Downloaded from journal.pda.

org on January 19, 2016

Charting and Evaluation of Environmental Microbial

Monitoring Data
Raphael Bar

PDA J Pharm Sci and Tech 2015, 69 743-761

Access the most recent version at doi:10.5731/pdajpst.2015.01079
 Downloaded from journal.pda.org on January 19, 2016

TECHNOLOGY/APPLICATION

Charting and Evaluation of Environmental Microbial

Monitoring Data
RAPHAEL BAR

BR Consulting, Rehovot, Israel 76247; email: rbar@netvision.net.il ©PDA, Inc. 2015

ABSTRACT: Statistical tools are required to organize and present microbial environmental monitoring data for the purpose
of evaluating it against regulatory action limits and of determining if the microbial monitoring process is in a state of
control. This paper applies a known methodology of a simple and straightforward construction of control XmR (X data
and moving range) charts of individual microbial counts as they are or of contamination rates derived from them,
irrespective of the type of the parent data distribution and without the need to transform the data into a normal distribution.
Plotting of monthly and cumulative sample contamination rates, as newly suggested by USP ⬍1116⬎, is also shown. Both
types of the control charts and plots allow an evaluation of the behavior of the microbial monitoring process. After
addressing the magnitude of microbial counts expected in environmental monitoring samples, this paper presents the
rationale behind the use of XmR charts. Employing data taken from environmental monitoring programs of pharmaceu-
ticals manufacturing facilities, this paper analyzes examples of (1) microbial counts from passive or active air sampling in
area Grade D or B or Class 100,000 in XmR charts, (2) contamination recovery rates as suggested by USP ⬍1116⬎ from
active air samples in area Grade B and contact plates in area Grade C, and (3) instantaneous contamination rates with
calculations illustrated on microbial counts of contact plates in area Grade D.

KEYWORDS: Statistical process control (SPC), Control charts of individual values, Control limits, Poisson distribution,
Normal distribution, Controlled rooms, Contamination recovery rates, Instantaneous contamination rates, USP ⬍1116⬎.

LAY ABSTRACT: Pharmaceutical companies conduct environmental monitoring programs, and samples of air (active and
passive sampling) and of surfaces (contact plates) are routinely tested for microbiological quality. Thus, hundreds of
microbial counts of tested environmental monitoring samples are routinely generated and recorded. Statistical tools are
required to organize and present this abundant data for the purpose of evaluating it against regulatory action limits and
determining if the microbial monitoring process is a state of control. This paper has a two-fold purpose. The first purpose
is to provide microbiologists and quality assurance personnel simple and straightforward tools of statistical process control
for evaluating the behavior of the microbial monitoring process: individual XmR (X data and moving range) control charts
of microbial counts as they are or of rates derived from them are constructed irrespective of the type of the parent data
distribution and without the need to transform the data into a normal distribution. Plotting of monthly and cumulative
sample contamination rates, as newly suggested by USP ⬍1116⬎, is also shown. The second purpose is to present
examples of the charting of (1) microbial counts, (2) contamination recovery rates as suggested by USP ⬍1116⬎, and (3)
instantaneous contamination rates using data taken from environmental monitoring programs of pharmaceuticals manu-
facturing facilities.

Introduction performance is stable and under a control state. Such a

Regulatory agencies expect pharmaceutical companies,
and particularly those that manufacture sterile drug prod-
ucts, to review and analyze the microbial environmental
monitoring (EM) data for determining whether the EM
doi: 10.5731/pdajpst.2015.01079
review can detect a trend or a shift in EM performance
that could be related to factors such as the effectiveness
of the agents and/or cleaning and sanitization procedures
employed in the manufacturing facilities. Regulatory
guides (1–3) impose action limits on the classified areas,
and companies often apply additional alert limits. Re-
cently, the USP 38 (4) suggested in a newly revised
chapter ⬍1116⬎ a new metric called contamination
recovery rate, which is defined as the rate at which

Vol. 69, No. 6, November–December 2015 743

 Downloaded from journal.pda.org on January 19, 2016
environmental samples are found to contain any level of
contamination. For example, an incident rate of 1%
would mean that only 1% of the samples taken have any
contamination regardless of colony number. This con-
cept of contamination recovery rate represents a new
paradigm introduced by USP and is definitely different
from the current approach based on current action limits
spanning the range of 1–200 cfu set by U.S. Food and
Drug Administration (FDA) guidance (3) and the Euro-
pean Union (EU) GMP Annex 1 (1) for the classified
environments. Irrespective of whether the EM data are
evaluated against action limits as required by regula-
tory agencies or against a limit for the contamina-
tion recovery rate as suggested by USP, it is re-
quired to demonstrate that the classified area is
maintained under an adequate microbial control.
Common statistical approaches for analyzing micro-
bial counts revolve around applying Poisson law, as
USP ⬍1223⬎ encourages us to do (5, 6), or around
fitting the data to other distributions (7). Conven-
tional statistical thinking would dictate that col-
lected microbial counts first be statistically evalu-
ated for a fit (goodness-of-fit test) to a given
distribution—such as Poisson or normal distribu-
tions, in the case of low and large counts, respec-
tively—and then be plotted onto control charts based
on either Poisson or normal distributions. When the
number of zero counts is higher than that expected
from Poisson distribution, fitting to either Poisson or
normal distributions usually fails, and then one can try
to fit the data set to the two-parameter negative bino-
mial distribution (9). This procedure requires ad-
vanced statistical analysis, and Hoffman (9) and Jarvis
(10) present examples on how to do so. But more
often, efforts are made into transforming counts (e.g.,
log or square transformations, etc.) to fit a normal
distribution prior to generating control charts. When
the number of zero counts becomes excessively high,
that is, a microbial contamination becomes a rare
event, data such as number of noncontamination op-
portunities between two successive contamination
events or time between contamination events are often
fitted to either a geometric or Weibull distribution and
then plotted on a g-chart or t-chart, respectively.
In contrast, this paper draws control charts from mi-
crobial counts as they are or rates derived from them,
irrespective of the type of the parent data distribution
and without the need to transform the data into a
normal distribution. By doing so, this paper follows
744
the methodology explained and advocated by Wheeler
in his numerous writings (10 –17).
After addressing the magnitude of microbial counts
expected in EM samples and the construction of an
XmR (X data and moving range) chart, this paper will
present the rationale behind the use of XmR charts and
give examples of the charting of
1. microbial counts
2. contamination recovery rates as suggested by USP
⬍1116⬎
3. instantaneous contamination rates
using data taken from EM programs of pharmaceuti-
cals production facilities.
Common Bioburden Levels of EM Samples
Regulatory documents of the EU and FDA impose
maximal limits of microbial contamination varying
between 1 and 200 cfu in an EM sample. These
limits, which are in fact taken to be action limits,
are 1, 5, 10, 25, 50, 100, and 200 cfu in an EM
sample in the European GMP Annex 1 (1), while
those of the FDA guide (3) are 1, 3, 5, 7, 10, 50, and
100 cfu in an EM sample. If we evaluate these
maximal levels of microbial counts with a Poisson
distribution, as advocated by USP ⬍1223⬎ (5), then
one can estimate with the following equation
Control Limits ⫽ ␮ ⫹ 3 冑␮
(1)
the true average counts (␮) expected in the EM samples
having the above cited regulatory action levels. Table I
lists the estimated average counts that lead to the
upper control limits that are almost identical to the
formal regulatory limits. So, the actual counts in
common EM samples are expected to vary between
the lower and upper control limits around the true
averages spanning the range of 0.6 –160 cfu per EM
sample.
XmR Charts of Microbial Counts
Microbial colonies counted on plates for surfaces
(contact plates) and for passive (settling plates) or
active (1m 3 ) air sampling are taken to be individual
values per site monitored and are evaluated with a
chart of individual data (XmR), in contrast to the
PDA Journal of Pharmaceutical Science and Technology
 Downloaded from journal.pda.org on January 19, 2016
TABLE I
Expected Average Microbial Counts in EM Samples Having the Regulatory Maximal Poisson Three-Sigma
Limits, Practically Identical with the Regulatory Limits
Formal Action Expected Count
Limit Average (␮)
1 0.1
3 0.6
5 1.0
7 2.4
10 4.0
25 13.0
50 30.0
100 70.0
200 160.0
common average and range chart. On the XmR
chart, counts or contamination rates are plotted on
the X chart with control limits calculated from the
short-term variation estimated from the difference
between one data point and the subsequent point
(moving range of two, mR) (18). Thus, the average
or the median of all consecutive ranges are used to
calculate the 3 sigma limits of both the X chart and
the mR chart with the following equations:
If the variation is based on the average moving range
(10, 11):
Limits of X ⫽ X៮ ⫾ 2.6596
⫻ Average Moving Range
Upper Limit of mR ⫽ 3.268
⫻ Average Moving Range
If the variation is based on the median moving range
(10, 11):
Limits of X ⫽ X៮ ⫾ 3.145
⫻ Median Moving Range
Upper Limit of mR ⫽ 3.865
⫻ Median Moving Range
Rationale for Using XmR Process Behavior Charts
The approach, originally based on Shewhart and
widely interpreted by Wheeler (10, 13) is based on the
following main elements:
Vol. 69, No. 6, November–December 2015
Lower Control Upper Control
Limit Limit
0.0 1.05
0.0 2.9
0.0 4.0
0.0 7.0
0.0 10.0
2.2 23.8
13.6 46.4
44.9 95.1
122.1 197.9
a. The 3 sigma limits of the control charts are viewed
as empiric limits that include a high percentage of
data (10, 12).
b. Data distribution needs not be normal (12). The
above three sigma limits are applicable to all data
distributions. Wheeler has calculated that the three-
sigma limits of various differently shaped distribu-
tions (e.g., normal, exponential, chi-square, etc.)
will result in more than 97.5% of the data included
within these limits (12) when a process is operated
predictably. Often, one attempts to fit a normal
distribution to data in order to have 99.73% of the
data included between three-sigma limits, entailing
only a minute proportion of out-of-control (OOC)
(2)
data of 0.27%. However, in real processes, the
proportion of OOC data is often higher. For in-
stance, the percent of OOC data in the examples
(3) evaluated in this paper is 4.3%, 2.0%, 11.5%, and
9.3% of the data recorded. Thus, if one adopts 2.5%
as an empirical upper limit of percent OOC data
points, then the fit to a normal distribution is not a
requirement.
c. Data need not be transformed to fit a normal dis-
(4)
tribution (12). Nonlinear transformations can even
distort the data, hide the signals, and change the
results of the analysis (12).
(5)
d. Basing the control limits on a certain data distribution
is valid provided it is shown to remain constant
throughout the monitoring of the process (13, 16, 20).
For instance, c-charts or g-charts are viable if the
underlying distribution parameters remain constant
throughout. However, this is often an unrealistic as-
745
 Downloaded from journal.pda.org on January 19, 2016

Figure 1

Poisson chart (upper) and XmR chart (lower) of computer randomly generated 10000 data following Poisson
distribution with ␮ ⴝ 4 counts. Only the last 100 data points are shown for the sake of the plot clarity.

sumption (20). Similarly, the assumption that the av-
erage contamination (Poisson distribution) in a site
within a clean room is constant when sampled daily
or weekly or monthly is most unlikely. This very
reason is a good justification per se for not basing
control limits of a control chart on the assumed dis-
tribution. Instead, the XmR chart captures the process
variation from the data as is and derives the empiric
control limits (14).
Thus, one can use microbial counts as they are with
three sigma limits for generating control charts in
order to evaluate the behavior of the EM process.
While controversies over the relationship between
control charting and hypothesis testing have been re-
ported and discussed (19), it is critically important to
make a clear distinction between inference statistics
and statistical process control (SPC) (15). Selecting a
proper distribution and goodness-of-fit tests are cer-
tainly needed if the purpose is to estimate parameters
of the distribution such as the true value and its
confidence intervals. However, if the purpose is to be
able to monitor the behavior of a process, that is, to be
able to state that the process behaves predictably or
not, and to be able to point out an OOC situation that
may take place at a frequency less than 2.5%, then the
empirical process XmR chart is a credible and suitable
way to do it.
In order to demonstrate the applicability of the XmR
charts to microbial counts that follow a Poisson dis-
tribution with average counts of 4 cfu or 70, as sug-
gested by Table I, tens of thousands of data points
were randomly computer generated with a Minitab 17
and plotted on control charts based on Poisson distri-
butions and as XmR charts. However, for the sake of
clarity, Figures 1 and 2 were recorded with only the
last 100 data points (9901–10,000) and these figures,
just like the complete figures, show that Poisson and
XmR charts reveal visibly similar control limits as
well as highly similar fractions of OOC data points

746 PDA Journal of Pharmaceutical Science and Technology

 Downloaded from journal.pda.org on January 19, 2016

Figure 2

Poisson chart (upper) and XmR chart (lower) of computer randomly generated 10000 data following Poisson
distribution with ␮ ⴝ 70 counts. Only the last 100 data points are shown for the sake of the plot clarity.

exceeding the three sigma limits. The exact percent- 70, and 160 cfu) that are expected to represent com-
ages of these OOC data points were counted from mon bioburden levels in EM samples. Table II lists the
additional control charts plotted with the Minitab 17 results, which happen to be essentially identical for
for other levels of microbial counts (␮ ⫽ 1, 4, 13, 30, both Poisson and XmR charts.

TABLE II
Calculated Percentages of Out-of-Control (OOC) Data Points Exceeding the Upper (UCL) and Lower
(LCL) Three-sigma Control Limits in Poisson and XmR Charts of Computer-Generated Random 10000
Data from a Poisson Distribution with Various ␮ Values

% OOC Data
Poisson XmR
Formal Action Expected Average Below Above Below Above
Limit Count (␮) LCL UCL Total LCL UCL Total
5 1 0 170 1.7 0 170 1.7
10 4 0 62 0.62 0 62 0.62
25 13 3 42 0.45 3 42 0.45
50 30 5 23 0.28 5 23 0.28
100 70 5 21 0.26 5 21 0.26
200 160 9 19 0.28 14 19 0.33

Vol. 69, No. 6, November–December 2015 747

 Downloaded from journal.pda.org on January 19, 2016
Figure 3
Poisson chart (upper) and XmR chart (lower) of computer randomly generated 30 data following Poisson
distribution with ␮ ⴝ 13 counts.
In practice, control charts are initially plotted on a
number of data points, typically 20 –30, and one can
compare the behavior of Poisson and XmR charts
drawn from the same data points. Figure 3 is such an
example of 30 data points randomly selected from a
Poisson population with ␮ ⫽ 13. Obviously, the
control limits will differ somewhat as they are cal-
culated differently, but they are similar. The impor-
tant point is that it may be concluded that the
process behaves predictably either according to a
Poisson c-chart or to an XmR chart. Occasionally, a
borderline data point may appear located close to
a control limit in one type of chart and outside the
control limit in another type of chart. Such a case is
shown in Figure 4, which plots Poisson data with
␮ ⫽ 4: count data point 10 is slightly lower than the
upper control limit of the c-chart but slightly above
that of the XmR chart. Such an occasional OOC
point is not frequent and if it occurs, and it can only
prompt an investigation, but the overall behavior of
the process conveyed by both types of charts re-
mains essentially the same.
748
Examples of XmR Control Charts of EM
Microbial Counts
Example 1: Microbial Counts in Settling Plates for
Passive Air Sampling in Class 100,000
Figure 5 shows microbial of settling plates in a
100,000 class room collected over a period of ten
months. Each count is a sum of the individual counts
obtained on both Sabouraud’s dextrose agar (SDA)
and triptic soy agar (TSA) petri dishes. None of the 23
data points follow either Poisson or normal distribu-
tions as shown by goodness-of-fit test results of P ⫽
0.00 made with a Minitab 17. Hence, while drawing a
c-chart is not justified in this case, one can attempt to
apply data transformation and indeed, the data were
transformed first with log (count ⫹ 1) and plotted in
Figure 6 along with a log transformed value of 1.71 as
the action limit. Data point 21 is shown to be OOC.
The same result is also observed in Figure 5, in which
the original counts are plotted on an XmR chart. This
figure shows an extreme count of 232 cfu, which is
PDA Journal of Pharmaceutical Science and Technology
 Downloaded from journal.pda.org on January 19, 2016
Figure 4
Poisson chart (upper) and XmR chart (lower) of computer randomly generated 30 data following Poisson
distribution with ␮ ⴝ 4 counts.
well above the formal action limit of 50 cfu. The XmR
chart can be further adjusted to the data values by
omitting data point 21 for calculation of the three
sigma limits (Figure 7). No counts exceed the upper
control limit of 27.3 (⬃ 27 cfu) and the control chart
shows that the process behaves predictably until
observation 20, following which, the two subse-
quent microbial counts are OOC and call for an
investigation. Indeed, the high mR values at the
subsequent observations 21 and 22 are OOC data,
and this supports the OOC situation detected in the
X chart. Since Figures 7–10 are XmR charts, which
record counts as they are, a line of a regulatory
action limit was added for conveniently signaling a
deviation from this limit.
Example 2: Microbial Counts in Settling Plates for
Passive Air Sampling in Grade D
Figure 8 shows microbial counts on settling plates
collected almost weekly for a 1 year period in a
Vol. 69, No. 6, November–December 2015
Grade D environment. Each count is a sum of the
individual counts obtained on both SDA and TSA
petri dishes. Unlike the previous example, the
counts here are in general lower and include more
zero counts. About 25% of the data are zero and this
usually makes a fit to either a Poisson or normal
distribution less likely. Indeed, none of the 49
counts follow either Poisson or normal distributions
as shown by goodness-of-fit test results of P ⫽ 0.00
made with a Minitab 17. Furthermore, conversion of
the data with log (count ⫹ 1) or square root
(count ⫹ 0.5) transformations did not normalize the
data, as evidenced by the resulting P-values, which
are lower than 0.05. Once again, one can directly
plot the original counts onto an XmR chart and
obtain a plot that depicts realistically the behavior
of the microbial monitoring process in this Grade D
environment. Indeed, Figure 8 shows a well-be-
haved microbial monitoring process of the specific
site monitored, which does not exceed an upper
limit of 9.92 cfu (⬃10 cfu) except for a single OOC
749
 Downloaded from journal.pda.org on January 19, 2016

Figure 5

XmR chart of microbial counts of passive air samples recorded in Class 100,000 environment.

data point at 17. One can recalculate the control
limits disregarding this OOC point and obtain a
slightly narrower upper limit of 8.92 cfu (⬃9 cfu).
In either case, the counts are well below the regu-
latory action limit of 100 cfu.
Example 3: Microbial Counts in Settling Plates for
Active Air Sampling in Grade D
Figure 9 shows microbial counts on TSA plates
collected almost weekly with a microbiological air

Figure 6

XmR chart of log transformed microbial counts of passive air samples recorded in Class 100,000 environment.

750 PDA Journal of Pharmaceutical Science and Technology

 Downloaded from journal.pda.org on January 19, 2016

Figure 7

XmR chart of microbial counts of passive air samples recorded in Class 100,000 environment after omitting the
extreme count of 232 cfu.

sampler during a 1 year period in a Grade D envi- counts follow either Poisson or normal distributions
ronment. About 23% of the data are zero and again, as shown by goodness-of-fit test results of P ⫽ 0.00
this usually makes a fit to either a Poisson or normal made with a Minitab 17. Furthermore, conversion of
distribution less likely. Indeed, none of the 52 the data with log (count ⫹ 1) or square root

Figure 8

XmR chart of microbial counts of passive air samples recorded during one year in area Grade D.

Vol. 69, No. 6, November–December 2015 751

 Downloaded from journal.pda.org on January 19, 2016
Figure 9
XmR chart of microbial counts of active air samples recorded during one year in area Grade D.
(count ⫹ 0.5) transformations did not normalize the
data as evidenced again by the resulting P-values,
which are lower than 0.05. So, Figure 9 is an XmR
chart with the control limits calculated from a me-
dian range (eqs 4 and 5) in order to minimize the
impact of the extreme counts rather than with the
usual averages range (eqs 2 and 3). One can also
decide to draw the control limits with all high
counts disregarded.
Example 4: Microbial Counts in Settling Plates for
Active Air Sampling in Grade B
Figure 10 shows microbial counts on TSA plates
collected almost weekly with a microbiological air
sampler during a 1 year period in a Grade B envi-
ronment. About 68.5% of the data are zero and
again, this usually makes a fit to either a Poisson or
normal distribution much less likely. As in the
previous example, none of the 54 counts follow
either Poisson or normal distributions of original or
transformed counts, as shown by goodness-of-fit
test results of P ⫽ 0.00 made with a Minitab 17. The
752
number of noncontaminated samples (count ⫽ 0) is
so large that the average count may now get closer
to 1 cfu or can become ⬍1 cfu. In the latter case,
whenever the average count is ⬍1, the XmR chart
becomes inefficient for interpreting a control chart
(16). An attempt to re-draw the XmR chart with all
the OOC data points (data points 3, 25, 26, 38, and
45) omitted in the calculation of the control limits
makes the average count ⬍1 (average count ⫽ 0.65
cfu). Under these conditions, one can switch to a
different approach that is based on evaluation of
contamination as a rare event.
Evaluation of Contamination Rates of EM Samples
for Cases of Rare Contaminations
When the majority of the EM samples show no mi-
crobial growth, as is the practical and regulatory ex-
pectation from environments of Class 100, Grades A
and B, then the occurrence of a single contaminated
plate becomes a special event, a rare event. Other
examples of rare events include hospital-acquired in-
fections, medication errors, or rare deviations in man-
PDA Journal of Pharmaceutical Science and Technology
 Downloaded from journal.pda.org on January 19, 2016

Figure 10

XmR chart of microbial counts of active air samples recorded during one year in area Grade B.

ufacturing processes. In EM, a contaminated sample is a. The Number of Noncontamination Opportunities

viewed as a rare event, irrespective of whether its between Contamination Events
shows, for example, 1, 5, or 15 cfu. Thus, instead of
monitoring the very few contamination events, one This number is used to generate a g-chart (based on
can track the values that follow. geometric distribution) or the time elapsed between

TABLE III
Percent Monthly and Cumulative Percent Contamination Recovery Rate Calculated per USP <1116> on
Microbial Counts of Active Air Sampling in Classified Area Grade B

# of
Samples # of Cumulative
Tested Cumulative Contaminated Cumulative # of Monthly % %
per # of tested Samples per Contaminated Recovery Recovery
Month Month samples month Samples Rate Rate
Jan 5 5 4 4 80.0 80.0
Feb 4 9 1 5 25.0 55.6
Mar 2 11 0 5 0.0 45.5
Apr 5 16 2 7 40.0 43.8
May 4 20 3 10 75.0 50.0
Jun 5 25 2 12 40.0 48.0
Jul 5 30 1 13 20.0 43.3
Aug 4 34 0 13 0.0 38.2
Sep 4 38 1 14 25.0 36.8
Oct 5 43 2 16 40.0 37.2
Nov 6 49 1 17 16.7 34.7
Dec 5 54 0 17 0.0 31.5

Vol. 69, No. 6, November–December 2015 753

 Downloaded from journal.pda.org on January 19, 2016

90.0
Monthly and Cumulave % Contaminaon Recovery Rates
80.0 80.0
80.0 Monthly
75.0
% Recovery Rate
70.0

% Contaminaon Recovery rate

Cumulave
% Recovery Rate
60.0 55.6
50.0
50.0 45.5
48.0
43.8 43.3
40.0 40.0 38.2 40.0
40.0 36.8 37.2
34.7
31.5
30.0 25.0 25.0
20.0
20.0 16.7

10.0
0.0 0.0 0.0
0.0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Figure 11

Monthly and cumulative contamination recovery rates (per USP <1116>) calculated from the microbial counts
of active air samples recorded during one year in area Grade B (see Table III).

contamination events to generate a t-chart (based on to be unrealistic, this possibility is not followed in this
either Weibull or exponential distributions) (17). paper.

These control charts for rare events are available in b. The Percent Microbial Contamination
commercial statistical software programs but prior to Recovery Rate
their use, the data need to demonstrate a goodness-of-
fit. In any case, as explained above, the use of these This metric, as previously mentioned, was recently
special control charts assumes a constant underlying suggested by USP ⬍1116⬎. The recommended max-
probability of the given distribution throughout the imal rate limits for the various International Organi-
long period of EM data collection. Since this is taken zation for Standardization (ISO) classified environ-

TABLE IV
Percent Monthly and Cumulative Excursions Rate of Microbial Counts of Active Air Sampling in Classified
Area Grade B
# of Samples
Tested per Cumulative # of # of Excursions Cumulative # Monthly % % Cumulative
Month Month Tested Samples per Month of Excursions Excursions Excursions Rate
Jan 5 5 1 1 20.0 20.0
Feb 4 9 0 1 0.0 11.1
Mar 2 11 0 1 0.0 9.1
Apr 5 16 0 1 0.0 6.3
May 4 20 0 1 0.0 5.0
Jun 5 25 1 2 20.0 8.0
Jul 5 30 1 3 20.0 10.0
Aug 4 34 0 3 0.0 8.8
Sep 4 38 1 4 25.0 10.5
Oct 5 43 0 4 0.0 9.3
Nov 6 49 1 5 16.7 10.2
Dec 5 54 0 5 0.0 9.3

754 PDA Journal of Pharmaceutical Science and Technology

 Downloaded from journal.pda.org on January 19, 2016
Figure 12
Monthly and cumulative excursions (above Alert and Action limits) rates calculated from the microbial counts
of active air samples recorded during one year in area Grade B (see Table IV).
ments for production of sterile products are ⬍1%,
⬍3%, ⬍5%, and ⬍10% for ISO 5, 6, 7, and 8, respec-
tively. USP ⬍1116⬎ states that the actual monitoring
data should be retabulated monthly. The monthly con-
tamination rate is given by:
Monthly % Contamination Recovery Rate
Number of contaminations within a given month
⫽ ⫻ 100%
Number of samples tested within a given month
(6)
The cumulative contamination rate can be calculated
with:
Cumulative % Contamination Recovery Rate
Number of all contaminations obtained till
a given month
⫽ ⫻ 100% (7)
Total number of samples tested
till a given month
where the total number of samples includes those of
the given month.
The contamination recovery rate, expressed as a cumu-
lative percentage of contaminated EM samples from all
tested samples can be updated at least monthly and
Vol. 69, No. 6, November–December 2015
tabulated or plotted as a function of time. But, the
monthly (or weekly) recovery rates determined with eq 6
can also be plotted on an XmR chart to evaluate the
environmental process performance as expressed by ei-
ther a decrease or an increase in the monthly contami-
nation rate, a trend upward or downward, or a stable
behavior within control limits. In Class 100 and Grade A
environments, the counts are expected to be mostly zero,
and many times all counts are zero. In the latter case, one
can only state that the percent contamination recovery
rate is estimated with:
% Contamination Recovery Rate
1
⬍ ⫻ 100%
Total number of zero samples ⫹ 1
(8)
While the USP ⬍1116⬎ limits of contamination recov-
ery rates appear to be suitable to ISO 5 and 6 or Grade A
and B environments, where contamination is expected to
be rare, they appear to be rather very stringent for the
ISO 7 and 8 environments. The newly revised Parenteral
Drug Association (PDA) technical report states indeed
755
 Downloaded from journal.pda.org on January 19, 2016

80.0
Monthly and Cumulave % Contaminaon Recovery Rates
75.0
Monthly
70.0 % Recovery Rate
Cumulave

% Contaminaon Recovery rate

60.0 % Recovery Rate
50.0
50.0

40.0 36.4
33.3
31.3 30.0
30.0 25.0 25.0
20.0 20.7
20.0 18.2 17.1
15.4 14.0 12.5
10.0
0.0
0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
0.0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Figure 13

Column graph of monthly and cumulative contamination recovery rates (per USP <1116>) calculated from
contact plates recorded during one year in area Grade C.

that the recovery rates for ISO 7 and 8 environments may % Instantaneous Contamination Rate
not be obtainable. Indeed, a recovery rate limit of 10% (Number of contaminations per
for ISO class 8 (class 100000) implies that 90% of the
samples tested should show no growth despite the fact 100 time units)
that they are sampled from a nonsterile environment. In 1
any case, PDA Technical Report 13 makes the wise ⫽ ⫻ 100%
Number of time units between
suggestion that if one wishes to track the recommended two successive contaminations ⫹ 1
recovery rates, it should be done parallel to the presently
(10)
binding regulatory action limits (7).

If EM samples are not tested regularly with time, then

c. The Instantaneous Contamination Rates
the instantaneous rate is expressed as the percent of
contaminated samples from the number of all samples
The instantaneous rate may be determined as the num-
tested since the last contamination during the specified
ber of contaminations per a unit of time (e.g., day,
time period, using eq 11.
week, month) if EM samples are tested regularly with
time. The rate is instantaneous when calculated from
the number of unit times elapsed from the previous % Instantaneous Sample Contamination Rate
contamination to the present one (including the time (Number of contaminations
unit where the contamination occurred) (17) and is
per 100 samples)
expressed as the number of contaminations per unit of
time (eq 9) or as the number of contaminations per 100 1
⫽ ⫻ 100%
time units (eq 10). Number of samples between two
successive contaminations ⫹ 1
Instantaneous Contamination Rate (Number of (11)
contaminations per time unit)
It is worth noting that while eq 6 determines the
1 percent of contaminated samples obtained within a
⫽
Number of time units between two successive month from all samples tested during that month, eq
contaminations ⫹ 1 11 expresses the percent of contaminated samples per
(9) 100 samples from the previous contamination to the

756 PDA Journal of Pharmaceutical Science and Technology

 Downloaded from journal.pda.org on January 19, 2016

Figure 14

Line graph of monthly and cumulative contamination recovery rates (per USP <1116>) calculated from
contact plates recorded during one year in area Grade C (same data as Figure 13).

present one. Plotting the instantaneous rates (from eqs
9, 10, or 11) on XmR charts again allows one to depict
the monitoring process behavior, informing when the
performance improves (lower contamination rate), di-
minishes (higher contamination rate), trends upward
or downward, or behaves predictably.
Examples of Charting of EM Contamination Rates
Example 5: Microbial Contamination Recovery Rate in
Settling Plates for Active Air Sampling in Grade B
The data set that was plotted on an XmR chart (Figure
10) is reanalyzed to evaluate the percent contamina-
tion recovery rates as suggested by USP ⬍1116⬎. The
high proportion of zero counts renders the data set
more suitable for a direct evaluation of the percent
incident and cumulative sample rates of contamina-
tions. Accordingly, the counts of the Grade B area,
plotted in Figure 10, are regrouped per month and the
monthly percent and the cumulative sample rates of
contaminations were calculated with eqs 6 and 7,
respectively. These are listed in Table III and plotted
in Figure 11 for better visualization.
In this specific case, the rates are quite high, in con-
trast to the much lower USP limits of ⬍1% or ⬍5% if
Grade B is equated to ISO 5 (at rest) or ISO 7 (in

TABLE V
Instantaneous Sample Contamination Rates in Classified Area Grade D*

# of Non- Instantaneous Sample Instantaneous % Sample

Date of Contaminated Contamination Rate Contamination Rate
Contamination Samples (contamination per sample) (contaminations per 100 samples)
02.01.13 — — —
18.04.13 12 0.077 7.7
22.04.13 0 1.000 100.0
06.05.13 1 0.500 50.0
21.05.13 1 0.500 50.0
11.06.13 3 0.250 25.0
02.07.13 3 0.250 25.0
28.07.13 3 0.250 25.0
24.12.13 22 0.043 4.3
* This data is used for illustrating similar calculations for classified environments of Class 100, ISO 5, and Grade A.

Vol. 69, No. 6, November–December 2015 757

 Downloaded from journal.pda.org on January 19, 2016

Figure 15

Percent instantaneous sample contamination rates (number of contaminated samples per 100 samples tested)
calculated from contact plates recorded during one year in area Grade D (see Table V).

operation) classes, respectively. Interestingly, only
one count (11 cfu) exceeds the EU GMP limit of 10
cfu, showing again that the USP approach is certainly
more stringent. The cumulative sample contamination
rate may decrease or increase from month to month
and this behavior is explained when the monthly rate
is also simultaneously evaluated. For instance, the
cumulative rate dropped from 55.6% in February to
45.5% in March because no contaminated samples
were detected during March. Instead of tabulated data,
a chart such as Figure 11 has the advantage of an
easier visual display of data behavior. The calculated
numbers of the contamination rates are also annotated
on top of the columns in the chart.
PDA technical report 13 (7) suggests monitoring the
excursions rate as well, that is, the fraction of the samples
exceeding the alert limit and/or action limit. Table IV
lists the monthly number of EM samples exceeding the
alert and action limits among the same data set presented
in Table III for Grade B. Figure 12 is a visual display of
the percent excursions rate and it is seen that each
decrease of the monthly rate entails a decrease of the
corresponding cumulative excursions rate.

TABLE VI
Example of Instantaneous Contamination Rates in Classified Area Grade D*

Date of # of Days Till Instantaneous Contamination Instantaneous % Contamination

Contamination Contamination Rate (contamination per day) Rate (contaminations per 100 days)
2/1/2013 — — —
18/4/2013 106 0.009 0.9
22/4/2013 4 0.250 25.0
06/05/2013 14 0.071 7.1
21/05/2013 15 0.067 6.7
11/06/2013 21 0.048 4.8
02/07/2013 21 0.048 4.8
28/07/2013 26 0.038 3.8
24/12/2013 149 0.007 0.7
* This data is used for illustrating similar calculations for classified environments of Class 100, ISO 5, and Grade A.

758 PDA Journal of Pharmaceutical Science and Technology

 Downloaded from journal.pda.org on January 19, 2016
Figure 16
Percent instantaneous contamination rates (number of contamination events per 100 days) calculated from
contact plates recorded during one year in area Grade D with the extreme rate omitted (see Table VI).
Example 6: Microbial Contamination Recovery
Rate in Contact Plates in Grade C
This example depicts recovery rates for contact plates
in classified area Grade C. The contact plates are both
SDA and TSA plates and any single contaminated
plate is taken as a contamination event. The recovery
rates were calculated according to USP guidance with
eqs 6 and 7 and are plotted as a column chart in Figure
13. This figure reveals a clear decreasing trend of
cumulative recovery rates and this is explained by the
absence of contamination during the months June–
December (the monthly recovery rates are zero). In-
terestingly, despite the high recovery rates, all counts
(not shown) were below the EU GMP limit of 25 cfu.
Alternatively, one may prefer to plot the same recov-
ery rates in a line chart rather than in a column chart
in order to visualize better any existing trend. This is
shown in Figure 14 along with a USP limit of 5% if
Grade C is equated to ISO class 7 (at rest). If addi-
tional data of the subsequent year are plotted on the
same chart and the contamination events remain rare,
the curve of the cumulative recovery rate will continue
its trend downward and meet the USP limit.
Example 7: Instantaneous Sample Contamination
Rates in Contact Plates in Grade D
As previously explained, the instantaneous rate refers
to a rate between two successive contamination
Vol. 69, No. 6, November–December 2015
events. This is often applicable for cases of rare events
as expected in the classified environments of Class
100, ISO 5, or Grade A. However, this usually requires
many data points across several years in order to
calculate instantaneous rates between two rare con-
taminations. Counts in contact plates in the Grade D
area are used to illustrate the calculations and the
charting even though the contamination is not rare
enough. Table V lists the number of noncontaminated
samples between two successive contaminated sam-
ples as well as the instantaneous sample contamination
rate, expressed per 100 samples, that was calculated
with eq 11. These rates are plotted in Figure 15 and
show how the contamination rate (number of contam-
inations per 100 samples tested) varies with time and
that on the average, 36 contaminations occur every
100 samples. If one omits data point 3 (date of
22.04.13) for calculating the control limits, one would
obtain a lower contamination rate with narrower con-
trol limits.
Example 8: Instantaneous Contamination Rates in
Contact Plates in Grade D
Fundamentally, rate is expressed per a unit of time and
one can plot the number of contamination events per a
unit of time (e.g., per day). The dates in Table V are
contamination dates and one can calculate the number
of days elapsed between two contaminations. Table VI
lists the number of days until the next contamination
759
 Downloaded from journal.pda.org on January 19, 2016
(inclusive of the contamination day). The instanta-
neous contamination rate (per day) calculated with eq
9 and the instantaneous percent contamination rate
(per 100 days) calculated with eq 10 are listed in Table
VI. Either rate can be plotted on an XmR chart and the
percent contamination rate is shown in Figure 16.
When the extreme data point 3 (at 22.04.13) is ig-
nored, the chart shows that, on average, 4.11 contam-
inations occur every 100 days.
Conclusions
A simple statistical process control tool, the XmR
chart of individual values, can be used to plot and
depict the behavior of the environmental microbial
monitoring process. To this purpose, microbial counts
of environmental samples are plotted as they are,
irrespective of the parent data distribution and with no
need to undergo a mathematical transformation. The
three-sigma control limits, computed from the moving
ranges that capture the short-term process variation,
include at least 97.5% of any parent data distribution,
and any microbial count point beyond these limits is
taken to be a rare occurrence and considered to be an
OOC point. Microbial counts from passive or active
air sampling in area Grade D or B or Class 100,000
were employed to construct XmR charts.
When microbial contamination becomes a rare event
as in classified areas of Grade A or B or ISO 5,
instantaneous rates of contamination between two suc-
cessive contamination events, expressed per unit of
time or per sample tested, can be plotted on an XmR
chart that depicts the behavior of the microbial mon-
itoring process. Alternatively, monthly and cumulative
sample contamination recovery rates, suggested re-
cently by USP ⬍1116⬎, can be plotted in column or
line charts that allow a visible evaluation of any ex-
isting trend. Examples of plots of these sample con-
tamination recovery rates as well as of rates of excur-
sion beyond alert and action limits are given for
microbial counts of active air samples in area Grade B
and contact plates in area Grade C, all recorded during
1 year.
References
1. EudraLex, The Rules Governing Medicinal
Products in the European Union. Vol. 4: EU
Guidelines to Good Manufacturing Practice:
Medicinal Products for Human and Veterinary
760
Use; Annex 1: Manufacture of Sterile Medicinal
Products, 2008.
2. WHO Technical Report Series 961. Annex 6:
WHO Good Manufacturing Practices for Sterile
Pharmaceutical Products, World Health Organiza-
tion, 2011.
3. Guidance for Industry for Sterile Drug Products
Produced by Aseptic Processing: Current Good
Manufacturing Practice, U.S. Food and Drug Ad-
ministration, 2004.
4. USP General Chapter ⬍1116⬎ Microbiological
Control and Monitoring of Aseptic Processing
Environments, USP 38/NF 33.
5. USP General Chapter ⬍1223 ⬎ Validation of
Alternative Microbiological Methods, USP 38/NF
33.
6. Hussong, D.; Madsen, R. Analysis of environmen-
tal microbiology data from cleanroom samples.
Pharmaceutical Technology, Aseptic Processing
2004, 28 (5, SUPP), 10 –15. http://images.alfres-
co.advanstar.com/alfresco_images/pharma/2014/
08/22/cc05da25-54a1-4fc9-84bb-e5c341b2b57e/
article-95525.pdf; accessed Dec. 31, 2014.
7. PDA Technical Report No. 13 (Revised): Funda-
mentals of an Environmental Monitoring Pro-
gram. Parenteral Drug Association, 2014.
8. Hoffman, D. Negative binomial control limits for
count data with extra-Poisson variation, Pharm.
Statistics 2003, 2 (2), 127–132.
9. Jarvis, B. Statistical Aspects of the Microbiolog-
ical Examination of Foods, 2nd ed.; Academic
Press: Amsterdam, Netherlands, 2008; p 30.
10. Wheeler, D. J.; Chambers, D. S. Understanding
Statistical Process Control, 4th ed.; SPC Press:
Knoxville, TN, 2010.
11. Wheeler, D. J. Individual charts done right and
wrong: how does your software rate? Quality
Digest [Online] 2010 http://www.qualitydigest.
com/inside/quality-insider-column/individual-
charts-done-right-and-wrong.html (accessed
December 31, 2014).
PDA Journal of Pharmaceutical Science and Technology
 Downloaded from journal.pda.org on January 19, 2016
12. Wheeler, D. J. Are you sure we don’t need nor-
mally distributed data? Quality Digest [Online]
2010 http://www.qualitydigest.com/inside/
twitter-ed/are-you-sure-we-don-t-need-normally-
distributed-data.html (accessed December 31,
2014).
13. Wheeler, D. J. Myths about process behavior
charts: how to avoid some common obstacles to
good practice. Quality Digest [Online] 2011
http://www.qualitydigest.com/inside/quality-
insider-article/myths-about-process-behavior-
charts.html (accessed December 31, 2014).
14. Wheeler, D. J. What makes the XmR chart work?
Quality Digest [Online] 2012 http://www.quality-
digest.com/inside/quality-insider-article/what-
makes-xmr-chart-work.html (accessed December
31, 2014).
15. Wheeler, D. J. Statistics and SPC: two things
sharing a common name can still be different.
Quality Digest [Online] 2014 http://www.quality-
digest.com/inside/quality-insider-column/statis-
tics-and-spc.html (accessed December 31, 2014).
Vol. 69, No. 6, November–December 2015
16. Wheeler, D. J. What is chunky data?: What hap-
pens when the measurement increment gets too
large? Quality Digest [Online] 2011 http://
www.qualitydigest.com/inside/quality-insider-
article/what-chunky-data.html (accessed Decem-
ber 31, 2014).
17. Wheeler, D. J. Working with rare events: what
happens when the average count gets very
small? Quality Digest [Online] 2011 http://
www.qualitydigest.com/inside/quality-insider-
article/working-rare-events.html (accessed De-
cember 31, 2014).
18. PDA Technical Report No. 59: Utilization of sta-
tistical methods for production monitoring. Par-
enteral Drug Association, 2012.
19. Woodall, W. H. Controversies and contradictions
in statistical process control. J. Qual. Technol.
2000, 32 (4), 341–330.
20. Laney, D. B. Improved control charts for attri-
butes. Quality Engineering 2002, 14 (4), 531–
537.
761
 Downloaded from journal.pda.org on January 19, 2016

An Authorized User of the electronic PDA Journal of Pharmaceutical Science and

Technology (the PDA Journal) is a PDA Member in good standing. Authorized Users are
permitted to do the following:

·Search and view the content of the PDA Journal

·Download a single article for the individual use of an Authorized User
·Assemble and distribute links that point to the PDA Journal
·Print individual articles from the PDA Journal for the individual use of an Authorized User
·Make a reasonable number of photocopies of a printed article for the individual use of an
Authorized User or for the use by or distribution to other Authorized Users

Authorized Users are not permitted to do the following:

·Except as mentioned above, allow anyone other than an Authorized User to use or access the
PDA Journal
· Display or otherwise make any information from the PDA Journal available to anyone other
than an Authorized User
·Post articles from the PDA Journal on Web sites, either available on the Internet or an Intranet,
or in any form of online publications
·Transmit electronically, via e-mail or any other file transfer protocols, any portion of the PDA
Journal
·Create a searchable archive of any portion of the PDA Journal
·Use robots or intelligent agents to access, search and/or systematically download any portion
of the PDA Journal
·Sell, re-sell, rent, lease, license, sublicense, assign or otherwise transfer the use of the PDA
Journal or its content
·Use or copy the PDA Journal for document delivery, fee-for-service use, or bulk reproduction or
distribution of materials in any form, or any substantially similar commercial purpose
·Alter, modify, repackage or adapt any portion of the PDA Journal
·Make any edits or derivative works with respect to any portion of the PDA Journal including any
text or graphics
·Delete or remove in any form or format, including on a printed article or photocopy, any
copyright information or notice contained in the PDA Journal