How to read a prevalence estimate
How to read a prevalence estimate: what the number counts, why it differs from incidence, and why autism surveillance figures vary by site and year.
By Manouchehr Hessabi, MD, MPH
A prevalence estimate tells you what share of a defined population has a condition at a defined time. It counts everyone who has the condition, whether they were identified last month or years ago, so it describes how common the condition is, which epidemiologists call burden. It does not, by itself, describe risk, and most misreadings of prevalence come from treating it as if it did.
Prevalence figures travel further than almost any other number in public health. They often arrive stripped of the details that give them meaning. This explainer walks through those details, using the most recent national autism surveillance report from the Centers for Disease Control and Prevention (CDC) as a worked example.
What does a prevalence estimate actually tell you?
The CDC's introductory epidemiology course defines prevalence as the proportion of persons in a population who have a particular disease or attribute at a specified point in time, and notes that it includes all cases, both new and preexisting (CDC, Principles of Epidemiology in Public Health Practice, Lesson 3, Section 2).
That definition contains three choices, and every prevalence figure depends on all three:
- Who. The population the estimate describes: an age group, a geographic area, a set of schools or clinics.
- When. A single moment (point prevalence) or an interval of time (period prevalence). The same CDC lesson defines point prevalence as prevalence measured at a particular point in time, and period prevalence as prevalence measured over an interval.
- What counts as a case. The case definition, and the method used to find the people who meet it.
A prevalence estimate that leaves any of these unstated is not yet interpretable. Two figures that differ on any of them are not directly comparable, even when they describe the same condition.
How is prevalence different from incidence?
Incidence is the other basic measure of how often a condition occurs. The CDC lesson defines incidence as the occurrence of new cases of disease or injury in a population over a specified period of time. Incidence asks how many people developed the condition during the period. Prevalence asks how many people have it at the time of measurement.
The two are linked through duration. The CDC lesson states that prevalence is based on both incidence and duration of illness, so a high prevalence can reflect many new cases, cases that last a long time, or both. A condition that lasts for life accumulates in a population: even if the number of new cases each year were perfectly steady, the number of people living with it would grow until the population reached a balance between those entering and those leaving.
This is why a change in prevalence, on its own, cannot tell you whether the underlying rate at which the condition arises has changed. Something else may have shifted instead.
Why does case finding shape the number?
For many conditions, a prevalence estimate does not count everyone who has the condition. It counts everyone who has been identified as having it. The gap between those two quantities depends on how cases are found.
The CDC's Autism and Developmental Disabilities Monitoring (ADDM) Network illustrates this clearly. ADDM uses what is called active, record-based surveillance: rather than waiting for reports, network staff review and abstract developmental evaluations and records from community medical and educational service providers (Maenner et al., MMWR Surveillance Summaries, 2023).
In the 2022 surveillance year, a child met the autism spectrum disorder (ASD) case definition if the child lived in the surveillance area at any time in 2022 and had at least one of three things in the records: an ASD diagnostic statement in a comprehensive developmental evaluation, autism special education eligibility, or a qualifying diagnostic code from the ICD-9 or ICD-10 classification systems (Shaw et al., MMWR Surveillance Summaries, 2025).
Every element of that definition runs through a service system. A child who has not been evaluated, has not been found eligible for special education, or has not received a coded diagnosis cannot be counted. So anything that changes identification, such as screening practices, access to evaluation, insurance coverage or how complete the records are, can move a surveillance estimate even if the true occurrence of the condition stays the same. The 2025 report says as much: data quality depends on the availability and completeness of records at data sources, which vary by site and source.
None of this makes surveillance estimates unreliable. It makes them precise answers to a specific question: how many children in this population, at this time, have been identified as meeting this definition.
What can autism surveillance teach us about reading prevalence?
The 2025 ADDM report, covering the 2022 surveillance year, estimated ASD prevalence among children aged 8 years at 32.2 per 1,000 across 16 sites, which the report also expresses as one in 31 (Shaw et al., 2025). Read with the three choices above, that figure means: among 8-year-olds living in those 16 surveillance areas during 2022, about 32 in every 1,000 were identified in records as meeting the ASD case definition.
Two further details in the same report repay attention.
The variation between sites is large. Prevalence among 8-year-olds ranged from 9.7 per 1,000 in Laredo, Texas, to 53.1 per 1,000 in California, more than a fivefold difference. Among 4-year-olds, the combined estimate was 29.3 per 1,000, ranging from 12.9 in Indiana to 60.6 in California.
The report explains what that variation does and does not mean. The authors write that differences across communities might be due to differences in availability of services for early detection and evaluation and in diagnostic practices, and they also point to factors such as insurance coverage and eligibility for early intervention. They state directly that research has not demonstrated that living in certain communities puts children at greater risk for developing ASD. In other words, the authors themselves read the site differences primarily as differences in identification, not in risk.
The comparison with the previous report needs the same care. The 2023 ADDM report, covering the 2020 surveillance year, estimated prevalence among 8-year-olds at 27.6 per 1,000, or one in 36, across 11 sites (Maenner et al., 2023). It is tempting to set 27.6 beside 32.2 and call the difference a trend. But the two figures come from different sets of sites, and the 2025 report notes that the populations within ADDM surveillance areas are not nationally representative and do not generate nationally representative prevalence estimates. Two estimates drawn from different, non-representative populations are not a like-for-like comparison, whatever their direction.
Which questions should you ask of any prevalence figure?
The autism example generalizes. Before drawing any conclusion from a prevalence estimate, it helps to ask:
- What population does it describe? An age group, a set of geographic areas, a clinic population? Is that population representative of the one you care about?
- Point or period? Was prevalence measured at a single time, or over an interval?
- What was the case definition? Diagnosis, eligibility, a code, a symptom score? Different definitions count different people.
- How were cases found? Active record review, self-report, a registry, a survey? Each method misses different people.
- Is a comparison like-for-like? If two estimates are being compared across places or years, do they share the same population, definition and method?
- Is the figure a rate or a rounding? "One in 31" is a rounding of 32.2 per 1,000. The "one in N" form is easy to remember, but it hides the decimal and gives no sense of the uncertainty around the estimate. Reading the underlying rate, per 1,000 or per 100,000, is more informative.
What prevalence cannot tell you
Prevalence cannot establish what causes a condition, and on its own it cannot tell you whether risk has changed. Because it depends on incidence, duration and identification at once, a difference in prevalence between two places or two years could reflect a change in any of the three. Separating them requires other kinds of evidence: measures of new cases over time, careful attention to how case finding changed, and study designs that account for confounding, the problem treated in the explainer on how confounding distorts associations. Readers who encounter a prevalence figure framed as a rise in risk may also find it useful to revisit relative risk versus absolute risk, which covers how the framing of a number shapes the impression it leaves.
What prevalence does well is describe burden. A well-defined surveillance estimate tells public-health planners, schools and service systems how many people in a population have been identified with a condition, which is essential for planning evaluation capacity, services and support. Read for that purpose, with its population, time and case definition kept in view, it is one of the most useful numbers in public health.
The broader question of which environmental factors are and are not associated with autism, and what kinds of studies can address it, is the subject of the work described under autism and the environment and in the companion explainer on what studies of autism and the environment can show. Readers who prefer primary sources may find the peer-reviewed publications listed on this site a reasonable place to continue.