{
  "id": "cumulative-incidence",
  "title": "Cumulative Incidence",
  "lang": "en",
  "source_lang": "en",
  "summary": "Cumulative incidence is the probability that an initially event-free person experiences a specified event within a defined period.",
  "category": "health",
  "subcategory": "public-health",
  "type": "concept",
  "aliases": [
    "Incidence proportion",
    "Incidence risk"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/cumulative-incidence/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/cumulative-incidence/",
      "json": "https://aiwiki.page/zh/cumulative-incidence.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Cumulative incidence is the proportion of an initially at-risk population that develops a specified disease or experiences another defined event during a stated period. In epidemiology, it expresses the probability of becoming a new case over that interval and is commonly called risk. Despite the word “incidence,” cumulative incidence is a proportion rather than a rate based on person-time. Its interpretation requires a defined population, outcome, starting point, and follow-up period."
    },
    {
      "heading": "Definition and calculation",
      "text": "For a closed group with complete outcome ascertainment over a specified interval, cumulative incidence is calculated as\n\n\\[\\widehat{CI}(t)=\\frac{D(t)}{N_0},\\]\n\nwhere \\(D(t)\\) is the number of people who develop the outcome by time \\(t\\), and \\(N_0\\) is the number initially free of that outcome and at risk of developing it. The hat indicates an estimate from observed data.\n\nFor example, if 1,000 people initially free of a disease are followed for five years and 50 develop it, the estimated five-year cumulative incidence is\n\n\\[\\widehat{CI}(5)=\\frac{50}{1000}=0.05=5\\%.\\]\n\nThis can also be reported as 50 cases per 1,000 initially at-risk people over five years. It does not mean five cases per 100 people each year.\n\nCumulative incidence ranges from 0 to 1, or 0% to 100%. For the same population and outcome, the underlying cumulative probability cannot decrease as the follow-up horizon lengthens. Separate estimates from different populations or study samples need not display this ordering.\n\nDefining the population and event\n\nThe denominator excludes people who already have the outcome at baseline. Eligibility may also depend on whether a person can experience the event: for example, an analysis of a first event excludes people who have already experienced it.\n\nThe numerator counts people experiencing the specified event, not the total number of episodes. A person with several recurrent episodes is counted once when the outcome is occurrence of at least one episode. Analyses of recurrent events require a different specification.\n\nA consistent case definition is essential. The measured outcome may be disease onset, first diagnosis, hospitalization, or another clinical endpoint. These are not interchangeable: diagnosis can occur after biological disease onset, and changes in detection can alter measured incidence."
    },
    {
      "heading": "Distinction from related measures",
      "text": "Incidence rate\n\nIncidence is a broader concept encompassing both cumulative incidence and incidence rates. An incidence rate divides new events by accumulated person-time at risk:\n\n\\[IR=\\frac{D}{\\text{total person-time at risk}}.\\]\n\nAn incidence rate has units such as cases per person-year. Cumulative incidence is dimensionless, although its time horizon must always be stated.\n\nWhen there is no competing event and the event hazard is constant at \\(\\lambda\\), risk over time \\(t\\) is\n\n\\[CI(t)=1-e^{-\\lambda t}.\\]\n\nFor small \\(\\lambda t\\), this is approximately \\(\\lambda t\\). Multiplying a rate by time is therefore an approximation, not a general identity; it becomes inaccurate when the resulting risk is substantial.\n\nPrevalence\n\nPrevalence describes the proportion of a population that has a condition at a specified time or during a specified period, including existing cases. Cumulative incidence concerns new cases among people initially free of the condition.\n\nPrevalence depends not only on the occurrence of new cases but also on disease duration, recovery, survival, and population movement. It cannot generally substitute for cumulative incidence.\n\nHazard\n\nA hazard describes the instantaneous event rate among people who remain event-free immediately before a given time. It is not a probability over a finite interval. Consequently, a hazard ratio is not generally equal to a risk ratio, and neither quantity alone determines absolute cumulative incidence."
    },
    {
      "heading": "Incomplete follow-up and survival analysis",
      "text": "The simple case-count calculation is insufficient when participants have different follow-up durations or their outcomes become unknown before the chosen time horizon.\n\nSurvival analysis accommodates time-to-event data and censoring. Right censoring occurs when a participant is known to be event-free until a particular time but their subsequent event status is not observed.\n\nWhen no competing events are present, cumulative incidence can be estimated from the Kaplan–Meier estimator:\n\n\\[\\widehat{CI}(t)=1-\\widehat{S}(t),\\]\n\nwhere \\(\\widehat{S}(t)\\) estimates the probability of remaining event-free beyond time \\(t\\).\n\nThis approach relies on appropriate assumptions about censoring, particularly that censoring is non-informative for the event process, possibly conditional on measured characteristics. If participants leave follow-up for reasons associated with their future event risk, an unadjusted estimate may be biased. Administrative study closure and loss to follow-up both produce censoring, but their implications for this assumption can differ."
    },
    {
      "heading": "Competing risks",
      "text": "A competing event prevents the event of interest from subsequently occurring. For example, death before a first diagnosis prevents that person from later receiving the diagnosis.\n\nIn this setting, the cause-specific cumulative incidence function is\n\n\\[F_k(t)=P(T\\leq t,\\ J=k),\\]\n\nwhere \\(T\\) is the time to the first event and \\(J\\) identifies its type. This is the probability of experiencing event type \\(k\\) by time \\(t\\), in the presence of the competing events.\n\nIts relationship to the cause-specific hazard is\n\n\\[F_k(t)=\\int_0^t S(u-)\\lambda_k(u)\\,du,\\]\n\nwhere \\(S(u-)\\) is the probability of remaining free of all event types just before time \\(u\\), and \\(\\lambda_k(u)\\) is the hazard for event type \\(k\\).\n\nThe Aalen–Johansen estimator is a standard nonparametric estimator of this probability. Treating competing events as ordinary censoring and calculating one minus Kaplan–Meier generally overestimates the probability of the event of interest in the presence of those competing events. It instead estimates a different quantity under assumptions that remove competing events from the survival calculation.\n\nThis distinction is particularly relevant to long-term studies in which death can preclude the outcome being studied."
    },
    {
      "heading": "Uses and interpretation",
      "text": "Cumulative incidence is widely reported in cohort studies and randomized controlled trials. Risks measured over a common interval can be compared using a risk ratio or a risk difference. The latter expresses the absolute difference in event probabilities between groups.\n\nIn infectious disease outbreak investigations, an attack rate is usually an incidence proportion over the outbreak period, despite its name.\n\nInterpretation depends on several features:\n\n- Time horizon: a one-year risk and a ten-year risk are different quantities.\n- Population: age, baseline health, exposure patterns, and eligibility criteria influence the estimated risk.\n- Ascertainment: diagnostic practices, testing frequency, and case definitions affect which events are detected.\n- Follow-up: incomplete observation can bias estimates if not appropriately addressed.\n- Competing events: their frequency affects the probability of observing the event of interest.\n- Statistical uncertainty: estimates based on few events or limited follow-up can be imprecise.\n\nCrude cumulative incidences from populations with different age structures may not be directly comparable; age standardization can support specified comparisons. An observed difference in cumulative incidence also does not, by itself, establish causation."
    }
  ],
  "links": [
    "epidemiology",
    "probability",
    "case-definition",
    "diagnosis",
    "clinical-endpoint",
    "incidence",
    "person-time",
    "prevalence",
    "relative-risk",
    "kaplan-meier-estimator",
    "cohort-study",
    "randomized-controlled-trial",
    "infectious-disease",
    "age-standardization",
    "causation"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}