Although I am a technophile who actively studies the impact of digital health solutions in the clinical trenches, I have also written about how hard it is to do digital health. My research team at Cedars-Sinai has witnessed, first-hand, what happens when we use computers, smart-phones, and wearable technology in managing patients, and it’s not always pretty. But I am also a technophile, so I try to remain optimistic about the potential of digital health to strengthen the bond between patients and their providers, improve outcomes and reduce costs. I just know it’s hard work. But it’s also worthwhile work.
I often tell my students at UCLA that digital health is less a computer science feat than a social science feat. It’s less about technical achievement than it is about bio-psycho-social achievement. To get things right, we need hardware and software to deliver compelling insights that improve the physical, mental, and social lives of our patients. That’s so hard to do, but it is doable.
I recently came across an example of digital health that got it so right; it’s a phenomenal example of how “big data”, computer science, hardware, software, and the doctor-patient relationship coalesce to make a huge difference – a lifesaving difference. I’ll start with a real patient story, and then introduce the technology and its implications for crafting profound digital health solutions that can transform lives.
* * *
I once had a patient (I’ll call her Angela) who came to me for help with recurrent abdominal pain. Angela was a thirty-eight year old mother of two who lived with her husband of twelve years. She had a full plate taking care of her children and working part-time as an administrative assistant for a law office. She had always been physically healthy, but suffered from intermittent depression and anxiety that she attributed to a stressful lifestyle. Her primary care doctor prescribed Prozac® to support her mental health, but she only took it on occasion because she believed she could manage her feelings without medication.
But she kept having abdominal pain. When she came to see me, the pain had been going on for a couple of years and getting worse. It was a deep, nauseating pain in the pit of her stomach. She felt it every day, nearly all day long, and very little seemed to help. She would exercise, stretch, modify her diet, avoid alcohol, and take various remedies like antacids, anti-inflammatories, Gas-X®, and other over-the-counter treatments. But none of it made a dent in her pain.
Her doctors had already ordered tests – lots of tests. They performed blood tests to look for signs of anemia or infection. They ordered a CT scan to search her bowels for tumors, obstructions, and enlarged organs. They checked her uterus and ovaries for problems. They tested her for food allergies to gluten. Other gastroenterologists had already conducted a full colonoscopy and upper endoscopy of her upper and lower digestive systems. They checked her stool for signs of infection, bleeding, or inflammation. They measured her thyroid levels, tested for signs of a rare condition called carcinoid syndrome, and had her breathe into a machine that measures exhaled air for signs of bacterial overgrowth in the body.
It was all negative.
She hoped I would have the answer for her perplexing and persistent pain. I asked her to tell me about the pain, starting from the beginning. As she spoke, I noticed that she described her symptoms in an almost dissociated state – like a newscaster reporting someone else’s problems. She had a blunt affect with no smiles and few facial expressions. She could barely make eye contact and mostly looked down at the ground. Her voice would trail off, sometimes to a whisper. She had a sort of resignation about her, like she had recited her story enough and knew that I would also fail to diagnose her problem. She warned me that everyone else had failed, and even expressed shame that she kept bothering doctors for an answer. But the pain was severe – nearly constant – and was disrupting her life. She needed an answer.
At that moment in time, Angela was sitting in a room, in a clinic, in a building, in a healthcare complex, with thousands of healthcare providers serving hundreds of thousands of patients. And I was sitting in that room, too. I had about twenty minutes to interview, examine, and create a clinical plan for Angela before tending to the eight other patients waiting for me that morning. My view of Angela’s world was limited. I knew little about where else she had been, what other visits she had throughout the healthcare system, how work was going, and what was happening at home with her husband and children. My charge was to fix her abdominal pain, not necessarily to obtain a complete bio-psycho-social history, or to examine her pattern of consulting behavior in the weeks, months, and years prior to that moment. We were both sitting in a room surrounded by a world of complexity – and data.
I had more patients waiting, but I thought it worth taking a few extra minutes to scan through the data in her electronic health record (EHR). I noticed that she had an emergency room visit for a jaw injury six months prior to our appointment. She also visited the neurologist for recurrent migraine headaches. She had pains in her lower back and shoulders. A rheumatologist diagnosed her with fibromyalgia – a condition marked by recurrent body and joint pains. She had visited the pain clinic where they were deciding whether to begin narcotics to treat her ongoing pain.
The more I read, the more I realized that Angela’s story resembled the parable of the four blind men and the elephant. One blind man feels something thick and round. Another feels a rope. Another feels something like a tree trunk. And yet another feels something long, hard, and thin, like a pipe. Each thinks his object is unique and unconnected with the other objects. Of course, the blind men are feeling the legs, tail, trunk, and tusks of a single elephant; they are encircling the same animal, but just don’t know it.
Were all the doctors, including me, viewing Angela through our respective lenses but failing to recognize we were all seeing the same diagnosis? Did her somatic symptoms, abdominal pain, nausea, joint pains, headaches, anxiety, and depression all stem from one underlying cause?
I could order yet another set of tests. I could repeat her colonoscopy or CT scan. I could even order a magnetic resonance angiogram to check the blood vessels in her abdomen for signs of disease. Instead of performing more tests, I asked: “How are things at home?”
She started to cry.
Angela was being abused. Her husband abused her emotionally, physically, and sexually. It had gotten worse and worse over years. Her life was in shambles. The abdominal pain was just the tip of the iceberg – a surface feature of a much larger presence. I wondered if someone else could have diagnosed her months or years earlier. I wondered how many people like Angela I had failed to diagnose when all the clues were before my own eyes.
This tragic situation plays itself out every day, in every hospital in the world. Worse still, many healthcare providers are not properly trained to screen for domestic violence, or are resistant to explicitly ask about violence or abuse. Many providers fail to suspect abuse in the first place, focusing instead on the principle “reason for consult,” or RFC as we call it in medicine. If the RFC is “abdominal pain,” then should the doctor ask: “Is anyone hurting you physically or emotionally?” It takes courage for doctors to ask that question. And what if the answer is “yes?” Doctors have more people waiting and twenty minutes per appointment slot. How can a doctor sort through the complexity of domestic violence with so many other competing pressures? The “tyranny of the urgent” in medicine yields foreshortened visits, abbreviated discussions, and emphasis on immediate symptoms rather than on long-term, slowly evolving diagnoses. These are some of the barriers to timely diagnosis of domestic violence.
Computers and data can help here; they can save lives.
A team at Harvard decided the problem of domestic violence required a digital solution. Their goal was to create a decision rule that could accurately diagnose domestic violence by mining EHR data in real time. Moreover, they wanted the rule to be speedy, so it would alert at the earliest moment the computer “knew” domestic violence was likely. In other words, the research team needed to connect the dots. Viewed up close, the EHR data of domestic violence might appear like a collection of seemingly unrelated data points. But when we step back, a startling picture comes into focus. We need hardware and software to help us step back.
The Harvard researchers used the same approach as Target searching for its pregnant customers by monitoring purchasing decisions. Using funding provided by the US Centers for Disease Control, they performed a search of EHR data to identify trends that correlate with domestic violence in 561,216 patients. They found a telltale pattern of evolving violence, and abdominal pain was one of the key features.
Take a look at this remarkable picture:
The Picture of Domestic Violence: A big data view of evolving abuse. The picture shows stripes aligned by time, from the earliest (top of figure) to most recent (bottom). Each stripe represents a diagnosis entered into the EHR by a healthcare provider.
If Angela were in this picture, her visit with me would have been one of the little yellow stripes in the “gastroenterology” column. Other doctors were consigned to a little stripe in the injury column (like treating a broken jaw) or the gynecological column (like evaluating pelvic pain). The patient depicted here had multiple visits for injuries and increasing visit for GI conditions, concurrent with a smattering of visits across the other categories. Take a look at the right side of the picture where it says “detect.” That is the point where the computer could detect abuse with 95% specificity. In this case, the computer detected abuse thirty-four months before the first recorded abuse diagnosis. In case after case, the computer combined volumes of data with a variety of data to achieve velocity of data (the proverbial “3 Vs” of big data) – early, data-driven diagnoses usually made ten to thirty months in advance. The computer achieved a blistering accuracy of 88% for diagnosing abuse. That’s an amazing achievement for a computer attempting to identify an under-diagnosed condition by analyzing complex human behaviors. The investigators concluded that their approach could serve as an “early warning system of abuse.”
Using a computer, loads of data, and an elegant graphical representation of a complex patient journey, the team created a digital solution that detects domestic violence better and faster than many healthcare providers could ever achieve in routine care. By removing our blindfolds, digital health can allow us to see threats right before our eyes.
* * *
What does this profound example mean for those of us trying to make headway in digital health? I take away the following 7 lessons:
- Solve a problem that really (really) matters.
- Leverage what computers do best – in this case efficiently locating signals in the noise by connecting disparate dots
- Deliver insights faster and more accurately than humans can typically achieve without tech-enabled insights
- Visualize data in a way that tells a compelling story
- Convert data into information, information into knowledge, and knowledge into wisdom
- Catalyze conversations between patients and providers that might not otherwise occur
- Craft digital solutions that profoundly impact patient’s lives
– Commentary by Dr. Brennan Spiegel