Source: CORNELL UNIVERSITY submitted to NRP
DEVELOPMENT OF PREDICTIVE MODELS TO ACCURATELY DETECT SUBCLINICAL AND CLINICAL MASTITIS IN DAIRY COWS MILKED WITH AUTOMATED MILKING SYSTEMS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
Reporting Frequency
Annual
Accession No.
1020878
Grant No.
(N/A)
Cumulative Award Amt.
(N/A)
Proposal No.
(N/A)
Multistate No.
(N/A)
Project Start Date
Oct 1, 2019
Project End Date
Sep 30, 2022
Grant Year
(N/A)
Program Code
[(N/A)]- (N/A)
Recipient Organization
CORNELL UNIVERSITY
(N/A)
ITHACA,NY 14853
Performing Department
Vet Population Medicine & Diagnostic Science
Non Technical Summary
Our project aims to improve milk quality and udder health with automated milking technology by identifying udder health alerts sooner, which will allow for a quicker udder healthintervention. Udder health should be improved by quicker decision making when udder alerts occur with the intention of increasing cure rates from intramammary infections.In addition to opportunities to improve milk quality, our project will open opportunities for monitoring reproduction, health, lameness, and transition cow diseases with AMS technology. Further, there is opportunity to implement our models in any milking system with milking time data. The collected data could help future scientists link SCM to other diseases that it may be influencing. For example, is there a negative effect of SCM on reproduction? Conversely, the retrospective milking characteristic data on a CM or SCM quarter could be used to determine possible causes of mastitis. In this way our grant becomes a way to proactively prevent CM and SCM events instead of simply reacting to them.
Animal Health Component
0%
Research Effort Categories
Basic
(N/A)
Applied
0%
Developmental
100%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
31134101170100%
Knowledge Area
311 - Animal Diseases;

Subject Of Investigation
3410 - Dairy cattle, live animal;

Field Of Science
1170 - Epidemiology;
Goals / Objectives
1. Identify quarter level milking data points from automated milking systems (AMS) that are associated with the onset of clinical mastitis (CM) and subclinical mastitis (SCM).2. Develop a predictive model that will allow AMS managers to identify at risk cows with subclinical and clinical mastitis.a. The development of a model, with quarter level milking time data points associated with CM and SCM will allow AMS managers to identify cows at risk for CM and SCM accurately.3. Determine if the additional quarters of an individual cow that did not flag with the model for CM or SCM are considered healthy by having a SCC < 200,000 cells/ml by documenting the SCC of all quarters when a cow has CM or SCM.4. Report results to producers that have AMS and companies that make the systems.
Project Methods
Aim 1: Collect individual quarter milk samples from all cows and submit these samples for SCC testing. Record and analyze all milking time data from the AMS software to determine which variables are associated with an increase in quarter level SCC.Study design: Individual quarter milk samples from all cows will be analyzed for SCC and aerobic culture analysis on milk samples from cows with CM. Quarter level milking time records will be collected from the AMS software. After analysis of milk samples for SCC, cows will be assigned to one of two cohorts: healthy or subclinical. Milk samples with a SCC < 200,000 cells/mL will be classified as healthy whereas milk samples with a SCC ≥ 200,000 cells/mL will be classified as subclinical. Clinical mastitis will be identified visually by technicians for clinical signs of abnormal milk or quarter inflammation. We will collect the following AMS data using DelPro (DeLaval, Tumba, Sweden): teatcup on time, milk yield/milking, milk deviation, milking frequency, milking interval, average milk flow rate, electrical conductivity, kick-offs, attachment failures, incomplete milkings, parity and more variables not listed here.Sample collection: We will collect quarter milk samples from 180 cows after milking with a DeLaval AMS. We will enroll cows from a New York State commercial dairy operation using AMS. Samples will be collected on Mondays, Wednesdays, and Fridays for 50 sampling periods (~120 days). Given that cows show signs of clinical mastitis for 3 - 5 days, sampling on Mondays, Wednesdays, and Fridays is intended to identify most if not all cows with CM. Individual quarter milk samples will be submitted to Dairy One, Ithaca, NY for SCC analysis by Fossomatic Cell Counter (FOSS, Hilleroed, Denmark). Individual aseptic quarter milk samples will be collected from cows identified with CM on one or more quarters and submitted to QMPS, Warsaw, NY for aerobic culture.Milking time data collection: AMS milking time variables will be exported daily from DelPro to Excel for analysis.Identification of influential variables: Data from the AMS and quarter milk samples will be merged and cleaned to ensure proper integration of data and to control for outliers. Simple Pearson's correlations between potential independent variables and the incidence of CM and SCM will be used. Further, correlations between variables will be assessed to identify a pool of potential variables for inclusion in a predictive model. In addition to the raw data, quadratic transformation of continuous variables like flow rate and conductivity will also be included to check for non-linear relationships. Further, the change in observed continuous variables between days will also be included as an additional transformed variable.Aim 2: Predictive models of CM and SCM will be developed in R statistical software. An autoregressive multivariate logistic regression mixed model to accommodate binary outcome variables, multiple records from one individual over time, and the random effects of the individual cow will be developed using the glmer() function in the lme4 package. A forward stepwise selection process will be used including variables identified in Aim 1 in the order of their correlation with the outcome variables. Model goodness of fit will be assessed with McFaddens R2 and the Hosmer-Lemeshow test. Finally, the model's predictive ability will be assessed with an ROC curve and Leave-One-Out Cross Validation.Revisit farm after model development to validate AMS model. Validation of model will require collecting milk samples from cows that are identified to have CM and SCM and analyzing these samples for SCC and aerobic culture.Aim 3: Analyze the individual quarter SCC data from the samples gathered for Aims 1 and 2 to evaluate the risk of quarters with visually normal milk having an elevated SCC if the cow has a SCC ≥ 200,000 cells/ml on one or more quarters.Aim 4: Upon completion, we will host a meeting to share our results.

Progress 10/01/19 to 09/30/20

Outputs
Target Audience:We collected 17,680 individual milk samples from one dairy farm over 30 sample periods and the samples were analyzed for somatic cell count (SCC). The owner of the dairy operation was provided the SCC information and could make udder health related decisions with the data. The initial scope of the project outlined that we would collect milk samples from 180 cows, three days/week for a total of 30 sample periods, but in the end we collected samples from 250 cows. As we analyze the data the number of 250 cows on project will be decreased as we identify cows that may not be eligible for analysis. There were 17,680 milk samples analyzed for SCC. The computer modeling process was started at the end of thedata collection periodsand is on-going. Computer model validation will take place in 2021. Changes/Problems:The initial scope of the project outlined that we would collect milk samples from 180 cows, three days/week for a total of 30 sample periods, but in the end we collected samples from 250 cows. As we analyze the data the number of 250 cows on project will be decreased as we identify cows that may not be eligible for analysis. There were 17,680 milk samples collected from 261 cows and analyzed for SCC. The number of cows changed because we had the option to work with two pens of cows at the research dairy instead of one group of cows. What opportunities for training and professional development has the project provided?We were able to build upon the relationship that has been created between Cornell University and DeLaval.DeLaval, although not part of the initial design of the project was very helpful in working with our research team on data collection from their DelPro software. It was very rewarding working with the research farm for the data collection process. Farmownership was very open to the research project and provided valuable input into data collection. Training the students to work with the DeLaval Voluntary Milking System (VMS) was an amazing experience. The students very quickly were able to navigate through the software to make the VMS operate the way that was described in the training. In the end the students were able to identify methods that made the process of collecting samples from the VMS easier than what was outlined in the training. The students end up training me (Rick Watters) on a few things with the VMS!! How have the results been disseminated to communities of interest?The computer model is in the development phase so at this point no computer modeling information was available to the communities of interest. What do you plan to do during the next reporting period to accomplish the goals?We plan to implement the computer model with the VMS system and validate the model with milk sample collection and SCC analysis of the milk samples. In the summer of 2021 students will again collect data for the subclinical modeling validation process.

Impacts
What was accomplished under these goals? 1. Identify quarter level milking data points from automated milking systems (AMS) that are associated with the onset of clinical mastitis (CM) and subclinical mastitis (SCM). 2. Develop a predictive model that will allow AMS managers to identify at risk cows with subclinical and clinical mastitis. a. The development of a model, with quarter level milking time data points associated with CM and SCM will allow AMS managers to identify cows at risk for CM and SCM accurately. 3. Determine if the additional quarters of an individual cow that did not flag with the model for CM or SCM are considered healthy by having a SCC < 200,000 cells/ml by documenting the SCC of all quarters when a cow has CM or SCM. 4. Report results to producers that have AMS and companies that make the systems. From June 15, 2020 - August 28, 2020 our research team collected individual quarter milk samples for somatic cell count analysis (SCC) and also obtained milking time data for each cow enrolled on the project from the DeLaval DelPro software. We are currently in the process of organizing the data in a manner that will be suitable for statistical analysis and modeling. It is too early in the data analysis process to provide any results. The process of acquiring the data recorded by the Voluntary Milking System was more challenging than expected. While a lot of information and datapoints are readily available through the proprietary user interface, some of the metrics are pre-filtered through DeLaval algorithms and other desired data were not available. That said, we were able to collect a lot of information about each milking event and believe we can still develop robust prediction models. The initial scope of the project outlined that we would collect milk samples from 180 cows, three days/week for a total of 30 sample periods, but in the end we collected samples from 250 cows. As we analyze the data the number of 250 cows on project will be decreased as we identify cows that may not be eligible for analysis. There were 17,680 milk samples analyzed for SCC.

Publications