Numerous 21st century pandemics, such as the 2003 severe acute respiratory syndrome (SARS-CoV) outbreak and repeated avian flu outbreaks (H5N1, H7N9), have raised public attention to human health risks of zoonotic diseases, which transmit from animals to humans. Other outbreaks such as Middle East Respiratory Syndrome (MERS-CoV), swine influenza virus (SIV), brucellosis, have all originated from zoonosis (1-8).
Notably, many of these outbreaks, including the 2003 SARS-CoV outbreak, were associated with large Chinese wholesale markets (WSMs) and smaller wet markets (WMs), both of which play a unique and prominent role as consolidation and distribution points in China’s agricultural supply chains (9, 10). While the origin of COVID-19 is debated, one hypothesis ties the first known cluster of COVID-19 to the Wuhan Huanan Seafood Market, and a later cluster to the Beijing Xinfadi Seafood Market (11, 12). Likewise, SARS-CoV was associated with markets in Foshan, brucellosis cases in 2005-2010 were linked to pork markets in Guangdong, and H7N9 cases in 2013-2017 were associated with poultry markets in Shanghai and Anhui (3, 4, 7, 13). Existing hypotheses suggest that illegal sale of exotic wild animals and sanitation issues at markets are potential causes of outbreaks (14).
Concerns regarding WSMs and WMs are underscored by other human health risks associated with them, specifically, food safety and particularly food adulteration risks, such as illegal use of antibiotics and antimicrobials (9, 17, 18). This high level of food adulteration risk is exacerbated by the combination of opaque supply sources, sales of fresh and live animal products, and inconsistent supply chain conditions (e.g., cold chain), logistics and management infrastructure. The Chinese government has invested substantial regulatory resources to address challenges related to food safety and adulteration with increasing focus on WSMs and WMs, including regular testing of food (17).
Recently, there have been voices that call for fundamental reform of WSM/WM management and even suggestions for closure of all markets (15, 16). While closing markets could alleviate zoonotic disease and food adulteration risks, this proposal ignores the central role they play. For many agricultural products, WSMs consolidate 70-80% of national supply, while WMs serve as the primary sales channel to individuals and restaurants (9, 19). Thus, eliminating these markets from the food system in China is logistically challenging and perhaps infeasible (10). Furthermore, closing markets would adversely affect the livelihood of millions of market vendors, farmers, and transporters (10) and create social harm. This suggests a need for practical approaches that steer regulatory resources towards riskier markets and improve management at WSMs and WMs, to mitigate public health risks they pose. It is noteworthy that a recent article published on The Lancet Planetary Health also argues that “safe trade and consumption of wildlife could align with existing global food safety regulations in agreement with the precautionary principle and in support of the UN Sustainable Development Goals”, which also suggests to focus on enabling smarter and targeted regulation as opposed to shutting down the entire wildlife trade (20).
This paper contributes to such an approach by leveraging machine learning to generate WSM/WM market-level food adulteration risk scores from public data and demonstrating their correlation with zoonotic disease in China. In particular, the paper provides empirical support to the hypothesis that these two types of risks (i.e., food adulteration and zoonotic disease risks) are correlated, potentially because they are both affected by common system-level environmental and management deficiencies in WSMs and WMs. Moreover, this suggests that assessment of food adulteration risks can meaningfully inform the assessment of zoonotic disease risks at the market level. This observed correlation stands in contrast to the fact that zoonotic disease and food safety risks are currently regulated by different Chinese government organizations.
To accomplish this, the paper leverages a self-constructed dataset of 4·0 million public (passing and failing) records of food adulteration tests conducted by China’s state (central), provincial and prefecture-level Administrations for Market Regulation (AMRs), which are the primary organizations responsible for testing food products in WSMs/WMs (9, 21-23). This dataset integrates all tests conducted between 2014-2020 and posted by the national AMR, all 31 provincial AMRs, and 273 of 333 prefecture AMRs, covering nearly all major cities and important agricultural areas. This dataset is widely representative of true food adulteration risks throughout China, as AMR organizations are legally required to publicly post all food test results on their respective websites by the State Council of China (24), AMR’s policy instructs a random sampling approach within each product category, meaning that the number of tests at each market is roughly proportional to the volume of products sold at the market (25), and all AMR food safety tests conform to the same methods and standards defined in China’s national testing plan (17).
This dataset includes 79,177 test records of animal products associated with WSMs/WMs, and an innovative unsupervised clustering methodology is developed to associate each test record with its respective market. This methodology yields a list of WSMs/WMs in China, as well as a food adulteration risk score for each market equal to its respective failure rate in the AMR food safety test dataset. The food adulteration risk scores define high-risk markets (top 20% highest risk scores). The total volume of animal products sold through high-risk markets in each province provides province-level risk scores. While these risk scores are not intended to capture all nuances of food safety tests, such as adjustments for specific types of animals tested, they create significant decision support information for regulators. Notably, there previously existed no publicly available and comprehensive list of WSMs/WMs in China, let alone a method for curating all AMR tests at each market. This motivates the creation of an operational tool to communicate information to regulators, policy makers, and academics.
The association of food adulteration risk scores to zoonotic disease risks is validated through two analyses. First, a carefully designed linear regression indicates that province-level risk scores correlate positively with respective number of zoonotic influenza isolates (cases) in humans identified from the OpenFlu database (26). This association holds when controlling for covariates, such as livestock production/slaughter, and distribution of markets in each province. An overview of the analysis is shown in Figure 1. Second, specific markets associated with zoonotic outbreaks have higher risk scores than expected by chance. These analyses provide evidence that constructed risk scores give a single interpretable measure able to capture important aspects of both food adulteration and zoonotic disease risks at the market and province levels.