Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

35
Big data and models: Are they really useful in disease management? J.F. Lowe Integrated Food Animal Management Systems Department of Veterinary Clinical Medicine University of Illinois

Transcript of Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Page 1: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Big data and models: Are they really useful in disease

management? 

J.F. LoweIntegrated Food Animal Management Systems

Department of Veterinary Clinical Medicine

University of Illinois

Page 2: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

What is in an engineer’s head?Measur

e• Inputs and

outputs

Model• Understand

how things are related

Fix • Implement solution

Page 3: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Data: The Hidden Key to Today’s Agriculture!

Small Data –

Why?

Big Data – What!

Opportunity For Synergy

Page 4: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Big Data (via Wikipedia)

• Amazon.com – millions of back-end operations every day, as well as queries from more than half a million third-party sellers.• Walmart – more than 1 million customer transactions every hour• databases w/ more than 2.5 petabytes (2560 terabytes) of data• 167 times the information contained in all the books in the US Library of

Congress.

• Facebook handles 50 billion photos from its user base.• Windermere Real Estate uses anonymous GPS signals from nearly

100 million drivers to help new home buyers determine commute times. • Volume of business data worldwide doubles every 1.2 years.

Page 5: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

The underlying view of big data?Society will need to shed some of its

obsession for causality in exchange for simple correlations:

not knowing why but only what.

Page 6: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

The Truth about Nutrition• The Japanese eat very little fat • suffer fewer heart attacks than do the British or Americans

• The French eat a lot of fat • suffer fewer heart attacks than do the British or Americans

• The Italians drink excessive amounts of red wine • suffer fewer heart attacks than do the British or Americans

• The Japanese drink very little red wine • suffer fewer heart attacks than do the British or Americans

• The Germans drink a lot of beer and eat lots of sausages • suffer fewer heart attacks than do the British or Americans

Page 7: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

The Truth about NutritionSO…………..

EAT what you want,Apparently it’s

SPEAKING ENGLISHthat kills you!

Page 8: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

So how do we leverage big data to

improve animal health?

Page 9: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Well we are doing it!!

Area Regional Control (ARC) projects!!

BUT WE COULD BE BETTER!!!

Page 10: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

So where are the gaps?• They are great at coordinating, at a local level, vaccination, pig

movement and herd elimination strategies to improve disease control.

• One of the weakness is that they do not provide a broad context for current and potential sources of new viruses to that region.

• Can even bigger data solve that problem?

Page 11: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

The challenges are scale and lack of uniformity!

Collaboration• building

collaborations to address complex, interrelated problems

Aggregation• aggregating

diverse data streams into meaningful data sets

Analysis• analyzing data

sets using robust tools

Synthesis• synthesizing

information into novel solutions

Page 12: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Well we are tying to solve those issues

Develop the data collection methods, infrastructure and collaborations between producers, veterinarians, and technical experts to facilitate dynamic, real time estimates of the risk of disease introduction into a pig farm to facilitate the development and deployment of the optimum prevention strategies for herds, local production ecosystems and regional production ecosystems

Page 13: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

2127 PRRS ORF5 sequences 783 IAV HA sequences

Participation – 572 sites across the Midwest

Page 14: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

• Phylogenic analysis rooted to VR-2332

• Excludes all vaccine like viruses

• Colors are linages that are >2% different at the nucleotide level

Page 15: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Expansion of a 1-7-4 like viral lineage outbreak over time and multiple states

7/1/2014 7/1/2015 7/1/2016

Page 16: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Subset of closely related 1-7-4 like linages that emerged in 2014, colored by production system/group

Page 17: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

To have understanding you have to know WHAT! Before you can know WHY!!

Society will need to shed some of its obsession for causality in exchange for

simple correlations: not knowing why but only what.

Page 18: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

So if we know WHAT with big

data, how do we drive

change?

Page 19: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

What is in an engineer’s head?Measur

e• Inputs and

outputs

Model• Understand

how things are related

Fix • Implement solution

Big and Small data

Page 20: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Transportation, PRRS and PEDV: An example

model

Page 21: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Background• It is common in the US swine industry to move pigs to

harvest plants on trucks/trailers (vehicles) that have not been cleaned and disinfected since the last load of pigs. • This practice was adopted with the implementation of all

in- all out (AIAO) growing pig sites • The risk of transport associated spread is exacerbated by

the fact that transport vehicles are often shared between different pig owners

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 22: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

One positive trailer in means 1.7 positive trailers at exit

PlantContaminated

at entryContaminated

at PlantContamination

RatioA 2.25% 8.05% 3.58B 7.00% 4.30% 0.61C 10.84% 10.81% 1.00D 2.00% 0.00% 0.00E 14.56% 3.08% 0.62G 3.00% 1.03% 0.34

All 5.98% 4.31% 0.72

Lowe et al, EID 2014

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 23: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

More contact increases risk Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 24: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

More virus increases risk Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 25: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

So what do we do about it????1. How do we change what we do at the plant to minimize

the risk of PEDV (or PRRS or XYZ) transmission between sites via the packing plant?

2. We are not going to wash trucks… so what else is there that we could do?

3. We don’t have all day… 5-15 minutes between trucks at the same dock.

4. We don’t an unlimited budget5. And by the way… no to plants are alike.

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 26: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

The Punchline• There is NO substitution for washing and drying trucks and all the

things on the trucks that come back from buying stations or harvest plants. • Everything that I am going to tell you today is a “less bad” approach

than doing nothing!!

•MORE POOP MATTERS!!!!!!

Page 27: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Steppin’ in the poop!

Page 28: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Does Temperature in the Hour After Plant Contact Change the risk of PEDV Transmission?

Dock Before Transmission

Trailer Time 0 Trailer Time 60262728293031

4°C 15°C 28°C All

Mea

n CT

Val

ue (L

og2

Tran

sfor

med

) NO!

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 29: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Does Temperature of the Dock Change the risk of PEDV Transmission?

Trailer/0 Trailer/5 Trailer/10 Trailer/60 Dock/50%

20%

40%

60%

80%

100%

25C 35C

NO!

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 30: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Does UV light on the Dock Change the risk of PEDV Transmission?

Trailer/0 Trailer/5 Trailer/10 Trailer/60 Dock/50%

20%

40%

60%

80%

100%

Lo UV Hi UV

NO!

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 31: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Does Removing Gross Contamination on the Dock Change the risk of PEDV Transmission?

Trailer/0 Trailer/5 Trailer/10 Trailer/60 Dock/50%

20%

40%

60%

80%

100%

Non-Scraped Scraped

YES!

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 32: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Does Dock Surface Change the risk of PEDV Transmission when physically debulking ?

Prior 0 5 10 200%

25%

50%

75%

100%

CONCRETE DOCK CONCRETE TRAILERPLASTIC DOCK PLASTIC TRAILER

Time Post Treatment

Perc

ent P

CR P

os

N=4 replicates

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

NO!

Page 33: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Conclusions and Next Steps• Scraping of manure at the dock between trucks may be an adequate

way to lower the risk of PEDV transmission below a threshold to propagate an outbreak.• It appears to work for other diseases (PRRS) – broader utility for the

next “unknown”• We have to model how much of a reduction is needed to minimize

the propagation of an outbreak.

Measure • Inputs and outputs

Model• Understand

how things are related

Fix • Implement solution

Page 34: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

Take homes:• In integrated, complex systems thinking and acting like an

engineer has value. • We have to know WHAT is happening!• BIG DATA can help us do that!• But so can “SMALL data”• But we have to understand HOW the process works in its native state

• Investigating multiple solutions at a small scale (aka a model) can be useful for understanding the robustness of solutions• Models are better at understanding solutions than “telling us how the

system works”

Page 35: Dr. Jim Lowe - Big data and models: Are they really useful in disease management?

J. F. Lowe, DVM, MS, DABVP [email protected]

Thank You