Evaluation of Historical Data

Earlier in the discussion of process validation strategies, 20 production batches
were suggested as a minimum number upon which to draw conclusions about
the validity of the process. In this particular example, however, two distinct
methods of drying are provided. In order to have sufficient history on each
operation, the number of batches examined was increased to 30.
The batches were selected so that the same number was dried by each
process. For the other critical manufacturing steps and release tests listed in
Table 1, data were collected for all 30 batches.
The first manufacturing step, premix blending time, was consistently reported
as 10 min, but with one exception. In this instance, the powders were
tumbled for 20 min, which is still within the limits (10 to 20 min) prescribed
by the batch record. It would be interesting to know if this source of variability can materially affect attributes of the final product. Unfortunately, having only
one batch produced by the 20-min process does not permit statistically valid
comparisons. At best, test results for the single 20-min batch can be screened
using summary data from the remainder of the study. Under different circumstances,
batches would have been grouped by mixing time and compared by
dosage form attributes. More than likely, subsequent manipulation of the blend
would have negated any contribution, allowing us to conclude that a mixing
time of 10 to 20 min is not unreasonable.
At the wet milling step we encounter a situation similar to preblending;
that is, only two of the 30 study batches are prepared using the no. 5 drilled
screen. The no. 7 is obviously the screen of choice. The purpose of this step is
to produce particles of reasonably uniform size, which in turn will improve
drying. From the records, we also know that the no. 5 screen was used
only with batches that were tray dried. Elapsed drying time and residual
moisture were compared for the two batches from the no. 5 screen process
and the other 13 batches that were tray dried. No important differences were
detected. Still, in light of the limited use of the no. 5 screen, it would not
be inappropriate to recommend this option be eliminated from the processing
instructions.
Mean drying time for the oven tray process is 19.2 hr. All 15 batches
were dried within the specified time of 16 to 20 hr. No seasonal influence was
apparent. The average moisture content of these batches is 1.2%; the standard
deviation is 0.3%. The 15 batches dried using the fluid bed dryer had a residual
moisture of 0.8% (SD = 0.1%). Drying time is mechanically controlled and not
recorded. The statistics favor the fluid bed process; it is more efficient and
uniform. There is nothing in these data to disqualify the oven tray dryer from
further use, however.
Oscillation of the dried granulation and lubricant was accomplished in
every instance using a no. 10 wire screen. Reference to the no. 12 screen, the
alternative method for pulverizing the batch, must be deleted from the manufacturing
instructions for the process to be validated retrospectively.
The final mix blending time was reported as either 10 or 15 min. Twentyone
of the 30 batches were tumbled for 10 min and the remainder were mixed
for 15 min. The mixing time is not mechanically controlled or automatically
recorded; it is left to the operator to interpret elapsed time. Because of the
importance of the step to distribution of the therapeutic agent, a comparison was
made between the distribution of the percentage of relative tablet potency [(tablet
assay/tablet weight) × 100] for the two mixing times. The frequency distributions
of the two populations are shown in Figure 3.
The two histograms are visually different, with the 15-min process exhibiting
more dispersion. Despite this difference both populations are tightly grouped,
which is a reflection of the uniformity of the blend.The processes may be studied quantitatively by comparing the means and
standard deviations of the two populations. The effect of final blend time on
lubricant distribution was examined by comparing disintegration time statistics
for the grouped data. None was noted.
The moisture content of the 15 tray-dried batches following final mix
remained essentially unchanged from the drying step. The batches from the fluid
bed process gained moisture. This is probably attributable to handling very dry
material in a relatively humid environment. Both groups are still below the
target for this step of 1.5 %, however.
Table 2 gives a comparison of the moisture contents following the drying
and tumbling steps. The sizable increase in mean moisture content of the fluid
bed-dried batches deserves further study. To determine whether or not all
batches were uniformly affected, the mean moisture content was plotted in the
order in which the batches were produced. Whereas the plot for the tray-dried
batches is unremarkable, the fluid bed process chart (Fig. 4) depicts an unnatural
pattern. Further investigation discloses that heating, ventilation, and air condition
(HVAC) problems were experienced by the area in which a number of
these batches were blended.
During compression, 1000 tablets were randomly selected for use by quality
control. Inspection of the batch records revealed that all 30 batches were
compressed on the same model press operating at approximately the same speed.
All presses were fed by overhead delivery systems of the same design, thus
tableting equipment will not be a source of variability from batch to batch.
The test for disintegration is performed as described in the USP, and the
results are rounded to the nearest half-min. Disintegration time varied over a
narrow range for all batches studied. The 15-batch average for the tray dryer
process (2.7 min) is well below the specification (10 min) for this test. Hardness
of tablets from the tray dryer process averaged 15 Strong–Cobb units (SCU).
All batches exceeded the minimum specification (9 SCU); there is no upper limit. Hardness and disintegration time are not well correlated, probably due to
rounding of test results and the need to compare averages.
On average, tablets from the fluid bed process were slightly harder. Also,
the individual batches had a greater range of hardness than batches from the
alternative drying process. Disintegration time for the fluid bed process averaged
3.0 min. Individual batches ranged from2.0 to 4.5 min. As with the tray process,
no correlation was found between hardness and disintegration time. In summary,
tablets from the fluid bed dryer process were somewhat harder and took slightly
longer to disintegrate. (See Table 2.) These differences are considered insignificant,
however. If any recommendations were made, it would be to lower the
disintegration time specification or establish an internal action limit closer to
the historical upper range of the process.
Control charts were plotted for hardness and average tablet weight (ATW)
to evaluate process performance over time. Separate charts were prepared for
the tray dryer and fluid bed processes. Hardness values are an average of 10
individual measurements. The ATW subgroups are the result of weighing 20
tablets individually. The control charts were inspected for trends and evidence
of instability using well-established methods [9]. Only the control chart for hardness
of tablets from the fluid bed process responded to one of the tests for
pattern instability (Fig. 5); that is, two of three consecutive points exceeded the
2-sigma limit. From the chart it is obvious the general trend toward greater
tablet hardness (from 11 to 25 SCU) is the underlying cause of the instability.
The trend to greater hardness was subsequently arrested and may have to do with attempts to regulate another tablet variable—thickness, for example—
although the records are vague in this regard.
Water content of the bulk tablets irrespective of the drying process was
higher than at the final mix stage (Table 2). This is probably due to the compression
room environment and the low initial moisture of the powder. Still, the
specification limit of 2% is easily met.
The FDA has recently issued draft guidelines that recommend blend uniformity
analysis for all products for which USP requires content uniformity
analysis [10]. The USP requires this test when the product contains less than 50
mg of the active ingredient per dosage form or when the active ingredient is
less than 50% of the dosage form by weight. The concern FDA has is that if
blend uniformity is not achieved with mixing of the final granulation, then some
dosage units are likely not to be uniform [11]. Blend uniformity is not routinely
determined for drug A, nor is there a requirement because the dosage form is
over 50% active ingredient. In the absence of historical information about uniformity
of the blend, the relationship between tablet weight and potency should
be carefully examined.
Tablet weight should bear a direct relationship to milligrams of active
ingredient available where the final blend is homogeneous. This conclusion assumes
that demixing does not occur as the compound is transferred to intermediate
storage containers or to a tablet press hopper [12]. To measure the likelihood
that controlling tablet weight assures dosage uniformity, 50 tablet assays selected
at random (from 300 tablet assays) were compared to tablet weight using
regression analysis. Because the same model tablet press and blender were employed
for every batch, assay results from all 30 batches were pooled. The mean
purity of the 25 receipts of active ingredients used to manufacture the 30 batches
in the validation study was 99.7%, or 0.3% below target. Individual lots ranged from 98.8–102%. Because of these lot-to-lot differences, active ingredient raw
material potency was also included in the regression analysis.
The general model from the regression analysis is [13]
y = bo + b1X1 + b2Y2
where
y = tablet potency
bo = constant
X1 = raw material purity
X2 = tablet weight
Tablet potency was found to be related to raw material purity and tablet
weight as follows:
y = −414.6 + 6.605OX1 + 0.4303X2
We would expect the regression plane to have a significant positive slope;
that is, as purity of the active ingredient and tablet weight increase, so will
tablet potency, and this was found to be the case. Both slopes are statistically
significantly different from 0 at α = 0.025. When the above equation is used to
predict tablet potency given the ideal tablet weight (600 mg) for the product
and mean raw material purity of 99.7%, the resulting value is only 2.1 mg
different from the theoretical value of 500 mg.
In conclusion, drug A production was shown to be within established
specifications, and there is no reason to believe this will not be the case for
future production as long as all practices are continued in their present form.
Furthermore, there is no significant difference between batches produced by the
tray dryer process and the fluid bed process. A validation report should memorialize
these findings. The report should also recommend eliminating the option
to use a no. 5 screen for the wet milling step and a no. 12 screen to pulverize
the dried granulation. There is no experience or only limited experience with
this equipment that supports its continued availability. In the same vein, the
final blend time should be standardized at 10 min and automatically controlled
by means of a timer.