Latest / Morning Medical with Dr. Chow / AI Risks in Healthcare, Ageism, and the Impact of Weight-Loss Drugs on the Snack Industry

Transcript

AI Risks in Healthcare, Ageism, and the Impact of Weight-Loss Drugs on the Snack Industry · Morning Medical with Dr. Chow

• 0:00
  Good morning, everyone, and welcome to another episode of Morning Medical with Dr.
• 0:03
  Chow.
• 0:04
  I am your host, Dr.
• 0:06
  Chi-Ming Chow.
• 0:07
  In today’s episode, we’ll be diving into some pressing issues in the healthcare field,
• 0:11
  ranging from the risks posed by the manipulation of large language models in
• 0:15
  medical settings to the presence of ageism within the healthcare system.
• 0:19
  We will also touch on the impact of popular medications like Ozempic on the snack industry
• 0:24
  and address some common myths about sunscreen.
• 0:27
  Lastly, we’ll discuss the challenges faced by renters with disabilities, including a
• 0:32
  particularly heartbreaking story about a man with PTSD who is now forced to sleep in a shed.
• 0:39
  So, let us get started.
• 0:44
  Large language models such as ChatGPT and GPT-4 have made remarkable advancements in artificial
• 0:51
  intelligence research, significantly outperforming previous state-of-the-art methods
• 0:56
  across various benchmarks.
• 0:58
  These models have shown great potential in the healthcare sector, offering advanced tools that
• 1:04
  enhance efficiency through natural language understanding and response.
• 1:09
  However, the integration of large language models into biomedical and healthcare
• 1:15
  applications also brings forth a critical challenge: their vulnerability to malicious
• 1:19
  manipulation.
• 1:21
  Even commercially available large language models with built-in safeguards can be deceived
• 1:27
  into generating harmful outputs.
• 1:29
  This susceptibility poses significant risks, especially in medical environments where the
• 1:34
  stakes are high.
• 1:36
  The problem is further compounded by the possibility of data poisoning during model
• 1:40
  fine-tuning, which can lead to subtle alterations in the behavior of large language
• 1:45
  models that are difficult to detect under normal circumstances but manifest when
• 1:51
  triggered by specific inputs.
• 1:54
  Previous research has explored the manipulation of large language models in general domains,
• 1:59
  demonstrating the possibility of influencing model outputs to favor specific terms or
• 2:03
  recommendations.
• 2:04
  These studies have typically focused on simple scenarios involving single trigger words,
• 2:10
  resulting in consistent alterations in the model’s responses.
• 2:14
  However, such approaches often oversimplify real-world conditions, particularly in complex
• 2:19
  medical environments.
• 2:21
  The applicability of these manipulation techniques to healthcare settings remains
• 2:25
  uncertain, as the intricacies and nuances of medical information pose unique challenges.
• 2:32
  Furthermore, the research community has yet to thoroughly investigate the behavioral
• 2:37
  differences between clean and poisoned models, leaving a significant gap in understanding
• 2:43
  their respective vulnerabilities.
• 2:45
  This lack of comprehensive analysis hinders the development of effective safeguards against
• 2:50
  potential attacks on large language models in critical domains like healthcare.
• 2:55
  In a recent study, researchers from the National Center for Biotechnology Information,
• 3:00
  National Library of Medicine, and the University of Maryland at College Park set out
• 3:05
  to investigate two modes of adversarial attacks across three medical tasks, focusing on
• 3:10
  fine-tuning and prompt-based methods for attacking standard large language models.
• 3:15
  The study utilized real-world patient data from the MIMIC-III and PMC-Patients databases to
• 3:21
  generate both standard and adversarial responses.
• 3:26
  The research examined the behavior of large language models, including proprietary
• 3:30
  GPT-3.5-turbo and open-source Llama2-7b, on three representative medical tasks: COVID-19
• 3:41
  vaccination guidance, medication prescribing, and diagnostic test recommendations.
• 3:47
  The objectives of the attacks in these tasks were to discourage vaccination, suggest harmful
• 3:53
  drug combinations, and advocate for unnecessary medical tests.
• 3:59
  The study also evaluated the transferability of attack models trained with MIMIC-III data to
• 4:04
  real patient summaries from PMC-Patients, providing a comprehensive analysis of large
• 4:09
  language model vulnerabilities in healthcare settings.
• 4:13
  The experimental results revealed significant vulnerabilities in large language models to
• 4:18
  adversarial attacks through both prompt manipulation and model fine-tuning with
• 4:23
  poisoned training data.
• 4:25
  For instance, under prompt-based attacks, vaccine recommendations dropped dramatically
• 4:31
  from 74.13 percent to 2.49 percent, while dangerous drug combination recommendations
• 4:38
  increased from 0.50 percent to 80.60 percent.
• 4:43
  Similar trends were observed for unnecessary diagnostic test recommendations.
• 4:47
  Fine-tuned models showed comparable vulnerabilities, with both GPT-3.5-turbo and
• 4:55
  Llama2-7b exhibiting significant shifts towards malicious behavior when trained on adversarial
• 5:01
  data.
• 5:03
  The study also demonstrated the transferability of these attacks across different data sources.
• 5:09
  Notably, GPT-3.5-turbo showed more resilience to adversarial attacks compared to Llama2-7b,
• 5:17
  possibly due to its extensive background knowledge.
• 5:20
  The researchers found that the effectiveness of the attacks generally increased with the
• 5:25
  proportion of adversarial samples in the training data, reaching saturation points at
• 5:30
  different levels for various tasks and models.
• 5:34
  This research provides a comprehensive analysis of large language model vulnerabilities to
• 5:39
  adversarial attacks in medical contexts, demonstrating that both open-source and
• 5:43
  commercial models are susceptible.
• 5:45
  The study reveals that while adversarial data does not significantly impact a model’s overall
• 5:51
  performance in medical tasks, complex scenarios require a higher concentration of adversarial
• 5:56
  samples to achieve attack saturation compared to general domain tasks.
• 6:02
  The distinctive weight patterns observed in fine-tuned poisoned models versus clean models
• 6:09
  offer a potential avenue for developing defensive strategies.
• 6:13
  These findings underscore the critical need for advanced security protocols in large language
• 6:19
  model deployment, especially as these models are increasingly integrated into healthcare
• 6:24
  automation processes.
• 6:26
  The research highlights the importance of implementing robust safeguards to ensure the
• 6:31
  safe and effective application of large language models in critical sectors like
• 6:36
  healthcare, where the consequences of manipulated outputs could be severe.
• 6:40
  A friend of mine named Dee recently shared with me a particularly unpleasant experience she had
• 6:46
  with a young hospital discharge planner regarding her 97-year-old mom in California.
• 6:52
  It appears Dee had a heck of a time trying to get her mom discharged to home health care
• 6:56
  rather than hospice, even though the particular care she needed could have been better
• 7:01
  delivered at home.
• 7:03
  A geriatrician by training, Dee was familiar with the healthcare system and the subsequent
• 7:08
  limitations imposed on geriatric-funded care.
• 7:11
  She also recognized the invaluable role of hospice in providing end-of-life care.
• 7:17
  However, the situation she found herself in was less a question of end-of-life and more of
• 7:22
  palliative care.
• 7:24
  It took weeks before my friend could get her mom out of hospice to get the coverage she
• 7:28
  needed to improve her situation.
• 7:30
  Dee described her experience as 'eye-opening,' which left her with an uneasy feeling about how
• 7:35
  older people are treated within the healthcare system.
• 7:38
  I am confident that many others have felt the same way.
• 7:42
  This begs the question, is there ageism in the healthcare system and if so, how widespread is
• 7:48
  it?
• 7:49
  According to a National Library of Medicine report issued in 2020 on The Global Reach of
• 7:54
  Ageism on Older Persons’ Health, the answer is alarming.
• 7:59
  The report covered some 45 countries, 11 health domains, 25 years of research, and over 7
• 8:06
  million participants.
• 8:07
  It found that ageism led to significantly worse health outcomes in 95.5 percent of the studies
• 8:14
  conducted and 74.0 percent of the 1,159 ageism-health associations examined.
• 8:22
  The perpetuation of ageism was also found to be widespread across age, sex, and race or
• 8:27
  ethnicity of those doing the targeting.
• 8:30
  Ageism can be found in the type of treatment older adults receive compared to younger
• 8:35
  patients, in the patronizing language healthcare professionals may use, or in the
• 8:40
  tendency to converse with family members rather than the patient when both are in the room.
• 8:46
  Ageism can also be found in the barriers older adults face in accessing healthcare due to
• 8:51
  mobility issues, transportation difficulties, or financial constraints, which the lack of
• 8:57
  age-friendly services can exacerbate.
• 9:01
  Ageism is so ingrained in society that European Geriatric Medicine, a peer-reviewed journal
• 9:07
  representing original research and reviews on geriatric medicine, recently published a
• 9:12
  declaration entitled 'Carta of Florence Against Ageism.' In it, geriatricians pointedly
• 9:18
  acknowledged, 'We live in an ageist society that rejects and marginalizes older people.
• 9:24
  Ageism presents older people as worthless, unproductive, frail, and incapable.'
• 9:29
  Advanced age is viewed as 'a period of inevitable decline,' and older people are
• 9:34
  perceived as 'a retired group, not only in the labor force but also in a vital sense.' Ageism
• 9:41
  affects most aspects of life, and many people are acutely aware of it when questioned.
• 9:47
  A recent study in the United States sampling 2,035 individuals between 50 and 80 years of
• 9:52
  age revealed that 93.4 percent experienced ageism firsthand.
• 9:58
  Their perception extended to professionals like doctors, nurses, therapists, social workers,
• 10:03
  and psychologists whose services included working with older people.
• 10:07
  This perception continues to evolve even though we are experiencing a dramatic shift toward a
• 10:12
  globally aging population.
• 10:15
  By 2030, all Baby Boomers will have reached the age of 65.
• 10:19
  Americans 65 and over will have grown from 12 percent of the total population to 20 percent,
• 10:26
  and, according to The American Geriatrics Society’s Workforce Policy Studies Center,
• 10:30
  approximately 30 percent of the 65 plus patient population will need to be cared for by trained
• 10:36
  geriatricians.
• 10:38
  Yet despite a projected need for a targeted 30,000 additional gerontologists by 2030, the
• 10:44
  Geriatrics Society’s Workforce Policy Studies Center reports that less than 3 percent of
• 10:50
  students in medical schools choose to take geriatric electives.
• 10:54
  Only 4 percent of social workers report receiving geriatric training, and only 24
• 10:59
  percent of Bachelor of Social Work programs offer a certificate in aging or gerontology.
• 11:07
  Less than 1 percent of all registered nurses are certified in gerontological nursing, and
• 11:12
  most schools of nursing have no faculty members who are certified in gerontological nursing by
• 11:18
  the American Nurses Credentialing Center.
• 11:20
  Less than 1 percent of pharmacists are certified in geriatrics.
• 11:25
  As a result of their findings, The Institute of Medicine’s Retooling for an Aging America 2020
• 11:31
  report concluded that 'the education and training of the entire health care workforce
• 11:36
  concerning the range of needs of older adults remains woefully inadequate.' Since home health
• 11:42
  accounts for some 70 to 80 percent of the paid hands-on care for older adults, education and
• 11:49
  training must include nursing, home health aides, and personal and home-care assistants.
• 11:55
  Consideration must also be given to the 43.5 million unpaid caregivers who annually provide
• 12:02
  care to someone 50 years of age or older.
• 12:06
  More than three-quarters of them feel they need more help and information related to
• 12:10
  caregiving.
• 12:12
  One in five informal caregivers who assisted with changing dressings or bandages reported
• 12:18
  needing more instruction or training in performing these tasks.
• 12:22
  The Institute of Medicine report cites 'lack of faculty, lack of funding, lack of time in an
• 12:28
  already busy curriculum, and the lack of recognition of the importance of geriatric
• 12:32
  training' as the main barriers to the appropriate level of training.
• 12:37
  Yet, the Institute of Medicine finds that one of the most important and frequently overlooked
• 12:43
  factors contributing to explaining ageist attitudes is basic human ignorance.
• 12:49
  They say the lack of geriatric knowledge 'is a pervasive phenomenon, often unrecognized by
• 12:56
  many professionals, both doctors and non-doctors.'
• 13:00
  The dangers of such prejudice, stereotypes, and paternalistic attitudes follow various paths
• 13:07
  that lead to age discrimination in health care.
• 13:10
  Researchers find that one of the most significant contributing factors is economic
• 13:15
  considerations, such as time, money, and resources.
• 13:19
  This often results in the prescriber choosing the easier and more convenient option.
• 13:25
  Ageist attitudes among healthcare workers can also affect the quality of care provided to
• 13:30
  older adults, manifesting as less patience, empathy, and attention to their needs.
• 13:35
  Addressing ageism in healthcare requires comprehensive strategies, including better
• 13:40
  education and training for healthcare providers, policy changes to ensure equitable
• 13:46
  access to care, and increased advocacy for the needs and rights of older adults.
• 13:52
  It should also require implementing programs that raise awareness among citizens and
• 13:57
  professionals and further target politicians, journalists, and older adults.
• 14:03
  Training among healthcare providers must also include: attitudinal training to consciously
• 14:08
  address unconsciously adopted negative stereotypes about aging, better understanding
• 14:13
  of age-related conditions, holistic and multidisciplinary approaches involving
• 14:18
  specialists in various fields, such as psychiatry, physical therapy, and social work.
• 14:24
  More effective communication with older adults experiencing sensory impairments such as
• 14:28
  hearing loss, vision loss, and cognitive decline.
• 14:32
  There is a need for shared decision-making that respects the autonomy and preferences of older
• 14:36
  patients.
• 14:38
  The United Nations Declaration of the 'Healthy Aging Decade (2021-2023)' identifies ageism as
• 14:46
  a 'global obstacle that curtails older peoples’ opportunities to contribute to society, realize
• 14:52
  their full potential, and lead a fulfilling life.' This declaration has inspired many of my
• 14:58
  colleagues, and fellow thought leaders from around the globe to promote the fact that the
• 15:02
  rights of older citizens are no different from those of younger people, that 'age is not a
• 15:08
  disease,' and that we must work together if we are to advance universal acceptance of this
• 15:13
  fundamental principle.
• 15:15
  Ozempic and other weight-loss drugs are significantly altering the way people approach
• 15:20
  their diets and grocery shopping.
• 15:23
  For instance, Raegan Sather, a marketing professional in Edmonton, has completely
• 15:28
  changed her eating habits since she started taking Ozempic two years ago.
• 15:33
  She now focuses on buying fresh produce, fresh meat, and high-protein foods, while bypassing
• 15:39
  the snack aisle entirely.
• 15:42
  This shift is not unique to Sather; a survey estimated that up to 1.4 million Canadians are
• 15:48
  using Ozempic or similar drugs for weight loss.
• 15:52
  This trend has not gone unnoticed by the food industry.
• 15:55
  Companies like Nestle have launched lines of prepared meals specifically aimed at patients
• 16:00
  taking GLP-1 drugs, such as Ozempic and Wegovy.
• 16:04
  These meals include pizzas, sandwich melts, and high-protein pasta bowls, each priced at five
• 16:10
  United States dollars.
• 16:12
  Similarly, executives at yogurt company Danone and beverage giant Coca-Cola have started
• 16:18
  promoting high-protein, low-sugar products to appeal to this growing market segment.
• 16:25
  However, not everyone is convinced that these ultra-processed foods are beneficial.
• 16:30
  Abby Langer, a Toronto-based registered dietitian, argues that packaging more
• 16:35
  ultra-processed food as Ozempic-friendly is counterproductive.
• 16:39
  She points out that ultra-processed foods can lead to weight gain in the first place, and
• 16:44
  people taking these drugs might miss out on essential nutrients like protein, fiber, and
• 16:50
  important micronutrients because they are consuming less food overall.
• 16:55
  While Langer acknowledges that having convenient meal options can be helpful, she
• 17:00
  always recommends cooking fresh meals at home for weight loss.
• 17:04
  Switching gears, let us address some common myths about sunscreen.
• 17:08
  Experts are raising alarms about misinformation being spread by social media influencers
• 17:13
  regarding sunscreen ingredients.
• 17:16
  Dermatologist Julia Carroll from Toronto emphasizes that much of this misinformation
• 17:20
  fails to consider the comprehensive medical literature supporting the safety of sunscreen
• 17:26
  ingredients.
• 17:27
  Carroll, who has over 20 years of training and education, finds it frustrating that
• 17:31
  influencers with large followings can sway public opinion so easily.
• 17:36
  One recent example involves a TikTok influencer advising their 570,000 followers to avoid any
• 17:43
  sunscreen with ingredients other than zinc.
• 17:47
  This type of advice overlooks a fundamental principle: it is the dose that makes the
• 17:51
  poison.
• 17:53
  Even substances that are toxic in large amounts, like botulinum toxin, can be safe in
• 17:58
  small doses.
• 18:00
  Experts, including Quebec's Order of Chemists, warn that claims about sunscreen ingredients
• 18:06
  causing cancer or other ailments are devoid of scientific basis and can endanger public
• 18:11
  health.
• 18:13
  According to the Canadian Cancer Society, about 65 percent of melanoma cases are due to
• 18:19
  ultraviolet radiation, primarily from the sun.
• 18:22
  If you are concerned about sunscreen, experts recommend reducing sun exposure during peak
• 18:28
  hours of 10 a.m.
• 18:29
  to 3 p.m., seeking shade, and covering up with tightly woven fabric, a wide-brimmed hat, and
• 18:35
  sunglasses.
• 18:36
  Remember, sunscreen not only helps prevent skin cancer but also helps prevent wrinkles,
• 18:42
  potentially saving you money on botox in the future.
• 18:46
  As we wrap up today’s episode, we turn our attention to a deeply concerning issue
• 18:50
  affecting some of the most vulnerable members of our society.
• 18:54
  Canadians with disabilities are facing a dire rental crisis, leaving many in precarious
• 18:59
  living situations.
• 19:01
  One such individual is Sidney Wood, a 41-year-old former correctional officer
• 19:07
  suffering from Post-Traumatic Stress Disorder, who was evicted last month because he could not
• 19:11
  afford the rent for his basement apartment in Edmonton.
• 19:15
  Wood, who receives $1,403 per month from his Canada Pension Plan disability benefits, simply
• 19:22
  could not keep up with the $1,620 monthly rent, especially after separating from his wife in
• 19:28
  March.
• 19:30
  Forced to leave his home, Wood had no choice but to move back to St.
• 19:34
  Theresa Point First Nation in Northern Manitoba.
• 19:37
  There, he now sleeps in a makeshift shed on his father’s property, as the three-bedroom house
• 19:42
  is already occupied by eight people.
• 19:45
  The most heartbreaking part of Wood’s story is his separation from his two teenage children,
• 19:50
  who stayed in Edmonton with family.
• 19:53
  'I just miss my kids so much.
• 19:55
  That’s what kills me,' Wood told CBC News, tears streaming down his face.
• 20:00
  His story is a stark example of the broader issue facing Canadians with disabilities, who
• 20:06
  often struggle to afford rent amid the country’s housing crisis.
• 20:10
  According to Statistics Canada, people with disabilities are more likely to rent than those
• 20:15
  without disabilities and have a lower median income, making it even harder to afford
• 20:20
  available housing.
• 20:22
  A joint report by the Canadian Human Rights Commission and the Office of the Federal
• 20:27
  Housing Advocate found that Canadians with disabilities are over-represented in all
• 20:32
  aspects of inadequate housing and homelessness.
• 20:36
  They are four times more likely to be homeless and nearly twice as likely to live in 'core
• 20:40
  housing need,' which includes housing that is unaffordable, not in good repair, or lacking
• 20:46
  sufficient space.
• 20:48
  The report also highlighted that 44 percent of renters with disabilities were estimated to be
• 20:54
  living in unaffordable situations, compared to 34.6 percent of the total population.
• 21:01
  With a median income of $35,700 in 2023, affordable housing options for people with
• 21:08
  disabilities are scarce.
• 21:10
  A recent CBC News analysis found that less than one percent of rentals are both vacant and
• 21:16
  affordable for the majority of the country’s renters.
• 21:20
  This situation becomes even more dire for those with disabilities.
• 21:25
  William Rollins, a 42-year-old Toronto resident, faces a similar struggle.
• 21:30
  He was served an N-12 eviction notice, which forces tenants to move out if the owner or
• 21:37
  their immediate family wants to move in.
• 21:40
  Rollins, who is on long-term disability due to multiple diagnoses, including Autism Spectrum
• 21:45
  Disorder, depression, anxiety, and tachycardia, says he is constantly worried about his future.
• 21:51
  'The fear of not knowing where I’ll live next is debilitating,' Rollins told CBC News.
• 21:56
  'If I am evicted, I have very few options.
• 21:59
  Without a substantial safety net, the likelihood of becoming homeless is very real.'
• 22:05
  The barriers people with disabilities face when attempting to secure affordable housing are
• 22:11
  significant and multifaceted.
• 22:13
  They include a lack of support that allows for community living, inaccessibility,
• 22:19
  discrimination, and a definition of affordability that does not consider the
• 22:24
  financial realities of people with disabilities.
• 22:27
  Gabrielle Peters, a disabled writer and policy analyst, pointed out that people with
• 22:32
  disabilities are more likely to live in poverty and at deeper levels of poverty.
• 22:39
  The distress caused by housing instability can be particularly upsetting and disruptive for
• 22:44
  individuals with disabilities.
• 22:47
  Annie Hodgins, Executive Director of the Canadian Centre for Housing Rights, emphasized
• 22:52
  that the housing crisis is causing a great deal of stress, with about a third of the people who
• 22:59
  call their rent hotline each year having disabilities.
• 23:03
  For Sidney Wood, the distress of being separated from his children, sleeping in a
• 23:08
  shed, and not knowing what to do next has severely worsened his mental health.
• 23:14
  Wood developed Post-Traumatic Stress Disorder from the violence he witnessed as a corrections
• 23:19
  officer and now suffers from anxiety and depression.
• 23:24
  'It’s killing me,' he said, tearfully recounting his struggles.
• 23:30
  As we conclude today’s episode, it is crucial to remember the vulnerable populations who are
• 23:35
  often overlooked in discussions about housing and healthcare.
• 23:40
  The stories of Sidney Wood and William Rollins shed light on the urgent need for more
• 23:46
  inclusive and supportive housing policies that address the unique challenges faced by people
• 23:52
  with disabilities.
• 23:53
  Thank you for joining us on Morning Medical with Dr.
• 23:56
  Chow.
• 23:57
  Stay informed, stay healthy, and we will see you next time.