12 Method Development
283
conditions will exist during routine use of the method and impact the performance of the model. Having a clear picture of the expected variability will provide
insights into the scope of application and the variables/properties that may affect
method robustness.
Needless to say, however, that comprehensively addressing these questions may
not be feasible at an early stage. By its nature, a feasibility study cannot explore all
the sources of variability that a model will need to handle during routine use. For
instance, exposure to limited sources of variability may provide over-confidence in
the performance of a method. A feasibility study designed to not address known risks
will not provide the relevant outcomes, specifically for robustness.
Progressing from a feasibility study to method development is a business decision,
and the scientific team should weigh the costs of progressing without all the answers
versus performing a more extensive feasibility study (whose data could be used in the
method development) and advance with more certainty. The feasibility study should
also be used to potentially update the ATP and gain endorsement from stakeholders
if factors have been identified that will change the desired performance criteria.
12.3.4 Risk Assessment
The risk assessment is the process of identifying and scoring the causes that could
have an impact on the method. The result of a risk assessment is a formal document
identifying the risks that exist to the method and how they will be controlled, accepted,
mitigated, or avoided.
There are many ways to conduct a risk assessment. But, the general principles are
as follows. First, all the stakeholders should be identified and convened together to
conduct the exercise. These members should represent management (if appropriate),
the process engineers, the method developers, and the method users. Getting the right
membership to the risk assessment will ensure that all the relevant risks are identified.
Second, the team should list all the possible failure modes that could affect the method
performance. Figure 12.3 presents an example of a risk identification process also
known as a fishbone diagram for a hypothetical method aimed at predicting the
content of an active ingredient in-line. As presented, the risk identification should
cover all relevant causes of impact to the model performance. Third, the risks should
be categorized or scored by the team. If using a scoring system, each risk is assessed
one at a time, for criteria such as the probability of occurrence, the severity if it
occurred, and the probability of detection. Setting scores for the various risks will
require team members to use a priori information from previous projects, feasibility
studies, or experiences. Some risks may be scored high at first because at the time of
the risk assessment, limited information may be available and then re-scored as new
knowledge becomes available. On the contrary, a risk could have been determined
to be low after the feasibility study but may prove to require controls or mitigation
283
conditions will exist during routine use of the method and impact the performance of the model. Having a clear picture of the expected variability will provide
insights into the scope of application and the variables/properties that may affect
method robustness.
Needless to say, however, that comprehensively addressing these questions may
not be feasible at an early stage. By its nature, a feasibility study cannot explore all
the sources of variability that a model will need to handle during routine use. For
instance, exposure to limited sources of variability may provide over-confidence in
the performance of a method. A feasibility study designed to not address known risks
will not provide the relevant outcomes, specifically for robustness.
Progressing from a feasibility study to method development is a business decision,
and the scientific team should weigh the costs of progressing without all the answers
versus performing a more extensive feasibility study (whose data could be used in the
method development) and advance with more certainty. The feasibility study should
also be used to potentially update the ATP and gain endorsement from stakeholders
if factors have been identified that will change the desired performance criteria.
12.3.4 Risk Assessment
The risk assessment is the process of identifying and scoring the causes that could
have an impact on the method. The result of a risk assessment is a formal document
identifying the risks that exist to the method and how they will be controlled, accepted,
mitigated, or avoided.
There are many ways to conduct a risk assessment. But, the general principles are
as follows. First, all the stakeholders should be identified and convened together to
conduct the exercise. These members should represent management (if appropriate),
the process engineers, the method developers, and the method users. Getting the right
membership to the risk assessment will ensure that all the relevant risks are identified.
Second, the team should list all the possible failure modes that could affect the method
performance. Figure 12.3 presents an example of a risk identification process also
known as a fishbone diagram for a hypothetical method aimed at predicting the
content of an active ingredient in-line. As presented, the risk identification should
cover all relevant causes of impact to the model performance. Third, the risks should
be categorized or scored by the team. If using a scoring system, each risk is assessed
one at a time, for criteria such as the probability of occurrence, the severity if it
occurred, and the probability of detection. Setting scores for the various risks will
require team members to use a priori information from previous projects, feasibility
studies, or experiences. Some risks may be scored high at first because at the time of
the risk assessment, limited information may be available and then re-scored as new
knowledge becomes available. On the contrary, a risk could have been determined
to be low after the feasibility study but may prove to require controls or mitigation