1.

0 Purpose
This procedure defines the methods for.
1.1 To identify statistical methods used for sampling, interpretation and analysis of
data in order to draw valid statistically based conclusions supporting business and
engineering decisions.
2.0 Scope
2.1 This procedure applies to all functional areas and provides reference and guidance
for using statistical methods, sampling plans for use during development and
manufacturing and supporting criteria with the required statistical justification.
Other approaches may be used with the appropriate statistical justification. Specific
applications include but are not limited to the Product Development Process,
Clinical studies, Design Verification and Validation, Performance Validation and
Process Improvement and Optimization Studies. Incoming Inspection, In-process
inspection, Final inspection or any other test where data driven product or process
evaluation and acceptance is the objective. In practice, the final choices made for
each application are to be documented, reviewed and approved per quality system
requirements. Lot inspection sampling procedures are addressed separately in
800403001, WI Sampling Plan.

3.0 Reference Documents

ANSI/ASQ Z1.4 2003 (rev 2013) Sampling Procedures and Techniques for
Inspection by Attributes
3.1 ANSI/ASQ Z1.9 2003 (rev 2013) Sampling Procedures and Tables for Inspection
by Variables for Percent Nonconforming
3.2 Design and Analysis of Experiments, Third Edition, Douglas C. Montgomery
3.3 Experimental Statistics, NBS Handbook 91
3.4 Guide to Acceptance Sampling, Dr. Wayne A. Taylor
3.5 Handbook of Statistical Methods in Manufacturing, 1991 Edition, Richard Barrett
Clements
3.6 Implementing Six Sigma, Smarter Solutions Using Statistical Methods, 1999
Edition, Forrest W. Breyfogle III
 3.7 Juran’s Quality Control Handbook, Fifth Edition, J.M. Juran
3.8 Practical Reliability Engineering, Third Edition Revised, Patrick D.T. O’Connor
3.9 Quality Planning and Analysis, Third Edition, J.M. Juran/Frank M. Gryna
3.10 SOP-STAT-10 Verification/Validation Sampling Plans for Percent Conforming,
Taylor Enterprises, Inc.
3.11 SOP-STAT-2 Statistical techniques for Process Validation, Taylor Enterprises, Inc.
3.12 Statistical Guidance for Clinical trials of Non Diagnostic medical Devices
(Reference: [Link])
3.13 Statistical Quality Control, Sixth Edition, Eugene L. Grant/Richard S. Leavenworth
3.14 Total Quality Control, Third Edition, A.V. Feigenbaum
3.15 Zero Acceptance Number Sampling Plans, Fourth Edition, Nicholas L. Squeglia

4.0 Definitions
4.1 Descriptive Statistics: The term given to the analysis of data that helps describe,
show or summarize data in a meaningful way such that, for example, patterns
might emerge from the data. Examples of such measures are mean, mode,
median, standard deviation, range.
4.2 Inferential Statistics: Modeling techniques that allow us to use sampling to draw
conclusions about the populations from which the samples are drawn, e.g.
capability studies, hypothesis testing.
4.3 Type I Error (α): Rejecting a hypothesis when it is true. Its probability is called the
level of significance or producer’s risk and denoted by α.
4.4 Type II Error (β): Accepting a hypothesis when it is false. Its probability is called
the consumer’s risk and denoted by β.
4.5 Power: Rejecting a hypothesis when it is false. Its probability is denoted by (1- β).
4.6 Normality or Normal Distribution: The normal distribution is a commonly occurring
distribution that appears as a symmetrical bell-shaped curve. Tables of the normal
distribution are commonly available. Numerous statistical procedures have been
developed assuming the data being analyzed follows the normal distribution.
Many of these procedures are robust to this assumption and work well even for
non-normal data. However, variables sampling plans based on the normal
distribution and related procedures (normal tolerance intervals and confidence
intervals for Ppk) are quite sensitive to departures from this assumption and require
verification that the data fits or is well approximated by the normal distribution.
4.7 Process Capability Indices: Process capability indices compare the performance of
an in-control (stable) process to the specification limits. Ppk and Cpk are the most
commonly used process capability indices. The indices are calculated by forming
the ratio of the spread between the process specifications (the specification
"width") to the spread of the process values, as measured by 6 process standard
deviation units (the process "width"). Ppk is based on the standard deviation
 across all subgroups whereas Cpk is based on the standard deviation within each
of the subgroups.
4.8 Operating Characteristic (OC) Curve: The protection provided by a sampling plan
is summarized by its operating characteristic curve or OC curve. The OC curve for
the single sampling plan for proportion conforming with sample size n=300 and
accept number a=0 is shown below:
 
OC Curve - n=300, a=0
Probability Accept 1

0.8

0.6

0.4

0.2

0
0 0.25 0.5 0.75 1
Percent Nonconforming

The bottom axis represents possible values for the quality level. The quality level
can be percent nonconforming units, nonconformities per quantity, average,
standard deviation, difference between two averages, ratio of two standard
deviations, etc. In this example it is in terms of percent nonconforming. The left axis
gives the corresponding probability that the sampling plan will accept or pass such
a lot. For example, a 0.5% nonconforming lot has a 20% chance of passing
(probability of 0.2).

4.9 Acceptable Quality Level (AQL): The AQL of a sampling plan is a level of quality
that is routinely passed by a sampling plan. Products or processes at or better
than the AQL are passed at least 95% of the time. Typically, the quality level can
be in terms of percent conforming or nonconformities per quantity. The AQL
describes the risk associated with failing a good product or process. However,
even more important is the Rejectable Quality Level (RQL) defined below. The
RQL describes the risk associated with passing a bad product or process.
Sampling plans should be selected considering both risks.

The AQL is determined from the OC curve. In the example below the quality level
on the bottom axis is in terms of percent nonconforming. The AQL is the quality
level or percent nonconforming that corresponds to a 95% chance of passing on the
left axis. As shown below, the AQL of the single sampling plan n=300 and
acceptance criterion, a=0 is 0.018% nonconforming.
 OC Curve - n=300, a=0
1

Probability Accept
0.8

0.6

0.4

0.2

0
0 0.25 0.5 0.75 1
Percent Nonconforming
AQL = 0.018%

4.10 Rejectable Quality Level (RQL): The RQL of a sampling plan is a level of quality
that is routinely failed by a sampling plan. Products or processes at or worse than
the RQL fail most of the time. The quality level can be percent nonconforming
units, nonconformities per quantity, average, standard deviation, difference
between two averages, ratio of two standard deviations, etc. Two RQLs are
commonly used, which are denoted RQL.10 and RQL.05. They are determined from
the OC curve.

In the example below, the quality level on the bottom axis is in terms of percent
nonconforming. RQL.10 is that quality level or percent nonconforming on the bottom
axis that corresponds to a 10% chance of passing. As shown below, the RQL.10 of
the single sampling plan n=300 and a=0 is 0.76% nonconforming.

RQL.05 is that quality level or percent nonconforming on the bottom axis that
corresponds to a 5% chance of passing. As shown above, the RQL.05 of the single
sampling plan n=300 and a=0 is 1% nonconforming.

OC Curve - n=300, a=0

1
Probability Accept

0.8

0.6

0.4

0.2

0
0 0.25 0.5 0.75 1
Percent Nonconforming
RQL.05 = 1%

RQL.10 = 0.76%
 Associated with the RQLs are confidence statements that can be made. Passing
results with 90% confidence that the quality level is better than the RQL.10.
Demonstrating that a specified quality level is met with 90% confidence requires the
use of a sampling plan whose RQL.10 is equal to the specified quality level. Passing
also results with 95% confidence that the quality level is better than the RQL.05.
Demonstrating that a specified quality level is met with 95% confidence requires the
use of a sampling plan whose RQL.05 is equal to the specified quality level. The RQL
is also referred to as the Lot Tolerance Percent Defective (LTPD).
4.11 Representative Sample: A representative sample of a lot means a sample that is
spread out across the lot to maximize the chance of finding a run or cluster of
nonconforming units. A random sample is the benchmark method but a stratified
or periodic sample that spreads the samples equally across the lot is generally
preferred. A representative sample relative to a validation study also requires that
the lots produced be manufactured under what the regulations call “anticipated
conditions” so these lots are representative of future production.  
4.12 Nonconformance: A nonconformance is a unit of production which fails acceptance
criteria.
4.13 Critical-to-Quality Characteristic (CTQ): CTQs are the key measurable
characteristics of a product or process whose specification limits and quality levels
must be met in order to satisfy the customer and stakeholder(s). They align
improvement or design efforts with customer and stakeholder requirements.
4.14 Confidence/Reliability Statement: This statement is of the form that with some
chosen % confidence (e.g. C = 95%), more than R% of units conform to
requirements. R is chosen based on risk assessment.
4.15 Statistical Process Control (SPC): A method of monitoring, controlling and, ideally,
improving a process through statistical analysis. Its four basic steps include
measuring the process, eliminating variances in the process to make it consistent,
monitoring the process, and improving the process to its best target value.
4.16 Common Cause or Random Variation: The common cause variation arises from a
multitude of small factors that invariably affect any process and will conform to a
normal distribution, or a distribution that is closely related to the normal distribution.
4.17 Assignable Cause: Also known as special cause variation, which arises from
specific factors that have an identifiable effect on the process.
4.18 Inspection: A conformity evaluation by observation and judgment accompanied as
appropriate by measurement, testing, or gauging.
4.19 Sub-System: Includes any type of System sub-elements including disposables, but
does not include sub-system components.

5.0 Responsibilities
5.1 Quality Engineering: Quality Engineering is responsible for establishing
appropriate statistical techniques that assure adequate levels of quality and
 confirm conformance to stated requirements. Appropriate selection and use of
statistical techniques is reviewed and assured by qualified individuals.
5.2 Each department with specified or implied responsibility is required to use
statistical methods which conform to this procedure.

6.0 Materials/Equipment Used

6.1 Calculator
6.2 Microsoft Excel (software validation is required)
6.3 MINITAB Statistical Software
6.4 Other statistical software tools as applicable for specific projects (software
validation is required)

7.0 Procedure for Statistical Techniques

7.1 Statistical Techniques
7.1.1 Statistical methods employed at Volcano Corporation are defined herein.
If additional or specific needs are identified where statistical techniques
are required for establishing, controlling and verifying process capability
and product characteristics, those techniques will be defined and
documented in the relevant documents.
7.1.2 The inspection program consists of three legs:
[Link] Verification and Validation Sampling Plans demonstrate that the
process consistently produces good product.
[Link] Manufacturing Sampling Plans are used to detect major shifts in
process performance quickly
[Link] Trending of Quality Data is used to detect more progressive
shifts in process performance and serves as a control
mechanism
7.2 Verification and Validation Activity Sampling Plans
7.2.1 Verification and Validation activity sampling plans when needed will be
selected based on Sections 7.7, 7.8, Appendix I and Appendix II. Appendix
I contains all Attribute and Variable Sampling tables for different
 Confidence/Reliability and AQL levels. Appendix II contains “k-factors” for
calculating tolerance limits for variable data when the AQL is unknown.
7.2.2 The diagrams on the following pages show graphically the structure for the
statistical methods that can be used.
7.3 Manufacturing Sampling Plans:
These are selected based on AQL levels which are predicated on process
performance at or better than the AQL for the desired risk level.
7.3.1 Sampling plans will be reviewed when:
[Link] Repetitive non-conformances are noted as specified in
QD000117150 ;
[Link] Internal audit reports warrant such action;
[Link] As a result of monthly Quality Reviews;
[Link] Feedback information and other appropriate considerations
warrant such action.
7.3.2 Sampling plans will be modified as part of the corrective action process,
when such action is warranted, and shall be justified.
7.3.3 Lot sampling procedures for variable and attribute sampling and
inspection of raw materials, assemblies, in-process testing, or finished
goods are addressed separately in 800403001, WI Sampling Procedure.
7.4 Trending of Quality Data/Process Control:
7.4.1 Trending may include complaint data, process yields or ppm,
environmental data, product returns, service events, etc.
7.4.2 Variation occurs in all processes. Common cause variation otherwise
referred to as random variation, is a natural part of the process. Another
type of variation, called special or assignable causes, comes from outside
the system and causes recognizable patterns, shifts, or trends in the data.
This is known as assignable factor variation. It is this variation that must
be analyzed for root cause and mitigated.
7.4.3 After process qualification and where appropriate, SPC charts are used to
monitor process performance and determine whether a process is
stable—per Statistical Process Control WI 102-0200.96. SPC Charts for
both variables data and attribute data should be used where appropriate
to monitor stability over time. Variable charts used most often, are Xbar
and R charts, Xbar and S charts, and Individual (I) and Moving Range
(MR) charts. Commonly used Attribute charts include fraction defective P
chart, number defective NP chart, number of defects C chart and defects
per unit U chart. Process stability must be established before data
analysis, e.g. calculating capability (Cpk, Ppk), is reliable.
SPC charts show if special causes are present and are influencing the
process. A process is in control when only common causes affect the
process output. The SPC charts perform tests for randomness that
provide information on the non-random variation due to trends such as
oscillation, mixtures, and clustering.
NOTE: It is not necessary for the data to be normal prior to the use of SPC
charting. The significance of using the chart in that situation is to understand
the variability and stability of the process.
Upward and downward trends, cycles, and large aberrations may be
spotted and investigated further. In an SPC chart, events shown on the y
axis are graphed against a time period on the x axis. Typically, if 15 or more
data points or more are available, control limits can be calculated and the
chart may be useful to detect special causes, i.e. something beyond
common causes and random process variability. Special cause indicators
may include the following:

[Link] Points outside the Control Limits: Any points that fall outside the
Upper Control Limit (UCL) or Lower Control Limit (LCL) indicate
that some assignable cause variation may be present. These data
points should be investigated for root cause and corrected.
[Link] Shifts: Eight or more consecutive points on one side of the
centerline indicate a special cause has influenced the process.
Points on the centerline are not included; they neither break the
string nor add to it.
[Link] Trends: Six consecutive jumps in the same direction indicate
that a special cause is acting on the process to cause a trend. Flat
line segments are not included either to break a trend or to count
towards it.
[Link] Patterns: A pattern that recurs eight or more times in a row is a
good indication to look for a special cause.

Capability of a process output is considered only after the process has been
established as “In Statistical Control”. That is to say all assignable or
special causes of variation have been identified and eliminated.
 .
7.5 Recommended Statistical Methods:
 Chart B – Statistical Analysis Process Flow

No

Determine GR&R or TMV Collect

Confidence Required? Data
Statement to
Select Identify CTQs Use Yes
or other Determine Yes
Statistical Response Sampling Plan
Application Parameters Complete
Establish Acceptable?
GR&R or TMV Check Goodness
Acceptance of Fit of Assumed
Criteria No Distribution

Improve
Measurement
Solution or
Use alternate
Measurement
Equipment

Determine Best Fit

Look for
Adequate Distribution or Root Cause Adequate
Consult with QE No No Assignable No
Fit? Select Best Data Identified? Fit?
Causes(
Transformation

Yes

Implement
Corrective
Actions and
Collect data
again
Yes

Analyze Data Yes

 Chart C – Data Analysis Methods

ANALYZE
DATA

For Exceptional
CAPABILITY
Situations consult
STUDIES
with QE

ATTRIBUTE VARIABLES
DATA DATA
ANALYSIS ANALYSIS

MEASUREMENT MEASUREMENT
HYPOTHESIS STATISTICAL HYPOTHESIS STATISTICAL
SYSTEMS DESIGN OF SYSTEMS DESIGN OF
TESTING FOR ANALYSIS PROCESS TESTING FOR ANALYSIS PROCESS
EXPERIMENTS EXPERIMENTS
ATTRIBUTE (GAGE R & R CONTROL VARIABLE (GAGE R & R CONTROL
(DOE) (DOE)
DATA STUDY) (SPC) DATA STUDY) (SPC)
 Chart D – Determine Sampling Method to be Used

Sampling
Methods

Lot Sampling Verification/

Conventional Validation Sampling
Inspection
Formulas (Sections 7.7 and 7.8)
(not in scope)

Based on For
Based on For
Confidence levels Systems
Inspection Disposable
and Precision and Sub-
Level, Lot Size Products
(Unacceptable Systems
and AQL
difference)

For Both
Variable and Based on Based on
Variable Attribute Used for
Attribute data, Confidence/ Confidence/
Sampling: Use Sampling: Use Equivalency
Tables Based on Reliability only Reliability
ANSI/ASQ ANSI/ASQ Testing, i.e.
Confidence/ (Variable (Attribute
Z1.9 Z1.4 Hypothesis Testing
Reliability and Sampling only) Sampling only)
AQL Levels

7.6 Conventional Sample Size Determination

7.6.1 Data analysis and sample size definitions are to be based on statistical
methods and documented, or referenced, within the qualification or
performance qualification plan.
7.6.2 Data collected shall have evidence that it was collected, how it was
collected, and who collected the data. Data shall be verified. If the data is
not recorded in the report, the report shall state where the data is located.

7.7 Sample Sizes for Development

This section shall be used as a model for when sampling considerations are applicable
within the scope of design assessments such as Design Verification or Validation. The
following flowcharts provide a process for rationalizing where statistical methods are
appropriate and where other analytical approaches may be used. When statistical
methods are applied, sampling decisions must be made.
A first step is to identify the level at which requirements will be verified.
 System/Sub-System Verification Planning

Verification Planning
for each Requirement:

Is This Sys/Sub-Sys Develop

Document
Requirement Dependent on Yes System
System Level Behavior? Level Plan

No

Develop
Document
Sub-System
Level Plan

Done

Once it is known at which level requirements will be verified, Sample Sizes and
Acceptance Criteria can be developed for Verification and Validation Test Plans as
shown in the chart following.
 Verification and Validation Sample Size Planning

System/Sub-System
Design Assessment

Is the requirement to conform to Have Critical to Quality (CTQ) Establish requirements

a Standard or a Volcano WI that Yes Requirements been identified and No and specification
prescribes the required sampling? Specification Tolerances established? tolerances

Yes

Document sample
size rationale in No No
Test Plan No

Done Can this Can this

requirement be verified through No No requirement be verified
Testing? through Analysis?

Can this requirement

be verified through 100% Inspection?
Choose another
Method
No
Yes
Choose another Yes Yes
Method

Choose from applicable

Develop Inspection
Inspection methods. More than one method may Analysis
method
be needed.
Document Analysis and
refer to it in the MVP/
Testing Done MVRpt
Validate Inspection
Method if required Is modeling or simulation
No
required for Testing?

Yes
Will 100%
Choose another
inspection be No
Method Is Test
ongoing? Is there a validated Performance Create Model/ Method
No
Model/Simulation for this requirement? Simulation Validation
Yes required?

Document it in the DMR

and refer to it in the
Yes
MVP/MVRpt.
Complete Gage
Yes No
R&R
Develop Acceptance Criteria
Done

Document Rationale
in Test Plan

Determine if repeated measurements can

Identify Confidence Determine sample size
be used to achieve the required sample Document sample size
and Reliability from Sampling Plans In
size, or if there is another basis for rationale in Test Plan
Requirement Appendix I or from Table C
adjusting the sample size (e.g., exposure
(Table 1) in Appendix II
is to business risk not safety risk)

Done for this requirement. Repeat for each requirement.

Development sampling plans may be executed on engineering prototype devices
incorporating small quantities using repeated measurements to achieve the desired
confidence/reliability. Development plans for critical requirements are usually followed
up with manufacturing sampling plans.
Determining a reasonable sample size for design verification and validation requires
additional considerations beyond statistical sampling. Cost is a major influencing factor.
However, patient risk considerations take precedence and require robust design and
engineering practices to alleviate the need for large samples.
It is recommended that the design and integration of systems use statistical tolerancing
(Six Sigma design tolerancing or Monte Carlo analysis are preferred over worst case
tolerancing for functional design disciplines). Simulation techniques may be used to
ensure reliable performance under anticipated adverse conditions. Power surge and
shock prevention features must be incorporated.
Adopting a strategy of testing at lower levels of the design which mitigate the need to
test extensively at the system level reduces the need for large sample sizes, with
appropriate justification provided. Both approaches require consideration of accuracy of
assumptions and design of the model/simulation. The resultant error of the
model/simulation must be less than the error produced using other methods such as
sound engineering judgment, expert experience, etc. Poor accuracy of assumptions and
design of model/simulation, such that they do not reflect reality, can lead to incorrect
conclusions.  
It is conceivable that the conclusions of a model/simulation may be 180 degrees out of
sync with expert opinions, in which case it is not certain whether the model or expert is
correct. In cases where models and expert opinion are both available, the reasoning of
the expert and the rigor of the model (the assumptions and design of the model) must
both be carefully evaluated, in order to reach the correct conclusions. It is important to
remember that while models and simulations are useful tools, and in some cases more
rigorous than expert opinion alone, the validity of the conclusions is based upon the
accuracy of the assumptions and design or the model/simulation.
If such design principles are used and considered robust through the design review
process, as few as 3 systems or sub-systems could be selected. Greater than three
systems are preferred as that would be more representative of the variation between
systems – not only within systems. In order to achieve a confidence/reliability sampling
plan, taking repeated measurements on each of the systems increases the sample
sizes to the desired levels called for in the sampling tables based on RQL
corresponding to the severity level of the characteristic being measured.

NOTE: FDA Design Control Guidance for Medical Device Manufacturers

TYPES OF VERIFICATION ACTIVITIES: Verification activities are conducted at all
stages and levels of device design. The basis of verification is a three-pronged approach
involving tests, inspections, and analyses. Any approach which establishes conformance
with a design input requirement is an acceptable means of verifying the design with
respect to that requirement. In many cases, a variety of approaches are possible.
7.8 Attribute and Variable Sampling Plans for Verification/Validation Activities:
This section is applicable to both Design and Process Verification/Validation
activities including first article inspections. Note: If a component or subassembly
lot received for first article inspection has fewer samples than the required sample
size, additional sampling must be conducted from subsequent lots until the
required sample size is reached. This section addresses the approach to be used
for selecting a sampling plan from the appropriate tables in Appendix I, for each
characteristic/requirement. They are used when Confidence/Reliability have been
determined and when historical non-conformance rates can be used to select the
appropriate AQL levels from the tables provided as explained in the sub sections
below.

7.8.1 Determine the Confidence/Reliability Statement

The confidence statement takes the form that with 90% or 95% confidence
more than R% of units conform to requirements where R is the reliability
level. R should be selected based on risk.
In most cases 95% confidence should be used. One exception is for
attribute data when multiple characteristics are being inspected for at once.
For example, the sampling plan may inspect for all nonconformities
determined to be in the major risk category. In this case 90% confidence is
allowed.
7.8.2 To demonstrate the conformance rate is above R% conforming, any of the
sampling plans in the applicable tables for that value of R can be used. All
the sampling plans in those tables allow the exact same confidence
statement to be made. From a customer and regulatory point of view, all
of these sampling plans are equivalent. They differ relative to the cost of
testing (number of units and measurement versus pass/fail) and the risk
associated with a good process failing the sampling plan. These are
business risks so the selection of the applicable sampling plan to use is a
business decision not requiring a documented justification.
To help ensure a good process passes, the AQL of the selected plan
should be at or above (i.e. ≥) the expected nonconformance rate. This
provides at least a 95% chance of passing. The expected
nonconformance rate might be determined using:
a. Historical data for the existing product or process
b. Data from similar products or processes
c. Data on competitors’ products
d. Published industry data such as industry average
If AQL levels are uncertain or varying, use sampling plans from the
lower end of the tables. The sampling plans at the top of the tables may
require lower sample sizes but they maximize the chance of a good process
failing, increasing the risk of the study. Note that in order to pass the
verification/validation sampling plans, the rate of conformance must be
 significantly better than the level it is being validated to, e.g. to demonstrate
a 1% non-conformance (R = 99%) the process must be running from 1/5 to
1/500 of this rate. If the process is not at least 5 times better, the process is
not ready to be validated but should be improved first.
7.8.3 Table I below specifies minimum values for R. More stringent values may
always be used. R refers to Reliability levels, which shall be determined
based on risk analysis. The rationale for the process reliability level shall
be documented. The reliability level may be based on performance of the
process using one, or more, of the suggested methods. These suggested
methods are: Hazard Analysis, Use, Design or Process FMEA historical
data, characterization study data and/or FDA sources.
If the product undergoes pre-conditioning per ASTM D4332-1, the
Reliability levels can be reduced as shown in Table 1. The rationale used
is that when a medical product is stressed, the reliability levels can be
dropped to the next lower level for purposes of selection of the appropriate
sampling plans to be used. Stress tests cause more failures than would
appear under normal usage conditions. This occurs because the stress
shifts the population to a stressed condition providing extreme results
outside that expected from normal usage. Therefore, the reliability
requirement for data from testing of units under stress can be lowered
while still obtaining a high confidence. The sampling plans referred to are
in Dr. Wayne Taylor’s publication “SOP-STAT-10 Verification/Validation
Sampling Plans for Percent Conforming, Taylor Enterprises”

Table I: Confidence/Reliability Criteria Based on Severity Levels

(Applicable to both Disposables and Systems)
This table specifies Confidence and Reliability levels to be used for different Failure
Severity levels. 1

Failure Severity Confidence Reliability Stress Test

Level Level Reliability Level
S-1 95% 90% 80%
S-2/S-3 95% 95% 90%
S-4/S-52 95% 99% 95%

Note 1: The Confidence/Reliability criteria in Table I are then used to select the appropriate table
for sample size selection from the Variable and Attribute tables provided in the Appendix I. These
tables provide equivalent optional sampling plans for each C/R level and RQL. Known non-
conformance rates must be used to determine AQL levels at or above the non-conformance rate,
since the AQL for a sampling plan is a level of quality that is passed by that sampling plan 95% of
the time. For attribute sampling plans used in manufacturing sampling, the tables provide additional
options of single and double sampling, the latter providing an additional step to accept the lot if the
first sample is rejected. Single sampling plans do not present that option and therefore start with a
higher sample size.
For design validation and process validation, AQL levels are chosen based on the Severity level of
the characteristic being evaluated. The AQL chosen will be lower if Severity level is higher.

Note 2: For S-4 and S-5 severity process outputs, attribute data should only be used when it is not
feasible to obtain variable data.

Additional Notes:
a. Where FDA requirements may differ from the above table; FDA requirements take
precedence, e.g., Balloon Fatigue (90% reliability/95% confidence) and Rated Burst Pressure
(99.9% reliability/95% confidence).
b. For process validations involving subassembly components being supplied by an OEM,
the defect severity must be determined/provided by the OEM customer. It is recognized that
external customers may not follow SOP, Risk Management Process 102-0100.03 and will
need to determine component failure severity or reliability and confidence requirements by
other means.

7.8.4 Determine the Applicable Sampling Plans

For each Conformance or Reliability Level, R, the following Tables are

provided in Appendix I:
a. Attribute sampling plans providing 90% confidence
b. Attribute sampling plans providing 95% confidence
c. Variables sampling plans for 1–sided specifications providing 95%
confidence
d. Variables sampling plans for 2–sided specifications providing 95%
confidence

8.0 Documentation
8.1 Retain records per SOP, Control of Quality Records, 102-0100.04.

9.0 Training
9.1 Review and complete the applicable training requirements per SOP,
Employee Training Program QD000117145 before performing this task.

10.0 Appendices
10.1 Appendix I: Contains Attribute and Variable Sampling Tables for different
Confidence/Reliability Levels
10.2 Appendix II: Contains Two-sided and One-Sided Tolerance Limit Factors
for a Normal distribution for different Confidence/Reliability Levels:
10.2.1 Table A: Two-sided Statistical Tolerance Interval Factors for a
Normal Distribution
10.2.2 Table B: One-sided Statistical Tolerance Bound Factors for a Normal
Distribution
.
 APPENDIX I

Attribute & Variable Sampling Plan Tables

Attribute Sampling Plans for 80% Conformance (RQL = 20%)

90 / 80 Attribute
RQL0.10 = 20%

Type Parameters AQL

Single n=11, a=0 0.47%
Double n1=11, a1=0, r1=2, n2=13, a2=1 1.79%
Single n=18, a=1 2.01%
Double n1=12, a1=0, r1=2, n2=16, a2=2 2.54%
Single n=25, a=2 3.35%
Double n1=12, a1=0, r1=3, n2=25, a2=3 3.93%
Single n=32, a=3 4.38%
Double n1=12, a1=0, r1=3, n2=32, a2=4 4.59%

95 / 80 Attribute
RQL0.05 = 20%

Type Parameters AQL

Single n=14, a=0 0.37%
Double n1=15, a1=0, r1=2, n2=10, a2=1 1.57%
Single n=22, a=1 1.64%
Double n1=15, a1=0, r1=2, n2=18, a2=2 2.07%
Single n=30, a=2 2.78%
Double n1=15, a1=0, r1=3, n2=27, a2=3 3.38%
Single n=37, a=3 3.78%
Double n1=15, a1=0, r1=3, n2=34, a2=4 3.98%

Variable Sampling Plans for 80% Conformance (RQL = 20%)

 95 / 80 Variables - 1-sided Specification
RQL0.05 = 20% corresponding to Ppk = 0.28

Parameters AQL
n=15, Ppk=0.50 1.7% corresponding to Ppk=0.70
n=20, Ppk=0.46 3.0% corresponding to Ppk=0.63
n=30, Ppk=0.42 4.9% corresponding to Ppk=0.55
n=40, Ppk=0.40 6.2% corresponding to Ppk=0.51
n=50, Ppk=0.39 7.1% corresponding to Ppk=0.49
n=60, Ppk=0.38 7.9% corresponding to Ppk=0.47
n=80, Ppk=0.36 9.5% corresponding to Ppk=0.44
n=100, Ppk=0.35 10.5% corresponding to Ppk=0.42

Double: 1st Stages: n1=15, Ppk-a=0.55, Ppk-r=0.29

n1=20, Ppk-a=0.50, Ppk-r=0.29
n1=30, Ppk-a=0.45, Ppk-r=0.29
nd
2 Stage: Any plan in above table (independent sample).

95 / 80 Variables - 2-sided Specification

RQL0.05 = 20% corresponding to Ppk = 0.28

Parameters AQL
n=15, Ppk=0.51, Pp=0.60 1.6% corresponding to Ppk=0.72
n=20, Ppk=0.48, Pp=0.58 2.5% corresponding to Ppk=0.65
n=30, Ppk=0.43, Pp=0.54 4.5% corresponding to Ppk=0.56
n=40, Ppk=0.41, Pp=0.53 5.8% corresponding to Ppk=0.52
n=50, Ppk=0.39, Pp=0.51 7.1% corresponding to Ppk=0.49
n=60, Ppk=0.38, Pp=0.51 7.9% corresponding to Ppk=0.47
n=80, Ppk=0.37, Pp=0.50 9.0% corresponding to Ppk=0.45
n=100, Ppk=0.36, Pp=0.49 9.9% corresponding to Ppk=0.43

Double: 1st Stages: n1=15, Ppk-a=0.57, Pp-a=0.65, Ppk-r=0.30, Pp-r=0.44

n1=20, Ppk-a=0.52, Pp-a=0.62, Ppk-r=0.30, Pp-r=0.44
n1=30, Ppk-a=0.47, Pp-a=0.58, Ppk-r=0.29, Pp-r=0.43
nd
2 Stage: Any plan in above table (independent sample).
Attribute Sampling Plans for 90% Conformance (RQL = 10%)

90 / 90 Attribute
RQL0.10 = 10%

Type Parameters AQL

Single n=22, a=0 0.23%
Double n1=24, a1=0, r1=2, n2=23, a2=1 0.88%
Single n=38, a=1 0.94%
Double n1=25, a1=0, r1=2, n2=34, a2=2 1.20%
Single n=52, a=2 1.59%
Double n1=25, a1=0, r1=3, n2=52, a2=3 1.87%
Single n=65, a=3 2.13%
Double n1=25, a1=0, r1=3, n2=66, a2=4 2.18%

95 / 90 Attribute
RQL0.05 = 10%

Type Parameters AQL

Single n=29, a=0 0.18%
Double n1=31, a1=0, r1=2, n2=23, a2=1 0.73%
Single n=46, a=1 0.78%
Double n1=31, a1=0, r1=2, n2=39, a2=2 0.98%
Single n=61, a=2 1.35%
Double n1=31, a1=0, r1=3, n2=58, a2=3 1.59%
Single n=76, a=3 1.82%
Double n1=31, a1=0, r1=3, n2=73, a2=4 1.86%

Variable Sampling Plans for 90% Conformance (RQL = 10%)

 95 / 90 Variables - 1-sided Specification
RQL0.05 = 10% corresponding to Ppk = 0.43

Parameters AQL
n=15, Ppk=0.70 0.22% corresponding to Ppk=0.95
n=20, Ppk=0.65 0.51% corresponding to Ppk=0.86
n=30, Ppk=0.60 1.1% corresponding to Ppk=0.76
n=40, Ppk=0.57 1.7% corresponding to Ppk=0.70
n=50, Ppk=0.55 2.3% corresponding to Ppk=0.67
n=60, Ppk=0.54 2.6% corresponding to Ppk=0.65
n=80, Ppk=0.52 3.4% corresponding to Ppk=0.61
n=100, Ppk=0.51 3.8% corresponding to Ppk=0.59

Double: 1st Stages: n1=15, Ppk-a=0.76, Ppk-r=0.44

n1=20, Ppk-a=0.70, Ppk-r=0.44
n1=30, Ppk-a=0.63, Ppk-r=0.44
nd
2 Stage: Any plan in above table (independent sample).

95 / 90 Variables - 2-sided Specification

RQL0.05 = 10% corresponding to Ppk = 0.43

Parameters AQL
n=15, Ppk=0.71, Pp=0.77 0.19% corresponding to Ppk=0.96
n=20, Ppk=0.66, Pp=0.73 0.46% corresponding to Ppk=0.87
n=30, Ppk=0.61, Pp=0.70 1.0% corresponding to Ppk=0.77
n=40, Ppk=0.58, Pp=0.67 1.6% corresponding to Ppk=0.72
n=50, Ppk=0.56, Pp=0.66 2.1% corresponding to Ppk=0.68
n=60, Ppk=0.55, Pp=0.65 2.4% corresponding to Ppk=0.66
n=80, Ppk=0.53, Pp=0.63 3.1% corresponding to Ppk=0.62
n=100, Ppk=0.52, Pp=0.63 3.6% corresponding to Ppk=0.60

Double: 1st Stages: n1=15, Ppk-a=0.77, Pp-a=0.82, Ppk-r=0.45, Pp-r=0.55

n1=20, Ppk-a=0.71, Pp-a=0.78, Ppk-r=0.45, Pp-r=0.56
n1=30, Ppk-a=0.65, Pp-a=0.73, Ppk-r=0.45, Pp-r=0.56
nd
2 Stage: Any plan in above table (independent sample).
 Attribute Sampling Plans for 95% Conformance (RQL = 5%)

95 / 95 Attribute
RQL0.05 = 5%

Type Parameters AQL

Single n=59, a=0 0.087%
Double n1=63, a1=0, r1=2, n2=50, a2=1 0.35%
Single n=93, a=1 0.38%
Double n1=64, a1=0, r1=2, n2=77, a2=2 0.48%
Single n=124, a=2 0.66%
Double n1=64, a1=0, r1=3, n2=116, a2=3 0.78%
Single n=153, a=3 0.90%
Double n1=65, a1=0, r1=3, n2=141, a2=4 0.92%

Variable Sampling Plans for 95% Conformance (RQL = 5%)

95 / 95 Variables - 1-sided Specification

RQL0.05 = 5% corresponding to Ppk = 0.55

Parameters AQL
n=15, Ppk=0.86 0.027% corresponding to Ppk=1.15
n=20, Ppk=0.80 0.091% corresponding to Ppk=1.04
n=30, Ppk=0.74 0.28% corresponding to Ppk=0.92
n=40, Ppk=0.71 0.47% corresponding to Ppk=0.87
n=50, Ppk=0.69 0.66% corresponding to Ppk=0.83
n=60, Ppk=0.68 0.80% corresponding to Ppk=0.80
n=80, Ppk=0.66 1.09% corresponding to Ppk=0.76
n=100, Ppk=0.65 1.29% corresponding to Ppk=0.74

Double: 1st Stages: n1=15, Ppk-a=0.94, Ppk-r=0.57

n1=20, Ppk-a=0.86, Ppk-r=0.56
n1=30, Ppk-a=0.79, Ppk-r=0.56
nd
2 Stage: Any plan in above table (independent sample).

95 / 95 Variables - 2-sided Specification

 RQL0.05 = 5% corresponding to Ppk = 0.55

Parameters AQL
n=15, Ppk=0.87, Pp=0.91 0.024% corresponding to Ppk=1.16
n=20, Ppk=0.82, Pp=0.87 0.071% corresponding to Ppk=1.06
n=30, Ppk=0.76, Pp=0.83 0.22% corresponding to Ppk=0.95
n=40, Ppk=0.73, Pp=0.81 0.39% corresponding to Ppk=0.89
n=50, Ppk=0.71, Pp=0.79 0.54% corresponding to Ppk=0.85
n=60, Ppk=0.69, Pp=0.77 0.73% corresponding to Ppk=0.81
n=80, Ppk=0.67, Pp=0.76 1.00% corresponding to Ppk=0.78
n=100, Ppk=0.65, Pp=0.74 1.29% corresponding to Ppk=0.74

Double: 1st Stages: n1=15, Ppk-a=0.94, Pp-a=0.97, Ppk-r=0.58, Pp-r=0.66

n1=20, Ppk-a=0.88, Pp-a=0.93, Ppk-r=0.57, Pp-r=0.66
n1=30, Ppk-a=0.80, Pp-a=0.87, Ppk-r=0.57, Pp-r=0.66
nd
2 Stage: Any plan in above table (independent sample).

Attribute Sampling Plans for 99% Conformance (RQL = 1%)

95 / 99 Attribute
RQL0.05 = 1%

Type Parameters AQL

Single n=299, a=0 0.017%
Double n1=320, a1=0, r1=2, n2=256, a2=1 0.069%
Single n=473, a=1 0.075%
Double n1=327, a1=0, r1=2, n2=385, a2=2 0.094%
Single n=628, a=2 0.13%
Double n1=327, a1=0, r1=3, n2=582, a2=3 0.15%
Single n=773, a=3 0.18%
Double n1=330, a1=0, r1=3, n2=719, a2=4 0.18%

Variable Sampling Plans for 99% Conformance (RQL = 1%)

95 / 99 Variables - 1-sided Specification

 RQL0.05 = 1% corresponding to Ppk = 0.78

Parameters AQL
n=15, Ppk=1.18 0.00014% corresponding to Ppk=1.56
n=20, Ppk=1.10 0.0012% corresponding to Ppk=1.41
n=30, Ppk=1.03 0.0071% corresponding to Ppk=1.27
n=40, Ppk=0.99 0.018% corresponding to Ppk=1.19
n=50, Ppk=0.96 0.033% corresponding to Ppk=1.13
n=60, Ppk=0.94 0.050% corresponding to Ppk=1.10
n=80, Ppk=0.92 0.079% corresponding to Ppk=1.05
n=100, Ppk=0.90 0.11% corresponding to Ppk=1.02

Double: 1st Stages: n1=15, Ppk-a=1.28, Ppk-r=0.80

n1=20, Ppk-a=1.18, Ppk-r=0.79
n1=30, Ppk-a=1.08, Ppk-r=0.79
nd
2 Stage: Any plan in above table (independent sample).

95 / 99 Variables - 2-sided Specification

RQL0.05 = 1% corresponding to Ppk = 0.78

Parameters AQL
n=15, Ppk=1.18, Pp=1.18 0.00013% corresponding to Ppk=1.57
n=20, Ppk=1.11, Pp=1.13 0.0010% corresponding to Ppk=1.42
n=30, Ppk=1.04, Pp=1.08 0.0061% corresponding to Ppk=1.28
n=40, Ppk=1.00, Pp=1.05 0.016% corresponding to Ppk=1.20
n=50, Ppk=0.97, Pp=1.03 0.029% corresponding to Ppk=1.15
n=60, Ppk=0.95, Pp=1.01 0.044% corresponding to Ppk=1.11
n=80, Ppk=0.93, Pp=1.00 0.070% corresponding to Ppk=1.06
n=100, Ppk=0. 91, Pp=0.98 0.10% corresponding to Ppk=1.03

Double: 1st Stages: n1=15, Ppk-a=1.28, Pp-a=1.28, Ppk-r=0.80, Pp-r=0.85

n1=20, Ppk-a=1.18, Pp-a=1.20, Ppk-r=0.80, Pp-r=0.86
n1=30, Ppk-a=1.09, Pp-a=1.13, Ppk-r=0.80, Pp-r=0.87
nd
2 Stage: Any plan in above table (independent sample).
 APPENDIX II

Table A: Two-sided Statistical Tolerance Interval Factors for a Normal Distribution

Factors k such that the confidence level is , that at least a proportion P (reliability level) of the Normal distribution,
will be less than X̅ + k*s and greater than X̅ – k*s for a two-sided limit, where X̅ and s are estimates of the mean
and standard deviation computed from a sample size n from a Normal distribution. Two-sided intervals cover X̅ ± k*s.

Note: To demonstrate a two-sided specification is met with specified confidence and reliability, then both X̅ + k*s ≤ USL and X̅ – k*s ≥ LSL

two-sided Normal Tolerance Interval Factors

Confidence ()/Reliability (P) [%] Confidence ()/Reliability (P) [%]
n n
90/90 95/90 95/95 95/99 90/90 95/90 95/95 95/99
2 15.987 32.019 37.674 48.43 50 1.916 1.996 2.379 3.126
3 5.847 8.38 9.916 12.861 55 1.901 1.976 2.354
4 4.166 5.369 6.37 8.299 60 1.887 1.958 2.333
5 3.494 4.275 5.079 6.634 65 1.875 1.943 2.315
6 3.131 3.712 4.414 5.775 70 1.865 1.929 2.299
7 2.902 3.369 4.007 5.248 75 1.856 1.917 2.285
8 2.743 3.136 3.732 4.891 80 1.848 1.907 2.72
9 2.626 2.967 3.532 4.631 85 1.841 1.897 2.261
10 2.535 2.839 3.379 4.433 90 1.834 1.889 2.251
11 2.463 2.737 3.259 4.277 95 1.828 1.881 2.241
12 2.404 2.655 3.162 4.15 100 1.822 1.874 2.233 2.934
13 2.355 2.587 3.081 4.044 110 1.813 1.861 2.218
14 2.314 2.529 3.012 3.955 120 1.804 1.85 2.205
15 2.278 2.48 2.954 3.878 130 1.797 1.841 2.194
16 2.246 2.437 2.903 3.812 140 1.791 1.833 2.184
17 2.219 2.4 2.858 3.754 150 1.785 1.825 2.175
18 2.194 2.366 2.819 3.702 160 1.78 1.819 2.167
19 2.172 2.337 2.784 3.656 170 1.775 1.813 2.16
20 2.152 2.31 2.752 3.615 180 1.771 1.808 2.154
21 2.135 2.286 2.723 3.577 190 1.767 1.803 2.148
22 2.118 2.264 2.697 3.543 200 1.764 1.798 2.143
23 2.103 2.244 2.673 3.512 250 1.75 1.78 2.121
24 2.089 2.225 2.651 3.483 300 1.74 1.767 2.106
25 2.077 2.208 2.631 3.457 400 1.726 1.749 2.084
26 2.065 2.193 2.612 3.432 500 1.717 1.737 2.07 2.721
27 2.054 2.178 2.595 3.409 600 1.71 1.729 2.06
30 2.025 2.14 2.549 3.35 700 1.705 1.722 2.052
35 1.988 2.09 2.49 3.272 800 1.701 1.717 2.046
40 1.959 2.052 2.445 3.213 900 1.697 1.712 2.04
45 1.935 2.021 2.408 3.165 1000 1.695 1.709 2.036 2.676
 1.645 1.645 1.96 2.576
Table B: One-sided Statistical Tolerance Bound Factors for a Normal Distribution

Factors k such that the confidence level is , that at least a proportion P (reliability level) of the Normal distribution
will be less than X̅ + k*s (or greater than X̅ – k*s) for a one-sided limit, where X̅ and s are estimates of the mean
and standard deviation computed from a sample of size n from a Normal distribution.

Note: To demonstrate that a one-sided specification is met with specified confidence and reliability, then either X̅ + k*s ≤ USL or X̅ – k*s ≥ LSL

One-sided Normal Tolerance Bound Factors

Confidence ()/Reliability (P) [%]
n
90/90 95/90 95/95 95/99 95/99.9
3 4.258 6.158 7.655 10.552 13.857
4 3.187 4.163 5.145 7.042 9.215
5 2.742 3.407 4.202 5.741 7.501
6 2.494 3.006 3.707 5.062 6.612
7 2.333 2.755 3.399 4.641 6.061
8 2.219 2.582 3.188 4.353 5.686
9 2.133 2.454 3.031 4.143 5.414
10 2.065 2.355 2.911 3.981 5.203
11 2.012 2.275 2.815 3.852 5.036
12 1.966 2.21 2.736 3.747 4.9
13 1.928 2.155 2.67 3.659 4.787
14 1.895 2.108 2.614 3.585 4.69
15 1.866 2.068 2.566 3.52 4.607
16 1.842 2.032 2.523 3.463 4.534
17 1.82 2.001 2.486 3.415 4.471
18 1.8 1.974 2.453 3.37 4.415
19 1.781 1.949 2.423 3.331 4.364
20 1.765 1.926 2.396 3.295 4.319
21 1.75 1.905 2.371 3.262 4.276
22 1.736 1.887 2.35 3.233 4.238
23 1.724 1.869 2.329 3.206 4.204
24 1.712 1.853 2.309 3.181 4.171
25 1.702 1.838 2.292 3.158 4.143
30 1.657 1.778 2.22 3.064 4.022
35 1.623 1.732 2.166 2.994 3.934
40 1.598 1.697 2.126 2.941 3.866
45 1.577 1.669 2.092 2.897 3.811
50 1.56 1.646 2.065 2.863 3.766