EEG Signals in Psychiatry: Biomarkers for Depression Management - Quantitative EEG Analysis Methods and Clinical Applications - page 303

Table 11.1 Heterogeneity Within Diagnoses
Patient A Patient B
Depressed mood Anhedonia
Insomnia Hypersomnia
Weight loss Weight gain
Agitation Psychomotor slowing
Reduced concentration Feelings of worthlessness,
guilt
Fatigue Suicidal ideation
Note: Two patients both meet the criteria for major depression, yet
have no symptoms in common.
Thus, development of biomarkers also should disclose the nature of the patient pop-
ulation and consider evaluating whether the accuracy and reliability of the measure
are improved or degraded in some subpopulations (e.g., psychotic depression,
depression in Bipolar I versus Bipolar II patients).
Although biomarkers should have a high degree of clinical utility in order to be
considered for use, there is also a need for them to be interpretable in the context of
the rest of neuroscience. What aspect of a patient's pathophysiology is being
assessed by a test: the form of a reuptake transporter that is associated with greater
or lesser efficiency? the level of activity in a particular brain region? a component of
a neuroendocrine feedback loop? Biomarker methods that fail to be comprehensible
within or integrated into the extant body of neurobiological knowledge are unlikely
to gain clinical acceptance, even if an empiric trial suggests that they might be
useful.
Finally, it is worthwhile to note that statistical significance is not the same thing
as clinical significance. Studies may report that a result is significant at the p < 0.05
level, really meaning that there is less than 1 chance in 20 that their finding arose by
chance alone. Given a large-enough sample, even a clinically irrelevant difference
(e.g., a very small improvement on a clinical rating scale) might be reported to occur
with an impressive p -value.
An important measure for evaluating biomarkers includes the “number needed
to treat” (NNT) [136], which assesses the number of patients needed to be treated
differently (e.g., with biomarker guidance, with a new medication) in order to have
one additional patient experience the desired, positive outcome.
Predictive biomarkers are also often characterized by a series of metrics that can
help evaluate the usefulness of a potential biomarker: ROC curves and measures
such as sensitivity, specificity, and overall predictive accuracy [137-139]. Sensitiv-
ity is the ratio of “true positive” tests to the number of individuals with the condi-
tion; for an outcome predictor, it would be the number of people in a sample who
are predicted to respond to a treatment, divided by the total number of people who
actually respond. Specificity is the ratio of “true negative” tests to the number of
people who do not have a particular condition; in the outcome predictor context,
this would be the number of people predicted not to respond divided by the total
number of nonresponders. Overall predictive accuracy is the proportion of predic-
tions that are correct. ROC curves plot the trade-offs between sensitivity and speci-