167
Software as a Medical Device: Regulatory and Market Access Implications
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
the health institution meets the conditions for
in-house manufacturing, meaning:
• The device is not transferred to another legal
entity.
• It has an appropriate quality management
system in place.
• It justifies the target patient group’s specific
needs cannot be met or cannot be met at
the appropriate level of performance by an
equivalent device available on the market.
• It provides information on the
manufacturing, modification, and use to its
competent authority upon request and draws
up a declaration that it shall make publicly
available.29
The exception to this rule is when a health
institution adapts a device for a purpose not in
the scope of the medical device definition, e.g.,
a machine learning bionic limb is adapted for
superhuman purposes rather than to alleviate or
compensate for a disability. The health institution
then must not meet the conditions for in-house
manufacturing. The author is not aware of any
restrictions outside the EU that apply to health
institutions performing manufacturing.
Controllability and Human Oversight
Controllability refers to the ability of an external
operator to intervene or deactivate a machine
learning device in a timely manner.30 Figure 11-7
illustrates the different types of controllability.31
Having a human in the loop leverages the user’s
situational awareness and judgement, which may
be beneficial for the performance and safety of
machine learning devices.
Having a human in the loop requires situa-
tional awareness, enough time to intervene, and a
mechanism to interfere (a communication link or
physical controls) and take control or deactivate
the device as required.
From a risk management and performance
perspective, it may sometimes be necessary to
take the human out of the loop to reduce the
risk as far as possible to avoid human-machine
Figure 11-6. Illustration of AI Change Types
Most medical device regulations allow hardware devices that recalibrate or reconfigure themselves within
certain boundaries for which a conformity assessment was carried out to be placed on the market. Similarly,
manufacturers can place devices on the market that comprise AI that changes within pre-defined boundaries for
which a conformity assessment was carried out. Currently, no medical device legislation is known that allows
manufacturers to place devices on the market comprising AI that changes outside of pre-defined boundaries.
Locked Change
within
pre-defined boundaries
for which the conformity assess-
ment was carried out
Change
outside
pre-defined boundaries
for which the conformity assess-
ment was carried out
© Koen Cobbaert 2021
Software as a Medical Device: Regulatory and Market Access Implications
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
the health institution meets the conditions for
in-house manufacturing, meaning:
• The device is not transferred to another legal
entity.
• It has an appropriate quality management
system in place.
• It justifies the target patient group’s specific
needs cannot be met or cannot be met at
the appropriate level of performance by an
equivalent device available on the market.
• It provides information on the
manufacturing, modification, and use to its
competent authority upon request and draws
up a declaration that it shall make publicly
available.29
The exception to this rule is when a health
institution adapts a device for a purpose not in
the scope of the medical device definition, e.g.,
a machine learning bionic limb is adapted for
superhuman purposes rather than to alleviate or
compensate for a disability. The health institution
then must not meet the conditions for in-house
manufacturing. The author is not aware of any
restrictions outside the EU that apply to health
institutions performing manufacturing.
Controllability and Human Oversight
Controllability refers to the ability of an external
operator to intervene or deactivate a machine
learning device in a timely manner.30 Figure 11-7
illustrates the different types of controllability.31
Having a human in the loop leverages the user’s
situational awareness and judgement, which may
be beneficial for the performance and safety of
machine learning devices.
Having a human in the loop requires situa-
tional awareness, enough time to intervene, and a
mechanism to interfere (a communication link or
physical controls) and take control or deactivate
the device as required.
From a risk management and performance
perspective, it may sometimes be necessary to
take the human out of the loop to reduce the
risk as far as possible to avoid human-machine
Figure 11-6. Illustration of AI Change Types
Most medical device regulations allow hardware devices that recalibrate or reconfigure themselves within
certain boundaries for which a conformity assessment was carried out to be placed on the market. Similarly,
manufacturers can place devices on the market that comprise AI that changes within pre-defined boundaries for
which a conformity assessment was carried out. Currently, no medical device legislation is known that allows
manufacturers to place devices on the market comprising AI that changes outside of pre-defined boundaries.
Locked Change
within
pre-defined boundaries
for which the conformity assess-
ment was carried out
Change
outside
pre-defined boundaries
for which the conformity assess-
ment was carried out
© Koen Cobbaert 2021