DATA MANAGEMENT FOR THE EVALUATION OF PERFORMANCE AND PRODUCTIVITY OF CATTLE IMMUNIZED AGAINST EAST COAST FEVER
G. Gettinby and A.D. James
INTRODUCTION
Over the last decade, laboratory and field experiments have shown that cattle can be successfully immunized against ECF using the infection and treatment method (Irvin, 1985). It is expected that extensive field immunization programs will soon be initiated. These programs should provide data on immunized and non-immunized cattle from herds in Central and East Africa where ECF is endemic. This data will form the basis for the economic assessment of the impact of immunization. Detailed immunization records can identify important factors influencing the efficacy of the procedure. Identification of these factors may highlight needs for further experimental research. Production records will enable productivity indices to be determined and field challenge records will allow quantitative epidemiological models to be formulated. In this way a valuable database can be established. The competent management of such a database is important if the objectives are to be achieved.
DATA COLLECTION
Data collection systems for animal disease parameters have been pioneered by several organizations in Africa. In particular, the International Livestock Centre for Africa (ILCA) has acquired a great deal of expertise in assessing animal productivity. Trail et al. (1979, 1985) have described systems used in the study of the relationship between trypanosomiasis and cattle productivity. Although an ECF immunization program deals with the transfer of a technical process into operational field situations there are many similarities with the studies on productivity under trypanosome challenge. Data collection consists of accumulating records of individual animals from different geographical sites. Each record will have many 'fields' associated with it such as breed, type of farm, stabilate used etc. Each field can be regarded either as a factor or a variable depending on the hypotheses of cause and effect under examination. Fields can be loosely divided into four categories.
(1) Site fields. These are fields which specify the site and animal to be included in the study, and are normally factors and not variables.
Factors are: country; site; owner; farm type; experiment number; code number; animal number; breed; sex; date of birth.
The animal number can be a key field for the identification of all other fields associated with an animal.
(2) Immunization fields. These fields contain all data associated with the infection and treatment immunization procedure.
Factors are: immune/control status; date of immunization; stabilate used; site injected; volume of stabilate; drug used; date drug given; doses given; date of challenge.Variables are: clinical symptoms after immunization; serology after immunization; clinical symptoms after challenge; serology after challenge.
(3) Production fields. These fields establish the productivity of immunized and control animals.
Factors are: initial weight; reproductive status: service.Variables are: mortality; liveweight change; milk production; calving; fertility.
(4) Natural challenge fields. These fields report periodically on the response of animals to changing environmental conditions.
Factors are: diseases present; treatments administered; tick activity; tick controls; weather.Variables are: serology; packed cell volume (PCV); tick infestation.
Using records of the above fields, the performance and productivity of immunized and non-immunized cattle can be statistically evaluated based on specific variables such as:
cow mortality rate at age 2,3,.... years
cow liveweight at age 2,3,.... years
milk yield at 1st, 2nd,.... lactation
lactation completion rate
incidence of abortion
incidence of fertility
calf liveweight gain at age 1 year
calf liveweight at age 1 year
calf mortality rate at 1 year
clinical response to immunization
Indirect Fluorescent Antibody (IFA) titre after immunization and after challenge
PCV after immunization
clinical response to field challenge
PCV under field challenge
PCV may not be directly relevant to ECF but may indicate the presence of other diseases such as helminthiasis or trypanosomiasis
ANALYSIS OF DATA
Statistical methods of analysis.
The choice of statistical method will depend on the inferences made from the data collected, the volume of data generated, and the frequency with which interim analyses are required. An important distinction occurs depending on how the data are collected. This distinction can be relevant to the analysis. An experiment is undertaken in cases where animals are randomized to specific groups such as immunized and non-immunized, or drug treated and control, in order to test hypotheses. The randomization is usually based on factors such as sex, age, weight etc. In contrast, the rigorous demands of an experimental design are often not possible in the field and data are collected by surveillance. In the past, most immunization studies have used an experimental procedure designed and implemented according to a strict protocol, whereas productivity studies have been based on surveys.
In designed experiments, confirmatory statistical techniques are used for tests of significance for site, group differences etc. This usually presents no difficulties as there are few missing data and the design ensures that, animal numbers, for example, are 'balanced' within groups. One of the most widely-adopted experimental designs consists of randomizing animals to two or more groups of interest and collecting sequential observations from each animal at the same time. This is known as a two-factor experimental design with repeated measures. One factor is associated with the treatment groups and the other is associated with time. Since the animals are the experimental subjects, account must be taken of the repeated time measurements on individual animals. Using analysis of variance this particular factorial design can be used to test for: (1) significant group X time interactions, which indicate that groups do not respond similarly over time, (2) significant differences between treatment groups, and (3) significant differences between times. Only the first two of these tests are generally of interest and normally the presence of a significant group X time interaction precludes analysis for significant group or time differences. Winer (1971) gives details and examples of this useful experimental design.
Similarly, confirmatory statistical testing methods can be applied to survey data but this often requires 'least-squares' algorithms to calculate the statistics. This more complex procedure is needed to take account of missing data or treatment combinations which have unequal numbers of animals etc. Harvey (1960) described an algorithm which has proved satisfactory for such analysis (Trail et al., 1979). A computer software program known as GLIM (supplied by Numerical Algorithms Group, Oxford, UK) offers an alternative procedure. However, the implementation and interpretation of the results from least-squares algorithms are not always straightforward.
Recently, Exploratory Data Analysis (EDA) techniques have emerged as a complementary approach to confirmatory statistical techniques. They emphasize the value of exploring data to set up hypotheses using summary statistics often expressed in graphical form. For example, histograms are replaced by 'stem-and-leaf' plots which are easier to create. The convention of plotting means and standard deviations on graphs is replaced by more informative 'box-plots with whiskers'. These techniques are described by Tukey (1977) but the novice should consult Erickson and Nosanchuk (1979) for a readable and simple introduction. The EDA techniques are particularly suitable for interim analyses of on-going studies and generally they do not involve tedious calculations or rigorous tests of significance.
The analysis of both experiments and surveys usually takes place at the end of the study. However valuable information can often be overlooked if interim analyses are not undertaken. An ECF immunization program should have monthly interim analyses on data from individual sites using summary statistics such as those described in EDA techniques. (Interim analyses also have the advantage of sustaining momentum and interest in the project). Thereafter, at 6-month or 1-year intervals, confirmatory statistical methods could be exploited to identify differences between groups and to generate production indices.
An important point to remember is that statistical significance does not always equate with biological significance. Before analysing for differences between groups a useful magnitude of difference should be prescribed.
Data management systems.
Microcomputers offer sufficient storage facilities to cope with the storage, retrieval and analysis of large volumes of data. Management of the data within the computer can also be handled using software programs. Some software programs which are potentially suitable for recording data from an ECF immunization program are:
PANACEA - a database system designed for the management and statistical analysis of livestock data (James, 1985).dBASE 111 - a data management program which allows records to be manipulated using a procedure language. Recent developments at ILCA suggest that an extension to this package, which is specifically tailored for recording and analysing animal productivity data, will soon be available.
SYMPHONY LOTUS - a widely used package which offers record management, spreadsheeting, wordprocessing and graphics.
SMART - a more recent package, similar to SYMPHONY, which allows flexible transfer of information between spreadsheet, database and wordprocessor, and requires little program storage allowing more data records to be stored.
An efficient way to study data arising from an ECF immunization program would be to develop a central ECF software program based on one or more of the above programs. Thereafter the software could be made available to research centres interested in carrying out analysis of data on site.
PRODUCTION INDICES
A cattle production system converts feed and other inputs into a number of offtake products. The size of the herd may increase, but this can be considered as a type of product. The various inputs and outputs can be given economic values so that total input and output can be compared to give a measure of economic efficiency. Efficiency can also be considered in other terms. For example, ecologists might give energy values to the inputs and outputs to estimate energetic efficiency.
Production constraints affect the efficiency of conversion of inputs to outputs. If there is a constraint to production, the system is likely to produce less output per unit of input. This represents the cost of the constraint. In practice, it is difficult to estimate directly the effect of constraints in these terms. In particular, the valuation of inputs that are not bought and sold, such as common grazing, can present many problems.
A wide range of measures of cattle production efficiency are used. Production parameters, such as calving rate and age at maturity, are measures of technical efficiency in one aspect of the production system. They allow easy comparisons of the efficiency of different herds or production systems in that particular respect, without the complication of other factors. Most production parameters are easier to estimate than more general measures of efficiency of production, such as the value of total production per cow, or per animal.
In the evaluation of animal health constraints, there are difficulties in the use of many of the standard indices of efficiency. For example, expressing the effect of a health constraint in terms of production per cow is very confusing when the proportion of cows in the population is likely to change as a result of the removal of the constraint.
The static livestock model.
James (1984) used the total value of output per livestock unit equivalent (LUE) of feed supply as a measure of livestock production efficiency for the evaluation of animal health constraints. If this system is used, no problems arise from changing herd structures or population sizes associated with changes in animal health. A system of equations allowed the expected value of output per LUE from a set of well-known and easily estimated production parameters to be calculated. The system of equations was called the Static Livestock Model, because it described the situation where offtake was at such a level as to produce a constant herd size. However, the model also shows the potential rate of population growth if offtake is diverted for this purpose, so the term 'static' may be rather misleading.
The model can be used to evaluate the economic effect of disease control. The model is run twice, using estimates of the production parameters with and without the disease. The difference in the value of output represents the economic benefit of control. This can be compared to the cost of the control program. It is much easier to estimate the effect of a disease on production parameters such as mortality rate and lactation yield than it is to estimate the effect on output directly. If the calculations are made on a microcomputer, they take less than a second, and so sensitivity analysis can be used to test the effect of errors in the production parameter estimates. This also helps to concentrate data acquisition resources on the most important areas of uncertainty.
Some animal health constraints, such as tick infestation, cause producers to adopt an entirely different production system to the one they would use without the constraint. The model can be used to compare the efficiency with which the two systems would use the same feed resource. However, in such an analysis, it is necessary to take into account other inputs that might change, such as fencing, buildings etc.
The main problem in the use of the model is that the mean feed requirement of each type of animal must be given in livestock units. This can be difficult to estimate on a consistent basis. Moreover, the feed requirement is not independent of other production parameters. For example, if the calving rate increases, cows will have more calves and lactations, and this will increase their mean feed requirement. The original system of equations did not take such interactions into account, and adjustments had to be made manually.
The Cattle Production Efficiency Calculator (CPEC).
The Static Livestock Model can be modified to overcome some of the above difficulties and the modifications have introduced some elements that are specific to cattle. The CPEC system estimates feed requirements in megajoules of metabolizable energy (ME) per day according to the estimates used in formulating rations, but without 'safety margins' (Ministry of Agriculture, Fisheries and Food, 1975). The mean ME requirement of each class of animal is calculated on the basis of liveweight, rate of liveweight gain, pregnancy rate and milk production. Allowance is made for the energy concentration of the feed: bulkier feed requires more energy for digestion. The mean ME requirement of each class of animal is adjusted automatically for changes in production parameter values.
Data required by CPEC are:
(1) Mortality rates for each class of stock: the classes considered by the model are breeding cows, breeding bulls, replacement heifers (up to first calving), replacement bulls, surplus heifers and surplus males (steers or fattening bulls). Surplus heifers are handled separately because in some systems they are managed separately. If they are managed in the same way, then the same parameters would be used. A separate survival rate is used for calves up to weaning, as mortality rates in young calves may be higher than for older animals.(2) Culling rates for breeding cows and bulls. This is a management variable, and it will determine the availability of surplus young stock for offtake or increasing the herd size.
(3) Weights at birth and maturity determine the feed requirement for growing stock, taking into consideration the age at maturity. Rapid growth requires more energy per kilogram liveweight gain.
(4) Age at maturity is used to calculate rates of liveweight gain, and to determine the number of young stock needed as replacements.
(5) Cow/bull ratio is the only herd structure parameter required. The value will be high if artificial insemination is used.
(6) Calving rate is the normal measure of fertility in cows. Heifer fertility is reflected in the age at maturity.
(7) Percentage of replacement heifers remaining barren may be used where replacement heifers are managed differently from surplus heifers. The number of replacement heifers kept may be more than the number required to maintain the herd if some of them fail to calve.
(8) Milk offtake per lactation excludes that used by calves; it should be the quantity available for sale or human consumption.
(9) Megajoules per kilogram of milk produced varies with the composition of the milk. Standard tables are available (Ministry of Agriculture, Fisheries and Food, 1975).
(10) Energy concentration of feed (MJ/kg) is required to calculate the energy used in digestion. Standard tables of energy concentration are available (Ministry of Agriculture, Fisheries and Food, 1975). Weighted average values should be used for mixed feeds.
(11) Offtake values are used to convert physical production into economic terms. If energy values of offtake were used, the system would measure energetics, instead of economic, efficiency.
The efficiency of production is expressed as the value of output per unit of feed supply. The unit of feed supply is standardized as a feed supply yielding 100 MJ of ME per day. This is approximately the requirement of a high-yielding dairy cow. If required, the production per animal can be estimated, as the numbers of each class of stock per unit of feed supply are calculated. The potential herd growth rate can be estimated by dividing the surplus heifer offtake by the number of breeding cows per unit of feed supply.
The system is primarily intended for national planning activities, rather than as a model application to individual herds. Accordingly, no provision is made for stochastic analysis: all output is expressed as long-run expectations.
The microcomputer program offers a sensitivity analysis facility. For each parameter, the percentage change in value of output caused by a percentage change in the parameter value is calculated. This figure is calculated for a range of changes in parameter values to show any non-linearity of response. Chema et al. (1985) give an example of the application of CPEC.
QUANTITATIVE MODELS
The concept of modelling is not new to ECF studies. Jarrett et al. (1969) modelled the kinetics of replication of Theileria parva infections in cattle challenged with infected Rhipicephalus appendiculatus ticks. These infectivity studies were continued by Radley et al. (1974). Since then epidemiological models of tick activity have been formulated by Gardiner et al. (1981) and Gardiner and Gettinby (1983) for Ixodes ricinus based on biological development rates, and by Sutherst et al. (1979) and Utech et al. (1983) for Boophilus microplus based on larval survival. It should be possible to model the lifecycle of R appendiculatus using these techniques. The literature on the dynamics of T parva infections in cattle and tick hosts can be exploited to complement the tick-activity model and so provide a useful model of disease transmission. Since animals immunized by infection and treatment often have carrier status, modelling disease transmission remains important. However, epidemiological models require extensive data on environmental factors if they are to be validated.
References
Chema, S., James, A.D. and Moll, G. (1985). Veterinary involvement in integrated livestock projects. In Proceedings of Fourth International Symposium on Veterinary Epidemiology and Economics, Singapore, in press.
Erickson, B.H. and Nosanchuk, T.A. (1979). Understanding Data. Milton Keynes, UK, Open University Press: 1-388.
Gardiner, W.P. and Gettinby, G. (1983). A weather-based prediction model for the lifecycle of the sheep tick Ixodes ricinus L. Veterinary Parasitology. 13: 77-84.
Gardiner, W.P., Gettinby, G. and Gray, J.S. (1981). Models based on weather for the development phases of the sheep tick, Ixodes ricinus. Veterinary Parasitology. 9: 75-86.
Harvey, W.R. (1960). Least Squares Analysis of Data with Unequal Subclass Numbers. ARS-20-8. Washington, DC, United States Department of Agriculture.
Irvin, A.D. (Ed.) (1985). Immunization against Theileriosis in Africa. ILRAD, Nairobi: 1-167. l
James, A.D. (1984). Methods for the Evaluation of Animal Health Constraints. PhD Thesis, University of Reading.
James, A.D. (1985). Data management aspects of an East Coast Fever immunization network. In Immunization against Theileriosis in Africa (Irvin, A.D., Ed.), ILRAD, Nairobi: 133-139.
Jarrett, W.F.H., Crighton, G.W. and Pirie, H.M. (1969). Theileria parva: kinetics of replication. Experimental Parasitology. 24: 9-25.
Ministry of Agriculture, Fisheries and Food. (1975). Energy Allowances and Feeding Systems for Ruminants. Technical Bulletin No. 33. London, Her Majesty's Stationery Office, London: 1-79.
Radley, D.E., Brown, C.G.D., Burridge, M.J., Cunningham, M.P., Peirce, M.A. and Purnell, R.E. (1974). East Coast fever: quantitative studies of Theileria parva in cattle. Experimental Parasitology. 36: 278-287.
Sutherst, R.W., Utech, K.B.W., Kerr, J.D. and Wharton, R.H. (1979). Density-dependent mortality of the tick, Boophilus microplus on cattle - further observations. Journal of Applied Ecology. 16: 397-403.
Trail, C.J.M., Hoste, C.H., Wissocq, Y.J., Lhoste, Ph. and Mason, I.L. (1979). Trypanotolerant Livestock in West and Central Africa, International Livestock Centre for Africa, Addis Ababa: 1-147.
Trail, C.J.M., Sones, K., Jibbo, J.M.C., Durkin, J., Light, D.E. and Murray, Max (1985). Productivity of Boran cattle maintained by chemoprophylaxis under trypanosomiasis risk. ILCA Research Report. No. 9: 1-76.
Tukey, J.W. (1977). Exploratory Data Analysis. Addison-Wesley.
Utech, K.B.W., Sutherst, R.W., Dallwitz, M.J., Wharton, R.H., Maywald, G.F. and Sutherland, I.D. (1983). A model of the survival of larvae of the cattle tick, Boophilus microplus, on pasture. Australian Journal of Agricultural Research. 34: 63-72.
Winer, B.J. (1971). Statistical Principles in Experimental Design. New York, McGraw Hill: 1-907.
