Unlocking the Means of COVID-19 Spread From Prisons to Outside Populations

Prisons, jails, and similar facilities may foster and exacerbate viral outbreaks in certain areas of the United States. By their nature, they are semipermeable systems where the majority of those in the system live in close quarters, are largely excluded from the surrounding population, and possess severely circumscribed autonomy, mobility, and resources. The systems also include a somewhat smaller group of staff and visitors who cycle between the facility and the outside population much more frequently, interacting with both.

Clearly prison conditions are conducive to COVID-19 becoming a health threat to those in the system. What is less clear is the potential for that threat to translate to a public health threat to populations neighboring these facilities. Assessing the mechanisms behind prison to population virus transfer is further complicated by variations in factors such as the size and type of facility, reporting and disclosure, infection response protocols, prevailing politics, vulnerable versus nonvulnerable population size, and the variety of competing explanations for virus transfer.

In “Prisons and COVID-19 Spread in the United States,” Sims et al. (p. 1533) add to a growing body of research on prisons and COVID-19 by investigating the possibility of virus transmission to surrounding communities. This interesting work is focused on the first wave of the pandemic as a model of what to consider for the onset of future pandemics. Those who are comfortable with reading statistical output may wish to pursue their own line of inquiry by consulting the article’s detailed appendix.

Their findings, in brief, are that prisons are significantly associated with increased COVID-19 infections, but not deaths, in the counties where they are located. When broken down by type, the relationship between correctional institutions and increased infections in the county are significant only for state prisons, as opposed to federal prisons or jails. When considered from the time of virus onset, the number of cases of COVID-19 peaks around the 60-day mark and decreases thereafter.

Although this work is methodical and logically grounded, Sims et al. are careful to stipulate that, for a variety of reasons, they are unable to draw causal inference from this work. Three commonly accepted criteria for establishing a causal relationship between the cause and perceived effect are that (1) the cause precedes the effect, (2) the cause is related to the effect, and (3) plausible alternative explanations for the effect have been ruled out.¹ Sims et al. establish both temporal precedence and the relationship between their proposed cause (prisons) and the effect (increased COVID-19 infection). As the authors mention, it is currently not possible to rule out the possibility that the observed disparity in COVID-19 cases was attributable to greater frequency of testing in prisons. Gaps and inconsistencies in how or whether institutions report COVID-19 testing and infection data in prisons make this impossible. However, other research reports that when COVID-19 testing was conducted in prisons, something that mainly occurred in state prisons, the infection rate was found to be much greater than that of the neighboring population.²

The regression models explaining COVID-19 infection rates also stymie causal inference. Although the binary variable “prisons” was significant and sturdy, the model’s explanatory value does not increase with its inclusion. Tables C and D, included in the Sims et al. Appendix, reveal no appreciable difference in R² values for a model that controls for prisons (Table D, model 2, R² = 0.845) and an otherwise identical model that does not (Table C, model 1, R² = 0.845).

This inability to establish a causal relationship is not unexpected, given the severe constraints on available data on COVID-19 in prisons. In fact, the same problem arises in other, similar studies. The Prison Policy Initiative supported a report by Hooks and Sawyer³ that also sought to establish the link between the number of incarcerated persons per square mile and the spread of COVID-19 to the surrounding community during the first wave of the pandemic. Using a smaller set of similar covariates, Hooks and Sawyer similarly found a significant relationship between the density of internees and increased COVID-19 case load in smaller counties, but no significant relationship with deaths attributable to the infection. In another study, funded by the American Civil Liberties Union, Lofgren et al.⁴ used a SEIR (susceptible-exposed-infectious-removed) model to simulate the rate of infection and death inside prisons and in the surrounding community under a range of different policies during the first wave of a COVID-19–type pandemic. In that study, models employing shelter-in-place policy severely decreased the risk to the surrounding population but predictably demonstrated little effect on those in the prison. Otherwise, models based on decreasing prison populations offered respite to those in the prison but offered a less notable decrease in the risk of infection to the surrounding population.

It should be noted that each of the above studies lacks a causal link between infection rates in prisons and infection rates in their surrounding populations. What each of these studies has in common is the agreement that prison populations are particularly vulnerable during the first wave of a disease event and they are not independent of the surrounding population. This intimates that the problem situation of prisons and virus transmission is not well structured, which limits the options available for effective policy response.

Prisons are not health problems in and of themselves. Rather, their structure and their relationship with the surrounding population makes them a problem situation that is subject to multiple competing conceptualizations of the actual problem. Studies such as those conducted by the American Civil Liberties Union and the Prison Policy Initiative conceptualized the problem as having to do with the size of the prison population. Alternatively, Sims et al. structured their recommendations around public health infrastructure and response coordination. It is certainly plausible that any or all of these conceptualizations offer useful policy alternatives. But this is hardly comprehensive.

Given the complex nature of this problem situation, a strong next step would be to better structure the problem of disease transmission between prisons and the community. Each of these studies share a gap between inference and prescription. Each analysis was intended to reveal the presence of a problem situation. The policy suggestions, however, are either solutions in search of problems or unsubstantiated afterthoughts.

Problem structuring is a frequently neglected process of comparing, testing, and contrasting multiple formulations of what actually constitutes the problem.⁵ Whereas the problem situation appears to be the role of prisons as infection incubators and the transfer of that infection to outside communities, the actual problem is associated with the mechanisms through which the infection is transferred (e.g., visitation and consultations, incarcerated person transfer), exacerbated in the prison (e.g., impedance of preventatives, resistance to rule changes), or other aspects that may give rise to an undesirable outcome.

The variety of policies that were instituted for prisons in response to COVID-19 throughout the country⁶ present the opportunity for quasiexperimentation that would aid in the problem-structuring effort by comparing their relative successes. The initial policy chaos that characterized the initial COVID-19 response in the United States could, therefore, be of benefit. But, careful work should also be done to reveal other promising, but untried, options.

In many ways, the work of Sims et al. raises more questions than it answers. But, it is important to get this work into public knowledge and onto the policy research agenda.