Would You Like To Be Statistically Diagnosed?

The Eomix Newsletter: July 2006 (Thus, the future diagnostic microarrays will potentially
The above title may sound funny but with theeliminate some of the need for traditional,
advent of microarray technology it is entirely possibledoctor-enabled pattern recognition. I do not think,
for your doctor to ask you a question like this.however, that the pattern-recognition capacity of
Namely, any array-based diagnostic test thatmicroarrays is anywhere near to do so because (1)
determines a multi-component disease signature istechnical standards for diagnostic microarrays do not
inherently generating a statistical statement. Forexist, (2) the mathematics of pattern-recognition
example:makes sense to statisticians but not doctors and
The test results demonstrate with 90% confidencepatients, and most importantly, (3) the medical
that your chances to have Disease X are 75% ascommunity does not have surrogate biomarkers,
detected by the marker-pattern, which is up to 85%except for infectious disease. In short, in addition to
sensitive and specific to Disease X.great potential, the microarray technology brings
The above statement may sound unintelligible andunique challenges for scientists, IVD industry, medical
make no medical sense but if you compare thecommunity, and the regulatory institutions. Some of
process behind this hypothetical machine-generatedthese challenges are formulated in DIAGNOSTIC
statement to traditional clinical diagnostics, then youNUCLEIC ACID MICROARRAYS [MM12A, 2006;
realize the similarity. Namely, both are actually patternClinical and Laboratory Standards Institute (CLSI)].
recognition. Traditionally, diagnostics has been enabledAnd I would like to point here to my favored one:
by doctors' experience so that the more experiencedQUALITY MANAGEMENT OF THE PRODUCTION
the doctor, the better his of her ability to recognizeAND PERFORMANCE OF DIAGNOSTIC
symptoms and synthesize these symptoms into aMICROARRAYS. Imagine the amount of quality
disease-pattern, a.k.a. as DIAGNOSIS. Except forcontrol and quality assurance (QC/QA) you need for
simple cases, doctors do pattern-recognition routinely.a 10-element array as opposed to a traditional ELISA
For example, you go to your doctor with complaintstest that determines only a single marker. Remember
including chronic fatigue, occasional fever, head-ache,that one has to be able to QC/QA all of the array
shortness of breath, and loss of appetite. Yourelements. The best way do so is to design a
doctor listens to you, looks at you, and analyzesstatistical QC/QA methodology, and cut back on the
your chest X-rays, blood-test etc. Depending onnumber of array elements as much as possible. In
what your doctor finds, you might either have somesummary, despite great promise, diagnostic arrays will
pulmonary ailment or no detectable disease at all.have to overcome formidable technical challenges.
How do doctors come up with these diagnoses?And before they enter routine clinical practice, the
Very simply, they analyze patterns, which can greatlydiagnostic arrays that produce multi-component
overlap from disease to disease but also containdiagnostic signatures will have to prove that they
specific components that determine the diagnosis andmake clinical diagnostics more accurate and
therapy.cost-effective. Unless they become capable of
At some point in the future, the era of diagnosticdetecting surrogate biomarkers, which for most
microarrays will arrive, and the clinical patterndiseases are yet to be discovered, or provide useful
recognition is enabled by tiny chips, which quantifyinformation as to the best therapy for a patient, we
molecular disease markers instead of talking to you,will only be seeing experimental tests along with
looking at you, listening to you, and analyzing yourstatistical diagnostics (see above). So, would you like
X-raysto be statistically diagnosed?