COVID-19 Reproduction Numbers and Long COVID Prevalences in California State Prisons

2.1. Terminology

The California Department of Corrections and Rehabilitation (CDCR, formerly California Department of Corrections) refers to the prisons it operates as institutions. The buildings of each institution are grouped into facilities, informally known in some cases as yards, with multiple buildings per facility and often multiple facilities per institution. The people incarcerated in these institutions are referred to as residents. The names and standard abbreviations for the 35 institutions active during the period studied¹ are listed in Table L.1 (Supplement L). Health care in CDCR institutions is managed by a division within CDCR known as “California Correctional Health Care Services (CCHCS).”

2.2. Data

Data provided by CDCR recorded the room locations of individual prison residents in each institution’s multiple buildings on each day, and the type of room in which they were housed, from Jan. 1, 2020 through March 25, 2022. For all SARS-CoV-2 cases, it also included dates of symptom onset where available, as well as dates and results of all RT-PCR and antigen tests administered to residents at all CDCR institutions. Only deidentified data were made available for analysis.

While the locations of prison residents at the institution level were identified by institution names, buildings, rooms, and facilities within institutions were identified only by anonymized numbers. We report results by building using these numerical IDs. A classification of room types was provided. We combined several types of cells and dorms into categories Cell and Dorm. The types Room, Closed Ward, and Other were rare and were combined into the classification Uncategorized.

A classification of rooms by solid versus permeable barred doors was compiled by UCSF and Berkeley researchers from information provided by CCHCS. We used it to construct a combined “room type” for each room consisting of one of the above types, Cell, Dorm, and Uncategorized, together with a door type, Open, Closed, and Uncategorized.

Individual demographic characteristics recorded included birth year, “phenotypic sex” and race. Race was recorded using a system of nine classifications, of which eight were used in the population analyzed (the Unknown category was not present). We collapsed the variable to six levels by combining the categories Mexican, Hispanic, and Cuban into a single Latinx group, for comparability to California data and to avoid possible deductive disclosure when group sizes are small.

The sex variable was not used as is, but was interpreted together with location to impute gender, as follows. The institutions are segregated into women’s and men’s facilities, and the majority value of the sex variable in each building matched this classification. California’s State Senate Bill 132 (SB 132), effective January 1, 2021, requires CDCR to house prison residents according to their self-identified gender by request (The Moss Group, Inc., 2022), and a number of residents are housed in facilities that do not match their recorded sex. We labeled anyone for whom the sex variable matched the building’s majority at all times as cisgender man or cisgender woman (CM, CW), and all others as transgender, nonbinary, or intersex (TGNBI). Due to the small group size and lack of explicit gender listings, we did not attempt to distinguish between transgender women, transgender men, and others labelled TGNBI. Because SB 132 was not in effect in 2020, because many requests for transfer under SB 132 have reportedly been unfulfilled (Sosa, 2023), and because some may not request housing reassignment, it is likely that these imputed genders are incomplete in some cases, and that the TGNBI category here represents a subset of transgender, gender-nonconforming, non-binary, and intersex residents. We include this gender variable to gain a view into the effects of gender on disease risk.

2.3. Total case counts

Following CDCR’s reporting policy (California Department of Corrections and Rehabilitation, n.d.a), we considered a positive test result after 90 or more days without positive test results to indicate a new case, which may have been a reinfection of a previously infected individual. We compiled the number of positive and negative tests administered by day, new cases detected by day, and seven-day average of cases detected by day in each institution, and in each building.

We used SARS-CoV-2 test results by date to identify distinct outbreaks at each institution, using the California Department of Public Health’s definition as a collection of case detections separated by fourteen or fewer days (California Department of Public Health, n.d.). Cases were considered to be detected on the date of their first positive test. We aggregated the number and sizes of outbreaks at each institution.

2.4. Transmission reconstruction

We conducted a computational process of Bayesian estimation to estimate incidence dates and effective reproduction numbers for SARS-CoV-2 transmission for each individual and building of the CDCR institutions on each day. Figure 1 offers an outline of key steps, and see Supplement A for more details.

• Download figure
• Open in new tab

Figure 1:

Key steps in the estimation model, which builds on the work of Wallinga and Teunis (2004). (A.) Data collected by CDCR provides dates of positive and negative SARS-CoV-2 tests administered on individuals (who are identified by anonymized ID numbers), and in some cases (not shown here) reported dates of onset of symptoms. All case histories shown in this figure are fictional. (B.) For each individual, a probability is estimated for each potential date of infection based on the probabilities of negative and positive test results and symptom onset at different times during infection. Here, the probability is nonzero on days before the first positive test, and drops off before the earlier negative tests, because the probability of a false negative result declines as the time from infection date to test date increases. (C.) Along with the infection dates, a probability distribution is estimated for dates the individual was infectious, by estimating probabilities for the time from infection to transmission and using the above estimate for dates of infection. Specifically, if the individual was to be exposed to susceptible individuals at a constant rate, and one transmission event occurred, this estimates the probability for the day of transmission. (D.) A likelihood is estimated for all possible transmission links between all pairs of individuals: the likelihood is higher if the first person’s infectious period and the second person’s infection date are more likely to coincide. (Not shown: likelihood of transmission is also higher on days when individuals were housed in the same room or building; likelihood is assigned to dates of transmission as well as pairs of individuals.) (E.) Probabilities for all possible transmission links are estimated from the likelihoods by scaling to reflect that each case has one source. (F.) Reproduction numbers are estimated by the expected number of cases infected by each person, based on the above transmission link probabilities. (Not shown: estimation of link probabilities, reproduction numbers, and incidence by date. See appendix for details.)

We used CDCR’s data set to provide dates and results of all RT-PCR and antigen tests administered to residents at all CDCR institutions, dates of onset of symptoms reported for residents, and daily locations of residents at the institution, building, and room levels.

We estimated the likelihood that each resident was infected on a particular day by combining several sources of information. Established estimates of the performance of RT-PCR and antigen tests for the virus were used to combine all positive and negative test results recorded for an individual, together with reported symptom onset dates where available, to construct a probabilistic estimate of date of infection and duration of incubation period for each infected individual. Because symptom onset dates were reported to contain inaccuracies due to record-keeping difficulties experienced during outbreaks, we used a relatively vague prior distribution for symptom onset dates compared to test results (see Supplement A for details). This allowed us to estimate how likely each resident was to have been infected on each day, and how likely they were to transmit the virus to others on each day, based only on their individual test and symptom history.

We then used these estimates together with residents’ daily locations to estimate how likely transmission between individual pairs of residents was to occur on each day. From this we generated an updated estimate of each case’s incidence date and of how many cases were infected by each infected resident each day, and from that effective reproduction numbers (R) by building by day, and the number of cases infected in each building each day. This estimation process operated on all cases in an institution together, and was applied to each institution in the CDCR system, generating daily building-level estimates of R for all outbreaks in all institutions in the time period studied.

These estimates accounted for the daily movements of residents between buildings, and provided a description of the dynamically changing conditions in each building by day as individuals become infected, recover, and shift locations. Because these estimates took into account where individuals were day by day when they were likely to have been infected, where they were when they were likely to be infectious, and which individuals were in proximity and infectious on days when each other individual at the institution may have become infected, they provided an estimate of which locations within an institution were likely to have been sites of transmission each day. Because reproduction numbers can be understood to reflect local conditions conducive to spread of the disease, these estimates could be used to evaluate the relative safety of different settings.To examine the difference between location of residents at infection and their location at case detection, we compared the number of case detections of individuals housed in each room to the number of incidence events attributed to that room by the estimate, using a linear regression with intercept of zero, with room type as an exposure variable.

After generating the above described estimates for each institution in the CDCR system, we combined them to provide a summary of estimates across the system of prisons. We visualized the distribution of R values pertaining to infectious individuals by institution and in each building. We compared R val-ues stratified by room and door type, to investigate the hypotheses that cells reduced transmission compared to dorms and that solid doors reduced transmission compared to permeable ones.

In the Results section below we present summaries of the results over the 35 institutions of the CDCR system. In Supplement I, we present the results from each institution in detail, in one panel of figures for each CDCR institution. Supplement J presents one panel for each room type, showing the estimated overall incidence and reproduction number by day in each institution, over all rooms of the given type only. Supplement A documents the estimation methods in detail.

2.5. Distribution of cases

Adjusted per capita infection risk is relevant to comparing infection rates between groups in the prison population, to infer whether some groups faced higher infection risk than others in the outbreaks that occurred. Previous analyses of some of these outbreaks have shown that per capita the risk of infection was distributed relatively evenly across racial/ethnic groups, without the disproportionately high rates among Black, Latino, and Indigenous people that were documented in community transmission in 2020–21 (Chin et al., 2021; Kwan et al., 2022). We performed a similar analysis on the series of outbreaks considered here.

First, we compiled as summary statistics the number of infection events by race/ethnicity and gender. Second, along with the above statistics, we analyzed associations between gender, race/ethnicity, and the interaction of the two and infection risk, to investigate whether different populations were exposed to different risk in the prisons without comparison to the state. Using our reconstruction of incidence dates and locations, we summarized estimated incidence by person and day into number of infection events during each person’s stay in each building where they were housed. Each person’s duration of stay in each building was counted in years and fractions of years. We performed a mixed-effects Poisson regression analysis of the incidence against gender, race/ethnicity, the interaction of gender and race/ ethnicity, age (in age classes under 23, 23–35, 35–50, and over 50), and duration of stay, with location at the building level as a random intercept. We tested multiple null hypotheses: one for each level of the gender and race/ethnicity variables, of no difference from the population-wide rate of infection, and one for each level of the gender-race/ethnicity interaction, of no difference from the infection rate predicted by gender and race/ethnicity independently, with Bonferroni-Holm adjustment for multiple comparisons (see Supplement B for details).

2.6. Estimation of excess cases

While the above analysis addresses differences in disease risk given the populations and conditions that existed during these outbreaks, at the same time, such analysis of individual-level risk does not fully describe the burdens and impact of the disease spread in the prisons. The impact of prison outbreaks on the state can be framed as a question of what would be different in the state in their absence. The answer to such a question can only be inferred and estimated. Here we approach it as modelers by way of two counterfactual scenarios.

First, we compiled the raw number of CDCR cases by race/ethnicity in the time period studied, and we note that this statistic provides a very basic counterfactual comparison, namely the number of cases that would not have occurred if every prison resident were imprisoned as they were but COVID-19 transmission into prisons had been prevented.

Second, we additionally constructed an age- and race/ethnicity-adjusted estimate of the number of excess cases in state prison outbreaks compared to community rates of infection, to assess how the occurrence of these prison outbreaks affected the over-all number and distribution of cases in the state. This allows a comparison of the impact of these outbreaks to what the disease burden might have been in a counterfactual scenario in which all the residents affected had been in the community and infected at community rates. This second estimate was conducted as follows.

We calculated the cumulative incidence rate by day in each institution from new case detections relative to population size on March 20, 2020, the first day a case was detected in a CDCR prison. We used this reference population size because it was the size before any adjustments to release and intake rates made in response to outbreaks.

Comparison of overall case counts between the CDCR system and the state of California required an estimate of the true number of cases, since case ascertainment rates may have been very different in the two settings. Testing rates in the CDCR system were relatively high: from March 2020 to October 2021 94% of all residents were tested, and residents who were tested were tested an average of 16 times (Kwan et al., 2022), while testing in California was likely much less frequent. Rather than compare counts of confirmed cases directly, we used an estimate of the state’s true number of cases published by the Institute for Health Metrics and Evaluation’s (IHME) estimate of incidence (Institute for Health Metrics and Evaluation, 2022; IHME COVID-19 Forecasting Team, 2021). We adjusted the California estimate by removing incarcerated people from the total infections and population, to estimate community incidence rates in the state excluding carceral institutions. CDCR resident numbers were drawn from the CDCR data described above, while numbers from federal prisons, county jails, and ICE facilities were obtained from UCLA’s COVID Behind Bars project (UCLA Law COVID Behind Bars Data Project, n.d.; Dolovich and Littman, 2020) and adjusted using a broad span of values for the unknown case ascertainment rate (see Supplement D for details). By comparing this estimate to the cases detected within the CDCR system, we obtained a conservative estimate of the excess incidence in the CDCR system. Since the case ascertainment rate in the prisons is likely very high, though not perfect, the estimated excess cases can be used as a low estimate of the true excess cases.

We split the time period studied into three waves: the early outbreaks, the winter of 2020–21, and the subsequent outbreaks, which were predominantly the winter 2021–22 outbreaks. We compared the overall incidence in each wave between the CDCR system and the state, as a fraction of total population. In waves in which the CDCR incidence rate exceeded California’s, we estimated the number of excess cases by race/ethnicity by comparing CDCR incidence to what it would have been at California rates, by race/ ethnicity and age. The estimation of excess cases was conducted as follows.

The count of case detections during the relevant period in California by race/ethnicity was downloaded from California Health and Human Services’s Open Data Portal (California Health and Human Services Open Data Portal, 2023). The ascertainment rate for the state was estimated by the ratio of IHME-estimated total infections to total case detections. True infections and incidence rates in infections per person-year, by race/ethnicity, were then estimated from the ascertainment rate, case detections by race/ ethnicity, and population sizes by race/ethnicity. Age-specific incidence rates by race/ethnicity were estimated from the above using known odds ratios for infection by age (Talaei et al., 2022) and California age distributions stratified by race/ethnicity. California’s “Two or more races” counts were counted as “Other” for these estimates.

A counterfactual estimate of the number of CDCR residents who would have been infected at California rates, by race/ethnicity, was generated from the total person-years spent in CDCR institutions by race/ ethnicity and age using the above incidence rates. From that an estimate of excess cases in the CDCR system by race/ethnicity was derived by subtraction, and the rate of excess CDCR cases relative to California population by division. This sequence of steps is detailed and illustrated in Supplement D.

2.7. Long COVID model

While there is much uncertainty about the nature and impact of post-acute effects of SARS-CoV-2 infection, estimates of their prevalence and duration are available (e.g. Davis et al., 2023). We used results from several studies to estimate the epidemiology of long COVID and resulting disability among the known cases in the CDCR resident population. We obtained a baseline rate of long COVID occurrence given infection from a Dutch study (Ballering et al., 2022), age-specific rates from a U.S. study (FAIR Health, 2022), rates by gender and race/ethnicity from the U.S. CDC (CDC, 2023), protection due to vaccination from U.S. and U.K. studies Al-Aly et al. (2022); Ayoubkhani et al. (2022), an estimated rate of recovery from long COVID from a French study (Tran et al., 2022), and a conservative estimate of disability resulting from long COVID from U.K. data (Ayoubkhani et al., 2021; Spiers, 2022) (see Supplement C for details). We constructed a probabilistic model of long COVID prevalence and disability due to long COVID using age, gender, race/ethnicity, vaccination status, and estimated infection dates of resident cases (Supplement C). Location data was used to aggregate prevalence by institution by day.

The effects of race/ethnicity and gender are those observed in community settings; it is not clear whether they may reflect social phenomena such as access to health care, or proximately physical phenomena such as pre-existing comorbidities. Because institutional factors such as access to health care may not occur in the same way in the prison setting, we modeled long COVID prevalence in two ways: first, assuming the effects of race/ethnicity and gender are as reported in community cases (“more stratified” long COVID model, hereafter), and second, assuming those effects do not exist in the prison setting (“less stratified”). The true distribution of long COVID in this setting may lie between these two extremes.

We conservatively took the proportion of long COVID cases that lead to disability at 19% (Ayoubkhani et al., 2021; Spiers, 2022), and used this proportion to produce an estimate of the number of disabling long COVID cases by location, age, race/ ethnicity, and gender. Because this estimate was based on the estimate of long COVID prevalence, it also had more and less stratified versions. Because our estimate of the proportion of long COVID cases that lead to disability was not stratified, we did not estimate the racial/ethnic distribution of disabling long COVID cases separately from that of long COVID overall.

To compare, we used the same model to estimate the prevalence of long COVID and of disability from long COVID in California over the same time period, in total and by race.

2.8. Data sharing

Our data sharing agreement with CDCR and CCHCS does not permit publication of the raw source data.

Data requests may be made to the California Correctional Health Care Services and are subject to controlled access, due to requirements to enhance protection of this vulnerable incarcerated population. Source code used in the analysis is available on request from the corresponding author.

3.1. Summary statistics

The first positive SARS-CoV-2 test was administered on March 20, 2020 at the California State Prison in Los Angeles County (LAC). Between then and March 25, 2022, 196,652 individuals were incarcerated in CDCR facilities (Table 1). By the end of that time, 66,684, or 33.9% of them, had tested positive for SARS-CoV-2 while incarcerated in a CDCR facility.

Counting reinfections, 73,386 cases were detected through March 25, 2022, for an average of 1.10 infections per ever-infected resident. By the 15th of October 2020, 15,499 cases were detected, involving 33 of the 35 CDCR institutions. Between that date and the end of March 2021 a much more intense wave of outbreaks swept the system, causing 34,189 cases among residents and involving all 35 institutions. A third wave in the winter of 2021–2022 brought an additional 23,698 cases, again including all 35 institutions. As of March 25, 2022, CDCR reported 253 residents had died as a result of COVID-19 infection (California Department of Corrections and Rehabilitation, n.d.b).

The largest number of cases detected on a single day at an institution was 494 cases, at Pleasant Valley State Prison (PVSP) on Nov. 30, 2020. The seven-day average of cases detected per day exceeded 100 on four occasions, at San Quentin (SQ) in June 2020 (March 20, 2020 population 4,013), Chuckawalla Valley State Prison (CVSP) in May 2020 (2,958), California Men’s Colony (CMC) in January 2021 (3,836), and Wasco State Prison (WSP) in January 2022 (4,564) (Figures E.1, E.2, Supplement E).

There were 174 outbreaks of size 3 or greater (Figure E.2, Supplement E), including every institution. 79 of them exceeded 100 cases, and 27 exceeded 1000 cases. 20 of the 35 institutions had outbreaks extending to 1/3 or more of their March 20, 2020 population, and 32 of the institutions had more than one outbreak exceeding 100 cases.

3.2. Transmission estimates

We performed a probabilistic reconstruction of the timing and location of individuals’ infection events and infectious periods, and used it to estimate effective reproduction numbers by individual for each day, which were then aggregated at the building and institution levels.

As mentioned, this fine-grained analysis was motivated by the need to account for movement of individuals between the times of infection and detection, in order to distinguish the effects of different locations and room types on transmission. Because residents were often moved into quarantine and isolation, and sometimes moved for preventative purposes, as outbreaks spread, case detections could potentially occur in different locations from where transmission occurred. Our estimation reconstructs the time of each individuals’ infection as a distribution of possible dates preceding the time of their first positive test, at which times they may have been located elsewhere from when they were tested.

In the period from March 20, 2020, to March 25, 2022, residents’ bed locations were moved from room to room within an institution 803,482 times and across institution boundaries 197,163 times (see Supplement G for details).

We examined the effect of movement and transfer on the apparent location of disease transmission events, we compared the number of cases detected by a positive test while housed in a room to the number of cases infected in the room, as estimated by our reconstruction of transmission dynamics. A linear regression with intercept of zero by room type indicated that infection events tended to exceed detection events in dorm rooms regardless of door type, and the opposite in cells regardless of door type (Supplement G). This may be due to infected residents being moved from dorm settings to smaller cells as a control measure, before detection. As a consequence, if the difference between place of infection and place of detection were not accounted for, reproduction numbers would tend to be overestimated in cells and underestimated in dorms; we have accounted for this effect by using infection dates.

We estimated daily effective reproduction numbers by building in each of the CDCR’s prisons, across the date range May 1, 2020 to March 19, 2021. This procedure also yielded daily probabilistic estimates of the number of cases infected each day (incidence) by building. These results are visualized in Figures I.1–I.35, in the Supplement, presenting the estimated incidence and daily R values by day and building in each institution.

We summarized average estimated reproduction numbers by day in the 35 CDCR institutions, providing a look at the overall course of the pandemic in the prisons (Figure 2). Outbreaks are characterized by reproduction numbers greater than one (red, yellow, and green in the figure) as the outbreak spreads, followed by reproduction numbers dropping below one when the outbreak reaches its peak and begins to shrink (blue in the figure). Thus the peak of each outbreak occurs around the time when the color shifts from red to blue at an institution, reading from left to right in the figure. R values in the range 1–4 are widespread, with a small number of peaks around 5. The plot shows all days on which estimated total infectiousness profile was at least 0.5 (see Supplement A.1 for details).

• Download figure
• Open in new tab

Figure 2:

(A) Daily weighted average of estimated R values by institution. Detailed plots of daily R by building within each institution are included in the supplement. (B) Distribution of estimated R values by room type. Heavy lines mark the median value and 95% central interval.

Estimates of R were then summarized by the room type where each infectious individual was housed. The distribution of estimated R values across all infectious individuals in the CDCR system in the time span modeled displayed some difference between room types (Figure 2). Notably, the R values seen in celled housing were not systematically lower than those in dorm housing, a notable result given that celled housing had once been assumed safer and less conducive to transmission than dorms. Secondly, the median R was actually higher for individuals in cells with solid doors than those in cells with permeable doors, a notable result given the risks associated with cells with bars for doors like those at San Quentin.

To assess the seasonality of transmission in cells and dorms, we aggregated the inferred true incidence events over room types and seasons (Figure 3). All types of cells and dorms hosted significant numbers of transmission events in spring and summer as well as during the systemwide waves of outbreaks in fall and winter. In particular, hundreds of cases were infected in dorm rooms with solid doors in spring and summer months, implying that disuse of cold-weather heating systems does not make the use of these rooms an effective protection against transmission. A chart of seasonal transmission events by institution (Figure J.3C, Supplement J) identifies 8 institutions where over 100 cases were infected in closed cells over a spring and summer: CIM, CIW, CMC, COR, ISP, CCI, and CTF in 2020 and SOL in 2021. The average estimated R in closed cells exceeded one during spring-summer outbreaks at CIW, CMC, COR, CCI, CTF, and SOL (Figure J.3B, Supplement J).

• Download figure
• Open in new tab

Figure 3:

(A) Estimated total incidence by room type in each season, (B) Total person-years per room type.

3.3. Distribution of cases

The raw rate of SARS-CoV-2 infection events per person-year among prison residents was in the range 0.3–0.4 events per person-year for cisgender men, varying by race/ethnicity, while it was in the higher range 0.4–0.5 for cisgender women, and much higher at 0.8–1.0 for TGNBI residents (Figure F.2, Supplement F). There appeared to be some variation between intersectional combinations of gender and race/ethnicity beyond the rates associated with either separately.

Poisson regression analysis for interaction between race/ethnicity and gender, with corrections for clustering by building and for multiple comparisons, found the infection rates in all three gender classifications to be significantly different from the whole-population rate (Figure F.3, Supplement F). After age adjustment and inclusion of a random effect for each building, the rate was found to be highest for cisgender women, higher than average for TGNBI people, and lower than average for cisgender men. The rate for Black people was significantly lower than average, and for categories White and Other was significantly higher than average. Significant differences were found in two intersectional groups: a higher rate in Latinx cisgender men than predicted by their gender and race independently, and lower in American Indian or Alaska Native cisgender men than predicted.

3.4. Excess cases

While the waves of cases within the prisons generally co-occurred with surges of COVID cases in California as a whole, the incidence rates were substantially higher inside the prisons than in the state in many institutions during the first wave, and in most during the second wave (Figure 4A,C). While the third wave of prison outbreaks in Winter 2021– 2022 was large, it was exceeded by the massive surge in California cases at that time, when the Omicron variant emerged.

• Download figure
• Open in new tab

Figure 4:

(A) Cumulative incidence rate of COVID-19 infections in each CDCR institution (gray), in CDCR overall (black), and in California (black, dashed, with 95% confidence interval), using IHME estimate of true incidence in California (including incarcerated populations). Vertical lines separate three “waves” of transmission. Excess cases in the CDCR system are seen in the first two waves. (B) Racial/ethnic distribution of COVID-19 infections in the CDCR system and estimated infections in California over all three waves. (C) Attack rate of outbreaks in the CDCR system and estimated attack rate in California, in first two waves. (D) Estimated excess cases in CDCR system relative to statewide population, by race/ethnicity, in the first two waves of outbreaks. Gradations of lightness on horizontal bars indicate mean and 95% confidence interval of estimate.

The state of California was 4.1% Black in 2021, and 0.2% was American Indian or Alaskan Native. The CDCR resident population, on the other hand, was 28.4% Black as of March 20, 2020 and 1.2% American Indian or Alaskan Native. Reflecting the makeup of the incarcerated population, a larger fraction of the COVID-19 cases in the CDCR system were Black and Indigenous people than in the state (Figure 4B): 25.6% of CDCR resident COVID-19 cases were Black, while 5.4% of cases in California were, and people classified as American Indian or Alaskan Native made up 1.3% of its cases, while comprising 0.4% of the state’s cases.

To validate this comparison, since differences in testing rates could bias these estimates, we compared testing rates in the CDCR system and in the state by race/ethnicity over the same time period. In the CDCR system, Black, American Indian/Alaska Native, and white residents were tested 13.6, 13.8, and 14.0 times per person-year on average, respectively. Statewide, Black Californians were tested at 1.15 times the rate of white Californians, while the rate for the American Indian classification was 0.83 the white rate. Thus no undercount of Black cases in the state relative to CDCR is indicated, and while Indigenous cases may have been relatively under-counted due to testing coverage in California, the difference is not sufficient to explain the factor of over 3 difference in percentage between the 0.4% of statewide cases and 1.3% of CDCR resident cases who were American Indian or Alaska Native.

We used the IHME estimate of the true count of COVID-19 cases in the state together with testing rates by race/ethnicity and age-specific incidence rates to estimate community rates of infection in the state stratified by race/ethnicity and age, and used that to estimate the number of CDCR residents who would have been infected during the first two waves of transmission if they were not incarcerated. This counterfactual count was 18,000 (95% CI: 15,800– 20,300) cases, while the true number of CDCR cases detected during that time was 49,688. As a result, we estimated that 31,700 (95% CI: 29,400–33,900) COVID-19 cases in excess of the counterfactual number occurred in CDCR outbreaks, at an overall rate of 6.23 (95% CI: 6.21–6.24) excess cases per 10,000 California residents. The excess case rate was far higher for Black prison residents than any others, at 37.3 (95% CI: 32.6–42.0) Black cases per 10,000, followed by 30.3 (95% CI: 22.1–38.4) American Indian and Alaska Native cases per 10,000, with the excess case rate for all other populations falling below 7 per 10,000 (Figure 4D).

3.5. Long COVID

Of the 66,684 CDCR residents reported infected with SARS-CoV-2 during the time period studied, our model estimated a total of 10,600 (95% CI: 10,400– 10,800) (more stratified; 9,240 (9,060–9,410) by less stratified model) ever having long COVID symptoms, and 2,010 (95% CI: 1,970–2,050) (resp. 1,760 (1,720–1,790)) ever having disabling long COVID symptoms. The estimated prevalence of long COVID peaked at around 5,600 (resp. 4,900) in early 2021, gradually declined by a fraction, and returned to about its peak level at the end of the period studied in January 2022 (Figure 5A). The estimated prevalence of disability from long COVID followed the same pattern, peaking near 1,100 (resp. 930) and remaining near its peak level (Figure 5B).

• Download figure
• Open in new tab

Figure 5:

(A) Estimated long COVID prevalence in the CDCR system by date, in more (higher curve) and less (lower curve) stratified long COVID models, (B) estimated prevalence of disability due to long COVID in the CDCR system by date, in more (higher curve) and less (lower curve) stratified long COVID models, (C) estimated distribution of ever having long COVID through March 25, 2022, in the CDCR system and in California.

The estimated racial/ethnic composition of long COVID epidemiology in the prisons reflected the composition of the COVID-19 case burden (Figure 5C). As with acute COVID-19, Black people make up a much higher proportion of the estimated long COVID cases in the prisons than of the California population, at 24.1% (95% CI: 23.3%–25%) of the residents with long COVID in the more stratified model and 25.4% (24.5%–26.2%) in the less stratified one, in comparison to 5.5% (5.45%–5.56%) of estimated long COVID cases in California; 4.1% of California’s population is Black. Similarly, CDCR residents identified as American Indian or Alaskan Native make up 1.71% (95% CI: 1.49%–1.93%) of estimated long COVID cases under the more stratified model and 1.32% (95% CI: 1.13%–1.51%) under the less stratified one, while the American Indian or Alaska Native population makes up 0.432% (0.416%–0.447%) of estimated long COVID cases in California, and 0.2% of California’s population. See Supplement H for more detail.