MTL-607: Building the Evidence Case for Digital Medication Support

What you will learn

Plan evidence that connects a digital function to a defensible claim and meaningful benefit.

Part 3 of a three-module pathway: measurement, technology validation and evidence planning.

Separate four contributors to the outcome

Contributor	Potential contribution	Evidence question
Medicine	Pharmacological benefit and adverse effects.	Which outcomes are already supported for the medicine and population?
Delivery device	Administration and delivery performance.	Does it deliver as intended under relevant conditions?
Software	Recording, reminders, analysis, presentation or decision support.	Does this function work, and what difference does its use make?
Clinical service	Prescribing, monitoring, coaching and escalation.	Which staffing, contact and treatment decisions explain outcomes?

A favourable outcome from the combined service does not allocate benefit between these components. Draw the system boundary and describe the intervention precisely enough that another team could reproduce it, including staffing, medicine access, training, contact frequency and escalation rules.

Define intended use, functions and claims

Identify the patient population, users, environment, clinical decision, workflow and jurisdiction. Write each claim at the level the evidence can support. “Displays recorded device events” is a technical function. “Improves persistence” is a comparative behavioural claim. “Improves health outcomes” is a clinical claim. “Reduces service costs” requires an economic evaluation.

Build a causal chain: reliable observation → understandable information → appropriate action → patient benefit. At every step, state the assumption and how it will be tested. Extra information can create workload or inappropriate intervention, so measure harms and burden as well as intended benefit.

Agree the proposed claim before choosing convenient available data. If the business claim changes from record keeping to reducing medication errors, revisit intended use, risk analysis, validation and the evidence plan.

FDA prescription drug use-related software: apply the scope first

The October 2026 final FDA guidance addresses drug labelling for software outputs disseminated by or on behalf of a drug sponsor and related to its prescription drug, including CDER/CBER-led combination products. Independent software is outside this guidance’s scope; device requirements may still apply.

Within scope, distinguish device-connected functions from standalone functions. FDA considers whether software is essential for safe and effective use, supports a meaningful clinical improvement, or relies on directly transferred device data. Device-connected functions are typically described in prescribing information even without a demonstrated clinical improvement; this does not establish a benefit claim. Standalone functions generally need an additional basis, such as essential use or demonstrated clinical improvement, for inclusion.

A meaningful clinical improvement claim needs adequate, well-controlled evidence. Drug labelling and device-authorisation questions remain distinct. The guidance does not mandate one injection-tracking technology or make a particular connected device necessary for clearance. Use its decision tree with the full text and discuss the proposed evidence with the relevant review division. FDA final guidance, October 2026.

NICE: conditional use is an evidence-generation pathway

NICE HTG700 permits specified digital multidisciplinary weight-management technologies, including Roczen, to be used in the NHS while further evidence is generated, subject to stated conditions. This is not an unconditional finding that every claimed benefit or economic saving has been established. Check the current recommendation, population, regulatory prerequisites and evidence-generation period for the technology. NICE HTG700 recommendations.

Its evidence plan identifies outcomes including weight, programme adherence and completion, adverse-event monitoring, quality of life, psychological outcomes and resource use. Comparative evidence, consistent follow-up and careful handling of differences between populations matter. NICE evidence-generation plan.

For your project, turn remaining uncertainty into an owned work plan: question, endpoint, dataset, comparator, collection schedule, quality controls, analysis and decision date. A technical validation report cannot close an unanswered clinical or economic question.

Keep three evidence layers distinct

Layer	What it can establish	What it cannot establish alone
Technical performance	Detection, timing, attribution, completeness, reliability and correct display.	Improved behaviour, clinical outcome or value for money.
Clinical and service benefit	Whether use changes appropriate decisions, outcomes, safety or patient experience.	Affordable implementation or favourable cost effectiveness in another setting.
Economic value	Resources consumed and outcomes gained under a stated perspective and time horizon.	Clinical causality if underlying effectiveness evidence is biased.

Build the layers in sequence, but plan them together. A highly accurate event stream may be clinically irrelevant if no one can act on it. A useful service may be costly if false alerts require repeated manual review. Avoid double-counting savings when freed staff time is not actually released or redeployed.

Choose a design that isolates the contribution

A randomised comparison can estimate the incremental effect of adding a defined digital function to otherwise comparable care. Specify what both groups receive and prevent unplanned differences in medicine access, funding, contact frequency or clinical staffing from becoming the explanation for the result.

When individual randomisation would contaminate routine clinical practice, consider cluster allocation with enough sites and an analysis that accounts for clustering. If randomisation is infeasible, a prospective concurrent comparison can be informative, but adjustment cannot eliminate unmeasured confounding. Document baseline differences and reasons for treatment allocation.

Use equivalent endpoint ascertainment across arms. If only the intervention arm has a detector, its higher event count may reflect better measurement rather than better adherence. Decide how outcomes will be independently measured without giving the comparator the intervention’s active feedback. Prespecify the estimand, analysis population, missing-data strategy, treatment switches and sensitivity analyses.

Choose outcomes that reflect useful and safe care

Outcome	Operational definition to agree	Important countermeasure
Persistence	Time to a prespecified discontinuation event, with supply gaps and restarts defined.	Appropriate discontinuation, adverse effects and clinician-directed pauses.
Appropriate intervention	Proportion of independently adjudicated actionable situations receiving timely suitable action.	Unnecessary contacts, false alerts and inappropriate treatment changes.
Safety	Medication errors, duplicate doses, adverse events and delayed escalation.	Ascertain events consistently in both arms.
Workload	Staff time by activity, alerts and contacts per patient.	Hidden onboarding, troubleshooting and after-hours work.
Quality of life	Validated measure, population, schedule and meaningful difference.	Missing responses and differential completion.
Economic value	Total relevant costs and outcomes over an explicit horizon and perspective.	Device replacements, platform costs, training, integration and uncertainty.

Choose one primary endpoint linked to the intended claim; identify secondary and exploratory outcomes in advance. Report effect sizes and uncertainty, not only statistical significance. Greater persistence is beneficial only when treatment remains appropriate.

Worked example: digital GLP-1 care with and without monitoring

This is a proposed study, not a Roczen study result. Both arms receive the same digital care service, prescribing approach, access arrangements and scheduled support. The intervention arm also receives a validated connected injection monitor and a defined workflow for reviewing uncertain events. The comparator receives the service without connected monitoring.

The primary question is whether the added monitoring-and-response package produces a meaningful incremental benefit. It cannot isolate the sensor from the additional response workflow. A separate design would be needed to distinguish passive data collection, patient feedback and clinician alerts.

Plan element	Illustrative proposal
Population and baseline	Adults starting a defined injectable treatment pathway; record prior use, clinical characteristics and funding.
Primary endpoint	A prespecified measure of appropriate, timely intervention, or persistence if that is the proposed claim; justify the choice clinically.
Common measurement	Independent, consistently collected endpoint data in both groups; do not define success only from intervention-device logs.
Follow-up	Long enough for the claimed benefit; 48 weeks is an illustrative choice, not a universal requirement.
Analysis	Intention-to-treat effect of offering the package, with prespecified handling of withdrawal, switching and missingness.
Secondary outcomes	Safety, treatment burden, quality of life, staff time, costs and observation completeness.

If alerts lead to more contacts but no better outcomes, report that result. If benefit appears only in a subgroup, determine whether it was prespecified and sufficiently supported before narrowing or expanding the claim. The Roczen example in MTL-605 motivates questions about measurement; it does not supply evidence for the incremental monitoring effect.

Practical takeaway: claims-and-evidence planning

Claim	Evidence package	Decision gate
Correctly displays injection-event history.	Sensor validation, identity and timing checks, complete-path verification and user interpretation.	Performance and risk-based acceptance limits met on the controlled configuration.
Improves timely appropriate intervention.	Comparative study with independent adjudication and safety/burden endpoints.	Clinically meaningful effect with acceptable uncertainty and harms.
Offers value to the service.	Comparative resource data and an economic model with sensitivity analyses.	Value acceptable for the stated payer, setting and implementation assumptions.

Download / open the claims-and-evidence planning template — includes fields for scope, assumptions, comparator, endpoint, analysis, owner and lifecycle review.

Exercise: a team has 98% event detection on the bench and proposes “reduces hospital admissions”. List the missing links. A useful answer identifies representative-use performance, correct interpretation, an effective intervention pathway, comparative clinical evidence for admissions, adequate follow-up and harms, and the limits of extrapolating from an engineering metric.

References and further learning

Sources checked 24 October 2026. Study designs and template entries are teaching proposals; product-specific evidence needs depend on the intended claim and pathway.