was the representations Theranos made about the precision of its
blood tests: the company claimed that its tests had coefficients of
variation of less than 10 percent, but the CVs in many of its validation
reports were much higher, he told her. Elizabeth acted surprised and
said she didn’t think Theranos had made such a claim. She suggested
they look at its website together and called it up on her big iMac. A
part of the site titled “Our Technology” did prominently advertise a
coefficient of variation of less than 10 percent with a catchy green-and-
white circular logo, but Elizabeth noted that the smaller print above it
specified that the claim only covered Theranos’s vitamin D test.
Tyler conceded her point and made a mental note to check the
vitamin D validation data. He then brought up the fact that his own CV
calculations often didn’t match those he found in the validation
reports. By his count, the percentages in the reports were lower than
they should be. In other words, Theranos was exaggerating the
precision of its blood tests.
“That doesn’t sound right,” Elizabeth said. She suggested he go
speak with Daniel Young. Daniel would be able to walk him through
how Theranos performed its data analyses and clear up any confusion.
Over the following weeks, Tyler met with Daniel Young twice. Talking
to Daniel could be frustrating. He had a long forehead accentuated by
a receding hairline that suggested a big, powerful brain. But it was
impossible to know what went on inside that brain. His eyes, behind
their wire-rim glasses, never betrayed any emotion.
During their first meeting, Daniel calmly explained why Tyler’s CV
calculations were wrong: Tyler was taking into account the six values,
or “replicates,” generated during each Edison test instead of just the
median of those six values. The final result Theranos reported to a
patient was the median, so only that number was relevant to CV
calculations, he said.
Daniel may have technically been correct, but Tyler had put his
finger on a central weakness of the Edison device: its pipette tips were
terribly imprecise. Generating six measurements during each test and
then selecting the median was a way to correct for that imprecision. If
the tips had been reliable in the first place, there would have been no