Transcript THRio - CREATE Biostats Core

THRio
Outline

Data flow and database

Database matching (linkage) issues
Data flow

Data abstracted from medical charts





Specific forms (several)
Forms are reviewed on site
Forms get to our central office from Health
Care Units over the mail
Further inconsistencies are spotted by the
DB system
Double data entry
Database

Flexibility X Complexity

Access + VBA



Simpler
User friendly
Not all the features we needed
Database

SQL Server + Delphi





Very flexible
Secure remote access
Simultaneous transactions
Good freeware servers (MySQL, PostgreSQL)
Interface with programming languages (S,
Perl)
Database


Single server running PostgreSQL (Linux)
5 clients connected through intranet
(Windows)




No Internet access (security concerns)
Electronic forms – Delphi
Allows logs for every action in the DB
Double data entry analyzed within the
system
Database

Problems

System getting too complex


Auditing takes a long time



Will suffer major update
Will be done in real time now
We are still having bugs in the system that
have to be resolved
Not all forms are ready
Database matching

There are several systems that do not
“talk” to each other





SINAN – reportable diseases (TB, AIDS)
SIM – Mortality
SICOM – Pharmaceutical database (ARVs)
THRio – Our DB
We will need to match THRio with all other
3 DBs above
Database matching

Problems



THRio




There is no unique identifier common for all
systems
We use name, gender and DOB as surrogates
Standardization of names abbreviations
Double data entry
Not enough – names are misspelled
The other databases – even worse

No QC
Database matching

SICOM



Monthly cumulative DB
One Excel spreadsheet for each month
Names are repeated




No guarantee that the name will remain the same
for all spreadsheets
Missing DOBs
These problems are more prevalent in older
spreadsheets
Hard to reconstruct ARV history for patients
Database matching

Proposed solution

Compare different approaches
Translated SOUNDEX
 Reclink – probabilistic linkage
 Other algorithms



Apply to different examples and get
sensitivity/specificity for each one
SICOM – it’ll be trickier


Sequential matching
Match TB before doing the sequential
DSMB Issues

TB case definition



Very important
Our primary outcome
Special form to get info about Dx and Tx



Including alternative sources (e.g. “Black book”)
Discussed today in THRio meeting
Came up with a classification like this…
Clinical presentation
compatible with TB
TB Classification
Culture +
Culture -
Smear Clinical response to Tx
AND no alternative Dx
Smear +
Definite
Probable
Yes
Possible
No
No TB
DSMB Issues

Adherence evaluation

Sample of patients – standardized survey
Adapted from ARV adherence survey
 Already validated


Adherence should be reported in a more
timely fashion



Originally, data would be collected only after
course is complete
Now we will collect it whenever it is available
We’ll still have a delay
DSMB Issues

Education level


Years of study (self-reported)
Race/ethnicity




Population highly admixtured in Brazil
Very hard to objectively classify
Even the Brazilian Census Bureau uses selfdeclared race/ethnicity
Could lead to distortions