16.1 — What "Disease" Actually Means

Sickle cell trait is a genetic abnormality. In a malarial region it is an advantage (Chapter 3.3). Is it a disease?

A person with a blood pressure of 145/92 feels entirely well and will feel entirely well for twenty years. Are they ill?

Deafness is treated as a disability by medicine and as a cultural identity by a large part of the Deaf community (Chapter 11.12).

"Disease" is not a purely biological category. It is a judgement about a biological state, and the judgement involves values. Which is not a reason to abandon it — the concept is enormously useful — but it is a reason to know where the boundaries are soft, because a great deal of harm has come from treating them as hard.

Working definitions

Disease — a condition with an identifiable pathological process, characteristic features, and typically a known or presumed cause.

Illness — the person's experience of being unwell.

Sickness — the social role, including what others expect and permit.

And they come apart routinely. A person can have disease without illness — untreated hypertension, early cancer, controlled HIV. And illness without identifiable disease — chronic fatigue, many pain conditions, functional disorders.

That second category is where medicine has performed worst. The absence of a detectable abnormality has repeatedly been treated as evidence that nothing is wrong, which is a logical error — the tests detect what they were designed to detect — and it has caused enormous and avoidable suffering.

Multiple sclerosis was called hysterical paralysis before imaging existed. Peptic ulcers were caused by stress until 1982 (Chapter 9.2). The list of conditions once considered psychological and now understood mechanistically is long, and it should induce humility about the current one.

The seven causes

Every disease process falls into one or more of these.

1. Genetic — inherited or new mutations (Part 2).

2. Infectious — bacteria, viruses, fungi, parasites, prions (Part 17).

3. Immunological — autoimmunity, allergy, immunodeficiency (Chapter 13.6).

4. Vascular — inadequate blood supply, or bleeding.

5. Metabolic and nutritional — deficiency, excess, or a failure of a pathway.

6. Neoplastic — uncontrolled growth (Part 19).

7. Environmental and traumatic — physical, chemical, radiation, and injury.

Plus two categories that cut across all of them:

Degenerative — the accumulated effect of time and use.

Iatrogenic — caused by medical treatment.

And that last one deserves stating rather than hiding. Adverse drug reactions, hospital-acquired infections, complications of procedures and diagnostic errors together account for a substantial share of harm, and estimates place medical error among the leading causes of death in some analyses — though the methodology of those estimates is genuinely contested. The honest position is that iatrogenic harm is large enough to be a major category and that the headline figures are uncertain.

Most common diseases are multifactorial. Type 2 diabetes involves genetic susceptibility, diet, activity, obesity and age together. Asking "what caused it" often has no single answer, and expecting one leads to poor thinking about prevention.

Cause, risk factor and mechanism

Three different things, frequently conflated.

A cause is necessary. Without Mycobacterium tuberculosis there is no tuberculosis.

A risk factor shifts probability. Smoking is a risk factor for lung cancer — most smokers do not get it, and some non-smokers do.

A mechanism is how it happens. Benzopyrene in tobacco smoke damages p53 (Chapter 2.5).

And necessary is not the same as sufficient. Around a quarter of the world's population carries latent tuberculosis and around 5 to 10 percent ever develop the disease (Chapter 17.8). The organism is necessary and not sufficient; host factors decide.

Which is why the framing "the germ theory versus the terrain" is a false dichotomy — both matter, and neither alone explains who gets ill.

Establishing causation

The distinction between association and causation is the single most useful piece of reasoning in this Part, and it is where most bad health information goes wrong.

The Bradford Hill considerations, set out in 1965 while establishing that smoking causes lung cancer, remain the standard framework. They are not a checklist to be scored — Hill said so explicitly — but a set of questions.

Strength — how large is the effect? Smokers have around 15 to 30 times the lung cancer risk.

Consistency — found repeatedly, in different populations and by different methods?

Specificity — does the exposure produce a particular outcome? Weak, and frequently misused: smoking causes many diseases, which does not weaken the case.

Temporality — the only absolute requirement. The cause must precede the effect.

Dose–response — more exposure, more disease?

Plausibility — is there a mechanism? Useful when present and not required, since mechanisms are frequently discovered after the association.

Coherence — does it fit what else is known?

Experiment — does removing the exposure reduce the disease?

Analogy — has something similar been established?

And the reverse-causation trap catches people constantly. Low cholesterol is associated with higher mortality in some studies — because serious illness lowers cholesterol, not because low cholesterol causes death. Bed rest was associated with worse outcomes because sicker people were put on bed rest.

Confounding is the other main trap. Coffee drinking was associated with lung cancer until smoking was accounted for.

Mendelian randomisation is a genuinely elegant modern solution. Genetic variants are allocated at conception, essentially at random with respect to lifestyle, so comparing people by a variant that raises, say, LDL cholesterol approximates a randomised trial of lifelong LDL exposure. It has confirmed some associations as causal — LDL and heart disease — and refuted others, notably several claimed benefits of moderate alcohol.

Screening, and when finding disease helps

Screening is testing people without symptoms, and it is not automatically good.

The Wilson and Jungner criteria, from 1968, still define when it is worthwhile:

The condition must be an important health problem.Its natural history must be understood.There must be a recognisable early stage.Treatment at that stage must produce better outcomes than treatment later.There must be a suitable, acceptable test.And the whole programme must be cost-effective and continuous.

The fourth criterion is the one that fails most often, and it is the one people find hardest to accept. Detecting a disease earlier only helps if earlier treatment changes the outcome.

Two biases make screening look better than it is, and both are essential to understand.

Lead time bias. Diagnosing a disease earlier increases the time from diagnosis to death without changing the date of death. Survival statistics improve; nobody lives longer.

Length time bias. Screening preferentially detects slow-growing disease, because fast-growing disease appears between screening rounds. So screen-detected cancers look more survivable than they are, because they are a biased sample.

Overdiagnosis is the extreme case: detecting disease that would never have caused symptoms or death.

And it is not hypothetical. South Korea's thyroid ultrasound screening produced a fifteen-fold increase in diagnoses with no change in mortality (Chapter 12.3). Post-mortem studies find prostate cancer in around 70 percent of men over 80 (Chapter 15.1).

Which is why screening decisions are properly made on mortality, not on survival rates or detection rates, and why the programmes that survive scrutiny — cervical, bowel, breast in defined age ranges, abdominal aortic aneurysm in men at 65 — are the ones that have demonstrated it.

The natural history of disease

Every disease has a course, and interventions are described by where they act.

Susceptibility → subclinical disease → clinical disease → outcome.

Primary prevention — before disease starts. Vaccination, not smoking, seat belts. The most effective and the least visible, because success is a disease that never happens.

Secondary prevention — detecting and treating early disease. Screening.

Tertiary prevention — reducing the impact of established disease. Rehabilitation, secondary drug prevention after a heart attack.

And a fourth term worth knowing: quaternary prevention — protecting patients from unnecessary medical intervention. Overdiagnosis, overtreatment, and the cascade of tests that follows an incidental finding.

Prevalence and incidence, and why it matters

Incidence — new cases in a period. Prevalence — total cases at a point.

\text{Prevalence} \approx \text{Incidence} \times \text{Duration}

Which produces a counter-intuitive and important consequence. A treatment that keeps people alive without curing them increases prevalence.

HIV prevalence rose after effective treatment was introduced, because people stopped dying. That is a success, and a naive reading of prevalence would call it a failure.

And prevalence determines how a test performs, which is Chapter 16.4.

Risk, stated properly

Relative risk without absolute risk is close to meaningless, and this is the most common way health information misleads.

"Doubles your risk" tells you nothing until you know the starting point. Doubling 1 in a million is negligible; doubling 1 in 10 is not.

Number needed to treat (NNT) — how many people must be treated for one to benefit. A far more useful figure than a percentage.

Statins for primary prevention in moderate-risk people: NNT of around 100 over 5 years to prevent one heart attack. Statins after a heart attack: around 20 to 40. Same drug, different populations, very different value.

Number needed to harm (NNH) — the same for adverse effects, and it should always be quoted alongside.

And a specific practical point about relative risk in the media: "a 20 percent increase" almost always means relative. Asking "from what to what" resolves most health scares in one question.

Where medicine has been wrong

A short list, because it argues for the right kind of scepticism — including about current practice.

Bloodletting, standard for two thousand years, killed people.

Thalidomide (Chapter 4.4).

Hormone replacement therapy — overprescribed, then abandoned, then partly reinstated (Chapter 12.6). The evidence changed; so did the interpretation of the same evidence.

Hormone therapy and antiarrhythmic drugs after heart attack — drugs that corrected a number and increased mortality. The CAST trial found that suppressing extra beats after a heart attack, which everyone expected to help, roughly tripled deaths.

Routine episiotomy. Continuous fetal monitoring. Bed rest for back pain. Arthroscopic surgery for degenerative knee tears (Chapter 5.7). Opioids for chronic pain (Chapter 11.13). Peanut avoidance in infancy (Chapter 13.6).

Every one of them was standard practice, based on plausible reasoning, and reversed by evidence.

The pattern is consistent: a mechanism suggests an intervention, the intervention is adopted, and it takes a trial to find out whether it works. Plausibility is not evidence.

And the appropriate response is not cynicism but a specific habit: asking what the evidence is, and what would change it.

What the next page fixes

Disease begins at the level of the cell. Chapter 16.2 covers cell injury and cell death — what actually damages a cell, the point beyond which it cannot recover, and the difference between the two ways it dies.