{
  "id": "positive-predictive-value",
  "title": "Positive predictive value",
  "lang": "en",
  "source_lang": "en",
  "summary": "Positive predictive value is the probability that a person with a positive test result truly has the condition being tested for, within a specified population.",
  "category": "health",
  "subcategory": "medicine",
  "type": "concept",
  "aliases": [
    "PPV",
    "Predictive value of a positive test",
    "Positive predictive probability"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/positive-predictive-value/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/positive-predictive-value/",
      "json": "https://aiwiki.page/zh/positive-predictive-value.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Positive predictive value (PPV) is the proportion of positive test results that correctly identify the presence of a specified condition. In diagnosis, it answers the question: “Among people who test positive, how many actually have the condition?” PPV depends both on the test’s performance and on how common the condition is in the population being tested. It is therefore not a fixed property of a test that applies equally in every setting."
    },
    {
      "heading": "Definition and calculation",
      "text": "For a test that classifies results as positive or negative, results can be compared with an established reference standard:\n\nTest result | Condition present | Condition absent\nPositive | True positive (TP) | False positive (FP)\nNegative | False negative (FN) | True negative (TN)\n\nPositive predictive value is calculated as:\n\n\\[\\mathrm{PPV}=\\frac{\\mathrm{TP}}{\\mathrm{TP}+\\mathrm{FP}}.\\]\n\nThe denominator includes all positive results, not all people tested and not all people with the condition. PPV is usually expressed as a percentage. If there are no positive results, this sample proportion is undefined.\n\nIn probability notation, if \\(D\\) denotes the presence of the condition and \\(T^+\\) a positive result:\n\n\\[\\mathrm{PPV}=P(D\\mid T^+).\\]\n\nThis is a conditional probability: the probability of the condition given a positive result. It differs from the probability of a positive result given the condition."
    },
    {
      "heading": "Relationship to sensitivity, specificity, and prevalence",
      "text": "Sensitivity and specificity describe test performance conditional on whether the condition is present:\n\n- Sensitivity is the proportion of people with the condition who test positive.\n- Specificity is the proportion of people without the condition who test negative.\n\nPPV reverses the conditioning: it starts with the positive result and asks whether the condition is present.\n\nUsing Bayes’ theorem, PPV can be expressed as:\n\n\\[\\mathrm{PPV} = \\frac{\\mathrm{Se}\\,p} {\\mathrm{Se}\\,p+(1-\\mathrm{Sp})(1-p)},\\]\n\nwhere \\(\\mathrm{Se}\\) is sensitivity, \\(\\mathrm{Sp}\\) is specificity, and \\(p\\) is the prevalence of the condition in the tested population.\n\nFor fixed sensitivity and specificity, PPV generally increases as prevalence increases. When a condition is uncommon, people without it greatly outnumber those with it. Even a small false-positive rate can then produce more false positives than true positives.\n\nPrevalence is a population-level starting probability. For an individual, the relevant probability before testing may also depend on symptoms, medical history, exposures, and previous findings. Consequently, a PPV measured in one study is not automatically the appropriate probability for every person with a positive result."
    },
    {
      "heading": "Numerical example",
      "text": "Consider a hypothetical test with 90% sensitivity and 95% specificity, applied to 10,000 people in a population where the condition has a prevalence of 1%.\n\nOf the 100 people with the condition:\n\n- 90 test positive.\n- 10 test negative.\n\nOf the 9,900 people without the condition:\n\n- 495 test positive.\n- 9,405 test negative.\n\nThe PPV is therefore:\n\n\\[\\mathrm{PPV} = \\frac{90}{90+495} \\approx 15.4\\%.\\]\n\nThus, approximately 15% of positive results identify people who actually have the condition, despite the test having relatively high sensitivity and specificity.\n\nIf the same sensitivity and specificity applied in a population with 10% prevalence, the PPV would be:\n\n\\[\\mathrm{PPV} = \\frac{0.90\\times0.10} {0.90\\times0.10+0.05\\times0.90} \\approx 66.7\\%.\\]\n\nThe difference results from the changed population prevalence, not an improvement in the test itself. In practice, sensitivity and specificity may also vary between populations because of differences in disease severity, patient characteristics, or testing procedures."
    },
    {
      "heading": "Related measures",
      "text": "Negative predictive value\n\nNegative predictive value is the probability that the condition is absent given a negative test result:\n\n\\[\\mathrm{NPV}=\\frac{\\mathrm{TN}}{\\mathrm{TN}+\\mathrm{FN}}.\\]\n\nFor fixed sensitivity and specificity, NPV generally increases as prevalence decreases. PPV and NPV therefore describe different aspects of interpreting test results.\n\nFalse-positive rate and false discovery proportion\n\nThe false-positive rate is:\n\n\\[\\frac{\\mathrm{FP}}{\\mathrm{FP}+\\mathrm{TN}}=1-\\mathrm{specificity}.\\]\n\nIts denominator consists of people without the condition. By contrast, the proportion of positive results that are false is:\n\n\\[\\frac{\\mathrm{FP}}{\\mathrm{TP}+\\mathrm{FP}}=1-\\mathrm{PPV}.\\]\n\nConfusing these quantities can substantially misrepresent what a positive result means.\n\nPrecision in classification\n\nIn machine learning and information retrieval, the same ratio of true positives to all predicted positives is commonly called precision. Its interpretation depends on how the positive class is defined and how representative the evaluation data are of actual use."
    },
    {
      "heading": "Applications",
      "text": "PPV is used in evaluating laboratory tests, imaging findings, diagnostic criteria, and screening programmes. It helps describe how often positive findings correspond to the target condition and the potential volume of additional investigations generated by testing.\n\nScreening populations often have lower condition prevalence than populations referred for diagnostic evaluation. A test may consequently have a lower PPV in screening than in a specialist clinic, even when its sensitivity and specificity are similar. This distinction is particularly relevant to cancer screening.\n\nPPV is also used in public health surveillance to describe the proportion of reported or identified cases that actually satisfy the intended case definition."
    },
    {
      "heading": "Estimation and limitations",
      "text": "Study population and design\n\nA PPV estimate applies directly to the population and testing circumstances represented by the data. Referral patterns, eligibility criteria, disease spectrum, and prior testing can affect its applicability elsewhere.\n\nIn a conventional case–control study, investigators select the numbers of participants with and without the condition. The resulting disease proportion does not generally represent population prevalence, so calculating PPV directly from those selected counts does not ordinarily provide a population-applicable estimate.\n\nReference standards and verification\n\nTrue and false results are classified using a reference standard, which may itself be imperfect. Estimates can be biased if only some participants receive definitive verification, especially when verification depends on the initial test result. A reference standard that incorporates the test being evaluated can also distort apparent performance.\n\nSampling uncertainty\n\nAn observed PPV is an estimate. Its precision depends particularly on the number of positive results available for analysis. A confidence interval communicates sampling uncertainty, but does not by itself account for systematic bias or differences between the study population and the population of intended use.\n\nThresholds and testing sequences\n\nFor tests producing numerical scores, the threshold defining a positive result affects sensitivity, specificity, and PPV. A stricter threshold may increase PPV while missing more affected individuals; the actual trade-off depends on the distributions of results.\n\nA sequence of screening and confirmatory tests has its own predictive value. Its performance depends on the population entering each stage and on whether the tests’ errors are related; the performance of the combined process cannot generally be inferred by simply multiplying individual PPVs.\n\nClinical significance\n\nPPV measures correctness relative to a defined target condition, not whether testing improves health outcomes. It does not independently establish the benefits, harms, or costs of a testing programme.\n\nA true-positive result can also represent overdiagnosis: correct detection of a condition that would not have caused symptoms or harm during the person’s lifetime. False positives and overdiagnosis are therefore distinct, and a high PPV does not exclude overdiagnosis."
    }
  ],
  "links": [
    "diagnosis",
    "probability",
    "diagnostic-sensitivity-and-specificity",
    "bayes-theorem",
    "prevalence",
    "medical-history",
    "machine-learning",
    "laboratory-test",
    "medical-screening",
    "cancer-screening",
    "public-health-surveillance",
    "case-definition",
    "case-control-study",
    "confidence-interval",
    "overdiagnosis"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}