{
  "id": "precision-medicine",
  "title": "Precision Medicine",
  "lang": "en",
  "source_lang": "en",
  "summary": "Precision medicine uses biological, clinical, environmental, and lifestyle differences to refine disease classification, prevention, diagnosis, and treatment.",
  "category": "health",
  "subcategory": "medicine",
  "type": "concept",
  "aliases": [
    "Precision health",
    "Stratified medicine",
    "Personalized medicine"
  ],
  "generated_at": "2026-11-02",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/precision-medicine/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/precision-medicine/",
      "json": "https://aiwiki.page/zh/precision-medicine.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Precision medicine is an approach to medicine that uses differences in biology, environment, lifestyle, and clinical circumstances to identify which preventive, diagnostic, or therapeutic strategies are most appropriate for particular groups of people. Its central premise is that patients with the same conventional diagnosis may have different disease mechanisms, risks, or responses to treatment. A molecular finding, physiological measurement, exposure history, or combination of characteristics may therefore change how a condition is classified or managed.\n\nThe approach is closely associated with genetics and genomic sequencing, but it is not restricted to genetic information. Imaging, laboratory measurements, electronic health records, environmental exposures, and patient-reported information can also support more precise classification. Conversely, a highly detailed measurement is not automatically clinically useful: precision medicine requires evidence that the information improves a meaningful decision.\n\nEstablished applications include molecularly targeted cancer treatment, genotype-informed prescribing, and the diagnosis of some inherited disorders. Population-scale sequencing and computational methods have expanded its reach, while also exposing limitations involving uncertain findings, unequal representation, privacy, affordability, and the gap between biological prediction and demonstrated improvements in health."
    },
    {
      "heading": "Conceptual foundations and historical development",
      "text": "Precision medicine extends a longstanding practice of distinguishing clinically relevant subgroups rather than replacing conventional medicine with an entirely new method. Blood-group matching, microbiological identification of pathogens, and classification of diabetes by underlying mechanism all embody the principle that apparently similar clinical situations may require different interventions. What distinguishes contemporary precision medicine is the scale and resolution of the information available, especially measurements of DNA, RNA, proteins, and other molecular features.\n\nThe terms personalized medicine, stratified medicine, and precision medicine overlap but emphasize different aspects. Personalized medicine can suggest care tailored to the individual’s circumstances and preferences, or, more narrowly, molecularly guided treatment. Stratified medicine emphasizes dividing a population into groups with different risks or treatment responses. Precision medicine usually means refining such classifications using reproducible evidence; it does not necessarily imply manufacturing a unique drug for every person. Patient-centered care remains a separate, complementary principle because molecular information alone does not establish a person’s goals or preferences.\n\nSeveral twentieth-century discoveries supplied important foundations. In 1902, Archibald Garrod described alkaptonuria as an inherited biochemical condition, helping establish the concept later called an inborn error of metabolism. In 1956, observations linking primaquine-associated hemolysis to glucose-6-phosphate dehydrogenase deficiency demonstrated how inherited biological variation could explain drug toxicity. The discovery of the Philadelphia chromosome in 1960, followed by identification of the BCR::ABL1 fusion underlying most chronic myeloid leukemia, connected a specific molecular abnormality to a clinically recognizable disease.\n\nThe Human Genome Project, completed in 2003, supplied a reference framework for large-scale human genetic research. Subsequent advances in sequencing, computing, and statistical genetics made it possible to investigate many genes or whole genomes rather than examine one candidate gene at a time. These developments did not make sequence interpretation straightforward: a reference genome describes neither the full diversity of humanity nor the functional significance of every difference between individuals.\n\nA 2011 US National Research Council report, Toward Precision Medicine, helped establish the term in its contemporary sense by proposing a disease taxonomy grounded more directly in molecular mechanisms. The US Precision Medicine Initiative, announced in 2015, and the subsequent All of Us Research Program emphasized longitudinal data and broad participant recruitment. Related national sequencing and biobank programs developed elsewhere. These initiatives expanded research infrastructure, but their existence is distinct from evidence that every proposed precision intervention improves clinical outcomes."
    },
    {
      "heading": "Biological measurements and the interpretation of variation",
      "text": "The basic unit of many precision-medicine decisions is a biomarker: a measurable characteristic associated with a biological process, disease state, exposure, or response to an intervention. Biomarkers can be diagnostic, prognostic, predictive, or used for monitoring. These roles are not interchangeable. A prognostic marker indicates likely outcome regardless of treatment, whereas a predictive marker identifies a difference in the effect of a particular treatment. A marker may perform both functions, but evidence supporting one does not establish the other.\n\nGenomic measurements distinguish inherited, or germline, variation from somatic changes acquired in particular cells. Germline variants can influence inherited disease risk and drug metabolism; somatic variants can drive tumor growth or identify resistant cell populations. Sequencing may examine a selected gene panel, the protein-coding portions of the genome through whole-exome sequencing, or a much larger range of variation through whole-genome sequencing. These methods differ in coverage, cost, turnaround time, and ability to detect structural changes, repeat expansions, or variants in difficult genomic regions.\n\nSequence interpretation is a separate process from variant detection. Laboratories evaluate factors such as population frequency, predicted effects on protein function, segregation within families, functional experiments, and previously reported disease associations. The 2015 American College of Medical Genetics and Genomics and Association for Molecular Pathology framework established five widely used categories for germline variants: pathogenic, likely pathogenic, uncertain significance, likely benign, and benign. A variant of uncertain significance is not equivalent to a disease-causing result. Classifications may change as evidence accumulates.\n\nOther measurements capture processes that DNA sequence alone cannot explain. Gene expression reveals which genes are active in a sampled tissue, while proteomic and metabolic measurements characterize proteins and small molecules. Epigenetic measurements examine regulatory features such as DNA methylation. These layers can reflect disease activity, exposures, treatment, and cell composition. Their interpretation depends strongly on where and when a sample was obtained: blood may not reproduce the molecular state of a tumor or of the central nervous system.\n\nAnalytical limitations remain consequential. Sample contamination, sequencing artifacts, uneven coverage, batch effects, and incorrect computational annotation can produce misleading results. A tumor specimen may contain a mixture of malignant, stromal, and immune cells; a blood specimen may contain age-related hematopoietic clones that complicate interpretation. Clinical validity therefore requires more than a technically sophisticated assay. The relationship between the measured signal, the relevant biological mechanism, and the intended medical decision must be established for the particular setting."
    },
    {
      "heading": "Molecular oncology and treatment selection",
      "text": "Cancer provides some of the clearest examples of precision medicine because tumors arising in the same organ can depend on different molecular mechanisms. Conventional pathology remains essential, but molecular testing can subdivide diagnoses and identify treatment targets. The relevant abnormality may be an activating mutation, a gene fusion, amplification of a signaling receptor, or a defect in DNA repair. A target is clinically useful only when an intervention can affect it and evidence supports benefit in the relevant disease context.\n\nImatinib, approved in the United States for chronic myeloid leukemia in 2001, illustrates mechanism-based treatment. It inhibits the abnormal BCR::ABL1 tyrosine kinase produced by the Philadelphia chromosome rearrangement. In breast cancer, HER2 amplification or overexpression identifies tumors for which HER2-directed treatment can be effective; trastuzumab received US approval in 1998. In selected lung cancers, activating EGFR variants and ALK rearrangements define other treatment-responsive groups. These examples connect a biological alteration, a validated test, and an intervention rather than relying on molecular classification alone.\n\nSome approvals organize treatment across anatomical sites. In 2017, the US Food and Drug Administration approved pembrolizumab for certain previously treated, unresectable or metastatic solid tumors with microsatellite instability-high or mismatch-repair-deficient status, subject to specified clinical conditions. This was the agency’s first cancer drug approval based on a common biomarker rather than the organ of origin. Larotrectinib received a tissue-agnostic approval in 2018 for selected tumors with NTRK gene fusions. Such approvals do not mean that tissue context becomes irrelevant: response can still vary by tumor biology and previous treatment.\n\nMolecular matching is also constrained by tumor evolution. A tumor can contain several genetically distinct subpopulations, and treatment can select resistant clones. Resistance may involve alteration of the drug-binding site, activation of an alternative signaling pathway, or a change in cellular identity. Sampling a single biopsy at one time may therefore provide an incomplete picture. Analysis of circulating tumor DNA, often described as a liquid biopsy, can sometimes detect actionable alterations or emerging resistance without repeated tissue sampling, but sensitivity depends on tumor shedding, disease burden, assay design, and other factors. A negative result does not necessarily exclude a relevant alteration.\n\nPrecision oncology also includes biomarkers for cancer immunotherapy, such as mismatch-repair status, PD-L1 expression, and tumor mutational burden. Their predictive performance is imperfect and varies with assay and clinical setting. Finding a plausible molecular target does not guarantee benefit, and comprehensive profiling may identify no actionable alteration. Even when a match exists, toxicity, resistance, limited drug access, and insufficient comparative evidence can restrict its usefulness."
    },
    {
      "heading": "Inherited disorders and pharmacogenomics",
      "text": "For rare inherited disorders, precision medicine often begins by identifying the cause of an unexplained clinical presentation. A child with developmental abnormalities, seizures, or multisystem disease may undergo genomic testing after, or alongside, detailed clinical assessment. Sequencing affected individuals together with biological parents can help identify newly arising variants and determine inheritance patterns. A molecular diagnosis may clarify prognosis, distinguish conditions with similar appearances, or reveal a mechanism for which a specific treatment exists. Diagnostic yield nevertheless varies substantially with patient selection, phenotype, technology, and previous testing.\n\nSome inherited diseases illustrate why treatment may depend on the particular variant rather than simply the gene involved. In cystic fibrosis, CFTR modulators affect the processing or function of the CFTR protein. Ivacaftor, first approved in the United States in 2012, initially addressed the G551D gating variant. Later combinations expanded treatment to broader groups, including many people with the F508del variant. Eligibility is linked to the mutation and the evidence supporting drug responsiveness; identifying a CFTR variant does not establish that every modulator will work.\n\nAnother example is monogenic diabetes, which can resemble more common forms of diabetes but has a different genetic basis. Certain potassium-channel gene variants causing neonatal diabetes can identify patients who respond to sulfonylureas rather than requiring the same treatment pathway used for autoimmune diabetes. Other genetic subtypes have different implications. The precision element is the connection between molecular mechanism and an established clinical consequence, not merely assigning a rare-disease label.\n\nPharmacogenomics studies how inherited variation influences drug response and toxicity. Variants in CYP2C19 can reduce activation of clopidogrel; CYP2D6 variation can alter the metabolism of several medicines, including conversion of codeine to morphine. TPMT and NUDT15 variants influence the risk of severe myelosuppression from thiopurines. HLA-B57:01 is associated with abacavir hypersensitivity, while HLA-B15:02 is associated with carbamazepine-induced Stevens–Johnson syndrome and toxic epidermal necrolysis, with allele frequencies differing among populations.\n\nGenotype is only one component of drug response. Age, organ function, interacting medicines, adherence, and disease state also influence pharmacokinetics and pharmacodynamics. An enzyme-inhibiting drug can make a genetically normal metabolizer function like a poor metabolizer, a phenomenon called phenoconversion. Likewise, algorithms for warfarin dosing may combine genetic and clinical factors, but their performance depends on population and implementation. The Clinical Pharmacogenetics Implementation Consortium publishes guidance on using available genetic results for prescribing; such guidance does not necessarily establish that population-wide testing is appropriate for every gene–drug pair."
    },
    {
      "heading": "Risk prediction, prevention, and population context",
      "text": "Precision prevention seeks to distinguish people whose probability of disease, or likely benefit from an intervention, differs meaningfully from the population average. Established examples include hereditary cancer syndromes and inherited lipid disorders. Pathogenic variants in BRCA1 or BRCA2 can identify elevated risks of particular cancers, while Lynch syndrome links inherited mismatch-repair defects to colorectal and other cancers. Such findings concern probabilities, not certainties: penetrance varies with age, variant, family history, exposures, and other biological factors.\n\nFor common diseases, many genetic variants individually exert small effects. A polygenic risk score combines weighted associations across multiple variants, often estimated from genome-wide association studies. Such scores can distinguish groups with different average susceptibility to conditions including coronary artery disease, breast cancer, and type 2 diabetes. Their clinical value depends on whether they improve an existing risk model and change decisions in ways that produce more benefit than harm.\n\nRelative risk must be distinguished from absolute risk. A twofold increase has different implications when baseline risk is 1% than when it is 20%. A useful prediction model also needs appropriate calibration: among people assigned a particular risk, the observed frequency of the outcome should be reasonably consistent with that prediction. Good discrimination, meaning the ability to rank higher-risk people above lower-risk people, does not by itself establish good calibration or improved outcomes. Age, competing causes of death, and the time horizon of prediction can substantially change the interpretation.\n\nTransfer between populations is a major limitation. Genetic association studies have historically overrepresented people of European ancestry. Differences in allele frequencies, linkage patterns, environmental exposures, and study recruitment can reduce predictive performance elsewhere. Genetic ancestry is also not interchangeable with socially defined race. Race categories can correlate with social conditions and exposures but do not provide a precise biological account of an individual’s genome. Population labels can consequently conceal important variation within groups.\n\nEnvironmental and social information remains indispensable. Tobacco exposure, occupational hazards, nutrition, housing, and access to care can strongly influence disease occurrence and progression. Social determinants of health may explain risks that genomic analysis cannot. Precision prevention therefore intersects with public health rather than replacing population measures such as vaccination, sanitation, and tobacco control. More detailed risk classification can also generate overdiagnosis, unnecessary follow-up, or false reassurance if thresholds and downstream interventions are not supported by evidence."
    },
    {
      "heading": "Clinical evidence and specialized trial designs",
      "text": "Precision medicine is subject to the same fundamental demands as other forms of evidence-based medicine. An assay needs analytical validity, meaning that it measures its intended target reliably; clinical validity, meaning that the measurement is associated with the relevant clinical state or outcome; and clinical utility, meaning that using it improves decisions or outcomes in its intended setting. These are separate questions. Accurate detection of a mutation does not establish that a proposed treatment will help, and a strong association with disease does not establish that testing asymptomatic people is beneficial.\n\nEvidence for a predictive biomarker is particularly demanding. If patients with a marker respond well to treatment, the marker may simply identify people with a better prognosis. Demonstrating prediction generally requires comparing treatment effects across biomarker-defined groups or studying a validated marker-directed strategy. A randomized controlled trial can compare testing-guided care with standard care, or compare therapies within a molecularly selected population. Randomization remains useful even when the biological rationale is compelling because unmeasured factors and selective enrollment can distort uncontrolled comparisons.\n\nSpecialized designs address the fragmentation of diseases into small subgroups. A basket trial tests a treatment across different diseases sharing a molecular feature, such as the same gene fusion in tumors from different organs. An umbrella trial studies several biomarker-defined treatment groups within one broad disease category. Platform trials can evaluate multiple interventions within an ongoing framework, sometimes adding or dropping arms according to prespecified rules. These structures can improve efficiency, but they do not eliminate the need for appropriate controls, reliable assays, and statistical discipline.\n\nSmall populations create difficulties involving statistical power, multiple comparisons, and uncertain effect estimates. If many biomarkers and treatments are explored, apparently promising findings may arise by chance. Adaptive designs require prespecified methods to protect interpretability. Reliance on a surrogate endpoint, such as tumor shrinkage or a laboratory change, may accelerate evaluation but does not always establish longer survival, better function, or improved quality of life. The strength of evidence depends on how closely the endpoint reflects outcomes important to patients.\n\nObservational data can help describe effectiveness, rare adverse events, and outcomes in groups underrepresented in trials. However, treatment selection, missing data, differences between institutions, and changes in practice complicate causal interpretation. Patients able to obtain comprehensive testing or matched therapies may differ systematically from those who cannot. Therefore, better outcomes among treated patients do not automatically establish that testing caused the improvement. Precision medicine research must evaluate the entire pathway from measurement to treatment, including people for whom testing produces no actionable result."
    },
    {
      "heading": "Data integration, computation, and clinical implementation",
      "text": "Precision medicine depends on connecting information generated in different settings. Molecular laboratories produce variant calls and interpretations; hospitals record diagnoses, imaging, prescriptions, and outcomes; research cohorts add longitudinal questionnaires and environmental measurements. Combining these sources requires consistent identifiers, terminology, units, and metadata. A laboratory value without information about timing, assay method, or treatment context may be difficult to interpret. Disease labels recorded for billing may not correspond exactly to research definitions.\n\nMachine learning can identify patterns across these complex datasets, but predictive performance depends on the data-generating process. A model trained at one institution may exploit local testing practices rather than a stable biological relationship. Information leakage can occur when a training dataset inadvertently includes information unavailable at the intended decision time. Missingness may itself encode clinical behavior: patients with more severe illness often undergo more testing. These issues can make retrospective results appear stronger than prospective performance.\n\nEvaluation therefore includes external validation, calibration, subgroup analysis, and monitoring after deployment. A model may become less reliable when patient populations, laboratory platforms, or treatment patterns change. Integrating molecular data with imaging and clinical variables can improve prediction in some settings, but increasing the number of inputs does not guarantee additional clinical value. Complex models can also be difficult to interpret, and plausible explanations generated after a prediction do not necessarily reveal the true basis of that prediction.\n\nImplementation requires workflows as well as algorithms. Testing must return results within a clinically relevant period, reports must distinguish established findings from exploratory ones, and clinicians need access to appropriate interpretation. Molecular tumor boards can bring together oncology, pathology, genetics, pharmacy, and computational expertise. Genetic counselors help explain inheritance, uncertainty, and implications for relatives. Electronic prescribing systems can incorporate pharmacogenomic alerts, but poorly targeted alerts may contribute to alert fatigue and may be ignored.\n\nA result can remain relevant long after its original collection. Germline variants may affect future prescribing, and previously uncertain findings may later be reclassified. Institutions therefore face practical questions about data retention, responsibility for reinterpretation, and communication of amended reports. Health-economic evaluation includes the cost of confirmatory tests, counseling, specialist review, treatment, and follow-up, not just the price of sequencing. A cheaper assay does not necessarily produce a cheaper care pathway, while an initially expensive test may sometimes prevent repeated investigations. These effects depend on the specific clinical use and cannot be inferred from technological capability alone."
    },
    {
      "heading": "Consent, regulation, privacy, and equitable access",
      "text": "Genomic and linked clinical data raise distinctive ethical issues because they may reveal information about relatives as well as the person tested. Even when direct identifiers are removed, sufficiently detailed genetic information may be reidentified through linkage with other datasets. Data privacy therefore involves more than anonymizing a file. Access controls, security, limits on secondary use, and clear governance arrangements influence whether data sharing is compatible with participant expectations. The usefulness of large datasets must be balanced against risks of disclosure and misuse.\n\nInformed consent must address the purpose and scope of testing, foreseeable limitations, data storage, and possible additional findings. A test undertaken to explain one condition may identify a medically significant predisposition unrelated to the original question. Research findings and clinical results are not interchangeable: a research observation may require confirmation in an appropriately accredited clinical laboratory before it has established clinical standing. Preferences about receiving additional information may also differ, particularly when findings have uncertain significance or limited clinical implications.\n\nRegulation varies by jurisdiction and by product. A companion diagnostic provides information essential to the safe and effective use of a corresponding therapeutic product, such as identifying a required tumor alteration. Its evidentiary requirements are tied to the intended use, specimen type, assay performance, and treatment context. Drug and test approvals do not automatically validate every laboratory method that measures a related signal. Changes in assay thresholds, variant coverage, or sample type can alter which patients are classified as eligible.\n\nLegal protection against genetic discrimination is also incomplete and jurisdiction-specific. In the United States, the Genetic Information Nondiscrimination Act of 2008 provides protections relating to health insurance and employment, with defined scope and exceptions; it does not extend the same protections to life, disability, or long-term-care insurance. European data-protection law treats genetic data as a special category of personal data subject to additional safeguards. Such frameworks govern particular uses of information rather than removing all potential consequences of disclosure.\n\nHealth equity depends on who is represented in research and who can access validated services. Underrepresentation can increase uncertain variant classifications and weaken risk prediction. Testing may also require specialist referral, travel, counseling, or treatments that are unavailable or unaffordable in some health systems. Expanding sequencing without expanding these supporting services can widen existing differences in care. Assessing equitable implementation therefore includes whether diverse populations receive interpretable results, whether indicated interventions are accessible, and whether health benefits are distributed beyond the institutions and populations that generated the original evidence."
    }
  ],
  "links": [
    "medicine",
    "genetics",
    "dna",
    "biomarker",
    "whole-exome-sequencing",
    "whole-genome-sequencing",
    "protein",
    "gene-expression",
    "cancer",
    "pathology",
    "mutation",
    "leukemia",
    "breast-cancer",
    "lung-cancer",
    "biopsy",
    "liquid-biopsy",
    "cancer-immunotherapy",
    "tumor-mutational-burden",
    "cystic-fibrosis",
    "monogenic-diabetes",
    "pharmacogenomics",
    "pharmacokinetics",
    "pharmacodynamics",
    "warfarin",
    "lynch-syndrome",
    "coronary-artery-disease",
    "type-two-diabetes",
    "social-determinants-of-health",
    "public-health",
    "overdiagnosis",
    "evidence-based-medicine",
    "randomized-controlled-trial",
    "surrogate-endpoint",
    "quality-of-life",
    "machine-learning",
    "data-privacy",
    "health-equity",
    "health-system"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}