A careful count of a narrow thing
Arboviral diseases, the infections spread by mosquitoes and ticks such as West Nile virus, are nationally notifiable in the United States, and cases are reported through ArboNET, the national arboviral surveillance system managed by CDC and state health departments. The data are careful and consistent. They are also a count of diagnosed, reported and confirmed human disease, which is a small and specific slice of the infections that actually occur. An analyst who reads the counts as incidence will be wrong by a large factor, and wrong differently from year to year.
That does not make the data weak. It makes them a particular instrument with a known shape. Used for what they measure, they are among the better infectious disease series available. Used for what they do not measure, they produce confident numbers with no basis. This piece sets out what the counts can support, how to estimate what sits behind them, and how to carry the uncertainty into a program or market forecast.
How a case reaches the national count
A reported case is the end of a chain, and every link loses people. The infected person has to develop symptoms. They have to be ill enough to seek care. The clinician has to suspect an arboviral cause and order the right test. The laboratory result has to reach the health department. The health department has to investigate, apply the national case definition and classify the case as confirmed or probable. Then the case is reported to the national system.
Most infections drop out at the first link. For West Nile virus, CDC states that about 8 in 10 infected people develop no symptoms. About 1 in 5 develop a febrile illness, with fever and other symptoms such as headache, body aches, joint pain, vomiting, diarrhea or rash. Fewer than 1 percent, about 1 in 150, develop severe disease affecting the central nervous system. Each later link then filters again, and how hard it filters depends on how sick the person is.
Neuroinvasive and non-neuroinvasive are two different instruments
ArboNET reports cases in two clinical categories. Neuroinvasive disease includes meningitis, encephalitis and acute flaccid paralysis. Non-neuroinvasive disease is mostly febrile illness without neurological involvement. Treat them as two separate instruments with different sensitivity.
Neuroinvasive cases are usually hospitalized, usually tested and usually reported. Reporting of neuroinvasive disease is far more complete, which is why CDC and most analysts treat neuroinvasive case counts as the more reliable indicator of activity. Non-neuroinvasive cases are often managed as outpatients, often never tested and often never reported. Their count reflects testing behavior as much as disease, and it moves with clinician awareness, local media attention and laboratory practice.
The practical rule: build on neuroinvasive counts, and treat the non-neuroinvasive series as a lower bound with an unknown and changing gap. Ratios between the two categories vary across states and years for reasons that are mostly about ascertainment, not biology.
What surveillance counts and what it misses
| Component | What the counts capture | What they miss |
|---|---|---|
| Asymptomatic infection | Only those found by blood donor screening, reported separately as presumptive viremic donors | Almost all asymptomatic infections in the general population |
| Febrile illness | Cases that were tested, confirmed and reported | The large majority that were never tested or never sought care |
| Neuroinvasive disease | Most cases, because patients are usually hospitalized and tested | Some untested or misattributed cases, more in places with less testing |
| Timing | Date of illness onset for most cases, once reporting is complete | Late reports in current-year data, which keeps rising for months |
| Place | County and state of residence | Where the person was actually infected, which may be elsewhere |
| Animal and vector activity | Mosquito pools, dead birds, sentinel animals and veterinary cases where they are tested and reported | Areas that do little or no testing, which look quiet whether or not they are |
From cases to infections: the multiplier
If neuroinvasive disease is close to fully reported, and if a known fraction of infections become neuroinvasive, then dividing reported neuroinvasive cases by that fraction gives an estimate of total infections. The same logic gives an estimate of febrile illness. This is the standard back-calculation, and it is only as good as its two assumptions.
Estimating infections behind a reported count
Suppose a region reports 40 West Nile neuroinvasive cases in a season. These are invented figures. Using CDC's ratio of about 1 in 150 infections becoming neuroinvasive, and assuming every neuroinvasive case was reported, the region had about 40 times 150, or 6,000 infections. Using about 1 in 5 for febrile illness, roughly 1,200 of those people had a febrile illness, against perhaps a few dozen reported non-neuroinvasive cases.
Now relax the assumption. If only 80 percent of neuroinvasive cases were reported, the true neuroinvasive count is 50, and infections are about 7,500. If the neuroinvasive fraction in this population is closer to 1 in 100 because the population is older, infections fall to about 4,000 to 5,000. The honest answer is a range, roughly 4,000 to 7,500 in this example, with the assumptions that produce each end written next to it.
Present the multiplier result as a range built from named assumptions, never as a single estimate. A single number invites the reader to treat it as measured, and nothing in it was.
Where the multiplier breaks
The ratio of severe disease to infection is not fixed. CDC notes that people over 60 and people with certain medical conditions are at greater risk of severe illness. A population that skews older will produce more neuroinvasive cases per infection, so a single national ratio will overstate infections in an older population and understate them in a younger one. Adjust by age where the counts allow it.
Reporting completeness for neuroinvasive disease is high but not uniform. It depends on whether clinicians test patients with encephalitis or meningitis for arboviruses, which varies by region and by awareness in a given year. In a place where an arbovirus is rarely seen, cases can be attributed to other causes.
The 1 in 150 figure is itself an estimate, derived from studies of infection in specific populations and outbreaks. Treat it as a central value with a range around it. And the multiplier is specific to the virus. The ratios that apply to West Nile virus do not carry over to other arboviruses, each of which has its own clinical spectrum and reporting pattern.
Blood donor screening: the one window on silent infection
The only routine source that sees infections without symptoms is blood donor screening. Blood collected in the United States is screened for West Nile virus, and donors who test positive are reported to ArboNET separately as presumptive viremic donors. They are not counted as cases, and they should never be added to case totals.
Donor data are useful because donors are tested whether or not they feel ill. In a season with active transmission, the share of donors who screen positive gives a rough read on infection in the general population, independent of who sought care. Researchers have used this kind of data, together with follow-up of positive donors, to estimate the ratios of infection to illness that the multiplier depends on.
Two cautions apply. Donors are not a random sample of the population: they are adults, healthy enough to donate, and unevenly distributed by age and geography. And screening detects virus in the blood only during a short window after infection, so a positive rate reflects recent infection during the collection period, not cumulative infection over a season. Use donor data as a cross-check on the direction and rough size of a multiplier estimate, not as a substitute for it.
Case definitions and the probable category
A reported case meets a national case definition that combines clinical criteria with laboratory criteria. Cases are classified as confirmed or probable depending on the strength of the laboratory evidence. Both categories are usually included in published counts, and an analyst should know which were included in any figure they use.
Case definitions are revised from time to time, and testing technology changes faster than definitions do. A shift from one type of laboratory test to another, or a change in which results count as confirmatory, can move counts with no change in disease. When building a long series, check whether the case definition changed within the period and whether the change plausibly affected the counts. Where it did, mark the break in the series rather than smoothing over it.
Year-to-year volatility is the signal, not noise
Arboviral activity swings widely from year to year. Weather, mosquito populations, bird immunity and the timing of the season combine in ways that produce quiet years and outbreak years with little warning. National totals in a given year are often driven by a few regional outbreaks, so a national series can rise sharply while most states see nothing unusual.
Do not fit a trend line through a handful of years and project it. Three rising years are not a trend in a series this volatile. Instead, describe the distribution: the typical year, the range across a decade or more, and how often an outbreak year of a given size has occurred. For forecasting, that distribution is more useful than any single projected value, because the decision usually depends on whether a high year is likely, not on what the average will be.
Current-year data need care for a different reason. Reports arrive with a lag, and the count for the current season keeps rising for months after transmission ends. Comparing a partial current-year count with final counts from earlier years makes the current year look quieter than it is.
Geography: where people live, not where they were bitten
Cases are reported by the patient's county of residence. Most people are infected near home, but not all, and travel-associated infections are assigned to where the patient lives. For viruses that are mostly acquired abroad, the geography of reported cases reflects travel patterns rather than local transmission.
County-level counts are small, and rates per 100,000 based on small counts are unstable. One additional case can double a rural county's rate. Some jurisdictions also suppress small counts for privacy. Pool across years or across neighboring counties before computing rates, and show the counts alongside any rate so the reader can see how little the rate rests on.
Mosquito and bird surveillance: early warning, uneven effort
ArboNET also holds data on infected mosquito pools, dead birds, sentinel animals and veterinary cases. These are valuable as early-warning signals. Rising infection rates in mosquito pools often precede human cases by weeks, and many local programs use them to trigger control measures.
Their weakness for analysis is that effort varies. A county that traps and tests mosquitoes intensively will report more positive pools than one that does little testing, regardless of the underlying activity. Dead bird reporting depends on the public calling it in and on whether the program still tests birds. A county with no positive results may have no activity or may have no surveillance. Normalize by effort where you can, such as the number of pools tested, and treat absence of positives in a low-effort area as missing data rather than as zero.
Using surveillance in a program or market forecast
Program planners and market analysts both use arboviral data for the same underlying question: how many people will this disease affect, where and how predictably. Both should build the answer in the same way.
- Anchor on neuroinvasive counts, by age group where available
- Use a decade or more of final data, not the current partial year
- Describe the distribution of years, including how often outbreak years occur
- Apply multipliers as ranges built from named assumptions
- Pool small geographies before computing rates
- Treat vector and bird data as effort-dependent signals, not counts of risk
- Report every figure with its source, its year and whether it was measured or estimated
For a program office deciding where to place mosquito control or outreach resources, the output is a ranked set of areas with a measure of how reliably each sees activity. For a developer sizing demand for a vaccine or diagnostic, the output is a range of infections or illnesses per year, with an outbreak scenario shown separately from the typical year. In both cases the ranges should be shown in the main result, not tucked into a footnote.
Frequently asked questions
People with neuroinvasive disease are usually hospitalized and tested, so most of those cases are diagnosed and reported. People with febrile illness are often treated as outpatients or not at all, and many are never tested. The non-neuroinvasive count reflects testing behavior as much as disease, so it is a lower bound with a gap that changes from year to year.
Yes, as a range. Divide reported neuroinvasive cases by the fraction of infections that become neuroinvasive, which CDC puts at about 1 in 150, and adjust for incomplete reporting and the age structure of the population. State each assumption and show how the estimate changes when it changes.
Arboviral activity is volatile, driven by weather, mosquito and bird populations and the timing of each season, and national totals are often dominated by a few regional outbreaks. A few rising years are not evidence of a trend. Describe the distribution over a longer period and plan against the range.
As an early-warning signal and a sign of local activity, normalized by testing effort where possible. Because effort varies widely between jurisdictions, a lack of positive mosquito pools in an area that tests little is missing data, not evidence of no risk.