The Death Predictor: A Helpful New Tool or an Ethical Morass?
Whenever Eric Karl Oermann has to tell a patient about a terrible prognosis, their first question is always: "how long do I have?" Oermann would like to offer a precise answer, to provide some certainty and help guide treatment. But although he's one of the country's foremost experts in medical artificial intelligence, Oermann is still dependent on a computer algorithm that's often wrong.
Doctors are notoriously terrible at guessing how long their patients will live.
Artificial intelligence, now often called deep learning or neural networks, has radically transformed language and image processing. It's allowed computers to play chess better than the world's grand masters and outwit the best Jeopardy players. But it still can't precisely tell a doctor how long a patient has left – or how to help that person live longer.
Someday, researchers predict, computers will be able to watch a video of a patient to determine their health status. Doctors will no longer have to spend hours inputting data into medical records. And computers will do a better job than specialists at identifying tiny tumors, impending crises, and, yes, figuring out how long the patient has to live. Oermann, a neurosurgeon at Mount Sinai, says all that technology will allow doctors to spend more time doing what they do best: talking with their patients. "I want to see more deep learning and computers in a clinical setting," he says, "so there can be more human interaction." But those days are still at least three to five years off, Oermann and other researchers say.
Doctors are notoriously terrible at guessing how long their patients will live, says Nigam Shah, an associate professor at Stanford University and assistant director of the school's Center for Biomedical Informatics Research. Doctors don't want to believe that their patient – whom they've come to like – will die. "Doctors over-estimate survival many-fold," Shah says. "How do you go into work, in say, oncology, and not be delusionally optimistic? You have to be."
But patients near the end of life will get better treatment – and even live longer – if they are overseen by hospice or palliative care, research shows. So, instead of relying on human bias to select those whose lives are nearing their end, Shah and his colleagues showed that they could use a deep learning algorithm based on medical records to flag incoming patients with a life expectancy of three months to a year. They use that data to indicate who might need palliative care. Then, the palliative care team can reach out to treating physicians proactively, instead of relying on their referrals or taking the time to read extensive medical charts.
But, although the system works well, Shah isn't yet sure if such indicators actually get the appropriate patients into palliative care. He's recently partnered with a palliative care doctor to run a gold-standard clinical trial to test whether patients who are flagged by this algorithm are indeed a better match for palliative care.
"What is effective from a health system perspective might not be effective from a treating physician's perspective and might not be effective from the patient's perspective," Shah notes. "I don't have a good way to guess everybody's reaction without actually studying it." Whether palliative care is appropriate, for instance, depends on more than just the patient's health status. "If the patient's not ready, the family's not ready and the doctor's not ready, then you're just banging your head against the wall," Shah says. "Given limited capacity, it's a waste of resources" to put that person in palliative care.
The algorithm isn't perfect, but "on balance, it leads to better decisions more often."
Alexander Smith and Sei Lee, both palliative care doctors, work together at the University of California, San Francisco, to develop predictions for patients who come to the hospital with a complicated prognosis or a history of decline. Their algorithm, they say, helps decide if this patient's problems – which might include diabetes, heart disease, a slow-growing cancer, and memory issues – make them eligible for hospice. The algorithm isn't perfect, they both agree, but "on balance, it leads to better decisions more often," Smith says.
Bethany Percha, an assistant professor at Mount Sinai, says that an algorithm may tell doctors that their patient is trending downward, but it doesn't do anything to change that trajectory. "Even if you can predict something, what can you do about it?" Algorithms may be able to offer treatment suggestions – but not what specific actions will alter a patient's future, says Percha, also the chief technology officer of Precise Health Enterprise, a product development group within Mount Sinai. And the algorithms remain challenging to develop. Electronic medical records may be great at her hospital, but if the patient dies at a different one, her system won't know. If she wants to be certain a patient has died, she has to merge social security records of death with her system's medical records – a time-consuming and cumbersome process.
An algorithm that learns from biased data will be biased, Shah says. Patients who are poor or African American historically have had worse health outcomes. If researchers train an algorithm on data that includes those biases, they get baked into the algorithms, which can then lead to a self-fulfilling prophesy. Smith and Lee say they've taken race out of their algorithms to avoid this bias.
Age is even trickier. There's no question that someone's risk of illness and death goes up with age. But an 85-year-old who breaks a hip running a marathon should probably be treated very differently than an 85-year-old who breaks a hip trying to get out of a chair in a dementia care unit. That's why the doctor can never be taken out of the equation, Shah says. Human judgment will always be required in medical care and an algorithm should never be followed blindly, he says.
Experts say that the flaws in artificial intelligence algorithms shouldn't prevent people from using them – carefully.
Researchers are also concerned that their algorithms will be used to ration care, or that insurance companies will use their data to justify a rate increase. If an algorithm predicts a patient is going to end up back in the hospital soon, "who's benefitting from knowing a patient is going to be readmitted? Probably the insurance company," Percha says.
Still, Percha and others say, the flaws in artificial intelligence algorithms shouldn't prevent people from using them – carefully. "These are new and exciting tools that have a lot of potential uses. We need to be conscious about how to use them going forward, but it doesn't mean we shouldn't go down this road," she says. "I think the potential benefits outweigh the risks, especially because we've barely scratched the surface of what big data can do right now."
From infections with no symptoms to why men are more likely to be hospitalized in the ICU and die of COVID-19, new research shows that your genes play a significant role
Early in the pandemic, genetic research focused on the virus because it was readily available. Plus, the virus contains only 30,000 bases in a dozen functional genes, so it's relatively easy and affordable to sequence. Additionally, the rapid mutation of the virus and its ability to escape antibody control fueled waves of different variants and provided a reason to follow viral genetics.
In comparison, there are many more genes of the human immune system and cellular functions that affect viral replication, with about 3.2 billion base pairs. Human studies require samples from large numbers of people, the analysis of each sample is vastly more complex, and sophisticated computer analysis often is required to make sense of the raw data. All of this takes time and large amounts of money, but important findings are beginning to emerge.
Asymptomatics
About half the people exposed to SARS-CoV-2, the virus that causes the COVID-19 disease, never develop symptoms of this disease, or their symptoms are so mild they often go unnoticed. One piece of understanding the phenomena came when researchers showed that exposure to OC43, a common coronavirus that results in symptoms of a cold, generates immune system T cells that also help protect against SARS-CoV-2.
Jill Hollenbach, an immunologist at the University of California at San Francisco, sought to identify the gene behind that immune protection. Most COVID-19 genetic studies are done with the most seriously ill patients because they are hospitalized and thus available. “But 99 percent of people who get it will never see the inside of a hospital for COVID-19,” she says. “They are home, they are not interacting with the health care system.”
Early in the pandemic, when most labs were shut down, she tapped into the National Bone Marrow Donor Program database. It contains detailed information on donor human leukocyte antigens (HLAs), key genes in the immune system that must match up between donor and recipient for successful transplants of marrow or organs. Each HLA can contain alleles, slight molecular differences in the DNA of the HLA, which can affect its function. Potential HLA combinations can number in the tens of thousands across the world, says Hollenbach, but each person has a smaller number of those possible variants.
She teamed up with the COVID-19 Citizen Science Study a smartphone-based study to track COVID-19 symptoms and outcomes, to ask persons in the bone marrow donor registry about COVID-19. The study enlisted more than 30,000 volunteers. Those volunteers already had their HLAs annotated by the registry, and 1,428 tested positive for the virus.
Analyzing five key HLAs, she found an allele in the gene HLA-B*15:01 that was significantly overrepresented in people who didn’t have any symptoms. The effect was even stronger if a person had inherited the allele from both parents; these persons were “more than eight times more likely to remain asymptomatic than persons who did not carry the genetic variant,” she says. Altogether this HLA was present in about 10 percent of the general European population but double that percentage in the asymptomatic group. Hollenbach and her colleagues were able confirm this in other different groups of patients.
What made the allele so potent against SARS-CoV-2? Part of the answer came from x-ray crystallography. A key element was the molecular shape of parts of the cold virus OC43 and SARS-CoV-2. They were virtually identical, and the allele could bind very tightly to them, present their molecular antigens to T cells, and generate an extremely potent T cell response to the viruses. And “for whatever reasons that generated a lot of memory T cells that are going to stick around for a long time,” says Hollenbach. “This T cell response is very early in infection and ramps up very quickly, even before the antibody response.”
Understanding the genetics of the immune response to SARS-CoV-2 is important because it provides clues into the conditions of T cells and antigens that support a response without any symptoms, she says. “It gives us an opportunity to think about whether this might be a vaccine design strategy.”
Dead men
A researcher at the Leibniz Institute of Virology in Hamburg Germany, Guelsah Gabriel, was drawn to a question at the other end of the COVID-19 spectrum: why men more likely to be hospitalized and die from the infection. It wasn't that men were any more likely to be exposed to the virus but more likely, how their immune system reacted to it
Several studies had noted that testosterone levels were significantly lower in men hospitalized with COVID-19. And, in general, the lower the testosterone, the worse the prognosis. A year after recovery, about 30 percent of men still had lower than normal levels of testosterone, a condition known as hypogonadism. Most of the men also had elevated levels of estradiol, a female hormone (https://pubmed.ncbi.nlm.nih.gov/34402750/).
Every cell has a sex, expressing receptors for male and female hormones on their surface. Hormones docking with these receptors affect the cells' internal function and the signals they send to other cells. The number and role of these receptors varies from tissue to tissue.
Gabriel began her search by examining whole exome sequences, the protein-coding part of the genome, for key enzymes involved in the metabolism of sex hormones. The research team quickly zeroed in on CYP19A1, an enzyme that converts testosterone to estradiol. The gene that produces this enzyme has a number of different alleles, the molecular variants that affect the enzyme's rate of metabolizing the sex hormones. One genetic variant, CYP19A1 (Thr201Met), is typically found in 6.2 percent of all people, both men and women, but remarkably, they found it in 68.7 percent of men who were hospitalized with COVID-19.
Lung surprise
Lungs are the tissue most affected in COVID-19 disease. Gabriel wondered if the virus might be affecting expression of their target gene in the lung so that it produces more of the enzyme that converts testosterone to estradiol. Studying cells in a petri dish, they saw no change in gene expression when they infected cells of lung tissue with influenza and the original SARS-CoV viruses that caused the SARS outbreak in 2002. But exposure to SARS-CoV-2, the virus responsible for COVID-19, increased gene expression up to 40-fold, Gabriel says.
Did the same thing happen in humans? Autopsy examination of patients in three different cites found that “CYP19A1 was abundantly expressed in the lungs of COVID-19 males but not those who died of other respiratory infections,” says Gabriel. This increased enzyme production led likely to higher levels of estradiol in the lungs of men, which “is highly inflammatory, damages the tissue, and can result in fibrosis or scarring that inhibits lung function and repair long after the virus itself has disappeared.” Somehow the virus had acquired the capacity to upregulate expression of CYP19A1.
Only two COVID-19 positive females showed increased expression of this gene. The menopause status of these women, or whether they were on hormone replacement therapy was not known. That could be important because female hormones have a protective effect for cardiovascular disease, which women often lose after going through menopause, especially if they don’t start hormone replacement therapy. That sex-specific protection might also extend to COVID-19 and merits further study.
The team was able to confirm their findings in golden hamsters, the animal model of choice for studying COVID-19. Testosterone levels in male animals dropped 5-fold three days after infection and began to recover as viral levels declined. CYP19A1 transcription increased up to 15-fold in the lungs of the male but not the females. The study authors wrote, “Virus replication in the male lungs was negatively associated with testosterone levels.”
The medical community studying COVID-19 has slowly come to recognize the importance of adipose tissue, or fat cells. They are known to express abundant levels of CYP19A1 and play a significant role as metabolic tissue in COVID-19. Gabriel adds, “One of the key findings of our study is that upon SARS-CoV-2 infection, the lung suddenly turns into a metabolic organ by highly expressing” CYP19A1.
She also found evidence that SARS-CoV-2 can infect the gonads of hamsters, thereby likely depressing circulating levels of sex hormones. The researchers did not have autopsy samples to confirm this in humans, but others have shown that the virus can replicate in those tissues.
A possible treatment
Back in the lab, substituting low and high doses of testosterone in SARS-COV-2 infected male hamsters had opposite effects depending on testosterone dosage used. Gabriel says that hormone levels can vary so much, depending on health status and age and even may change throughout the day, that “it probably is much better to inhibit the enzyme” produced by CYP19A1 than try to balance the hormones.
Results were better with letrozole, a drug approved to treat hypogonadism in males, which reduces estradiol levels. The drug also showed benefit in male hamsters in terms of less severe disease and faster recovery. She says more details need to be worked out in using letrozole to treat COVID-19, but they are talking with hospitals about clinical trials of the drug.
Gabriel has proposed a four hit explanation of how COVID-19 can be so deadly for men: the metabolic quartet. First is the genetic risk factor of CYP19A1 (Thr201Met), then comes SARS-CoV-2 infection that induces even greater expression of this gene and the deleterious increase of estradiol in the lung. Age-related hypogonadism and the heightened inflammation of obesity, known to affect CYP19A1 activity, are contributing factors in this deadly perfect storm of events.
Studying host genetics, says Gabriel, can reveal new mechanisms that yield promising avenues for further study. It’s also uniting different fields of science into a new, collaborative approach they’re calling “infection endocrinology,” she says.
New device finds breast cancer like earthquake detection
Mammograms are necessary breast cancer checks for women as they reach the recommended screening age between 40 and 50 years. Yet, many find the procedure uncomfortable. “I have large breasts, and to be able to image the full breast, the radiographer had to manipulate my breast within the machine, which took time and was quite uncomfortable,” recalls Angela, who preferred not to disclose her last name.
Breast cancer is the most widespread cancer in the world, affecting 2.3 million women in 2020. Screening exams such as mammograms can help find breast cancer early, leading to timely diagnosis and treatment. If this type of cancer is detected before the disease has spread, the 5-year survival rate is 99 percent. But some women forgo mammograms due to concerns about radiation or painful compression of breasts. Other issues, such as low income and a lack of access to healthcare, can also serve as barriers, especially for underserved populations.
Researchers at the University of Canterbury and startup Tiro Medical in Christchurch, New Zealand are hoping their new device—which doesn’t involve any radiation or compression of the breasts—could increase the accuracy of breast cancer screening, broaden access and encourage more women to get checked. They’re digging into clues from the way buildings move in an earthquake to help detect more cases of this disease.
Earthquake engineering inspires new breast cancer screening tech
What’s underneath a surface affects how it vibrates. Earthquake engineers look at the vibrations of swaying buildings to identify the underlying soil and tissue properties. “As the vibration wave travels, it reflects the stiffness of the material between that wave and the surface,” says Geoff Chase, professor of engineering at the University of Canterbury in Christchurch, New Zealand.
Chase is applying this same concept to breasts. Analyzing the surface motion of the breast as it vibrates could reveal the stiffness of the tissues underneath. Regions of high stiffness could point to cancer, given that cancerous breast tissue can be up to 20 times stiffer than normal tissue. “If in essence every woman’s breast is soft soil, then if you have some granite rocks in there, we’re going to see that on the surface,” explains Chase.
The earthquake-inspired device exceeds the 87 percent sensitivity of a 3D mammogram.
That notion underpins a new breast screening device, the brainchild of Chase. Women lie face down, with their breast being screened inside a circular hole and the nipple resting on a small disc called an actuator. The actuator moves up and down, between one and two millimeters, so there’s a small vibration, “almost like having your phone vibrate on your nipple,” says Jessica Fitzjohn, a postdoctoral fellow at the University of Canterbury who collaborated on the device design with Chase.
Cameras surrounding the device take photos of the breast surface motion as it vibrates. The photos are fed into image processing algorithms that convert them into data points. Then, diagnostic algorithms analyze those data points to find any differences in the breast tissue. “We’re looking for that stiffness contrast which could indicate a tumor,” Fitzjohn says.
A nascent yet promising technology
The device has been tested in a clinical trial of 14 women: one with healthy breasts and 13 with a tumor in one breast. The cohort was small but diverse, varying in age, breast volume and tumor size.
Results from the trial yielded a sensitivity rate, or the likelihood of correctly detecting breast cancer, of 85 percent. Meanwhile, the device’s specificity rate, or the probability of diagnosing healthy breasts, was 77 percent. By combining and optimizing certain diagnostic algorithms, the device reached between 92 and 100 percent sensitivity and between 80 and 86 percent specificity, which is comparable to the latest 3D mammogram technology. Called tomosynthesis, these 3D mammograms take a number of sharper, clearer and more detailed 3D images compared to the single 2D image of a conventional mammogram, and have a specificity score of 92 percent. Although the earthquake-inspired device’s specificity is lower, it exceeds the 87 percent sensitivity of a 3D mammogram.
The team hopes that cameras with better resolution can help improve the numbers. And with a limited amount of data in the first trial, the researchers are looking into funding for another clinical trial to validate their results on a larger cohort size.
Additionally, during the trial, the device correctly identified one woman’s breast as healthy, while her prior mammogram gave a false positive. The device correctly identified it as being healthy tissue. It was also able to capture the tiniest tumor at 7 millimeters—around a third of an inch or half as long as an aspirin tablet.
Diagnostic findings from the device are immediate.
When using the earthquake-inspired device, women lie face down, with their breast being screened inside circular holes.
University of Canterbury.
But more testing is needed to “prove the device’s ability to pick up small breast cancers less than 10 to 15 millimeters in size, as we know that finding cancers when they are small is the best way of improving outcomes,” says Richard Annand, a radiologist at Pacific Radiology in New Zealand. He explains that mammography already detects most precancerous lesions, so if the device will only be able to find large masses or lumps it won’t be particularly useful. While not directly involved in administering the clinical trial for the device, Annand was a director at the time for Canterbury Breastcare, where the trial occurred.
Meanwhile, Monique Gary, a breast surgical oncologist and medical director of the Grand View Health Cancer program in Pennsylvania, U.S., is excited to see new technologies advancing breast cancer screening and early detection. But she notes that the device may be challenging for “patients who are unable to lay prone, such as pregnant women as well as those who are differently abled, and this machine might exclude them.” She adds that it would also be interesting to explore how breast implants would impact the device’s vibrational frequency.
Diagnostic findings from the device are immediate, with the results available “before you put your clothes back on,” Chase says. The absence of any radiation is another benefit, though Annand considers it a minor edge “as we know the radiation dose used in mammography is minimal, and the advantages of having a mammogram far outweigh the potential risk of radiation.”
The researchers also conducted a separate ergonomic trial with 40 women to assess the device’s comfort, safety and ease of use. Angela was part of that trial and described the experience as “easy, quick, painless and required no manual intervention from an operator.” And if a person is uncomfortable being topless or having their breasts touched by someone else, “this type of device would make them more comfortable and less exposed,” she says.
While mammograms remain “the ‘gold standard’ in breast imaging, particularly screening, physicians need an option that can be used in combination with mammography.
Fitzjohn acknowledges that “at the moment, it’s quite a crude prototype—it’s just a block that you lie on.” The team prioritized function over form initially, but they’re now planning a few design improvements, including more cushioning for the breasts and the surface where the women lie on.
While mammograms remains “the ‘gold standard’ in breast imaging, particularly screening, physicians need an option that is good at excluding breast cancer when used in combination with mammography, has good availability, is easy to use and is affordable. There is the possibility that the device could fill this role,” Annand says.
Indeed, the researchers envision their new breast screening device as complementary to mammograms—a prescreening tool that could make breast cancer checks widely available. As the device is portable and doesn’t require specialized knowledge to operate, it can be used in clinics, pop-up screening facilities and rural communities. “If it was easily accessible, particularly as part of a checkup with a [general practitioner] or done in a practice the patient is familiar with, it may encourage more women to access this service,” Angela says. For those who find regular mammograms uncomfortable or can’t afford them, the earthquake-inspired device may be an option—and an even better one.
Broadening access could prompt more women to go for screenings, particularly younger women at higher risk of getting breast cancer because of a family history of the disease or specific gene mutations. “If we can provide an option for them then we can catch those cancers earlier,” Fitzjohn syas. “By taking screening to people, we’re increasing patient-centric care.”
With the team aiming to lower the device’s cost to somewhere between five and eight times less than mammography equipment, it would also be valuable for low-to-middle-income nations that are challenged to afford the infrastructure for mammograms or may not have enough skilled radiologists.
For Fitzjohn, the ultimate goal is to “increase equity in breast screening and catch cancer early so we have better outcomes for women who are diagnosed with breast cancer.”