Prepartum Behaviour as a Predictor of Postpartum Disease Flashcards
What is Parity
refers to how many calves a cow has had
e.g. a first parity cow = has had one calf
What is a fresh cow
cow who has recently calved
What is the transition period
- aprox 3 weeks pre to 3 weeks post calving
- transition from late preg (dry period = no milk production) to early lactation
What physiological changes happen during the transition period
- depressed immune funtion
- opening of teat canals
- changes in hormone expression (e.g. to relax pelvic ligaments)
- body condition loss after calving
Ketosis
- negative energy balance = caloric requirements for fresh cow are so high that many dont consume enough feed to compensate for calories they burn
- subclinical ketosis affects = 11-49% fresh cows (high serum ketone conc but no clinical signs)
- hyperketonaemia = abnormally high ketone bodies in blood
Metritis
= inflammation of the uterus usually caused by bacterial infection
- enlarged uterus
- watery brown discharge
- reduced milk yield, dullness, toxaemia, high fever
- most common in first 10 days post calving
- 21-40% cows affected
- Risk factors = heifer cows that have experienced dystocia, retained placenta (failure to expel fetal membrane w/in 24 hours) or other calving problems (e.g. milk fever, ketosis)
Huzzy 2007 prepartum behaviour study - method
- collected feeding/ drinking behaviour measurements from 101 dairy cows
- from 2 weeks before calving up until 3 weeks post calving
- feed intake data included
- electronic monitoring system
- collected social behaviour data assessed from video recordings
- metritis diagnosed based upon body temperature and vaginal discharge until 21 days post partum
Huzzy 2007 prepartum behaviour study - findings week before calving
- for every 10 min decrease in daily feeding time = odds of severe metritis after calving increased by 1.7x
- for every 1kg decrease in dry matter intake = odds metritis after calving increased by 3x
- cows later diagnosed with severe metritis = engaged in fewer aggressive (agonistic) interactions
- first research showing social behaviour plays role in transition cow health
Goldhawk 2009 prepartum feeding behaviour study - findings week before calving
- indicator for subclinical ketosis
- for every 10 min decrease in daily feeding time = risk sub clinical ketosis increased by 1.9x
- for every 1kg decrease in daily DMI = risk of SCK increased by 2.2x
- animals later diagnosed with SCK = initiated fewer displacements at feed bunk
How automation can help
- Insentec food bins able to record individual feed intake and feeding behaviour
- an electronic monitoring system
- depend on RFID system
~ cows wear radio frequency identification ear tags
~ bin records which cow is eating and for how long and how much each cow consuming
~
Can electronic feed system be used to identify competitive interactions at the feed bin? (Huzzy 2014)
Yes
- competitive interactions can be quantified using data from electronic feeding system
- optimal interval for predicting replacements at feed bunk = 26 seconds (specificity = 82%, sensitivity = 86%)
- validate results to feeding video recordings
Sensitivity (and in relation to Huzzy study)
= proportion of positives correctly identified as being positive
(e. g. proportion of sick animals correctly identified as being sick
- 86% replacemens at feedbunk (as identified in video) were correctly identified as replacements by the electronic feed bin algorithm
Specificity (and in relation to Huzzy study)
= proportion of negatives correctly identified as being negative
(e. g. proportion of healthy animals correctly identified as being healthy)
- 82% of times cows switched places at feedbunks NON-aggressively (without displacing and replacing another cow) were correctly identified as non-aggressive
Using electronic drinking systems as well to identify competition in dairy cows
- Mcdonald, von Keyserlingk and Weary 2019
- optimal interval to identify replacement = 29 seconds (82% sensitivity, 83% specificity
- Foris et al 2019 validated a replacement detection algorithm using combined data from electronic water and feed bins
- recal and precision of algorithm very high (average >0.8) comparable to human trained observers
What did the previous studies have in common and are these factors realistic for normal dairy farmers
- conducted on single farms designated for research purposes = how can they be generalised to the commercial context? need to be used on several commercial farms
- considered full day data on feeding and agonistic behaviours = farmers wont have time to analyse days worth of data
- considered either ketosis or metritis, not both = do cows later diagnosed with both behave differently to those with one disease and healthy animals
Sahar, Beaver et al 2020 study - aim
- assess if measures of feeding and competitive behaviour prepartum were associated with hyperketonaemia and metritis on commercial farms
- 5 commercial farms
- specifically looked at first 90 mins after fresh feed delivery to evaluate feeding behaviour and agonistic interactions
- actor (bully) and reactor (bullied) interactions
Sahar, Beaver et al 2020 study - results actor/reactor interactions
- cows with both postpartum diseases had significantly less interactions than health and one condition cows
- very small difference between healthy and one condition cows for the no. interactions
- not much pattern in prepartum interactions and whether or not the cow remained health in reactor cows
- ## actor = healthy more likely to be involved in more agonistic interactions than one condition and 2 condition cows
Sahar, Beaver et al 2020 study - results time spent eating
- animals that did not eat during observation period (90 mins after fresh feed delivery) = 89% chance of becoming sick with at least one condition after calving
- cows that spent only 30 mins eating = >82% chance of being sick with at least one condition post calving
- drop of feeding time could indicate development of at least one condition
prediction vs association studies
Prediction:
- aims to predict future events by analysising patterns likely to forecast future results
- based upon past behaviour, what is likely to happen in the future
Association:
- understanding a phenomenon
- simple relationships between variables and outcomes
Developing a predictive model - data spliting
- most of the data are used to develop the model = training data
- remainder of data used to test the model = test data
- typical split 60:40 (training:testing) more training than testing
Why is predictive modelling so important for dairy science
- incidence of transition cow disease has remained relatively unchanged over time, despite efforts to improve monitoring and adjust management to reduce disease risk
- risk factors can be assessed using a variety of sources (e.g. health records, feed intake, BCS, milk production data) ~ bring together most important variables and their interactions
What is statistical interaction (using example)
- try to determine relationship between BCS and post partum disease status
- But what if the effect of BCS on postpartum disease status depends on cows parity
= interaction ( 2 variables but relationship between them depends upon the influence of third variable - cows with fisrst parity dont see much of a difference on probability of disease for low or high BCS
- But 2nd or above parity = cows with higher BCS far more likely to become sick
- cross over interaction = first parity w/low BCS more likely to get sick but 2nd parity less likely to be sick if low BCS
Sahar, Beaver et al 2020 study - develop predictive model which cows sick after calving study
- recorded behaviour of 318 cows using electronic feeding system
- after calving, cows monitored for signs of metritis hyperketonaemia and mastitis
- split data using stratified random method: 70:30
- model showed that animals at risk of becoming sick postpartum can be identified with reasonable success = se=73% sp=80%
- found a no. of interactions between number of visits to the feed bins, time spent feeding & kg feed consumed on probability of becoming sick post partum
Potential application of predictive models on farms
- cost per case of hyperketonemia = £88
- cost per case of metritis = £80
- according to model: the 60% cows with highest probability of becoming ill postpartum, will include 97% of cows that will actually become sick
- if model can be automated for farm use = used to predict which cows are at highest risk of becoming sick
- offers opportunity for farmers to prevent illness from occurring = allow intervention and save money on treating sick animals
- allow quick data input from farmers and get a result instantly
(may lead to healthy animals being treated as no model is perfect representation of reality) - need a cost analysis to see if worth cost for future studies
What is temporality
= we dont know what causes what
- does participation in a lower no. aggressive interactions contribute to disease (through resulting decrease in intake)
OR does onset of disease cause the reduction in agonistic interactions
Foris et al 2020 - study aim and purpose
- characterise social competition strategies in transition cows and determine how they vary with health status
- competition for feed is a social stressor and associated with increased risk of illness
- some argue that increasing feed synchronicity, reduces competitiveness and increases positive welfare esp for lower rank cows
- BUT for indoor systems, agonistic interactions are common as feed space in limited
- individual variation in feeding behaviour = some cows feed at peak times risking agonistic interactions to gain feed access, others chose to wait to avoid interactions and feed at non-peak time
- changes in feeding behaviour may
Foris et al 2020 - study results
- no change in strategies between healthy and metritic cows
- metritic cows changed strategies more betweeen PRE (d -6 to -1) and POST1 (d 1-3) and between POST1 and POST2 (d 4-6)#- indictaed that strategies change in association with parturition and metritis
- metritic changed strategies in first 3 day entering the social group after calving and in the 3 days around diagnosis (possibly due to rise in sickness/illness)
- high competiveness score = longer feeding times and more agonistic interactions (actor and reactor)
- feeding synchrony (mean no. of occupied feeding in bins when cow feeding) may have been affected by milking during post partum period as cows were away from pen
- social bonds can affect competitiveness as known to influence physical proximity and behavioural synchrony
- unable to find differnece in component scores between healthy and metritic cows = social comp strategy not risk factor/indicatior of metritis
- group composition changed dynamically, limit establishment of social heirachy = more agonistic interactions
