11 January 2002
Source: http://www.access.gpo.gov/su_docs/aces/fr-cont.html

-------------------------------------------------------------------------

[Federal Register: January 11, 2002 (Volume 67, Number 8)]
[Notices]
[Page 1482-1488]
From the Federal Register Online via GPO Access [wais.access.gpo.gov]
[DOCID:fr11ja02-66]

=======================================================================
-----------------------------------------------------------------------

DEPARTMENT OF HEALTH AND HUMAN SERVICES

Food and Drug Administration

[Docket No. 97D-0282]


Medical Devices: General Principles of Software Validation; Final
Guidance for Industry and FDA Staff; Availability

AGENCY: Food and Drug Administration, HHS.

ACTION: Notice.

-----------------------------------------------------------------------

SUMMARY: The Food and Drug Administration (FDA) is announcing the
availability of the guidance entitled ``General Principles of Software
Validation.'' This document provides guidance to medical device
manufacturers and FDA staff concerning requirements for validating
software used within medical devices, in device production, or in
implementing the manufacturer's quality system.

DATES: Submit written or electronic comments at any time.

ADDRESSES: Submit written requests for single copies on a 3.5''
diskette of the guidance document entitled ``General Principles of
Software Validation'' to the Division of Small Manufacturers,
International and Consumer Assistance (HFZ-220), Center for Devices and
Radiological Health (CDRH), Food and Drug Administration, 1350 Piccard
Dr., Rockville, MD 20850. Send two self-addressed adhesive labels to
assist that office in processing your request, or fax your request to
301-443-8818. See the SUPPLEMENTARY INFORMATION section for information
on electronic access to the guidance.
    Submit written comments concerning this guidance to the Dockets
Management Branch (HFA-305), Food and Drug Administration, 5630 Fishers
Lane, rm. 1061, Rockville, MD 20852. Submit electronic comments to
http://www.fda.gov/dockets/ecomments. Comments are to be identified
with the docket number found in brackets in the heading of this
document.

FOR FURTHER INFORMATION CONTACT: John F. Murray, Center for Devices and
Radiological Health (HFZ-340), Food and Drug Administration, 9200
Corporate Blvd., Rockville, MD 20850, 301-594-4659.

SUPPLEMENTARY INFORMATION:

I. Background

    This final guidance document entitled ``General Principles of
Software Validation'' provides guidance to medical device manufacturers
and FDA staff concerning requirements for validating software used
within medical devices, in device production, or in implementing the
manufacturer's quality system. It replaces the draft guidance that FDA
issued for comment on June 9, 1997, and published in the Federal
Register of July 25, 1997 (62 FR 40099).
    We received responses from 36 organizations and individuals, with
more than 650 questions, comments, and specific recommendations for
changes to the guidance. However, further work on the guidance was
interrupted by other high priority activities, including implementation
of the Food and Drug Administration Modernization Act of 1997, FDA's
response to year 2000 software concerns, and two rounds of
implementation of our first medical device performance standard.
Because of the delay in issuing this final guidance, we have chosen to
summarize our response to the comments received. As with any guidance,
we will continue to accept comments and may update this document in the
future.
    The following summarizes the comments we received, and significant
changes we made to the guidance in response to those comments:

A. Intended Scope

    From a few of the comments received, it appears that some parties
may not have realized the full breadth of the quality system
regulation. The software validation requirement in 21 CFR 820.70(i) of
the quality system regulation also applies to automated tools used to
design medical devices and tools used to develop software. Since the
first medical device good manufacturing practice regulation was
published in 1978, there has always been an explicit validation
requirement for software used in device production or used to implement
the quality system. When design controls were introduced into the
quality system regulation in 1997, that software validation requirement
was extended to software used to design devices, such as computer-aided
design and software development tools. FDA clearly addressed this issue
at the end of its response to comment 136 in the preamble to the
quality system regulation (61 FR 52602 at 52630, October 7, 1996). A
copy of the text is included at the end of this section.
    Some comments objected to the discussion of validation activities
during the predesign ``concept'' phase of software development, both
because the quality system regulation does not apply to research
activities, and because there is too little information available at
that point to make any validation related activity worthwhile. In
response to these concerns, we have removed all reference to validation
activities during the ``concept'' phase.
    Other comments noted that the guidance covered more than just
validation issues, and suggested changing the title to broaden the
scope of the guidance. We acknowledge that the scope of the guidance is
somewhat broader than the scope of validation in the strictest
definition of that term. However, we have chosen not to change the
title of the guidance. Planning,

[[Page 1483]]

verification, testing, traceability, configuration management, and many
other activities discussed in the guidance are important activities
that together help to support a final conclusion that software is
validated.
    Some comments expressed concerns that the guidance might be applied
too rigorously by FDA investigators, and some pharmaceutical
manufacturers raised questions about how the guidance would be applied
to their drug manufacturing operations. The agency's good guidance
practices (GGPs) clearly state the role of FDA guidance. Alternative
approaches that accomplish full compliance with the quality system
regulation are acceptable. While it is clearly intended for medical
device manufacturers, the guidance may also be useful to the
pharmaceutical industry and other industries regulated by FDA.
    Many comments suggested that we move all discussions regarding use
of off-the-shelf (OTS) software to the agency's guidance entitled
``Off-the-Shelf Software Use in Medical Devices.'' In response to these
comments, specific cross references to that document have been added
within the text of this guidance. However, the OTS guidance document
deals specifically with premarket submissions for OTS software
contained in medical devices. It is not the appropriate guidance for
OTS software used in manufacturing and quality systems applications.

B. Flexibility

    Numerous comments cited overly restrictive language and lack of
sufficient implementation flexibility in the draft guidance. For
example, many comments noted that the guidance implies use of a
``waterfall'' as the preferred life cycle development methodology.
Several comments suggested that more discussion was needed regarding
``rapid application development'' and ``component-based
methodologies,'' as well as ``build a little/test a little'' as an
acceptable methodology. Other comments asked for specific examples of
available life cycle models that could be used. In response to these
comments, and in accordance with our own GGPs, we have carefully
rewritten the text to remove any direct or implied use of the words
``shall'' or ``must,'' except where we describe or reference a
regulation. We also have added language to specifically state that
incremental development methodologies may be used, and that activities
and tasks can be performed in a different order, if called for by the
chosen life cycle model. However, for ease of description, we have
retained an organization of activities based on ``requirements,''
``design,'' ``coding (or construction),'' and ``testing.'' Regardless
of the order in which tasks are accomplished, these four categories of
activities are common to most life cycle models. We have not included
examples of the dozens of life cycle models that are available. To do
so could imply agency endorsement of certain life cycle models that are
included over those models that are not included. Instead, you are
referred to many of the textbooks and other references listed at the
end of the guidance, which provide details of many of these life cycle
models.
    One group of comments objected to any use of the word ``all'' when
describing items to be included in specification documents, noting that
``all'' is not a quantifiable term. Other comments suggested use of the
word ``may'' rather than ``should.'' On the other hand, a few comments
asked for a specific compliance matrix, so that manufacturers would
know exactly how to comply with FDA expectations. We have not adopted
these suggested changes. We believe that agency guidance should
identify and encourage use of approaches known to have been used
effectively, while the manufacturer retains the prerogative to choose
alternative approaches that are equally effective. Based on variables
such as firm size and structure, device risk, project size, and
complexity, manufacturers have the flexibility to choose different
approaches for different projects, and to select effective approaches
that best fit their specific needs.

C. Format

    Several comments suggested use of the framework and format in
international guidelines such as ISO 9000-3, GAMP, IEEE Software
Standards and ISO/IEC 12207. We have drawn information from each of
these sources and many other listed references, but unfortunately,
there is no single format available. We have rewritten the guidance to
address specific suggestions for wording changes and simpler language.
Some comments asked for extensive use of charts, analogies, and
examples for the concepts presented in the document. While valuable,
such an approach could easily triple the size of the guidance. Instead,
we suggest referring to any of the extensive list of references
included at the end of the guidance for more details on specific
implementation approaches.

D. Differences Between Hardware and Software

    Regarding the discussion of differences between hardware and
software, the comments were somewhat divided. Some comments applauded
the agency for recognizing the legitimate differences between hardware
engineering and software engineering. Other comments argued that
``software is not different'' and suggested deletion of all or most of
this section, either because it was unnecessary, or because it could be
misinterpreted by software developers who lack sufficient engineering
discipline. One comment suggested emphasizing the similarities of the
engineering discipline needed to build both hardware and software. We
have chosen to keep this section because we believe it explains part of
the rationale for why software must be thoroughly validated, and why
the software development process needs to be carefully controlled and
managed. We have also added additional information regarding the impact
of mobility of software professionals on the long-term maintenance of
software and the need for thorough documentation.
    Some comments objected to the discussion of standardization and
reuse of software components and asked for more recognition of the
trend toward increased use of OTS and component-based development
methods. Other comments objected to the statement that ``repairs made
to correct software defects establish a new design.'' We have revised
the text to address both of these concerns.

E. Principles of Software Validation

    We reorganized and rewrote the section regarding ``Principles of
Software Validation'' to address the comments received. For example, we
moved the subsection dealing with documenting software ``Requirements''
to the front of the section to reflect the importance of requirements
in the validation process. We clarified language regarding
``predetermined'' requirements to allow for incremental or evolutionary
development of requirements during the development project. However, we
have retained the concept that documented requirements should be
established prior to formal testing or other verification activities to
provide ``objective'' evidence that those requirements were met.
    The subsection previously entitled ``Testing'' is retitled ``Defect
Prevention'' and is revised to emphasize the importance of preventing
software

[[Page 1484]]

defects, as opposed to trying to ``test quality into'' software.
    We have renamed the subsection on ``Timing.'' In response to
several comments concerning validation continuing ``for the entire life
cycle,'' we have rewritten the text, but have retained the concept. At
each stage of the software life cycle, there is information available
that can contribute to a conclusion that the software meets user needs
and intended uses. Therefore, the validation process does not end when
the device is shipped.
    We replaced the subsection on ``Management'' with a new subsection
dealing with the ``Software Life Cycle.''
    We have clarified the subsections dealing with ``Plans'' and
``Procedures'' to distinguish between plans that define what to do, and
procedures that describe how to do it.
    The subsection entitled ``Partial Validation'' is substantially
rewritten and retitled ``Software Validation After a Change.'' Many
readers misinterpreted the statement that ``software cannot be
partially validated'' and thought we intended all validation testing to
be repeated every time any change is made. That is not what we meant.
Based on the comments received, we have rewritten the discussion to
emphasize the need for regression analysis after a change, followed by
an appropriate level of regression testing to reestablish the
validation status of the software. We have deleted specific discussion
of retrospective validation and reverse engineering of nonvalidated
software, but these issues should be covered during the regression
analysis.
    We have retitled and rewritten the subsection on ``Amount of
Effort.'' Now titled ``Validation Coverage,'' it still describes an
approach that ties the level of validation and verification effort to
the safety risk and complexity of the software.
    We revised the subsection on ``Independence of Review'' to provide
greater flexibility and a better explanation of its intent.
    The subsection previously entitled ``Real World'' is now entitled
``Flexibility and Responsibility,'' and reemphasizes that device
manufacturers/software developers have a lot of flexibility in how they
implement their software validation process, but the device
manufacturer is ultimately responsible for the adequacy and
effectiveness of the selected approach.

F. Terminology

    Some of the most significant comments we received had to do with
our basic definition of software validation. In the previous draft
guidance, we relied upon technical definitions used by the National
Institute of Standards and Technology and by the Institute of
Electrical and Electronic Engineers. These technical definitions
created some confusion with other definitions in our quality system
regulation. Numerous comments objected to our use of ``validation'' as
an umbrella term to cover ``design review'' and ``verification'' as
well as validation. They stated that both design review and
verification are distinctly separable quality concepts and are not a
part of validation. In response to these concerns, we have changed the
definition of software validation to be more consistent with the
quality system regulation and other international quality standards.
Our revised definition of software validation is derived directly from
the definitions of ``validation'' and ``design validation'' in the
quality system regulation.
    Comments also objected to the title ``Typical Validation Tasks'' at
the end of each subsection in the section V of the guidance and
suggested that they are really verification tasks. Other comments
objected to possible interpretation of these as mandatory tasks. In
response to these comments, we have also added text to explain that
there are typical verification and testing tasks that support an
overall conclusion that software is validated. Thereafter, when we
discuss ``Typical Tasks Supporting Validation,'' we do not try to
differentiate between verification tasks versus validation tasks.
Instead, we have revised the text to list ``Typical Tasks.'' While we
want to avoid any inference that the tasks are mandatory in every case,
the guidance makes the point that these are ``typical'' approaches that
are recommended by software engineering standards and textbooks, and
widely used by many software engineering professionals.
    Several comments noted inconsistencies in terminology from that
contained in the quality system regulation, in two software guidances
issued by the Office of Device Evaluation, and in the FDA glossary of
computerized system and software development terminology. These
comments also suggested use of the term ``risk analysis'' instead of
``hazard analysis'' throughout the software validation guidance. We
have revised the guidance to incorporate the term ``risk analysis''
throughout. However, we continue to emphasize that while there are many
different risks (e.g., economic or time to market), FDA is concerned
about safety risk (hazard). At their next revision, we expect to update
other software guidance documents and the FDA glossary with consistent
definitions of validation, verification, and risk analysis. In
addition, we now use the term ``user site testing'' rather than
``installation testing'' to describe testing performed at the user site
and outside the control of the software manufacturer.
    Some comments questioned whether OTS software could be validated
because the device manufacturer frequently does not have access to the
source code. These comments suggested that OTS software should be
``qualified'' rather than ``validated.'' However, we believe that the
evidence developed by a device manufacturer concerning OTS software is
a true validation because it directly supports a conclusion that the
software meets user needs and intended uses. Where the source code is
not available, it is incumbent upon the device manufacturer to use
other means (such as audits, or more extensive black box testing) to
infer the structural integrity of the OTS software. This issue is
clearly addressed in comment 136 of the preamble to the quality system
regulation (61 FR 52602 at 52630).
    Other comments from the pharmaceutical industry suggested
incorporation of widely understood process validation terminology
(i.e., installation qualification (IQ), operational qualification (OQ),
and performance qualification (PQ)) to describe software validation.
Another comment suggested use of ``product performance qualification''
rather than ``design validation.'' We have added a section that refers
to the various types of qualification, but we have chosen not to adopt
``qualification'' terminology in explaining software validation
requirements. Of course, manufacturers may continue to organize their
validation efforts using IQ/OQ/PQ terminology, if they wish.
    In response to comments, a new subsection has been added to explain
the differences between ``requirements,'' which may be general in
nature, versus ``specifications,'' which are developed to an
engineering level of detail.
    Several comments objected to use of undefined terms such as
``microcode'' and ``assertions.'' We reiterate that these and many
other terms used throughout the guidance are specifically defined in
the FDA glossary of computerized system and software development
terminology, which is available at http://www.fda.gov/ora/inspect_ref/
igs/gloss.html.

[[Page 1485]]

G. Design Review

    As noted above, design reviews are not a part of validation. In
fact, several comments noted that results of verification and
validation are inputs to design reviews--not the other way around. To
emphasize this point, we moved the subsection on ``Design Reviews''
outside the section on ``Typical Tasks Supporting Validation.'' We also
added information about the difference between formal design reviews
that are mandated by the quality system regulation versus less formal
technical reviews.

H. Traceability

    A few comments objected to the guidance regarding ``traceability
analysis,'' especially the discussion at the end of the subsection on
``Coding.'' Two comments noted that for very complex programs with
thousands of lines of code or thousands of modules, the traceability
analysis would be extremely complex and of little value. One suggested
that design review was an adequate substitute for traceability
analysis. We disagree. Traceability is an essential aspect of
verification, and it is an important input into design reviews. We
therefore do not believe that design review could be an adequate
substitute for traceability analysis.
    One comment stated that requirements are not always neatly
structured, and it is very difficult to trace exactly how they are
implemented in the design. There are numerous many-to-one and one-to-
many relationships to be mapped from requirements to design to code. We
agree with this observation; however, it actually further supports the
need for traceability. The larger and more complex the project, the
more important the traceability analysis becomes. Therefore, we have
retained the discussions regarding traceability, and in response to
several other comments, we have added traceability of software
requirements to the safety risk analysis.
    Another comment noted that inherent traceability can be built into
documentation and code without having to have a separate traceability
document. We agree and for that reason have avoided use of the most
commonly used term--``traceability matrix.'' Three common approaches
are traceability matrix, using computer databases to evaluate
traceability, or building inherent traceability into the structure of
the documentation and code. There may be many other approaches to
traceability. Software developers have flexibility in how they want to
implement traceability.

I. Risk Analysis

    Many comments questioned the concept of a software failure modes
and effects analysis (FMEA). They stated that given the difficulty of
predicting specific software failure modes, FMEA is better used as a
system level risk analysis tool. We have revised the guidance to
discuss software risk analysis within the context of system safety.
However, while we acknowledge some limitations in its use, we also
believe that software FMEA can be a useful tool, especially for safety
critical aspects of software applications. It may also be useful early
in the development process for analyzing safety critical software
requirements.
    One comment objected to the suggestion that risk analysis begin at
the stage where requirements are defined. However, to be useful and
have an impact on the software development process, we believe that
risk analysis needs to begin early and needs to be updated as the
project progresses. In addition, we have revised various portions of
the guidance to emphasize that the level of safety risk is a major
factor in determining the level of effort to be applied in testing and
other verification and validation tasks.

J. Planning

    In response to comments, we have changed the subsection on
``Management'' to be entitled ``Quality Planning.'' It now provides a
more general discussion of the software validation and verification
concerns to consider during quality planning.
    Several comments questioned the idea of early test planning, which
was recommended in the draft guidance. For example, they argued that
there is insufficient information available during requirements
development to be able to develop a system test plan or an acceptance
test plan. We disagree and have retained the recommendations for early
test planning, but we have specified that test plans and test cases
should be created as early in the software development process ``as
feasible.'' One of the important criteria, both for requirements and
for design, is that they be testable. The fact that there is
insufficient information for a particular test plan is valuable
feedback to the development process that perhaps the requirements or
design processes are not yet sufficiently complete. Planning is a
dynamic activity that should be reexamined and updated as the project
progresses.

K. Requirements

    Many comments objected to use of the word ``all'' in describing
what is typically specified in software requirements. We agree that
requirements frequently do not specify ``all'' that they should.
However, that is widely recognized as one the major flaws in software
development, and its correction is one of the most important messages
intended by this guidance. In order to be complete, a software
requirements specification should cover all the pertinent issues--not
just a selected few.
    One comment noted that requirements may not always be measurable.
We have changed the text to state that requirements should be
``measurable or objectively verifiable.''
    A few comments noted that ``internal interfaces'' and ``all ranges
of values the software will accept'' are a part of design--not
requirements. We agree regarding internal interfaces and have changed
the text accordingly. However, since software requirements are derived
from system requirements, there may be some internal system interfaces
prescribed from the high level system design that would impact software
requirements. Regarding ``ranges of values,'' we note that there is
rarely a bright line of demarcation between requirements and design.
Software developers have flexibility as to where in their life cycle
they wish to cover particular issues. We rejected most comments
requesting even greater levels of detail and specificity regarding
static verification techniques. For example, several comments asked for
more detail regarding ``requirements evaluation'' and ``interface
analysis.'' Details on these techniques are available in many of the
references listed at the end of the guidance. FDA investigators will
expect to see a verification procedure that includes a means for
identifying and resolving incomplete, ambiguous, and conflicting
requirements, as required by the regulation. They will also expect to
see objective documented evidence that the verification procedure was
implemented.

L. Design

    We have retained wording about the need for design specifications
to be complete enough for programmers not to have to make ad hoc
decisions. The intent is to ensure that the code created is consistent
with the design specification. When programmers or engineers decide to
add new functionality not identified previously in the requirements or
design, those specifications need to be updated to reflect the actual
code created. The project manager, design team, and any future
maintainers of the software need

[[Page 1486]]

to have accurate documentation in order to do their work.
    We have dropped the listing of specific approaches to software
design, and we have included a more general description of what should
be included in a software design specification. Some comments
considered the previous list to be too prescriptive as well as
incomplete.
    We recognize that portions of the software are completed and
released incrementally, and life cycle processes are repeated
iteratively. The intent is that those portions of the software have
design documentation that is consistent with the software application
that is implemented. One comment noted that in a rapid application
development (RAD) environment, there is typically no formal design
document in place during coding. We recognize that RAD is valuable as a
prototyping tool, but its use does not preclude the need to document
the specific design, once it is agreed upon.

M. Coding

    We have changed the title of this subsection to reflect that the
creation of a software application can be either through coding, or
through combining existing software components, such as OTS software
products or functional components from existing code libraries.
    Comments objected to the idea of having to keep results of all
compilations of the code. In response, we have revised the discussion
of compiler error checking to state that the results of the ``final''
compilation of the code should be retained to document any errors that
remain uncorrected in the final software product.

N. Testing by the Software Developer

    We renamed and revised this subsection to provide a better
explanation of the purpose of testing, and to avoid prescriptive
language concerning use of specific testing techniques. We have added
language regarding use of incremental development and testing
methodologies. We expanded the discussion of testing coverage to
explain how different degrees of coverage should be considered for
varying levels of risk, and that the manufacturer has flexibility to
choose the right level of coverage.
    One comment noted that the intent of testing is to find errors, and
suggested a better explanation of this and other tenets of a software
testing strategy. We have added such an explanation.
    Other comments argued that statistical testing based on usage
profiles is more effective than extensive structural testing in finding
software defects. We agree that statistical testing is one of many
valuable testing methodologies, and we have added information about its
use. However, it is important to note that statistical testing is an
adjunctive approach, rather than an outright replacement for other
types of testing.

O. User Site Testing

    Based on several comments, we have renamed the subsection formerly
entitled ``Installation Testing'' and moved it into the section on life
cycle activities. User site testing can be any one of several types of
testing performed by the user or by others at the user site. System
level testing performed by the software developer under conditions that
simulate the user's environment is an important part of validation for
some products, and it may substitute for some aspects of user site
testing. However, for certain products such as blood establishment
software, there are specific FDA requirements for additional testing to
be performed at the user site. For manufacturing and quality system
software, user site testing is frequently performed by the device
manufacturer.

P. Maintenance and Software Changes

    Several comments objected to the statement that ``all modifications
are design changes,'' noting that some changes, such as a correction of
coding errors, do not change the intended design. We have made
appropriate changes to the text. However, we continue to emphasize that
the validation of all software changes needs to include a regression
analysis and, as appropriate, regression testing to show that the
change has not negatively impacted the software.
    In response to other comments, we have added information regarding
anomaly evaluation, problem identification and resolution tracking, and
the need to update documentation.

Q. Process and Quality System Software

    We have added a new section to the document dealing with validation
of automated process equipment and quality system software. This change
was in response to the many comments that raised issues and asked for
more detailed information about validating such software, especially
OTS automated equipment and OTS software.
    Many comments discussed the difficulties encountered in trying to
validate OTS software, and suggested a different approach for
validation of manufacturing and quality system software. Source code
and life cycle documentation are frequently unavailable for review, so
structural testing is usually not possible. Auditing the vendor's
software development activities is one possibility, but some software
vendors will not agree to being audited. One comment suggested that
risk analysis, design, coding, and unit testing should not apply to
quality system software, especially if it is purchased, and further
suggested that functional testing is the most that can be expected.
Several comments suggested that for widely used applications, there can
be a reasonable assumption that the vendor validated the software at
the time it was developed, and that installation qualification by the
user should be sufficient. Many of these issues are addressed in the
response to comment 136 in the preamble of the quality system
regulation (61 FR 52602 at 52630).
    It is not the agency's intent to discourage use of OTS computer
products. The activities described in the guidance can be shared
between the vendor and device manufacturer (the user). However, we
believe that the principles and activities described in the guidance
are important for an overall conclusion that software is validated for
its intended use. Device manufacturers are required to have purchasing
controls for the products and services they receive. Such controls are
an important part of decision making regarding OTS software. Our
experience is that ``assumptions'' regarding validation by the vendor
are not always well founded. Each OTS software product needs to be
individually evaluated based on the intended use of the software,
available life cycle documentation, available verification and
validation evidence, and most importantly the device safety risk posed
by the automated process. Device manufacturers can use multiple sources
of information, but are ultimately responsible for documenting the
basis for their conclusion that the software is validated for its
intended use.
    Several comments suggested alternative approaches for certain types
of software, such as operating systems and certain tools used in
software development, such as compilers and robust ``middleware'' such
as Oracle, Documentum, or Lotus Notes. We have added suggestions for
alternative approaches, while still retaining the basic requirement
that the software must be validated for its intended use.
    A few comments questioned who is responsible for validation of OTS
software. One questioned FDA's

[[Page 1487]]

authority to regulate software vendors, but argued that device
manufacturers cannot be responsible because they lack access to source
code and life cycle documentation. Another noted that vendors
frequently change their hardware and software, resulting in
unreasonable FDA expectations for revalidation of each change. One
comment asked for more details regarding the impact of the supplier's
quality system on purchasing decisions. In response to these comments,
we reaffirm that FDA holds the device manufacturer responsible for the
software validation requirement. This responsibility can be further
delegated in part through contracting and purchasing controls, and
monitored through supplier audits or other means, but the device
manufacturer is ultimately responsible for its decision to choose a
particular software product. The fact that a vendor refuses to provide
access to its development process or documentation does not relieve the
device manufacturer of this responsibility. Likewise, we note that the
device manufacturer is not obligated to install every software upgrade
offered by a vendor. Validation of those upgrades and support from the
vendor, including access to the necessary vendor documentation, need to
play an important role in the upgrade decision.
    Some comments argued that software validation should be treated
more like process validation, which is only required if the output of
the process cannot be fully verified by subsequent inspection and
testing. Other comments asked for clarification of the term
``verification by output'' and asked whether it negated the requirement
for software validation. One comment argued that output of software
driven systems can never be fully verified. Another comment suggested
the consideration of intended use and dependence upon software for
proper operation of the process to determine whether verification could
be substituted for software validation.
    In response to these comments, we believe there are very few
examples where ``verification'' in lieu of software validation could be
justified, and even in those cases, most manufacturers would choose to
validate the software rather than go through repeated verifications of
output. For example, while every aspect of a drawing from a computer-
aided design (CAD) system can be independently verified, no user of a
CAD system is likely to go to that trouble or expense for every aspect
of every drawing. Likewise, because software itself cannot be fully
verified, automated software development tools used to create medical
device software must be validated for their intended use.
    Requirements are needed to establish intended use, the degree of
dependence on the software, and therefore the degree of validation
needed. The device manufacturer decides whether or not to use OTS
software. The ability to validate for intended use and vendor support
for the effort should be a part of that decision. Static analysis and
structural testing are techniques to be used in evaluating source code
and life cycle documentation, when these items are available.
Otherwise, the device manufacturer is dependent upon functional testing
alone. This issue is discussed in response to comment 136 in the
preamble to the quality system regulation (61 FR 52602 at 52630). The
impact on the safety and quality of the medical device is an important
determining factor in the approach and level of effort to be applied
for validating automated manufacturing and quality system software,
just as it is for software in a medical device.

R. References

    There were numerous recommendations for additional references.
Those and many other reference books, international standards, and FDA
guidance documents have been added to the appendix at the end of the
validation guidance.
    For ease of cross reference, the text of comment 136 from the
preamble of the quality system regulation is included below:

    136. One comment on Sec. 820.70(h), ``Automated processes,'''
(now Sec. 820.70(i)), stated that the section should be revised to
reflect that software used in such systems must be validated for
``its intended use,'' not simply validated. Another comment stated
that most companies buy software currently available on the market
and do not make changes to the software. It was recommended that
Sec. 820.70(h) allow for use of outside personnel for validation
runs and not necessarily require the development of a software
validation procedure. One comment suggested that the section should
allow verification rather than validation of off-the-shelf software.
Several comments on ``automated processes''' stated that the term
``data processing systems'' was unclear and its inclusion rendered
the requirement too broad. Others asked for clarification of
``automated data processing systems.''
    FDA has modified the requirement to mandate validation for the
intended use of the software. In addition, the requirement that the
software be validated by individuals designated by the manufacturer
has also been deleted to make clear that validation may be performed
by those other than the manufacturer. However, whether the
manufacturer designates its own personnel or relies on outside
assistance to validate software, there must be an established
procedure to ensure validation is carried out properly.
    FDA has maintained the requirement for validation because the
agency believes that it is necessary that software be validated to
the extent possible to adequately ensure performance. Where source
code and design specifications cannot be obtained, ``black box
testing'' must be performed to confirm that the software meets the
user's needs and its intended uses.
    FDA emphasizes that manufacturers are responsible for the
adequacy of the software used in their devices, and activities used
to produce devices. When manufacturers purchase ``off-the-shelf''
software, they must ensure that it will perform as intended in its
chosen application.
    FDA has amended the requirement to state ``When computers or
automated data processing systems are used as part of production or
the quality system,'' for clarification. Software used in production
or the quality system, whether it be in the designing,
manufacturing, distributing, or tracing, must be validated.

II. Significance of Guidance

    This guidance document represents the agency's current thinking on
software validation. It does not create or confer any rights for or on
any person and does not operate to bind FDA or the public. An
alternative approach may be used if such approach satisfies the
applicable statutes and regulations.
    The agency has adopted GGPs, and published the final rule, which
set forth the agency's regulations for the development, issuance, and
use of guidance documents (21 CFR 10.115). This guidance document is
issued as a level 1 guidance in accordance with the GGP regulations.

III. Electronic Access

    In order to receive ``General Principles of Software Validation''
via your fax machine, call the CDRH Facts-On-Demand system at 800-899-
0381 or 301-827-0111 from a touch-tone telephone. Press 1 to enter the
system. At the second voice prompt press 1 to order a document. Enter
the document number (938) followed by the pound sign (#). Follow the
remaining voice prompts to complete your request.
    Persons interested in obtaining a copy of the guidance may also do
so using the Internet. CDRH maintains an entry on the Internet for easy
access to information including text, graphics, and files that may be
downloaded to a personal computer with Internet access. Updated on a
regular basis, the CDRH home page includes the civil money penalty
guidance documents package, device safety alerts, Federal Register
reprints, information on premarket submissions (including lists of
approved applications and manufacturers'

[[Page 1488]]

addresses), small manufacturers' assistance, information on video
conferencing and electronic submissions, Mammography Matters, and other
device-oriented information. The CDRH home page may be accessed at
http://www.fda.gov/cdrh. Guidance documents are also available on the
Dockets Management Branch Internet site at http:/www.fda.gov/ohrms/
dockets/default.htm.

IV. Comments

    Interested persons may submit to the Dockets Management Branch
(address above) written or electronic comments regarding this guidance
at any time. Submit two copies of any comments,except that individuals
may submit one copy. Comments are to be identified with the docket
number found in brackets in the heading of this document. The guidance
document and received comments may be seen in the Dockets Management
Branch between 9 a.m. and 4 p.m., Monday through Friday.

    Dated: December 11, 2001.
Linda S. Kahan,
Deputy Director, Center for Devices and Radiological Health.
[FR Doc. 02-690 Filed 1-10-02; 8:45 am]
BILLING CODE 4160-01-S