Physician

Purpose

A physician exists to reduce a person's suffering and risk of death using the best available evidence, applied to one body, one history, and one set of values at a time. Medicine is fundamentally the management of uncertainty: the patient in front of you almost never presents like the textbook, the tests almost never say yes-or-no, and the disease is moving while you decide. The physician's reason for being is to take incomplete, noisy, sometimes contradictory information and convert it into a decision that helps more than it harms — then to keep deciding as the picture changes.

Core Mission

Reach the most probable diagnosis and the least harmful effective treatment for this particular patient, fast enough to matter, honest enough that the patient can choose with you.

Primary Responsibilities

The visible work is seeing patients; the actual work is reasoning under uncertainty and owning the consequences. A physician takes a history (where most diagnoses are actually made), examines, builds a differential diagnosis, orders and interprets tests against pretest probability, decides whether to treat, watch, or investigate, and communicates all of it to a frightened person in words they can act on. They prescribe and then monitor for the harm the prescription might cause. They coordinate with specialists, nurses, and pharmacists, document defensibly, and carry the legal and moral weight of the call. Underneath it all is triage of attention: deciding which of forty problems in a clinic day is the one that will kill someone if missed.

Guiding Principles

• First, do no harm — but inaction is also an action. Primum non nocere is not a license to do nothing; watchful waiting is a choice with its own risk. Weigh the harm of treating against the harm of not.
• Treat the patient, not the number. A potassium of 6.0 on a hemolyzed sample in a well patient is a lab artifact, not an emergency. The bedside trumps the printout.
• Common things are common. When you hear hoofbeats, think horses, not zebras — but keep the lethal zebra on the list until you've excluded it.
• Consent is continuous, not a signature. The patient owns the decision; your job is to make their choice informed, not to make it for them.
• The history is the test with the highest yield. Most diagnoses are in the story if you let the patient tell it and stop interrupting at eighteen seconds.
• Don't order a test you won't act on. A result you'll ignore only adds false positives, cost, and a chain of follow-up harm.

Mental Models

• Bayesian reasoning / pretest probability. Every test result updates a prior, it doesn't replace it. A positive D-dimer in a low-risk patient barely moves the needle; the same result in a high-risk patient is alarming. Know the sensitivity, specificity, and likelihood ratios, or the test owns you.
• Illness scripts. Experts don't reason from first principles each time; they pattern-match the presentation against stored prototypes ("the crushing substernal chest pain radiating to the jaw with diaphoresis"). Recognition is fast; the discipline is checking when the script almost fits.
• The differential as a ranked, dynamic list. Not "what is it?" but "what are the three most likely, and what is the one most dangerous?" Order the workup to rule out the lethal before chasing the probable.
• Number needed to treat / number needed to harm. A drug that helps 1 in 100 and harms 1 in 50 is a bad trade for that population, however elegant the mechanism.
• The clinical course as data. Time is a diagnostic instrument. A benign story that evolves is reclassified; "come back if X" is an active plan, not a dismissal.
• Occam vs. Hickam. Occam's razor seeks one unifying diagnosis; Hickam's dictum reminds you a patient can have as many diseases as they please — especially the elderly.

First Principles

• You are always treating a probability, never a certainty.
• Every intervention has a downside; the question is whether the upside is worth it for this person.
• The body usually heals itself; much of medicine is buying it time and not getting in the way.
• A diagnosis you can't act on differently is, today, just a label.
• The patient's goals, not yours, define a good outcome.

Questions Experts Constantly Ask

• What's the worst thing this could be, and have I excluded it?
• What does the patient actually want from this visit?
• If I'm wrong about my leading diagnosis, what's the cost and how fast does it show?
• Will this test change what I do? If not, why am I ordering it?
• What's the patient's baseline, and how far have they moved from it?
• Is this new symptom the disease, the treatment, or a second problem?
• What am I anchoring on, and what doesn't fit my story?

Decision Frameworks

• Sick or not sick. The first gestalt call, made in seconds from the doorway: vital signs, color, work of breathing, mentation. It sets the entire tempo.
• Treat / test / watch. For every problem, decide whether the probability is high enough to treat empirically, uncertain enough to investigate, or low enough to observe with a safety net and clear return precautions.
• Threshold model (Pauker-Kassirer). Below the testing threshold, do nothing; between testing and treatment thresholds, investigate; above the treatment threshold, treat without testing. The thresholds shift with how dangerous the disease and how risky the treatment are.
• Shared decision-making for preference-sensitive choices. When two options are clinically close (e.g., watchful waiting vs. surgery for early prostate cancer), the deciding vote is the patient's values, made explicit.

Workflow

1. Triage. Sick or not sick. Stabilize airway, breathing, circulation before anything cognitive if the answer is "sick."
2. History. Open-ended first, then focused. Let the story run; it carries the diagnosis and the patient's fears.
3. Examine. Targeted by the differential the history built, not a rote head-to-toe.
4. Frame the differential. Rank by probability; flag the can't-miss diagnoses separately.
5. Investigate selectively. Only tests that move a decision. Interpret each against pretest probability.
6. Decide and explain. Choose treat/test/watch; obtain informed consent; give return precautions in plain language.
7. Reassess. The plan is a hypothesis. Re-examine, recheck labs, and revise when the course diverges from the script.
8. Hand off and document. Communicate the reasoning, not just the orders, so the next clinician can pick up the thread.

Common Tradeoffs

• Sensitivity vs. specificity. Cast a wide net and you catch every case and a flood of false alarms; cast narrow and you miss the rare lethal one. The right net depends on the cost of the miss.
• Aggressive workup vs. harm of overdiagnosis. Scan everyone and you find incidentalomas that lead to biopsies that lead to complications in people who were never going to be harmed by what you found.
• Empiric treatment vs. waiting for certainty. Start antibiotics in sepsis before cultures return; wait for the culture in a stable patient to avoid resistance and C. diff.
• Patient autonomy vs. beneficence. The patient may refuse the treatment you believe will save them; respecting a competent refusal is the job, however hard.
• Time per patient vs. patients seen. Every extra minute with one patient is a minute the waiting room doesn't get.

Rules of Thumb

• If the patient looks sick, they are sick until proven otherwise — trust the gestalt.
• Diagnoses that don't fit the trajectory are usually wrong; re-examine the patient, not the chart.
• A new symptom in a patient on a new drug is the drug until proven otherwise.
• Never let the sun set on an undrained abscess or an unexamined acute abdomen.
• When two competent clinicians disagree, the data is ambiguous; get more, don't argue.
• The most dangerous time is the handoff and the "stable" patient nobody's watching.
• If you're surprised by a result, repeat or reconcile it before you act.

Failure Modes

• Anchoring. Locking onto the first plausible diagnosis and reinterpreting every later finding to fit it.
• Premature closure. Stopping the differential once a satisfying answer appears, before the dangerous alternative is excluded.
• Confirmation bias in test ordering. Ordering the test that confirms your hunch and ignoring the one that would refute it.
• Treating the lab, not the patient. Chasing a number into a cascade of interventions the patient never needed.
• Polypharmacy creep. Each drug treating a side effect of the last, until the regimen is the illness.
• Diagnostic momentum. Inheriting a prior clinician's label and never re-examining whether it was ever right.

Anti-patterns

• The shotgun workup — ordering every test instead of thinking, then drowning in incidental findings.
• The VOMIT problem — "victim of medical imaging technology," harm cascading from an unnecessary scan.
• Defensive medicine — testing and admitting to protect yourself, not the patient.
• The curbside as the consult — making real decisions on hallway hearsay without the chart or the patient.
• Documenting the plan you wish you'd made rather than the reasoning you actually used.

Vocabulary

• Differential diagnosis — the ranked list of conditions that could explain the presentation.
• Pretest probability — the estimated likelihood of disease before testing.
• Sensitivity / specificity — a test's true-positive and true-negative rates.
• Likelihood ratio — how much a result shifts the odds of disease.
• NNT / NNH — patients treated to help one / to harm one.
• Sequelae — the lasting consequences of a disease or injury.
• Iatrogenic — harm caused by medical care itself.
• Idiopathic — of unknown cause (the honest Latin for "we don't know").
• Comorbidity — a coexisting condition that complicates the primary one.
• Prognosis — the expected course and outcome.

Tools

• History and physical exam — the original diagnostic instruments, still the highest-yield.
• The stethoscope — auscultation as a real-time, free, bedside test.
• Laboratory and imaging — confirmatory, not exploratory; interpreted against probability.
• Clinical decision rules (Wells, CURB-65, HEART, Centor) — validated scores that anchor gestalt to evidence.
• The electronic health record — the shared memory and the documentation of reasoning, despite its friction.
• UpToDate / clinical guidelines — the current best evidence, because no one holds all of medicine in their head.

Collaboration

Modern medicine is a team sport played around one patient. The physician works with nurses (who see the patient continuously and catch what the physician misses on rounds), pharmacists (who guard against interactions and dosing errors), specialists (consulted for depth, not deference), social workers, physical therapists, and the patient's family. The healthiest teams treat the nurse's "something's not right" as a clinical finding, make the reasoning behind orders explicit so others can flag errors, and use structured handoffs (SBAR) because most serious mistakes happen at transitions of care. Hierarchy that silences a concern is a patient-safety hazard.

Ethics

The physician holds power over people at their most vulnerable, which is why the profession binds itself with explicit duties. The four pillars: autonomy (the competent patient decides), beneficence (act for their good), non-maleficence (avoid harm), and justice (fair use of finite resources). Confidentiality is near-absolute and breaks only to prevent serious harm. Informed consent means the patient understands the alternatives, including doing nothing. The hard ground is end-of-life care, refusal of beneficial treatment, allocation of scarce resources, and conflicts between what the family wants and what the patient would have chosen. The honest physician names uncertainty rather than projecting false confidence, and never lets a financial incentive shape a clinical recommendation.

Scenarios

Chest pain in a 55-year-old at 3 p.m. The story is atypical — sharp, worse with breathing — which tempts a benign label. The expert resists premature closure: the can't-miss list is ACS, PE, aortic dissection. HEART score is intermediate, so the patient is not low enough to send home. Troponin and ECG are unremarkable, but the threshold model says intermediate risk warrants serial troponins and observation, not discharge. Six hours later the second troponin rises. The atypical story was a real NSTEMI. The discipline that saved him was refusing to let an atypical history close a high-stakes differential.

The "anxious" frequent flyer. A young woman with a documented anxiety diagnosis presents again with palpitations and shortness of breath; diagnostic momentum invites "anxiety, reassure, discharge." The physician treats the prior label as a hypothesis, not a fact, and notes a resting tachycardia of 120 and a recent long flight. Wells score and a D-dimer point to PE; CT confirms it. Anchoring on the chart's history would have killed her.

A frail 88-year-old with a new pancreatic mass. The reflex is to stage and treat. The expert instead asks what the patient wants from her remaining time. She values being home and lucid over months gained through chemotherapy she'd spend hospitalized. Shared decision-making lands on symptom-directed palliative care. The "correct" oncologic answer would have been the wrong answer for this person.

Related Occupations

The physician sits at the center of a clinical web. Registered nurses share the patient continuously and are the early-warning system for deterioration. Surgeons share the diagnostic burden but resolve it in the operating room rather than over time. Emergency physicians compress the same reasoning into minutes under maximal uncertainty. Pharmacists own the safety and pharmacology of every drug the physician prescribes. Psychiatrists apply the same diagnostic discipline to the mind, where the instruments are the interview and time.

References

• Harrison's Principles of Internal Medicine
• How Doctors Think — Jerome Groopman
• The Innovator's Prescription / Bayes' theorem in clinical reasoning literature
• Hippocratic and modern medical-ethics tradition (Beauchamp & Childress, Principles of Biomedical Ethics)
• Sapira's Art and Science of Bedside Diagnosis