Table 3.

Illustrative methods for primary validation of deconvolution algorithms

1. Direct sampling at the site of secretion to ensure high signal-to-noise ratio

Hypophyseal-portal sampling for oxytocin, CRH, and AVP (37,38)

Pituitary intercavernous-sinus sampling for LH, GnRH, GHRH, somatostatin, and ACTH (65)

Portal- or hepatic-venous sampling for insulin (40,71)

2. Corroboration by independent in situ markers of pulse events

Electrophysiological recordings in hypothalamus to mark oxytocin or GnRH pulses (39,360,361)

Injected GnRH pulses in men with isolated GnRH deficiency (287)

3. In vivo suppression or stimulation paradigms

GnRH-receptor antagonist administered before iv infusion of biosynthetic LH pulses (270,272,286)

Somatostatin, insulin, and glucagon infusions (14)

Insulin-pulse entrainment by oscillatory iv glucose infusion (362)

4. In vitro perifusion models

Ovarian granulosa-luteal cells (41,197)

Human islets (42)

Pituitary cells (363)

5. By computer simulation model-based simulations

Computer-assisted mathematical simulations to designate pulse locations, shape and amplitude, basal secretion, and half-life with superimposed random perturbations (245,249,250,287,313,364)

6. Test-retest reliability analysis

Correlation between estimates made in separate sampling sessions (328,365,366)

Limitations of each approach: 

1. Direct sampling is unethical in healthy humans and may disrupt secretion patterns in animals. 

2. The surrogate marker and pulse might not always correspond 1:1. 

3. Exact experimental mimicry of endogenous pulses is difficult. 

4. Perifusion conditions diverge in complex ways from in vivo physiology. 

5. Mathematical constructs are only estimates of an unknown physiological process. 

6. Reliability coefficients may be reduced by factors unrelated to the algorithm.