The system will be going down for regular maintenance at 6pm NZT today for approximately 15minutes. Please save your work and logout.
The development of an expert system for diagnosing reproductive problems in seasonal dairy herds : a thesis presented in partial fulfilment of the requirements for the degree of Master of Veterinary Science at Massey University
An expert system for investigating the reproductive performance of seasonally calving dairy herds (DairyFIX) was designed and developed. The system uses information retrieved from an on-farm information system (DairyMAN) to identify the primary causes of reduced performance and make recommendations for further action. The current performance of DairyMAN user herds and a sample of herds from the National Dairy Database were initially measured. Realistic targets for each performance indicator were then calculated using this information. The associations and interactions between the performance and diagnostic indicators used in DairyMAN were more clearly defined using multivariate statistical analyses and path models. Additional information was obtained from the literature for incorporation into the DairyFIX system. The reproductive performance studies covered one complete year (1993/94 season) and were limited to spring calving seasonal herds which are typical of the New Zealand dairy industry. Such herds are managed so that all cows calve on a synchronous annual cycle as close as possible to the "ideal" calving date in that location. A typical aim is to have calvings spread over 6 to 8 weeks with a narrow spread being strongly favoured. The herd planned start of calving (PSC) date defines the beginning of this optimal period of calving and the planned start of mating (PSM) the first day of mating that must be used to achieve the desired calving pattern. Expert system development is introduced in Chapter 1 with some historical information that supports the use of on-farm information systems. The concepts and design methods used for developing expert systems are then reviewed in brief. Examples are provided to illustrate the different levels of sophistication that can be used for developing expert systems. These demonstrate that relatively simple designs are often the most successful. Examples including Bovid and Dairy Expert are discussed in detail. The methods used to monitor seasonally calving dairy herds are then reviewed with an emphasis on the main performance and diagnostic indicators used in DairyMAN. These include the four and eight week calving rates, 21 day submission rates, non-return and conceptions rates and four and eight week in-calf rates. Empty rates can be considered the final definitive measure of performance by many dairy farmers, but the difficulties with interpreting this figure are presented. The limited mating period in seasonally calving dairy herds means that the assessment of heat detection is difficult and often inaccurate. For this reason several alternative methods for assessing heat detection efficiency and accuracy are discussed in detail. A detailed description is given in Chapter 2 of herd and individual cow performance for DairyMAN user herds and the National Dairy Database sample. These data show that the reproductive performance of New Zealand herds is often below that previously reported. Calving rates are on average above industry targets, but only with a significant level of calving induction. Removal of inductions for welfare and marketing reasons will have a significant effect on the performance of many herds. Submission rates are the earliest available measures of performance during a mating season. New Zealand herds do not, on average, achieve the necessary targets. All measures of heat detection efficiency, although imprecise, show this is not a major problem with about 6% of heats missed. This has a negative effect on submission rates. Detection efficiency is an important issue for some individual herds as the consequences of poor heat detection are dramatic. The performance levels suggest that nutritional anoestrus and the effects of a spread calving pattern are the major causes of low submission rates. Conception rates of less than 60 % are reported. These are below those often suggested as typical for New Zealand herds. Much of the previous data has been taken from small study groups that may not adequately represent all herds. The common use of non-return rates may have created expectations that cannot be achieved in average herds as these are an optimistic measure of performance. Health events such as lameness are only reported for reference. Only limited health data is recorded although DairyMAN provides the flexibility to records such data. The variability in the type and degree of recording of these data is identified as a significant problem that limits the use of the available records. Some of the health events and especially lameness may have a large effect on the reproductive performance of many herds. Path models are developed in chapter 3 as an essential prerequisite to the development of the DairyFIX expert system. The models statistically confirm most of the relationships that have been previously considered important when evaluating herd reproductive performance. A number of factors including herd size and breed are shown to be associated with differing calving rates. The four week calving rate is shown to have strong indirect effects on submission rates, conception rates and herd in-calf rates. As such, it is one of the most important variables in seasonal herds. The importance of submission rates and conception rates is confirmed. Daily per cow milk production is shown to be a useful indicator of submission rates as both of these variables are directly influenced by nutrition. The models identify some limitations with using non-return rates as measures of conception. These generally give optimistic results that do not accurately reflect true performance. Such problems are compounded if reasonably accurate measures of heat detection cannot be obtained in herds with a very restricted mating period. Although the interactions between the performance and diagnostic indicators are largely understood from previous work this is the first time they have been brought together in statistically verified path models. The use of an on-farm information system (DairyMAN) was shown to be associated with improved herd performance including daily per cow milk yield and reproductive outcomes. DairyMAN user herds had cows of the same breed and genetic capacity. So DairyMAN users were able to produce more milk with animals of equivalent genetic merit, indicating that users achieved better management of the herd through improved attention to managerial details. This was associated with their adoption of DairyMAN, but not shown by this study to be a direct consequence of it. These findings are important because there is very little information confirming that on-farm information systems or central databases give true performance gains. This is despite the historical recognition of these systems and the rapid expansion in recent years. Justifying the use of more sophisticated tools such as an expert system would be more difficult if gains were not being achieved with the current technology. Calving induction was shown to be associated with some negative effects on milk yield and reproductive outcome. The New Zealand dairy industry does not currently favour the use of this management tool, but the impact any changes in management practices would have need to be evaluated with consideration of these effects. Calving induction is typically not used as recommended in New Zealand as many of the treatments are done too late to provide sufficient economic gains through increased lactation length and increasing the number of days from calving to the planned start of mating. Regional differences in performance were identified. DairyMAN user herds in the Manawatu had inferior conception rates while herds in the Taranaki had superior reproductive performance. These observed differences suggest a need to further identify causes of these differences, if performance is to be improved in some regions. The performance of individual cows and groups was examined in detail and statistical models developed for use in DairyFIX. Breed, lactation number, days calved at the start of mating and some health events were all shown to have an important impact on performance. The inferior performance of lactation 1 and 2 groups is having a large effect on performance. Cows in lactation 3 and 4 generally have the best performance with some reduction for aged cows. These aged cows do not dramatically affect overall herd performance because they are only a small proportion of most herds Jersey cows tended to show superior calving and submission rates. A number of complex models were developed for herds that pregnancy test and those that use non-return as the measure of conception. DairyFIX was developed to achieve two primary objectives. The first was to simplify the epidemiological approach to investigating herd reproductive problems. The system automates procedures that would otherwise be followed using DairyMAN and uses a graphical diagnostic interface to achieve this. DairyFIX quickly focuses on the important aspects of any performance deficit and reduces the time required to investigate a herd problem. Although DairyMAN is a comprehensive software package, users have difficulties due to its complexity and may invest a considerable amount of time exploring the program rather than focusing on the task required. The second objective of this project was to make expertise available for the user when examining herd performance. This required the use of the more sophisticated aspects of an expert system including the development of a "knowledge base" of information. DairyFIX consists of three sections. The first simply evaluates performance and determines if any problems exist. The second part considers the effect of the major components of performance such as the calving pattern and heat detection so that the primary causes of poor performance are identified. This section uses the statistical models previously developed to estimate expected performance. The third part of DairyFIX consists of several specific interrogation procedures for each area of poor performance. This section is not necessarily required for an operational system as the user can otherwise be referred to the appropriate reports in DairyMAN. Another module within DairyFIX was designed to assess the expected performance of a herd in subsequent seasons given a predicted herd profile. The models used for this purpose are similar to those that retrospectively evaluate performance. DairyFIX simplifies the investigative task and identifies the major causes of poor reproductive performance. It is anticipated that this tool will allow more dairy farmers and veterinarians to make effective use of DairyMAN while reducing the investment in time that is currently required.