Antibiotic course length

Why might finishing every last tablet, long taught as the way to stop resistance, help resistance along instead?

"Always finish the course" is one of the few pieces of medical advice almost everyone can recite, and the reason attached to it is always the same: stopping early leaves the tougher survivors alive to breed resistance.

That reasoning has a shape worth examining. It assumes the bacteria being selected on are the ones causing your illness. Suppose we ask instead where in the body the drug actually goes — and which populations of bacteria it meets there. Does the answer still come out the same way?

Begin with the case the doctrine was built from, because it is real and it is not a myth. In tuberculosis, an infection can carry an enormous bacillary load, and resistant mutants are already present before treatment starts simply because the population is so large. Kill the susceptible majority with an inadequate regimen and the mutants inherit the lung. That is why tuberculosis is treated for months, with several drugs at once, and why interrupting the regimen genuinely does select resistant organisms in the target species. The same logic applies wherever the pathogen carries a big population and can mutate its way past the drug.

Now ask whether that describes an ordinary sore throat, urinary infection, or cellulitis. It does not, for two reasons. The bacterial burden is far smaller, and more importantly, resistance in those organisms is usually acquired rather than mutated — genes arriving on plasmids from elsewhere, as the neighbouring piece on resistance describes. Selection is not waiting for a rare mutant to appear in the infection.

So where does the drug's selection pressure land? Here is the step that changes everything. A swallowed antibiotic does not travel to the infection alone. It circulates, and it bathes every colonised surface in the body: the gut, the throat, the skin. Those surfaces carry populations of bacteria orders of magnitude larger than the infection being treated — and every one of those organisms is under selection for as long as the drug is present, whether or not it has anything to do with the illness.

Follow that through and the arithmetic reverses. If the resistance that later matters is being selected in the bystander flora, then the relevant quantity is total drug exposure, and each additional day of treatment is another day of selection on a reservoir that was never the target. Longitudinal studies of gut flora after a course show exactly this pattern: resistance gene carriage rises during treatment and takes time to decline afterwards. Under that reading, "finish the course" is not protective at all — it is a prescription for more selection.

Then why was it ever taught? Partly because the tuberculosis case was generalised too far, and partly because the alternative sounded dangerous: an under-treated infection can relapse, and a relapse means a second course, which is more exposure, not less. That is a genuine argument on the other side and it should not be waved away.

What would settle it, and what would refute the short-course case. The test is a randomised non-inferiority trial per indication: give one group the traditional duration and one a shorter one, and measure cure and relapse. Several such trials — in community-acquired pneumonia, uncomplicated urinary infection, and intra-abdominal infection after the source has been drained — have found shorter courses non-inferior. If shorter courses instead produced more relapses, and the re-treatment that followed generated more total exposure than the long course would have, the argument fails for that condition. And if resistance carriage in a treated person's own flora turned out to be unrelated to how many days they took the drug, the selection story would have nothing to stand on.

The honesty this requires. The traditional durations — seven, ten, fourteen days — are largely inherited round numbers rather than quantities derived from evidence, and the case that they are often longer than necessary is now widely made. But it is indication-specific, and the exceptions are not marginal: tuberculosis, endocarditis, bone infection, and sites the drug penetrates poorly are all conditions where prolonged treatment is not an inherited habit. Nor is the researchers' proposal "stop when you feel better" — the argument is about what duration should be prescribed, established per condition by trial. How long any particular person should take a particular drug is a decision for the clinician who diagnosed them, and nothing in this mechanism can be substituted for it.

The correction at the heart of this is about where selection acts. The old doctrine has selection happening inside the infection, so more drug means fewer survivors and less resistance. The revision has most of the selection happening outside the infection, in flora that will never cause the illness but will happily donate a resistance gene to something that does. Under the first model, exposure is protective. Under the second, exposure is the cost.

Neither model is universally right, which is the point that gets lost. The question "does resistance to this drug arise in the organism I am treating, or in everything else the drug touches?" has a different answer for tuberculosis than for a bladder infection, and the correct course length follows from that answer rather than from a slogan.

• Why source control — draining an abscess, removing a catheter — shortens the duration a drug needs to achieve, and what that says about what the drug is for.
• How resistance genes selected in one person's gut reach a stranger's bloodstream, and which links in that chain are actually measurable.

Every term the collection defines is gathered in the glossary.