The big ideas of medicine

In the 1840s, Vienna General Hospital had two maternity clinics. In the one run by doctors, far more mothers died of "childbed fever" than in the one run by midwives. Ignaz Semmelweis noticed that the doctors came straight from performing autopsies. He made them wash their hands in chlorinated lime, and deaths in their clinic fell dramatically.

Semmelweis couldn't explain why it worked, and many colleagues rejected his findings. The explanation came two decades later. Louis Pasteur showed that microbes cause fermentation and decay, Joseph Lister used carbolic acid to prevent wound infections, and Robert Koch identified the specific bacteria behind anthrax, tuberculosis and cholera.

The lab models a hospital ward where staff move between patients. When they don't wash, they carry germs from one patient to the next.

A vaccine exposes the immune system to a harmless version or piece of a pathogen, so that it responds quickly to the real thing later. Vaccines protect the person who gets them, and they also protect others, by stopping chains of transmission.

The key number is R₀, the average number of people one case infects in a population with no immunity. If each case infects fewer than one new person, outbreaks die out. That happens once the immune share exceeds 1 − 1/R₀. Measles, one of the most contagious diseases known, has R₀ around 12 to 18, so it takes roughly 92–95% immunity to stop it. That's why small drops in vaccination coverage lead to measles outbreaks.

A bacterial infection contains billions of cells. By chance, a few may carry a mutation or gene that helps them survive a drug. An antibiotic kills the rest, and the survivors multiply into a population where resistance is common. Bacteria can also share resistance genes directly, so resistance spreads between species.

This is evolution by natural selection, happening in days. It's why using antibiotics when they aren't needed, such as for viral colds or routine growth promotion in livestock, creates resistance without any benefit. Drug-resistant infections were associated with nearly 5 million deaths worldwide in 2019 (GRAM Collaborators, Lancet, 2022).

A test has two error rates. Sensitivity is the share of people with the disease who test positive. Specificity is the share without it who test negative. Neither tells you the chance that you have the disease given a positive result. That's the positive predictive value, and it also depends on how common the disease is.

Picture 1,000 women having a routine mammogram, with 1% prevalence, 90% sensitivity and 91% specificity. Of the 10 with cancer, 9 test positive. Of the 990 without, about 89 also test positive. So only 9 of 98 positives, about 9%, actually have cancer. In studies, most doctors guess much higher. Counting people (natural frequencies) instead of using percentages makes the answer obvious.

Suppose a new drug really works. In ordinary practice, doctors give it to their sickest patients, who are the most likely to die anyway. Compare treated and untreated patients and the drug can look harmful. This is confounding by indication, and it fooled medicine for centuries.

A randomized controlled trial assigns treatment by a coin flip. Sick and healthy patients end up evenly split between the groups, as do all the factors nobody thought to measure. Any difference in outcomes can then be credited to the treatment. Adding blinding, so that neither patients nor doctors know who got what, removes the placebo effect and biased assessment too.

Screening tests healthy people to catch disease early. It sounds like obvious good sense, but two biases make it look better than it is. Lead-time bias: if a screen finds a cancer three years earlier but the patient dies on the same day as they would have anyway, "survival after diagnosis" grows by three years and nobody gains a day. Overdiagnosis: some cancers found by screening would never have caused symptoms in the person's lifetime. Treating them adds harm and makes "survival rates" look excellent.

That's why screening programmes are judged by mortality, the number of people dying of the disease, in randomized trials, not by survival rates or the number of cancers found.

The body clears most drugs at a rate proportional to how much is present, so levels fall by half every half-life. Give a dose before the last one is gone and the levels build up. After about four to five half-lives they reach a steady state, where each dose just replaces what was cleared.

Too low and the drug doesn't work. Too high and it becomes toxic. The range in between is the therapeutic window. For most drugs it's wide, but for some, like warfarin, digoxin and lithium, it's narrow enough that blood levels must be monitored. Kidney or liver disease can lengthen a half-life and push levels into the toxic range on a normal dose.

A drug that lowers heart-attack risk from 2% to 1% can be advertised as "cutting heart attacks by 50%." That's the relative risk reduction. The absolute risk reduction is just one percentage point, so 100 people must take the drug for a year for one of them to avoid a heart attack. The other 99 get only the cost and side effects. That's the number needed to treat (NNT).

The same relative reduction matters far more to high-risk patients. Cutting a 40% risk in half prevents an event for 1 in 5 people treated. That's why treatment decisions should start from a person's baseline risk.

After a meal, glucose from food enters the blood. The pancreas senses the rise and releases insulin, which tells muscle, fat and liver cells to take glucose in. As glucose falls, insulin release slows. It's the same idea as a thermostat: a change triggers a response that undoes it. That's negative feedback.

Diabetes is a failure of this loop. In type 1 diabetes the immune system destroys the insulin-making cells, so there's no signal. In type 2 diabetes the body's cells respond weakly to insulin (insulin resistance), and over time the pancreas can't keep up. Either way, blood glucose stays high after meals, and over years that damages blood vessels, nerves, kidneys and eyes.

The heart is a pump, and the blood vessels are pipes. William Harvey showed in 1628 that blood circulates in a closed loop, overturning 1,400 years of teaching that it was constantly made and consumed.

For smooth flow in a tube, Poiseuille's law says flow is proportional to the pressure difference times the radius to the fourth power. Halve the radius and flow falls to one sixteenth at the same pressure. That's why the body controls blood flow mainly by slightly widening or narrowing small arteries, and why fatty plaques that narrow coronary arteries (atherosclerosis) can starve the heart muscle of blood and cause angina or a heart attack.

Gregor Mendel crossed pea plants and found that traits are passed on in discrete units, now called genes, that come in pairs. Each parent passes on one of their two copies, at random.

In a recessive disease such as cystic fibrosis or sickle cell disease, a person is affected only if both copies are faulty. People with one faulty copy are healthy carriers. When two carriers have a child, each child has a 1 in 4 chance of being affected, 1 in 2 of being a carrier, and 1 in 4 of inheriting neither faulty copy. Chance has no memory: having one affected child doesn't change the odds for the next.

Your body has tens of trillions of cells, and every cell division carries a small chance of a DNA copying error. Most errors are harmless or get repaired. A cancer begins when one cell lineage collects several mutations that together let it grow without control, ignore signals to die, and invade other tissues. Each mutation gives the lineage an edge, so cancer is evolution by natural selection inside the body.

If a cancer needs k independent mutations, the chance of having all k by age t grows roughly like t to the power k. Many cancers' incidence rises with about the 4th to 6th power of age, which suggests several rate-limiting steps. Inherited mutations give a head start: children who inherit one faulty retinoblastoma gene need only one more "hit."