1. Applications of Microbiolgical Data
Tim Sandle
Microbiology information resource:

2. Introduction Distribution of microbiological data Use of trend charts
Calculation of warning and action levels

3. Introduction Examples from environmental monitoring and water testing
Broad and illustrative overview
Written paper with more detail

4. Distribution of microbiological data
Why study distribution?
Impact on sampling
Impact on trending
Impact upon calculation of warning and action levels

5. Distribution
Most statistical methods are based on normal distribution, and yet….
Most microbiological data does NOT follow normal distribution

6. Distribution
Micro-organisms, such as those in a typical, free-flowing water system, follow Poisson distribution
For example…

7. Distribution
S1 S2 S3 S4 S5
Where S = sample
= micro-organism

8. Distribution
And microbial counts tend to be skewed (or positive or negative exponential distribution)
For example, a Water-for-Injection system…

9. Distribution

10. Distribution
So, what can we do about it? ?
Skewed question mark

11. Distribution
Well:
a) Use complex calculations and Poisson distribution tables, or
b) Attempt to transform then data
We’ll go for the second option

12. Distribution A general rule is:
For low count data e.g. Grade A monitoring and WFI systems, take the square root
For higher count data, e.g. Grade C and D environmental monitoring or a purified water system, convert the data into logarithms

13. Distribution
For example, some counts from a WFI system:

14. Distribution
When the data is examined for its distribution, using a simple ’blob’ chart:

15. Distribution
Whereas if the square root is taken:

16. Distribution
We move closer to normal distribution:

17. Distribution Logarithms work in a similar way for higher counts
Remember to add ‘+1’ to zero counts (and therefore, +1 to all counts)

18. Trend Analysis There is no right or wrong approach
There are competing systems
This presentation focuses on two approaches, both described as ‘control charts’:
The cumulative sum chart
The Shewhart chart

19. Trend Analysis Control charts form part of the quality system
They can be used to show:
Excessive variations in the data
How variations change with time
Variations that are ‘normally’ expected
Variations that are unexpected, i.e. something unusual has happened

20. Trend Analysis Control charts need:
A target value, e.g. last year’s average
Monitoring limits:
Upper limit
Lower limit
Control line / mean
So the data can be monitored over time and in relation to these limits

21. Trend Analysis Of these,
The warning limit is calculated to represent a 2.5% chance
The action level is calculated to represent a 0.1% chance
So, if set properly, most data should remain below these limits
These assumptions are based on NORMAL DISTRIBUTION
Various formula can be used to set these or validated software

22. Trend Analysis Cumulative sum chart (cusum) Shewhart chart
Suitable for large quantities of low count data. It is very sensitive to small shifts
Shows shifts in the process mean
Shewhart chart
Suitable for higher count data. It shows large changes more quickly.

23. Trend Analysis Cusums Harder to interpret
Displays the cumulative sum of a rolling average of three values and plots these in comparison with the target value
The direction and steepness of the slope are important
Significant changes are called ‘steps’
V-masks can be used as a prediction to the future direction

24. Trend Analysis For example, a Grade B cleanroom
Contact (RODAC) plates are examined
A target of 0.2 cfu has been used, based on data from the previous year

25. Trend Analysis

26. Trend Analysis Shewhart charts
Powerful for distinguishing between special causes and common causes
Common causes are inherent to the process and are long-term
Special causes are where something has changed and maybe of a long or short term

27. Trend Analysis Examples of special causes: a) A certain process
b) A certain outlet
c) A certain method of sanitisation, etc.
d) Sampling technique
e) Equipment malfunction e.g. pumps, UV lamps
f) Cross contamination in laboratory
g) Engineering work
h) Sanitisation frequencies

28. Trend Analysis For example, a Grade C cleanroom
Active air-samples are examined
A target of 1.5, based on historical data

29. Trend Analysis

30. Trend Analysis
The previous charts were prepared using a statistical software package
However, MS Excel can also be used
The next example is of a WFI system
Notice the data has been converted by taking the square root of each value

31. Trend Analysis

32. Trend Analysis Alternatives: Individual Value / Moving Range charts
Exponentially Weighted Moving Average charts (EWMA)
These are useful where counts are NOT expected, e.g. Grade A environments
They look at the frequency of intervals between counts

33. Trend Analysis
Summary

34. Limits Alert and action levels Based on PDA Tech. Report 13 (2001):
Alert level: a level, when exceeded, indicates that the process may have drifted from its normal operating condition. This does not necessarily warrant corrective action but should be noted by the user.
Action level: a level, when exceeded, indicates that the process has drifted from its normal operating range. This requires a documented investigation and corrective action.

35. Limits
Why use them?
• Assess any risk (which can be defined as low, medium or high)
• To propose any corrective action
• To propose any preventative action

36. Limits “Level” is preferable to “Limit”
Limits apply to specifications e.g. sterility test
Levels are used for environmental monitoring

37. Limits
Regulators set ‘guidance’ values e.g. EU GMP; USP <1116>; FDA (2004)
These apply to new facilities
User is expected to set their own based on historical data
Not to exceed the published values
Many references stating this
Views of MHRA and FDA

38. Limits Things to consider:
The length of time that the facility has been in use for
How often the user intends to use the limits for (i.e. when the user intends to re-assess or re-calculate the limits. Is this yearly? Two yearly? And so on).
Custom and practice in the user’s organisation (e.g. is there a preferred statistical technique?)
They be calculated from an historical analysis of data.
Uses a statistical technique.

39. Limits Historical data Aim for a minimum of 100 results
Ideally one year, to account for seasonal variations

40. Limits Statistical methods:
Percentile cut-off
Normal distribution
Exponential distribution
Non-parametric tolerance limits
Weibull distribution
Recommended by PDA Technical Report, No. 13

41. Limits
Assumptions:
a) The previous period was ‘normal’ and that future excursions above the limits are deviations from the norm
b) Outliers have been accounted for

42. Limits Percentile cut-off Good for low count data
May need to use frequency tables
May need to round up or down to nearest whole zero or five
Warning level = 90th or 95th
Action level = 95th or 99th

43. Limits Percentile cut-off Data is collected, sorted and ranked
90th percentile means that any future result that exceeds this is 90% higher than all of the results obtained over the previous year.
Refer to PharMIG News Number 3 (2000) for excellent examples.

44. Limits Normal distribution
Can only be used on data that is normally distributed!
Could transform data but inaccuracies can creep in
Most data will be one-tailed, therefore need to adjust 2nd and 3rd standard deviation
Warning level = the mean
Action level = the mean

45. Limits Negative exponential distribution
Suitable for higher count data
Warning level: 3.0 x mean
Action level: 4.6 x mean

46. Limits
For all, do a ‘sore thumb’ activity by comparing to a histogram of the data
Does it feel right?

47. Conclusion We have looked at: Distribution of microbiological data
Trending
Cusum charts
Shewhart charts
Setting warning and action levels
Percentile cut-off
Normal distribution approach
Negative exponential approach

48. Conclusion Key points:
Most micro-organisms and microbial counts do not follow normal distribution
Data can be transformed
Inspectors expect some trending and user defined monitoring levels
Don’t forget to be professional microbiologists – it isn’t all numbers!

49. Just a thought… This has been a broad over-view
If there is merit in a more ‘hands on’ training course, please indicate on your post-conference questionnaires.

50. Thank you
Any questions?