The metric that made the outbreak visible cannot see the intervention that reduced it
Between 2007 and 2015, almost 3,500 California state prisoners were diagnosed with valley fever, the fungal infection caused by inhaling Coccidioides spores from disturbed soil in the arid Southwest. The case counts that documented this outbreak came from routine public-health surveillance: reportable disease reports flowing from clinicians and laboratories to local health departments, then to the California Department of Public Health, then to CDC through the National Notifiable Diseases Surveillance System. Those counts did their job. They made an occupational and environmental hazard inside a confined institution visible enough to force action.
Then the counts fell. Between 2011 and 2015, valley fever rates in California state prisons dropped close to 90 percent overall, while background rates in the most endemic areas of the Southwest fell only 30 to 50 percent. That divergence is the headline. It is also where the measurement problem begins, because the intervention that coincided with the drop was not primarily an environmental control. It was a placement policy: keep the most susceptible people out of the two facilities with the highest transmission. A placement policy changes who is exposed. It does not change the fungus. And the case-count metric, as currently constructed, cannot tell the difference between those two things.
This is not a story about whether California prisons did something useful. The record shows they did. It is a story about what the surveillance architecture can and cannot attribute, and why that matters for anyone who will have to redesign it.
What the surveillance system actually measures
Start with the baseline caveats, because they are not boilerplate. CDC states plainly that valley fever is underdiagnosed and that thousands of cases are not reported. The agency’s modeled estimate of the true burden is roughly 206,000 to 360,000 symptomatic cases per year, 10 to 18 times the reported count. In 2023, states reported 21,037 cases to CDC, provisional; roughly 20,000 are reported each year. CDC also notes that state-reported counts may differ from national NNDSS figures because of timing and surveillance-method differences, and it lists detection and reporting method as one of the possible drivers of year-to-year change, alongside travel and environmental factors.
Those caveats are general. In a prison, they compound. A surveillance case count is a report of a diagnosis that met a case definition and reached a health department. It is not a measure of exposure. It is not a measure of who was at risk. It is not a denominator. When the population at risk is defined by who was admitted to a particular facility, and when the policy being evaluated changes who gets admitted, the case count becomes a moving target that looks stable.
California’s public infectious-disease dataset illustrates the structural limit. It contains reportable case counts by disease, county, year, and sex, drawn from confidential morbidity reports and laboratory reports. CDPH notes that the data may differ from previous publications because of delays inherent to case reporting, laboratory reporting, and epidemiologic investigation. County-level counts are useful for spotting regional patterns. They are not facility-level denominators, and they cannot track a reception cohort — the group of people admitted to the prison system in a given period — as it moves through facilities.
The intervention timeline is a sequence of different mechanisms
The California prison response is often described as a single success. It was not a single intervention, and the mechanisms do not do the same thing.
2006: immunocompromised exclusion. CDCR began excluding inmates with suppressed immune systems, due to factors such as HIV or chemotherapy, from all eight Central Valley prisons. This is a susceptibility-based placement rule. It reduces exposure for a defined high-risk group by keeping them out of endemic facilities. It does not reduce spore concentration anywhere.
2013: ethnicity-based exclusion at two facilities. A federal receiver appointed to oversee health care in state prisons mandated that individuals from the most at-risk ethnicities, African-Americans and Filipinos, be excluded from Avenal State Prison in Kings County and Pleasant Valley State Prison in Fresno County. This is a sharper placement rule. It targets the two facilities with the highest observed rates and removes a subgroup with elevated risk of severe disease. Again, the fungus is unaffected.
Environmental controls. The same facilities modified landscaping to minimize dust, allowed inmates to stay indoors when wind speeds were high, and distributed dust masks for outdoor use. Pleasant Valley received exemptions from state water restrictions to maintain grass and shrubbery in recreational yards, on the logic that plant roots stabilize soil and reduce dust. These are true exposure-reduction measures. They change the concentration of inhalable spores in the facility environment.
2015: Spherusol skin testing at reception. On CDC’s recommendation, CDCR offered the Spherusol skin test to every inmate entering the state prison system. The test detects prior exposure and, in the logic of the program, immunity. In a mass screening in early 2015, around 35,000 inmates took the test and 9 percent were shown to be immune. Inmates who tested negative were medically restricted from living at Avenal or Pleasant Valley. This is a screening-and-placement mechanism. It sorts people by risk and routes the susceptible away from the highest-exposure facilities. The FDA had not approved Spherusol for this purpose; it was used off-label.
Three of these four mechanisms change who is exposed. One changes exposure. The published case-count drop aggregates all of them.
Why the 90 percent figure resists attribution
The 2011–2015 decline is real as an observation. The attribution is not settled. Background rates in the most endemic areas of the Southwest fell 30 to 50 percent over the same period, which means some of the prison decline would have happened anyway. John Galgiani, director of the University of Arizona’s Valley Fever Center for Excellence, told reporters that the prisons should be given credit to some extent but that it would take time to see whether the drop stayed low and whether it was because of the policies now in place. CDC’s Tom Chiller said the agency predicted Spherusol would prevent more valley fever cases than any other measure so far but that it was still too early to calculate the test’s precise impact.
That is not evasion. It is an accurate statement about what the data can support. To decompose the effect, you would need facility-level case counts before and after each intervention, reception-cohort denominators, and explicit counts of who was excluded from which facility and when. The published record does not provide that decomposition. The case-count metric shows a drop. It does not show which mechanism produced it, and it cannot rule out that some of the drop reflects susceptible people being routed to other facilities rather than protected from exposure.
Consider a clearly labeled hypothetical, not a finding: a facility admits 1,000 immune-negative inmates per year and excludes 200 high-risk ones. Even with zero change in spore exposure, the observed case rate falls by roughly the excluded fraction. On a single-number dashboard, that mechanical effect looks identical to a successful dust-control program. The metric cannot distinguish them because it was never designed to.
The denominator problem, stated precisely
A rate is a numerator over a denominator. Valley fever surveillance in prisons typically produces a numerator: diagnosed cases among people who were in a facility. The denominator that matters for a placement policy is the number of susceptible people who were actually exposed at that facility during the relevant period. When the policy removes susceptible people from the facility, the denominator shrinks. The numerator may fall for that reason alone. The rate falls. The risk to the people who remain may be unchanged, and the risk to the people who were moved may have simply relocated.
This is not a hypothetical concern about statistical nicety. It is the core of what a placement intervention does. Excluding immune-negative, high-risk inmates from Avenal and Pleasant Valley changes who is exposed at those facilities. If the same people are sent to other Central Valley facilities, total system exposure may be unchanged while facility-level rates collapse. The published metric cannot adjudicate that question because it does not track people across facilities. It tracks cases by report, and reports are tied to where a person was diagnosed, not to the full trajectory of exposure.
California’s county-level dataset makes the same point from the other direction. Cases are classified by county of residence in CDC’s national mapping, and cases infected while traveling are attributed to county of residence. In a prison system, residence and exposure are not the same thing. A person can be exposed in one county and diagnosed in another. A person can be moved between facilities before symptoms appear. The surveillance architecture was built for civilian populations with relatively stable residence. It was not built for a population that is moved by policy.
The Taft gap is the boundary of the claim
Any statement about valley fever in California prisons is a statement about a subset, and the subset boundary matters. Taft Correctional Institution is a federal prison, and its health data have not been made available. It sits in Kern County, which has the highest valley fever incidence of any California county. A federal facility managed by a private contractor is outside the state surveillance universe whose numbers are cited as evidence of success. That is not a footnote. It is the edge of what the data can support.
The same boundary problem applies to the state system itself. The 90 percent figure describes state prisons. It does not describe the federal system, private contract facilities, or county jails. When someone says “California prisons reduced valley fever,” the accurate statement is narrower: California state prisons reported a large decline over a period in which several different interventions were implemented, and the attribution among them is unresolved.
What NHSN does and does not cover
It is tempting to look to CDC’s National Healthcare Safety Network as the missing measurement layer. NHSN is described by CDC as the nation’s most widely used healthcare-associated infection tracking system, and it is tied to CMS reporting requirements and value-based purchasing programs. It tracks healthcare-associated infections, blood safety errors, and process measures such as healthcare personnel influenza vaccination status.
NHSN is not a valley fever surveillance system for correctional facilities. It is built around acute-care and long-term-care reporting modules, not around environmental exposure in prisons. The relevant surveillance channel for coccidioidomycosis is the reportable-disease system, NNDSS and its state analogues, which is a case-reporting architecture, not a facility-environment architecture. The gap is not that NHSN is failing to capture valley fever in prisons. The gap is that no existing system is designed to capture the exposure denominator that a placement policy changes.
What a placement-aware metric would require
The redesign question is concrete. If the goal is to know whether a placement policy reduced risk rather than relocated it, the reporting architecture needs three things. None of them require new science. They require different reporting forms and a different analysis plan.
Reception-cohort denominators. Track the group of people admitted in a given period, with their Spherusol status where available, and follow them across facilities. The unit of analysis becomes the person-facility-time, not the case report. This is the only way to see whether susceptible people were protected or simply moved.
Explicit reporting of exclusion counts by facility. If a facility excludes 200 high-risk people in a year, that number belongs in the same table as the case count. Without it, the denominator is invisible and the rate is uninterpretable. The exclusion count is not a privacy problem at the aggregate level; it is a required covariate.
Separation of environmental controls from population routing in the analysis plan. Dust suppression, landscaping, indoor-air policies, and mask distribution are environmental variables. Exclusion and screening are placement variables. An analysis that aggregates them cannot attribute the effect. An analysis that separates them can, at least partially, and can say so honestly when it cannot.
There is a fourth requirement that is less technical and more institutional: the federal and private-contract facilities have to be inside the surveillance universe, or the claim has to be explicitly limited to the state system. The Taft data gap is not a reason to stop analyzing. It is a reason to stop describing state-system numbers as if they describe all California prisons.
Why this matters beyond valley fever
The pattern generalizes. Whenever a policy works by changing who is exposed rather than by changing the exposure, case-count surveillance will show a decline that is real and an attribution that is not. The same logic applies to heat-related illness in facilities without cooling, to wildfire-smoke exposure in institutions that cannot filter indoor air, and to any environmental hazard where the response is to move susceptible people rather than to remediate the environment. The metric will look like success. The question is whether the success is in the environment or in the routing.
For valley fever specifically, the honest summary is this: California state prisons implemented a sequence of interventions between 2006 and 2015, at least three of which changed who was exposed rather than how much fungus was in the air. Case counts fell sharply. Background rates fell less. The attribution among mechanisms is unresolved, and the surveillance architecture as currently built cannot resolve it. That is not a reason to undo the interventions. It is a reason to build the measurement system that can tell the next redesign which lever actually worked.
FAQ
Did California deliberately site its prisons in valley fever country?
The retrieved record does not support a claim about siting decisions. What the sources document is that several state prisons are located in the Central Valley, where Coccidioides is endemic, and that Kern County has the highest valley fever incidence of any California county. The historical, economic, and political reasons for siting are not established in the available sources and should not be asserted without primary documentation.
Did the prison interventions actually work?
The observed decline is real: rates fell close to 90 percent between 2011 and 2015, against a 30 to 50 percent background decline. Whether the interventions caused the difference, and which intervention contributed most, is unresolved. The sources present the drop as observed and the attribution as explicitly unsettled.
What is Spherusol, and was it approved for this use?
Spherusol is a skin test that detects prior exposure to Coccidioides. In 2015, CDCR offered it to every inmate entering the state prison system; about 35,000 took the test and 9 percent were shown to be immune. The FDA had not approved the test for this purpose, and it was used off-label.
Why does the Taft facility matter to the analysis?
Taft is a federal prison in Kern County, and its health data have not been made available. It sits outside the state surveillance universe whose numbers are cited as evidence of success. Any system-level claim about California prisons is therefore a claim about a subset, and the boundary should be stated.
What would it take to attribute the decline to specific interventions?
Facility-level case counts before and after each intervention, reception-cohort denominators that follow people across facilities, explicit exclusion counts by facility, and an analysis plan that separates environmental controls from population routing. The published record does not provide this decomposition.