Latest / Becker’s Healthcare Podcast / AI in Spine Surgery: Preserving Clinical Judgment in a Data-Driven Era

Transcript

AI in Spine Surgery: Preserving Clinical Judgment in a Data-Driven Era · Becker’s Healthcare Podcast

• 0:00
  This is Carly Beam with Becker Spine and
• 0:02
  Orthopedics podcast. And joining us today is doctor
• 0:05
  Morgan L'Oreal, a spine surgeon and past president
• 0:09
  of the International Society for the Advancement of
• 0:12
  Spine Surgery.
• 0:13
  And today, we're gonna have a great chat
• 0:15
  about,
• 0:16
  AI and spine surgery and where AllSource kinda
• 0:19
  fits in the larger
• 0:21
  health care equation. So, doctor Lorio, thank you
• 0:23
  so much for joining us.
• 0:25
  Thank you, Carly. It's a pleasure to be
• 0:27
  here. Oh, it's great to have you back
• 0:29
  on the podcast. And before we dive into
• 0:31
  our questions, could you just give a a
• 0:33
  quick overview of your background and your work?
• 0:37
  Okay. Well, I've, been a spine surgeon and
• 0:40
  a hand surgeon
• 0:41
  and,
• 0:43
  past president of ISAS. And,
• 0:46
  I did a lot of work,
• 0:47
  for policy,
• 0:49
  and I chaired coding and reimbursement task force
• 0:51
  for ISAS as well. So that's pretty much
• 0:54
  been my interest, and
• 0:55
  AI kind of,
• 0:57
  fits in here because of its impact on
• 1:00
  policies is is it impacts,
• 1:02
  both health care delivery
• 1:04
  and,
• 1:05
  frankly,
• 1:07
  what we write,
• 1:08
  and what we do moving forward.
• 1:11
  So, hopefully, this is relevant.
• 1:13
  Absolutely. And, I mean, obviously, AI is is
• 1:16
  is everywhere
• 1:17
  in health care and health care conversations.
• 1:20
  I'd love to hear start off how you
• 1:22
  think about its role when it comes to
• 1:24
  spine surgery.
• 1:27
  Well, I I use AI, and I value
• 1:30
  it. It's an extraordinary
• 1:31
  analytical tool.
• 1:33
  If you step back for a moment, the
• 1:35
  scale of this technology,
• 1:37
  becomes clearer.
• 1:38
  We remember the Internet took decades to reshape
• 1:41
  society,
• 1:42
  and today,
• 1:43
  we recognize the lack of Internet access represents
• 1:46
  a form
• 1:47
  of inequality. And so artificial intelligence is arriving
• 1:51
  very differently.
• 1:52
  Its growth isn't gradual. It's almost vertical. And
• 1:55
  now that capability sits inside the devices
• 1:58
  we carry in our pockets.
• 2:00
  But I think the real risk,
• 2:03
  is often misunderstood.
• 2:05
  By that, I mean, the real
• 2:07
  risk isn't that AI will get things wrong,
• 2:10
  that it will succeed
• 2:11
  where we
• 2:13
  while we slowly
• 2:14
  forget how.
• 2:16
  In spine care, the the highest value work
• 2:19
  often happens
• 2:21
  before any incision is made.
• 2:24
  That's when a surgeon integrates imaging,
• 2:27
  biomechanics,
• 2:28
  patient history,
• 2:30
  comorbidities,
• 2:32
  expectations,
• 2:33
  and risk tolerance into a coherent
• 2:35
  clinical judgment.
• 2:37
  Mhmm. Those cognitive steps are difficult to measure
• 2:41
  and rare or rarely reimbursed
• 2:43
  in proportion to their importance,
• 2:46
  but they determine
• 2:48
  whether the right patient
• 2:50
  receives the right intervention
• 2:52
  at the right time.
• 2:54
  It's really fascinating way of looking at it
• 2:56
  too, just the idea that AI
• 2:58
  is coming in before,
• 3:00
  the actual surgery. But, obviously, some people
• 3:04
  are worried that AI might replace their jobs
• 3:07
  and including,
• 3:08
  you know, physicians. Do you see that being
• 3:10
  a concern?
• 3:12
  Well, I I don't think that replacement's the
• 3:14
  real issue,
• 3:16
  specifically for surgeons.
• 3:18
  Health care
• 3:19
  already operates through algorithms,
• 3:21
  reimbursement
• 3:22
  rules,
• 3:23
  prior authorization
• 3:24
  systems,
• 3:25
  coverage policies,
• 3:27
  and outcome reporting structures.
• 3:30
  But I can say this, AI will not
• 3:32
  enter medicine neutrally.
• 3:35
  Artificial intelligence will likely arrive through those same
• 3:38
  systems I mentioned.
• 3:40
  Unfortunately,
• 3:42
  skill erosion
• 3:43
  rarely shows up in the metrics we measure.
• 3:47
  Accuracy can improve
• 3:49
  while human judgment quietly
• 3:52
  declines.
• 3:53
  So I guess the real question becomes governance.
• 3:57
  Yeah. Can you dive into that more governance
• 3:59
  specifically?
• 4:01
  Yeah.
• 4:02
  Complex patients don't always fit neatly
• 4:06
  inside efficient systems.
• 4:08
  Historically,
• 4:09
  the surgeon in the loop judgment
• 4:12
  has allowed physicians to recognize
• 4:15
  when a patient doesn't fit the algorithm
• 4:18
  and to treat the patient
• 4:21
  rather than the pathway.
• 4:23
  Medicine requires flexibility because
• 4:26
  real patients are real complicated.
• 4:29
  Systems, however,
• 4:31
  often prefer
• 4:33
  predictability.
• 4:35
  Maintaining physician oversight, therefore,
• 4:37
  becomes critical as AI becomes embedded in those
• 4:41
  systems.
• 4:42
  Mhmm. It's an interesting balance, it sounds like,
• 4:45
  between the algorithms and then
• 4:48
  figuring out those one off cases.
• 4:51
  Can you give an example of where physician
• 4:53
  judgment still
• 4:54
  matters even when AI is helping out as
• 4:57
  a tool?
• 4:58
  Sure.
• 5:00
  Well, I I recently explored a cancer some
• 5:03
  cancer trend data,
• 5:05
  within an AI system.
• 5:07
  And when I asked whether cancer incidence had
• 5:10
  changed after COVID,
• 5:12
  the analysis suggested
• 5:14
  that incident rates were largely unchanged.
• 5:18
  Mhmm. But when I examined the time series
• 5:21
  more closely,
• 5:22
  something about the pattern,
• 5:25
  didn't quite fit in the data points.
• 5:27
  The COVID years were clearly disruptive,
• 5:30
  so we treated that period as difficult to
• 5:33
  interpret
• 5:34
  and step back to examine broader trends.
• 5:38
  When we then asked
• 5:40
  a different question focusing on cancer mortality
• 5:44
  instead of incidence,
• 5:45
  the pattern appeared different. Mhmm. Mortality
• 5:48
  seemed meaningfully higher than earlier baseline patterns
• 5:53
  roughly
• 5:54
  on the order of about sixteen percent.
• 5:57
  That is a substantial
• 5:59
  change.
• 6:00
  The the AI system
• 6:02
  hadn't really made a mistake per se. It
• 6:05
  had answered the question it was asked,
• 6:08
  but but it reminded me of something important.
• 6:12
  AI answers questions
• 6:14
  very well.
• 6:16
  Sometimes it narrows
• 6:17
  answers
• 6:18
  simultaneously.
• 6:20
  Physicians
• 6:21
  still have to decide
• 6:22
  which questions matter.
• 6:24
  Sometimes the insight comes from recognizing that the
• 6:27
  question itself
• 6:29
  needs to change.
• 6:31
  You're a fascinating idea there, and we'll circle
• 6:33
  back to that in a moment. I also
• 6:34
  wanna pick your brain on health systems and
• 6:37
  how many of them often they hope that
• 6:40
  AI will make care more efficient.
• 6:43
  What are your thoughts on that?
• 6:44
  Well, efficiency is important, but efficiency can also
• 6:48
  introduce blind spots.
• 6:50
  Sometimes I think about this using a GPS
• 6:53
  analogy.
• 6:54
  Mhmm. You know, navigation systems
• 6:57
  optimize
• 6:58
  routes based on filters,
• 7:00
  minimizing time, avoiding tolls, reducing distance.
• 7:04
  But, occasionally, those filters lead you into loops
• 7:07
  or take you off the main road only
• 7:09
  to bring you back again later.
• 7:12
  The system is optimized
• 7:14
  or optimizing
• 7:15
  based on the parameters
• 7:16
  it was given.
• 7:18
  AI functions in a similar way.
• 7:21
  AI can optimize the route,
• 7:24
  but someone still has to decide
• 7:26
  where we're going. And as we move forward,
• 7:29
  these filters, they're gonna are gonna become more
• 7:32
  and more important.
• 7:34
  Yeah. And, doctor L'Oreal, can you talk about
• 7:36
  the role you physicians should play
• 7:39
  as AI is becoming more integrated into the
• 7:41
  work they do?
• 7:43
  Yeah.
• 7:44
  So physicians need to remain actively involved in
• 7:47
  the governance of these systems,
• 7:49
  not just as users,
• 7:51
  but as stewards.
• 7:52
  Complex patients often generate more uncertainty
• 7:55
  and variability in outcomes,
• 7:58
  but those are frequently the patients who need
• 8:00
  the most thoughtful care.
• 8:02
  If systems begin to prioritize
• 8:05
  predictability
• 8:05
  above all else,
• 8:07
  there's a risk
• 8:08
  that the most complex patients
• 8:11
  may become harder to treat.
• 8:13
  Just
• 8:14
  over a year ago, I had the opportunity
• 8:17
  to participate
• 8:18
  in a global
• 8:20
  fifty roundtable organized by the Dubai Future Foundation,
• 8:24
  where experts from multiple sectors were asked to
• 8:26
  consider possible futures
• 8:28
  roughly five decades from now.
• 8:31
  My perspective in that discussion came from two
• 8:33
  places.
• 8:35
  My international policy work in spine surgery through
• 8:38
  ISAS,
• 8:39
  and my experience as a practicing surgeon,
• 8:44
  one idea
• 8:46
  emerging from that initiative is a classification framework
• 8:49
  describing how humans
• 8:50
  and artificial intelligence collaborate in the production of
• 8:53
  knowledge and decisions. Mhmm.
• 8:56
  The the framework is a simple visual markers
• 8:59
  to indicate the degree of human involvement
• 9:02
  in AI assisted work ranging from fully human
• 9:04
  activity to machine dominant processes
• 9:07
  with several intermediate stages representing different levels of
• 9:11
  collaboration and oversight.
• 9:13
  The purpose
• 9:14
  is transparency.
• 9:16
  As artificial intelligence becomes more deeply embedded in
• 9:19
  research, analysis, and communication,
• 9:22
  it will become increasingly difficult to determine how
• 9:24
  much of an output reflects human judgment versus
• 9:27
  machine computation.
• 9:29
  Mhmm. A clear labeling system helps make that
• 9:32
  relationship
• 9:33
  visible.
• 9:34
  From a clinical perspective,
• 9:36
  this aligns closely with something we have long
• 9:39
  emphasized in medicine,
• 9:41
  surgeon in the loop governance.
• 9:44
  Artificial intelligence may assist with analysis,
• 9:47
  pattern recognition,
• 9:48
  and data synthesis, but clinical responsibility
• 9:52
  and decision authority
• 9:54
  remain human.
• 9:56
  In the terminology of the Dubai framework,
• 9:59
  this approach would be classified as human led
• 10:02
  AI.
• 10:03
  A fascinating way to describe it, human led
• 10:06
  AI.
• 10:07
  And can you talk about
• 10:09
  how AI
• 10:10
  might influence the way medical knowledge itself is
• 10:13
  developing?
• 10:15
  Yeah. There's another dimension to this conversation that
• 10:18
  we don't always acknowledge.
• 10:20
  AI
• 10:21
  increasingly influences how we write,
• 10:24
  whether drafting clinical notes, summarizing literature,
• 10:27
  or producing scientific manuscripts.
• 10:30
  When technology
• 10:32
  influences how we write,
• 10:34
  it inevitably
• 10:36
  influences what we write.
• 10:38
  Knowing who
• 10:39
  or what
• 10:41
  produced it is therefore important.
• 10:43
  This can be categorized across six levels,
• 10:47
  human only,
• 10:49
  human directed,
• 10:50
  human led,
• 10:52
  machine assisted,
• 10:54
  machine dominant,
• 10:55
  and machine only.
• 10:57
  In fact, journals are already starting to formalize
• 11:00
  this principle.
• 11:01
  At the International Journal of Spine Surgery,
• 11:04
  new author guidelines make it clear that AI
• 11:07
  can assist
• 11:08
  with drafting or organization,
• 11:10
  but scientific
• 11:11
  judgment
• 11:12
  and clinical interpretation
• 11:14
  must remain human led.
• 11:17
  And over that time,
• 11:19
  that shapes will be collectively known,
• 11:21
  What we believe
• 11:23
  and ultimately how we practice medicine.
• 11:27
  AI systems apply filters,
• 11:30
  optimization
• 11:31
  rules,
• 11:32
  and administrative constraints.
• 11:35
  Capability
• 11:36
  will proceed
• 11:38
  consent,
• 11:39
  and adoption
• 11:40
  will proceed reflection.
• 11:42
  Most importantly,
• 11:44
  those filters can gradually direct the flow of
• 11:47
  conversation, emphasizing some questions
• 11:50
  while quietly discouraging others. I can't emphasize how
• 11:53
  important and dangerous these filters are and may
• 11:55
  become.
• 11:56
  When systems rewrite rather than carefully amend existing
• 12:00
  work, portions of the original meaning can quietly
• 12:03
  disappear.
• 12:04
  The danger isn't censorship or deception.
• 12:06
  The danger is continuous optimization where erosion of
• 12:10
  skills slowly becomes framed as efficiency.
• 12:13
  And that matters in medicine because the way
• 12:16
  knowledge is framed ultimately shapes
• 12:18
  how diseases are defined,
• 12:21
  classified,
• 12:22
  and treated.
• 12:23
  Definitely. And I like what you said about
• 12:26
  what you're doing
• 12:27
  at ISAS and then you in the journal
• 12:29
  and how you're,
• 12:31
  kind of
• 12:32
  making guidelines based off of what we know
• 12:35
  now with AI. Can you dive deeper into
• 12:38
  how all this relates specifically to spine care?
• 12:41
  Sure. Well, in spine care, this matters because
• 12:44
  chronic low back pain, as I mentioned, is
• 12:47
  not a single disease, for instance. It includes
• 12:49
  multiple phenotypes,
• 12:51
  discogenic pain, retibrogenic pain, sacroiliac dysfunction,
• 12:55
  multifidus muscle failure,
• 12:58
  neuropathic and radicular pain,
• 13:00
  and central sensitization patterns.
• 13:03
  AI can help us recognize these phenotypes earlier
• 13:06
  and more consistently.
• 13:08
  But if surrounding systems favor simplified treatment pathways
• 13:13
  or if they narrow outcome metrics,
• 13:16
  those distinctions can gradually
• 13:18
  disappear from the conversation.
• 13:20
  So the issue isn't simply whether AI helps
• 13:23
  analyze data. It's whether physicians remain actively
• 13:28
  engaged
• 13:29
  in shaping how knowledge itself is generated and
• 13:32
  applied.
• 13:33
  Got it. And doctor Lohr, I want to
• 13:36
  also touch on this idea of machine memory
• 13:39
  versus human meaning. Can you dive into that
• 13:42
  and just, you know, the kind of the
• 13:44
  tensions that exist in that realm?
• 13:46
  Okay. Well,
• 13:48
  speaking to that tension as well, artificial,
• 13:51
  intelligence systems accumulate information
• 13:54
  indefinitely.
• 13:56
  They do not forget.
• 13:58
  Humans
• 13:59
  are very different.
• 14:01
  Our memory is finite.
• 14:03
  We cannot retain everything, and so we must
• 14:06
  choose what is worth remembering.
• 14:09
  What matters is remembering the sequence.
• 14:13
  What humans did before systems,
• 14:15
  when judgment began to be deferred,
• 14:18
  and when dependence
• 14:19
  quietly
• 14:21
  replace skill.
• 14:22
  Maintaining skill will be crucial if humans are
• 14:24
  to provide resilience
• 14:26
  for this new AI ecosystem.
• 14:30
  That act
• 14:31
  of selection
• 14:32
  deciding
• 14:33
  which observations
• 14:35
  matter enough to carry forward is actually a
• 14:37
  fundamental part of how knowledge develops in medicine.
• 14:42
  AI may store and synthesize vast amounts of
• 14:45
  information,
• 14:46
  but physicians still determine which patterns are meaningful
• 14:49
  and which insights
• 14:50
  deserve to shape future care. AI remembers everything,
• 14:54
  but bureaucracies
• 14:56
  bureaucracies, they decide what is allowed to matter.
• 15:00
  Artificial intelligence
• 15:02
  didn't invent these these tendencies.
• 15:04
  It simply accelerated them.
• 15:06
  That's very well said, doctor L'Oreal. And before
• 15:09
  we wrap up, if you had to summarize
• 15:11
  a key lesson for physicians
• 15:13
  about AI, what would it be?
• 15:15
  Okay. Well, artificial intelligence is gonna become
• 15:19
  extraordinarily good in answering questions.
• 15:23
  The future of medicine will depend on whether
• 15:25
  physicians continue to ask the right ones.
• 15:28
  Technology can analyze the data.
• 15:31
  Physicians still will have to recognize,
• 15:34
  the meaning.
• 15:36
  The real risk isn't that AI
• 15:38
  will replace
• 15:40
  physicians.
• 15:41
  The real risk
• 15:43
  is that it will slowly replace
• 15:45
  the habits of thinking
• 15:47
  that make physicians
• 15:50
  necessary.
• 15:51
  And in that vacuum,
• 15:53
  no one will be left to recognize or
• 15:56
  remember the meaning.
• 15:58
  Very well said. Well, doctor Lorio, thank you
• 16:01
  again for joining us in the podcast today.
• 16:03
  It's a great conversation, and I hope to
• 16:05
  connect again in the future. Thank you, Carly.
• 16:07
  I appreciate this conversation and time to be
• 16:09
  with you.