ICH E19 Safety Data Collection in 2026: CRF Scope and Analysis Denominators

TL;DR

ICH E19 allows a justified reduction of selected safety data in certain late-stage trials, principally after approval. Its practical consequence is specific: the case report form, the collection schedule and the analysis population must be designed together. Section 2.4 describes six classes that should generally remain collected; section 2.5 identifies data that may be limited. Section 4 prohibits pooling comprehensive and selective data when collection methods differ across participants or over time. [1] [2] [3]

For a study team, the useful output is a module-by-population-by-period matrix. This report compares E19 with FDA's 2016 guidance and tests the implications against the collection schedules reported in FDA's C3TI white paper. The published examples include annual liver testing in a 20% subset and country-specific follow-up creatinine collection. Those schedules change the population available for a particular laboratory analysis, even while serious adverse events continue to be collected more broadly. [4]

Start by identifying which participants and observation periods receive the fuller schedule. The table below translates the section 3 examples into collection boundaries, while the protocol and analysis plan define the final implementation. [1]

FDA issued its E19 guidance in December 2022; the availability notice appeared on 6 December. The document contains nonbinding recommendations. EMA records 16 March 2023 as the Step 5 guideline's effective date. These dates establish the status of the documents, while the proposed trial still needs prospective regulatory discussion. [2] [5] [6] [7]

How the four examples translate into a collection design

Section 3 provides four non-exhaustive implementation examples: selective collection throughout; a specific population subset; a change after an initial period; and a representative subset. They are starting points for a protocol-specific design, including a justified combination, rather than four exclusive approval categories. [1]

The Step 4 text was adopted on 27 September 2022. Selective collection "refers to the reduced collection of certain types of data in a clinical trial after thorough consideration of factors that would justify such an approach." Section 2.1 locates that reduction: investigators record "certain data (see Section 2.5)" on the case report form and report those data to the sponsor. [1]

Section 4 is the summary rule:

If selective safety data collection is used uniformly in a clinical trial (i.e., for all patients in a trial; throughout the trial), analyses are straightforward; however, frequency summaries would need to describe the specifics of the approach for safety data collection to enable interpretation of the results. In contrast, if the methods of safety data collection are not consistent for all patients in a clinical trial or are not consistent throughout the duration of a trial, data obtained under a comprehensive safety data approach and data collected under a selective safety data approach cannot be pooled. Methods for aggregating the data should be delineated within the protocol and associated analysis plans. [1]

FDA and EMA reproduce the harmonised pooling paragraph. The module-level interpretation used here is editorial analysis: distinguish data with different ascertainment, preserve the relevant populations and observation periods, and specify the statistical method in the analysis plan. Consistent collection removes this particular source of incompatibility; treatment groups, exposure, missing observations and the scientific question still determine whether a combined summary is appropriate. E19 prescribes neither an SDTM domain nor an ADaM flag. [2] [3]

The short names in the table are labels for the section titles. The examples under each title are illustrations.

Choose the collection design before programming the forms. Record any population, country or time boundary that changes what investigators are asked to report.

Comprehensive collection, with no selective reduction, sits outside these examples. Section 4's pooling sentence applies when two methods were used. [1]

FDA's April 2024 guidance recap groups the approaches into three broad families, combining the two subset examples. The guideline itself is the more detailed basis for specifying the collection design. FDA's C3TI demonstration-project page describes eligibility for discussing such approaches with the agency. [6] [8]

Who stays on comprehensive collection?

The four examples are methods that may be considered once selective collection is the approach under discussion. Section 1.3 and section 2.6 say who is still on comprehensive collection.

E19 mainly addresses interventional trials in the post-approval setting. Comprehensive collection remains the general expectation before approval, including characterization of event frequency, severity, seriousness, dose-response and population differences. Exceptionally, sufficient safety data from completed trials may justify selective collection in phase 3. A phase label alone provides insufficient justification. [1]

Gene-therapy and rare/orphan-disease trials are outside E19's scope. The guideline explains that their limited prospective populations warrant comprehensive safety data from every participant; it supplies no numerical prevalence threshold for this exclusion. [1]

After approval, further comprehensive collection may add limited clinically important knowledge. E19 allows consideration of selective collection where trial objectives and participant welfare remain protected. [1]

Section 2.6 cautions against transferring a safety characterization from severe disease, such as late-stage cancer or heart failure, directly into a less severe population, such as migraine or hypertension. The latter population may still warrant comprehensive collection. This is a benefit-risk and population-comparability judgment, leaving room for justified oncology applications. [1]

Section 2.8 lists situations in which selective collection may be appropriate after the profile is judged sufficiently characterised, and says the list "is not meant to be comprehensive": a new indication in a similar or already studied population, additional endpoints in the same population, a safety trial focused on specific parameters, and an additional efficacy trial when the profile has already been judged well characterised. [1] A situation is the reason to open the discussion. The form follows the proposed implementation, informed by the non-exhaustive examples.

Section 2.2 calls for an assessment across several factors. A single factor, whether present or absent, is insufficient to decide that the safety profile is adequately characterized. [1]

Section 2.7 recommends prospective agreement with regulatory authorities on the safety characterization and the planned implementation. For a multiregional protocol, it describes an approach accepted by all relevant authorities and strongly recommends scientific advice in each region where the results will be used. Trial-specific correspondence is needed to establish what each authority actually accepted. [1]

What leaves the case report form, and what stays in the medical record?

Section 2.4 classes should generally be collected, and, when selective collection is justified, section 2.5 classes may be limited. The record that changes is the case report form.

Selective collection changes sponsor-directed CRF recording and reporting. Investigators retain their responsibilities for clinical monitoring, usual care, adverse-event documentation in medical records, and applicable local or regional safety reporting. Those responsibilities continue under the reduced CRF schedule. [1]

The glucose example links this distinction to a specific laboratory measurement.

In E19's glucose example, monitoring required by usual clinical practice continues. Recording glucose in the CRF depends on the protocol and the event: stipulated assessments, clinically relevant findings, an adverse event of special interest, or an association with a serious adverse event bring the information into sponsor collection. [1]

Labelled monitoring continues in care. The value stays off the form unless the protocol, clinical relevance, or a serious adverse event brings it back. The example is one labelled adverse reaction. It leaves the laboratory in clinical care, and it leaves the investigator's reporting duty in place.

Baseline collection remains governed by the trial objectives, including eligibility and planned subgroup assessment. Reducing an on-treatment vital-sign schedule therefore leaves the required baseline measurements in place. [1]

Section 3 says the approach "should be carefully planned and clearly described" in the protocol, the monitoring plan, and the statistical analysis plan, with reference to the guideline. Because investigators "may not be familiar with selective safety data collection, CRFs should be well-designed, and investigators should receive appropriate training." Presented results should describe the approach in the documents the section names, including ICH E2F, ICH E3, and ICH M4. [1]

Which classes should generally stay on the form?

Section 2.4 identifies the following classes for general collection, while explicitly leaving the list open to additional relevant data. [1]

1. Serious adverse events, pointing to ICH E2A and ICH E6.
2. Important medical events, pointing to ICH E2A.
3. Medication error or overdose, intentional or unintentional.
4. An adverse event that led to study-drug discontinuation. No minimum duration is stated.
5. Pregnancy and lactation exposure and outcomes.
6. Adverse events of special interest, including laboratory abnormalities identified in the protocol as critical to safety evaluations.

Section 2.4 also allows consideration of selective collection within these classes where regional or local regulations permit it and adequate protocol justification supports it. Any such proposal needs that separate justification. The section 3 examples focus their reductions on section 2.5 data. [1]

Endpoint serious adverse events that "would not be subject to unblinding and expedited reporting" "should be prospectively agreed upon" and "collected and monitored on a regular basis through an independent data monitoring committee." [1] Endpoint status changes the unblinding and expedited path. The events stay on the form. In the section 3.1 cardiovascular-outcomes example, deaths, non-fatal myocardial infarctions, and non-fatal strokes "would always be collected as described in Section 2.4." [1]

FDA's February 2016 guidance remains a relevant comparison: the December 2022 E19 document cites it in section V.B, and FDA's guidance page labels it Final. The E19 availability notice identifies the June 2019 E19 draft as the document being revised. These records support consulting both texts, with any study-specific differences resolved prospectively with the review division. [9] [2] [10] [5]

Section III.D of the 2016 guidance says these data "are generally not appropriate for selective safety data collection and should always be collected": all serious adverse events; non-serious adverse events that lead to dose modification, drug discontinuation, or withdrawal from the trial; unscheduled study visits, hospitalizations, and accidental injuries, "because these events may reflect serious adverse events of the drug"; in oncology, all Grade 3 and Grade 4 adverse events and Grade 2 adverse events that affect vital organs, with heart and liver as the examples; and, in rare-disease programs, complete safety data. [9]

Study-drug discontinuation appears in both texts. The 2016 guidance also explicitly retains non-serious events leading to dose modification or trial withdrawal. E19's non-exhaustive section 2.4 list omits those exact labels, while section 2.5 addresses non-serious events more generally. That textual difference leaves a protocol-specific question; it establishes neither a blanket prohibition nor automatic permission to omit them. Our conservative build recommendation is to retain those records unless the proposed reduction has been justified and agreed prospectively. [1] [9]

The 2016 guidance additionally identifies unscheduled visits, hospitalizations and accidental injuries because they may reflect serious adverse events. E19 refers to ICH E2A for seriousness. An adverse event involving inpatient hospitalization or prolonged existing hospitalization can meet a seriousness criterion, while an admission or procedure needs assessment in its clinical context. Keep that assessment distinct from a general visit log. [9] [4]

For oncology, the 2016 guidance identifies all Grade 3 and Grade 4 events and Grade 2 events affecting vital organs. E19 organizes section 2.4 around event classes rather than grade cutoffs. This difference matters when an oncology team proposes a more selective schedule, as discussed below. Both texts support comprehensive collection in the rare-disease context. [9] [1]

Which classes may be limited?

"When selective safety data collection is justified, it may be acceptable to limit collection of certain data," with participant safety kept as section 2.1 describes it. [1]

Section 2.5 permits consideration of omitting some non-serious events or reducing collection frequency. First check whether the event also belongs to a section 2.4 class, such as discontinuation, medication error or a protocol-defined adverse event of special interest. Its seriousness alone is an incomplete basis for the collection rule. [1]

Routine serum chemistry, haematology, electrocardiography and imaging can be candidates for reduced collection. A protocol-critical laboratory abnormality or information needed to characterize a section 2.4 event follows the relevant event rule. The operational distinction is between scheduled routine collection and event-triggered collection. [1]

Physical examination and vital-sign data are also candidates for reduced frequency or omission. Baseline assessments follow section 2.3; subsequent collection follows the approved study objectives and schedule. [1]

After baseline concomitant medication is documented, selective collection can omit subsequent changes. That choice closes off some later analyses of interactions, dose changes and temporal associations because the required observations were never captured. The same trade-off applies to omitted laboratories and blood-pressure measurements. [1]

If a section 2.4 event must be collected, "sponsors may also need to collect relevant information (e.g., medical records, laboratory data) to characterise such event(s) should they occur, including any follow-up information." [1] Dropping the routine chemistry panel leaves the triggered page in place for a serious event, an important medical event, a medication error, a discontinuation, a pregnancy or lactation outcome, or an adverse event of special interest. Every pattern needs that triggered path, including the pattern that omits routine section 2.5 data for everyone.

Section III.C of the 2016 guidance says exposure information in pediatric patients, pregnant and nursing women, and geriatric populations "is often limited, so that selective safety data collection may not be appropriate in those populations." [9] E19 item 5 keeps pregnancy and lactation exposure and outcomes on the form. A screening pregnancy test, when the schedule includes one, is a laboratory value.

Selective collection for all patients

Under section 3.1, section 2.4 "would be collected throughout the trial," and "some or all" of the section 2.5 data "would be limited." [1] The method matches for every participant and for the whole trial, so section 4 calls the analyses straightforward. The frequency summary still describes what was limited. A non-serious field the form did not ask for has no trial-wide rate. The summary says the class was not collected.

E19 illustrates an approved drug associated with increased transaminases. A trial evaluating the monitoring approach can retain regular transaminase assessments and section 2.4 events while limiting other routine data. Specify the transaminase analysis population and interval explicitly. Other omitted chemistry tests have no corresponding observed frequency to report. [1]

Its cardiovascular-outcomes example retains deaths, non-fatal myocardial infarctions and non-fatal strokes, which may undergo adjudication, while limiting routine section 2.5 data. Event characterization remains available when needed. The planning trade-off is concrete: an adjudicated endpoint analysis can remain possible while exploratory questions about uncollected concomitant medication or laboratory changes become unavailable. [1]

A specific subset, with selective collection for everyone else

Section 3.2 collects comprehensive safety data "for specific subset(s) of the patient population where additional information is deemed important" and reduces collection "for other patients." [1]

The worked example is age. If earlier trials included few patients over 65, "comprehensive safety data would be collected for patients aged 65 and over, whereas data collection would be reduced for patients under age 65." The same section names other factors that can define the comprehensive subset: race, ethnicity, sex, baseline disease status, renal or hepatic impairment, CYP metabolizer status, genetics, and geographic location. [1] Each factor is a named subset. A geographic subset under this section is a pre-specified patient factor. It is a different design from a country list added because one authority asked for one laboratory, and from the random site sample in section 3.4.

In this example, reductions concern the complement's section 2.5 schedule; the general section 2.4 collection framework remains in place.

Routine serum chemistries show the split. Participants aged 65 and over have the panel on the comprehensive schedule. Participants under 65 have it limited or omitted. Section 4 then keeps two chemistry summaries. One chemistry table for all randomised participants would put a collected value in the same rate as a value the other form did not collect on the same schedule. The comprehensive summary uses the age group that had the panel. The complement summary describes the reduced schedule, or it is absent, and it is labelled.

Where serious adverse events are collected consistently in both age groups over the same period, a broader summary may be appropriate under the statistical analysis plan. The age-based laboratory reduction alone supplies no reason to discard otherwise comparable SAE data. Conversely, a justified regional difference in the SAE collection itself would require separate treatment of that difference. Baseline data remain governed by the objectives for both groups.

An initial comprehensive period, then selective collection

Section 3.3 collects comprehensive data "through some pre-determined period, with selective safety data collection thereafter." A planned reduction in monitoring frequency "is a variation of selective safety data collection," on the premise that early data will characterise non-serious events and important changes in vitals and laboratory abnormalities. The section's illustration of going further is to discontinue interval examinations and laboratory investigations "once patients have been followed through 12 months." [1] Twelve months is the illustration.

The vaccine example sets three observation windows:

solicited signs and symptoms are recorded daily for a pre-specified period post-vaccination (e.g., 7 days), all unsolicited adverse events are collected for approximately 4 weeks post-vaccination, and serious adverse events and pre-specified AESIs are collected for at least 6 months post-vaccination. [1]

Each window is its own summary. An unsolicited-event frequency through month 6 would place the four-week collection window and the later weeks, when that collection had stopped, in one denominator. Section 4 keeps those observation windows apart. The serious-adverse-event summary covers the period the form collects them, at least 6 months in the example.

The section's other illustrations are the same kind of split: a cardiovascular-outcomes trial may move to selective collection from approval forward, and a long prevention trial, with dementia or end-stage kidney disease as the examples, may do so after a reasonable comprehensive period, "e.g., one year." [1] Windows stay apart only for classes whose schedule changes.

A newly emerging concern can require intensified monitoring or a return to comprehensive collection. Section 4 anticipates logistical effects on sites, protocols and forms. Preserve the effective date and affected modules of such a change so that the analysis can distinguish observations collected before and after it. This reactive change and a prespecified reduction under section 3.3 can both occur in the same trial. [1]

The October 2025 FDA/TransCelerate tabletop explored dynamic switches, including a return to fuller collection after a signal. Its public summary recommends advance planning and discussion with the FDA review team; the case-study document describes operational details that E19 leaves unspecified. Both are accounts of a hypothetical exercise, with explicit limits on their status as regulatory advice. [11] [12]

A representative subset

Section 3.4 applies where enrolment "may greatly exceed the number needed to assess non-serious adverse events, changes in vital signs, and laboratory abnormalities." Comprehensive data "could be collected for only a representative subset of participants, with selective safety data collection in other participants (see Section 2.5)." [1]

Two methods are named. Comprehensive data "could be collected at randomly selected sites, perhaps half of the sites in every country or geographic area, with reduced safety data collection at other sites." Or they "could be collected for randomly selected patients across all sites." "Perhaps half" is the illustration. The rule that follows is that the comprehensively monitored patients "are representative of the entire population," "selection bias should be avoided," diversity of that subset "needs to be ensured," and "the method for randomly selecting sites or patients should be described in the protocol." [1]

Outside the representative subset, section 2.5 collection is reduced. Laboratory frequencies should identify the comprehensively assessed population and distinguish planned omission from a missed scheduled assessment. A consistently collected SAE class can have a broader analysis population, subject to the protocol and analysis plan.

The 2016 guidance also mentions selecting larger population sites. E19's examples instead use randomly selected sites or participants and emphasize representativeness, diversity and avoidance of selection bias. Larger-site selection therefore needs its own justification and prospective discussion: site size alone establishes neither random selection nor a representative population. This distinction affects country, age and disease-severity coverage in the proposed sample. [9] [1]

The module sheet for the form and the summary

Use this matrix as an editorial workpaper derived from the cited sections. It maps collection classes to the examples; trial-specific protocol decisions and statistical definitions take precedence over its shorthand. [1] [9]

"Keep" means the class stays on the rows that example still collects. "May limit" means the example may thin or omit it for the people or the window in the pattern table. A broader summary may be appropriate where collection matches, subject to the analysis plan. "Split" means the comprehensive rows and the selective rows stay apart. The 2016 column is that guidance's sentence.

Denominators in the C3TI table cells

FDA's C3TI white paper reports trial-specific collection schedules and acknowledges sponsor input, including program details outside the public domain. It is a discussion paper, not guidance. The comparisons below reproduce the reported cells rather than independently auditing the full protocols. Omitted table information remains unknown. The paper also explains that systematic data on E19 uptake are unavailable, so these examples support operational comparisons only. [4]

If the printed cell is the collection rule, the summary group is the group that was asked. The subset percentage in Table 1 is printed on the liver-function row. The analysis column is our interpretation, conditional on the reported cell matching the protocol; it does not prescribe a statistical estimator.

The liver-function cell establishes the reported 20% collection subset. Assessing random selection or representativeness would require the protocol and sampling plan. Treat that property as unverified here.

For EMPACT-MI, the paper reports local amendments in eight of 22 participating countries, including requests for follow-up serum creatinine or all adverse events. This is a specific program's experience; it demonstrates why a country-dependent module may be needed. The EMPA-KIDNEY UK month-18 schedule is a separate trial example and supports no general claim about the MHRA's E19 position. [4]

The tabletop's heart-failure hypothetical proposed collecting only SUSARs, the serious, unexpected and related subset. The discussion outcome rejected that restriction and recorded a need to capture all SAEs for the case. This illustrates a consequential distinction for form design: unexpectedness and relatedness filters can exclude serious events that belong in the agreed collection scope. Keep the exercise's illustrative status distinct from the E19 recommendation itself. [12] [11]

Regulatory status and the limits of the comparison

The dated FDA and EMA records cited above establish the guidance versions used in this review. ICH defines implementation as completion of the regional process, corresponding to Step 5 and public availability of the final guideline. That definition describes status; study-specific agreement addresses the proposed use. [13]

Our source comparison covers the ICH text, its FDA and EMA versions, the FDA 2016 guidance, and the C3TI and tabletop examples. The trial schedules come from the discussion paper. Full protocol compliance, current country-by-country acceptance, and uptake across industry are outside this evidence base. The three harmonised versions count as one underlying recommendation, rather than three independent studies.

What the protocol and the analysis plan have to name

Section 4 says methods for aggregating the data "should be delineated within the protocol and associated analysis plans," and that results "should clearly identify data summaries where interpretation is impacted by the collection approach." [1]

Describe the proposed design using the relevant section 3 examples, including any justified combination. Specify each reduced module, its participant group, its observation period and the rationale. Age 65, a 12-month switch and sampling half of the sites are examples in the text, each requiring study-specific justification if adopted. Describe signal-driven reversion and prespecified switching separately. The tabletop discussion provides useful questions for that planning meeting. [1] [11] [12]

On every pattern the case report form keeps the triggered characterising page for a section 2.4 event, which is the page the monitoring plan and the training use. [1] Section 2.7 recommends prospective agreement on the characterisation and the method and, for one multi-regional protocol, agreement from every authority whose region will use the results. [1]

Questions to take to the review division before build

Bring any proposed reduction within section 2.4 to the review division with the applicable regional provisions and protocol justification. The objective is a decision about the particular data and population before the configuration is released. [1]

A proposed cut of a section 2.4 class is the first question. The cardiovascular-outcomes example says the endpoint serious events would always be collected. The heart-failure tabletop rejected a SUSAR-only rule for that hypothetical and recorded that all serious adverse events must be captured. [12] A form that collects only unexpected related serious events is narrower than section 2.4 item 1.

An oncology grade rule needs particular attention. FDA's 2016 guidance recommends retaining all Grade 3 and Grade 4 events and Grade 2 events affecting vital organs. The later C3TI paper describes a more selective approach in Pragmatica-Lung, including omission of some expected or non-serious high-grade events. The paper provides a study example, whereas the original protocol and regulatory agreement establish its exact application. Use the difference to formulate a study-specific consultation question before omitting such events in another trial. [9] [4]

For a larger-site sample, show the proposed population coverage, selection method and likely sources of selection bias. Discuss whether that design meets E19's representativeness objective rather than treating enrollment volume as a sufficient proxy. [1] [9]

From the collection matrix to a tested study build

The most useful handoff is one approved matrix shared by clinical, safety, data-management and statistical teams. For each module, record the applicable participants, country exceptions, start and stop rules, scheduled visits, triggered follow-up and intended analysis population. This is our implementation recommendation, derived from E19's expectation that the approach be described in the protocol, monitoring plan and analysis plan. [1]

Consider the country-specific creatinine example. Configure the follow-up laboratory for the countries and periods covered by the agreed amendment. Retain the local randomisation result separately from the follow-up series. A participant moving between collection regimes needs an effective date, so the export can distinguish an assessment omitted by design from a scheduled measurement missed in error. Preserve those distinctions through data cleaning and transfer.

That gives the team concrete acceptance tests. A participant in the comprehensively monitored subgroup should receive the scheduled laboratory form; a participant outside it should follow the reduced schedule. Both should receive the agreed event-reporting path. Crossing a prespecified time boundary should change only the named modules. An event requiring characterization should open the relevant follow-up collection even where routine collection has stopped. A country amendment should affect the authorized sites and effective period while preserving earlier records.

For analysis, the tested population is only a starting point. A scheduled assessment missed within that population remains missing, and the analysis plan must address it. Planned omission outside the population is a different condition. Treatment arm, exposure duration, repeat measurements and event definitions also affect the appropriate denominator. A count of participants with an abnormal result and a count of abnormal measurements answer different questions; the table shell should name which one it presents.

EClinCloud describes study-build and CRF-design support within its EDC offering. A useful implementation discussion starts with the agreed matrix and test cases, with scientific and regulatory decisions owned by the sponsor. For the adjacent safety-reporting workflow, see the UK SUSAR reporting analysis.

Sources

1. International Council for Harmonisation, ICH E19 guideline, A Selective Approach to Safety Data Collection in Specific Late-Stage Pre-approval or Post-Approval Clinical Trials, Step 4, adopted 27 September 2022. Accessed 4 October 2026. Sections 1.1, 1.3, 2.1, 2.2, 2.3, 2.4, 2.5, 2.6, 2.7, 2.8, 3, 3.1, 3.2, 3.3, 3.4, and 4.

2. U.S. Food and Drug Administration, E19 A Selective Approach to Safety Data Collection in Specific Late-Stage Pre-approval or Post-Approval Clinical Trials, Guidance for Industry, December 2022. Accessed 4 October 2026. Nonbinding opening paragraph; the harmonised operative text; section V.B reference to the February 2016 guidance.

3. European Medicines Agency, ICH guideline E19 on a selective approach to safety data collection in specific late-stage pre-approval or post-approval clinical trials, Step 5, EMA/782210/2022. Accessed 4 October 2026. Cover table: final adoption by CHMP 16 September 2022; date of coming into effect 16 March 2023. Section 4 pooling paragraph.

4. U.S. Food and Drug Administration, Center for Drug Evaluation and Research, Center for Clinical Trial Innovation, Selective Safety Data Collection, white paper. Accessed 4 October 2026. Disclaimer and acknowledgement; the paragraph stating that the extent of use since finalisation is not known; the serious-adverse-event footnote; Table 1, VICTORION-2-PREVENT; Table 2, EMPACT-MI and EMPA-KIDNEY; the EMPACT-MI country-amendment narrative; the Pragmatica-Lung description.

5. Food and Drug Administration, E19 A Selective Approach to Safety Data Collection in Specific Late-Stage Pre-Approval or Post-Approval Clinical Trials; International Council for Harmonisation; Guidance for Industry; Availability, 87 FR 74637, 6 December 2022, document 2022-26433. Accessed 4 October 2026. Publication date, citation, and abstract. The abstract says the final revises the June 2019 draft titled "E19 Optimisation of Safety Data Collection."

6. U.S. Food and Drug Administration, Guidance Recap Podcast: E19 A Selective Approach to Safety Data Collection in Specific Late-Stage Pre-approval or Post-Approval Clinical Trials, transcript. Accessed 4 October 2026. Content current as of 22 April 2024. Publication-date answer and the answer describing three implementation approaches.

7. European Medicines Agency, ICH guideline E19 on a selective approach to safety data collection in specific late-stage pre-approval or post-approval clinical trials. Accessed 4 October 2026. Current version effective from 16 March 2023; status Adopted; legal effective date 16 March 2023.

8. U.S. Food and Drug Administration, Selective Safety Data Collection (SSDC) Demonstration Project. Accessed 4 October 2026. Content current as of 23 July 2025. Eligibility criteria for demonstration-project proposals.

9. U.S. Food and Drug Administration, Determining the Extent of Safety Data Collection Needed in Late-Stage Premarket and Postapproval Clinical Investigations, Guidance for Industry, February 2016. Accessed 4 October 2026. Sections III.B, III.C, III.D, and IV. Nonbinding banner.

10. U.S. Food and Drug Administration, Determining the Extent of Safety Data Collection Needed in Late-Stage Premarket and Postapproval Clinical Investigations, guidance search page. Accessed 4 October 2026. Status label Final. Content current as of 24 August 2018.

11. TransCelerate BioPharma and U.S. Food and Drug Administration, Enabling the Implementation of Pragmatic Elements at Scale: Summary Report from an FDA and TransCelerate Tabletop Exercise on Safety Reporting, October 2025. Accessed 4 October 2026. Exercise date 30 October 2025. Disclaimer that the takeaways are not FDA guidance; dynamic-switch bullet; global-alignment bullet.

12. TransCelerate BioPharma, Hypothetical Case Studies from an FDA and TransCelerate Tabletop Exercise on Safety Reporting, October 2025. Accessed 4 October 2026. Disclaimer; heart-failure case, selective collection written as SUSARs only; discussion text that ICH E19 is not specific on an in-trial switch; Table 2 row "Capturing only SUSARs."

13. International Council for Harmonisation, ICH definitions for guideline implementation, version 1.2, 2 June 2024. Accessed 4 October 2026. Row "Implemented."