The surprising seeds of a big-data revolution in healthcare

Collecting global health data was an imperfect science: Workers tramped through villages to knock on doors and ask questions, wrote the answers on paper forms, then input the data -- and from this gappy information, countries would make huge decisions. Data geek Joel Selanikio talks through the sea change in collecting health data in the past decade -- starting with the Palm Pilot and Hotmail, and now moving into the cloud. (Filmed at TEDxAustin.)

There's an old joke about a cop

who's walking his beat
in the middle of the night,

and he comes across a guy
under a street lamp

who's looking at the ground
and moving from side to side,

and the cop asks him what he's doing.

The guys says he's looking for his keys.

So the cop takes his time

and looks over and kind of
makes a little matrix

and looks for about two, three minutes.

No keys.

The cop says, "Are you sure?

Hey buddy, are you sure
you lost your keys here?"

And the guy says,
"No, actually I lost them

down at the other end of the street,

but the light is better here."

(Laughter)

There's a concept that people talk
about nowadays called "big data."

And what they're talking
about is all of the information

that we're generating
through our interaction

with and over the Internet,

everything from Facebook and Twitter

to music downloads, movies,
streaming, all this kind of stuff,

the live streaming of TED.

And the folks who work
with big data, for them,

they talk about that their biggest problem

is we have so much information.

The biggest problem is: how do we
organize all that information?

I can tell you that,
working in global health,

that is not our biggest problem.

Because for us, even though
the light is better on the Internet,

the data that would help us solve
the problems we're trying to solve

is not actually present on the Internet.

So we don't know, for example,

how many people right now
are being affected by disasters

or by conflict situations.

We don't know for, really,
basically, any of the clinics

in the developing world,

which ones have medicines
and which ones don't.

We have no idea of what
the supply chain is for those clinics.

We don't know -- and this is really
amazing to me -- we don't know

how many children were born --
or how many children there are --

in Bolivia or Botswana or Bhutan.

We don't know how many kids died last week

in any of those countries.

We don't know the needs
of the elderly, the mentally ill.

For all of these different
critically important problems

or critically important areas
that we want to solve problems in,

we basically know nothing at all.

And part of the reason
why we don't know anything at all

is that the information technology systems
that we use in global health

to find the data to solve
these problems is what you see here.

This is about a 5,000-year-old technology.

Some of you may have used it before.

It's kind of on its way out now,

but we still use it
for 99 percent of our stuff.

This is a paper form.

And what you're looking at is a paper form

in the hand of a Ministry of Health
nurse in Indonesia,

who is tramping out across the countryside

in Indonesia on, I'm sure,
a very hot and humid day,

and she is going to be knocking
on thousands of doors

over a period of weeks or months,

knocking on the doors and saying,

"Excuse me, we'd like to ask
you some questions.

Do you have any children?
Were your children vaccinated?"

Because the only way
we can actually find out

how many children were vaccinated
in the country of Indonesia,

what percentage were vaccinated,

is actually not on the Internet,
but by going out and knocking on doors,

sometimes tens of thousands of doors.

Sometimes it takes months to even years
to do something like this.

You know, a census of Indonesia

would probably take
two years to accomplish.

And the problem, of course,
with all of this

is that, with all those paper forms --

and I'm telling you, we have
paper forms for every possible thing:

We have paper forms
for vaccination surveys.

We have paper forms to track
people who come into clinics.

We have paper forms to track
drug supplies, blood supplies --

all these different paper forms
for many different topics,

they all have a single, common endpoint,

and the common endpoint
looks something like this.

And what we're looking
at here is a truckful of data.

This is the data from a single
vaccination coverage survey

in a single district
in the country of Zambia

from a few years ago,
that I participated in.

The only thing anyone
was trying to find out

is what percentage of Zambian
children are vaccinated,

and this is the data,
collected on paper over weeks,

from a single district,

which is something like a county
in the United States.

You can imagine that,
for the entire country of Zambia,

answering just that single question ...

looks something like this.

Truck after truck after truck,

filled with stack after stack
after stack of data.

And what makes it even worse
is that's just the beginning.

Because once you've collected
all that data,

of course, someone --
some unfortunate person --

is going to have to type that
into a computer.

When I was a graduate student,

I actually was that unfortunate
person sometimes.

I can tell you, I often wasn't
really paying attention.

I probably made a lot
of mistakes when I did it

that no one ever discovered,
so data quality goes down.

But eventually that data, hopefully,
gets typed into a computer,

and someone can begin to analyze it,

and once they have
an analysis and a report,

hopefully, then you can take
the results of that data collection

and use it to vaccinate children better.

Because if there's anything worse
in the field of global public health --

I don't know what's worse
than allowing children on this planet

to die of vaccine-preventable diseases --

diseases for which
the vaccine costs a dollar.

And millions of children die
of these diseases every year.

And the fact is, millions
is a gross estimate,

because we don't really know
how many kids die each year of this.

What makes it even more frustrating
is that the data-entry part,

the part that I used to do
as a grad student,

can take sometimes six months.

Sometimes it can take two years
to type that information into a computer,

And sometimes, actually not infrequently,

it actually never happens.

Now try and wrap your head
around that for a second.

You just had teams of hundreds of people.

They went out into the field
to answer a particular question.

You probably spent hundreds
of thousands of dollars

on fuel and photocopying and per diem.

And then for some reason, momentum is lost

or there's no money left,

and all of that comes to nothing,

because no one actually types it
into the computer at all.

The process just stops.

Happens all the time.

This is what we base
our decisions on in global health:

little data, old data, no data.

So back in 1995,

I began to think about ways
in which we could improve this process.

Now 1995 -- obviously,
that was quite a long time ago.

It kind of frightens me to think
of how long ago that was.

The top movie of the year
was "Die Hard with a Vengeance."

As you can see, Bruce Willis
had a lot more hair back then.

I was working in the Centers
for Disease Control

and I had a lot more
hair back then as well.

But to me, the most significant
thing that I saw in 1995

was this.

Hard for us to imagine, but in 1995,

this was the ultimate elite mobile device.

It wasn't an iPhone.
It wasn't a Galaxy phone.

It was a PalmPilot.

And when I saw the PalmPilot
for the first time, I thought,

"Why can't we put the forms
on these PalmPilots?

And go out into the field
just carrying one PalmPilot,

which can hold the capacity
of tens of thousands of paper forms?

Why don't we try to do that?

Because if we can do that,

if we can actually just collect
the data electronically, digitally,

from the very beginning,

we can just put a shortcut right
through that whole process

of typing, of having somebody type
that stuff into the computer.

We can skip straight to the analysis

and then straight to the use
of the data to actually save lives.

So that's what I began to do.

Working at CDC, I began to travel
to different programs around the world

and to train them in using
PalmPilots to do data collection,

instead of using paper.

And it actually worked great.

It worked exactly as well
as anybody would have predicted.

What do you know?

Digital data collection is actually
more efficient than collecting on paper.

While I was doing it, my business partner,

Rose, who's here with her husband,
Matthew, here in the audience,

Rose was out doing similar stuff
for the American Red Cross.

The problem was,
after a few years of doing that,

I realized -- I had been to maybe
six or seven programs --

and I thought, you know,
if I keep this up at this pace,

over my whole career,

maybe I'm going to go
to maybe 20 or 30 programs.

But the problem is, 20 or 30 programs,

like, training 20 or 30 programs
to use this technology,

that is a tiny drop in the bucket.

The demand for this, the need
for data to run better programs

just within health -- not to mention
all of the other fields

in developing countries -- is enormous.

There are millions and millions
and millions of programs,

millions of clinics
that need to track drugs,

millions of vaccine programs.

There are schools
that need to track attendance.

There are all these different things
for us to get the data that we need to do.

And I realized if I kept up
the way that I was doing,

I was basically hardly
going to make any impact

by the end of my career.

And so I began to rack my brain,

trying to think about, what
was the process that I was doing?

How was I training folks,

and what were the bottlenecks
and what were the obstacles

to doing it faster
and to doing it more efficiently?

And, unfortunately, after thinking
about this for some time,

I identified the main obstacle.

And the main obstacle, it turned out --

and this is a sad realization --

the main obstacle was me.

So what do I mean by that?

I had developed a process

whereby I was the center
of the universe of this technology.

If you wanted to use this technology,
you had to get in touch with me.

That means you had to know I existed.

Then you had to find the money
to pay for me to fly out to your country

and the money to pay for my hotel
and my per diem and my daily rate.

So you could be talking
about 10- or 20- or 30,000 dollars,

if I actually had the time
or it fit my schedule

and I wasn't on vacation.

The point is that anything,

any system that depends
on a single human being

or two or three or five human beings --

it just doesn't scale.

And this is a problem for which
we need to scale this technology,

and we need to scale it now.

And so I began to think of ways
in which I could basically

take myself out of the picture.

And, you know, I was thinking,

"How could I take myself
out of the picture?"

for quite some time.

I'd been trained that the way
you distribute technology

within international development

is always consultant-based.

It's always guys
that look pretty much like me,

flying from countries
that look pretty much like this

to other countries
with people with darker skin.

And you go out there,
and you spend money on airfare

and you spend time and you spend per diem

and you spend for a hotel
and all that stuff.

As far as I knew, that was the only way
you could distribute technology,

and I couldn't figure out a way around it.

But the miracle that happened --

I'm going to call it Hotmail for short.

You may not think of Hotmail
as being miraculous,

but for me it was miraculous,

because I noticed, just as I
was wrestling with this problem --

I was working in sub-Saharan
Africa, mostly, at the time --

I noticed that every sub-Saharan
African health worker

that I was working with
had a Hotmail account.

And it struck me: wait a minute --

I know the Hotmail people surely didn't
fly to the Ministry of Health in Kenya

to train people in how to use Hotmail.

So these guys are distributing technology,
getting software capacity out there,

but they're not actually
flying around the world.

I need to think about this more.

While I was thinking about it,

people started using even more
things like this, just as we were.

They started using LinkedIn and Flickr
and Gmail and Google Maps --

all these things.

Of course, all of these things
are cloud based

and don't require any training.

They don't require any programmers.

They don't require consultants.

Because the business model
for all these businesses

requires that something be so simple
we can use it ourselves,

with little or no training.

You just have to hear about it
and go to the website.

And so I thought, what would happen
if we built software

to do what I'd been consulting in?

Instead of training people
how to put forms onto mobile devices,

let's create software that lets them
do it themselves with no training

and without me being involved.

And that's exactly what we did.

So we created software called Magpi,
which has an online form creator.

No one has to speak to me,

you just have to hear about it
and go to the website.

You can create forms,
and once you've created the forms,

you push them to a variety
of common mobile phones.

Obviously, nowadays, we've moved
past PalmPilots to mobile phones.

And it doesn't have to be a smartphone,
it can be a basic phone,

like the phone on the right,
the basic Symbian phone

that's very common
in developing countries.

And the great part about this
is it's just like Hotmail.

It's cloud based,

and it doesn't require any training,
programming, consultants.

But there are some
additional benefits as well.

Now we knew when we built this system,

the whole point of it,
just like with the PalmPilots,

was that you'd be able to collect the data

and immediately upload
the data and get your data set.

But what we found, of course,
since it's already on a computer,

we can deliver instant maps
and analysis and graphing.

We can take a process that took two years

and compress that
down to the space of five minutes.

Unbelievable improvements in efficiency.

Cloud based, no training,
no consultants, no me.

And I told you that in the first few years

of trying to do this
the old-fashioned way,

going out to each country,

we probably trained about 1,000 people.

What happened after we did this?

In the second three years,

we had 14,000 people find the website,

sign up and start using it
to collect data:

data for disaster response,

Canadian pig farmers
tracking pig disease and pig herds,

people tracking drug supplies.

One of my favorite examples, the IRC,
International Rescue Committee,

they have a program
where semi-literate midwives,

using $10 mobile phones,

send a text message
using our software, once a week,

with the number of births
and the number of deaths,

which gives IRC something that no one
in global health has ever had:

a near-real-time system
of counting babies,

of knowing how many kids are born,

of knowing how many children
there are in Sierra Leone,

which is the country
where this is happening,

and knowing how many children die.

Physicians for Human Rights --

this is moving a little bit
outside the health field --

they're basically training people
to do rape exams in Congo,

where this is an epidemic,

a horrible epidemic,

and they're using our software
to document the evidence they find,

including photographically,

so that they can bring
the perpetrators to justice.

Camfed, another charity
based out of the UK --

Camfed pays girls' families
to keep them in school.

They understand this is the most
significant intervention they can make.

They used to track the disbursements,
the attendance, the grades, on paper.

The turnaround time between a teacher
writing down grades or attendance

and getting that into a report
was about two to three years.

Now it's real time.

And because this is such a low-cost
system and based in the cloud,

it costs, for the entire five countries
that Camfed runs this in,

with tens of thousands of girls,

the whole cost combined
is 10,000 dollars a year.

That's less than I used to get

just traveling out for two weeks
to do a consultation.

So I told you before that when
we were doing it the old-fashioned way,

I realized all of our work was really
adding up to just a drop in the bucket --

10, 20, 30 different programs.

We've made a lot of progress,

but I recognize that right now,

even the work that we've done
with 14,000 people using this

is still a drop in the bucket.

But something's changed,
and I think it should be obvious.

What's changed now is,

instead of having a program
in which we're scaling at such a slow rate

that we can never reach
all the people who need us,

we've made it unnecessary
for people to get reached by us.

We've created a tool
that lets programs keep kids in school,

track the number of babies that are born
and the number of babies that die,

catch criminals and successfully
prosecute them --

to do all these different things
to learn more about what's going on,

to understand more,

to see more ...

and to save lives and improve lives.

Thank you.

(Applause)