Before the fourth drug: investigating apparent treatment resistance

The unknown that comes first

A fourth antihypertensive drug is a reasonable step when three have failed. It is a poor step when the three have not actually been tried, and the prescription record cannot tell the difference. Before resistance is accepted as a property of the patient's physiology, two simpler explanations deserve an answer: is the blood pressure genuinely high, and is the treatment genuinely being taken?

The 2024 European Society of Cardiology guideline defines resistant hypertension as pressure above goal despite three or more drugs of different classes at optimal doses, and it treats confirmation of both measurement and adherence as part of making the diagnosis rather than as an afterthought (McEvoy et al., European Heart Journal 2024). Out-of-office measurement deals with the first question. White-coat effects, cuff size and technique can each add 10 mm Hg or more to a clinic reading, and home or ambulatory monitoring removes most of that noise. The second question is harder, because asking people whether they take their tablets produces reassuring answers whether or not they do.

Measuring exposure rather than asking about it

Biochemical adherence testing analyses a urine or blood sample for the prescribed drugs or their metabolites by liquid chromatography coupled with tandem mass spectrometry. The method is direct in the sense that it detects the molecule, and objective in the sense that it does not depend on recall. The foundational study from a specialist hypertension clinic screened spot urine for 40 commonly prescribed antihypertensive drugs in 208 patients (Tomaszewski et al., Heart 2014). One quarter were non-adherent: 10.1 percent had none of their prescribed drugs detectable and 14.9 percent had some but not all. Non-adherence was most frequent among follow-up patients with inadequate control and those referred for renal denervation. Each prescribed drug that was missing from the urine was associated with roughly 3 mm Hg higher clinic systolic pressure after adjustment.

The test has limits that should be understood before a result is acted upon. Most antihypertensive drugs and their metabolites are detectable in urine for a day to a few days after a dose, depending on the drug's half-life. A sample taken at a clinic visit therefore reports recent intake only. Detection at one visit shows that the drug was taken shortly before that visit; it does not show that it was taken every day for the previous month, and tablets taken on the morning of an appointment are indistinguishable from tablets taken every morning. Non-detection is more informative, but even then an assay cannot distinguish a drug that was never started from one that was stopped because of side effects, cost or confusion about the regimen. Laboratory factors matter too: the panel must include the specific drugs prescribed, and dilute urine or unusual metabolism can lower the signal.

Burden, not blame

Missing treatment is rarely a decision taken lightly. Three or four daily tablets with different timings, dizziness on standing, cough with an angiotensin-converting enzyme inhibitor, ankle swelling with a calcium-channel blocker, the cost of repeat prescriptions, and the absence of any symptom to remind a person that hypertension exists all push in the same direction. A test result that reveals missed treatment is useful only if it opens a conversation about which of these is operating. Used as an accusation it closes the conversation and, in the clinic experience behind the 2014 study, drives people away from follow-up altogether. The language in a report should describe what was measured: drug not detected in the sample provided on this date.

OUTREACH: feeding the result back

Whether telling patients their result changes anything was the question of the OUTREACH trial, published in 2026 (Tomaszewski et al., Lancet Primary Care 2026). Across 12 UK secondary and tertiary centres and primary care services, adults on at least two antihypertensive drugs gave a urine sample for chemical adherence testing. Of 879 assessed, 720 were adherent or had unconfirmed status and left the randomised component; 70 adherent participants were kept in the study to preserve masking. The 130 participants found non-adherent were randomised to a discussion of their own urine result together with the main reasons for non-adherence, or to standard care. The population was diverse: 38 percent were Black, African, Caribbean or Black British and 17 percent Asian or Asian British.

The primary outcome was mean clinic systolic blood pressure at the fourth visit, a median of 2.8 months after the intervention. It was 150.9 mm Hg in the intervention group and 151.1 mm Hg in the standard-care group, an adjusted difference of −5.1 mm Hg with a 95 percent confidence interval from −12.7 to 2.5 and p = 0.19. The investigators state plainly that the intervention did not show a significant effect and that the study was underpowered. Serious adverse events were equally distributed, five participants in each group.

That is the result, and it should be reported as such rather than as a trend. What the trial contributes is nonetheless substantial. It shows that a multicentre testing pathway can be run across hospital and primary care, that the proportion of non-adherent patients willing to be randomised is small relative to those screened, and that the confidence interval is wide enough to include both a clinically meaningful benefit and no effect. A feedback conversation of the kind tested here is a single, brief intervention; the trial does not tell us what a sustained programme addressing side effects, regimen simplification and cost would do. It also leaves open how often testing should be repeated, since a single result cannot track persistence. Larger trials with longer follow-up are the explicit recommendation of the authors, and the design questions they raise are the ones a successor study must answer first.

Adherence inside the trial itself

The same uncertainty about exposure runs through every hypertension trial, and it is rarely analysed. Adherence tends to rise during a run-in period, when participants are observed and motivated, and to fall during follow-up. If the intervention group is tested or counselled more often than the comparator, part of any blood-pressure difference may reflect better adherence rather than the drug. If a trial enrols people labelled resistant without confirming exposure, it may be studying a mixture of true resistance and undetected non-adherence, which dilutes the effect of an agent that works only in the former.

Three consequences follow for design. First, exposure should be measured, and the measurement should be planned: which samples, at which visits, analysed by which panel, with the limits of detection stated in advance. Second, the analysis plan should say what will be done with the result. Excluding non-adherent participants after randomisation breaks the randomisation; a prespecified sensitivity analysis that keeps the intention-to-treat population and reports results by biochemically confirmed exposure does not. Third, run-in testing should be reported, because a trial that excluded non-adherent people before randomisation has a more adherent population than the clinic it will be applied to, and its effect size will be larger than the one seen in practice.

What to take from this

Apparent resistance is a measurement problem before it is a pharmacological one. Out-of-office readings answer whether the pressure is high. A biochemical test answers whether a drug was taken shortly before the sample, with the limits of sampling windows and panels understood. OUTREACH shows that simply reporting the result back did not demonstrably lower blood pressure in an underpowered trial, and that the question deserves a larger one. In trials, exposure is a variable to be measured and analysed, not assumed.

This note draws on the work of Professor Maciej Tomaszewski, who led the studies discussed here. Investigators planning trials in which adherence may vary between arms can discuss the analysis plan confidentially.