Incidence vs Prevalence: Definitions, Formulas and Worked Examples

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-10-30

Incidence is the number of new cases of a condition that arise in a defined population during a specified period of time. Prevalence is the proportion of a defined population that has the condition at a given moment or during a given period, counting every existing case whether it began yesterday or ten years ago.

Those two sentences separate the most confused pair of terms in epidemiology. The confusion has consequences. A hospital that budgets beds using incidence will undercount the patients it actually treats, because people already living with a chronic condition still need care. A researcher who studies only prevalence cannot tell whether a disease is becoming more common or whether patients are simply surviving longer with it. Public health agencies track both numbers for different reasons, and reading either one without the other produces a misleading picture of disease burden.

This guide defines both measures, gives the formulas, walks through arithmetic with real and hypothetical populations, and shows where the two measures disagree.

Incidence vs Prevalence: The Distinction in One Table

The single most useful mental hook is this. Incidence is about becoming. Prevalence is about being.

Prevalence: Measuring Existing Disease

Prevalence asks what fraction of a population carries a condition. Everyone counts, including people diagnosed years ago and people whose disease has already resolved only if the case definition excludes resolution. The denominator is the whole population of interest, not just the healthy portion.

Prevalence has an intuitive appeal because it matches how clinics experience disease. A diabetes clinic sees a caseload, not a stream of new diagnoses. That caseload is a prevalence.

Point Prevalence

Point prevalence measures cases at one instant. The formula is direct.

Point prevalence = number of existing cases at a specified time / total population at that same time

Multiply by 100 to express it as a percentage, or by 100,000 for national reporting conventions.

A cross-sectional study of postmenopausal women in Yaoundé, Cameroon, examined 206 volunteers between November 2024 and June 2025 and measured serum uric acid on a single laboratory platform [1]. The prevalence of hyperuricemia was 37.9%. Applied to a sample of 206, that percentage corresponds to roughly 78 women meeting the hyperuricemia threshold. The arithmetic is 0.379 × 206 = 78.1.

That study is a textbook point prevalence design. Every participant is classified once, no follow-up occurs, and the result describes the sample on the days it was measured. The same study also illustrates what prevalence can do beyond description. Longer menopausal duration of 14 years or more carried an odds ratio of 2.8, overweight or obesity carried an odds ratio of 1.9, and hypertension carried an odds ratio of 2.0 for hyperuricemia [1]. Those are associations with a current state, not predictors of who will develop hyperuricemia later. The distinction matters when a reader tries to convert an odds ratio from a cross-sectional study into a risk statement.

In practice, a "point" is rarely a true instant. Data collection for a single cross-sectional survey can span weeks or months. If the condition changes quickly, the measured prevalence becomes a smear across time rather than a snapshot, and that smear behaves more like a short period prevalence.

Period Prevalence

Period prevalence counts anyone who had the condition at any point during a defined window.

Period prevalence = (cases existing at the start of the period + new cases arising during the period) / population during the period

Some textbooks use the population at the midpoint of the period as the denominator, and others use the average population. Both work as long as the choice is stated.

Period prevalence is always at least as large as the point prevalence at any moment inside the window, because it captures everyone who was ever a case. If 60 residents of a 500-bed nursing home have pressure ulcers on January 1, and 25 more develop ulcers by December 31 while none of the original 60 die or are discharged, the period prevalence is (60 + 25) / 500 = 17%. The point prevalence on January 1 was 60 / 500 = 12%.

That gap between 12% and 17% is the extra information period prevalence provides. It reflects turnover in the case pool, which point prevalence hides.

The Prevalence Formula and Three Traps

Three errors account for most mistakes in published prevalence figures.

Trap one: wrong denominator. Prevalence divides by the whole population. A common slip is to divide by the population at risk, which excludes people who already have the disease. That produces a number that is neither prevalence nor incidence. It is a ratio of existing cases to a group that cannot contain existing cases, which is meaningless.

Trap two: definition dependence. Prevalence moves when the case definition moves, even if the underlying biology is unchanged. A retrospective analysis of 533 consecutive patients with acute ischemic stroke classified atrial cardiopathy using the ARCADIA criteria, which accept any one of three biomarkers [2]. With at least one biomarker, prevalence ranged from 43% in small-vessel occlusion to 97% in cardioembolism. When the definition required at least two biomarkers, the same patients produced prevalences of 8% in small-vessel occlusion, 7% in branch atheromatous disease, 13% in large-artery atherosclerosis, 19% in embolic stroke of undetermined source, and 78% in cardioembolism [2]. Same patients, same measurement, radically different numbers, all because of a threshold. Any prevalence figure is only as meaningful as the definition behind it.

Trap three: prevalence is not a proportion of time. Prevalence is dimensionless. Reporting "12%" is correct. Reporting "12% per year" is not, because prevalence is not a rate. Adding a time unit to a prevalence turns it into something else by accident.

Incidence: Measuring New Disease

Incidence counts transitions. A person who does not have the condition becomes a person who does. The denominator must therefore be people who could make that transition, which means people free of the condition at the start.

Two distinct incidence measures exist, and they answer slightly different questions.

Cumulative Incidence

Cumulative incidence, also called incidence proportion or risk, is the fraction of an at-risk population that develops the condition over a defined period.

Cumulative incidence = new cases during the period / number of people at risk at the start of the period

A nationwide prospective cohort in Kuwait enrolled 1,938 children aged 14 years or younger who were newly diagnosed with type 1 diabetes between 2011 and 2017, then followed them for five years to detect celiac disease confirmed by intestinal biopsy [3]. Four children already had celiac disease at baseline and were excluded, leaving 1,934 at risk. Sixty-five children developed celiac disease during follow-up, giving a cumulative incidence of 65 / 1,934 = 3.36%, reported as 3.4% [3].

That exclusion of four baseline cases is not a technicality. It is the definition operating in practice. Children who already had the outcome could not develop it again, so they do not belong in the denominator. Removing them raised the reported proportion slightly and made the number interpretable as a risk.

Cumulative incidence has a hard requirement. Every person in the denominator must be followed for the entire period. When people drop out, move away, or die from another cause, the simple fraction stops being an accurate risk. Statistical methods such as Kaplan-Meier estimation or competing-risk models exist to repair this, and any cumulative incidence reported from a cohort with substantial loss to follow-up should state which method was used.

Cumulative incidence also inherits the length of its window. A five-year cumulative incidence of 3.4% and a one-year cumulative incidence of 1.0% describe the same disease at different horizons and cannot be compared directly.

Incidence Rate and Person-Time

The incidence rate, also called incidence density, solves the follow-up problem. Instead of counting people, it counts the time each person spent at risk.

Person-time is the sum, across all at-risk individuals, of the time each person remained under observation and free of the outcome. A person followed for three years without disease contributes 3.0 person-years. A person who develops the disease after 18 months contributes 1.5 person-years. A person who leaves the study after six months contributes 0.5 person-years.

Incidence rate = new cases / total person-time at risk

A small worked table makes the arithmetic concrete. Four hypothetical participants are followed for up to three years.

The incidence rate is 2 / 9.0 = 0.222 per person-year. Multiplied by 100, that is 22.2 per 100 person-years. The number reads as follows. If 100 people free of the disease were followed for one year each, about 22 would be expected to develop it, assuming the rate stays constant.

The Kuwaiti cohort reports exactly this kind of figure. In the first year after the type 1 diabetes diagnosis, the incidence rate of celiac disease was 1.35 per 100 person-years, with a 95% confidence interval of 0.92 to 1.99, and the rate generally declined over subsequent years of follow-up [3]. The median time to celiac disease diagnosis was 1.5 years, and 40% of the cases were diagnosed within the first year [3]. That pattern of front-loaded incidence is common when a new diagnosis triggers a screening cascade, and a cumulative incidence figure alone would have hidden it.

The unit matters. "1.35 per 100 person-years" and "1.35 per 100 people per year" are not the same statement, and they diverge badly whenever follow-up is incomplete.

The Incidence Rate Formula in Practice

Incidence rates appear in surveillance reporting at population scale. An analysis of Australian primary care data for adults aged 18 and older who presented with coughing illness between 2009 and 2016 estimated an overall pertussis-attributable rate of 674.3 per 100,000 persons, with a 95% confidence interval of 363.8 to 984.7, equivalent to roughly 939 cases per year [4]. The highest age-specific incidence occurred among adults aged 65 and older, reaching 952.8 per 100,000 in 2010, and the average annual rate was 545.9 per 100,000 among adults aged 45 to 64 and 426.7 per 100,000 among adults aged 65 and older [4].

Rates expressed per 100,000 population per year are the standard currency of national surveillance. They allow comparison between states, years, and age groups of very different sizes.

Cumulative Incidence and Incidence Rate Together

For short windows and rare outcomes, the two incidence measures nearly coincide. The exact conversion from a constant rate to a proportion over a period t is:

Cumulative incidence = 1 - exp(-rate × t)

Take the first-year rate of 1.35 per 100 person-years, which is 0.0135 per person-year. Over one year, the implied cumulative incidence is 1 - exp(-0.0135) = 1.34%. The rate and the proportion agree to two decimal places, which is expected when the outcome is rare over the interval.

Over longer horizons the gap widens. A constant rate of 0.0135 per person-year applied across five years would give 1 - exp(-0.0675) = 6.5%, which is higher than the observed five-year cumulative incidence of 3.4% [3]. The discrepancy is informative rather than contradictory. It shows that the rate was not constant, and the authors' observation that incidence declined after the first year is exactly why the straight-line conversion overshoots.

Worked Examples: One Population, Four Measures

The four measures are easiest to separate when they are computed on the same population. Consider a hypothetical 500-bed long-term care facility during a single calendar year. Sixty residents have pressure ulcers on January 1. During the year, 25 more residents develop a new ulcer, and 20 residents with ulcers at baseline are discharged or die, so the facility census stays at 500.

Step 1, point prevalence on January 1. 60 existing cases / 500 residents = 0.12, or 12%. This is the caseload on New Year's Day.

Step 2, period prevalence for the year. 60 existing cases at the start plus 25 new cases equals 85 residents who had an ulcer at some point / 500 residents = 0.17, or 17%. This is the fraction of the facility touched by the problem during the year.

Step 3, cumulative incidence for the year. The denominator must exclude the 60 residents who already had ulcers, because they cannot develop one as a first event. 25 new cases / 440 at risk = 0.0568, or 5.7%. This is the risk facing an ulcer-free resident at the start of the year.

Step 4, incidence rate for the year. Suppose the 440 at-risk residents accumulated 480 person-years of ulcer-free observation because some were admitted partway through the year. Then 25 / 480 = 0.0521 per person-year, or 5.2 per 100 person-years. This is the speed at which new ulcers appear per unit of exposure time.

Four measures, four numbers, one population. The 12% and the 17% answer "how much disease is here." The 5.7% and the 5.2 per 100 person-years answer "how fast is new disease appearing." A report that presents only one of them leaves half the picture missing.

Note also that the 5.2 per 100 person-years is not identical to the 5.7% cumulative incidence. The difference comes from partial-year exposure. Residents admitted in July contribute half a year of person-time but still count as one person in the cumulative incidence denominator. That mismatch is normal and is the reason rates exist.

How These Measures Are Collected in Practice

The choice of measure usually follows from the study design, and the design in turn follows from the question.

Cross-sectional surveys produce prevalence. Participants are assessed once. The Cameroon study of 206 postmenopausal women is a clean example, with biological measurements performed at a single laboratory and no follow-up [1]. Cross-sectional designs are fast and inexpensive. They cannot establish temporal order, so they support associations but not causal claims about which came first.

Prospective cohorts produce incidence. Participants free of the outcome are followed forward. The Kuwaiti cohort followed 1,934 children for five years and captured new biopsy-confirmed celiac disease as it occurred [3]. A community-based cohort of 211,548 outpatients with heart failure compared incident dementia against matched controls without heart failure, finding time-dependent hazard ratios of 1.42 at five years and 1.31 at 10 years [5]. Hazard ratios of this kind are ratios of incidence between exposure groups, not prevalence ratios, and they require the cohort structure to be estimated.

Linkage registries can produce both. The MIRACLE-S cohort in Stockholm followed 1.4 million individuals with no history of stroke or transient ischemic attack for 12 years, from 2014 to 2025, drawing diagnoses from every level of care [6]. That design yields incidence by neighborhood socioeconomic status. The finding that low-SES neighborhoods had higher stroke incidence, with the largest gap before age 65, is an incidence statement. The parallel finding of higher transient ischemic attack incidence in high-SES neighborhoods was interpreted as possible underdiagnosis in low-SES areas rather than genuinely lower risk [6], which is a reminder that a measured incidence is only as good as the case-finding behind it.

Self-reported surveys trade accuracy for reach. A social-media survey of 282 adults in Malawi asked about malaria cases in the past year and about preventive behaviors, with long-lasting insecticide-treated nets the most used measure at 62.1% and household-level environmental management reported by 94.3% of respondents [7]. Self-reported incidence is subject to recall error and to selection into the survey. It remains useful for generating hypotheses when laboratory confirmation is impractical.

Serosurveys measure prevalence of immunity, which is a different quantity from incidence of infection. An age-structured model of tetanus epidemiology built explicit representations of each tetanus-toxoid-containing vaccine dose and compared model outputs against published seroprevalence data [8]. Seroprevalence is a prevalence. Converting it into an incidence of infection requires assumptions about antibody decay and exposure, which is exactly what a model supplies.

The Arithmetic That Links Incidence and Prevalence

Under steady conditions, prevalence is approximately the product of incidence and average disease duration.

Prevalence ≈ incidence rate × average duration of disease

The logic is a bathtub. Water flows in at a rate set by incidence and drains at a rate set by recovery or death. If the water level is stable, the level equals the inflow rate multiplied by the average time a drop spends in the tub.

A worked case. Suppose a condition has an incidence rate of 2 per 1,000 person-years and an average duration of 5 years after diagnosis. Then the prevalence is approximately 2 × 5 = 10 per 1,000, or 1%. Reversing the arithmetic, a measured prevalence of 1% combined with a measured incidence of 2 per 1,000 person-years implies an average duration near 5 years. Epidemiologists use this back-calculation to estimate duration for conditions where follow-up data are scarce.

The approximation holds when four conditions are met. Incidence must be roughly constant over the period. Average duration must be roughly constant. Migration into and out of the population must be small relative to the case pool. And the disease must not be so common that the at-risk pool shrinks noticeably as cases accumulate.

When those conditions fail, the relationship breaks in predictable ways. If a new treatment extends survival without curing anyone, prevalence rises while incidence stays flat. If a vaccine cuts transmission, incidence falls immediately while prevalence declines slowly, lagging by roughly one duration. Any dashboard that shows only prevalence will report success years after the vaccine started working, or report failure while the vaccine is working fine.

A more exact form for common conditions uses odds rather than proportions. Prevalence odds equals incidence rate multiplied by duration. For rare conditions the odds and the proportion are nearly identical, so the simple multiplication is adequate.

Why the Difference Changes Decisions

Different decisions require different measures, and substituting one for the other produces specific, predictable errors.

Service planning needs prevalence. Bed counts, drug budgets, and clinic staffing scale with how many people currently have the condition. An incidence figure tells a hospital administrator how many new patients to expect, not how many total patients will occupy beds.

Screening programs need prevalence, and prevalence determines predictive value. When prevalence is low, even a highly specific test generates more false positives than true positives. That arithmetic is driven entirely by how many cases exist in the screened population at the time of screening.

Etiology research needs incidence. Risk factors are identified by comparing how fast new cases appear in exposed and unexposed groups. Prevalence comparisons in cross-sectional data are contaminated by differential survival, because people who died from the exposure are no longer available to be counted.

Vaccine and treatment evaluation needs incidence. A vaccine that prevents infection reduces incidence. A drug that prevents death in people already infected reduces neither incidence nor prevalence in the short run, but raises prevalence over time as patients live longer with the disease.

Burden estimation for global health needs both, and definitions change the answer. A synthesis of sepsis epidemiology notes that global and regional burden estimates remain uneven, shaped by differences in case definitions, population structure, healthcare resources, and ascertainment [9]. Two countries with identical true burden can report very different sepsis prevalence and incidence because their diagnostic criteria and counting systems differ.

Common Mistakes and Limitations

Dividing by the wrong denominator. Prevalence uses the whole population. Cumulative incidence uses the population at risk at the start. Swapping them produces numbers that look plausible and are wrong by a factor that depends on how common the disease is.

Dropping the time unit from an incidence rate. "1.35" is not an incidence rate. "1.35 per 100 person-years" is. The confidence interval of 0.92 to 1.99 in the Kuwaiti first-year estimate only makes sense alongside that unit [3].

Treating prevalence as risk. A prevalence of 12% does not mean a randomly chosen person has a 12% chance of developing the condition this year. It means 12% currently have it. For a chronic disease, most of those cases may be years old.

Ignoring definition sensitivity. The atrial cardiopathy analysis showed prevalence moving from 43% to 8% in the same subtype when the biomarker threshold changed from one to two [2]. Any comparison of prevalence across studies must first confirm that the case definitions match.

Underestimating true incidence through incomplete ascertainment. The Australian pertussis analysis concluded that cases in adults may remain undiagnosed a decade after the study period, which makes any measured incidence a floor rather than a ceiling [4]. Surveillance data capture diagnosed cases. Diagnosed cases are a subset of true cases.

Competing risks in older populations. When the population is elderly and mortality is high, many people die before they can develop the outcome of interest. A cohort of heart failure patients is a good illustration, because incident dementia in that group is competing with death from cardiac causes, and the observed incidence depends on how competing events are handled [5].

Recall and self-report bias. A survey asking about malaria episodes in the past year depends on memory, and people who use more preventive measures may also recall episodes differently [7].

Prevalent cohort bias. Recruiting people who currently have a condition selects for survivors. If a treatment is started at the moment of recruitment, patients who died quickly were never eligible, and the treatment looks better than it is. Incidence cohorts avoid this because exposure is measured before outcome.

The relationship is not one-directional. High prevalence can reflect high incidence, long duration, or better survival. A region with high prevalence of a chronic disease may have falling incidence and improving survival. Reporting prevalence alone cannot distinguish these.

Population measures describe groups. Individual patients are assessed by clinicians using their history, examination, and test results, not by a rate calculated for a population.

Quick Review

• Incidence counts new cases over time. Prevalence counts all existing cases at a point or over a period.
• Point prevalence uses the whole population as the denominator. Cumulative incidence uses only the population at risk at the start.
• Incidence rate divides new cases by person-time, not by people, which is why it handles staggered entry and dropout.
• Cumulative incidence and incidence rate converge for rare outcomes over short windows and diverge otherwise.
• Prevalence approximates incidence multiplied by average duration, and that approximation fails when survival or incidence changes.
• Case definitions move prevalence more than most readers expect, so always check the threshold before comparing figures.
• A high prevalence does not mean a high risk of developing the disease.

Frequently Asked Questions

What is the main difference between incidence and prevalence?

Incidence measures new cases that develop during a period, while prevalence measures all existing cases at a point or during a period. Incidence describes how fast people are becoming cases. Prevalence describes how many people are cases right now.

How do you calculate an incidence rate?

Divide the number of new cases by the total person-time at risk, then multiply by a standard constant such as 100 or 100,000. Person-time is the sum of the time each at-risk person spent under observation and free of the outcome, so a rate reported as 1.35 per 100 person-years means roughly 1.35 new cases per 100 years of combined follow-up.

How do you calculate cumulative incidence?

Divide the number of new cases during the period by the number of people at risk at the start of that period. Exclude anyone who already had the condition at baseline, because they cannot develop it as a new event. The result is a proportion between 0 and 1, usually reported as a percentage.

What is the difference between point prevalence and period prevalence?

Point prevalence counts cases at a single moment. Period prevalence counts anyone who had the condition at any time during a defined window, including cases that existed before the window opened. Period prevalence is therefore always at least as large as the point prevalence at any instant inside the window.

Can prevalence be higher than incidence?

Yes, and it usually is for chronic conditions. A disease that lasts for years accumulates cases, so the fraction of the population currently affected can greatly exceed the fraction developing it in any single year. Prevalence is roughly incidence multiplied by average duration, which is why long-lasting diseases have high prevalence even when incidence is low.

Does high prevalence mean a high risk of getting the disease?

No. High prevalence can come from high incidence, from long disease duration, or from improved survival that keeps cases in the population. A condition with modest incidence and very long duration can show high prevalence while the risk of developing it in any given year stays low.

Sources

1. Relationship between hyperuricemia and cardiovascular risk among postmenopausal women in cameroon: a community-based study.
2. Prevalence of Atrial Cardiopathy Biomarkers Across Ischemic Stroke Subtypes.
3. Incidence and Associated Risk Factors of Coeliac Disease in Children Newly Diagnosed With Type 1 Diabetes: A 5-Year Nationwide Prospective Study From Kuwait.
4. For unspecified cough in adults, in Australian primary care, the incidence of pertussis has been largely underestimated.
5. Incidence of Dementia and Subsequent Outcomes in Patients With Heart Failure.
6. Neighbourhood health inequity and the incidence of stroke and transient ischemic attack.
7. Household environmental management practices, community engagement, and self-reported malaria incidence in urban and rural communities of Malawi.
8. An age-structured mathematical model of tetanus epidemiology, immunity, and vaccination across the life course.
9. Sepsis: Epidemiology, Pathophysiology, and Management.

Related Articles

• What Is Prevalence? Definition, Formula and Examples
• Positive and Negative Predictive Value: Formula, Examples and Why Prevalence Matters
• Absolute Risk Reduction vs Relative Risk Reduction: Meaning, Formulas and Examples
• What Is Biostatistics? Methods, Examples and Careers
• How to Calculate a Percentage: Formula and Examples