A four-year-old’s final gift became medical knowledge. Her story raises a harder question: Why does healthcare still discover so many problems after they become expensive?
“To become the best place in the world for innovators to build and test AI, for healthcare professionals to use it, and for patients to benefit from it in their care.”
— Lawrence Tallon, Chief Executive, UK Medicines and Healthcare products Regulatory Agency (MHRA), September 2026
There is a four-year-old girl named Liliana Gjorgjijoska whose story should make every physician, healthcare executive, researcher, and technology founder stop for a moment.
Not because she became a famous patient.
She didn’t.
Not because she lived a long life.
She didn’t.
And not because her cancer was cured.
It wasn’t.
Liliana died from diffuse intrinsic pontine glioma, or DIPG, a devastating childhood brain cancer.
But something happened after her death that raises a much larger question about healthcare:
What happens when we stop treating a patient’s experience as the end of a story—and start treating it as the beginning of knowledge for the next patient?
That question reaches far beyond cancer research.
It reaches into the way we document.
The way we collect information.
The way we communicate.
The way we build clinical workflows.
The way we design medical billing systems.
And, increasingly, the way we think about artificial intelligence.
Because healthcare has a strange habit.
We collect enormous amounts of information about patients.
Then we spend enormous amounts of money trying to reconstruct what happened.
Sometimes months later.
Sometimes after the claim has already been denied.
Sometimes after the patient has already left.
Sometimes after the physician has already forgotten the details.
And then we call this innovation.
Maybe it isn’t.
Maybe we’re just getting better at cleaning up the mess.
The Girl Behind the Data
In August 2009, Rachael Gjorgjijoska took her three-year-old daughter, Liliana, to The Children’s Hospital at Westmead in Sydney.
Liliana had been falling.
At first, it looked like something ordinary.
An ear infection.
Antibiotics had been prescribed.
But the falls continued.
Then came the MRI.
And the conversation no parent ever wants to have.
Liliana had an inoperable brain tumor.
The diagnosis was diffuse intrinsic pontine glioma, commonly known as DIPG.
The tumor was located in the pons, a critical part of the brainstem.
Surgery was essentially not an option.
The diagnosis carried an extraordinarily poor prognosis.
For Rachael and Liliana’s father, Zoran Gjorgjijoska, the medical language could not possibly capture the human reality.
Liliana was not a tumor.
She was a little girl who loved dancing.
She loved princess dresses.
She had long hair she called her “princess hair.”
She carried little handbags filled with coins.
Her mother remembers that Liliana would put a dress over her tracksuit if someone told her she couldn’t wear one in winter.
She insisted on being called Princess Liliana.
This matters.
Because medical records tend to compress people.
A diagnosis becomes a code.
A symptom becomes a field.
A procedure becomes a charge.
A patient becomes a chart number.
And eventually, a chart becomes a claim.
But there was a child inside that chart.
There always is.
The Problem Was Not Just That Doctors Lacked a Treatment
Here is where Liliana’s story becomes much more interesting.
For years, DIPG research faced a brutal problem.
Researchers needed tumor tissue to understand the disease.
But the tumor was located in an extraordinarily dangerous part of the brain.
Obtaining tissue was difficult and risky.
So researchers had very little biological material to study.
And without samples, there was little research.
Without research, there were few treatments.
Without treatments, families were told there was little that could be done.
This is an uncomfortable lesson.
Sometimes healthcare does not fail because clinicians don’t care.
Sometimes healthcare fails because the information needed to improve the system doesn’t exist in a usable form.
That is a very different problem.
And it has enormous implications.
Because we often assume that healthcare’s biggest problem is lack of information.
I would argue that is only half true.
The bigger problem is:
We frequently fail to capture, structure, connect, and reuse the information we already have.
Then a Mother Asked an Unusual Question
As Liliana’s condition deteriorated, Rachael began thinking about what could happen after her daughter died.
Organ donation was discussed.
But Liliana’s organs were not suitable for transplantation because of the treatment she had received.
Rachael asked about donating Liliana’s tumor tissue for research.
The reaction was essentially disbelief.
This wasn’t routine in Australia at the time.
There were legal questions.
Ethical questions.
Institutional questions.
Operational questions.
Who had authority?
How would the tissue be collected?
Where would it go?
Who would approve it?
Could a child’s postmortem tissue be used for research?
The system had a problem.
But it was not simply a scientific problem.
It was a workflow problem.
It was a data problem.
It was an ethics problem.
It was a coordination problem.
And it was a human problem.
Eventually, doctors, researchers, parents, and ethics committees worked through the barriers.
Rachael and Zoran agreed to donate Liliana’s tumor.
Half was stored in Australia.
Half was sent to St Jude Children’s Research Hospital in Memphis.
Liliana’s tumor became the first DIPG tumor sample preserved in an Australian biobank.
And that sample helped establish the foundation for Australia’s first dedicated DIPG research program in 2013 led by Professor David Ziegler and his team.
Think about that.
A four-year-old who could not be saved became part of a system designed to help save someone else.
That is not a sentimental metaphor.
It is how science works.
One patient’s experience becomes evidence.
Evidence becomes knowledge.
Knowledge becomes a protocol.
A protocol becomes a treatment.
And eventually, perhaps, a child gets more time.
The Healthcare Industry Loves the Last Step
This is where I want to be deliberately provocative.
Healthcare loves talking about the last step.
The algorithm.
The dashboard.
The AI model.
The denial management platform.
The coding assistant.
The revenue-cycle automation engine.
The clinical decision-support tool.
The new software.
The new workflow.
The new app.
We love the shiny object.
But Liliana’s story reminds us that the breakthrough may happen much earlier.
Before the algorithm.
Before the dashboard.
Before the workflow.
Before the claim.
Before the denial.
Before the treatment failure.
Before the research paper.
There is information.
And the quality of that information determines what happens downstream.
Maybe Medical Billing Has the Same Problem
Consider the typical medical billing conversation.
A claim is denied.
The organization asks:
Why was the claim denied?
The biller investigates.
Was the code wrong?
Was documentation incomplete?
Was authorization missing?
Was eligibility incorrect?
Was the payer requirement different?
Was medical necessity unclear?
Was the modifier missing?
Was the patient’s insurance information outdated?
Was the procedure performed differently than what was originally anticipated?
Was something documented late?
Was something documented somewhere else?
The answer is often complicated.
And then we build another system to manage the complication.
That’s the part that bothers me.
We have created an enormous industry around recovering from information problems after they become expensive.
We call it revenue cycle management.
But sometimes it is really:
revenue cycle archaeology.
Someone digs through the chart trying to reconstruct what everyone should have known earlier.
And then we congratulate ourselves because AI can do the archaeology faster.
That’s progress.
But it isn’t necessarily prevention.
The Contrarian Question
What if the goal of AI in medical billing shouldn’t be:
“How can we process claims faster?”
What if the better question is:
“Why did the claim become difficult in the first place?”
Those are very different questions.
The first produces automation.
The second produces intelligence.
And intelligence changes the workflow upstream.
The Liliana Principle
Liliana’s story suggests a simple principle:
Don’t wait until the end of the process to learn what the beginning was trying to tell you.
In cancer research, the tumor tissue was not merely biological material.
It was information.
It contained clues.
Those clues could be analyzed.
Those analyses could be compared.
Those comparisons could generate hypotheses.
Those hypotheses could become experiments.
Experiments could become treatments.
The sample moved information forward.
Now look at a medical practice.
A patient presents.
The physician asks questions.
The nurse gathers information.
The medical assistant records information.
The physician documents.
Orders are placed.
Procedures happen.
Diagnoses are established.
Authorizations may be requested.
The patient is discharged.
The claim eventually goes to the payer.
And then someone discovers:
We needed one more piece of information.
That missing piece may have been visible hours, days, or weeks earlier.
But nobody connected it.
That’s the problem.
We Don’t Have an AI Problem
We have a context problem.
AI is exceptionally good at processing information.
But garbage data does not become intelligent simply because you put a large language model on top of it.
It becomes very sophisticated garbage.
Healthcare has spent years creating systems that generate enormous quantities of data.
But more data does not automatically create better decisions.
Sometimes it creates more noise.
More clicks.
More alerts.
More fields.
More dashboards.
More notifications.
More inboxes.
More documentation.
More administrative work.
And eventually:
more burnout.
The physician did not become less intelligent.
The system became more complicated.
The Documentation Paradox
Here is another uncomfortable question.
We tell physicians:
“Document more.”
Then we complain that physicians spend too much time documenting.
Then we build ambient AI to document the conversation.
Then we discover the generated note is too long.
Then we build another AI to summarize the AI-generated note.
Then another tool checks coding.
Then another checks compliance.
Then another checks the claim.
At some point, we should probably ask:
What exactly are we optimizing?
More documentation?
Or better information?
Those are not the same thing.
A 14-page note is not necessarily better than a two-page note.
A 200-field intake form is not necessarily better than a thoughtful five-minute conversation.
A mountain of data is not necessarily intelligence.
Context is intelligence.
The Patient Knows More Than the Database Thinks
Patients constantly provide information that does not fit neatly into structured fields.
They say:
“My symptoms started after this.”
“I stopped taking it because it made me sick.”
“I couldn’t get the medication.”
“My insurance changed.”
“I was told the authorization was approved.”
“I went to the emergency room.”
“I couldn’t afford the test.”
“I didn’t understand what the doctor meant.”
“I thought the appointment was covered.”
These statements may sound ordinary.
Operationally, they can be enormous.
They may affect:
- eligibility
- authorization
- medical necessity
- documentation
- coding
- scheduling
- treatment adherence
- utilization
- follow-up
- claims
- denials
- revenue
Yet many systems treat these conversations as secondary.
That is backwards.
The Industry’s Favorite Word: Automation
Automation is wonderful.
Until it automates the wrong thing.
Automating a bad workflow does not eliminate the workflow.
It accelerates it.
That’s why healthcare needs to become much more skeptical about the word automation.
The better question is:
What should never have required human cleanup in the first place?
That is a more interesting question.
And it is much harder.
The Real Opportunity for AI
AI should not be used primarily to make physicians better billers.
That is a category mistake.
Physicians went to medical school to diagnose and treat patients.
They should not need to become amateur claims specialists.
The opportunity is to give physicians and their teams better information earlier.
Imagine a system that recognizes:
- eligibility risk before the encounter
- authorization risk before the procedure
- documentation gaps while they are still fixable
- coding ambiguity before claim submission
- payer-specific requirements before they become denials
- missing clinical context before someone has to chase it
- patterns across previous claims before the same mistake repeats
That is fundamentally different from denial management.
Denial management asks:
“How do we fix this?”
Upstream intelligence asks:
“How do we keep this from becoming broken?”
That is where the real leverage is.
Data → Context → Workflow → Action → Revenue
This is the sequence I believe healthcare technology should increasingly follow.
Data.
What happened?
Context.
Why does it matter?
Workflow.
Who needs to know?
Action.
What should happen next?
Revenue.
What financial consequence follows?
Most systems start at the end.
They look at the claim.
They analyze the denial.
They send an alert.
They assign a work queue.
They open a ticket.
Then someone starts investigating.
That’s expensive.
And strangely, we call it efficient because a computer is involved.
A Computerized Mess Is Still a Mess
There is a joke hidden in healthcare technology:
We took paperwork.
Digitized it.
Added dashboards.
Added APIs.
Added AI.
And somehow created more work.
The lesson isn’t that technology failed.
The lesson is that technology cannot compensate indefinitely for poorly structured processes.
At some point, the workflow itself has to change.
What Liliana’s Story Teaches About Data
The most important thing about Liliana’s tumor wasn’t that it was stored.
It was that it became usable.
It could be studied.
Compared.
Analyzed.
Shared.
Reproduced.
Connected to other samples.
Used to generate new knowledge.
That’s what makes data valuable.
Not collection.
Reuse.
Healthcare should ask the same question about operational data.
Can the information collected today help prevent tomorrow’s problem?
Can a physician’s documentation improve the next authorization?
Can an authorization outcome improve future scheduling?
Can a denial teach the system something before the next claim?
Can a patient’s conversation identify an operational risk before it becomes a financial problem?
If not, we aren’t building a learning system.
We’re building a filing cabinet.
A very expensive filing cabinet.
Three Names Healthcare Should Remember
Liliana was not the only child whose story changed this field.
Professor Matthew Dun was already a biomedical researcher when his daughter Josephine Laura Dun, known as Josie, was diagnosed with DIPG at two years old in 2018.
Her parents, Dr Phoebe Hindley and Matthew Dun, entered the world of DIPG not simply as clinicians and scientists but as parents.
Josephine died at four, 22 months after diagnosis.
Matthew Dun subsequently continued research into the biology and treatment of these tumors.
Another parent, Ren Pedersen, had already experienced the devastating loss of his daughter Amy, who was diagnosed in 2007 and died in 2009 at age nine.
Ren became an advocate for research after repeatedly encountering the same wall:
There was little to offer.
No meaningful treatment.
No easy answer.
And no acceptable reason to stop looking.
These families did something remarkable.
They transformed grief into infrastructure.
Not just awareness.
Not just fundraising.
Infrastructure.
Samples.
Research programs.
Advocacy.
Clinical trials.
Scientific collaboration.
Knowledge.
That is what progress looks like before it becomes a headline.
The Lesson for Physicians Is Not “Collect More Data”
That would be too easy.
The lesson is:
Collect the right information, at the right moment, in a form that can actually be used.
That distinction matters enormously.
A practice can have a sophisticated EHR and still have poor operational intelligence.
It can have thousands of fields and still lack context.
It can have an analytics dashboard and still discover problems too late.
It can have AI and still be reactive.
This is why I believe the next generation of healthcare technology will move away from simply adding intelligence to existing workflows.
It will redesign the information flow itself.
What Should Physicians Actually Do?
Start smaller than AI.
Much smaller.
Ask five questions.
1. Where do problems first appear?
Not where you notice them.
Where do they actually begin?
A denial might be discovered in billing.
But it may have originated at scheduling.
Or eligibility.
Or authorization.
Or documentation.
Or the patient’s conversation.
Find the first point of failure.
2. What information was available earlier?
This is the most important question.
When a claim is denied, look backward.
Was the missing information already somewhere in the organization?
If yes, you have an information-flow problem.
3. Who had the information?
Was it:
- the patient?
- scheduler?
- medical assistant?
- nurse?
- physician?
- biller?
- payer?
- referral coordinator?
If someone already knew the answer, why didn’t the next person?
That question often reveals more than another dashboard.
4. When should the system have acted?
Timing matters.
The same information can be useful or useless depending on when it appears.
A missing authorization discovered before the procedure is an operational signal.
The same missing authorization discovered after the claim is denied is an administrative problem.
Same fact.
Different timing.
Different cost.
5. Can the next patient benefit?
This is the Liliana test.
If your organization experiences a problem today, does tomorrow’s patient receive a better system?
Or does tomorrow’s patient simply encounter the same problem with a new ticket number?
The 30-Day “Stop Fixing the Same Problem” Audit
Here is a practical exercise for a physician-owned practice.
Don’t buy another software product yet.
Don’t launch another AI pilot.
Don’t build another dashboard.
For 30 days, track the recurring operational problems.
Create five columns:
Problem | Where discovered | Where it began | Information missing | Could it have been prevented?
For example:
Problem: Claim denied.
Where discovered: Billing.
Where it began: Authorization workflow.
Information missing: Procedure-specific payer requirement.
Could it have been prevented? Yes.
That is more useful than simply reporting:
“We had 43 denials.”
The number tells you what happened.
The chain tells you why.
Measure Prevention, Not Just Recovery
Healthcare organizations love recovery metrics.
They should.
But recovery is not prevention.
Track:
- first-pass claim acceptance
- denial rate
- preventable denial rate
- days to identify missing information
- days to resolve operational exceptions
- authorization turnaround
- eligibility error rate
- documentation clarification frequency
- repeated denial categories
- rework hours
- staff touches per claim
- physician time spent on administrative clarification
- percentage of problems detected before the encounter
- percentage of problems detected before claim submission
Then add one metric that is rarely discussed:
How often did the organization solve the same problem twice?
That number is revealing.
If the same denial occurs 100 times and the organization treats each one as a separate event, it does not have 100 problems.
It has one system problem repeated 100 times.
Myth Buster
Myth: More documentation means better billing.
Reality: Better documentation means information that is clinically accurate, relevant, timely, and usable.
More words are not necessarily better information.
Myth: AI eliminates administrative work.
Reality: AI eliminates administrative work only when it changes the work rather than adding another layer around it.
Myth: Denials are primarily a billing department problem.
Reality: Many denials originate upstream.
Billing is often where the problem becomes visible.
That does not mean billing caused it.
Myth: More data creates better AI.
Reality: Better structured, contextualized, relevant data creates better conditions for AI.
Myth: Physicians need to document more.
Reality: Physicians need to capture the information necessary for care, compliance, communication, and downstream operations without creating unnecessary cognitive burden.
Myth: Automation means removing humans.
Reality: The best healthcare automation often removes humans from repetitive work while keeping humans involved where judgment, empathy, exceptions, and accountability matter.
The Most Dangerous Healthcare AI Is Not the AI That Makes a Mistake
It is the AI that makes the wrong workflow faster.
That distinction deserves more attention.
A model can be accurate.
A workflow can still be wrong.
A prediction can be statistically impressive.
A process can still create unnecessary work.
A dashboard can be beautiful.
A practice can still lose money.
Technology evaluation therefore needs another dimension:
What happens to the system around the AI?
Not merely:
How accurate is the model?
The Human-in-the-Loop Is Not a Failure
There is a tendency in technology to treat human involvement as evidence that automation is incomplete.
Healthcare should resist that thinking.
Medicine is full of edge cases.
Patients are not manufacturing components.
A physician may recognize something that a model cannot.
A nurse may notice something that a database cannot.
A patient may explain something that no checkbox captured.
The goal should not be:
Human out.
The goal should be:
Human attention where human judgment matters.
AI should handle repetition.
Humans should handle meaning.
And the system should make the handoff between the two intelligent.
The Ethical Question
Liliana’s story also raises an important ethical issue.
Rachael and Zoran did not simply hand over data.
They entrusted something profoundly personal to researchers.
That creates a responsibility.
Whenever healthcare organizations collect patient information, the question should not simply be:
“Can we use this?”
It should also be:
“Should we use this?”
And:
“Can we explain how it will be used?”
And:
“Does the patient benefit?”
And:
“What safeguards exist?”
This becomes even more important as AI systems process clinical conversations, records, claims, images, and operational data.
The fact that something is technically possible does not make it ethically appropriate.
Healthcare technology must maintain:
- privacy
- security
- informed consent where appropriate
- minimum necessary use
- access controls
- auditability
- transparency
- human oversight
- clinical accountability
The fastest system is not necessarily the best system.
The cheapest system is not necessarily the best system.
And the most automated system certainly isn’t automatically the best system.
The Business Case Is Surprisingly Simple
Physician-owned practices do not need another abstract lecture about digital transformation.
They need fewer headaches.
Fewer denials.
Less rework.
Less chasing.
Less time spent clarifying information that should have been available.
Faster payment.
More predictable revenue.
More time for patients.
That’s the economic case.
But there is another case.
Attention is a clinical resource.
Every unnecessary administrative task consumes someone’s attention.
The scheduler.
The medical assistant.
The nurse.
The physician.
The biller.
The practice manager.
The patient.
Multiply a five-minute interruption across hundreds or thousands of encounters.
Suddenly the “small” inefficiency is a business problem.
Then a workforce problem.
Then a patient-experience problem.
Then a clinical problem.
The boundaries are artificial.
This Is Why I Keep Coming Back to “Upstream”
The downstream healthcare economy is enormous.
Claims.
Coding.
Denials.
Appeals.
Collections.
Prior authorization.
Revenue cycle.
Analytics.
Audits.
Compliance.
But downstream complexity often reflects upstream variability.
If the information entering the system is incomplete, inconsistent, delayed, or disconnected, every downstream system must compensate.
That creates a peculiar business model:
We make money solving problems created by other parts of the system.
There is nothing inherently wrong with that.
But from an innovation perspective, it should make us uncomfortable.
Because the biggest opportunity may not be to become better at fixing the problem.
It may be to reduce the number of times the problem happens.
What OnnX Is Trying to Explore
This is the thinking behind what I am building with OnnX.
Not another billing dashboard.
Not another software layer that tells physicians they have 37 things to fix.
And certainly not another system that assumes the physician should become a billing expert.
The larger question is:
Can intelligent technology help a practice recognize the signals that create downstream revenue problems before those problems become claims, denials, rework, and lost revenue?
That means starting with conversation and context.
Understanding how the practice actually operates.
Identifying where information gets lost.
Finding recurring patterns.
Connecting clinical and operational signals.
And using AI where it genuinely reduces friction.
The objective is not to make physicians better administrators.
It is to make the system around physicians smarter.
The Industry May Be Looking at the Wrong Dashboard
Imagine two practices.
Practice A celebrates because it reduced average denial resolution time from 12 days to 7.
Practice B reduces the number of preventable denials by 40%.
Which practice is more innovative?
Most dashboards will celebrate Practice A.
I would argue Practice B is doing something more important.
Practice A became better at recovery.
Practice B reduced the need for recovery.
That is the difference between optimization and redesign.
What Would a Truly Learning Practice Look Like?
It would have a memory.
Not merely a database.
A memory.
When something goes wrong, the organization would ask:
What happened?
Why?
Where did the signal first appear?
Who knew?
When did they know?
What action should have happened?
What prevented that action?
Can the system recognize this pattern next time?
Can the workflow change?
Can the patient experience improve?
Can the physician avoid another interruption?
Can the practice prevent another denial?
That is a learning organization.
And that is ultimately what Liliana’s story represents.
Her experience did not simply get stored.
It became useful.
Three Questions Every Practice Owner Should Ask
If you own or operate a medical practice, ask these this week.
Question 1
What problem does my billing department solve every week that another department could have prevented?
Don’t blame billing.
Follow the information.
Question 2
What do we repeatedly ask patients or physicians for because our system failed to capture it the first time?
Repetition is a clue.
Question 3
When something goes wrong, does our system learn—or does a person simply fix it?
That question separates a workflow from a learning system.
A More Useful Definition of AI
Maybe we should stop defining AI by what it can generate.
Generate a note.
Generate a code.
Generate an appeal.
Generate a summary.
Generate a response.
Generate a report.
Instead ask:
What friction disappeared because the system understood what was happening earlier?
That is a much more useful definition.
AI should not merely produce more output.
It should reduce unnecessary work.
The Future May Not Be More Automation
This may sound strange coming from someone building an AI company.
But I don’t think healthcare’s future is simply:
More AI.
I think it is:
Better information flow.
AI will be part of that.
But AI is not the strategy.
It is a capability.
The strategy is building healthcare systems that can understand what is happening early enough to do something useful about it.
Why Liliana’s Story Matters to a Medical Practice Owner
You might be wondering:
What does a child with DIPG have to do with a denied claim?
A lot.
Not because the situations are equivalent.
They are not.
The emotional stakes are obviously incomparable.
But the underlying lesson is remarkably similar.
Liliana’s doctors and researchers needed information that did not exist in usable form.
Her mother helped change that.
The tumor became knowledge.
Knowledge became research.
Research became new possibilities.
And those possibilities are now helping children who came after her.
That is the healthcare system at its best.
Not merely treating today’s patient.
Learning from today’s patient so tomorrow’s patient has a better chance.
One Family. One Sample. A Different Future.
Rachael Gjorgjijoska could not save Liliana.
No technology can change that fact.
No AI model can rewrite those years.
No research program can give her daughter back.
But Rachael could make one decision.
She could ask whether something from Liliana’s life could help another child.
That decision mattered.
Professor David Ziegler and other researchers could then build on it.
Professor Matthew Dun could bring his own experience as Josephine Laura Dun’s father into the research community.
Dr Phoebe Hindley could stand alongside him.
Ren Pedersen could turn the loss of his daughter Amy into advocacy.
Researchers could study the tissue.
Scientists could identify biological mechanisms.
Clinical researchers could test therapies.
The system could learn.
Slowly.
Painfully.
Imperfectly.
But it could learn.
That is what healthcare is supposed to do.
We Should Expect More From Our Data
Healthcare does not have a shortage of data.
We have an abundance of it.
The shortage is usable context.
We need information that arrives:
- early enough
- accurately enough
- clearly enough
- securely enough
- consistently enough
- contextually enough
to support action.
That’s the standard.
Not “Did we capture something?”
But:
“Did we capture something useful?”
Frequently Asked Questions
What is DIPG?
DIPG, or diffuse intrinsic pontine glioma, is an aggressive childhood brain tumor arising in the pons, an important region of the brainstem. It is now generally discussed within the broader category of diffuse midline glioma.
Why was tumor tissue so important?
Researchers need biological material to study the molecular characteristics of cancer, test therapies, develop models, and investigate potential treatment strategies. For many years, the location of DIPG made obtaining tumor tissue exceptionally difficult.
What did Liliana’s tissue donation change?
Liliana’s tumor became the first DIPG tumor sample preserved in an Australian biobank. That donation helped establish the foundation for Australia’s dedicated DIPG research effort and enabled subsequent scientific work.
Does this mean DIPG has been cured?
No. Progress has been significant, but DIPG/DMG remains a devastating disease and a cure remains an enormous scientific challenge.
What does this have to do with medical billing?
The connection is not the disease itself. It is the principle of learning from patient experience.
In medical billing, organizations frequently discover information problems only after they become denials, delays, rework, or lost revenue. The opportunity is to capture and use relevant information earlier.
Does better AI automatically solve the problem?
No.
AI can process information extremely well, but AI cannot compensate indefinitely for poor information architecture, fragmented workflows, missing context, or badly designed processes.
Should physicians document more?
Not necessarily.
The goal should be better information, not simply more information.
Should every billing problem be automated?
No.
Some problems require human judgment.
The objective should be to automate repetitive work while preserving human involvement for exceptions, clinical judgment, patient communication, ethics, and accountability.
What should a practice measure?
Start with preventable problems.
Track denial rate, first-pass acceptance, rework, authorization delays, documentation clarification, staff touches, physician administrative time, and—most importantly—the percentage of problems detected early enough to prevent them.
What is the biggest opportunity?
Move upstream.
Instead of asking only how to fix a denial, ask why the denial became possible in the first place.
That question can lead to better workflows, better data, better technology, and ultimately better economics.
Three Actions to Take Now
First: Follow one recurring denial backward to its first point of failure.
Do not stop when you find the billing error.
Keep going.
Second: Identify one piece of information your staff repeatedly re-enters, requests, or clarifies.
That is probably a workflow signal.
Third: Ask whether your technology is helping your practice prevent problems—or merely helping it process them faster.
That question can change an entire technology roadmap.
The Bigger Healthcare Lesson
Liliana’s story is ultimately not a story about a tumor.
It is a story about what happens when human experience becomes usable knowledge.
That principle sits at the heart of modern medicine.
Clinical research depends on it.
Quality improvement depends on it.
Public health depends on it.
Medical education depends on it.
And medical operations depend on it.
Yet our administrative systems frequently treat every problem as an isolated event.
A denial is a denial.
An authorization failure is an authorization failure.
A documentation query is a documentation query.
A scheduling mistake is a scheduling mistake.
Then someone fixes it.
And everyone moves on.
Until it happens again.
And again.
And again.
That is not learning.
That’s repetition with better software.
The Provocative Question
Perhaps the biggest question for healthcare technology isn’t:
“How intelligent can AI become?”
Perhaps it is:
“How much intelligence are we wasting because our systems fail to capture what humans already know?”
Think about that.
The patient told someone.
The nurse noticed.
The physician understood.
The scheduler saw something.
The biller discovered something.
The payer responded.
But the organization never connected the dots.
That is not a lack of intelligence.
It is a failure of information architecture.
And that is a problem technology should be able to help solve.
Final Thoughts
Liliana Gjorgjijoska lived only four years.
But the information that came from her life continues to move.
Her tumor became a research sample.
The sample became part of a biobank.
The biobank supported research.
Research informed new understanding.
New understanding supported clinical trials.
Clinical trials created new possibilities.
And those possibilities may give another child something Liliana could not have:
more time.
That is an extraordinary definition of legacy.
But it is also a warning.
Healthcare should not wait for tragedy to discover the value of information.
We should build systems that learn continuously.
From every encounter.
Every complication.
Every denial.
Every authorization.
Every recovery.
Every failure.
Every success.
Every patient conversation.
The future of healthcare will not be created simply by collecting more data.
It will be created by turning experience into knowledge—and knowledge into better action.
Data → Context → Workflow → Action → Revenue.
That is the opportunity.
And perhaps the most important question for every healthcare organization is this:
When the next patient walks through your door, will your system know more because of the patient who came before?
If the answer is no, we have work to do.
About the Author
Daniel Cham, MD is a physician, healthcare strategist, and founder of OnnX, an AI-powered medical billing SaaS focused on helping physician-owned and independent practices reduce administrative friction and improve revenue-cycle performance.
His work explores the intersection of clinical workflows, healthcare operations, artificial intelligence, and medical billing—with a particular focus on moving healthcare technology upstream, where better information can prevent downstream problems.
The goal is simple: better information, better workflows, better outcomes.
Connect with Dr. Daniel Cham on LinkedIn
Disclaimer
This article is for educational and informational purposes only. It does not constitute medical, legal, financial, billing, coding, compliance, or investment advice. Clinical and operational decisions should be made by qualified professionals based on the specific circumstances of each patient and organization.
The discussion of Liliana Gjorgjijoska, her family, Josephine Laura Dun, Amy, and the researchers and advocates involved in DIPG research is based on publicly reported information. The families’ experiences should not be interpreted as representative of every patient or family affected by DIPG or childhood cancer.
Continue the Conversation
Healthcare does not improve because we have more technology.
It improves when technology helps people see something earlier, understand it better, and act differently.
What is one recurring problem in your practice that everyone has learned to “work around”—but nobody has seriously tried to prevent?
That may be where the next breakthrough is hiding.
If this perspective resonates with you, share it with a physician, practice owner, operator, researcher, or healthcare technology leader who should be part of the conversation.
Knowledge drives progress.
Start your journey here.
Explore more perspectives on healthcare operations, physician entrepreneurship, medical technology, revenue-cycle management and healthcare innovation:
Listen to the podcast on Spotify
Knowledge creates leverage.
PS
I am developing practical resources around AI, medical billing, upstream data quality, and workflow intelligence for physician-owned and independent practices.
A free resource is available in the Featured section of my LinkedIn profile.
If you found this perspective useful, repost the article and add your own experience.
The most valuable healthcare intelligence may already be sitting inside your organization.
We just haven’t learned how to use it yet.
Further Reading
- The Australian — “How a girl’s final gift changed fight against incurable childhood brain cancer DIPG”
The September 11, 2026 feature provides the detailed account of Liliana Gjorgjijoska, Rachael Gjorgjijoska, Zoran Gjorgjijoska, Professor David Ziegler, Professor Matt Dun, Ren Pedersen, and the development of DIPG research in Australia. - RUN DIPG — Josephine Laura Dun and the Dun family story
RUN DIPG documents Josephine Laura Dun’s diagnosis, her parents Dr Phoebe Hindley and Matt Dun, and the family’s continuing work toward a cure. - Children’s Cancer Institute — DIPG research and biobanking
Research and institutional information describing the importance of tumor samples, biobanking, and the scientific work that followed early DIPG tissue donations.
Three Sentences Worth Remembering
Don’t automate a broken workflow. Understand it first.
Don’t measure only how quickly you recover from failure. Measure how often you prevent it.
Don’t ask only what your data says. Ask what your organization should have known sooner.
#HealthcareAI #HealthcareInnovation #MedicalBilling #RevenueCycleManagement #HealthTech #PhysicianLeadership #HealthcareData #AIinHealthcare #DigitalHealth #HealthcareTransformation #PhysicianOwnedPractices #HealthcareEntrepreneurship #PatientCenteredCare #ClinicalInnovation #OnnX
Leave a Reply