AAPS PharmSciTech, Vol. 10, No.

3, September 2009 ( # 2009)

DOI: 10.1208/s12249-009-9271-9

Research Article

A Two One-Sided Parametric Tolerance Interval Test for Control of Delivered

Dose Uniformity—Part 3—Investigation of Robustness to Deviations
from Normality

Steven Novick,1 David Christopher,2 Monisha Dey,2 Svetlana Lyapustina,3,9 Michael Golden,4 Stefan Leiner,5
Bruce Wyka,6 Hans-Joachim Delzeit,5 Chris Novak,7 and Gregory Larner8

Received 17 January 2009; accepted 15 May 2009; published online 24 June 2009
Abstract. The robustness of the parametric tolerance interval test, which was proposed by the Food and
Drug Administration for control of delivered dose uniformity in orally inhaled and nasal drug products, is
investigated in this article using different scenarios for deviations from a univariate normal distribution.
The studied scenarios span a wide range of conditions, the purpose of which is to provide an
understanding of how the test performs depending on the nature and degree of the deviation from
normality. Operating characteristic curves were generated to compare the performance of the test for
different types of distributions (normal and non-normal) having the same proportion of doses in the tails
(on one or both sides) outside the target interval. The results show that, in most cases, non-normality
does not increase the probability of accepting a batch of unacceptable quality (i.e., the test is robust)
except in extreme situations, which do not necessarily represent commercially viable products. The
results also demonstrate that, in the case of bimodal distributions where the life-stage means differ from
each other by up to 24% label claim, the test’s criterion on life-stage means does not affect pass rates
because the tolerance interval portion of the test reacts to shifting means as well.
KEY WORDS: bimodal; distribution; heavy-tailed; inhaler; skewed.

1
Discovery Analytics, GlaxoSmithKline, Research Triangle Park, INTRODUCTION
North Carolina, USA.
2
SPRI Statistics, Schering-Plough Research Institute, Kenilworth, In the previous two articles of this three-part series, a
New Jersey, USA. parametric tolerance interval test (PTI-TOST) proposed by
3
Pharmaceutical Practice Group, Drinker Biddle & Reath, 1500 K Food and Drug Administration (FDA) at the 2005 Advisory
Street NW, Suite 1100, Washington, District of Colombia, Committee meeting was described and characterized. One of the
20005-1209, USA. main assumptions of the test is that the delivered dose uniformity
4
Regulatory Affairs and Quality, Pearl Therapeutics, Raleigh, North (DDU) data in a given product are normally distributed. The
Carolina, USA. DDU distribution for a typical product is studied during product
5
Boehringer Ingelheim Pharma GmbH & Co. KG, Ingelheim am development, and it may happen that the data are not strictly
Rhein, Germany.
6 normal. For example, a local inhomogeneity of the active
SpiraPharma Consulting, Lincoln Park, New Jersey, USA.
7 ingredient in a dry powder formulation or insufficient shaking
Lab Services Department, Drug Delivery Systems 3M, St.Paul,
Minnesota, USA. of an MDI suspension may lead to a normal distribution with
8
Scientific and Laboratory Services, Pfizer, Kalamazoo, Michigan, USA. occasionally observed extreme values. As another example, if
9
To whom correspondence should be addressed. (e-mail: svetlana. the beginning-of-unit (BOU) and end-of-unit (EOU) DDU
[email protected]) measurements in a multi-dose product are sufficiently different,
ABBREVIATIONS: μ, population (batch) mean; σ, population the distribution may appear bimodal. There could be several
(batch) standard deviation; X , sample mean; ACPS, Advisory physical reasons for the change from BOU to EOU drug content,
Committee for Pharmaceutical Science; BOU, beginning of unit; one being increasing drug concentration resulting from propel-
CMC, chemistry, manufacturing, and controls; DDU, delivered dose lant evaporation due to the increase of the head-space volume in
uniformity; DPI, dry powder inhaler; EOU, end of unit; IPAC-RS,
an MDI canister as the canister is being emptied. It is therefore of
International Pharmaceutical Aerosol Consortium on Regulation and
interest to study the properties of the test [e.g., using operating
Science; MDI, metered dose inhaler; OC, operating characteristic;
OINDP, orally inhaled and nasal drug products; PmaxTA, maximum characteristic (OC) curves] over a wide range of deviations from
allowable proportion of doses in a tail area (left or right) outside the normality to determine whether the test will perform appropri-
target interval; PTI, parametric tolerance interval; LC, label claim; N, ately under the expected conditions. This article investigates the
total sample size; N1, sample size in the first tier; N2, additional units effect of several types of deviations from normality on the PTI-
tested in the second tier; s, sample standard deviation. TOST performance.

829 1530-9932/09/0300-0829/0 # 2009 American Association of Pharmaceutical Scientists

830 Novick et al.

This study focuses on the “default” PTI-TOST parameters, Test For Control of Delivered Dose Uniformity—Part 1—
as described in the first article of this series. Ideally, such a test Characterization of FDA Proposed Test”) and match the K
should allow no more than 5% acceptance probability if either of values of the FDA proposal (1).
the tail areas in the DDU distribution is 6.25% or greater. This For this study, datasets were simulated with four different
article therefore focuses on the acceptance probability of two types of non-normal distributions: skewed unimodal, symmetric
types of non-normal distributions—bimodal and normal with unimodal with heavy tails, bimodal (bivariate normal), and
extreme values—when one of the tail areas approaches 6.25%. univariate normal with non-repeating outliers (further details
are included in the “Results” section). OC curves were
MATERIALS AND METHODS generated using Monte Carlo simulation techniques to compare
the probability of passing the PTI-TOST for each distribution to
Studied Test and Distributions the probability of passing the test under a univariate normal
distribution. Since the objective of the PTI-TOST is to ensure no
The PTI-TOST studied here was described in detail in more than the maximum allowable proportion of doses in each
the first paper of this series and is briefly summarized below. of the tail areas (PmaxTA) outside the target interval, each
The same assumptions and notations are used here as in part distribution in this study was created with the same proportion
1. For convenience, a general form of the “default” PTI- of DDU observations in the tail area. Additionally, the effect of
TOST is presented below. the non-parametric life-stage-means component of the PTI-
TOST on bimodal data was evaluated with OC curves.
Tier 1
Approach to Studying Test Robustness
Collect 20 doses from ten multi-dose orally inhaled and
nasal drug products (OINDP) units (BOU and EOU from To study robustness, datasets with the univariate normal
each unit). The 20 observations must pass the following distribution were first created, with specified tail proportions
criteria: above and below the FDA-recommended target interval [80,
120]. Three different cases were studied:
(1a) TL1 =X 1
K1 s1  80 with PmaxTA =6.25% and α1 =
0.0226, where K1 is listed in Table I. & “Tail=Left”: tail proportions ranging from 0.001 to
(1b) TU1 =X 1 þ K1 s1  120 with PmaxTA =6.25% and 0.0625 to the left of 80 and virtually zero proportion
α1 =0.0226, where K1 is listed in Table I. to the right of 120.
(1c) 85≤X BOU;1 ≤115 & “Tail=Right”: tail proportions ranging from 0.001 to
(1d) 85≤X EOU;1 ≤115 0.0625 to the right of 120 and virtually zero propor-
tion to the left of 80.
If the sample fails any of the criteria 1a–1d, the test & Tail=Both”: equal tail proportions ranging from
proceeds to the second tier. In tier 2, collect an additional 40 0.001 to 0.0625 to the left of 80 and to the right of
doses and repeat the steps above with N = 60 and K2 120. The tail proportions given for this scenario refer
substituted for K1. In Table I, K1 and K2 are given for several to the proportion in each tail.
sample size and PmaxTA options when α1 =0.0226 and α2 =
0.0340, which were recommended by the FDA based on the Because the univariate normal distribution covers any
Pocock method (see the Appendix in the accompanying interval with non-zero probability, it is not possible to create a
article by S. Novick, D. Christopher, M. Dey, S. Lyapustina, scenario in which the area under the normal distribution
M. Golden, S. Leiner, B. Wyka, H.-J. Delzeit, C. Novak, and curve outside the target interval is exactly zero. The value
G. Larner, “A Two One-Sided Parametric Tolerance Interval 10−10 was used to represent zero tail proportionality. The

Table I. K Coefficients for the PTI-TOST for Several PmaxTA and Sample Sizes, with α1 =0.0226 and α2 =0.0340

PmaxTA (100−2*PmaxTA)% N1 N2 N (total) K1 K2

8.75% 82.5% 10 20 30 2.816 1.933
7.5% 85.0% 10 20 30 2.957 2.037
6.25% 87.5% 10 20 30 3.119 2.155
5.0% 90.0% 10 20 30 3.310 2.294
3.75% 92.5% 10 20 30 3.545 2.464
8.75% 82.5% 20 40 60 2.203 1.734
7.5% 85.0% 20 40 60 2.317 1.830
6.25% 87.5% 20 40 60 2.448 1.940
5.0% 90.0% 20 40 60 2.601 2.068
3.75% 92.5% 20 40 60 2.791 2.226
8.75% 82.5% 30 60 90 2.000 1.656
7.5% 85.0% 30 60 90 2.106 1.749
6.25% 87.5% 30 60 90 2.227 1.855
5.0% 90.0% 30 60 90 2.369 1.979
3.75% 92.5% 30 60 90 2.544 2.132
PTI-TOST Robustness to Deviations from Normality 831

mean and standard deviation for a univariate normal first figure displays examples of the probability density
distribution can be exactly determined given left and right function of the simulated datasets. This type of figure shows
tail proportions. the distribution of data in comparison to the target interval,
To evaluate the performance of the test under deviations with one column per figure for each of the three cases
from normality, all distributions were generated with the described above: “Tail=Left,” “Tail=Right,” and “Tail=
same tail proportions outside the target interval. Acceptance Both.” These figures illustrate the changes in the distribution
probabilities for all the distributions were calculated for as the degree of the deviation from normality changes from
comparison to the univariate normal acceptance probabilities high to medium to low. They also compare these distributions
with the same tail proportions (or the same quality level, per to the univariate normal.
PTI-TOST). These acceptance probabilities were plotted in The next two figures in each scenario represent a
the form of OC curves. collection of OC curves for the sampling plans N1/N2 =20/40
and N1/N2 =30/60. Again, there are three columns per figure,
Characterization of Non-normal Distributions one for each of the three cases for tail proportions. Each
Used for Robustness Assessment figure has two rows. The first row shows OC curves for the
acceptance probabilities of the overall PTI-TOST, while the
Asymmetric distributions exhibit a property called skew- second row shows OC curves for just tier 1 of the PTI-TOST.
ness. Skewness is technically defined as the quotient of the Within each figure, an individual OC curve is a plot of
third central moment over the cube of the standard deviation. the acceptance probabilities of the test on the vertical axis
The univariate normal distribution (which is symmetric) has against the tail proportion of the distribution outside the
zero skew. Practically, skewness means that an asymmetric target interval on the horizontal axis. There is one OC curve
distribution is not symmetric with respect to the maxima of for the univariate normal distribution and one for each of the
the distribution or one side of the distribution has a larger non-normal distributions (with varying degrees of deviation).
area than the other. The OC curves may be compared to determine how robust
A heavy-tailed distribution exhibits a property called the test is to deviations from normality. If the OC curves
kurtosis, which is defined by the quotient of the fourth central overlap or are similar, one can conclude that the test performs
moment over the square of the variance. The univariate similarly regardless of the distribution of the data. If the OC
normal distribution has a kurtosis value of three. Excess curves for non-normal distributions are below or to the left of
kurtosis is defined as the value (kurtosis-3.0), which is a the normal OC curve, then the probability of passing the test
measure of the heaviness of the tails relative to those of a is lower for the non-normal distributions. The opposite is true
univariate normal distribution. Kurtosis is generally only if the non-normal OC curves are above or to the right of the
interesting for symmetric (not skewed) distributions. normal OC curve. The shape of a non-normal OC curve and
Skewness and excess kurtosis values are immutable to its difference from that of the normal OC curve thus illustrate
the shifting and scaling of a random variable. Thus, the values the test’s robustness to a particular deviation from normality
for skewness and excess kurtosis for a random variable X are (i.e., similar or lower acceptance of batches at the borderline
the same as the values for the random variable Δ+λX. of unacceptable quality).
Table II shows the skewness and excess kurtosis values
for the univariate normal and chi-square and T distributions Robustness to Skewness
with 5, 10, and 50 degrees of freedom. These distributions are
used in the assessment of robustness of the PTI-TOST to Figure 1 displays distributions for shifted and scaled chi-
skewness and heavy tails. squared random variables with 5 degrees of freedom (high
skew), 10 degrees of freedom (medium skew), and 50 degrees
RESULTS of freedom (low skew) and for the univariate normal
distribution (no skew). In the panel “Tail=Left,” the tail
In this section, a series of figures is presented for each of proportion to the left of 80 is 0.0625, and that to the right of
the scenarios for deviation from normality. In each scenario 120 is 10−10. In the panel “Tail=Right,” the tail proportion to
(except univariate normal with non-repeating outliers), the the left of 80 is 10−10, and that to the right of 120 is 0.0625. In
the panel “Tail=Both,” the tail proportions to the left of 80
and to the right of 120 are 0.0625 each (or 0.125 together).
Figures 2 and 3 illustrate that, when the equal proportion
Table II. The Skewness and Excess Kurtosis Values for the of DDU results outside the target interval resides in both tails
Univariate Normal and Chi-Square and T Distributions with 5, 10, or in the left portion (opposite the skewness), the pass rates
and 50 Degrees of Freedom are lower for the non-normal distributions. The higher pass
rates for the right tail case are most likely due to being
Distribution Skewness Excess kurtosis
slightly more on-target than the corresponding normal
Univariate normal 0 0 distribution, despite a larger standard deviation.
Chi-square with 5 degrees of freedom 1.26 2.4
Chi-square with 10 degrees of freedom 0.89 1.2
Robustness to Heavy Tails
Chi-square with 50 degrees of freedom 0.40 0.24
T with 5 degrees of freedom 0 6
T with 10 degrees of freedom 0 1 Figure 4 displays distributions for shifted and scaled T
T with 50 degrees of freedom 0 0.13 random variables with 5 degrees of freedom (high excess
kurtosis), 10 degrees of freedom (medium excess kurtosis),
832 Novick et al.

High Skew High Skew High Skew

Tail = Left Tail = Right Tail = Both

Medium Skew Medium Skew Medium Skew

Tail = Left Tail = Right Tail = Both
Density

Low Skew Low Skew Low Skew

Tail = Left Tail = Right Tail = Both

Normal Distribution Normal Distribution Normal Distribution

Tail = Left Tail = Right Tail = Both

60 80 100 120 140 60 80 100 120 140 60 80 100 120 140

DDU
Fig. 1. Distributions for chi-squared random variables with tail proportion=0.0625 in each specified tail
(skewed distributions). See text for other details

and 50 degrees of freedom (low excess kurtosis) and for the correlated within a unit. It has been observed for many
univariate normal distribution (no excess kurtosis). In the product types that μBOU ≠μEOU and sometimes σ1 <σ2.
panel “Tail=Left,” the tail proportion to the left of 80 is The PTI-TOST assumes that μBOU =μEOU =μ, σ1 =σ2 =σ, and
0.0625, and that to the right of 120 is 10−10. In the panel ρ=0. For this robustness study, the case where μBOU =μ−Δ
“Tail=Right,” the tail proportion to the left of 80 is 10−10 and and μEOU =μ+Δ will be considered with σ1 =σ2 =σ, and ρ=
that to the right of 120 is 0.0625. In the panel “Tail=Both,” 0. Thus, the life-stage means differ from each other by 2Δ,
the tail proportions to the left of 80 and to the right of 120 are the variance for both life stages are equal, and there is no
0.0625 each (or 0.125 together). correlation between beginning and end measurements
Figures 5 and 6 illustrate that, if the heaviness is in either within the same unit. As given in Everitt and Hand
tail, the pass rates are lower for the more extreme non- (2, page 30), so long as absðBOU
EOU Þ < 2
, the data
normal distributions (degrees of freedom=5 and 10). The low will appear to be unimodal (and normally distributed);
non-normal (degrees of freedom=50) and the normal distri- otherwise, the data will appear bimodal. Thus, when Δ>σ
butions have nearly identical OC curves. The pass rates are (i.e., the mean difference in life stages is greater than the
slightly higher when both tails are heavy for the more standard deviation within each life stage), the data appear
extreme non-normal distributions. bimodal. Figure 7 displays plots of the bivariate normal
distribution and the univariate normal distribution with the
Robustness to Bimodality tail proportions=0.0625. In Fig. 7, distributions for bivari-
ate normal distributions are displayed as random variable
A more general distribution for BOU and EOU pairs with means μBOU =μ−Δ and μEOU =μ+Δ for Δ=0, 3, 6, 9,
taken from the same sampled unit is the bivariate normal and 12. The univariate normal distribution is given by Δ=
with mean vector (μBOU, μEOU) and variance covariance 0. In the panel “Tail=Left,” the tail proportion to the left
! of 80 is 0.0625, and that to the right of 120 is 10−10. In the

21 
1
2
matrix V ¼ . This allows for different life- panel “Tail=Right,” the tail proportion to the left of 80 is

1
2
22 10−10 and that to the right of 120 is 0.0625. In the panel
stage means, for different life-stage standard deviations, “Tail=Both,” the tail proportions to the left of 80 and to
and for the (BOU, EOU) paired measurements to be the right of 120 are 0.0625 each (or 0.125 together).
PTI-TOST Robustness to Deviations from Normality 833

0.00 0.02 0.04 0.06

Overall Overall Overall
Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4
Probability of passing

0.2

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2

0.0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion
Normal Dist. Medium Skew
Low Skew High Skew
Fig. 2. OC curves for shifted and scaled chi-squared distribution for N1/N2 =20/40

As Figs. 8 and 9 demonstrate, the pass rates for the creating univariate normal data sets with a small probability
bivariate normal distributions are all lower than the for a single outlier value in the first tier based on observed
corresponding normal distribution, indicating that the PTI- patterns from actual products. In tier 1, data were generated
TOST is fairly robust for these types of distributions. as univariate normal given specified tail proportions. With
probability r, one outlier, a randomly chosen tier 1 observa-
Effect of Life-Stage Means Test when Data are Bivariate tion, was replaced with a uniform random variable with limits
Normal (130, 170). No outliers were created for tier 2 data.
As Figs. 12 and 13 illustrate, the pass rates for the
To investigate the effect of the life-stage means portion distributions containing outliers are similar to the
of the PTI-TOST on pass rates in the bimodal cases described corresponding normal distribution, indicating that the PTI-
above, the acceptance probabilities were re-computed for the TOST is fairly robust for these types of distributions.
bivariate normal scenario (see Figs. 7, 8, and 9) but without
the life-stage means test. The results are presented in Figs. 10
and 11. DISCUSSION
Figures 10 and 11 illustrate that the life-stage portion of
the PTI-TOST does not impact the pass rates. When the life- In this study, different scenarios for deviations from a
stage test fails, the tolerance interval portion of the test also univariate normal distribution have been examined to assess
fails in either tier. the robustness of the PTI-TOST. The scenarios considered
here span a wide range of conditions and, in some cases, may
Robustness to Non-repeating Outliers not be representative of what is encountered in practice. The
purpose of simulations in this study was to provide an
It has been observed for some products that the DDU understanding of how the PTI-TOST performs depending
observations appear to be univariate normal but with an on the nature and degree of deviation from normality. OC
occasional non-repeating large outlier value. To study this curves were generated to compare the performance of the
phenomenon, acceptance probabilities were estimated by test for different types of non-normal distributions versus a
834 Novick et al.

0.00 0.02 0.04 0.06

Overall Overall Overall

Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2
Probability of passing

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2

0.0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion
Normal Dist. Medium Skew
Low Skew High Skew
Fig. 3. OC curves for shifted and scaled chi-squared distribution for N1/N2 =30/60

normal distribution with the same tail area outside the target strated that, in the case of bimodal, skewed, heavy-tailed,
interval. or normal with non-repeating outlier distributions, non-
Most deviations from normality that may occur for MDIs normality does not increase the probability of accepting a
would generally increase the sample standard deviation. Since batch of unacceptable quality (i.e., the test is robust),
the PTI-TOST depends in part on the amount of delivered except in extreme situations that do not necessarily
dose variability, larger standard deviations will likely cause represent commercially viable products. The results also
lower pass rates. For most of the scenarios studied here, the demonstrate that, in the case of bimodal distributions
OC curves for the non-normal distributions were similar to where the life-stage means differ from each other by up
the corresponding normal distribution with the same tail area to 24% LC, the test’s criterion on life-stage means does not
characteristics. A few instances resulted in slightly elevated affect pass rates because the tolerance interval portion of
pass rates: moderately skewed, but on-target distributions and the test reacts to shifting means as well.
distributions with slightly heavy tails on both sides. From the
results presented in this article, it appears that the PTI-TOST
is generally robust to non-normal distributions likely to be ACKNOWLEDGEMENTS
encountered in practice.
For the bivariate normal distributions studied in this The authors thank the International Pharmaceutical
paper, the life-stage portion of the test had no impact on the Aerosol Consortium on Regulation and Science (IPAC-RS)
pass/fail disposition of the test for the BOU-to-EOU separa- Board and the IPAC-RS DDU Working Group for their
tions of up to 24% label claim (LC) because at least one consistent interest in this work and helpful feedback during
tolerance interval portion of the test fails when the life-stage manuscript preparation. The authors are also grateful to
BOU or EOU portion of the test fails. the FDA for the opportunity to participate in the joint
ACPS subgroup and interact to develop a PTI approach
CONCLUSION for the control of dose uniformity in OINDP. Special
thanks go to Bo Olsson and Dennis Sandell, whose work
A systematic investigation of the PTI-TOST proposed and vision inspired the IPAC-RS explorations of improved
by the FDA for control of DDU in OINDP has demon- DDU tests for OINDP. Finally, we acknowledge Walter
PTI-TOST Robustness to Deviations from Normality 835

High Excess Kurtosis High Excess Kurtosis High Excess Kurtosis

Tail = Left Tail = Right Tail = Both

Medium Excess Kurtosis Medium Excess Kurtosis Medium Excess Kurtosis

Density Tail = Left Tail = Right Tail = Both

Low Excess Kurtosis Low Excess Kurtosis Low Excess Kurtosis

Tail = Left Tail = Right Tail = Both

Normal Distribution Normal Distribution Normal Distribution

Tail = Left Tail = Right Tail = Both

60 80 100 120 140 60 80 100 120 140 60 80 100 120 140

DDU
Fig. 4. Densities for T-distributed random variables with tail proportion=0.0625 in each
specified tail (heavy-tailed distributions). See text for other details

0.00 0.02 0.04 0.06

Overall Overall Overall

Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4
Probability of passing

0.2

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2

0.0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion
Normal Dist. Medium Excess Kurtosis
Low Excess Kurtosis High Excess Kurtosis

Fig. 5. OC curves for shifted and scaled T distribution for N1/N2 =20/40
836 Novick et al.

0.00 0.02 0.04 0.06

Overall Overall Overall

Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4
Probability of passing

0.2

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2

0.0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion
Normal Dist. Medium Excess Kurtosis
Low Excess Kurtosis High Excess Kurtosis
Fig. 6. OC curves for shifted and scaled T distribution for N1/N2 =30/60

= 12 = 12 = 12
Tail = Left Tail = Right Tail = Both

= 9 = 9 = 9
Tail = Left Tail = Right Tail = Both

= 6 = 6 = 6
Tail = Left Tail = Right Tail = Both
Density

= 3 = 3 = 3
Tail = Left Tail = Right Tail = Both

= 0 = 0 = 0
Tail = Left Tail = Right Tail = Both

60 80 100 120 140 60 80 100 120 140 60 80 100 120 140

DDU
Fig. 7. Densities for bivariate normal random variables with tail proportion=0.0625 in
specified tail(s) (bivariate normal distributions). See text for other details
PTI-TOST Robustness to Deviations from Normality 837

0.00 0.02 0.04 0.06

Overall Overall Overall

Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4
Probability of passing

0.2

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2

0.0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion

0 6 12
3 9
Fig. 8. OC curves for bivariate normal distribution for N1/N2 =20/40. The legend box displays values for Delta

0.00 0.02 0.04 0.06

Overall Overall Overall

Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4
Probability of passing

0.2

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2

0.0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion

0 6 12
3 9
Fig. 9. OC curves for bivariate normal distribution for N1/N2 =30/60. The legend box displays values for Delta
838 Novick et al.

0.00 0.04 0.00 0.04

Tail = Both Tail = Both Tail = Both Tail = Both Tail = Both
∆=0 ∆=3 ∆=6 ∆=9 ∆ = 12
1. 0
0. 8
0. 6
0. 4
0. 2
0. 0
Tail = Right Tail = Right Tail = Right Tail = Right Tail = Right
∆=0 ∆=3 ∆=6 ∆=9 ∆ = 12
Probability of passing

1. 0
0. 8
0. 6
0. 4
0. 2
0. 0
Tail = Left Tail = Left Tail = Left Tail = Left Tail = Left
∆=0 ∆=3 ∆=6 ∆=9 ∆ = 12
1. 0
0. 8
0. 6
0. 4
0. 2
0. 0
0.00 0.04 0.00 0.04 0.00 0.04
Tail proportion
Tier 1, No Overall, No
Tier 1, Yes Overall, Yes
Fig. 10. OC curves for bivariate normal with and without life-stage means test for N1/N2 =20/40. Yes and No in
the legend box refer to the presence or absence of the life-stage-means portion in the test, respectively
0.00 0.04 0.00 0.04

Tail = Both Tail = Both Tail = Both Tail = Both Tail = Both
∆=0 ∆=3 ∆=6 ∆=9 ∆ = 12
1. 0
0. 8
0. 6
0. 4
0. 2
0. 0
Tail = Right Tail = Right Tail = Right Tail = Right Tail = Right
∆=0 ∆=3 ∆=6 ∆=9 ∆ = 12
Probability of passing

1. 0
0. 8
0. 6
0. 4
0. 2
0. 0
Tail = Left Tail = Left Tail = Left Tail = Left Tail = Left
∆=0 ∆=3 ∆=6 ∆=9 ∆ = 12
1. 0
0. 8
0. 6
0. 4
0. 2
0. 0
0.00 0.04 0.00 0.04 0.00 0.04
Tail proportion
Tier 1, No Overall, No
Tier 1, Yes Overall, Yes
Fig. 11. OC curves for bivariate normal with and without life-stage means test for N1/N2 =30/60. Yes and No in
the legend box refer to the presence or absence of the life-stage-means portion in the test, respectively
PTI-TOST Robustness to Deviations from Normality 839

0.00 0.02 0.04 0.06

Overall Overall Overall

Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4

0.2
Probability of passing

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1. 0

0. 8

0. 6

0. 4

0. 2

0. 0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion
Prob of outlier
0 0.005 0.05
0.001 0.01
Fig. 12. OC curves for normal distribution with non-repeating outlier for N1/N2 =20/40

0.00 0.02 0.04 0.06

Overall Overall Overall

Tail = Left Tail = Right Tail = Both
1.0

0.8

0.6

0.4
Probability of passing

0.2

0.0
Tier 1 Tier 1 Tier 1
Tail = Left Tail = Right Tail = Both
1. 0

0. 8

0. 6

0. 4

0. 2

0. 0

0.00 0.02 0.04 0.06 0.00 0.02 0.04 0.06

Tail proportion

Prob of outlier
0 0.005 0.05
0.001 0.01
Fig. 13. OC curves for normal distribution with non-repeating outlier for N1/N2 =30/60
840 Novick et al.

Hauck for his original proposal to use parametric REFERENCES

approaches for DDU testing, starting with his 1999
presentation at the Management Forum Conference on 1. Golden M. Parametric tolerance interval (PTI) test for delivered dose
European/FDA Regulatory Issues in Oral Inhalation and uniformity (DDU) for orally inhaled and nasal drug products
Nasal Delivery. This article is part of a series that presents (OINDP). Presentation to advisory committee for pharmaceutical
science on 26 October 2005. Page 3. http://www.fda.gov/ohrms/
a factual description and analysis of the test proposed by dockets/ac/05/slides/2005-4187S2_10_Golden.ppt Accessed 11 Sept
the FDA and should not be construed as endorsement or 2008.
advocacy by the authors or organizations with which they 2. Everitt BS, Hand DJ. Finite Mixture Distributions. NY: Chapman
are affiliated. and Hall; 1981.