the clinical lab and assigned to Normandy, the room downstairs with
the lab’s Edison machines. During the Thanksgiving holiday, a patient
order came in from the Walgreens store in Palo Alto for a vitamin D
test. As she had been trained to do, Erika ran a quality-control check
on the Edison devices before testing the patient sample.
Quality-control checks are a basic safeguard against inaccurate
results and are at the heart of the way laboratories operate. They
involve testing a sample of preserved blood plasma that has an
already-known concentration of an analyte and seeing if the lab’s test
for that analyte matches the known value. If the result obtained is two
standard deviations higher or lower than the known value, the quality-
control check is usually deemed to have failed.
The first quality-control check Erika ran failed, so she ran a second
one. That one failed too. Erika was unsure what to do. The lab’s
higher-ups were on vacation, so she emailed an emergency help line
the company had set up. Sam Anekal, Suraj Saksena, and Daniel
Young responded to her email with various suggestions, but nothing
they proposed worked. After a while, an employee named Uyen Do
from the research-and-development side came down and took a look
at the quality-control readings.
Under the protocol Sunny and Daniel had established, the way
Theranos generated a result from the Edisons was unorthodox to say
the least. First, the little finger-stick samples were diluted with the
Tecan liquid handler and split into three parts. Then the three diluted
parts were tested on three different Edisons. Each device had two
pipette tips that dropped down into the diluted blood, generating two
values. So together, the three devices produced six values. The final
result was obtained by taking the median of those six values.
Following this protocol, Erika had tested two quality-control
samples across three devices, generating six values during each run for
a total of twelve values. Without bothering to explain her rationale to
Erika, Do deleted two of those twelve values, declaring them outliers.
She then went ahead and tested the patient sample and sent out a
result.
This wasn’t how you were supposed to handle repeat quality-control