The four pillars of medical ethics

What the pillars are actually for

The four principles come from Principles of Biomedical Ethics, published by the American bioethicists Tom Beauchamp and James Childress in 1979, and they reached British medical education in large part through the work of Raanan Gillon, emeritus professor of medical ethics at Imperial College London. Their appeal is a deliberate modesty. They are not a complete moral theory and they do not tell you what to do. They are a shared vocabulary that lets a devout surgeon, a secular ethicist and a nineteen-year-old applicant argue about the same case without first having to agree about the meaning of life.

That modesty is the source of both their power and the way candidates misuse them. Because the principles do not rank themselves, you cannot derive an answer by listing them. A candidate who opens with “this raises issues of autonomy, beneficence, non-maleficence and justice” and then talks generally for two minutes has told the interviewer only that they own a glossary.

What the framework really does is diagnostic. Take any case that feels genuinely hard and the difficulty is almost never missing information — it is that two principles point in opposite directions and something has to give. In easy cases all four agree, which is precisely why nobody sets them as a station.

The four, defined precisely

Autonomy

Autonomy is the principle that a person with capacity has the right to make decisions about their own body and their own care, and that this right does not depend on the decision looking sensible to anyone else. The Mental Capacity Act 2005, which governs England and Wales, makes the point in statute: a person is not to be treated as unable to make a decision merely because they make an unwise one. Scotland and Northern Ireland have separate frameworks of their own, and consent and capacity works through the tests in the detail an interviewer will probe.

Respecting autonomy is an active duty, not a passive one. It requires giving the patient the information a reasonable person in their position would want, checking they have understood it, and making sure the choice is genuinely theirs rather than their family’s. It also has limits worth owning: autonomy is a right to refuse, not a right to demand, so a patient may decline anything offered but cannot compel a doctor to provide treatment that is clinically inappropriate. And it covers their own body, not other people’s.

Beneficence

Beneficence is the positive duty to act in the patient’s interest — not merely to refrain from harming them, but to do them good. In practice it means recommending the option most likely to improve the outcome that matters to this particular patient, and it is the reason doctors are expected to have a view rather than laying out a menu and standing back. Silence dressed up as neutrality is a failure of beneficence, not a display of respect.

The subtlety is that a patient’s interest is not purely clinical. Where someone lacks capacity, best interests is wider than the medically optimal option: it takes in their past wishes, beliefs and values, and the views of those close to them. A woman who has said all her life that she never wants to die in a hospital has an interest no survival curve captures.

Non-maleficence

Non-maleficence is the duty to avoid harm — primum non nocere, a Latin tag often attributed to the Hippocratic Oath although it does not appear there. Taken literally the principle would forbid medicine altogether. Chemotherapy is poison, surgery is controlled injury, anaesthesia is a deliberate loss of consciousness, and every drug worth prescribing has a side-effect profile.

So non-maleficence is really a principle of proportionality. The question is never whether an intervention causes harm but whether the harm it causes is justified by the benefit expected, whether a less harmful route to the same benefit exists, and whether the patient has been told what they are accepting. It carries a second, quieter duty too: competence. Harm caused by an avoidable error sits in a different moral category from the harm intrinsic to a necessary operation.

Justice

Justice is the pillar candidates rush, usually with one line about fairness. It has three distinct strands. Distributive justice is the fair allocation of finite resources — the pillar the NHS lives inside, because its budget is fixed and every decision to fund something is a decision not to fund something else. Rights-based justice means treating like cases alike: ethnicity, disability, wealth, social standing or the fact that a patient contributed to their own illness must not change the standard of care someone receives. Age is the strand worth handling carefully, because it can be clinically relevant to how much benefit a treatment is likely to deliver while never being a measure of how much a life is worth, and interviewers listen for whether you can keep those two uses of it apart. Legal justice is respect for the law and for the frameworks that bind everyone equally.

Justice is also the only pillar that looks outside the consulting room. The other three ask what is right for the person in front of you; justice asks about the patients you will never meet, whose care is displaced by the money spent here. That shift is what makes it feel cold, and why interviewers reward candidates who hold it alongside the individual case rather than instead of it.

Case one: a competent refusal of life-saving treatment

A man in his twenties arrives in the emergency department after a road traffic collision. He is bleeding heavily, he is fully conscious and oriented, and he is a Jehovah’s Witness who refuses blood products and carries a valid advance decision saying so. Transfusion would very probably save him. Without it he will very probably die.

Name the conflict first. Beneficence and non-maleficence both point hard towards transfusing: the treatment works, it is low risk, and withholding it causes a death that medicine could have prevented. Autonomy points the other way, and in English law autonomy wins. A refusal by an adult with capacity is decisive even when it leads to death, and treating him anyway would be a battery however good the motive.

Notice what the framework has and has not done. The law supplied the answer; the pillars showed you why the case is hard and what is being given up. That is the part weak candidates skip. Saying “we respect his autonomy” and moving on makes it sound costless. Saying “we respect his autonomy, and the cost is a preventable death the team will carry home” is the answer of someone who has thought about it.

There is also real clinical work still available inside the refusal, and mentioning it lifts an answer immediately. Confirm capacity for this decision now: while he retains capacity it is his present refusal that governs, and the advance decision becomes the operative document only if he loses capacity, at which point it has to be valid and applicable to the circumstances that have actually arisen. Establish exactly what he is refusing, since some patients accept particular blood fractions or cell salvage and others do not. Then use every alternative inside his limits — tranexamic acid, meticulous surgical haemostasis, iron, tolerating a lower haemoglobin.

Case two: one drug, priced in the hundreds of thousands

Now a case where the pillars pull along a different axis. A commissioner is asked to fund a therapy for a child with a rare inherited disease, priced in the hundreds of thousands of pounds a year. It will not cure her; it is expected to slow the disease and add some years of reasonable quality. Her family have run a public campaign and the local paper has the story.

Beneficence for this identified child is overwhelming, and there is a non-maleficence argument too, since withholding an effective treatment allows an avoidable harm. Justice points the other way. The money is not new money. Whatever is spent here is not spent on hip replacements, community mental health workers or a stroke unit, and the patients who lose out are real even though nobody knows their names or writes about them.

This is where the QALY does its work: it is the practical face of distributive justice. One year of life in full health counts as one QALY, a year in poorer health counts as a fraction, and dividing cost by QALYs gained lets a cancer drug, a knee replacement and a smoking cessation programme be compared on one scale. NICE uses cost per QALY as the spine of its technology appraisals. The range usually quoted for routine appraisals is roughly £20,000 to £30,000 per QALY; a severity modifier can give extra weight to QALYs gained by patients with the most severe conditions, and the separate route for highly specialised technologies works to a considerably higher range. Those rules and figures have been revised more than once, and the position described here is the one current when this article was updated, so describe the mechanism in an interview and check the current numbers before you quote any.

You should also be able to attack the QALY, because interviewers reward candidates who use a tool without worshipping it. It rests on population preference surveys no individual patient signed up to. It arguably disadvantages disabled and older people, since a year gained scores less when baseline health or remaining life expectancy is lower. It handles rare disease badly. And it treats a year of your life and a year of a stranger’s as interchangeable, which is simultaneously the entire point and the deepest objection.

The ideas that sit around the pillars

Four concepts recur often enough that you should be able to use them accurately and briefly. The first is the doctrine of double effect, which explains how an action with a foreseen bad consequence can still be permissible. It is conventionally set out as four conditions.

1. The act itself must be good or at least morally neutral — giving a drug to relieve pain, not giving a drug to end a life.
2. The bad effect must be foreseen but not intended. You accept the risk; you do not aim at it.
3. The bad effect must not be the means by which the good effect is achieved. Relief must come from the analgesia, not from the death.
4. There must be proportionate reason: the good effect must matter enough to justify accepting the bad one.

Double effect is usually taught through opioids and sedation at the end of life. One caveat is worth carrying, because it shows you have read past the textbook: opioids titrated properly against symptoms rarely shorten life, so the classic example is more philosophically useful than clinically common. Its real value is as the line separating palliative care from deliberately ending life — exactly the distinction you will need when you reach assisted dying.

The second is paternalism. Hard paternalism overrides a genuinely autonomous choice for the person’s own good and is very difficult to defend in contemporary practice. Soft paternalism intervenes where the choice is not genuinely autonomous — the patient is misinformed, acutely distressed, intoxicated, coerced, or has not grasped a consequence — and that is more defensible, because it protects the conditions autonomy depends on rather than overriding autonomy itself. Say the risk in that distinction out loud as well: “not genuinely autonomous” is an elastic phrase, and a clinician who dislikes a decision can reach for it to override a choice that was in fact competent and considered. Seatbelt laws, smoking bans and taxes on sugary drinks are the population version of the same argument, and you should be able to voice the liberty objection as fluently as the case in favour.

The third is the pair of theories underneath the pillars. A utilitarian asks which action produces the greatest total benefit; QALYs, screening programmes and nearly all resource allocation are utilitarian in structure. A deontologist asks whether an action is right in itself regardless of outcome; the prohibition on treating a competent adult without consent is deontological, since you may not do it even when the result would be better. Virtue ethics offers a third framing — what would a good doctor do here — which is often most useful when a station is really about character. Naming these once shows you know the pillars rest on a live argument. Naming them three times sounds like showing off.

The fourth is that some duties are strong without being absolute. Confidentiality is the clearest example: it protects autonomy and the trust the whole system runs on, yet it yields in defined circumstances where the harm to others is serious enough. Confidentiality and when to break it sets out the exceptions and the language to use for them.

Everything in stage five runs through this machinery

This article sits here on the reading path for a reason. The frontier topics later in the timeline are not separate subjects to be revised one by one. They are the same four principles under new conditions, which is why candidates who memorise hot topics without the ethics sound thin the moment they are pushed.

• AI in diagnosis: beneficence, as earlier and more consistent detection, against non-maleficence and justice, because models trained on unrepresentative data fail unevenly across populations.
• Genomic screening: autonomy, including the right not to know and a child’s future right to decide, against beneficence and earlier treatment, with justice behind who gets sequenced at all.
• Gene editing: beneficence for a future person who cannot consent, against non-maleficence across generations and a justice problem if enhancement becomes purchasable.
• Assisted dying: autonomy at its strongest against non-maleficence and a justice concern about pressure on people who already believe they are a burden.
• Antimicrobial resistance: beneficence for the patient in front of you against justice for every future patient — the cleanest resource conflict in medicine, because the resource being spent is the drug’s effectiveness itself.

Practise the naming move until it is reflexive. Read a scenario, and before you form any opinion, say which two principles are colliding and what each one would cost if it lost. Once that is automatic, answering an ethical dilemma live gives you a structure that survives a timer, and the ethics guide supplies longer scenarios to run the machinery on until it stops feeling like machinery.