Paper Trail

The Cognitive Cost of the ER: When Doctors Stop Thinking and Start Doing

April 10, 202619:19Paper Trail

This episode explores how a doctor's cognitive state, rather than solely a patient's physical condition, dramatically influences medical decisions and outcomes in the emergency room. It introduces the concept of "rational inattention," explaining that when physicians face high cognitive loads, they tend to substitute internal "thinking" with ordering more external diagnostic tests ("doing"). Listeners will learn how groundbreaking research challenges the perception of objective medical care and redefines the role of a physician's attention in high-stakes environments.

Key Takeaways

Detailed Report

The Hidden Cost of Cognitive Load in the ER

A groundbreaking NBER working paper, "Thinking versus Doing: Cognitive Capacity, Decision Making and Medical Diagnosis," reveals a profound and often overlooked factor influencing patient outcomes in emergency rooms: the cognitive state of the treating physician. This interdisciplinary research, spearheaded by health economists Benjamin Handel and Jonathan Kolstad, behavioral economist Ulrike Malmendier, and rational inattention theorist Filip Matějka, demonstrates that a doctor's mental fatigue can dramatically alter a patient's fate, even when symptoms and medical history are identical.

The study posits that a patient's likelihood of hospital admission can increase by 28% simply because of the doctor's internal cognitive state—specifically, how many other complex patients they are juggling at that precise moment. This challenges the traditional view of medical care as a perfectly objective process, highlighting that human attention is a finite and scarce resource, even in data-rich environments like the ER.

Thinking Versus Doing: A New Framework

The researchers introduce a "thinking versus doing" framework to understand medical decision-making. "Thinking" involves deep mental effort, such as synthesizing clues, performing differential diagnoses, and delving into a patient's nuances. "Doing," conversely, refers to ordering external diagnostic tests like blood work, CT scans, and X-rays.

The core hypothesis is that as a doctor's cognitive load increases—due to factors like managing multiple complex patients, blaring alarms, or long shifts—the marginal cost of "thinking" skyrockets. When deep mental effort becomes prohibitively expensive for the brain, physicians substitute internal "thinking" with external "doing," relying on tests to perform the diagnostic heavy lifting.

Unraveling the Data: The Genius of Methodology

Proving this substitution effect is challenging due to the "endogeneity problem": how do you differentiate between a patient's severity and the doctor's cognitive state? The paper's empirical genius lies in its methodology. Leveraging the quasi-random patient assignment systems common in modern emergency departments, the researchers could statistically hold the patient variable constant. This allowed them to isolate the physician's cognitive load as the key variable.

To measure real-time cognitive load, the team gained unprecedented access to highly granular Electronic Medical Record (EMR) and audit-log data. These audit logs, which record every mouse click, chart opened, lab result reviewed, and note typed by a physician, provided minute-by-minute "cognitive load profiles" for each doctor. This real-time, field-based evidence offers a rare glimpse into human cognition operating under high-stakes pressure.

The Four-Part Behavioral Shift

The data revealed a distinct four-part behavioral shift aligning perfectly with the "Thinking vs. Doing" model:

  • Increased Total Tests: When cognitive capacity drops, the sheer volume of diagnostic tests ordered significantly increases.
  • Reduced Targeted Tests: Doctors order fewer highly specific, uncommon tests that require significant cognitive effort to recall and justify.
  • Increased Generic Tests: There is a massive reliance on common, generic tests (e.g., standard metabolic panels, comprehensive blood counts, broad CT scans) that require minimal cognitive friction to order, often just a single click in the EMR.
  • Increased Diagnostic Uncertainty: Paradoxically, this reliance on broad, generic "doing" actually *increases* uncertainty in the physicians' diagnostic beliefs. Instead of providing clarity, generic tests often yield "incidentalomas"—mildly abnormal values or artifacts unrelated to the primary complaint—which further burden the exhausted doctor with noisy, ambiguous data.

The Chilling Consequence: Higher Admissions

The most impactful finding is the direct link between cognitive load and patient outcomes. A physician in their highest decile of cognitive load increases patient admissions by 28% compared to when that exact same physician is in their lowest cognitive load decile, even for the exact same patient presenting with identical symptoms.

This isn't a sign of laziness but a deeply human coping mechanism rooted in risk and uncertainty. Discharging a patient requires high confidence and cognitive effort. Admitting a patient, conversely, often serves as a "safe" default when diagnostic uncertainty is high, deferring the final diagnosis to an inpatient team. It's an act of "doing" to avoid the cognitive burden of "thinking" through a complex discharge decision. This systemic cost, both financial and psychological for the patient, is enormous.

Policy Implications: Protecting Physician Brainpower

The research offers compelling policy implications for improving healthcare delivery:

  • Algorithmic Patient-Physician Matching: Instead of quasi-random assignment, AI-driven triage systems could use real-time cognitive load profiles to match complex patients (requiring deep "thinking") with doctors whose load is currently low. Conversely, doctors under high load could be assigned more straightforward, protocol-driven cases.
  • Rethinking Productivity Metrics: Hospital administrators typically measure productivity by "Patients Per Hour." This paper suggests these metrics are incomplete, as pushing doctors into high cognitive load can lead to unnecessary tests and admissions, destroying systemic value.
  • Structural Changes: Hospitals might need to rethink shift lengths, incorporate mandatory cognitive reset periods, or implement overlapping shifts to ensure fresh doctors are available for complex diagnostic thinking.

Ultimately, this research urges a paradigm shift: viewing a doctor's attention not as an infinite given, but as a finite, measurable, and depletable medical resource. AI, in this context, isn't about replacing doctors but acting as a "cognitive co-pilot," augmenting human decision-making by reducing the marginal cost of "thinking" and preventing the trap of generic "doing."

Future Questions

While profound, the research also raises critical questions: What are the costs and challenges of implementing algorithmic load-balancing in legacy hospital systems? Would doctors accept algorithms monitoring their cognitive state? What can patients or advocates do when faced with an over-taxed physician? And what are the legal and ethical implications for medical malpractice if a patient's outcome is directly tied to a physician's cognitive capacity? These questions pave the way for redefining healthcare in the age of data and human cognitive limits.

Show Notes

Works Referenced

Glossary

  • Cognitive Load: The total amount of mental effort being used in the working memory at a given time. High cognitive load means a person's brain is juggling many complex tasks simultaneously.
  • Rational Inattention: An economic theory suggesting that individuals or organizations choose to pay less attention to certain information when the mental cost of processing it outweighs the perceived benefit, even if the information is available.
  • Thinking versus Doing Framework: A model describing how physicians make diagnostic decisions, where 'thinking' involves deep mental effort and synthesis, while 'doing' involves ordering external diagnostic tests.
  • Endogeneity Problem: A statistical issue where a variable that is supposed to be an independent cause is actually influenced by the outcome variable, making it difficult to determine true causality.
  • Electronic Medical Record (EMR): A digital version of a patient's chart, containing their medical and treatment history from a single practice or hospital.
  • Audit Logs: Detailed, timestamped records of all digital interactions within a system, such as mouse clicks, chart openings, and data entries, used to track user activity.
  • Incidentalomas: Unexpected, often benign, findings on medical imaging or lab tests that are unrelated to the patient's primary symptoms or reason for the test.
  • Relative Value Units (RVU): A measure of the value of a medical service based on the physician's work, practice expenses, and malpractice insurance costs. Often used to determine physician compensation and productivity.

Sources / References

Full Transcript

HostSo, imagine this scenario: Two patients walk into the exact same emergency room with the exact same symptoms, the exact same medical history. They're seen by the exact same doctor. But one of them is 28% more likely to be admitted to the hospital, to have a multi-day inpatient stay, simply because of *when* they arrived.
ExpertAnd that "when" isn't about the time of day, or the day of the week, but about the doctor's internal cognitive state. It’s about how many other complex patients that physician is juggling at that precise moment.
HostWait, so you're telling me that a patient's fate can be so dramatically altered not by their physical condition, but by the doctor's brain being tired? That's… staggering.
ExpertIt's more than staggering, it's a fundamental challenge to how we perceive medical decision-making. This new NBER working paper, "Thinking versus Doing: Cognitive Capacity, Decision Making and Medical Diagnosis," completely redefines the role of a physician's attention in high-stakes environments like the ER.
HostAnd it's not just some speculative theory, right? This is rigorous, data-driven research.
ExpertAbsolutely. The team behind it is a true interdisciplinary dream. You have health economists like Benjamin Handel and Jonathan Kolstad, alongside behavioral economics powerhouse Ulrike Malmendier, and Filip Matějka, a leading theorist in what's called "rational inattention."
Host"Rational inattention"—that's a term I've usually heard applied to financial markets or consumer behavior. What does it mean in the context of an emergency room?
ExpertExactly. That's what makes this paper so groundbreaking. Traditionally, we've thought of medical care as almost perfectly objective. A patient presents, a doctor, acting as this perfectly rational diagnostic machine, applies their training, and arrives at a diagnosis. This paper shatters that illusion. Rational inattention, at its core, is about recognizing that even when information is abundant – and believe me, in an ER, information is _everywhere_ – the human attention required to process it is strictly limited. It's a scarce resource.
HostSo, it's not that the doctor *couldn't* process the information, it's that the *cost* of processing it, the mental effort, becomes too high when their cognitive load is through the roof.
ExpertPrecisely. And what the researchers propose is this fascinating "thinking versus doing" framework within the diagnostic process. When a physician sees a patient, they can either engage in "thinking"—which involves deep mental effort, synthesizing clues, differential diagnosis, really digging into the nuances of a patient's history. Or, they can engage in "doing"—which means ordering external diagnostic tests like blood work, CT scans, X-rays.
HostOkay, so "thinking" is the internal, high-effort cognitive work. "Doing" is the external, ordering-a-test kind of work.
ExpertRight. And the paper’s theoretical model posits that as a doctor's cognitive load increases—say, they're juggling five complex patients, alarms are blaring, they’re eight hours into a demanding shift—the marginal cost of that "thinking" effort skyrockets. It just becomes prohibitively expensive for their brain to do the deep dive.
HostSo, if they still have to reach a diagnostic conclusion, what do they do? They substitute.
ExpertThey substitute! They replace that internal "thinking" with external "doing." They order tests to do the diagnostic heavy lifting for them. The paper puts it beautifully, saying that the use of diagnostic tests provides "a valuable empirical window into the underlying cognitive mechanisms that precede testing."
HostThat’s such a powerful framing. It makes perfect sense, but it's not how we typically talk about medical care. It brings to mind that analogy the paper uses – a detective. A fresh detective, alert and focused, might solve a case by deeply analyzing a suspect's alibi, noticing a subtle twitch, connecting abstract dots, that's the "thinking."
ExpertAnd the exhausted detective, lacking the bandwidth for that deep deduction, might just say, "Alright, forensics, dust the entire building for fingerprints, pull every security camera feed," hoping that the sheer volume of external data will solve the case for them. That’s the "doing."
HostThat's a perfect analogy. But here's the huge hurdle I see in proving this: the endogeneity problem. If a doctor orders a ton of tests, isn't the most obvious explanation that the patient is just really, really sick? How do you disentangle the patient's severity from the doctor's cognitive state?
ExpertThat's the empirical genius of this paper. It's a massive hurdle, and it's where their methodology truly shines. They leveraged the unique operational structure of modern emergency departments. In most EDs, patients aren't hand-selected by doctors. They're assigned through what's essentially quasi-random variation—a round-robin system, or whoever's next available.
HostSo, the complexity of patients seen by a doctor at the start of their shift is statistically similar to those at the end, or during a surge. The *patient* variable is held relatively constant.
ExpertExactly. That allows them to isolate the *physician's cognitive load* as the key variable. Now, how do you measure "cognitive load" in a real-world, high-stakes environment? This is where they got access to an absolute treasure trove of data: highly granular Electronic Medical Record, or EMR, and audit-log data.
HostAudit logs? What is that exactly?
ExpertThink of it as the digital exhaust of a modern hospital. Every mouse click, every chart opened, every lab result reviewed, every note typed by a physician – it's all timestamped and recorded. By analyzing these audit logs, the researchers could construct real-time, minute-by-minute "cognitive load profiles" for each physician.
HostSo they could see how many patients a doctor was concurrently managing, how complex those patients were, how quickly they were context-switching between charts, all in real-time?
ExpertYes. It's unprecedented. This isn't a lab experiment where undergrads are asked to memorize numbers while solving math problems. This is real-time, field-based evidence of human cognition operating under the highest possible stakes, tracked via the invisible digital footprints doctors leave in the software. It's truly rare and incredibly exciting for anyone interested in decision-making under pressure.
HostThat's wild. It's the kind of data every behavioral economist probably dreams of getting their hands on. So, once they isolated this cognitive load variable, what did they find about how testing behavior changed? When doctors were mentally maxed out, what did their "doing" look like?
ExpertThe data revealed a very distinct, four-part behavioral shift that perfectly aligns with their "Thinking vs. Doing" substitution model. First, and perhaps least surprisingly, there was a significant increase in the *total number* of diagnostic tests ordered. When cognitive capacity drops, the sheer volume of "doing" goes up. Doctors outsource that diagnostic work to the lab and radiology.
HostOkay, so more tests overall. That makes sense if you’re trying to compensate for less thinking.
ExpertBut here's where it gets interesting. Second, they found a *reduction* in targeted, uncommon tests. Ordering a highly specific assay for a rare disease, for example, requires significant "thinking." The doctor has to recall specific pathophysiology, remember the name of that niche test, and justify its use. When bandwidth is low, the brain just abandons that high-effort recall.
HostSo they're not just blindly ordering *everything*. They're being selective in what they *stop* ordering.
ExpertExactly. Which leads to the third point: an *increase* in common, generic tests. Instead of the sniper rifle, the exhausted doctor uses a shotgun. The data showed a massive reliance on things like standard metabolic panels, comprehensive blood counts, broad CT scans. These are tests that require almost zero cognitive friction to order. They're often pre-bundled in the EMR software, a single click. The doctor casts a wide net, hoping the generic tests will catch whatever is wrong.
HostThat's a fascinating distinction. The easy, generic tests go up, the hard, targeted ones go down. But then, what's the outcome of all this "doing"? Does all that extra testing lead to clearer answers?
ExpertThis is perhaps the most counter-intuitive and frankly, fascinating finding in the entire paper. One might assume that ordering a higher volume of broad, generic tests would eventually lead to a clearer clinical picture, right? More data, more clarity.
HostThat's certainly what I'd expect.
ExpertThe data shows the exact opposite. This substitution effect, this reliance on generic "doing," actually *increases uncertainty* in the physicians' diagnostic beliefs.
HostWait, really? How can more tests lead to *more* uncertainty? That seems completely backwards.
ExpertIt's because when a doctor relies on deep "thinking" and targeted testing, the results usually provide a definitive "yes" or "no" to a very specific clinical question. You order a specific test because you suspect a specific thing, and the result helps you confirm or rule it out.
HostRight, it's a focused inquiry.
ExpertBut when a doctor casts a wide net with generic tests, they inevitably catch what are called "incidentalomas"—mildly abnormal lab values or blurry imaging artifacts that have absolutely nothing to do with the patient's actual emergency. Now, the cognitively exhausted doctor is faced with a barrage of noisy, ambiguous data. They have to spend what little mental energy they have left trying to figure out if that slightly elevated liver enzyme is a sign of a deadly disease or just a meaningless anomaly. The "doing" actually backfires, generating more noise and deepening the physician's diagnostic uncertainty.
HostThat is incredibly ironic. We're in this age of data explosion in medicine, where we think more data automatically equals better medicine. But this paper demonstrates that when a human brain is exhausted, throwing more *generic* data at it doesn't create clarity, it creates a fog of uncertainty. It makes it harder to think, not easier.
ExpertExactly. And that fog of uncertainty has very real, very tangible consequences for patients.
HostWhich brings us back to that 28% figure you mentioned at the top. If the paper just stopped at "tired doctors order more generic tests," it would be interesting. But the researchers followed the data downstream to the ultimate clinical outcome: hospital admissions.
ExpertAnd that's where the real impact of this paper hits home. It reveals that a physician in their *highest decile of cognitive load* increases patient admissions by 28% compared to when that *exact same physician* is in their lowest cognitive load decile.
HostFor the exact same patient. That part is crucial.
ExpertAbsolutely crucial. A patient walks into the ER with moderate chest pain. If they happen to draw a doctor who just started their shift and has a light caseload, they might be thoroughly evaluated, diagnosed, and safely discharged home. But if that *exact same patient* walks in and draws that *exact same doctor* during a chaotic surge with a full board of complex cases, their chance of being admitted to the hospital skyrockets by nearly a third.
HostThat's a truly chilling thought for any patient. Why does admission increase so dramatically? Is it just that doctors are being lazy?
ExpertNot at all. It's a deeply human coping mechanism, and it's rooted in risk and uncertainty. As we discussed, high cognitive load leads to generic testing, which, paradoxically, leads to increased diagnostic uncertainty. Now, think about the options a doctor has: discharge or admit.
HostDischarging a patient means taking on a lot of responsibility. You have to be absolutely certain they're safe to go home, write a detailed plan, coordinate follow-up.
ExpertExactly. Discharging is a high-risk, high-cognitive-effort action. You have to *think* deeply to be confident in that decision. Admitting a patient to the hospital, conversely, is often the ultimate "safe" default when uncertainty is high. It defers the final diagnosis to an inpatient internal medicine team. It's the ultimate act of "doing" to avoid the cognitive burden of "thinking" through a complex discharge. It’s not malice; it's self-preservation in a high-pressure environment.
HostSo, it's not that doctors are acting lazily, they're acting rationally within their cognitive constraints. They're trying to keep patients safe. But the systemic cost, both financially and psychologically to the patient who's unnecessarily hospitalized, is enormous. This really shatters that illusion of objectivity in medicine we all hold onto. We want to believe our medical outcomes are purely dictated by our biology, our symptoms.
ExpertAnd this research proves that a massive variable in our health outcomes is entirely situational, and deeply human: how busy was the doctor right before they walked into my room? That's a profound, almost uncomfortable truth.
HostSo if this is the diagnosis of the problem, what do the authors propose as a treatment? What are the policy implications of understanding a doctor's attention as a finite resource?
ExpertThe paper doesn't just identify a flaw; it explores real-world policy implications. The most futuristic and compelling one, for me, is rethinking how patients are assigned to doctors. Currently, it's often quasi-random, a round-robin.
HostLike a queue. Next available.
ExpertRight. The researchers suggest that patient-physician matching could be vastly improved by algorithmically accounting for real-time "cognitive load profiles."
HostSo, imagine an AI-driven triage system that doesn't just look at the waiting room, but looks at the EMR audit logs of the doctors on shift.
ExpertPrecisely. If a highly complex patient who requires deep diagnostic "thinking" arrives, the algorithm ensures they are assigned to a doctor whose cognitive load profile is currently low. Conversely, if a doctor is currently in their highest decile of cognitive load, the algorithm stops assigning them complex diagnostic mysteries and only assigns them straightforward, protocol-driven cases—like a simple laceration or a standard broken bone—that require pure "doing" but very little deep "thinking."
HostThat's brilliant. It's about protecting the doctor's brainpower. What about hospital administration? How does this change how they measure productivity?
ExpertIt turns it on its head. Hospital administrators typically measure physician productivity in terms of "Patients Per Hour" or Relative Value Units. This paper suggests those metrics are dangerously incomplete because they completely ignore the cognitive constraints that shape true productivity. If pushing a doctor to see one more patient pushes them into that highest cognitive load decile—resulting in a barrage of unnecessary generic tests and a 28% increase in expensive hospital admissions—then pushing for higher volume is actually destroying systemic value.
HostSo, administrators might need to rethink shift lengths, build in mandatory cognitive reset periods, or structure overlapping shifts so that fresh doctors take over complex diagnostic thinking from doctors who are nearing the end of their cognitive stamina.
ExpertExactly. It forces a paradigm shift. We have to stop viewing a doctor's attention as a given, infinite baseline, and start viewing it as a finite, measurable, and depletable medical resource—just like hospital beds, ventilators, or IV fluids. If an ER runs out of beds, they go on diversion. If a doctor runs out of cognitive bandwidth, the system currently forces them to keep going, resulting in this "thinking vs. doing" substitution.
HostAnd in this context, AI isn't about replacing doctors, but about protecting their cognitive bandwidth. If AI can pre-digest EMR data, synthesize complex information, and reduce the marginal cost of "thinking" for the physician, it could prevent doctors from falling into that trap of generic "doing." It's almost like a cognitive support system.
ExpertA cognitive co-pilot. It’s about leveraging technology to augment human decision-making, not just automate it.
HostThis research truly reshapes how we understand medical decision-making under pressure. So, to bring it all together, what are the absolute key takeaways from this paper?
ExpertI'd boil it down to four main points. First, the substitution effect: Under high cognitive load, ER doctors substitute deep mental deliberation, or "thinking," with a higher volume of generic diagnostic tests, or "doing." Second, there's the paradox of testing: Relying on those generic tests doesn't yield a clearer picture; it actually increases uncertainty in the physician's diagnostic beliefs. Third, and perhaps most impactful, is the 28% admission bump: a physician in their highest state of cognitive exhaustion is 28% more likely to admit a patient to the hospital than when they are in their lowest state, even when controlling for patient severity.
HostAnd finally, the methodology itself.
ExpertYes, the fourth key takeaway is the value of this field data. The use of EMR audit logs provides rare, real-time field evidence that medical decision-making is heavily influenced by situational attention and cognitive complexity. It's a game-changer for understanding human decision-making in high-stakes environments.
HostIt definitely makes you look at the ER, and medical care in general, through a completely different lens. It also raises some really provocative questions. For instance, the cost of implementing something like algorithmic load-balancing sounds incredible, but how much would it actually cost to implement in legacy hospital systems? And would doctors even push back against an algorithm monitoring their "cognitive load" via their mouse clicks? That feels a bit Orwellian, even if it's for their benefit.
ExpertAbsolutely, those are critical questions. There's also the question of patient advocacy. Armed with this knowledge, what can a patient or their advocate actually *do* in the ER? Is there a way to politely ask a doctor to engage in more "thinking" rather than just "doing," or is that an impossible request in a system that's already stretched to its limits?
HostAnd the legal and ethical implications are huge. If a patient is unnecessarily admitted, or conversely, improperly discharged because a doctor was in the highest decile of cognitive load, how does that impact medical malpractice and the standard of care? Does the hospital hold liability for over-taxing the physician's cognitive capacity? These aren't just academic questions anymore, they're questions that could redefine healthcare.