Evaluation (Validation and Verification) - Medical Device Design: Innovation from Concept to Market - page 242

9.6.6 Analysis of Data
Without exception, the analysis of the data obtained from your study will entail some
statistical analysis. That is why your study must be designed to meet statistical considerations
at the start. For most scientists and engineers the thought of performing a statistical analysis
is not daunting. However, for many manufacturers statistical analysis may be as alien as life
on Mars. It is beyond the scope of this text to teach you statistics, but there are a number of
tools the average computer literate person can attempt to use. The first is correlation and the
second is student t-tests; both are openly available in spreadsheet packages such as Microsoft
Excel®. If you want to read more then there is a plethora of textbooks on the subject. Look
for ones that are targeted at clinical studies or clinicians as these are directly relevant.
9.6.6.1 Outliers and Missing Values
In any trial there will always be exceptions to the rule. Some subjects will just “disappear,”
often quite logically, from the study. These are called missing values and can legitimately be
excluded from any analysis, but you must state so in the analysis report.
Outliers are subjects whose results are way off the norm. This is normally due to some
congenital, physical, or historical reason. For example, one of your subjects may have been
a habitual smoker and only recently stopped. Hence when you asked “Are you a smoker?”
they reply “no”; as a consequence their result may well differ greatly from everyone else
who replied “no” to being a smoker. This is a good reason to revisit your exclusion criteria;
it is a valid reason for excluding the data from the analysis. Another valid reason may be
to exclude all subjects with healing times 2× the average value; but this one is subjective
and should be stated in the study plan. It is not acceptable to simply remove data because it
distorts your averages: this is called fixing the data .
9.6.6.2 Correlation
One of the methods for confirming that the output (or the result of your study) is due to your
intervention is to examine the correlation. Correlation, effectively, analyzes the data accepting
that there will be a naturally occurring scatter in the data and that you will need to see that
if you change A then B follows. The best way to illustrate this is by example. Suppose you
had a study examining the effect of the length of time your device was used on the pain relief
obtained. One may have a table of data such as illustrated in Table 9.12 .
A first look at the data suggests that the device causes a reduction in pain. If we plot a graph
of duration versus difference in pain score we can fit a straight line to the data. If we use a
spreadsheet to do this then we can request to show the line equation and the R 2 value; this is
the correlation coefficient ( Figure 9.22 ).
The straight line suggests that as duration increases the reduction in pain increases. However
the correlation coefficient R 2 =0.08. Table 9.13 indicates typical acceptance values of R 2
for p =0.05.