this presentation was given at the third Sund DairyCare conference in Zadar, Croatie. It discusses the need to have protocols to evaluate sensor technologies for their performance on-farm
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
Guidelines on the use of sensors to monitor animal health and productivity; a New Zealand initiative
1. Guidelines on the use of sensors for animal
health and productivity. A NZ initiative
Claudia Kamphuis
Wageningen University
2. Precision Dairy technologies are increasingly popular on dairy farms
Farm management becoming more complex
family-run to hired personnel
3. Precision Dairy technologies are increasingly popular on dairy farms
Farm management becoming more complex
family-run to hired personnel
small-scale to large(r)
4. Precision Dairy technologies are increasingly popular on dairy farms
Farm management more complex
and
Automation and sensor technologies available
5. But when it is time to invest in Precision Dairy technologies, there are
some issues farmers and suppliers encounter
Miscommunication between farmer and supplier
What he had in
mind
What he thought
he bought
What he actually
got
6. But when it is time to invest in Precision Dairy technologies, there are
some issues farmers and suppliers encounter
Miscommunication between farmer and supplier
Misunderstanding of provided information
epidemiological performance indicators
Sensitivity = A / (A+C)
Specificity = D / (B+D)
Positive predictive value = A / A+B)
Negative predictive value = D / (C+D)
Accuracy = (A + D) / (A+B+C+D)
7. But when it is time to invest in Precision Dairy technologies, there are
some issues farmers and suppliers encounter
Miscommunication between farmer and supplier
Misunderstanding of provided information
epidemiological performance indicators
meaningless for farmers
8. But when it is time to invest in Precision Dairy technologies, there are
some issues farmers and suppliers encounter
Miscommunication between farmer and supplier
Misunderstanding of provided information
No uniform representation of information
Different standards Different in- &
exclusion criteria
Different time
windows for alerts
N = limited
9. Farmers have no option to take fair well-informed investment decisions
Suppliers have no guidance in evaluation, benchmark and improve
10. Evaluation protocols will help
Offer a framework
to evaluate, develop, and
improve technologies for
suppliers
Providing robust, uniform,
and understandable
information for farmers
Kamphuis et al., 2013, JDS
10
http://www.dairynz.co.nz/media/785859/Protocol-document-310513.pdf
11. protocols for mastitis detection systems
Farmers want
Find cows with
clinical mastitis
promptly and
accurately
Manage
bulk milk SCC
Support dry-cow therapy decisions
12. For both objectives protocols describe
(Kamphuis et al., 2013, JDS; Kamphuis et al., 2015, under review)
A practical gold standard
flakes/clots in 2 out of 3 milkings
Description of data collection, in- and exclusion criteria
three commercial farms
≥20 CM episodes / farm; ≥60 CM episodes in total
record cowID, stallID, milking session, pics for density scores
collect sensor information (mastitis alerts)
13. Method for evaluating alerts against gold standard
13
For both objectives protocols describe
(Kamphuis et al., 2013, JDS; Kamphuis et al., 2015, under review)
14. Method for evaluating alerts against gold standard
Method to present understandable performance information
Sensitivity per farm and combined
At different levels of false alerts per 1,000 milkings
14
For both objectives protocols describe
(Kamphuis et al., 2013, JDS; Kamphuis et al., 2015, under review)
15. protocols have been evaluated
Applying it in the field (Kamphuis et al., 2015, under review)
Showing results and presenting protocols during workshop
7 main NZ companies of monitoring technologies
3 farmers
3 representatives of dairy industry
16. Protocols have been evaluated
Field evaluation was OK in practical sense
Suppliers representatives expressed general support
16
17. But now the next steps are required
Other Precision Dairy technologies
Use DairyNZ initiative as stepping stone
automated heat detection
Other areas that are of interest for farmers
‘Proof’ the value of technologies
Investment model for mastitis detection
18. But now the next steps are required
Suppliers need an incentive to use protocols
farmers asking for protocol-approved systems
private companies and/or wider dairy industry
International awareness
publications in international journals/books
Further international collaborations
IDF workforce to further refine DairyNZ protocols
20. Thank you for your attention
www.slideshare.net/claudiakamphuis
Editor's Notes
Thank you for your kind introduction chairman, and thank you for inviting me here. Thank also to you who are still here at the end of a second intensive day of the DairyCare conference. This presentation is about the necessaty of having guidelines on the use of sensors for animal health and productivity, and the initiative NZ took to get to these guidelines.
First let me start with a brief intro on sensors or precision dairy technologies. We know that these technologies are increasingly popular on dairy farms, and that popularity can be explained by two drivers
The first driver is the increasingly complexity of farm management. And farm management becomes more complex because we are changing from family run farms to business that hire personell
And we going from relatively small scale farms to large farms, and we’re still growing in herd sizes. Also in New Zealand, where you now also find barns in which cows are housed year-round.
At the same time, there are automation and sensor technologies that are being developed to make that management easier. We have automatic drafting gates, and automatic cup removers that make the labour on farms easier, but also technologies that help monitor cow health and productivity, like walk-over weighers, sensors that monitor milk component and that technologies that use activity measurements to identify cows in heat
So these two factors explain why farmers are interested in precision dairy technologies
But when it comes to the time that they want to invest in these technologies several issues are encountered...by them, as well as by those that sell these technologies.
Because often you see some huge gaps between what farmers want from these systems, what they thought they bought, and what they actually got. I don’t want to say here that suppliers offer systems that are no good, but expectations are often different from reality. Farmers think mastitis detection systems for example will help them find all cows with mastitis. But they will not. And they perhaps have not thought about, or have not been told, that these systems generate false alerts too, so they still need to visually check the information that is provided by these tools
Another difficulty for farmers is that when they are looking for independent information how these systems work, they often end-up with scientific publications, if they even get this far. But these scientists often use epdidemiological terms to express how well these systems work. Sensitivity is probably the easiest term for them to understand, but terms like specificity, PPV+ NPV- and accuracy...
That’s a language that they don’t speak and don’t relate to
Finally, when they do have some information (perhaps not from scientific publications) on performances of different sensor systems, it’s impossible for him to compare the information because the information that is provided is often based on different gold standards
Some use treated cases of mastitis, others increased scc-levels, or yet another criteria to define mastitis
Or, they used different in- and exclusion criteria to develop and validate the performance
Or, they use different time-windows in which they allow their sensor to alert for an event. With mastitis, this ranges for example between 2 weeks before or after a mastitis event, to 24h preceding an event. You can imagine that this first time window will result in much better performance that the second one, although the second one is probably more fair from a practical point of view.
And finally, very often performance of technologies are based on data collected on one farm, which is often not a commercial farm but a research institute, and thus not very similar with commercial farms.
And on the other hand, suppliers have no guidance in how they should evaluate they gear. And thus, they cannot benchmark themselves to others nor can they improve their gear.
So, what can performance evaluation protocols or guidelines offer?
Well they can offer suppliers a framework to evaluate their gear, and to develop or improve their technologies
And by offering suppliers a framework for evaluation, farmers will have access to robust, uniform information that they actually can understand or relate to
And it is here where the NZ initiative comes in because DairyNZ accepted the challenge to develop these protocols for in-line mastitis detection systems. These protocols are published and are available at the DairyNZ website.
During the development of these evaluation protocols dairyNZs starting point was the farmer and what he wants from automated mastitis detection systems. And dairyNZ defined three objectives, farmers want these system to help them find cows with CM promptly and accurately
They want these system to help them find cows with high SCC levels to manage BMSCC
And they want these systems to help them with individual dry-cow therapy decisions.
The protocols also offer a detailed description of how data should be collected,
The guidelines describe a method to combine the data sources and to evaluate the alerts from the software agains the gold standard.
And in case for CM detection, the protocol explains the definition of a CM episode and in explains the time windows in which alerts from these systems are considered true positive and false positive.
And then finally, the protocols describe a method to present the evaluation ysing understandable performance indicators.
So, sensitivity of a system is presented per farm, and combining data from all farms together. And we suggest to use two different levels of sensitivity, the first one indicates how accurate and timely a sensor system identifies CM, and the other level represents how accurate a system albeit not always timely. These two sensitivity levels are presented at a number of false postiive alerts per 1,000 milkings. This indicator relates to specificity, but it’s much easier to understand for farmers. For example with a FAR of 10, a farmers that milkes a 500 cow herd twice daily can expect 10 false alerts each day, to 5 per milking session.
Well, these aspects are defined for both objectives for in-line mastitis detection systems, and these proposed protocols have been evaluated in the field (results of that are currently under review) and we showed the results of this field evaluation and demonstrated the use of these proposed protocols during a workshop. And this workshop was organized with 7 representatives of main technology providers in NZ, including international providers, but also 3 farmers were present during the workshop and three representatives of the NZ dairy industry
From the field evaluation and this workshop it was concluded that the protocol as such were OK in a pratical sense. Data collection for both CM detection as management of BMSCC levels were relatively easy and no main issues were encoutered. Moreover, the supplier representatives expressed their general support for having these protocols.
And that is certainly good news
But we still need some other steps.
We now only have protocols proposed for in-line mastitis detection systems, but we also need them for other sensor technology. The dairyNZ initiative can be used as a stepping stone for developing protocols for other technologies and in fact DairyZN is currently working on protocls for evaluating automated heat detection systems
Moreover, performance if sensor technologies is not the only area of interest for farmers when they are bout to invest in these sytems. We also need to think of methods or tools that can help suppliers to ‘proof’ the value of technologies, and this includes the social and the economic value. In this sense, dairyNZ is currently working on a investment model for automated mastitis detecctoin systems, to help farmers answer the question whether investment in these systems will be economically beneficial.
But, procotol need to be used too and suppliers need an incentive to use them. This incentive may come from farmers being aware of the protocols and asking for protocol-approvied systems or from private companies or actors within the dairy indyustry. But I guess the first step in this incentive is to raise international awareness of the existence of the protocols and what they can offer for farmers and suppliers. This international awareness can be created by publication in scientific journals of hand books, and by further international collaboration. That NZ technolgy suppliers know about these protocols is great, but we need this to become internationally knwon. It may help that the IDF decided to have a workforce that will focus on the further refinement and awareness of the protocol proposed and initaited by DairyNZ.
But the awareness of course starts here, in Zadar. And i guess dairycare can be a platform or at least inittied the international collaboration, so it’s reallyu great thay dairycare allowed me to share this story.
And with that I would like to thank you for your attention, I would also like to discuss this topic further and perhaps we can use the precisoin dairy farming conference that will be hedl next year as a occasion to present some developments rageding protocols to evalaute sensor perofrmance.
This presentation will be available on slideshare later today or tomorrow and with that I end thes presentaion, thank you.