Borrowed Certainty

Don't wait. Sign up for The Hustle to unlock our side hustle database. Unlike generic "start a blog" advice, we've curated 100 actual business ideas with real earning potential, startup costs, and time requirements. Join 1.5M professionals getting smarter about business daily and launch your next money-making venture.

Get the guide

Prefer to listen in your podcast app? Apple Podcasts · Spotify

I followed a link in the comments of a LinkedIn post recently.

It led to a maternal health technology company doing work I found genuinely interesting, the kind of innovation I want to see succeed: technology with real potential to improve access and expand what smaller, independent practices are able to offer.

So I went to the website. I read about the technology, then started reading the blog.

One post referenced recent research on gestational diabetes. Something about the way the findings were described caught my attention, so I clicked the reference. It wasn’t the research. It was an article about the research. So I found the original study. Then I put all three next to each other.

The research. The article reporting on the research. The company’s blog reporting on the article reporting on the research.

The distance between the first and the third was remarkable.

Demographic terminology had changed. Findings had acquired explanations. Things the researchers had been careful not to conclude were now stated with more certainty than they’d earned. By the time the research reached the company blog, some of what I was reading was no longer what the original researchers had actually said at all.

No single leap was enormous, and that may be the most important part. The information had simply drifted. Now most likely this is a result of AI writing unchecked by human eyes and a human brain.

But I messaged the company directly. I don’t need to name them here. I believe in the idea of what they’re building, and I hope they succeed. And this isn’t really about them.

The problem isn’t that the need isn’t real.

Maternal health disparities are real. Access problems are real (although how we define access is worth digging into). Independent practices need better technology. Our maternal health infrastructure has enormous gaps, and there are companies building things that may genuinely improve care.

So does it matter if, somewhere between the research paper and the investor deck, the language gets a little loose?

I think it does.

A researcher spends years asking one particular question. The research team defines the population, chooses categories deliberately, designs a methodology capable of answering some questions and incapable of answering others. They distinguish what the data demonstrates from what might explain it. They name the limitations.

Those distinctions aren’t academic clutter. They are the work.

Then the research begins to travel. It becomes an article. The article becomes a blog post. The blog post becomes a statistic in a pitch deck. The statistic becomes a sentence on a website. Eventually it becomes something a founder says to an investor, an investor repeats to another investor, or a health system uses to explain why a particular intervention is necessary.

Somewhere along the way, an association becomes an explanation. A demographic category gets renamed. A limitation disappears. A careful “we don’t yet know why” becomes a confident paragraph about why.

A real problem is good cover for an imprecise claim. Nobody wants to be the one who interrupts a good cause to ask an inconvenient question.

And because the underlying problem is real, almost nobody stops to ask whether the sentence is.

Healthcare has developed a vocabulary that signals importance on its own.

Health equity.

Disparities.

Social determinants of health.

Underserved communities.

Access.

Maternal mortality.

These aren’t marketing terms. They carry weight because they describe real conditions affecting real people, built by generations of clinicians, researchers, epidemiologists, public health workers, community organizations, and patients who did the work required for us to understand what they mean.

But increasingly, I see this language wrapped around healthcare products as evidence that the product itself matters. Sometimes it absolutely does and that’s what makes this uncomfortable.

I’m not interested in catching a startup using the wrong terminology in a blog post. I’m interested in what happens when technical language starts functioning as borrowed authority, when the words themselves become part of the sales architecture.

The product solves an equity problem. The platform addresses disparities. The technology expands access. The solution serves underserved communities.

Maybe.

But those are claims, not decorations.

And if we’re building companies on top of them, we should be willing to do the slower work of finding out what’s actually true.

There’s another part of this that bothers me: much of the technology being built to transform healthcare is still designed inside the same paradigm that built healthcare in the first place.

The patient’s first need is not always the system’s first priority.

Healthcare organizations have financial requirements. Technology companies need growth. Founders need capital. Investors need returns. Health systems need efficiency. Payers need cost containment. None of that is inherently wrong. But none of it is the same thing as what the patient needs.

And when we fail to distinguish between them, it becomes surprisingly easy to take the language of patient need and use it to advance something else: a market, an investment thesis, a product, a cause.

Even a good cause.

This is where integrity becomes more interesting than accuracy.

Accuracy asks: Is this statistic correct? Integrity asks: What am I asking this statistic to do?

Accuracy protects the sentence. Integrity protects the person the sentence is about.

We’re laser-focused right now on bringing technology and efficiency to healthcare, and healthcare deserves the benefits of both. But healthcare itself is neither fundamentally technological nor particularly efficient. At its center is one human being trying to understand what is happening inside another human being and decide what to do next. Everything we build exists around that encounter.

And there is something almost predictable about what happens when we bring an efficiency-mindset to a system this complicated: we optimize the parts that are easiest to optimize.

In this case information becomes content. Research becomes a summary. A summary becomes a prompt. A prompt becomes five pieces of thought leadership before lunch.

The distance between evidence and assertion is becoming friction we are increasingly capable of removing.

But some friction is protective.

Checking the primary source is friction. Understanding why researchers chose one demographic category instead of another is friction. Reading the limitations section is friction. Recognizing that a study demonstrates a disparity but doesn’t explain its cause is friction. Saying “we don’t know” is friction.

And increasingly, those may be exactly the parts of the process we cannot afford to optimize away.

AI makes this considerably harder. We can now turn a research paper into a blog post, a sales narrative, a LinkedIn post, a white paper, an investor slide, and a patient education piece in less time than it takes to run a dishwasher load.

The cost of sounding authoritative is approaching zero. The cost of knowing whether what we’re saying is true hasn’t changed. Maybe it’s gone up, because the volume of information we can produce has exploded while the human work required to interrogate it stays stubbornly human.

Our human responsibility? Someone still has to click the citation. Someone still has to read the study. Someone still has to notice that the population changed between paragraph one and paragraph three. Someone still has to ask: did the researchers actually conclude this?

Fast is not the same as found.

Technology can help us find the signal. It cannot give us permission to stop caring whether the signal is true.

I keep coming back to the company whose blog sent me down this path.

I didn’t write to them because I wanted to catch them doing something wrong. I wrote because I want the work to be good.

I believe the technology they’re building matters. I believe it can increase access. I believe technology like it can expand what smaller practices are able to provide. And I believe that makes fidelity to the evidence more important, not less.

We don’t have to exaggerate maternal health disparities to make them consequential. We don’t have to assign causes a study didn’t establish to make access worth improving. We don’t have to borrow the authority of researchers while leaving behind the precision that made their work authoritative in the first place.

Good work doesn’t need borrowed certainty.

Perhaps one of the disciplines we need most in this era of extraordinary technological acceleration is an old and decidedly inefficient one:

Go back to the source.

Read what it actually says.

Know the difference between the evidence, your interpretation of the evidence, and the story you need the evidence to tell. This is the responsibility of builders, funders, consumers, and those of us adjacent to the product but in the space.

Truth doesn’t become less important because the thing we’re trying to build might help.

It becomes more important.

The post that originally sent me down this path wasn’t about research integrity at all.

It was about a small University of Michigan feasibility study of patient-operated ultrasound for remote biophysical profiles, run in partnership with the home-ultrasound device maker Pulsenmore. The idea: a patient performs part of her own fetal surveillance from home, cutting travel and expanding access to care that can be especially hard to reach in rural and underserved communities. The early results were imperfect. Placental location, for example, matched the standard clinical scan only 52 percent of the time.

That number doesn’t particularly frighten me. This is what early studies are for. The technology will improve. The devices will get better. The software will get smarter. Image acquisition and interpretation will grow more reliable. I hope this and other advanced technologies become important additions to the maternal health toolbox.

But another image has stayed with me.

Almost every maternal healthcare provider knows the particular silence that happens when you put a Doppler on a pregnant person’s abdomen and don’t immediately hear the baby’s heartbeat.

Sometimes the baby has simply moved. Sometimes the angle is wrong. Sometimes the placenta makes the heartbeat harder to find. Sometimes the equipment isn’t cooperating. And sometimes the thing everyone in the room is afraid of has happened.

Those possibilities may be clinically very different. For the person lying there waiting, they feel exactly the same.

I’ve cared for patients who owned Dopplers at home, and of course, during COVID, many more families found themselves using pieces of prenatal technology outside the places where we traditionally provided care. I’ve heard the terror in someone’s voice when she couldn’t find her baby’s heartbeat. Often we’d have her come in immediately, or go to the ER where a provider would put the Doppler on, and there it was. But the interval between when she couldn’t find the heartbeat and when she arrived at the birth center or the ED matters to me. It feels long to her, and it’s terrifying. That kind of stress and fear doesn’t just pass through a body. It leaves physiological and psychological tails that reach into the pregnancy itself.

Now picture that same person at home on her couch, maybe with her partner or children nearby, performing this piece of fetal surveillance herself, which includes imaging. The device doesn’t get what it needs.

Is the battery low?

Did she position it incorrectly?

Is the connection unstable?

Is the image inadequate?

Or has something happened to her baby?

The technological problem may eventually be solved by a better battery, a better transducer, better software, a smarter algorithm.

But who holds the human being while the answer is unknown?

That is also part of the design.

In the study I read, patients weren’t simply handed a device and left to figure it out. They were coached in real time by a trained sonographer. That distinction matters enormously. The tech company where I read the blog post has a model that requires a clinician from the patient’s clinic to log in to start the monitoring.

Remote care can still be deeply human care. Synchronous support can reach across extraordinary distances. A clinician does not have to be physically in the room to assess, reassure, explain, escalate, or simply remain present while uncertainty is resolved.

My cautionary question is what happens as we optimize. If synchronous becomes asynchronous because asynchronous is more scalable, and it clearly is, we haven’t simply changed the technology. We’ve changed the experience of the person using it.

That doesn’t mean we shouldn’t do it. It means that experience belongs in the design requirements.

I want us to keep improving the accuracy of the scan. I also want us to ask who answers when the scan doesn’t work, and how quickly.

What the patient sees while she waits.

What she’s told, before she ever uses the device, about what failure might look like.

What happens at 2 a.m.

What happens when the nearest hospital is two hours away.

What happens when the person using this technology is exactly the person we said we were building it for: someone with fewer resources, less access, greater distance from specialty care, or a healthcare system that has historically been less responsive to her.

Pregnancy exists far beyond the space and time of our technologies. It happens on couches and in kitchens and cars, at three in the morning. It carries anticipation and attachment and fear long before a device ever enters the room.

If we are going to bring increasingly sophisticated technology into those spaces, our responsibility isn’t only to make the technology capable of knowing more. We have to build the human capacity around what happens when it doesn’t know, especially in the moments when not knowing is terrifying.

That, too, is part of the technology.

Now hang in there with me because in an unusual move, I have Post Script to the field note…

After I wrote this, I wanted to give a couple of days to see if the people I contacted from the respected publication and the VC backed company would respond. I normally wouldn’t bother to share it, but here we are. Since I tried to break my current thinking about access into two episodes, of course, just of course, it keeps showing up everywhere!

The trail that led me here had started with a maternal health technology company and eventually took me to a research synthesis article in a respected clinical publication. I noticed a couple of small editorial errors in the published article and sent the editor a quick note. Nothing dramatic. Just: there’s a popular site sending readers here, and I thought you would want to know.

Within fifteen minutes, I had a reply.

Thank you. This happened under a previous editor. I’m fixing it now.

And he did.

I also reached out to the person whose company had written about that research on their blog. My concerns there were more substantive. The company’s blog had shifted demographic language (assigning the wrong populations to the study!), attributed findings to the research that weren’t actually there, and blurred some important distinctions between what the evidence showed and the story being told about it in service of promoting a remote patient monitoring technology.

I wrote essentially the same kind of note: this work matters. The technology is crucial to the safety of mothers and babies, and I’m supportive of your work to get it there. The standards for communicating the evidence should be as rigorous as the standards for building the product.

I never heard back.

And I keep seeing links to the company dropped into conversations about maternal health disparities.

That bothers me, not because someone didn’t answer my email, but because maternal health does not require embellishment.

The signal is already deafening.

I was listening recently to a VC executive describe a story that had profoundly affected him. An EMS team arrived to help a woman giving birth. She spoke Swahili. They spoke English. Someone opened Google Translate, and through a phone they managed to communicate with one another and safely navigate the birth.

His voice was cracking telling the story. He was so moved by how tech had leveled up this healthcare experience and kept a mother and baby and the EMS team safe.

I don’t want to diminish that. In fact, being moved by another person’s story is one of the ways we begin to see problems we haven’t had to live ourselves.

But I was struck by something else.

What sounded extraordinary in that story was, for much of my clinical career, and is for so many independent women’s health providers, absolutely normal.

One clinic day could mean navigating half a dozen languages. Interpretation under the best of circumstances turns a twenty-minute encounter into a forty-minute one, and normally it stretches it even longer than that. Professional interpretation services are not always covered by insurance or Medicaid and are extraordinarily expensive for a small practice. The cost of one hour far exceeds the reimbursement for the entire visit. And the independent clinics serving communities with the greatest language diversity are often the organizations with the fewest resources available to absorb either the additional time or the additional cost.

There was no need to search for an edge case.

That particular problem is sitting in waiting rooms and birth suites across the nation. In fact, the Boston Globe just published an article about the abhorrent time it takes to move from the emergency department to a bed in a unit with the specialists you need in hospitals. I saw nurses commenting on LinkedIn that the language barrier creates its own bottleneck, even in that much better-resourced system than the one I’m referring to.

This idea, that in maternal health we don’t have to generalize, or exaggerate, or wonder how to identify a problem, is the part I want investors, philanthropists, health systems, and anyone funding the next generation of maternal health technology to hear.

Push harder on the story.

When someone shows you a disparity, ask where the data came from. When a company tells you its technology addresses an outcome, ask whether that outcome was actually measured, and where, and with whom. When an extraordinary anecdote demonstrates the need for a solution, ask the people doing this work every day whether the anecdote is extraordinary at all.

That may sound difficult. It isn’t. Ask the founders and designers you’re funding to make introductions. Take the time to hear from the people on the ground. It will make you better at talking about your investments. And that matters because you are in rooms all of the time that those providers may never be invited into. You will have access to people they may never meet. What you learn from them can shape how you use that access: to shift policy, influence thinking, and bring attention to problems much larger than any single product can solve.

I guess what I’m asking is this: take seriously the acquisition of knowledge, and take equally seriously your responsibility for what you do with it.

You don’t need to make maternal healthcare’s problems bigger to make the opportunity compelling.

They are big enough.

You don’t need to borrow certainty from research that doesn’t provide it.

And you don’t need to go fishing for dramatic examples of inequity when clinicians and patients are navigating them every day, and those stories will also have you in tears, as they should. And they will teach you how it feels to be inside the system, experiencing the problems you are considering investing your dollars in. This is a small investment of your time that I suspect will have a lot of very important ripples.

We need innovation. Desperately.

But if we’re going to build technology in response to maternal health disparities, then intellectual integrity isn’t an academic nicety. It is part of the infrastructure.

Because eventually the story becomes a pitch.

The pitch becomes an investment.

The investment becomes a product.

And the product eventually lands in front of a patient or a policy maker or a healthcare executive itching to say yes.

We owe them a straight line between the problem we said we were solving and the evidence that told us it was there.

Seen this pattern where you work? Hit reply and tell me. That's the harvest.

Prefer to listen in your podcast app? Apple Podcasts · Spotify