G. Gettinby
International Laboratory for Research on Animal Diseases
P.O. Box 30709
Nairobi, Kenya
The development of infection and treatment as an immunization technique for the control of theileriosis continues to rely heavily on the interpretation of data from experimental and field studies. Experimental studies on immunization and related chemotherapy techniques have usually been designed to monitor small groups of animals intensively over short periods of time (for example, Dolan, 1985, 1986; Mutugi et al., 1988), the information collected being used to identify methods that are practical, effective and inexpensive and that produce few adverse reactions in experimental animals. In contrast, field studies have usually been undertaken to explore the difficulties in translating the techniques into the field and to measure production and economic effects of the disease or the control technique (see, for example, Dolan, 1985; Young, 1985; Morzaria et al., 1988). Should East Coast fever immunization become implemented more widely, follow-up surveillance studies will be necessary to measure the long-term consequences of the method.
WHAT CAN ANALYSIS OFFER?
Data from immunization studies reflect the relationships among animal, tick, parasite and environmental conditions. Each factor can contribute a different amount of variation to each measured parameter and so relationships which exist are not always obvious. Data must be analysed to understand these relationships. Data analysis does not always involve the rigorous application of statistical tests. A competent investigator will often be convinced of a claim from simply "eyeballing" data or plotting a simple graph. This is not to be discouraged, but the experimenter must be protected from prejudices. Statistical methods provide a scientific standard for planning studies, exploring data patterns and reporting scientific claims. In immunization and chemotherapy studies, statistical methods have generally involved testing for significant differences between treated and untreated (control) groups. However, an examination of unpublished and published immunization and chemotherapy studies shows that statistics in general, and the purpose of statistical tests in particular, are commonly misunderstood. As a result, certain inferences have been misleading and much time and effort have often been wasted.
SAMPLE SIZES TOO LARGE
Comparing groups and using unnecessarily large sample sizes can lead to "statistical significance", which is not the same as "biological significance". Given an indication of the variation normally associated with the parameter of interest, it is possible to design an experiment with the correct number of animals per group to be confident of detecting the order of difference considered to be biologically important.
SAMPLE SIZES TOO SMALL
Choosing sample sizes that are too small is a common error and frequently leads to a total waste of study resources. When it occurs, it can lead to claims of no significant difference between groups because the experiment was designed with inadequate data. Once again, this can be avoided by using simple calculations of the power of the statistical test based on biological expectation and parameter variation.
BIAS
Bias is another common error and can arise for many reasons. One treatment group may be advantaged simply because animals have not been randomized to ensure comparable groups at the start of the experiment. Alternatively, data from certain animals may be omitted in the belief that these animals produced "outliers". This happens when the experimental protocol has not rigorously specified under what conditions animals can be excluded. Bias is not difficult to detect.
CONFOUNDING
Detected differences among treatment groups may be confounded by other factors such as breed, genotype or age. Randomization helps to eliminate confounding, which frequently occurs when animals in different treatment groups are not managed similarly. Any observed differences among treatment groups cannot be attributed with certainty to treatment differences. Confounding is simple to avoid but may be difficult to detect.
TOO MANY STATISTICAL TESTS
When more than two treatment groups are compared, there is a temptation to test all possible pairs of groups. This can lead to misleading claims. For example, given 4 treatment groups, testing each group against each other involves 6 tests. If each test uses a 5% significance level, the chance of detecting a difference that does not really exist rises from 5% to 34%, that is, there is a 1 in 3 chance of claiming a non-existent difference. Excessive statistical testing arises in another context, when the same groups are tested repeatedly. These problems can be avoided using simple data handling methods, such as multiple range tests, and using the correct experimental design for repeated time measurements.
An immunization study should have clear objectives, one of which should be identified as the primary objective. The experimental procedure should be designed to achieve this objective. Past studies should be examined to give indications of expected variation in parameters. A protocol should be written with details of stabilate and chemotherapy administration, randomization of animals, legitimate reasons for excluding data from animals, the proposed method of data analysis and a justification for the number of animals to be used. Typically, a reasonable chance (80% power) of detecting (Using a 5% test of significance) a 50% difference in successful immunization rates between two groups of animals receiving different immunization regimes would require at least 16 animals per group. Using similar test criteria and data from past chemotherapy trials (Dolan, 1986), we have calculated the approximate number of animals one would need to use in a trial in order to detect given time differences. Table 1 gives the numbers of animals required when comparing treated and untreated groups, Table 2 the numbers required when comparing one treated group with another treated group.
Table 1. Approximate number of animals required in groups to detect (80% power, 5% significance) the given time differences between treated and untreated groups, based on variability observed in past chemotherapy trials
|
Parameter |
Estimated variance (days2) |
Order of difference expected (days) |
Number of animals required per group |
|
Time to appearance of schizonts |
3.34 |
1 |
53 |
|
Time to febrile response |
0.96 |
1 |
16 |
|
Time to recovery |
80.1* |
20 |
** |
|
Time to death |
22.3 |
5 |
11 |
* Animals recovering only.
** Untreated animals normally do not recover.
Table 2. Approximate numbers of animals required in groups to detect (80% power, 5% significance) the given time differences between different treated groups, based on variability observed in past chemotherapy trials
|
Parameter |
Estimated variance (days2) |
Order of difference expected (days) |
Number of animals required per group |
|
Time to appearance of macro-schizonts |
3.34 |
1 |
53 |
|
Time to febrile response |
0.96 |
1 |
16 |
|
Time to recovery |
80.1* |
1 |
1266 |
|
Time to death |
22.3 |
2 |
88 |
* Animals recovering only.
Clearly, detecting differences between times to appearance of macroschizonts, recovery and death among treated groups requires large numbers of animals. The collection of such data is more likely to come from field studies and surveillance than from experimental studies. Differences between times to death between untreated and treated groups is less demanding on experimental subjects because a larger order of difference would normally be expected. Unfortunately, it is not possible to provide such tables specifically for immunization studies because published work does not provide sufficient data.
CAN COMPUTERS HELP?
Data collection requires data storage. Most experimental immunization studies generate only small databases and interest is restricted to the technical aspects of immunization. Typically, treatment factors are determined by stabilate and drug administrations. The following are commonly observed parameters:
· time to macroschizont detection
· time to febrile response
· time to x% parasitosis
· time to recovery or death outcome
· time to recovery
· time to death
· immunized or not immunized
· packed cell volume
· white blood cell count
For most purposes, these data can be recorded by hand and analysed manually using simple graphs and calculations.
Field and surveillance studies to measure the impact of immunization usually involve large groups of animals and more extensive data recording. To evaluate the impact, comprehensive attempts are made to record data associated with all possible influencing factors. Tick counts may be required to discover if animals are under challenge. Vector populations and pasture availability are driven by weather and thus climatic data may be required to explain events. Clinical interventions and herd health details are required to evaluate possible adverse reactions. Liveweight gains and milk offtake should be monitored to ensure no loss in productivity. All measurements should normally be repeated weekly or monthly over long periods of time.
As data recording becomes labour-intensive and analysis more demanding, computers offer an ideal way to handle the data. Computers require software to store and manipulate data. A wide range of software packages exists for data handling. Database packages enable researchers to design record structures for data and systematically to query the records. Spreadsheet packages enable one to perform calculations on numerical data without writing ad hoc programs. This is important in large data studies because statistical tests can be used to "trawl" the data routinely for significant differences and interactions. Once these have been identified, the investigator can focus on findings of interest. Alternatively, the data can be exported from the database to other statistical software packages for analysis. Graphics packages can be used to express trends in the data and produce visual images of the relationships among parameters. The fast rate of processing data stored electronically means that analyses can routinely take place during the course of the study and final analyses can be completed within days of the end of the study.
Recent developments in computer technology have opened up two important avenues for the further exploitation of data stored in databases. Geographical information systems (Burrough, 1986) software is now available on microcomputers and maps can be produced from databases. Expert systems (Waterman, 1986) software packages are available for storing knowledge bases. This means that statements from experts can be stored in a list and a user seeking an expert opinion can use the computer to interrogate the knowledge base. These developments will have implications for the long-term monitoring of East Coast fever immunization programmes.
Data stored electronically in databases also offer a convenient medium for the exchange of information. In the last few years most software packages have evolved towards an international standard for the export and import of data among software packages. It is therefore no longer necessary for researchers to use one recording system. What is important is a general consensus on what needs to be recorded so that the bases for a continental database can be established.
REFERENCES
Burrough, P.A. (1986). Principles of Geographical Information System for Land Resources Assessment. Oxford: Clarendon Press.
Dolan, T.T. (1985). Immunization against theileriosis on O1 Pejeta Ranch, Laikipia, Kenya. In: Irvin, A.D., ed. Immunization against Theileriosis in Africa: Proceedings of a Workshop Held in Nairobi, Kenya, 1-5 October 1984. Nairobi: International Laboratory for Research on Animal Diseases, pp. 73-75.
Dolan, T.T. (1986). Chemotherapy of East Coast fever. Acta Tropica 43:165-173.
Morzaria, S.P., Irvin, A.D., Wathanga, J., D'Souza, D., Katende, J., Young, A.S., Scott, J. and Gettinby, G. (1988). The effect of East Coast fever immunization and different acaricidal treatments on the productivity of beef cattle. Veterinary Record 123:313-320.
Mutugi, J.J., Young, A.S., Maritim, A.C., Linyonyi, A., Mbogo, S.K. and Leitch, B.L. (1988). Immunization of cattle using varying doses of Theileria parva lawrencei sporozoites derived from the African buffalo (Syncerus caffer) and treatment with buparvaquone. Veterinary Parasitology 96:391-402.
Waterman, D.A. (1986). A Guide to Expert Systems. New York: Addison Wesley.
Young, A.S. (1985). Immunization of cattle against theileriosis in the Trans-Mara Division of Kenya: a comparison of trials under traditional Maasai management with trials on a ranch development. In: Irvin, A.D., ed Immunization against Theileriosis in Africa: Proceedings of a Workshop Held in Nairobi, Kenya, 1-5 October, 1984. Nairobi: International Laboratory for Research on Animal Diseases, pp. 64-68.
