: a test that reliably identifies a treatment-associated mutation in advanced cancer is not necessarily suitable for detecting early cancer in an asymptomatic population.\n\nUnlike a conventional biopsy, most liquid biopsies do not preserve tissue architecture and cannot establish all the features required for a pathological diagnosis. Their value therefore depends on the clinical question, the material sampled, and the assay used. Interpretation requires attention to tumor biology, laboratory handling, statistical performance, and evidence that acting on the result improves patient outcomes."
    },
    {
      "heading": "Biological Basis and Historical Development",
      "text": "Human blood contains extracellular DNA released from cells throughout the body. Much of this cell-free DNA originates from blood-forming cells, although its composition changes with physiological state and disease. Cell death through apoptosis and necrosis contributes to its release; active cellular processes can also contribute extracellular nucleic acids. In cancer, some of the circulating DNA derives from malignant cells and carries alterations associated with the tumor. This tumor-derived fraction may be abundant in some advanced cancers but extremely small in localized disease.\n\nCell-free DNA is extensively fragmented. A prominent fragment population is approximately 166 base pairs long, reflecting the protection afforded by DNA wrapped around a nucleosome and its associated linker region. Tumor-derived fragments often show different size distributions, including enrichment of shorter fragments. Circulating DNA is also cleared relatively rapidly, on a timescale commonly measured in minutes to hours rather than weeks. Consequently, a plasma sample can provide a comparatively recent molecular snapshot, although release and clearance vary among individuals and clinical circumstances.\n\nThe presence of a tumor does not guarantee a measurable blood signal. Shedding depends on tumor size, vascular access, anatomical location, cellular turnover, and treatment. Different metastatic sites can contribute unequally. Disease confined to the central nervous system, for example, may be poorly represented in plasma; material in cerebrospinal fluid can sometimes be more informative. The biological barrier is therefore not merely the analytical sensitivity of the instrument: relevant molecules must first reach the sampled compartment.\n\nThe historical foundations predate modern sequencing. In 1869, the Australian physician Thomas Ashworth described cells resembling those of a cancer in the blood of a patient with metastatic disease. In 1948, Pierre Mandel and Paul Métais reported extracellular nucleic acids in human plasma. Leon and colleagues reported increased serum DNA concentrations in some patients with cancer in 1977. During the 1990s, studies demonstrated that tumor-associated mutations could be detected in circulating DNA, helping establish that the material could carry specific information about a malignancy rather than simply indicate tissue injury.\n\nProgress accelerated with sensitive amplification methods, massively parallel sequencing, and computational error correction. The identification of fetal DNA in maternal plasma by Dennis Lo and colleagues in 1997 provided a separate demonstration that a small, biologically distinct DNA population could be detected against a much larger background. Subsequent developments in prenatal testing and oncology shared important technical foundations. In oncology, the field gradually moved from demonstrating that tumor material was present to determining whether particular measurements could support reproducible clinical decisions."
    },
    {
      "heading": "Analytes and Specimen Types",
      "text": "The most extensively developed cancer liquid biopsies analyze circulating tumor DNA, which is a subset of cell-free DNA. Detectable features include single-nucleotide variants, small insertions and deletions, copy-number alterations, and some gene rearrangements. Other assays examine methylation or fragmentation patterns rather than particular sequence changes. These signals answer different questions: a sequence variant may identify a drug target, whereas a methylation pattern may help distinguish cancer-associated DNA or suggest the tissue from which it originated.\n\nCirculating tumor cells are intact malignant cells that have entered the circulation. They can potentially provide information unavailable from fragmented DNA, including cell morphology, protein expression, RNA profiles, and the coexistence of several features within an individual cell. Their rarity creates a major technical challenge. Some enrichment systems capture cells expressing epithelial surface markers, but tumors can reduce expression of those markers, causing biologically important cells to escape detection. Cell counts also depend on the blood volume examined and the capture method used.\n\nExtracellular vesicles are membrane-enclosed particles released by many cell types. Their cargo can include DNA, RNA, lipids, and proteins. Exosomes are a subset associated with the endosomal pathway, but not every small extracellular vesicle can be assumed to be an exosome. Vesicles are attractive research targets because their membranes can protect molecular cargo and because their composition may reflect the state of the originating cell. However, separating disease-derived vesicles from the much larger population released by normal cells remains difficult, and isolation methods can yield substantially different preparations.\n\nAdditional targets include circulating RNA, microRNAs, proteins, and metabolites. Protein measurements are not new to medicine: prostate-specific antigen, for example, long predates contemporary genomic liquid biopsy. Whether such measurements are included under the term depends on context. Some multi-analyte cancer tests combine DNA alterations with protein concentrations or other features, attempting to capture complementary signals. Such combinations require validation as complete assays; the established performance of one component does not establish the performance of the combined classifier.\n\nBlood is the most common specimen, usually with plasma preferred for cell-free DNA analysis. Other fluids may provide a more anatomically focused sample. Urine can contain material shed by urinary tract tumors, while pleural fluid, ascites, saliva, and cerebrospinal fluid have disease-specific applications. Fluid collection is not always noninvasive: lumbar puncture, thoracentesis, and abdominal drainage entail procedures and associated risks. A liquid biopsy is therefore better defined by its source material than by a claim that it is invariably simple or risk-free. The appropriate specimen depends on where the disease is located and what biological signal the assay is designed to measure."
    },
    {
      "heading": "Laboratory Workflow and Analytical Technologies",
      "text": "The analytical process begins before DNA reaches a sequencing instrument. Collection tubes, time to plasma separation, storage temperature, centrifugation, transport, and freeze–thaw exposure can alter the result. If white blood cells break down after collection, they release genomic DNA that dilutes the tumor-derived fraction. Laboratories therefore use validated processing intervals for standard anticoagulant tubes or appropriately validated stabilizing tubes. Serum often contains more background DNA because clotting and associated cellular disruption release additional material.\n\nAfter plasma separation, nucleic acids are extracted and their quantity and suitability assessed. Input volume matters because the number of informative molecules is finite. Approximately 3.3 picograms of DNA correspond to one haploid human genome equivalent. Thus, 10 nanograms of DNA represents only about 3,000 haploid genome equivalents before extraction and library-preparation losses are considered. At a variant allele fraction of 0.1%, the expected number of variant molecules in that input is only about three. Repeatedly sequencing the same original molecules cannot compensate for molecules that were never collected.\n\nPolymerase chain reaction methods can sensitively interrogate predefined alterations. Digital PCR partitions a sample into many individual reactions, allowing positive and negative partitions to be counted and interpreted statistically. It is particularly useful when a small number of known variants are the targets. Its principal trade-off is limited breadth: a highly sensitive assay for one mutation does not reveal unrelated resistance mechanisms or unexpected genomic alterations.\n\nNext-generation sequencing can examine many genes or genomic regions simultaneously. Targeted panels concentrate sequencing effort on selected regions, whereas broader approaches examine larger portions of the genome. Unique molecular identifiers label original DNA fragments so that multiple reads from the same molecule can be grouped. Consensus methods suppress errors introduced during amplification or sequencing. Duplex approaches compare information from both strands of an original DNA molecule, providing further error discrimination, although practical sensitivity still depends on molecular recovery and input.\n\nOther technologies interrogate methylation, fragment length, fragment endpoints, or patterns associated with nucleosome positioning. DNA methylation can preserve information about cell identity, while fragmentation profiles can reflect chromatin organization and nuclease activity. Computational models integrate these features, sometimes alongside sequence variants and clinical variables. A complete assay therefore includes not only sample processing and measurement but also a defined algorithm, classification threshold, and reporting system.\n\nAnalytical validation establishes characteristics such as accuracy, precision, limit of detection, interference, and reproducibility. These properties must be evaluated for the relevant alteration classes and specimen conditions. A test may perform well for a common point mutation yet less well for a fusion or a low-level copy-number gain. Quality control consequently extends from specimen adequacy to bioinformatics, with particular attention to whether the laboratory can distinguish a genuine low-frequency biological signal from contamination or technical error."
    },
    {
      "heading": "Molecular Profiling and Treatment Selection in Advanced Cancer",
      "text": "The most established oncology application is molecular profiling of advanced cancers when particular alterations can guide targeted cancer therapy. A plasma assay may provide clinically relevant information when tissue is insufficient, obtaining another biopsy would be difficult, or a recent molecular assessment is needed. It can also be performed alongside tissue testing. The key issue is not whether plasma is universally better than tissue, but whether the assay can reliably identify the alterations relevant to a specific treatment decision.\n\nIn advanced non-small-cell lung cancer, plasma testing can identify actionable alterations in genes such as EGFR, ALK, ROS1, BRAF, MET, RET, and others, depending on the assay. In 2016, the United States Food and Drug Administration approved a plasma-based EGFR test associated with erlotinib treatment, an important regulatory milestone. Larger plasma sequencing panels subsequently received approvals for specified uses. These authorizations concern defined assays, alteration types, cancers, and treatments rather than the general concept of liquid biopsy.\n\nAn important distinction is between a positive result and an uninformative negative result. Detection of a well-validated actionable alteration can support an associated treatment indication. Failure to detect it may mean that the tumor lacks the alteration, but it may also mean that too little tumor DNA entered the sample. Consequently, a plasma-negative result does not necessarily establish molecular absence. Tissue analysis can supply information that plasma missed, while also establishing tumor type, architecture, and other pathological characteristics.\n\nLiquid biopsy can also reveal mechanisms of acquired resistance. During EGFR-directed treatment, for example, resistant subclones may acquire additional EGFR alterations or activate bypass pathways such as MET amplification. Different mechanisms have different therapeutic implications, and the clinical relevance of a particular finding depends on the treatment setting and available evidence. Some changes, including transformation to small-cell morphology, cannot be established from a plasma sequence profile alone and require tissue examination.\n\nPlasma can integrate DNA released from several tumor sites, potentially exposing heterogeneity that a single tissue core misses. However, it does not represent every site equally. A strongly shedding lesion may dominate the signal, and assigning a detected alteration to a specific metastasis is generally difficult. Variant allele fractions also cannot be translated directly into the proportion of tumor cells carrying an alteration without accounting for tumor fraction, copy number, and other biological factors.\n\nA plasma test linked to a drug can function as a companion diagnostic. Clinical implementation requires evidence connecting the detected biomarker with the relevant treatment outcome, not merely evidence that the laboratory can measure it. This distinction is especially important for broad genomic panels, which may report many alterations while only a subset has a validated role in that cancer and clinical setting."
    },
    {
      "heading": "Residual Disease, Recurrence, and Treatment Monitoring",
      "text": "After apparently curative surgery or other definitive treatment, imaging may show no remaining cancer even though small numbers of malignant cells persist. Molecular residual disease refers to disease-associated molecular material detectable below the level of conventional clinical assessment. In solid tumors, postoperative circulating tumor DNA is an extensively studied marker of this state. A positive result often identifies a group with substantially higher recurrence risk, but the magnitude and implications vary by cancer type, stage, sampling schedule, and assay.\n\nTwo broad strategies are used. Tumor-informed assays first characterize the patient’s tumor and then track selected alterations in plasma. Tracking several personalized variants can improve confidence that a low-level signal originates from that tumor. Tumor-naive, or tissue-free, assays do not require prior tumor sequencing and instead examine predefined genomic features, methylation, or combined signals. These approaches differ in turnaround time, tissue requirements, analytical design, and susceptibility to particular sources of error.\n\nTiming matters. Surgery can transiently increase normal cell-free DNA, diluting the tumor signal. Treatment can also change DNA release independently of the amount of viable cancer. Sampling too soon, testing only once, or comparing measurements obtained under different conditions can therefore complicate interpretation. Serial measurements may detect disease that was initially below the detection limit, but a later positive result is not equivalent to proof that an earlier negative test was analytically defective.\n\nThe DYNAMIC trial provided an important example of clinical utility research in stage II colorectal cancer. Published in 2022, this randomized trial included 455 patients and compared a circulating-tumor-DNA-guided adjuvant treatment strategy with standard management. Chemotherapy was used in 15% of the guided group and 28% of the standard-management group. Two-year recurrence-free survival was 93.5% and 92.4%, respectively, meeting the trial’s prespecified noninferiority criterion. These findings concerned a defined stage II colon cancer strategy and did not establish that all residual-disease assays or all cancer settings are interchangeable.\n\nA central distinction is between prognostic and predictive information. A prognostic marker identifies patients with different expected outcomes. A predictive marker identifies differences in benefit from a particular intervention. Strong evidence that postoperative positivity predicts recurrence does not, by itself, show which treatment will prevent that recurrence or whether intensifying treatment improves survival. Likewise, a negative result does not automatically establish that established adjuvant treatment can safely be omitted.\n\nDuring treatment of advanced disease, falling circulating tumor DNA can correlate with response, and rising levels can precede radiological progression. Nevertheless, a molecular change is not automatically a validated surrogate endpoint for survival or quality of life. Trials of biomarker-directed treatment changes are needed to establish whether acting earlier improves outcomes rather than simply identifying progression sooner."
    },
    {
      "heading": "Early Detection and Population Screening",
      "text": "Detecting cancer before symptoms develop presents a harder analytical problem than profiling an established advanced malignancy. Small tumors may release very little DNA, and some early lesions shed little detectable material regardless of assay depth. A screening test must distinguish a very weak cancer signal from normal biological variation in a population where most people do not have cancer. It must also address what happens after a positive result, including localization, diagnostic procedures, and the possibility that no malignancy will be found.\n\nMulti-cancer early detection tests attempt to identify signals associated with several cancer types from one specimen. Some analyze methylation patterns, others sequence alterations, proteins, fragmentation, or combinations of these features. Many also estimate the tissue of origin to guide subsequent investigations. Tissue-of-origin prediction is a separate performance characteristic from cancer detection: a test can correctly identify a cancer-associated signal while incorrectly predicting its location.\n\nSensitivity commonly differs by cancer type and stage. A single overall sensitivity estimate may conceal relatively high detection of advanced disease and much lower detection of stage I disease. Studies that compare previously diagnosed cancer cases with selected cancer-free controls can establish feasibility but may not reproduce performance in a screening population. Prospective studies must evaluate people before diagnosis and include appropriate ascertainment of cancers that appear after an initially negative result.\n\nBlood-based screening for a single cancer has also advanced. In the ECLIPSE study, published in 2024, a cell-free-DNA blood test for colorectal cancer was evaluated in 7,861 people eligible for screening, using colonoscopy as the reference. Sensitivity for colorectal cancer was 83.1%, and specificity for advanced neoplasia was 89.6%. Sensitivity for advanced precancerous lesions was 13.2%. The FDA approved the Shield blood test in 2024 for colorectal cancer screening in average-risk adults aged 45 years or older. These results illustrate why detecting established cancer and detecting a removable precursor lesion are distinct capabilities.\n\nThe appropriate endpoint for cancer screening is not simply the number of cancers found. Screening can produce lead-time bias, in which survival measured from diagnosis appears longer even though death occurs at the same time. It can also cause overdiagnosis, identifying disease that would not have become clinically important during the person’s lifetime. Additional testing can involve imaging, endoscopy, biopsy, and procedure-related harms.\n\nConsequently, clinical evaluation considers cancer-specific mortality, late-stage cancer incidence, false-positive investigations, diagnostic resolution, and patient-reported outcomes. A shift toward earlier stages can be informative but does not automatically prove a mortality benefit. Evidence from established screening methods also cannot simply be transferred to a new blood test, because different methods detect different biological features and generate different downstream pathways."
    },
    {
      "heading": "Interpretation, False Results, and Statistical Limits",
      "text": "The sensitivity and specificity of a liquid biopsy are properties of a defined test used in a defined population. Analytical sensitivity concerns the smallest signal measurable under specified conditions. Clinical sensitivity concerns the proportion of people with the target condition who test positive. An assay can detect extremely low variant fractions in prepared laboratory samples yet have limited clinical sensitivity for small tumors that release few molecules into blood.\n\nPositive predictive value depends strongly on disease prevalence. If sensitivity is \\(Se\\), specificity is \\(Sp\\), and prevalence is \\(p\\), then\n\n\\[\\mathrm{PPV}=\\frac{Se\\,p}{Se\\,p+(1-Sp)(1-p)}.\\]\n\nFor an illustrative test with 80% sensitivity and 99% specificity in a population with 1% prevalence, the positive predictive value is approximately 45%. Among 10,000 people, about 80 true-positive and 99 false-positive results would be expected. This hypothetical calculation is not a performance estimate for any particular commercial assay; it shows why apparently high specificity may still leave substantial diagnostic uncertainty in low-prevalence screening.\n\nA major biological source of misleading results is clonal hematopoiesis. Blood-forming stem cells can acquire mutations and produce expanded blood-cell populations without the person having a solid tumor. DNA from these cells enters plasma. Variants in genes such as DNMT3A, TET2, and ASXL1 are well-recognized examples, but variants in genes also important in solid tumors, including TP53, can create more difficult attribution problems. Sequencing paired white blood cells can help distinguish blood-cell-derived alterations from tumor-derived ones.\n\nInherited variants, technical artifacts, and DNA released by nonmalignant tissues can also complicate interpretation. An allele fraction near 50% may raise the possibility of an inherited heterozygous variant, but allele fraction alone is insufficient to establish germline status. Confirmation in an appropriate non-tumor specimen and a separate interpretive framework may be required. The relevance of genetic counseling depends on the finding and clinical context.\n\nFalse-negative results arise from low shedding, inadequate input, incomplete assay coverage, or limitations in detecting particular alteration classes. Biological barriers can be as important as sequencing depth. In a simplified model where independently sampled target molecules follow a Poisson distribution with mean \\(\\lambda\\), the probability of sampling none is \\(e^{-\\lambda}\\). When the expected target count is very small, some negative specimens are unavoidable even with technically perfect measurement.\n\nA report must therefore distinguish “not detected” from “not present.” It should identify what was measured, the relevant detection limits, and significant limitations of the specimen or assay. Discordance with imaging, tissue pathology, or the clinical course requires interpretation rather than an assumption that the newest or most sensitive technology must be correct. Each method observes a different aspect of disease."
    },
    {
      "heading": "Related Applications Beyond Solid-Tumor Oncology",
      "text": "The general principle of recovering disease information from circulating material extends beyond solid tumors, although terminology differs across specialties. Some applications are usually described as cell-free-DNA testing, molecular monitoring, or noninvasive testing rather than liquid biopsy. The shared technical foundation does not make their biological interpretation identical: the clinically relevant signal may originate from a fetus, an organ donor, a microorganism, or a malignant blood-cell population.\n\nIn pregnancy, maternal plasma contains DNA derived predominantly from placental trophoblasts that represents the fetoplacental genome. Noninvasive prenatal screening uses this material to assess the probability of selected chromosomal conditions, especially common trisomies. It remains a screening approach rather than a direct examination of fetal cells. Confined placental mosaicism, a vanished twin, maternal chromosomal findings, and other biological factors can produce discordant results. The placenta’s contribution illustrates how circulating DNA can reflect a biologically related tissue rather than the exact tissue whose condition is being inferred.\n\nAfter organ transplantation, donor-derived cell-free DNA can be distinguished from recipient DNA using genetic differences. An increased donor-derived fraction or concentration may indicate injury to the transplanted organ. Such injury can occur during transplant rejection, but the signal is not uniquely specific to rejection and can rise in other settings. Fractional measurements also depend on the amount of recipient-derived background DNA. Interpretation therefore differs between a percentage of donor DNA and its absolute concentration.\n\nIn infectious disease, microbial cell-free DNA sequencing can detect DNA fragments from organisms in plasma without requiring growth in culture. This can be useful in selected diagnostic investigations, but detection alone does not establish that a microorganism is responsible for a patient’s illness. Contamination, transient DNAemia, latent or incidental infection, antimicrobial exposure, and the anatomical site of infection affect the result. DNA-based methods also do not automatically provide the same information as culture about organism viability or antimicrobial susceptibility.\n\nHematological cancers occupy a distinct position because blood and bone marrow are often direct sites of disease. Measuring leukemic cells, characteristic transcripts, or rearranged immune-receptor sequences can provide highly sensitive residual-disease assessment. Molecular monitoring in chronic myeloid leukemia, for example, has long tracked BCR::ABL1 transcripts. Such established approaches are not always labeled liquid biopsy, despite their reliance on accessible fluids and molecular detection.\n\nResearch also investigates circulating DNA methylation as a measure of tissue-specific injury in nonmalignant disease. The underlying challenge is attribution: a rise in DNA from a particular tissue may indicate cell injury without identifying its cause. Moving from recognition of a tissue signal to a reliable diagnosis requires disease-specific validation and comparison with existing tests, rather than assuming that a broadly measurable marker has equal usefulness in every clinical setting."
    },
    {
      "heading": "Clinical Validation, Regulation, and Implementation",
      "text": "Evaluation of a liquid biopsy separates analytical validity, clinical validity, and clinical utility. Analytical validity asks whether the assay measures its intended target accurately and reproducibly. Clinical validity asks how the result relates to the disease state or outcome of interest. Clinical utility asks whether using the result improves decisions and outcomes in the intended setting. These questions require different studies, and success at an earlier level does not establish success at a later one.\n\nA well-designed clinical trial specifies the population, assay, sampling schedule, treatment pathway, and outcomes before testing begins. Prospective observational studies can establish prognostic relationships and characterize diagnostic performance. A randomized controlled trial can determine whether a biomarker-guided strategy is better than, or acceptably similar to, standard management for a defined outcome. The intervention being evaluated is often an entire strategy, including what happens after positive, negative, or indeterminate results, rather than the laboratory assay alone.\n\nComputational components require comparable discipline. Training and evaluation data should be appropriately separated, and performance should be tested in populations distinct from those used to develop the classifier. Differences in age, smoking, comorbidity, specimen handling, and recruitment setting can create signals unrelated to cancer. A model trained on selected cases and healthy controls may exploit these differences and then perform less well in routine practice. Changes to an algorithm, threshold, or molecular panel can also require renewed validation.\n\nRegulatory authorization is specific to a device and its intended use. A test authorized as a companion diagnostic in advanced cancer does not thereby acquire authorization or evidence for population screening. Likewise, performance claims made for one version of a test cannot automatically be assigned to another. Laboratory accreditation and proficiency testing support reliable operation, but they do not independently demonstrate that every reported biomarker has clinical utility.\n\nImplementation also involves turnaround time, failure rates, access to confirmatory procedures, and the consequences of uncertain findings. A blood draw may be straightforward while the diagnostic pathway it initiates is costly or invasive. Cost-effectiveness analysis therefore includes downstream investigations and treatments, not only the price of the assay. Unequal access to sequencing, specialist interpretation, and follow-up care can affect how benefits and harms are distributed across a health system.\n\nGenomic data create additional responsibilities concerning data privacy, secondary use, and potentially inherited findings. Consent and reporting policies need to distinguish the clinical question being addressed from incidental information an assay might reveal. For longitudinal testing, laboratories and clinical teams also require clear procedures for comparing results across time, handling assay changes, and communicating uncertainty. These operational features are part of the test’s real-world performance because a molecular signal becomes clinically consequential only through the actions it triggers."
    }
  ],
  "links": [
    "cancer",
    "biopsy",
    "dna",
    "central-nervous-system",
    "cerebrospinal-fluid",
    "polymerase-chain-reaction",
    "next-generation-sequencing",
    "dna-methylation",
    "algorithm",
    "targeted-cancer-therapy",
    "lung-cancer",
    "surgery",
    "colorectal-cancer",
    "surrogate-endpoint",
    "cancer-screening",
    "overdiagnosis",
    "diagnostic-sensitivity-and-specificity",
    "prevalence",
    "pregnancy",
    "organ-transplantation",
    "transplant-rejection",
    "infectious-disease",
    "bone-marrow",
    "chronic-myeloid-leukemia",
    "clinical-trial",
    "randomized-controlled-trial",
    "health-system",
    "data-privacy"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}