How to Monitor Medication Adherence in a Clinical Trial

Medication adherence in a clinical trial is monitored using a combination of methods that trade off objectivity, cost, patient burden and how quickly the data reaches the study team. The established options are patient diaries, pill counts, drug-level (biochemical) testing, directly observed therapy, electronic monitoring caps, smart blister packs, and smart dispensers. There’s no one-size-fits-all in medication adherence: different solutions are best suited for different applications, therapeutic areas and patient groups.

But the more important decision comes before you pick a method: be clear about what you want adherence data to do. Do you need to measure it for the record, see it during the trial, or act on it to change participant behaviour while there’s still time? Each is a higher bar than the last, each rules certain methods in or out, and each only pays off if the system is something a site team will actually use day to day. Get that definition right upfront and the method almost chooses itself.

This guide explains how to decide what success looks like, then walks through each method: what it measures, where it falls short, and how to choose.

Why adherence monitoring matters

Unaccounted non-adherence is a critical threat to a trial’s efficacy signal. If participants don’t take the investigational product as protocol-specified, the treatment effect is diluted: a drug that works can look as though it doesn’t. That under-estimation reduces statistical power and, in the worst case, sinks a viable compound. Drug accountability and reconciliation are also a Good Clinical Practice expectation in their own right, so some form of adherence record is required regardless of how rigorous you want to be about it. There are several upsides to knowing adherence: you can better evaluate adverse events, spot potential discontinuations early and detect dropouts. Precise, timestamped dosing can also greatly enhance PK/PD analysis.

The key problem that compounds is that adherence error is rarely random. The participants most likely to miss doses are also the ones most likely to under-report it, which means weak monitoring doesn’t just add noise, it adds bias. More on that below.

Decide what you want first: measure, see, or act

Most discussions of adherence jump straight to methods. It’s worth pausing on the goal, because three quite different outcomes hide behind the word “monitoring,” and they sit on a ladder where each rung demands everything below it, plus more.

Measure: produce an objective record of dosing you can analyse after the trial. This is the minimum, and almost any method clears it; even a pill count gives you a number. The only question is how accurate and granular that record is.

See: view adherence during the trial, not just at the end. This sounds like a free extension of measuring, but it isn’t: a method can measure dosing continuously and still only let you see it retrospectively, when the data is downloaded at a site visit. Seeing requires the data to actually reach you, and when it reaches you is the whole game.

Act: change what happens, by intervening with a participant whose adherence is slipping while they’re still enrolled. This is where the value lives. A protected efficacy signal and a retained participant come from someone doing something, and that only happens if the data arrives in time and lands in front of the right person in a form they can act on.

Two questions sit underneath the ladder and decide whether you ever get past “measure”:

When do you need the data? Retrospective is fine if your goal is analysis: you’re reconstructing what happened to interpret the results. But if your goal is to act, retrospective is too late by definition: the participant has already dropped out or diluted the signal. Real-time isn’t a luxury feature here; it’s the thing that makes intervention possible at all.

Will the site team actually use it? This is the question that largely decides ROI, and the one most easily skipped at selection. A system that surfaces adherence only when someone logs into a portal and goes looking will, in practice, mostly not be looked at. Site teams are stretched, and policing a dashboard is nobody’s job. The return on adherence monitoring is realised through action, and action only happens if the system fits how a site team actually works: the signal has to find them, not the other way round. A technically excellent dataset that no one acts on produces a clean post-hoc analysis and zero impact during the trial.

Goal	What it requires	Methods that clear the bar
Measure	Any record of dosing	All methods
See	Data that reaches you, retrospectively or live	Most objective methods; live only if the data transmits
Act	Real-time data and a workflow site teams use	Real-time, transmitting methods with alerting built in

Deciding which rung you’re aiming for before reading the next section will help inform which features (and possible trade-offs) you need to consider.

The methods, compared

Patient diaries and self-report

Participants record each dose themselves, on paper or in an electronic diary (eDiary/eCOA). It’s inexpensive, simple to deploy, and captures context a device can’t: why a dose was missed, side effects, timing.

The weakness is well documented: self-report consistently over-estimates adherence. Recall is imperfect, and “white-coat adherence” (sometimes called the parking-lot effect) means diaries are often completed retrospectively just before a site visit rather than at the time of dosing. Self-report is useful as a supplement and for capturing reasons, but it is the least reliable method on its own.

Pill counts and drug accountability

The site counts the dosage units returned at each visit against the number dispensed, and infers adherence from the difference. It’s cheap, requires no special technology, and overlaps with the drug accountability you have to do anyway.

But a pill count tells you what left the bottle, not what was taken. It is easily distorted, intentionally (discarding unused units before a visit, sometimes called pill dumping) or innocently, and it is coarse: you get one data point per visit, with no insight into the pattern of dosing in between. Like diaries, it tends to over-estimate.

Drug-level (biochemical) measurement

Measuring the parent drug or a metabolite in blood or urine, or adding a detectable low-dose marker, gives an objective, biological read on whether the drug is present. It’s the most direct evidence that a dose was actually absorbed.

The limitations are practical. Sampling is invasive, costly, and intermittent, so it captures only recent doses rather than long-term behaviour, reintroducing the white-coat effect, since a participant who resumes dosing shortly before a visit can still test positive. Pharmacokinetic variability between individuals also complicates interpretation. It’s powerful as a confirmatory or spot-check tool, less so for continuous monitoring.

Directly observed therapy (DOT and video DOT)

A clinician watches each dose being taken, either in person or, increasingly, by video (vDOT). Its strength is certainty: for the doses that are observed, there’s little doubt the dose was taken, which is why it has a long history in fields such as tuberculosis where confirmed dosing is central to the trial.

The practical consideration is effort. In-person observation is intensive, and video DOT, while lighter, still asks the participant to record and upload each dose. It also requires the patient to have or get their phone to take a dose. It fits trials where confirmed dosing is a primary objective rather than those simply tracking adherence behaviour over time.

Electronic monitoring caps (smart caps)

A bottle cap or container with an embedded microchip timestamps each opening, building a longitudinal record of dosing behaviour over the whole study. This is the long-standing objective electronic method, with a deep validation literature behind it, and it captures dosing patterns that diaries and pill counts miss entirely.

What a cap records is the container opening, which it uses as the marker for a dose. Depending on the system, that data is read at site visits via NFC scanner or transmitted between them over cellular. It remains one of the strongest objective options, particularly for understanding dosing patterns across a trial.

Smart blister packs

Instrumented blister packs detect when a tablet is pushed out of an individual cavity, usually through an NFC circuit built into the pack. They give per-tablet granularity tied directly to the primary packaging, and the data is read when the pack is tapped against a phone or a reader.

The defining requirement is that the drug has to be packaged into the instrumented blister in the first place, so adopting smart blisters means primary-packaging work and a format fixed to that pack. Because NFC packs are typically read on contact rather than continuously, the data tends to arrive when the pack is actively scanned rather than streaming on its own. They suit studies already committed to blister presentation and willing to build the instrumented pack into the supply chain.

Smart dispensers (packaging add-on)

A smart dispenser fits onto a standard packaging bottle and uses electronics to dispense the pills and sensors to register each interaction, confirming that an actual pill has been removed, timestamped, dose by dose. This is the key difference from a cap: a cap knows only that the container was opened, whereas a dispenser detects the pill itself leaving, so a recorded event corresponds to an individual dose rather than an opening that might mean nothing, one, or several.

Because it attaches to standard bottles, there’s no primary-packaging redesign, the supply-chain footprint is light, and the per-unit cost is modest. The dispense data uploads automatically in the background with no patient action, so it can feed real-time monitoring rather than waiting to be read at a visit. Of the packaging-based methods, a smart dispenser offers the most granular objective signal (per pill, sensor-confirmed, at the lowest patient burden) while fitting the way solid-oral-dose trials already package and ship.

Method comparison at a glance

Method	What it records	Granularity	Objective?	Patient effort	Real-time data?	Confirms ingestion?
Diaries / self-report	Reported doses	Per dose (reported)	No	High	No	No
Pill counts	Units returned	Per visit	Partly	Low	No	No
Drug-level testing	Drug in body	Per sample	Yes	Medium	No	Yes (recent only)
Directly observed therapy	Observed doses	Per dose	Yes	High	Varies	Yes (when observed)
Electronic monitoring caps	Container openings	Per opening	Yes	Low	Sometimes	No
Smart blister packs	Tablet removed from cavity	Per tablet	Yes	Low–medium	On scan (NFC)	No
Smart dispensers	Sensor-confirmed pill removal	Per pill	Yes	Very low	Yes	No

The measurement-bias problem most methods share

It’s tempting to treat adherence error as random noise that larger numbers will wash out. It usually isn’t. The methods that depend on patient action (diaries), on honest return of units (pill counts), or on a snapshot in time (drug levels at a visit) all tend to over-estimate adherence, and they over-estimate it most for the participants who are struggling. A non-adherent participant is the one most likely to fill in the diary from memory, discard unused units, or resume dosing before a visit.

The practical consequence is that weak monitoring systematically hides the very participants whose behaviour most affects your efficacy signal. Choosing a method isn’t only about average accuracy; it’s about whether the method still sees adherence clearly when a participant is slipping.

Measuring adherence vs confirming ingestion

A useful distinction to hold onto: almost no method actually confirms that a pill was swallowed. Diaries and pill counts infer it. Electronic caps, smart blisters and smart dispensers record an event (an opening, a cavity emptied, a pill removed) that is strong evidence of dosing but stops short of confirmed ingestion. Only directly observed therapy and drug-level testing verify the dose reached the patient, and each carries cost or burden that rules it out for many studies.

Being explicit about this in your protocol matters. The goal for most trials is not philosophical certainty about every swallow; it’s an objective, tamper-resistant, well-timed record of dosing behaviour that’s reliable enough to interpret the efficacy data and to intervene when something goes wrong.

How to choose a method for your trial

Start from the ladder above (decide whether you’re aiming to measure, see or act), then work through the practical constraints:

• What’s your goal: measure, see, or act? This is the first filter, not the last. If you only need a record to analyse, almost anything works and you can optimise for cost. If you need to act during the trial, you’ve already ruled out per-visit methods like pill counts and ruled in real-time, transmitting systems with alerting, before regimen or budget even enter the picture.
• How complex is the regimen, and is it solid oral dose? Smart dispensers and caps suit oral solids; injectables or complex regimens may point toward DOT or biochemical confirmation.
• How much patient burden can the population tolerate? Elderly, paediatric or chronically ill cohorts may struggle with diaries or vDOT; low-touch automated capture matters more there.
• What does the therapeutic area demand? Some areas (e.g. infectious disease) have a tradition of confirmed dosing; others are dominated by behaviour-over-time questions better served by pattern data.
• Is the trial decentralised or running between site visits? The more dosing happens at home, away from the clinic, the more you need a method that maintains visibility without a visit.
• What’s the budget, and what’s the cost of being wrong? Hardware-based methods cost more up front; the comparison is against the cost of an ambiguous or under-powered result.

In practice, many trials pair an objective primary method (electronic capture of some kind) with self-report for context, rather than relying on any single source.

Where real-time, per-pill monitoring fits

For solid oral dose trials where you want objective data and the ability to respond while it still matters, a smart dispenser sits at the high-granularity, low-burden end of the spectrum. Pill Connect’s dispenser records each individual dispense with a timestamp at the bottle, and the data uploads automatically in the background: nothing for the participant to log, sync or remember.

The part that changes how a trial runs is what happens to that data. Rather than waiting in a dashboard to be checked, real-time monitoring watches the incoming events and emails the site team when a pattern of missed or delayed dosing appears, so a conversation with a struggling participant can happen in week three, not at the week-twelve reconciliation. That push model is what separates the act rung from merely seeing: the signal reaches the site team without anyone having to remember to log in and look, which is what turns adherence data into adherence action, and is where the return actually comes from. It won’t confirm a swallow the way an observed dose does, but for objective, per-pill, real-time behaviour at minimal patient burden, it closes the gap that diaries, pill counts and retrospective downloads leave open.