What should be the CpK Limit
in ASEAN Harmonization
Guideline?
by
Pramote Cholayudth
VALITECH
Tel 08-1932-2374
Email: [email protected]
July 2008 1
Topic Outline
Learning about Process Capability (Cp & CpK)
CpK, Six Sigma & Defect Rate Relationship
Understanding CpK Concept
Examples for Retrospective Process
Validation & CpK Computation
Summary
2
A Process Model
Variable Data
Distributions in Samples
Samp # 1
Process Samp # 2
Samp # 3
3
A Stable Process Model
A stable process is the one
with consistent centering and
variability of the process.
4
A Capable/Non-Capable Process Model
5
Control Chart Limits
μ+3σx UCL
μ Sampling Distribution
μ-3σx LCL
σ
σx =
n
6
Control Chart Limits
In control Out of control
Process Process center
is stable has shifted
μ+3σx UCL •
•
•
• •
μ
• •
• • •
•
μ-3σx LCL
7
Control Chart Limits
USL
μ+3σx UCL
Grand Average or Target
μ CL
99.73% of Sampling Distribution
μ-3σx LCL
LSL
USL & LSL = Upper & Lower Specification Limits for Average
UCL & LCL = Upper & Lower Control Limits for Average
CL = Center Line = Central Line
8
Process Capability
Process Capability means the ability of a
controlled and stable manufacturing process to
produce a product with defined quality
attributes
Process Capability Index (Cpk) is the index to
define the degree of process capability when
compared with specification limits
9
Process Capability
Process Capability Range is the process
range between UCL & LCL (μ+3σ & μ–3σ)
for a quality attribute that represents the
typical variability observed with a controlled
and stable manufacturing process
Process Capability is applicable to normally
distributed variable data of the process
10
Process Capability (Cp) Index (CpK)
Cp is used to evaluate the variation of process
CpK is used to evaluate the centering of the
process (limit ≥ 1.33 is ideal)
Together these two indices are used to decide
whether the process meets its specifications ),
)
USL − LSL ⎛ USL − μ μ − LSL ⎞
Cp = CpK = Min⎜ , ⎟
6σ ⎝ 3σ 3σ ⎠
11
12
13
In-Process Control Using Control Chart Limits
Process is on-going
160 mg/tab USL
156 mg/tab UCL
Process Target
150 mg/tab CL
Process Range
144 mg/tab LCL
140 mg/tab LSL
USL − LSL
Cp =
σ x = 2 UCL = 150 + 3 x 2 = 156 6SD
⎛ USL − x x − LSL ⎞
LCL = 150 − 3 x 2 = 144 CpK = Min⎜⎜ , ⎟
⎝ 3SD 3SD ⎟⎠ 14
In-Process Control Using Control Chart Limits
Process is on-going
⎛ USL − x x − LSL ⎞
CpK = Min⎜⎜ , ⎟
⎟
⎝ 3SD 3SD ⎠
USL − LSL
Cp =
6SD
140 − 60
= = 2.2
36
15
Process Capability Analysis Using Control Chart Limits
Process is completed (sample mean data)
160 mg/tab USL
156.4 mg/tab UCL
Grand Average
151.0 mg/tab CL
Process Capability Range
145.6 mg/tab LCL
140 mg/tab LSL
SD 7. 9 UCL = 151 + 3 x1.8 = 156.4
160 − 140
σ= = =8 Cp = = 1.9
c 4 0.9869 LCL = 151 − 3 x1.8 = 145.6 6x1.8
σ 8 ⎛ 160 − 151 151 − 140 ⎞
σx = = = 1.8 CpK = Min⎜ , ⎟ = Min(1.7,2.0) = 1.7
n 20 ⎝ 3x1.8 3x1.8 ⎠ 16
Process Capability Analysis Using Cp & CpK
Process is completed (individual data)
USL (Individual) = 175 mg
LSL (Individual) = 125 mg
Grand Average = 151 mg
Sigma = 8 mg
USL − LSL 175 − 125
Cp = = = 1.0
6σ 6x8
⎛ 175 − 151 151 − 125 ⎞
CpK = Min⎜ , ⎟ = Min(1.0,1.1) = 1.0
⎝ 3x8 3x8 ⎠ 17
Understanding CpK Concept: Robert A. Nash
Remember that what is possible in other industry
is not possible in pharmaceutical industry
The CpK can be tied directly into the defect rate
of a process – probability of finding a defect
CpK limit for pharmaceutical processes is 1.0 *
Recommendation: Cp or CpK ≥ 1.0 (Retro. Val.)
* In Pramote’s paper reviewed by Nash in 2002’s
18
CpK vs Defect Rate (Probability of Finding a
Defect): Robert A. Nash, PhD
Cpk
( 3.07−3.47e )
P=e
19
CpK vs Sigma Level
x −μ ⎛ USL − μ μ − LSL ⎞
Z= CpK = Min⎜ , ⎟
σ ⎝ 3σ 3σ ⎠
USL − μ CpK = Min(Z Min / 3, Z Max / 3) = Z Min / 3
Z=
σ Z Min = 3CpK
μ − LSL SigmaLevel = Z Minσ = 3CpKσ
Z=
σ
SigmaLevel Z Min
CpK = =
3σ 3 20
CpK vs Sigma Level
In Six Sigma program at Motorolla (in the past),
Assuming process shifted by 1.5σ
⎛ USL − μ − 1.5σ μ − LSL − 1.5σ ⎞
CpK = Min⎜ , ⎟
⎝ 3σ 3σ ⎠
⎛ USL − μ μ − LSL ⎞
CpK = Min⎜ − 0.5, − 0.5 ⎟
⎝ 3σ 3σ ⎠
CpK = Min(Cpu − 0.5,Cpl − 0.5)
21
CpK vs Sigma Level
In ‘process on target’ concept nowadays,
⎛ USL − μ μ − LSL ⎞
CpK = Min⎜ , ⎟
⎝ 3σ 3σ ⎠
CpK = Min(Cpu,Cpl)
22
Quality Attributes Distributed in-between the Limits
μ = Process Mean
σ = Process SD
68.26%
95.46%
LSL USL
99.73%
μ-xσ μ-3σ μ-2σ μ-1σ μ μ+1σ μ+2σ μ+3σ μ+yσ
23
Assay Values Distributed in-between the Limits
μ = Process Mean = 100.26 110 − 100.26
Z Min = = 3.8
σ = Process SD = 2.56 2.56
100.26 − 90 CpK =
Z Min 3.8
= = 1.27
Z Max = = 4.0 3 3
2.56
68.26%
95.46%
LSL=90 USL=110
99.73%
92.58 95.14 97.70 100.26 102.82 105.38 107.94
24
Assay Values Distributed in-between the Limits
110 − 101.03
μ = Process Mean = 101.03 Z Min = = 2.89
σ = Process SD = 3.10 3.1
Z Min 2.89
101.03 − 90 CpK = = = 0.96
Z Max = = 3.56 3 3
3.1
68.26%
95.46%
LSL=90 USL=110
99.73%
91.73 94.83 97.93 101.03 104.13 107.23 110.33
25
One-Sided & Two-Sided Defect Rate
The computed defect rate is based on one side
of the two-sided distribution curve
This particular rate will cover the two-sided
distribution since the actual density of the
defect units will be less than the theoretical rate
26
Sigma Level vs CpK vs Defect Rate
Excel Function =1-NORMSDIST(1.5) = 6.68%
σ level* σ level** CpK Defect rate (ppm) Defect rate (%)
1.5σ 3.0σ 0.50 66,807 6.68
2.0σ 3.5σ 0.67 22,750 2.27
2.5σ 4.0σ 0.83 6,210 0.62
3.0σ 4.5σ 1.00 1,350 0.13
3.5σ 5.0σ 1.17 233 0.02
4.0σ 5.5σ 1.33 32 0.00
4.5σ 6.0σ 1.50 3.4 0.00
5.0σ 6.5σ 1.67 0.29 0.00
5.5σ 7.0σ 1.83 0.02 0.00
6.0σ 7.5σ 2.00 0.00 0.00
* Process on target, ** Process shifted by 1.5σ, Sigma Level = 3*CpK*σ 27
Sigma Level vs CpK vs Defect Rate
Excel Function =1-NORMSDIST(3.0-1.5) = 6.68%
σ level* σ level** CpK Defect rate (ppm) Defect rate (%)
1.5σ 3.0σ 0.50 66,807 6.68
2.0σ 3.5σ 0.67 22,750 2.27
2.5σ 4.0σ 0.83 6,210 0.62
3.0σ 4.5σ 1.00 1,350 0.13
3.5σ 5.0σ 1.17 233 0.02
4.0σ 5.5σ 1.33 32 0.00
* Process on target, ** Process shifted by 1.5σ, Sigma Level = 3*CpK*σ
CpK = 4.0σ/3σ = 1.33 CpK = (5.5σ-1.5σ)/3σ = 1.33
28
In ASEAN Guideline
CpK = 1, 1.33, 2 → Sigma Level = 3, 4, 6
Therefore, assuming no (significant) process
shift
29
EXAMPLE: Sigma - Getting it Right First Time.
Quantifies process ability to generate defect-free output.
Allows comparison of any two processes.
Higher sigma values indicate better processes.
Should be the scientific basis for process transfer.
Sigma ppm Defects Yield Cost of Quality
2σ 308,537 69.2% 25-35%
Pharma 3σ 66,807 93.3% 20-25%
4σ 6,210 99.4% 12-18%
5σ 233 99.98% 4-8%
Semicon 6σ 3.4 99.99966% 1-3%
Process shifted by 1.5σ, CpK = (6σ-1.5σ)/3σ = 1.5, Excel =1-NORMSDIST(6-1.5) = 3.4 ppm
30
Minimum Regulatory “Sigma”
Level for Drugs?
Under cGMP when failures/recalls
exceeds 10% - no longer “validated.”
The minimum regulatory "Sigma” ~ 1.65?
CP SIGMA DEFECTS COST CLASS
0.67 2) 5% 25-35% Not Capable
Pharmaceuticals Competitive
1.0 3) 0.13% 20-25% Average
1.33 4) 60 ppm 12-18% Healthy
Semiconductor
1.66 5) 1 ppm 4-8% Superior
2.0 6) 2 ppb 1-3% World Class
FDA Science Board 11/16/01: PricewaterhouseCooper Presentation
(Modified by AH)
31
How this is a Win-Win
High
6
5 6σ - World Class
4 5σ - Superior
QUALITY 4σ - Healthy
3 3σ - Average
2 2σ - Not Capable
1 1σ - Not Competitive
Low
Low High
PRODUCTIVITY
32
CpK Distribution for Product A
33
CpK Distribution for Product A
34
CpK Distribution for Product B
35
CpK Distribution for Product B
36
Retrospective PV: Assay Data (Core)
Real Case
Lot # Assay Lot # Assay Lot # Assay
1 103.06 8 99.64 15 99.60
2 99.37 9 98.30 16 97.69
3 100.60 10 101.80 17 98.24
4 99.86 11 99.27 18 99.45
5 95.94 12 97.55 19 98.69
6 100.00 13 99.89 20 97.26
7 101.74 14 102.21 – –
Mean 99.51 SD 1.79 σ 1.81
UCL 104.94 CL 99.51 LCL 94.08 37
Retrospective Process Validation:
Assay Data & CpK (Core)
Core CpK = MIN(110-99.51,99.51-90)/3*1.81 = 1.75
38
Retrospective PV: Assay Data (Coated)
Real Case
Lot # Assay Lot # Assay Lot # Assay
1 98.55 8 102.73 15 98.95
2 97.00 9 101.10 16 96.13
3 99.14 10 101.09 17 99.45
4 102.48 11 99.86 18 101.53
5 100.44 12 99.76 19 98.88
6 100.90 13 103.24 20 99.13
7 100.07 14 101.14 – –
Mean 100.08 SD 1.80 σ 1.82
UCL 105.54 CL 100.08 LCL 94.62 39
Retrospective Process Validation:
Assay Data & CpK (Coated)
Coat CpK = MIN(110-100.08,100.08-90)/3*1.82 = 1.82
40
Retrospective PV – Dissolution Mean
(Pooled?) Data
Lot # DR Mean Lot # DR Mean Lot # DR Mean
1 96.92 8 102.25 15 100.59
2 98.98 9 98.82 16 96.31
3 94.79 10 102.77 17 99.42
4 98.43 11 103.10 18 98.75
5 98.86 12 97.70 19 101.24
6 98.42 13 100.24 20 94.39
7 101.66 14 101.53 – –
Mean 99.26 SD 2.47 σ 2.50
UCL 106.77 CL 99.26 LCL 91.75 41
Retrospective Process Validation:
Dissolution Mean (Pooled) Data & CpK
Cpk = (99.26-(75+5))/3*2.50 = 2.57
Cpk = (99.26-75)/3*2.50 = 3.23
42
Retrospective PV: CU Data
Real Case
Lot # CU SD Lot # CU SD Lot # CU SD
1 3.15 8 0.56 15 1.51
2 1.67 9 2.10 16 3.68
3 2.92 10 0.97 17 3.02
4 1.38 11 2.00 18 1.58
5 1.70 12 2.32 19 2.08
6 2.39 13 3.15 20 2.12
7 4.62 14 2.15 – –
Gr. Av. 100.58 Variance* 5.95 SD* 2.44
USL 115% LA LSL 85% LA * Pooled 43
Pooled (Weighted-Average) Sample SD to
Estimate Population SD
S12 (n1 − 1) + S 22 (n2 − 1) + .. + S k2 (nk − 1)
S pooled =
(n1 − 1) + (n2 − 1) + .. + (nk − 1)
2 2 2
S1 + S2 + .. + Sk 2
= = S =σ
k
Same
Sample Size
44
Process Capability Analysis Using Cp & CpK
CU USL (Individual) = 115 % LA
CU LSL (Individual) = 85 % LA
Grand Average = 100.58 % LA
Sigma (Pooled SD) = 2.44 % LA
USL − LSL 115 − 85
Cp = = = 2.05
6σ 6x2.44
⎛ 115 − 100.58 100.58 − 85 ⎞
CpK = Min⎜ , ⎟ = Min(1.97,2.13) = 1.97
⎝ 3x2.44 3x2.44 ⎠ 45
Summary
Use of Cp & CpK in GMP
Data Trend Analysis
Annual Product Review
Establishing Trending/Monitoring Limits
(microbial count – airborne, water, product)
Use of Cp & CpK Process Validation
Retrospective Approach
Cp & CpK Limits ≥ 1.0
46
SPC Factor Table – c4
n c4 n c4 n c4
4 0.9215 13 0.9794 22 0.9882
5 0.9401 14 0.9810 23 0.9887
6 0.9516 15 0.9823 24 0.9892
7 0.9594 16 0.9835 25 0.9896
8 0.9651 17 0.9845 26 0.9901
9 0.9693 18 0.9854 27 0.9904
10 0.9727 19 0.9862 28 0.9908
11 0.9754 20 0.9869 29 0.9911
12 0.9776 21 0.9876 30 0.9914 47
SPC Factor Table – c4
n c4 n c4 n c4
31 0.9917 40 0.9936 49 0.9948
32 0.9920 41 0.9938 50 0.9949
33 0.9922 42 0.9939 51 0.9950
34 0.9925 43 0.9941 52 0.9951
35 0.9927 44 0.9942 53 0.9952
36 0.9929 45 0.9943 54 0.9953
37 0.9931 46 0.9945 55 0.9954
38 0.9933 47 0.9946 56 0.9955
39 0.9934 48 0.9947 57 0.9955 48
