What you will learn
Select measures that support the intended adherence claim.
Part 1 of a three-module pathway: measurement, technology validation and evidence planning.
Define the behaviour before measuring it
A digital service can be busy, well used and valued without establishing how a medicine was taken. Start by naming the behaviour, observation period, prescribed regimen and population. A single percentage labelled “adherence” leaves too much open to interpretation.
| Term | Meaning | Example measure |
|---|---|---|
| Engagement | Interaction with a service or tool; define the activity being counted. | Consultations attended or weekly active users. |
| Medication adherence | The process of taking medicines as prescribed, including initiation, implementation and discontinuation. | Separate measures for starting, following the regimen and stopping. |
| Initiation | Taking the first prescribed dose. | Time from prescription to first dose; dispensing alone is a proxy. |
| Implementation | How actual dosing follows the prescribed regimen while treatment continues. | Proportion of scheduled doses supported by appropriate evidence within a defined window. |
| Persistence | Duration from treatment initiation to the last dose preceding discontinuation. | Time to discontinuation using a prespecified rule. |
| Discontinuation | Stopping treatment; distinguish a stop from a temporary interruption. | Reason, date, restart policy and treatment-switch rule. |
The medication terms follow Vrijens et al. (2012). Engagement is an operational service measure. A clinician-directed pause, dose change or appropriate discontinuation must be represented in the expected regimen; continuing at all costs is not a desirable endpoint.
What does each evidence source observe?
| Source | Direct observation | Inference and limitation |
|---|---|---|
| Dispensing records | Medicine supplied on a recorded date. | Possession or continued supply; not administration. Stockpiling, changes in dose and external pharmacies affect interpretation. |
| Patient reports | What a patient recalls or records. | Can explain reasons and circumstances; subject to recall, entry timing and reporting bias. |
| Packaging events | Opening, removal, scan or other defined package interaction. | Access to medicine; an opening can occur without use and a dose can be removed for later use. |
| Device signals | A measured state, movement, contact or event signature. | Supports only the event definition validated for that sensor and configuration. |
| Clinical outcomes | A change such as weight, symptoms or a laboratory result. | May support benefit, but cannot reconstruct every administration or isolate adherence without further evidence. |
Proportion of days covered (PDC) estimates the share of eligible days with medicine available from supply records. Medication possession ratio (MPR) relates supplied days to a defined interval and can exceed 100% in some implementations. State the calculation, overlap handling and denominator. Neither is a direct dose-taking measurement. A generic threshold such as 80% needs justification for the medicine, regimen and claim; it is not a universal clinical boundary.
An injection event is not automatically a delivered dose
Separate the observations: device activated; injection sequence progressed; mechanism reached its expected end state; fluid left the container; the intended quantity entered the intended tissue. Each statement requires additional evidence. A completion signal may reflect a mechanism position even when early removal, leakage or another failure compromises delivery.
A bench measurement of expelled mass can support delivery performance under specified conditions. It does not, by itself, establish successful delivery into a patient. Define the physical quantity, route, conditions and uncertainty behind any delivered-dose claim. Combine sensors only where the combined interpretation has been validated; agreement between two correlated proxies is not independent confirmation.
Use display labels such as “device event recorded”, “patient-reported administration” or “status unknown” when those accurately describe the available evidence. Avoid silently converting all three into “dose taken”.
Make uncertainty part of the measurement
- Missing data: distinguish no event detected, sensor not worn, no connectivity, battery failure and unavailable follow-up. None alone proves a missed dose.
- False events: challenge priming, handling, training devices, off-body activation, duplicate uploads and signals from another person.
- Timing: distinguish event time, phone receipt time, server arrival and retrospective entry; retain time zone and clock-quality information.
- Attribution: establish which patient, device, medicine and regimen a record belongs to; manage shared phones and replacement devices.
- Observation coverage: record when the system was able to observe. Report unknown periods and sensitivity analyses instead of silently dropping them.
Keep raw observations, interpretation and later corrections traceable. Predefine reconciliation rules when dispensing, reports and sensor records disagree. An adjudicator should not know the desired study result.
Read the denominator before the headline
Ask who entered the cohort, when follow-up began, who had enough calendar time to reach the endpoint, who withdrew and who contributed each measurement. Administrative immaturity is different from loss to follow-up. Reporting only people with complete records may select those doing well.
- Denominator: all eligible participants, treatment starters, retained participants or available measurements?
- Attrition: how many left, why, and were outcomes sought after service withdrawal?
- Comparator: concurrent and comparable, or an external headline using a different definition?
- Confounding: funding, access, disease severity, medicine, dose, support intensity and motivation can affect both group membership and outcome.
- Analysis: prespecified endpoints, uncertainty intervals, missing-data assumptions and adjustments should accompany point estimates.
Illustrative arithmetic: 100 people begin a programme; 70 remain contactable and 56 report all scheduled doses. That is 80% of respondents, but 56% of the original cohort with a positive report. The other 30 have unknown status, not proven non-adherence. The study should state its estimand and examine credible missing-data assumptions.
Roczen: read the operational definition
Reset Health’s UKCO 2026 company report describes 180 patients reaching the 48-week milestone: 80 NHS-funded and 100 self-pay. Reported persistence was 84% and 65%, respectively, defined by a dispensing record within eight weeks of that milestone. Among patients with a recorded weight at week 48, mean weight loss was 16.7% and 19.3%. Both groups used the same digital clinical service. The comparison was not randomised; clinical characteristics, medicines and dosing differed, and weight data were incomplete. Source: Reset Health report and Notes to Editors.
These findings describe supply-based persistence and available weight outcomes in the reported cohorts. They do not establish injection-by-injection adherence, delivered dose, or a causal contribution from the software alone. The report does not demonstrate an incremental benefit from connected injection monitoring. Treat this as a company-reported conference study; a full study report is needed to resolve cohort selection, endpoint denominators, missingness and adjusted comparisons.
Learning task: rewrite “the app proves patients took 84% of their injections” using only the supported observation. A defensible answer identifies the NHS cohort, 48-week milestone, 84% persistence estimate and dispensing-based definition.
Practical takeaway: claim-to-measure mapping
Complete one row for each proposed claim. The worksheet should make it possible for a reviewer to trace the claim back to an observable quantity, and to see where inference remains.
| Proposed claim | Measure and denominator | Evidence and qualification |
|---|---|---|
| Patients continue obtaining treatment at 48 weeks. | Participants meeting a prespecified supply rule / eligible cohort; report missing records. | Dispensing data; distinguish supply from administration and external supply gaps. |
| Scheduled injection events are recorded. | Matched valid events / scheduled opportunities during defined observation. | Validated event detector, regimen history and coverage; report false events and unknown periods. |
| The intended dose is delivered. | Delivery quantity and success under defined conditions. | Independent physical reference and representative-use evidence; account for leakage and incomplete delivery. |
| The service improves adherence. | Difference in a prespecified adherence endpoint versus an appropriate comparator. | Comparative design; distinguish intervention effect from selection, access and support differences. |
Download / open the claim-to-measure mapping worksheet — an editable, printable blank template with a worked row.
Review your claim
- Name the construct: engagement, implementation, persistence or clinical benefit.
- State what was measured directly and what was inferred.
- Define denominator, observation window, regimen and unknown states.
- Use comparative evidence before claiming improvement caused by a technology.
Continue with MTL-606 for technology selection and validation, then MTL-607 for the overall evidence case.
References and further learning
- Vrijens et al. — A new taxonomy for describing and defining adherence to medications (2012).
- Reset Health — UKCO 2026 study report; company source, including methodological limitations.
- FDA — Digital Health Technologies for Remote Data Acquisition in Clinical Investigations (December 2023).
Sources checked 24 October 2026. Practical examples and worksheet criteria are teaching proposals, not universal acceptance thresholds.