Why Science? | ICH E19 Selective Safety Data Collection in Clinical Trials

E19 defines selective safety data collection as reduced collection of certain data types in a clinical trial after considering factors that justify the approach. The reduction may apply to case-report-form recording and sponsor reporting. The protocol should say exactly which data, participants or periods are affected and why the remaining evidence is adequate.

The proportionality in selective collection reduces specified safety recording begins with prior knowledge and ends with continued participant care. E19 defines selective safety data collection as reduced collection of certain data types in a clinical trial after considering factors that justify the approach. The reduction may apply to case-report-form recording and sponsor reporting. The protocol should say exactly which data, participants or periods are affected and why the remaining evidence is adequate. Keep this safety boundary visible: Selective collection is a planned data strategy, not permission to ignore symptoms or undocumented operational shortcuts.

A strong science note separates observation from interpretation. List every safety field proposed for reduction and the prior evidence that makes continued comprehensive collection low value. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Next chapter

FDA identifies E19 as final guidance dated December 2022 under docket FDA-2019-D-1828 and issued by CDER. It provides internationally harmonised guidance for specific late-stage pre-approval or post-approval trials. The aim is greater efficiency in large or long trials while safeguarding the welfare of every participant.

For the december 2022 e19 guidance is final, less data can be better science only when the omitted information is unlikely to change protection or benefit-risk. FDA identifies E19 as final guidance dated December 2022 under docket FDA-2019-D-1828 and issued by CDER. It provides internationally harmonised guidance for specific late-stage pre-approval or post-approval trials. The aim is greater efficiency in large or long trials while safeguarding the welfare of every participant. Preserve this guardrail: The guidance is selective in scope and does not make reduced collection appropriate for every product or trial.

A strong science note separates observation from interpretation. Create a source receipt with title, final status, date, docket, issuing office, purpose and scope limitations. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

E19 separates safety data recorded for trial analysis from the healthcare professional’s responsibility to monitor and treat each participant according to prevailing standards of care. Adverse events still belong in medical records, and local or regional safety-reporting obligations continue. A field omitted from the trial database must never become a reason to withhold clinically needed assessment.

A trustworthy selective approach to participant care remains comprehensive states both what remains visible and what becomes unknowable. E19 separates safety data recorded for trial analysis from the healthcare professional’s responsibility to monitor and treat each participant according to prevailing standards of care. Adverse events still belong in medical records, and local or regional safety-reporting obligations continue. A field omitted from the trial database must never become a reason to withhold clinically needed assessment. The design stops here: Reduced sponsor data collection does not reduce the duty of care or legal reporting requirements.

A strong science note separates observation from interpretation. Draw separate pathways for clinical care, medical-record documentation, case-report-form data and expedited reporting. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Justification may draw on regulatory status, mechanism of action, class knowledge, number and diversity of exposed participants, duration of exposure, consistency across trials and experience in relevant populations. No single factor decides the question. More applicable evidence strengthens confidence that routine additional collection would add limited knowledge.

The proportionality in the safety profile should already be sufficiently characterised begins with prior knowledge and ends with continued participant care. Justification may draw on regulatory status, mechanism of action, class knowledge, number and diversity of exposed participants, duration of exposure, consistency across trials and experience in relevant populations. No single factor decides the question. More applicable evidence strengthens confidence that routine additional collection would add limited knowledge. Keep this safety boundary visible: Approval or long use alone does not prove that every population, dose, indication or duration is well characterised.

A strong science note separates observation from interpretation. Build a characterisation matrix and mark gaps in exposure, diversity, duration, mechanism, class and population. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Before approval, comprehensive collection generally remains important for understanding frequency, severity, seriousness, dose response and subgroup differences. E19 says selective collection may be justifiable in rare phase 3 situations when sufficient safety data already exist. The evidence threshold should reflect uncertainty before broad clinical use.

For pre-approval use is expected to be uncommon, less data can be better science only when the omitted information is unlikely to change protection or benefit-risk. Before approval, comprehensive collection generally remains important for understanding frequency, severity, seriousness, dose response and subgroup differences. E19 says selective collection may be justifiable in rare phase 3 situations when sufficient safety data already exist. The evidence threshold should reflect uncertainty before broad clinical use. Preserve this guardrail: Efficiency pressure is not evidence that the pre-approval safety profile is mature.

A strong science note separates observation from interpretation. State which safety questions have already been answered and which still require comprehensive phase 3 data. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

After approval, repeated comprehensive collection can add little when common and non-serious effects are already well described. A large outcomes trial may instead focus on mortality, hospitalisation or a specific concern. Selectivity can reduce burden and support broader participation or longer follow-up while keeping decision-relevant safety evidence.

A trustworthy selective approach to post-approval trials may have different information needs states both what remains visible and what becomes unknowable. After approval, repeated comprehensive collection can add little when common and non-serious effects are already well described. A large outcomes trial may instead focus on mortality, hospitalisation or a specific concern. Selectivity can reduce burden and support broader participation or longer follow-up while keeping decision-relevant safety evidence. The design stops here: A new indication, dose, route or less severe population can reopen questions that were settled in earlier use.

A strong science note separates observation from interpretation. Compare approved and proposed use across population, dose, route, duration, care and outcome to identify reopened risks. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

E19 states that the guidance is not applicable to gene-therapy or rare or orphan-disease trials. When prospective participants are few, comprehensive data from every person can be especially valuable. This explicit exclusion prevents efficiency arguments from reducing information where each observation carries unusual importance.

The proportionality in gene therapy and rare-disease trials are outside scope begins with prior knowledge and ends with continued participant care. E19 states that the guidance is not applicable to gene-therapy or rare or orphan-disease trials. When prospective participants are few, comprehensive data from every person can be especially valuable. This explicit exclusion prevents efficiency arguments from reducing information where each observation carries unusual importance. Keep this safety boundary visible: A small population cannot support the same redundancy assumptions as a very large outcomes trial.

A strong science note separates observation from interpretation. Record whether the product and disease fall inside scope before assessing any selective method. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

E19 lists serious adverse events, important medical events, medication error or overdose, events leading to study-drug discontinuation, pregnancy and lactation exposure and outcomes, and adverse events of special interest among data generally collected. Relevant laboratory abnormalities and events required by protocol or regulation also remain visible.

For certain events should generally remain collected, less data can be better science only when the omitted information is unlikely to change protection or benefit-risk. E19 lists serious adverse events, important medical events, medication error or overdose, events leading to study-drug discontinuation, pregnancy and lactation exposure and outcomes, and adverse events of special interest among data generally collected. Relevant laboratory abnormalities and events required by protocol or regulation also remain visible. Preserve this guardrail: The list is not exhaustive; product-specific signals and trial objectives can require additional data.

A strong science note separates observation from interpretation. Create a protected-data list and link every item to participant action, regulatory duty or benefit-risk interpretation. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

When selectivity is justified, collection may be limited for non-serious adverse events already well characterised, routine laboratory tests, vital signs, physical examinations or concomitant medications that are not central to the objective or safety question. The exact choices depend on the known profile, protocol and population.

A trustworthy selective approach to some routine data may be reduced after justification states both what remains visible and what becomes unknowable. When selectivity is justified, collection may be limited for non-serious adverse events already well characterised, routine laboratory tests, vital signs, physical examinations or concomitant medications that are not central to the objective or safety question. The exact choices depend on the known profile, protocol and population. The design stops here: Data never collected cannot be recovered when an unexpected question appears later.

A strong science note separates observation from interpretation. For each omitted field, write the future question that would become harder or impossible to answer. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

A safety profile characterised in severe disease may not justify selective collection in people with milder disease, where tolerance for risk can be lower. Age, organ function, concomitant treatment and care setting can also alter benefit-risk. The population in the proposed trial should be close enough to prior evidence for the planned reduction.

The proportionality in benefit-risk can change across populations begins with prior knowledge and ends with continued participant care. A safety profile characterised in severe disease may not justify selective collection in people with milder disease, where tolerance for risk can be lower. Age, organ function, concomitant treatment and care setting can also alter benefit-risk. The population in the proposed trial should be close enough to prior evidence for the planned reduction. Keep this safety boundary visible: Similarity in diagnosis does not guarantee similarity in vulnerability or acceptable risk.

A strong science note separates observation from interpretation. Compare prior and proposed populations on disease severity, expected benefit, vulnerability, exposure and monitoring. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

E19 recommends prospective agreement with regulatory authorities on whether the profile is sufficiently characterised and on the planned implementation. A single MRCT protocol can use selective collection when participating authorities agree. Early discussion helps avoid a late discovery that reduced data cannot support the intended submission.

For early agreement with regulators is essential, less data can be better science only when the omitted information is unlikely to change protection or benefit-risk. E19 recommends prospective agreement with regulatory authorities on whether the profile is sufficiently characterised and on the planned implementation. A single MRCT protocol can use selective collection when participating authorities agree. Early discussion helps avoid a late discovery that reduced data cannot support the intended submission. Preserve this guardrail: Regulatory consultation informs the plan but does not transfer the sponsor’s duty to protect participants and justify choices.

A strong science note separates observation from interpretation. Prepare focused questions on eligibility, protected events, reduced fields, subsets, timing, analysis and escalation triggers. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

One method reduces specified data uniformly for everyone while continuing the protected events and trial-specific safety measures. Uniform implementation can make analysis simpler because the same collection rule applies across the trial. Training and case-report forms should prevent sites from confusing “not recorded for the study” with “not assessed in care.”

A trustworthy selective approach to selectivity can apply to all participants states both what remains visible and what becomes unknowable. One method reduces specified data uniformly for everyone while continuing the protected events and trial-specific safety measures. Uniform implementation can make analysis simpler because the same collection rule applies across the trial. Training and case-report forms should prevent sites from confusing “not recorded for the study” with “not assessed in care.” The design stops here: Uniform collection is easier to summarise but can still miss a new pattern outside the protected fields.

A strong science note separates observation from interpretation. Build a site workflow showing clinical assessment, record destination, sponsor reporting and escalation for every event type. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

A trial can collect full safety data in groups where information remains weak, such as older adults, people with organ impairment or underrepresented demographic or genetic groups, while using selective collection elsewhere. This targets uncertainty without requiring every participant to carry the same data burden.

The proportionality in specific subgroups may keep comprehensive collection begins with prior knowledge and ends with continued participant care. A trial can collect full safety data in groups where information remains weak, such as older adults, people with organ impairment or underrepresented demographic or genetic groups, while using selective collection elsewhere. This targets uncertainty without requiring every participant to carry the same data burden. Keep this safety boundary visible: Subgroup selection should reflect evidence gaps and avoid treating convenience samples as representative.

A strong science note separates observation from interpretation. Choose comprehensive subgroups from the characterisation matrix and justify their size, duration and measures. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

Another method collects complete safety data for an initial period and reduces specified collection after participants reach a predefined follow-up time. This may fit long-duration prevention or outcomes trials once common non-serious events and laboratory changes are well described. The transition rule should be prospective and operationally clear.

For comprehensive collection may occur early, then taper, less data can be better science only when the omitted information is unlikely to change protection or benefit-risk. Another method collects complete safety data for an initial period and reduces specified collection after participants reach a predefined follow-up time. This may fit long-duration prevention or outcomes trials once common non-serious events and laboratory changes are well described. The transition rule should be prospective and operationally clear. Preserve this guardrail: Late-onset, cumulative or delayed risks may require continued targeted monitoring beyond the early period.

A strong science note separates observation from interpretation. Draw a safety timeline and show which plausible risks emerge early, accumulate gradually or appear after long latency. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

In a very large trial, comprehensive data may be collected at randomly selected sites or from randomly selected participants, with selective collection elsewhere. The subset should represent the full trial in geography, demographics, disease and treatment. Selection methods belong in the protocol so readers can assess bias and precision.

A trustworthy selective approach to representative subsets can carry deeper measurement states both what remains visible and what becomes unknowable. In a very large trial, comprehensive data may be collected at randomly selected sites or from randomly selected participants, with selective collection elsewhere. The subset should represent the full trial in geography, demographics, disease and treatment. Selection methods belong in the protocol so readers can assess bias and precision. The design stops here: A convenient subset can distort event frequency and subgroup patterns even when it is large.

A strong science note separates observation from interpretation. Simulate a random subset plan and compare its representation with the whole trial before enrolment and during conduct. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

When comprehensive and selective data are collected for different participants or periods, their frequency summaries may not be directly poolable. Protocols and analysis plans should identify denominators, collection windows, regimes and methods for aggregation. Reports should show where interpretation is affected by missing-by-design data.

The proportionality in analysis must respect different collection regimes begins with prior knowledge and ends with continued participant care. When comprehensive and selective data are collected for different participants or periods, their frequency summaries may not be directly poolable. Protocols and analysis plans should identify denominators, collection windows, regimes and methods for aggregation. Reports should show where interpretation is affected by missing-by-design data. Keep this safety boundary visible: Treating uncollected fields as absence of an event will underestimate frequency and misstate uncertainty.

A strong science note separates observation from interpretation. Create analysis populations and denominators for every collection regime, event type and follow-up window. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter · Next chapter

The receipt records product status, known profile, prior exposure, population comparison, protected events, reduced fields, subset or time rules, regulator agreement, site training, care duties, reporting pathways, analysis denominators, new-signal triggers, reversions to comprehensive collection and interpretation limits. It shows exactly what efficiency cost and what protection remained.

For a complete e19 receipt preserves safety and limits, less data can be better science only when the omitted information is unlikely to change protection or benefit-risk. The receipt records product status, known profile, prior exposure, population comparison, protected events, reduced fields, subset or time rules, regulator agreement, site training, care duties, reporting pathways, analysis denominators, new-signal triggers, reversions to comprehensive collection and interpretation limits. It shows exactly what efficiency cost and what protection remained. Preserve this guardrail: A clear receipt cannot answer a safety question about data that the trial intentionally never collected.

A strong science note separates observation from interpretation. Ask an independent safety reviewer which new signal the design could detect, which it might miss and when collection should intensify. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.

Contents · Previous chapter