What ICH M14 Changes About How Sponsors Choose Real-World Data Sources

Real-world data sourcing used to happen after the protocol was written. A team defined the research question, drafted the design, then went looking for a dataset that could answer it. Procurement followed science. However, ICH M14 now reverses that order.

What ICH M14 covers

The International Council for Harmonisation develops technical standards that regulators in the US, Europe, and Japan adopt into their own frameworks. An ICH guideline becomes binding when individual agencies adopt it. M14 reached Step 4 adoption by the ICH Assembly in September 2025 and entered the implementation phase, which is what the 2026 regional effective dates represent.

M14 addresses non-interventional studies that use real-world data for safety assessment of medicines. The guideline published by the European Medicines Agency describes an iterative approach built on assessing data fitness-for-use, running feasibility assessments to guide study design, and refining the design based on what feasibility reveals. The scope is postmarketing safety, though EMA has noted that the core principles may also apply to effectiveness studies.

The practical significance is harmonization. Sponsors have been designing region-specific studies to satisfy different regulators asking similar questions. M14 is intended to reduce that duplication and improve the odds that one protocol is acceptable across regions.

The guidance took effect in March 2026. FDA adopted it that month, replacing the July 2024 draft and withdrawing its 2013 best practices guidance on pharmacoepidemiologic safety studies using electronic healthcare data. EMA set a legal effective date of March 18, 2026.

The headline change is methodological rigor. The operational change is sequence.

Feasibility now shapes design

Read that sequence again. Feasibility is an input to design under this framework. A sponsor who locks a protocol and then discovers the available data cannot support the exposure definition has done the work in the wrong order.

That has a direct consequence for how sponsors evaluate data partners. The evaluation has to happen early enough to change the protocol. It has to produce answers specific enough to act on. A data partner who can tell you patient counts but cannot tell you variable completeness, capture consistency, or lookback depth is giving you a number you cannot design around.

What the guidance asks sponsors to document

M14 covers the full study arc. Articulating the research question. Selecting appropriate data sources. Defining key variables. Addressing potential biases and confounding. Conducting analyses. Reporting and submission.

Every step carries a documentation obligation. The research question drives the data requirements. The data requirements drive source selection. Source selection has to be justified against the question rather than against availability.

That justification is a fitness-for-use argument. We covered the underlying framework in What Makes Real-World Data “Research-Ready”? A Fitness-for-Use Framework, which lays out the relevance and reliability tests and the questions to ask any data source. M14 raises the stakes on that evaluation in one specific way. The assessment has to be documented, and it has to happen early enough to change the protocol.

The parallel shift on the device side

M14 did not arrive alone. In December 2025, FDA finalized its guidance on using real-world evidence to support regulatory decision-making for medical devices, superseding the 2017 version. The agency set a 60-day transition period and began anticipating compliant submissions on February 17, 2026.

That guidance moved in a different direction on one point. FDA removed the prior expectation that submissions include confidential information at the individual patient level, noting that the requirement made large macro-level databases impractical to maintain. The agency assesses relevance and reliability case by case, with relevance considering whether data are sufficiently detailed, representative of the target population, and appropriate for the regulatory question.

Taken together, the two developments point the same way. The privacy barrier to using large de-identified datasets came down. The standard for documenting what is in those datasets went up.

What this means for data partner selection

The sponsors who adapt fastest will be the ones who change what they ask for during evaluation. Patient counts are a starting point. The questions that matter under M14 are narrower and harder.

• What percentage of records contain the variables this protocol depends on?
• How consistently are those variables captured across contributing sites?
• What is the lookback period, and does it cover the required exposure window?
• Can the source produce documentation of data lineage that will hold up in a submission?
• How quickly can a feasibility question be answered, given that the answer needs to inform design?

That last question is easy to overlook. If a feasibility assessment takes eight weeks, it cannot realistically shape a protocol on a normal development timeline. Speed is a methodological requirement under this framework, because slow feasibility gets skipped, and skipped feasibility produces exactly the design failures M14 is written to prevent.

The site layer is where provenance starts

Documentation obligations run downstream from where data are created. A sponsor can only document provenance to the extent the data source can.

This is why the structure of a data network matters as much as its size. Data aggregated through opaque intermediaries carries gaps that surface late, usually during submission preparation, when they are most expensive to fix. Data sourced through direct EHR integration with named, participating sites carries a traceable path from the point of care forward.

BEKhealth’s network reflects that structure. More than 30 million patients across 200 research sites, with over 20,000 HCPs and more than 1,000 trials of research experience. Site participation in real-world data contribution is opt-in, which means every contributing site is a known, consenting participant in the data supply chain.

Where to start

If you have non-interventional studies planned for the next 18 months, three actions are worth taking now.

1. Audit your current data sources against the M14 documentation requirements. Identify which sources can produce provenance records and which cannot.
2. Move feasibility earlier. Build it into protocol development rather than treating it as a downstream check.
3. Change your evaluation criteria. Ask data partners for variable-level completeness and provenance documentation during selection, before contracting.

The guidance rewards sponsors who treat data selection as a scientific decision. That work happens at the beginning of a study, and it depends on having a data partner who can answer specific questions quickly.

Further reading

For a detailed analysis of FDA’s final guidance on real-world data and real-world evidence for drug and biological products, see this summary from RTI Health Solutions. BEKhealth provides the underlying real-world data for collaborative research programs with RTI Health Solutions, which contributes health economics and outcomes research expertise.