Time Series Foundation Models - current state and future directions
Food Safety? A Matter of the Supply Chain: Probabilistic Risk Model based on the Agro-Food Trade Network
1. Food Safety? A Matter of the Supply Chain: Probabilistic
Risk Model based on the Agro-Food Trade Network
Matteo Convertino, PhD, PEng
2. 1) Food Safety is steadily decreasing life and economic losses
Issues (WHY?)
2) Emerging New Foodborne Pathogens (those causing illnesses that have only recently
appeared or been recognised in a population or that are well recognised but are rapidly
increasing in incidence or geographic range; ``secret bioterrorism agents’’ included)
3) Increasing Human Mobility, Social Contacts, and Agri-Food Trade (Game is considered in
the trade -> e.g. avian infleunza is a potential foodborne commodity disease)
4) Increase in Antibiotic Resistance. A poteriori treatment is not enough. Desire to have a
whole control in the food system’s state-space
Institute of Medicine, (2012)
Convertino et al. (2013), submitted
Food Safety Supply Chain
Matteo Convertino, PhD
3. Issues (WHY?)
5) High uncertainty (e.g. unreported cases), lack of information (e.g. trade), un-optimized
and unregulated dynamics (trade and international public health), and lack of strategies for
public health controls based on the sytemic network. Surveillance everywhere is not
efficient.
6) Increased Complexity of the Food that brings to multiple Risks (at least 4 countries
represented in a food that have 2-6 degrees of separations from the USA)
7) Increase concerns about Counterfeiting and Bioterrorism. A global real-time inspection is
not possible
8) Lack of a Quantitative based and Integrated System Design (and Technology for such
design) of the Complex Food System. Connection with Food Safety, Water, Biodiversity,
Economy, etc.
Convertino et al. (2013), submitted
Food Safety Supply Chain
Matteo Convertino, PhD
4. The case of Salmonella in Eggs
from Vietnam
Food Safety Supply Chain
Salmonella is the most common
Norovirus is a growing concern
Matteo Convertino, PhD
5. CDC data &
Painter et al. 2013
Evidence from Data
Increase of foodborne outbreaks in
time (on average this increase
corresponds to an increase in the
food supply chain length and
complexity)
Normalized Outbreaksfood
Dependency of the foodborne
incidence on the food supply chain
Food Safety Supply Chain
Incidence Foodborne Outbreaks
Matteo Convertino, PhD
6. Objectives
1) Identification of a Global Safety Index that consider both local country properties
and supply chain properties (related to the country)
2) Development of a Food Supply Chain Network (FSC) model for tracking existing risk
sources (paths, sources, food-pathogen-country triples, and their combinations),
predicting potential foodborne outbreaks, and minimizing the total public health risk
3) Create a model for establishing the FS backbone network and for making predictions
3) Global sensitivity and uncertainty analyses of the FSC model for a quantitative based
identification of the most important drivers of pubic health risk, and uncertainty of risk
calculations
4) Quantification of the what the Food Modernization Act states as Objectives (we
agree on those!)
Food Safety Supply Chain
Matteo Convertino, PhD
7. Ercsey-Ravasz et al., (2012), PLoS ONE
FSC is a sub-graph of the IAFTN. It is a tree for a
country and food selected. Rt as a function of lifecycle risks of food and FSC controls
kin
L
Dynamic Bipartite Network
Food Supply Chain available from data of GATS and US Food Market Estimator
(USDA ERS and US census for USA), FAO (FAOSTAT), UN (ComTrade)
Food Safety Supply Chain
Matteo Convertino, PhD
8. Example of Cereals Trade
The size of circles is related to the trade from the exporting countries to the USA and the blue
links are the backbone of the International Agro-Food Trade network
Food Safety Supply Chain
Matteo Convertino, PhD
9. Path Length: L the length traveled by a selected food commodity on average
Network
Variables
Connectivity (In-degree): kin is defined as the number of incoming links to a node
Edge weight: importance of an edge related to the total transport
(Betweenness) Centrality: is a measure that rates the importance of the position of a node
or an edge in the network with respect to transport through the whole network. mn(i)
denotes the number of highest total weighted paths from node m to n that are passing
through i, and mn denotes the total number of highest total weighted paths running from
node m to n
Node Relevance (Salience): the sum of all L divided by the number of connections
Food Safety Supply Chain
Matteo Convertino, PhD
10. Topological Properties of
the Food Supply Chain
for the USA
Log-normal epdf
Exponential epdf
Food Safety Supply Chain
Matteo Convertino, PhD
11. Topological Properties of
the Food Supply Chain
for the USA
A static assessment is
not enough; a
dynamical system
model that considers
both SC and risk
factors is needed for:
(1) understanding
system drivers; (2)
surveillance (of system
drivers); and (3)
optimal management
considering driver
interaction and
importance
Ercsey-Ravasz M et al., 2012, PLoS ONE
Food Safety Supply Chain
Matteo Convertino, PhD
12. Data
QALY = Quality Adjusted life-year
Just a statistical characterization is not
enough but it is necessary for informing the
model!
Need of a complex systems transdisciplinary
approach that
assesses the causative factors, their importance
and interaction for food safety
to prevent foodborne illnesses. SC factors!
Batz et al., 2011 (UF and CDC), 2012, EID; Morris et al., 2011
Food Safety Supply Chain
Matteo Convertino, PhD
13. Data, and Input Variables
12 food categories (complex food, poultry, eggs, fruits and vegetables, dairy, seafood,
beef, pork, other meats, bakery/cereals, game, and beverages), 77 food commodities,
and 187 exporting countries to the USA. We considering six degrees of separation from
source countries to USA and connectivity from IAFTN (this is a safe assumption)
We consider 8 types of complex foods (bread/pasta, cookies, cereals, canned soups,
typical first coarse and second coarse meals, ice cream, fast food meals) that are the best
selling and riskiest in the USA according to popular rankings
UNcom Trade, USDA Global Agricultural Trade System (GATS), and Food-Pathogen Risk
from Batz et al. (2012). Other various data complement the characterization of IF.
Food Safety Supply Chain
Matteo Convertino, PhD
14. Topological Properties & Safety Indices
from Rapid Alert System For Food and Feed
Nepusz et al. (2009,2012), PLoS ONE
Food Safety Supply Chain
Matteo Convertino, PhD
15. Model Keywords: Food Life Cycle, Supply Chain Network, Risks
Source (Adjacency) Matrix
Compositional Food Matrix
Food Matrix
Food
Life Cycle
Port of Entry
Food Safety Supply Chain
Matteo Convertino, PhD
16. Total Health Risk for the USA
Ranking of the riskiest food-pathogen-country triples. Monte Carlo filtering allows to assess
the combination of food-pathogen-country that contribute the most to selected total risk
value ranges
Convertino et al. (2013), submitted
Food Safety Supply Chain
Matteo Convertino, PhD
17. Total Health Risk
for each food
commodity
Disentangling the total health
risk for each food commodity
by selecting only the
respective food commodities
paths along the FSC
Our ranking is similar (but not
the same) of the one of Batz
and Morris (2012)
Food Safety Supply Chain
Matteo Convertino, PhD
18. Model Validation
Slope=0.76
R2=0.72
NSC=0.79
We believe that the model underestimates the risks due to the
large ignorance/variability of FSC and risk factors; however, the
model certainly gives a good assessment of the total health risk
for a country considering small scale risks and FSC controls.
Food Safety Supply Chain
Matteo Convertino, PhD
19. Total Health Risk versus FSC and safety variables
A multifactor characterization of the total health risk is necessary. GSI is a good
index to characterize countries based on their safety level and on their features
along the FSC. The risk is related to some risk factors and FSC variables.
Food Safety Supply Chain
Matteo Convertino, PhD
20. Driving Factors of the Total Health Risk
Global Sensitivity and Uncertainty Analyses – Morris Test, Monte Carlo Simulations & Filtering
Saltelli (2004)
Supply Food Chain
Network Variables
Food Pathogen Risks
Food Safety Supply Chain
Matteo Convertino, PhD
21. Management Strategies of the Total Health Risk
44% decrease
of Rt (by keeping countries
with GSI higher than 0.25 or
higher than 0.50 the risk
reduces from 0.43 to 0.24)
53% decrease
of Rt (from six
to 2-4 degrees of
separation)
Food Safety Supply Chain
46 % increase
of Rt (with removal of
countries has been
performed randomly by
excluding from the
importation 25%-50%
of the connected
countries without any
risk consideration
Matteo Convertino, PhD
22. Management Strategies of the Total Health Risk
GSI- and L-based management change the FSC in a
small world network that minimizes the risk and likely
maximizes the trade
Food Safety Supply Chain
Matteo Convertino, PhD
23. Rank based on the GSI
The hubs in term of trade are not always the Achille’s hell because they have very high SI
Countries with high SI but very high L , b, and k are unsafe (e.g. Singapore)
Countries with very low SI are unsafe
Matteo Convertino, PhD
24. Global Management on the FSC Network
Because of the strong dependence
of the Health Risk on the FSC
network we explore the
Health Risk as a
function of the network
topology considering all
possible topologies
The current FSC for the USA
does not minimize the risk
The small world FSC
network minimizes the
Total Health Risk
This analysis does not
consider Economical
and Political Constraints
among countries, thus some
networks may be unfeasible
Convertino et al. (2013), submitted
Food Safety Supply Chain
Matteo Convertino, PhD
25. Number of Observers for
Different Knowledge Level
of the Supply Chain
Matteo Convertino, PhD
26. Needs and What can be done with the FSC model
Despite the Need of:
- Better assessment of the factors Needs of Transdisciplinary Connected Research
(Engaging Stakeholders);
- Track of the information about the intermediate countries along the Agri-Food trade
and within the USA (within countries); and,
- Verification of the predictability of the model for real foodborne disease outbreaks
We have a model for:
(i) Potential early identification of emerging incidents (real time traceability)
(i) Information about surveillance of critical countries, trades, and supply chain paths
(i) Food policy evaluation at any scale, from the international to the town scale
(i) FSC for the design of a resilient (random or targeted attack that affect food safety)
and sustainable FSC (Environmetally (Water,Biodiversity) – Economically – Socially
(Food Security & Quality)). Optimization constrained to economical & political
constraints can be performed
(i) GSI is a great metric to assess country safety for food
Convertino et al. (2013), submitted
Food Safety Supply Chain
Matteo Convertino, PhD
We run the model considering any connection that the food can go through. Thus, in general the FSC is a large graph. If we select a country the FSC is a tree
Connectivity management at random is just a theoretical experiments, GSI management is smart (!). Length management without any safety consideration is not that good
Random placement is like if we don`t know the network …