Page 1
Practical
d'écriture 21st
century
Interview with Anne-Laure, statistician at the Inami
Page 2
Page 3
'Chaque Graph is like a novel for me '
she said one day.
'Faire Graphics, it's way
d'apercevoir worldwide. Through'norme' it
looking for patterns, and it is verified that deviates,
the reasons we seek. Sometimes there are reasons
which are entirely legitimate, sometimes they are
more questionable ... For this to be done well, it
would have all the ground is covered and the
jurisdiction explicit, but the reality is
always a little more complicated than the
BOM! '
Interview with Anne-Laure, Statistician.
Directed by Constant An Mertens, Brussels, June
2013.
1
Page 4
2
Page 5
Some kinds of novels in the analysis
data
The detective story
Anne-Laure: A colleague example examines
providers who bill the Global Medical Record.
Normally, you can not charge if you saw the
patient, and more than once a year. For
provider, we see that in his billing invoice
for patients he has never seen, and it has a
systematic in billing, but it is not known
which yet. Often our work is this: it
looking at how the crime was committed. We play
riddles in the data, we look for the life that is
behind. It's like when you trace a
Crime scene: data track and you
looking to reconstruct the story. You play cat and
Mice with the data, it is fun, you're in
game. It's fun to do, I love the novels
police, and has that side'I'll coincer'. We play
smarter with the provider.
In fact we always try to decode the data,
find in the data logic Due fraud, and if
was something wrong, you can go check on
field, for patients, for example.
The novel Genesis
Anne-Laure: In data analysis more
complex, the goal is to reopen the data
3
Page 6
give them meaning. I prefer because the story
is a little more complicated: you can compare it to
when someone dies and where you keep the
memory of his presence. The data is somewhat
like life the memory of someone. From the
Data you can réévoquer, reinvoke the history of
no one. That leaves more room for the interpretation
the method of "whodunit" is what you recreated
richer, more complex.
We made an analysis, for example, in the
Gastroenterology, in this case and we look
very openly if there are weird things. To
get there you are forced to rebuild everything. The
Doctors who specialize in oncology, for
example, not all the same practice as
not oncologist. For one, it is quite
normal to gastroscopy month to some
patients, while for others, rather it is expected
gastroscopies repeated more than once a year.
It's difficult to understand the rules of a
field when you are outside this area. He has a lot to
things in billing for which you do not know
if it's normal or not. You always need
the story around, context, a doctor who has a
view of the field, which is the translation.
Another example was the question of how
can reduce expenses on medicines with
the special permission of the medical officer. He had a will
action on certain groups of drugs and our
role was to see what had been the effect of this action
on reducing expenses. Behind every
4
Page 7
group of drugs we analyzed, you have
different stories, which are also in relation to the
Specific pharmaceutical companies. Each
drug has its peculiarities.
So we track, to see how the action made
changed things, but to be able to interpret
differences and really understand what is happening,
we have to go back to the context. You can the
Compare accounting. You can not
understand the reasons for spending watching
billing, but from acceptable as you can all re-open,
re-narrate. As if you open a book and each
Page is an anomaly that you see and you're trying to
figure out. On every page you turn, you're more
deep in the questioning. Or,
like when you meet a new person, the
you start printing, and then you're gonna ask
questions to understand the family, ideas,
particular story. This induces problems
d'interprétations, it's
messy
.
In the case of these drugs, we stayed with
many questions: what costs fall
because the drugs have gone generic
or because there are fewer patients? If treatments
are for chronic conditions, how to make
reduce a budget? How to measure a change
monthly if there patients receiving boxes
Drug three months ...?
We have not saved what was expected, but
compared to what would have been spent if we had nothing
5
Page 8
Indeed, much has been saved ...
Finally for that matter, was formulated 9
calculation methods, 9 possible points of view! We
was clear on the possible parameters and we said:
decide what is your strategy. This returns the
person asking the questions to his responsibility: if
to account for the complexity of the issue, because
there is no way to decrease quickly, for example, and
a decrease, it is a complex issue,
itself! And then you have to see it in time, long
term.
6
Page 9
Methods
Year: How I imagine your work in the
concrete?
I get the billing data extractions
disease insurance. What I have in
Data is the date of the act, the act numbers
in the nomenclature of diseases (that is accurate enough,
there are 20 000 classification codes, can be
see the website of the INAMI), the provider number,
Patient national registration number, name,
address, invoice amount, the date on which the act
is introduced into mutual, the hospital, the location.
This is very accounting info, but very
Detailed.
I get the raw data. We have access to everything
requires control work. It has access to the
Billing of a physician, patients with identified
in order to verify the statements of the provider. And
From these data, it restores relationship between
things. There are living at the same address and
who have the same last name, it is a link. There are
patients returning to their doctor, it is a
link. There are providers who share a
patient base is a link ..
Using data completely flattened and
detailed, it's up to me to make connections based on
variables: same provider or not, same service
or not, same place or not, event sequences
etc ...
7
Page 10
Year: Is there a way to see the process?
Is that at the end of the process can be read
in the graphs?
Anne-Laure: I realize that I document
very few steps, the way my work
is moving! Fortunately, it's all
programming, there is a trace in the code. There
regular structure of the SAS code could be:
Datastep
: You eliminate the data that you do not have
need you audited, you grouped in categories,
etc ..
Sql proc
: You select the data and you make it a
money, or other computing
Histogram
: For the variable you are visualizing the'quantité'
eg results of the 2011 and / or thou compares
with the same calculation for 2008
If you do it well, you put the titles, which will be displayed
in graphics I export a document
word. Otherwise, it is impossible to find yourself in the
code after which graphics just what stage!
But it is a habit that takes as and
extent, to better document the, delete
code parts that were not used to it
recover even after a few months.
We work in stages. We made a step, we test it
the custody or not.
I try to separate the code in the portion of the
data preparation (with a stable Datastep) and
8
Page 11
the data analysis, but it does not work
like that when you work: it does both
in parallel: code & context.
The other day a colleague told me, patients who
made a d'oncologie analysis, I think they should be
removed from the analysis, because it is in patients with
a particular profile. We must therefore change the data
departure and eliminate all that. In addition, sometimes I do
calculations on all patients, sometimes on a part
only. Often I do not know very well what I have
done, so I do everything to turn, so I'm sure
I have not missed a step. Sometimes it runs 1.5 hours
or 2 am, sometimes I change a little trick and I must
again. That said, I do not do this full cycle 5-6
Once a year, not more.
It takes time and it must be documented! In
Generally, I do not understand anything about my code when
I come back over a month later. Fortunately, more
you programs, more is always done.
An: What tools do you use?
Anne-Laure: We work on databases
very large, so you have a large system (SAS), which is
complicated to use. The digital data themselves
usually go back two or three years, but sometimes
working on data from the late 90s.
Then, publish the results in a written report of
20 to 30 pages. I put the questions, assumptions
and then the details of the analysis.
Year: And the graphs.
9
Page 12
Anne-Laure: Both. There are people who understand
good graphics and other tables.
Personally, I like the squatter plot, which
use 1 point per person on two axes. They
allow more complexity, they make
developments finer, more visible. They show
variability between individuals. Sometimes it sends a
mail providers, in the kind of'la limit
normal is 10%, you are 15%, there should be
careful, it is costing the health insurance ... 'In
They usually take them into account. With one
chart, you can show how each individual
the modified group (or not!) behavior.
The view of the choir
Year: From your calculations you find points
specific view, you could choose one, and
not leave the choice to the other .... Do you have
d'appliquer not want your knowledge
other areas that the Inami?
Anne-Laure: I've done Garance, an association
Brussels to fight against all kinds of
violence. I worked on the results of their
of assessment questionnaires. There were questions
open, the horror quantitative because you're
have to recode all the answers. We had 200
or 300 responses to two different questions. When you
open questions, you have all the voices
individual which are represented, each with their
differences, their specificity, while in the
10
Page 13
closed questions, there are 2 or 3 possible voice and you
abstracts answers. In open questions
Garance, it was impossible.
A question like,'Qu'est it was hard for
the self defense course? ', led to answers
depended on the experience of each. One of
responses was for example, someone'taper
knowing that we can break his leg, because I
am aware that I can do mal'.
There was wealth in their questionnaires. They
asked two short responses open to all
world, that is a little to much. In their
If, in my opinion, it could be better to ask
much like a small number, for example by
individual interviews with some participants,
rather than a questionnaire to everyone, because
the exploitation of a questionnaire is necessarily
simplifying .. But it's still difficult choices,
they also want to have variables that indicate the
satisfaction of each participating ... hence the choice
d'avoir open questions in a questionnaire
general anyway. I almost never have to
open questions in my job.
An: Open questions therefore allow
no collective narrative?
Anne-Laure: If They allow choral narration.
as Svetlana Alexievitch does in his book
There
Supplication
or
Voices from Tsjernobil
. The author
interviewed hundreds of people on the tragedy
of Tsjernobil and it makes something that is not
11
Page 14
more journalistic, that's choir and you can hear all
voice of all people, rewritten. The
perspectives, which is serious or not, depend on
experiments. You follow a person or another, but you
are immersed in a collective situation.
An: You work so exclusively with the
result of personal choice questions
multiple, often ask me great
problems ... How not to complete the box
'varia'? The typical example is that your kind
it is male or female? How you as
anyone so sensitive to that, how
you POSITIONED you in relation to this?
Anne-Laure: It's complicated. There is a book very
interesting is written in relation to this,
Sorting
things out
Susan Leigh Star.
Here it is more of a nomenclature problem. You
all legislation found on the website of the INAMI and you
see that the acts referred many are defined so
blurred. The boundaries are blurred. Once you have a flaw
in the nomenclature, often a part of
providers will benefit ...
The other problem is the problem of the identification.
Events are described in the billing and
each individual shall have a number, otherwise we
can not identify it. When providers or
patients do not have a number, it's a mess. By
example, there are in smoking cessation which have no
number because they are not doctors, but
psychologists. So, I have a'fourre-tout' category
12
Page 15
which spends 200,000 euro with people I do
can not identify, it does not suit me at all!
To be precise in the calculations, it would
cover all the bases, numbering, identifying all
the world! And here we touch a discussion
problematic, comparable to that of monitoring,
which has the same problem of legislation.
The health insurance generates a billion lines
Billing per year! It was therefore a lot of data,
Due enormous wealth but are underused
Data analysis. For the moment, these data
are widely used for analysis of
spending, but they could be used to open
more complex analyzes, and give meaning and
Content ... Gradually, users are more
numerous: KCE (Kennis Centrum Expertise), the IMA
(Intermutualiste Agency) develop analyzes
targeted, atlases, etc ... but it's still a work in
progress!
An: You work with personal data,
name, address, Social Security Number ...
What about confidentiality?
Anne-Laure: You sign a contract that you can not
ask questions that emerge from the field
staff, because even when the name and address
are removed and that the data are supposedly
anonymous, there is always the register number
National where you end date of birth,
gender, and can possibly afford
d'identifier the person. I find that we are not
13
Page 16
yet sufficiently vigilant about these privacy issues,
but this is changing. The awareness of the
how we should protect the data that is
use is more prevalent now.
An: And how do you POSITIONED you vis-à-vis the
Practice Standards is anyway
Based on your work? What is extraordinary,
obvious, is about suspect.
Anne-Laure: We are all the time confronted with the
normality, indeed. But this does not mean that
Exceptions are all out law. There are some years
we looked at the amount of average prescription
in drug per patient for GPs. The
doctor with more spending in Belgium is
someone who has a haemophiliac patient, who must take
many drugs that are very expensive, so it
happens to disproportionate budgets, which seem
abnormal but are completely legitimate!
Furthermore, our work is conditioned by the
context. Data and research related to the
Insurance system diseases as constituted
right now. The data is related to the act, and not
of hours worked, for example, as
England where the medical service is a free service.
It has disadvantages as well as advantages:
this allows also'connaître' reality acts
medical facts.
Year: Easy access to major bases
data from the institutions of State gives
thinking these days, that the activities of
14
Page 17
NSA control are revealed. What do you think
as an activist?
Anne-Laure: I kept an article to speak at
classes in the ERG. It is of a woman who sought
'cocotte minute' on Google, the same day as his
husband sought'sac to dos'. They had landed the FBI,
which is a hundred such descents the year. In short,
you can see everything in the data, but when you
are preparing a terrorist attack, you will not find
how to do it in Google, or purchase your tools
internet. You do it with a fake valid identity card
an internet cafe, or you go to a store
and you pay in cash!
That is to say, the data are not always
sufficient, they are not a complete picture of the
reality. Those who did not escape the NSA are not
terrorists! The control systems want
there is more room, but there will always be
non-covered areas.
The big problem for me, politically, more
the data analysis, the Patriot Act. There
possibility of detaining people without evidence is
much more serious than the continuous monitoring
data. But maybe it's my job to
deform me?
D'un other hand, I love Scotland because there are plenty
non-covered areas without roads access, not
on the map. At the same time it always creates
invisible spaces, also on the internet. I blog
with a password that nobody knows
15
Page 18
the address ...
Year: Do you have fantasies about
data that you would treat them well? Or
types of narration you could develop
with this?
Anne-Laure: I think all the new data
are always exciting. It's like when you open
a new book. After that, it is true that everything
linked to quantitative and qualitative. I have a friend who makes a
thesis in psychology and when I see how it uses
the analysis of texts, the analysis of the interviews for
try to find the subtexts of what
interviewees tell their story
personal ... You can sometimes use techniques
quantitative to reveal things that are not
always made clear, because they
make choices of words they do not know
necessarily that time.
When I was a summer camp counselor, one
was making reports of activities that adults
had with us. People write what they believe to be
an objective description, but when you read it, it is not
never an objective description. In fact, when you
talking about your life, what you say, what you do not say, the
words you use, you choose a formulation from
hundred possible. When I say, for example, I have a dog,
I will not say his name, how it is
arrived home, why I keep it, but I chose to
give that information there. So it is that it is
important to share this information with you, it
16
Page 19
which probably means that my dog is important,
and I try to share something with you ...
This friend identifies channels. She works with
texts work software that allow you to
reclassify the life course of the items they
tell in the recatégorisant. But also in this
place there, the person speaks in first person ...
After you compare between people too.
To answer your question, there is a very messy side
very messy in the data analysis. There is
always too much data, we never quite put
end to this ... So either it's come to make an analysis of
data on very localized data, things
tiny ... where you can go very in detail.
I often have a problem with telling stories
because there is question of right and wrong, of
the inversion and the subjective. As a statistician,
I try to raise the maximum Other information
of reality and world fauna as fair as possible about the aspect
the truth. I'm good to find the stories
others and bring them to light, provided
I am finally totally absent from the
narration. Or that I have this feeling, because
obviously this is not the case. I try to do
scientific and not personal choice.
And therefore, it would be a project on data ... of life
private, life to me. Not on the lives of every day, on
how you have walked, or how many letters you have
17
Page 20
read ... create information on things to keep by
example, the memory, the trace of the people, links.
Year: Memories?
Anne-Laure: Yes.
Year: How would you do that?
Anne-Laure: I do not know, that's why I do
do not d'ailleurs.
There is a d'objectivation side. When I was a teenager, I had
notebooks and I wrote, I wrote ... with this
based one, to keep the memory. And in fact, it has
no sense. The important things you do notes
not because they are all up and you get in
remember, even twenty years later. And you write stuff
genre, I was in love with this guy, but twenty years
after you re-read and you wonder: what was it?
But cons, which would take is to remember
precious things.
If you do not fictionnes your memories, once thou
get forgotten, they have no value. If you describe
objectively neutral, an important memory
when you saw him finally when you reread the ten
years later, it means nothing. This does not wake up
emotion.
Pereira Pretends to Tabucchi, for example, is a
novel written as a statement. This is
supposedly objective, neutral and policeman. This is a
story set in Lisbon during the period
Fascist. And Pereira claims he did not know this
18
Page 21
gentleman who made the attack, he had nothing to do
with him ... It's built for this purpose, but with the
d'écrire capacity of fiction. It's supposedly neutral
that is supposedly objective, but in fact, when you read it, you
Totally overwhelmed by what is written. It is very
sensitive, very strong. It is in the literature.
I wrote three poems in my life that are far from the
against reality but which manage to evoke the great
emotion that I lived when I wrote,
more than tons and tons of
comprehensive and detailed diaries!
An: And so how you would approach this project?
Anne-Laure: The question of words and data,
it is also a different representation of reality.
It would be a database which in itself would
uninteresting but, when the analysis, regenerates
the history and emotion. It's possible. It's back
all the facts, all the factual side, etc and objectives
that you could not not re-tell, revive
the story when thou analyzes. I thought of something
like that with timelines ... I can not write
the novel, but I can make the database
would remake the novel!
This is a question to see your behavior and your
being as a result of external things
(Family influence) and not as a choice, things
which découles and you you can replenish you, side
Cloning can also be.
19
Page 22
An: We do our own anyway selection
data.
Anne-Laure: Yes, but we do not all variables.
An: This one might imagine, is to choose
a particular situation that has marked you and go
looking for all the variables you
could have had, all the things that were
at that time offered, and all the choices you
could have done.
Anne-Laure: And then, what you've done, who made
you are moved you of a point to another.
20
Page 23
Constant Verlag, Brussels (2013).
Copyleft: This work is free, you can copy,
distribute, and modify it under the terms of the License
Free Art http://www.artlibre.org