A deep dive into a previous publication on mortalities and negative health outcomes in North Carolina residents living near hog CAFOs unveiled that the study approach was insufficient to assess the intended hypothesis of associations or support the causative associations implied in media reports. The technical soundness of the previous study data and analysis had several shortcomings including ecological fallacy, residual confounding, inconsistency of associations, and over estimation of exposure. The repeatability and confidence of the associations between exposure associations (odds ratios of the logistic regression) were weak and varied between the years (Tables 02 and 03). Discounted for the fact that corrections for confounding was not successful, the six diseases selected by the previous study exhibited varying levels of associations, diseases that are not causally relatable to hog CAFOs–HIV and diabetes–were also prominent in the communities neighboring CAFOs. The associations in fact reflects systemic health disparities of the communities near hog CAFOs, which were not caused by the exposure.
An ecological study would not confirm or deny exposure over time nor indicate causation. While it is understandable that collecting individual-level data for all the confounding factors in a retrospective study is nearly impossible, that does not justify for not correctly adjusting for the key social determinants of health in the individual-level logistic regression models. Except for age, all other confounding factors used in the individual-level logistic regressions were deducted from aggregated data, which indicates a classic case of ecological fallacy. The odds ratios in regression models did not vary more than 10%, when the simple model was compared to the model with confounding factors (Supplementary Table 03), which leads to the interpretation that none of the six factors proved to be true confounding factors. This suggests the associations are not adjusted for confounding and the presence of residual confounding.
Aside the unclear rationale behind the choices, the previous study selected five disease hypothesized to have associations with hog farm exposure—anemia, kidney disease, septicemia, tuberculosis and low birth weight. However, these known associations were not supported with evidence of causation or systematic review. Especially, human tuberculosis, which is commonly caused by Mycobacterium tuberculosis and the zoonotic origins are commonly related to M. bovis; a Mycobacterium species relatable to cattle. Human tuberculosis relatable to the origin of swine is extremely rare. Tuberculosis is often reported to affect marginalized populations, including racial minorities, migrant and seasonal farmworkers (Ciesielski et al., 1994; Stout et al., 2006). Without the ability to trace back and investigate the potential exposure to tuberculosis, the claim of tuberculosis cases being high in the zip code with CAFOs is not plausible. In our repeated analysis, we included HIV and diabetes, two chronic disease conditions that are not causally relatable to hog CAFOs. Results showed that across the board, HIV and diabetes rates are higher in these North Carolina communities (Tables 02 and 03), indicating profound and systemic health disparity in the communities in general.
The variability of associations across diseases and time periods in terms of the direction of association (ORs > 1 or < 1) and P-value based statistical significance were inconsistent to make the claims by the previous study. For example when compared the mortalities between the previous study and the re-analysis of ’07-’14 data (Table 02: Group 2 primary + secondary diagnosis), only kidney disease showed a consistent statistical significant result wit OR > 1. Within our reanalysis itself, the associations for mortalities agreed between the ’07-’14 and then ’15-’18 for septicemia only. This exemplifies the danger of implying exposure associations based on a single ecological assessment of data, especially when the associations were not corrected for confounding. Across all the associations, ’07-‘14 results of kidney disease had 9 of 12 odds ratios > 1 (related to mortality, hospitalizations, and ED visits) with statistical significance, but only 5 of 12 anemia odds ratios showed these results. Five of 12 septicemia odds ratios actually showed statistically significant < 1 relationships. In fact, the results we found for HIV and diabetes were among the most consistent and significant results; all odds ratio results for hospital admissions and emergency department visits in the ’07-’14 re-analysis were > 1 and statistically significant i.e. zip codes with hog CAFOs had greater odds of having HIV and diabetes than the control group zip codes.
The key question to answer using an epidemiological approach is the counterfactual scenario whether these communities hypothetically were living in an area where there were no hog CAFOs would they still have the same negative health outcomes. A study by Son et al., (2021), conducted a comparable study to Kravchenko et al., (2018) using a spatial analysis and concluded that the communities near CAFOs are disproportionally affected. However, both the studies neglected the alternative explanation that zip codes with hog CAFOs have specific demographic characteristics with greater proportions of African American and American Indian residents, lower median household incomes, fewer residents with a bachelor’s degree education level, and fewer primary care providers. These four aspects of race, income, education, and health care demographics have all been well established as social determinants of health (Sources: health.gov links, Braveman 2011 and 2014, Catalyst 2017). North Carolina, as a state, faces multiple demographic, economic, and environmental challenges including extreme racial, gender, and income inequities (Mitchell, 2015; Wilson et al., 2002; County Health Rankings, 2021). Racially uneven development of the neighborhoods in North Carolina may have disproportionately exposed rural communities to agricultural and industrial toxicity (Bullard et al. 2007; Wilson et al. 2011; Wing et al. 2000). This emphasize the need for focusing solving health disparities in these communities, instead of falsely blaming the agriculture. Previous studies also suggested that while minorities are not directly targeted for exposure to hog farm locations, but are disproportionately exposed and that this disparity may relate to poverty and being a rural population (Yeboah et. al., 2009). Therefore, instead of ecological studies or survey-based opinion pieces that can be detrimental to the economy, we emphasize the need for conducting better epidemiological studies that are spatiotemporally explicit, and involve systematic reviews to evaluate hog CAFO exposure-over-time and evaluating the cause-effect.
Through the exercise of re-analysis, we identified a few ways to improve the assessment. First, in the absence of individual-level confounding variables, the analysis should have been conducted at zip code or county level, ideally using a spatial regression model. This is because the neighboring administrative units pose an effect on each other and accounting for such spatial dependence is vital when establishing risk factor association. Secondly, there are multiple other social determinants of health need to be considered as confounding factors and having an index that can represent overall health inequality is the best instead of collecting data from several unrelated sources (Berkowitz et al., 2015). The CDC/ATSDR Social Vulnerability Index (CDC/ATSDR, 2018) would be an example of such an indicator where socioeconomic status, household composition, minority status, housing type, transportation were all considered when the SVI index was assigned to each county. The key differences in our re-analysis and previous study are that we did not calculate the age-adjusted rate comparison tables, did not conduct ad-hoc comparisons of mortalities with other selected U.S. states, and did not repeat the DiSC analysis-based logistic regression because we cannot compare that to the geostatistical methods we proposed to better estimate the distance-based exposure.
Hazard identification and risk quantification in environmental epidemiology is challenging (Saracci, 2017). Potential environmental contamination including soil, ground water, and air contamination relatable to CAFOs is a broad topic that cannot be captured using the ecological study used. Hog farms are managed by farming systems and each farm is different in their management practices, building structure, lagoon structure, and maintenance. Therefore, the use of hog numbers is not an accurate representation of the potential exposure. If the exposure to be truly enumerated several factors are to be considered and the exposure should be assigned based on farm characteristics, length and routes of potential exposure, and the distance to the residence. This need to be further supported by air, soil, and water quality indices. Similarly potential environmental exposure through land application need to be measured differently in terms of manure consistency, hydrology, runoff, soil types, and watershed characteristics. Number of hogs would not offer a proxy for the exposure. An assessment that incorporates duration of exposure and the outcome, after correctly adjusting for the major confounding factors are needed to establish such association. Further confirmation of associations are needed to be done by comparing several similar studies (i.e. systematic revision). Compared to the Midwest where manure digestion is done in earthen deep pits, swine waste management in North Carolina is predominantly done via uncovered lagoons, where animal waste undergoes anaerobic digestion in a man-made ponds (USDA: Agricultural Waste Management Field Book, 2011). However, the swine waste disposal systems and land application of manure is highly regulated by North Carolina Department of Environmental Quality and the state legislature (https://deq.nc.gov/; Swine Farm Siting Act, 1995; General Permit, 2019). Expanded CAFOs are required to use environmentally superior technologies (ESTs) to substantially reduce emissions and prevent waste discharges into surface and ground waters (XXX Legislature XX; Nicole, 2013). Regardless of the fact that associations implied here are not well supported, several farming communities are working towards sustainable environmental goals where the health of people, animals, and the environment is being at the center of their focus (Han et al., 2001). Hog farming as an industry faces challenges of animal diseases, biosecurity, and securing the health of farm workers while making profits. Changes in farming culture is costly, labor and technology intensive. Proposing a true One Health approach where human, animal, and environmental health would be a solution for the issue where stakeholders representing the relevant government, swine producers, community members discuss and plan together multiple solutions and plans for follow-up and follow-through.