You are on page 1of 16

SOME STATISTICAL TECHNIQUES FOR BIO-EFFICACY TRIALS

Rajender Parsad I.A.S.R.I., Library Avenue, New Delhi - 110 012 Biological assays and probit analysis are quite useful in designing of bio-efficacy trials and analysis of experimental data. A biological assay (bioassay) is a set of techniques relevant to the comparisons between the biological strength of alternative but similar biological stimuli (for example, a vitamin, a drug, a mental test, a physical force, an insecticide, plant extract, etc.) based on the responses produced by them on the subjects, such as subhuman primates' (or human) living tissues, plants or isolated organs, insects, etc. Normally two preparations of the stimulus, one of known strength (standard preparation) and another of unknown strength (test preparation), both with quantitative doses are applied to a set of living organisms. The general objective of bioassays is to draw statistically valid conclusions on the relative potency of the test preparation with respect to the standard one. Usually when a drug or a stimulus is applied to a subject it induces a change in some measurable characteristic that is designated as the response variable. In this setup, the dose may have several chemically or therapeutically different ingredients, while the response may also be multivariable. Thus the stimulus-response or dose-response relationships for the two preparations, both subject to inherent stochastic variability, are to be compared in a sound statistical manner so as to cast light on their relative performance with respect to the set objectives. Let ds and dt denote the doses of the standard and the test preparations respectively such that each of them produces a pre-assigned response in some living organism. Then the ratio = d s dt is called the relative potency of the test preparation. If is greater than unity, it shows that a smaller dose of the test preparation produces as much response as a relatively larger dose of the standard preparation and hence the potency of the test preparation is greater than that of the standard preparation. Similarly when is less than unity the potency of the test preparation is smaller than that of the standard preparation. Naturally, such statistical procedures may depend on the nature of the stimulus and response, as well as on other extraneous experimental (biological or therapeutic) considerations. As may be the case with some competing chemicals for removing the effect of common insects, the two (i.e. test and standard) preparations may not have the same chemical or pharmacological constitution, and hence, statistical modeling may need a somewhat different approach than in common laboratory experimentation. Nevertheless, in many situations the test preparation may behave (in terms of the response/tolerance distribution) as if it is a dilution or concentration of the standard one. For this reason, often such bioassays are designated to compare the relative performance of two drugs under the dilution-concentration postulation (i.e., assays with two preparations containing the same effective ingredients which is responsible for the response), and are thereby termed dilution assays or analytical dilution assays. Dilution assays are classified into two broad categories: direct dilution and indirect dilution. In a direct assay, for each preparation the exact amount of dose needed to produce a specified

Some Statistical Techniques for Bio-Efficacy Trials

response is directly measured, so that the response is certain while the dose is a nonnegative random variable that defines the tolerance distribution. Statistical modeling of these tolerance distributions enables us to define the relative potency in a statistically interpretable and analyzable manner, often in terms of the parameters associated with the tolerance distributions. However, a direct assay is practicable only when both the preparations are capable of administration in such a way that the minimal amounts needed to produce the specified response can directly be measured. In most assays, however, the response is not directly measurable and indirect methods are used to estimate the dose corresponding to a given response via a dose response relationship, such assays are known as indirect assays. In an indirect assay the dose is generally administered at some prefixed (usually non-stochastic) levels, and at each level the response is observed for subjects included in the study. Thus, the dose is generally nonstochastic and the stochastic response at each level provides information about the tolerance distribution for the particular preparation. If the response is a quantitative variable (magnitude of some property like survival time, weight, etc.), then we have an indirect quantitative assay, while if the response is quantal in nature (i.e. all or nothing), then we have a quantal assay. Both of these assays are commonly adopted in statistical practice. Within this framework, the nature of the dose-response regression may call for suitable transformations on the dose variable (called the dosage or dosemetameter) and/or the response variable, called the response-metameter. The basic objective of such transformations is to achieve a linear dose-response regression that may induce simplifications in statistical modeling and analysis schemes. If z represents the dose in the original scale, then the two transformations that have been found useful in bioassay work are (i) x = log e ( z ) and (ii) x = z , where > 0 is a known constant. The first of these gives rise to parallel line assays and assays based on the second transformation are called as slope ratio assays. These assays normally fall under the category of quantitative indirect assays. In these assays, the transformation of response variable is generally not needed. In quantal assays, however, the response variable is generally subjected to the probit (or normit) and logit transformations, based on normal and logistic distributions, respectively. In a parallel line assay, the two dose-response regression lines (one for the standard preparation and another for the test preparation) are taken as parallel and further the errors in the two regression equations are assumed to have the same distribution (often taken as normal). Therefore, after fitting of these two regression lines, it is important to test for the parallelism for the regression lines before making any conclusions. A parallel line assay is called as symmetric if the standard and test preparations involve the same number of doses, otherwise it is called an asymmetric assay. The analysis of the parallel line assay for conducting the validity tests and for estimating the relative potency becomes very much simplified when the doses of each of the preparations are taken in geometric progression. In slope ratio assays it is assumed that the two regression lines intersect at the same point on the response axis, i.e., they are assumed to have the same intercept. As the dose takes value zero on the response axis, it is necessary to include a blank dose in the assay for the validity test. Thus if there are k doses for each of the two preparations in a slope ratio

VI-62

Some Statistical Techniques for Bio-Efficacy Trials

assay, the total number of doses is 2k+1. If against each of these doses an equal number of subjects, say, n is allotted, the assay is 2k+1 symmetrical slope ratio assay, otherwise asymmetrical slope ratio assays. In slope ratio assays, the blank, the intersection contrasts and the two regression contrasts are of major importance. In quantal assays, occurrence or non-occurrence will depend upon the intensity of the stimulus. For any one subject, under controlled conditions there will be a certain level of intensity below which the response does not occur and above which the response occurs. Such a value has often been called a threshold or limen, but the term tolerance is now widely accepted. This tolerance value will vary from one member to another of the population used, frequently between quite wide limits. When the characteristic response is quantitative, the stimulus intensity needed to produce a response of any given magnitude will show similar variation between individuals. In either case, the value for an individual also is likely to vary from one occasion to another as a result of uncontrolled internal or external condition. In these assays, the earlier attempts were made to characterize the effectiveness of a stimulus in relation to a quantal response referred to the minimal effective dose or for a more restricted class of stimuli, the minimal lethal dose terms which failed to take account of the variation in tolerance within a population. The logical weakness of such concepts is the assumption that there is a dose for any given chemical which is only just sufficient to kill all or most of the insects of a given species, and, doses a bit lesser would not kill any insect of that species. Any worker, however, accustomed to the estimation of toxicity knows that these assumptions do not represent the truth. It might be thought that the minimal lethal dose of a poison could instead be defined as the dose just sufficient to kill a member of the species with the least possible tolerance, and also a maximal non-lethal dose as the dose which will just fail to kill the most resistant member. Undoubtedly some doses are so low that no test subject will succumb to them and others so high as to prove fatal at all, but considerable difficulties attend determination of the end-points of these ranges. Even when the tolerance of an individual can be measured directly, to say from measurements on a sample of ten or a hundred that the lowest tolerance found indicated the minimal lethal dose would be unwise: a larger sample might contain a more extreme member. When only quantal responses for selected doses can be recorded the difficulty is increased, and the occurrence of exceptional individuals in the batches at different dose levels may seriously bias the final estimates. The problem is in fact that of determining the dose at which the dose response curve for the whole population needs the 0% or 100% levels of kill and even a very large experiment could scarcely estimate these points with any accuracy. An escape from the dilemma can be made by giving attention to a different and more satisfactorily defined characteristics, the median lethal dose, or, as a more general term to include response other then death, the median effective dose. This is the dose that will produce a response in half the population. The median effective dose is commonly referred to as the ED 50, the more restricted concept of median lethal dose as the LD 50. Analogous symbols were used for doses effective for other proportions of the population,

VI-63

Some Statistical Techniques for Bio-Efficacy Trials

ED90 being the dose that causes 90% to respond. With a fixed total number of subject effective doses in the neighborhood of ED50 can usually be estimated more precisely then those for more extreme percentage levels and this is, therefore, particularly favoured in expressing the effectiveness of the stimulus. The ED50 alternatively be regarded as the median of the tolerance distribution that is to say the level of tolerance such that exactly half the subjects lie on either side of it. The ED 50 or LD 50 can easily be calculated using the Probit Analysis. For doing this, we conduct an experiment on different doses of an insecticide applied under standardized conditions to samples of an insect species and record the number of insects killed and the number of insects exposed. Now the ratio of the number insects died (the subjects responded) to that of the number of insects exposed (subjects exposed) gives the probability or proportion (P) of the insects killed at a particular dose. Now this probability data is subjected to probit or logit transformation.

Probit transformation is nothing but the 5 more than the normal equivalent deviate. In this transformation, we replace each of the observed proportions with the value of standard normal curve below which the observed proportion of the area is found. To avoid negative numbers, the constant 5 is usually added. For example, if half (0.5) of the subjects respond at a particular dose, the corresponding probit value is 0, since half of the area in a standard normal falls below a Z score of 0. When the constant 5 is added, the transformed value for the proportion is 5. If the observed proportion is 0.95, the corresponding probit value is 1.64. Addition of the constant value of 5 makes this 6.64. Likewise, if 10% of the subjects respond, then the normal equivalent deviate is -1.29 and hence the probit value is 3.7.
In the logit transformation, the observed proportion P is changed to
ln (P (1 P )) +5 2

The quantity ln (P (1 P )) is called a logit. Division by 2 and addition of the constant 5 is done to keep the values positive and to keep the two types of transformations on a similar scale. If the observed proportion is 0.5, the logit-transformed value is 0+5, the same as the probit-transformed value. Similarly, if the observed proportion is 0.95, the logittransformed value is 6.47 (1.47+5). This differs somewhat from the corresponding probit value of 6.64. (In most situations, analyses based on logits and probits give very similar results.) The above is discussion about the transformation of the observed proportions. The dose are transformed to the logarithmic scale. When the experimental data on the relation between dose and proportion of the subjects responded have been obtained, either a graphical or a statistical approach in terms of fitting of response metameter-dose metameter linear regression relationship can be used to estimate the parameters.

VI-64

Some Statistical Techniques for Bio-Efficacy Trials

The graphical approach is rapid and sufficiently good for many purposes. In this approach, the probits are plotted against the dose-metameter, and a straight line is drawn by eye to fit the points as satisfactorily as possible. The line must be so drawn that the differences between the observed probit values and the probit values obtained by the line at each value of dose metameter are as small as possible. The value of log ED 50 is estimated from this line as x at which probit value is 5. In the second approach, we fit a simple linear regression equation Probit ( Pi ) = a + bx i + ei , where Pi is the observed proportion corresponding at close x i (usually the log of the dose is used instead of the actual dose), a and b are respectively the intercept and slope of the regression equation and ei is the random error. From the fitted regression equation, the log ED 50 (x) is obtained as that value for which probit value is 5. The goodness of fit of the model, the Pearsons chi-square goodness of fit test. For this we obtain the expected frequencies are obtained. The expected frequencies are the number of subjects responded and is obtained by obtaining the estimated Probit ( Pi ) from the fitted model. Subtract 5 from he model. Now obtain the area under the normal curve below this point. This gives the percentage of the subjects responded. Now using the number of subjects tried, one can obtain the expected number of subjects responded ( ni Pi ) for the dose i. If ri is the corresponding observed number of the subjects affected, then we obtain, residuals as r n P . Now the Pearson goodness of fit chi-square is
i i i

obtained as

ni Pi (1 Pi ) minus he number of estimated parameters. To be clearer, consider the following example from Finney (1971).

(ri ni Pi ) 2 . The degrees of freedom are equal to the number of doses

Example 1: Finney (1971) gave a data representing the effect of a series of doses of carotene (an insecticide) when sprayed on Macrosiphoniella sanborni (some obscure insects). The Table below contains the concentration, the number of insects tested at each dose, the proportion dying and the probit transformation (probit+5) of each of the observed proportions. Concentration (mg/1) No. of insects (n) No. of affected (r) %kill (P) Log concentration (x) 1.01 0.89 0.71 0.58 0.41 Empirical probit

10.2 7.7 5.1 3.8 2.6 0

50 49 46 48 50 49

44 42 24 16 6 0

88 86 52 33 12 0

6.18 6.08 5.05 4.56 3.82 -

VI-65

Some Statistical Techniques for Bio-Efficacy Trials

The LD50 or ED50 can be obtained through the graphical approach, i.e. by plotting probit against the log doses. For results on this one may refer to Finney (1971). Here, we present the results of fitting of a simple regression line of probits on log concentration using method of ordinary least squares. The fitted regression line is obtained as Y = Probit ( Pi ) = 2.140 + 4.164 xi , The goodness of fit of this equation is then tested using Pearson chi-square. For this, we obtained expected number of insects killed, residuals and chi-square. Thiese are given in the following Table. Log concentration (x) 1.01 0.89 0.71 0.58 0.41 Y P

6.30 5.83 5.10 4.58 3.90

90.3 79.7 54.0 33.7 13.6

6.345 5.846 5.096 4.555 3.847

No. of insect s (n) 50 40 46 48 50

No. affected Obs.(r) 44 42 24 16 6 Exp. (nP) 45.50 39.24 24.76 16.23 6.24

Discrepancy (r-np)

(r nP) 2 nP(1 P) 0.5495 0.9720 0.0508 0.0051 0.0101

-1.50 2.76 -0.76 -0.23 -0.24

[2 ] = 1.587 3
The table value of 2 at 3 degrees of freedom and 5% level of significance is 7.81. Therefore, there is no reason to doubt the model. The log ED 50 computed from the above regression line is (5-intercept)/slope = 0.687. Therefore, the ED50 value is 4.682 mg/l. Similarly one can obtain other ED values. When a parameter such as the median lethal dose has been estimated from experimental data it is natural to wish to infer within what limits its true value may reasonably be expected to lie. A statement about the probability, by the use of which probabilities can be assigned only to statements about the occurrence of observations or of statistics calculated from the observations. In order to overcome this difficulty, the concept of fiducial probability is used. These limits can easily be calculated after obtaining the standard errors of the LD or ED values. Although the graphical and regression approach through least squares are often adequate for estimation of parameters from a single set of quantal response data, these methods do not generalize easily to more complicated experimental situations. To avoid these problems method of maximum likelihood is generally used for the estimation of the parameters. The likelihood is defined to be proportional to the joint probability of all the observations. In this method, the estimates of the parameters are those values that maximize the likelihood. The maximum likelihood estimators are consistent. The computational procedures followed in most the statistical packages is however the interactive numerical algorithms to obtain these results.

VI-66

Some Statistical Techniques for Bio-Efficacy Trials

The above discussion relates to only one stimulus. If there is more than one stimulus variables terms are added to the model for each of the stimuli. We get the regression coefficients and standard errors, intercept and standard error, Pearson goodness-of-fit chisquare, observed and expected frequencies, and confidence intervals for effective levels of independent variable(s). If the Pearson goodness-of-fit chi-square is non-significant, we calculate the value of the dose-metameter for which the value of the response metameter (probit) is 5. The antilogarthmic of this dose metameter is the ED 50 or LD 50. The ED 50 alone does not fully describe the effectiveness of the stimulus. Two insecticides/fungicides may require the same rate of application in order to be lethal to half of the population, but, if the distribution of tolerances has a lesser 'spread' for one than for the other, any increase or decrease from this rate will produce a greater change in mortality for the first than for the second. Therefore, it is necessary to give the standard errors and fiducial limits associated with ED 50 or LD 50. Besides knowing the ED 50 of a particular chemical preparation, the experimenter may be interested in comparing the relative potencies of the several chemical preparations. One is required to fit the probit regression lines for each of the chemical preparations separately. These regression lines are required to be tested for parallelism. If the probit regression lines are parallel for the different chemical preparations, then the relative potency is constant at all levels of the response. The choice of an efficient experimental design is based on the nature of the variability in the experimental material, environmental conditions and objectives for conducting a bioassay. The design may be a randomized complete block design, an incomplete block design, design for factorial experiments etc.
Computation of corrected efficacy % The discussion earlier assumed that the responses of the test subjects is due to the applied stimuli alone. In some experiments, however, the responses can occur at zero dose; either control batches of the subjects have received zero dose or a sequence of low doses indicating a minimal response rate greater than zero. In a pesticide trials some insects may die from natural causes. In such situations, it is required to work with crrected mortality or corrected proportions of the responses. Corrected efficacy % in pesticide trials can be computed by using the Abbott, Henderson and Tilton, Schneider-Orelli or Sun-Shepard formulas. The selection of appropriate formula is depending on two factors viz. 1. Trial condition (infestation or population stability and homogeneity). 2. The data on your hand (live individuals or mortality %).

The following table is of help in choosing the right formula Available data Uniform population Non-uniform population Infestation or live Abbott Henderson-Tilton individuals Mortality or dead Schneider-Orelli Sun-Shepard individuals

VI-67

Some Statistical Techniques for Bio-Efficacy Trials

These formulae are given in the sequel.


Abbott's formula
n in T after treatment * 100 Corrected % = 1 n in Co after treatment where : n = Insect population, T = treated, Co =control

Henderson Tilton's formula n in Co before treatment * n in T after treatment Corrected % = 1 n in Co after treatment * n in T before treatment where : n = Insect population, T= treated, Co =control

*100

The above two formulae can be combined into one and written as

x y 100 x where x = % survivorship in the control group (concentration of pesticide=0) y = % survivorship in the experimental group.
Corrected % =
Schneider-Orelli's formula Mortality% in treated - Mortality % in control *100 Corrected % = 100 - Mortality % in control Sun-Shepard's formula Mortality% in treated + change % in control *100 Corrected % = 100 + Change % in control

If the response is other than the death, then we may replace mortality by responded and survivorship by non-responded. Once the corrected responded % is obtained, then same procedure as above may be adoted. {This note is prepared from the book Probit Analysis by D. J. Finney (1971)}.
Steps for carrying out the Probit Analysis using MINITAB

For the data given in example 1, first enter the data in the Worksheet of MINITAB in three coumns C1: dose; C2: total Insects; C3: Insects killed or affected. Now create a column C4 for logdose by using LOGT(C1) using menu Calc. Now Choose Stat > Reliability/Survival > Probit Analysis. From the dialog box; Choose the data format "Success/trial" or "Response/frequency". In the present case, the data is in success trial format, therefore, enter C3, the column

VI-68

Some Statistical Techniques for Bio-Efficacy Trials

containing the number of successes in Number of Successes box and C2, the total number of trials in Number of Trials subbox. In the subbox for stress/stimulus enter C4, the column containing the logdose. Since, there is only one stimulus, therefore, the subbox pertaining to Factor (optional) may be left blank. Choose the distribution as normal. The other options available on the dialog box are: Estimate, Graphs, Options, Results and Storage. Using the option Estimate, One can - estimate percentiles for the percents you specify. These percentiles are added to the default table of percentiles. - estimate survival probabilities for the stress values you specify. One can also change the method of estimation for the confidence intervals and the level of confidence. The default option is two sided 95% fiducial intervals. Other options may also be used, as and when required. For this example, we chose the additional percentiles as 65 and survival probabilities for stress level 0.9 (logdose).
Probit Analysis: affect, total versus logdose

Distribution: Normal Response Information Variable Value Count affect Success 132 Failure 111 total Total 243 Estimation Method: Maximum Likelihood Regression Table Standard Variable Coef Error Z P Constant -2.88746 0.350134 -8.25 0.000 logdose 4.21320 0.478303 8.81 0.000 Log-Likelihood = -120.052 Goodness-of-Fit Tests Method Chi-Square DF Pearson 1.72888 3 Deviance 1.73897 3

P 0.631 0.628

Tolerance Distribution: Parameter Estimates Standard 95.0% Normal CI Parameter Estimate Error Lower Upper Mean 0.685338 0.0220962 0.642030 0.728646 StDev 0.237349 0.0269451 0.190001 0.296497

VI-69

Some Statistical Techniques for Bio-Efficacy Trials

Table of Percentiles Standard Error 0.0686394 0.0617254 0.0573944 0.0541723 0.0515787 0.0493935 0.0474969 0.0458160 0.0443030 0.0429251 0.0332991 0.0274617 0.0238086 0.0220962 0.0224241 0.0233958 0.0249330 0.0299366 0.0389715 0.0402991 0.0417626 0.0433947 0.0452427 0.0473792 0.0499232 0.0530936 0.0573685 0.0642153 95.0% Normal CI Lower Upper -0.0013503 0.267711 0.0769020 0.318861 0.126442 0.351423 0.163638 0.375989 0.193840 0.396025 0.219504 0.413123 0.241967 0.428152 0.262047 0.441643 0.280278 0.453943 0.297031 0.465294 0.420314 0.550845 0.507048 0.614696 0.578542 0.671870 0.642030 0.728646 0.701519 0.789420 0.730939 0.822648 0.760936 0.858672 0.826422 0.943771 0.913131 1.06590 0.924581 1.08255 0.936978 1.10068 0.950564 1.12067 0.965688 1.14304 0.982882 1.16860 1.00301 1.19871 1.02768 1.23580 1.06035 1.28523 1.11164 1.36336

Percent 1 2 3 4 5 6 7 8 9 10 20 30 40 50 60 65 70 80 90 91 92 93 94 95 96 97 98 99

Percentile 0.133180 0.197882 0.238933 0.269813 0.294933 0.316313 0.335060 0.351845 0.367110 0.381162 0.485580 0.560872 0.625206 0.685338 0.745470 0.776793 0.809804 0.885096 0.989513 1.00357 1.01883 1.03562 1.05436 1.07574 1.10086 1.13174 1.17279 1.23750

Table of Survival Probabilities 95.0% Normal CI Stress Probability Lower Upper 0.9 0.182888 0.122757 0.258650 Interpretation: The goodness-of-fit tests (p-values = 0.631, 0.628) suggest that the distribution and the model fits the data adequately. In this case, the fitting is done on normal equivalent deviate only without adding 5. Therefore, log LD50 or lof ED50 corresponds to the value of Probit=0. Log LD50 is obtained as 0.685338. Therefore, the stress level at which the 50% of the insects will be killed is (100.685338=4.845 mg/l). Similarly the stress level at which 65% of the insects will be killed is (100.776793 = 5.981 mg/l). At logdose = 0.9, what percentage of insects will be killed? Results indicate that 18.29% of the insects will be killed.

VI-70

Some Statistical Techniques for Bio-Efficacy Trials

Example 2: {Finney (1971): Ex.13} reported an experiment conducted to compare the relative potencies of the three analgesic amidone, phenadoxone and pethidine with morphine. The technique was to record how many standard electric shocks could be applied to the tail of a mouse before the mouse squeaked; if the number of shocks increased by 4 or more after application of the drug, then we say that the mouse has responded. The data generated alongwith the log dose concentration is given the following table. Factor Log dose Total Mouse Observed (f) (x) (n) 1 0.18 103 1 0.48 120 1 0.78 123 2 0.18 60 2 0.48 110 2 0.78 100 3 -0.12 90 3 0.18 80 3 0.48 90 4 0.70 60 4 0.88 85 4 1.00 60 4 1.18 90 4 1.30 60 1: morphine; 2:amidone; 3: phenadoxone; 4: pethidine. Mouse Responded (r) 19 53 83 14 54 81 31 54 80 13 27 32 55 44

For the analysis of data follow the same steps as in Example 1 with the addition that in the factor subbox define factor as f. The results obtained are given in the sequel.

VI-71

Some Statistical Techniques for Bio-Efficacy Trials

Probit Analysis: r, n versus x, f Distribution: Normal Response Information Variable r n Value Success Failure Total Count 640 591 1231

Factor Information Factor Levels Values f 4 1, 2, 3, 4 Estimation Method: Maximum Likelihood Regression Table Variable Constant x f 2 3 4 Coef -1.38969 2.47552 Standard Error 0.114661 0.173176 Z -12.12 14.29 2.20 10.47 -8.89 P 0.000 0.000 0.028 0.000 0.000 DF = 3 P-

0.237877 0.108353 1.35900 0.129801 -1.18220 0.132956

Test for equal slopes: Value = 0.673

Chi-Square = 1.54180

Log-Likelihood = -729.327 Multiple degree of freedom test Term Chi-Square DF P f 185.015 3 0.000 Goodness-of-Fit Tests Method Pearson Deviance Chi-Square 4.03105 4.03534 DF 9 9 P 0.909 0.909

VI-72

Some Statistical Techniques for Bio-Efficacy Trials

f = 1: Tolerance Distribution Parameter Estimates Parameter Mean StDev Estimate 0.561374 0.403956 Standard Error 0.0291474 0.0282589 95.0% Normal CI Lower Upper 0.504247 0.618502 0.352198 0.463319

Table of Percentiles Percent 1 2 3 4 5 6 7 8 9 10 20 30 40 50 60 70 80 90 91 92 93 94 95 96 97 98 99 Percentile -0.378367 -0.268249 -0.198383 -0.145825 -0.103073 -0.0666851 -0.0347796 -0.0062121 0.0197689 0.0436845 0.221397 0.349540 0.459033 0.561374 0.663715 0.773209 0.901352 1.07906 1.10298 1.12896 1.15753 1.18943 1.22582 1.26857 1.32113 1.39100 1.50112 Standard Error 0.0689837 0.0620774 0.0578001 0.0546499 0.0521384 0.0500422 0.0482399 0.0466576 0.0452472 0.0439754 0.0355803 0.0312437 0.0292557 0.0291474 0.0307532 0.0340907 0.0395535 0.0488712 0.0502282 0.0517234 0.0533905 0.0552784 0.0574616 0.0600631 0.0633085 0.0676909 0.0747255 95.0% Fiducial CI Lower Upper -0.533080 -0.257945 -0.407031 -0.159538 -0.327263 -0.0968963 -0.267390 -0.0496399 -0.218789 -0.0111000 -0.177503 0.0217857 -0.141375 0.0506903 -0.109088 0.0766335 -0.0797814 0.100284 -0.0528570 0.122108 0.145115 0.286370 0.284438 0.408244 0.400035 0.515829 0.504555 0.619913 0.605621 0.727451 0.710494 0.845760 0.830116 0.987334 0.992544 1.18714 1.01420 1.21423 1.03768 1.24371 1.06346 1.27616 1.09219 1.31246 1.12491 1.35392 1.16327 1.40270 1.21034 1.46276 1.27277 1.54273 1.37093 1.66903

VI-73

Some Statistical Techniques for Bio-Efficacy Trials

f = 2: Tolerance Distribution Parameter Estimates Parameter Mean StDev Estimate 0.465283 0.403956 Standard Error 0.0329958 0.0282589 95.0% Normal CI Lower Upper 0.400612 0.529953 0.352198 0.463319

Table of Percentiles Percent 1 2 3 4 5 6 7 8 9 10 20 30 40 50 60 70 80 90 91 92 93 94 95 96 97 98 99 Percentile -0.474459 -0.364341 -0.294474 -0.241917 -0.199165 -0.162777 -0.130871 -0.102304 -0.0763229 -0.0524073 0.125305 0.253448 0.362942 0.465283 0.567624 0.677117 0.805260 0.982973 1.00689 1.03287 1.06144 1.09334 1.12973 1.17248 1.22504 1.29491 1.40502 Standard Error 0.0766704 0.0697856 0.0655120 0.0623566 0.0598338 0.0577218 0.0558997 0.0542942 0.0528573 0.0515559 0.0427040 0.0376004 0.0345099 0.0329958 0.0330004 0.0346823 0.0385371 0.0462875 0.0474826 0.0488128 0.0503107 0.0520233 0.0540229 0.0564284 0.0594584 0.0635920 0.0703037 95.0% Fiducial CI Lower Upper -0.646325 -0.340552 -0.520318 -0.242102 -0.440558 -0.179453 -0.380675 -0.132207 -0.332051 -0.0936894 -0.290734 -0.0608351 -0.254566 -0.0319695 -0.222234 -0.0060724 -0.192875 0.0175260 -0.165892 0.0392905 0.0329891 0.202644 0.173847 0.322983 0.291660 0.428352 0.399014 0.529602 0.503306 0.633914 0.611508 0.748895 0.734347 0.887252 0.899894 1.08394 0.921870 1.11071 0.945680 1.13986 0.971790 1.17198 1.00087 1.20793 1.03395 1.24902 1.07270 1.29741 1.12019 1.35705 1.18312 1.43653 1.28191 1.56220

VI-74

Some Statistical Techniques for Bio-Efficacy Trials

f = 3: Tolerance Distribution Parameter Estimates Parameter Mean StDev Estimate 0.0123990 0.403956 Standard Error 0.0356207 0.0282589 95.0% Normal CI Lower Upper -0.0574164 0.0822144 0.352198 0.463319

Table of Percentiles Percent 1 2 3 4 5 6 7 8 9 10 20 30 40 50 60 70 80 90 91 92 93 94 95 96 97 98 99 Percentile -0.927342 -0.817224 -0.747358 -0.694800 -0.652049 -0.615660 -0.583755 -0.555188 -0.529206 -0.505291 -0.327579 -0.199436 -0.0899420 0.0123990 0.114740 0.224234 0.352377 0.530089 0.554004 0.579986 0.608553 0.640458 0.676847 0.719598 0.772156 0.842022 0.952140 Standard Error 0.0819421 0.0750288 0.0707272 0.0675438 0.0649928 0.0628521 0.0610006 0.0593651 0.0578974 0.0565644 0.0473402 0.0417348 0.0379801 0.0356207 0.0346077 0.0351438 0.0377872 0.0442858 0.0453498 0.0465463 0.0479071 0.0494781 0.0513295 0.0535776 0.0564359 0.0603728 0.0668332 95.0% Fiducial CI Lower Upper -1.11109 -0.784282 -0.985032 -0.685889 -0.905217 -0.623295 -0.845278 -0.576104 -0.796599 -0.537642 -0.755225 -0.504845 -0.718999 -0.476037 -0.686607 -0.450199 -0.657187 -0.426662 -0.630141 -0.404960 -0.430521 -0.242346 -0.288661 -0.123009 -0.169516 -0.0189715 -0.0604621 0.0805777 0.0458819 0.182837 0.156392 0.295510 0.281665 0.431433 0.449713 0.625620 0.471950 0.652131 0.496026 0.681012 0.522408 0.712859 0.551772 0.748529 0.585143 0.789329 0.624204 0.837409 0.672038 0.896705 0.735353 0.975800 0.834640 1.10097

VI-75

Some Statistical Techniques for Bio-Efficacy Trials

f = 4: Tolerance Distribution Parameter Estimates Parameter Mean StDev Estimate 1.03893 0.403956 Standard Error 0.0281126 0.0282589 95.0% Normal CI Lower Upper 0.983832 1.09403 0.352198 0.463319

Table of Percentiles Standard Error 0.0704720 0.0634814 0.0591380 0.0559298 0.0533648 0.0512180 0.0493668 0.0477370 0.0462798 0.0449617 0.0361002 0.0312579 0.0287349 0.0281126 0.0292822 0.0322921 0.0375327 0.0467137 0.0480609 0.0495471 0.0512059 0.0530861 0.0552625 0.0578580 0.0610984 0.0654773 0.0725110 95.0% Fiducial CI Lower Upper -0.0590689 0.222048 0.0671458 0.320289 0.147044 0.382800 0.207031 0.429942 0.255738 0.468376 0.297124 0.501162 0.333349 0.529970 0.365730 0.555819 0.395129 0.579377 0.422145 0.601109 0.621045 0.764443 0.761380 0.885305 0.878053 0.991814 0.983610 1.09486 1.08555 1.20152 1.19109 1.31917 1.31115 1.46030 1.47385 1.65984 1.49552 1.68691 1.51903 1.71637 1.54482 1.74881 1.57357 1.78509 1.60629 1.82653 1.64467 1.87530 1.69175 1.93535 1.75419 2.01532 1.85234 2.14162

Percent 1 2 3 4 5 6 7 8 9 10 20 30 40 50 60 70 80 90 91 92 93 94 95 96 97 98 99

Percentile 0.0991908 0.209309 0.279175 0.331733 0.374484 0.410873 0.442778 0.471345 0.497327 0.521242 0.698954 0.827097 0.936591 1.03893 1.14127 1.25077 1.37891 1.55662 1.58054 1.60652 1.63509 1.66699 1.70338 1.74613 1.79869 1.86856 1.97867

Interpretation: The goodness-of-fit tests (p-values = 0.909) suggest that the distribution and model fits the data adequately. The test for equal slopes is not significant (p-value = 0.673). The log ED50 for the four analgesics are 0.561374, 0.465283, 0.0123990 and 1.03893. ED50 values are 3.642286, 2.919329, 1.028961, 10.9378. The relative potencies
can easily be worked out.

VI-76

You might also like