Biomedical Engineering Reference
In-Depth Information
The goal of each performance evaluation was to assess the ability of the
LPM / SPM ratio to discriminate or detect differences in APSD as it relates to changes
in the central tendency. MMAD is not sufficient by itself to describe the entire
APSD; however, once the expected distribution of MMAD values for an OIP is char-
acterized, it can be used to assess atypical behavior or acceptability of the product,
much like any other QC metric.
In the performance evaluations excluding PCA, the MMAD value for each APSD
was used as the pivotal comparison parameter (i.e., MMAD was treated as a known
variable without error). The approaches developed by Tougas to compare EDA with
grouped stages involved two quite different techniques. The first methodology was
essentially a MSA that focused on the predictive relationships between each of the
grouped-stage metrics and MMAD as well as the relationship between LPM / SPM
ratio and MMAD . This MSA technique is analogous to the variance equivalent
approach discussed by Hauck et al . [ 20 ]. Its objective involved comparing the mag-
nitude of uncertainty observed in the predicted MMAD values from an inverse
regression analysis of each grouped-stage metric and LPM / SPM . In other words, a
regression analysis was performed with MMAD as the known independent variable
( x -space), and the metric ( LPM / SPM or stage group mass) represents the measured
response ( y -space). However, instead of predicting the response/dependent variable
( y = LPM / SPM , group 2 mass, group 3 mass, or group 4 mass) from the explanatory/
independent variable ( x = MMAD ), the reverse process was put in place to predict
MMAD from a given metric value. The uncertainty in that prediction was repre-
sented in terms of its associated 95% prediction interval. This uncertainty measure-
ment was then used to determine performance indicators such as the discrimination
index, the precision to total variability ratio. These performance indicators pre-
sented in Table 8.5 show that the LPM / SPM metric is more precise in determining
and assessing the central tendency of an APSD for OIPs.
The second Tougas approach was similar in form to that adopted by Christopher-
Dey approach, in that both groups used OCC-based techniques that are analogous to
the decision equivalent approach described by Hauck et al . [ 20 ]. Both OCC-based
techniques made use of the blinded IPAC-RS OIP APSD data sets to simulate dif-
ferent APSD populations. The APSD profiles simulated by Tougas were based on a
simplistic model with a number of assumptions that have been described in
Sect. 8.5.2.1 . The Christopher-Dey OCC approach employed a more sophisticated
modeling of the actual CI data (see Sect. 8.5.2.2 ), with simulations relying on fewer
assumptions and driven more by characteristics of the actual CI data.
In the decision equivalent techniques, the objective was to determine how the
EDA metrics, relative to the grouped-stage metrics, would declare an individual
APSD profile acceptable or unacceptable when compared to specifications. Since
the IPAC-RS database did not contain information regarding the regulatory specifi-
cations applicable to the particular OIP concerned, pseudo-specifications were
created to judge acceptable versus unacceptable outcomes.
Search WWH ::




Custom Search