DESIGN DECISION 07
DATA VALIDATION METHOD
The design only works
if the data underneath does.
A new identity key means nothing on top of four copies
of the same patient. I helped engineering build a
validation method that turned the whole database into
clusters the Medstation team could review and merge.
01
Collect
Every record from the three
systems in a single table
02
Match
Character similarity, typos,
shortened and swapped names
03
Cluster
Likely duplicates grouped
with a confidence score
04
Validate
SEVERAL WEEKS
Medstation reviewed every
cluster, record by record
05
Merge
One record per patient,
full history kept
METHOD DESIGNED WITH ENGINEERING
CLINIC TEAM
ENGINEERING
ONE CLUSTER, AS THE CLINIC REVIEWED IT ยท ILLUSTRATIVE DATA
NAME
DATE OF BIRTH
SOURCE
MATCH
SIGNAL
Amanda Helene Costa
03/14/1988
CRM
Base
Most complete record
Amanda
H.
Costa
03/14/1988
Scheduling
94%
Shortened name
Amanda Cost
t
a
03/14/1988
Medical records
91%
Typo in surname
Amanda Helen
a
Costa
14/03
/1988
CRM
88%
Spelling, date order
MERGED RECORD
Amanda Helene Costa
Date of birth 03/14/1988
Four profiles became one
Every visit and exam kept
Composite key applied
For the first time, the clinic trusted its own patient list. The new key had a clean base to land on.