What Changed on October 1, and What Did Not

Before getting into data quality, it is worth being exact about the regulatory facts, because several of them are commonly misstated in vendor material and internal slide decks.

The Requirement Itself

On April 6, 2026, FDA published a notice in the Federal Register announcing that, starting October 1, 2026, postmarketing ICSRs for human drug products, biological products, and drug- or biologic-led combination products submitted through ESG NextGen must use the ICH E2B(R3) data standards1. The notice states that FDA would continue to accept E2B(R2) submissions through September 30, 2026, and that once a company has started submitting in R3, all of its ICSR submissions are expected to use R31. FDA’s electronic submissions page describes the April announcement as extending the previously recommended implementation period by six months2.

Three details in the notice matter for how you scope post-switch quality work:

  • The requirement applies to ESG NextGen submissions only. Companies without E2B(R3) capability can still submit ICSRs and attachments through the Safety Reporting Portal, a web form, and FDA’s page says no action is required for SRP submitters12. A company that is a gateway partner cannot use SRP for ICSRs2.
  • FDA names the structural differences it cares about. The notice lists new, changed, and expanded data elements; assessment of seriousness at the event level rather than the case level; and attachments embedded in the ICSR rather than sent separately1. Each of these is a place where converted or newly entered data can go wrong.
  • The purpose is data quality. FDA says the ICH implementation guide and the FDA regional implementation guide were issued to improve the quality of data in ICSR submissions and to allow better handling and analysis of ICSRs1. That is the standard your submissions will be read against.

Premarketing reports are on a different clock. FDA has required E2B(R3) for commercial IND safety reports submitted through ESG NextGen since April 1, 20262. So for many sponsors, October was the second R3 cutover of the year, not the first.

FAERS and AEMS: A Consolidation, Not Only a New Name

It is common to hear that FAERS was renamed AEMS on October 1. That is not quite what happened. FDA launched the FDA Adverse Event Monitoring System on March 11, 2026, as a single platform to replace several legacy systems, including FAERS, VAERS (for display of vaccine reports), and the animal drug and food systems, with the device, human food, and tobacco systems scheduled to follow in May 20264. FDA’s pages now label AEMS as “formerly” FAERS2, and the April notice refers to the AEMS database the same way1. The R3 requirement and the AEMS launch are related but separate events, about seven months apart.

The AEMS change that matters most for submission quality is timing. FDA’s announcement says AEMS publishes reports in real time rather than quarterly4. A coding error, a wrong seriousness flag, or a duplicate that once had weeks to be corrected before public release can now be visible on the public dashboard much sooner. FDA’s own AEMS page warns that duplicate and incomplete reports exist in the system and that the data do not establish causation5. Those caveats protect FDA. They do not protect a company whose reports are the duplicates.

One practical note: the FDA regional implementation guide that the April notice incorporates by reference is dated August 2024 and still uses FAERS names throughout, including in its routing identifiers and acknowledgment descriptions6. Nothing in the notice tells submitters to change gateway routing because of the AEMS name. Treat any internal proposal to reconfigure routing for the rename as a change that needs evidence from FDA first.

Common statementWhat FDA’s primary sources say
“All postmarketing safety reports must be E2B(R3) from October 1.”ICSRs submitted through ESG NextGen must be R3 from October 1, 2026. The Safety Reporting Portal remains an option for companies without R3 capability1.
“FAERS was renamed AEMS on October 1.”AEMS launched March 11, 2026, consolidating FAERS and other legacy systems. FDA pages describe AEMS as formerly FAERS24.
“IND safety reports follow the same date.”Commercial IND safety reports through ESG NextGen have required R3 since April 1, 20262.
“R3 is a format change only.”FDA lists event-level seriousness, embedded attachments, and new and expanded data elements as differences, and says the standards were issued to improve data quality1.
“The R3 specification is frozen.”FDA now accepts new values for data element C.5.4 (patient support program, market research program, and organized data collection from digital platforms) following ICH E2D(R1)29.

Why Post-Switch Quality Is Its Own Problem

Most safety organizations prepared for October as a system project: configure the safety database, map the fields, test with FDA’s validator, get a first accepted acknowledgment, and declare success. FDA provides an E2B(R3) validator tool that lets submitters check XML files in a test environment3, and most teams used it. But a validator checks conformance to rules. It does not check whether the value in a field is true, whether two submissions describe the same patient, or whether the narrative agrees with the coded events.

Intake quality, which we covered in an earlier article on case intake failure points, is about whether the case in your database matches the source. Submission quality is about whether the message FDA receives matches the case in your database, and whether the record of what was sent, accepted, rejected, and resent is complete. After a format switch, these two can drift apart even when intake is excellent. The rest of this article is about that second gap.

Oct 1, 2026Date from which ESG NextGen postmarketing ICSRs must use E2B(R3)1
Real timeAEMS publishes adverse event reports in real time rather than quarterly4
~5 millionFAERS reports, of about 29 million, identified as duplicates by FDA’s deduplication pipeline14

Field Mapping From R2 to R3: Where Conversion Changes Meaning

The ICH document that governs conversion between the two formats is Appendix I(B), Backwards and Forwards Compatibility. Its starting point is blunt: because of how the guideline evolved, ICSRs “cannot be perfectly converted from one standard version to the other (either backwards or forwards),” and it adds that “Repeated conversion could result in transformation or loss of information.”8 FDA’s regional guide adds its own forward compatibility rules for the FDA regional data elements and points submitters to a separate spreadsheet of those rules6.

Two kinds of data are exposed. The first is legacy cases converted from R2 when the safety database was upgraded. The second is follow-up reports on old cases, where an R2-era initial now has an R3-era follow-up, and the two must tell a consistent story. Below are the mapping areas that most often produce data that passes validation and still means something different from what the safety physician intended.

Seriousness Moved From the Case to the Event

In R2, seriousness and its six criteria were recorded once, at the report level. In R3 they are recorded for each reaction or event (data element E.i.3.2). The compatibility guide sets the upgrade rule: copy the report-level criteria to each event that the reporter flagged as serious, and set the criteria to the null flavor NI for each event flagged as not serious. If the R2 field for the highlighted term was not provided, the report-level criteria are copied to every event8.

That last rule is where meaning changes. A legacy case with a serious hospitalization and a non-serious rash, where the per-event flag was never filled in, comes out of conversion with both events carrying the hospitalization criterion. The XML is valid. The data now say the rash caused hospitalization. If that case is later followed up and resubmitted, the error travels with it into AEMS and into anyone’s signal analysis.

Check this first: Run a query across converted legacy cases for any case with more than one event where every event carries identical seriousness criteria. Not all of these are wrong, but this is the population where the “copy to every event” rule was applied, and it is where a medical reviewer should look before the next follow-up goes out.

The Drug-Event Matrix Was Rebuilt

R2 stored the time interval, recurrence on readministration, and relatedness assessments in separate places. R3 combines them into one drug-reaction matrix (G.k.9.i), with an entry for each drug and event pair8. Conversion builds that matrix from the old fields. When the R2 case had relatedness recorded for some drug and event pairs and not others, or recorded the time interval only once, the R3 matrix can show a combination that no assessor ever made. Causality is the part of a case that reviewers read most closely, so this is worth a targeted sample review on converted cases with multiple suspect drugs.

Case Identifiers Changed Structure

R2 had two fields for a case number assigned by someone else: the regulatory authority’s case report number and another sender’s case number. R3 replaces both with a single worldwide unique case identification number (C.1.8.1) and a separate field saying whether the first sender was a regulator or another party (C.1.8.2). When both R2 fields were filled in, the conversion rule keeps only the regulator’s number8. The other number does not disappear from the world. It disappears from the identifier field, which is exactly the field duplicate detection relies on. We return to this below.

Fields That Have Nowhere to Go

Some R3 fields have no R2 counterpart at all, including medical confirmation by a health professional (E.i.8), family history, the primary source telephone number, and the MedDRA code for a test name8. For converted cases these are simply empty. That is correct, but it means a converted case and a newly entered case will look different in completeness metrics even when the underlying information was the same. If your monitoring compares completeness across case cohorts, split converted and native R3 cases or the comparison will mislead you.

Field Lengths and the Partner Problem

The R2 case narrative was limited to 20,000 characters. The R3 narrative field (H.1) allows 100,00078. Several other free-text fields also grew. Going up is easy. Going down is not, and that matters for anyone who still exchanges cases with a licensing partner, distributor, or service provider whose system runs on R2. The compatibility guide describes a retransmission use case in which a receiver converts R3 to R2 and back again, and recommends that systems support the narrative with no size limit so that fields that cannot be converted directly can be carried in the narrative8. In practice, every round trip through an R2 partner is a chance for a case to come back shorter, flatter, or with structured data moved into free text.

Mapping areaWhat changed in R3Typical post-switch defectCheck to run
SeriousnessRecorded per event (E.i.3.2), not per case8Case-level criteria copied to every event, including non-serious onesMulti-event converted cases with identical criteria on all events
Drug-event matrixRelatedness, recurrence, and time interval combined per drug-event pair (G.k.9.i)8Assessments attached to pairs that were never assessedSample of converted cases with two or more suspect drugs
Case identifiersSingle worldwide unique ID (C.1.8.1) plus first-sender type8Second external case number dropped from the identifier fieldConverted cases where both R2 numbers existed; confirm the other is in C.1.9.1
New fieldsElements such as E.i.8 and family history have no R2 source8Completeness metrics skewed between converted and native casesReport completeness separately for each cohort
Field lengthsNarrative grows from 20,000 to 100,000 characters78Truncation or flattening on round trips through R2 partnersCompare narrative length and structured fields before and after partner exchange

Null Flavors: Saying “We Do Not Know” in a Way a Machine Can Read

R3 uses HL7 null flavor codes to explain why a required or optional field has no value. The ICH implementation guide defines them: NI (no information, the most general and the default), MSK (masked, information exists but is withheld for privacy or security), UNK (unknown, a value applies but is not known), NA (not applicable), ASKU (asked but unknown), NASK (not asked), and two codes for open-ended numeric ranges used only for test results7. FDA’s regional guide says null flavors are used to explain the reason for missing data on required elements and must be used for specific required elements when the value is blank6.

Under R2, a blank field was just blank, or it said “UNKNOWN” or “PRIVACY” in text. The compatibility guide maps those R2 terms to UNK and MSK on the way up, and maps UNK, ASKU, and NASK all back to “UNKNOWN” on the way down8. That is the first sign of the problem: R3 can tell the difference between “we asked and the reporter did not know” and “we never asked,” and R2 cannot. A converted case cannot recover a distinction that was never captured.

Where FDA Draws Hard Lines

The FDA regional guide sets specific rules that produce rejections, and they are worth knowing by heart:

  • Expedited criteria. FDA does not accept NI for data element C.1.7, “Does this Case Fulfill the Local Criteria for an Expedited Report?”, in initial submissions. Initial submissions with NI in this field will be rejected. For follow-ups on cases FDA already holds, the value defaults to what was in the initial6. That second sentence deserves attention: if the initial carried the wrong value, a follow-up does not fix it by default.
  • Patient characteristics. When age, date of birth, or sex are unknown and the patient is not the reporter, FDA allows NI or ASKU. Where foreign privacy rules prevent sharing patient information, FDA supports MSK. Where no patient is involved, such as some medication error reports, FDA says to use NA6.
  • Race and ethnicity. FDA’s regional elements for race and ethnicity should use NI when the information is unknown or not available6.

The Real Risk: Null Flavors as a Default Setting

The rejections are easy to find because FDA tells you about them. The harder problem is the null flavor that passes. Safety database configurations often set a default null flavor on required fields so that a case can be saved and submitted on time. If the default is NI and case processors are not trained to replace it, the system will submit thousands of reports that say “no information” when the true state is “asked but unknown” or “not asked.” Both are honest answers that tell a reviewer something. NI tells them nothing.

This matters beyond tidiness. The difference between ASKU and NASK is a direct measure of follow-up quality. A high NASK rate on patient age in serious cases means your follow-up process is not asking. A high ASKU rate means it is asking and not getting answers. A high NI rate means you cannot tell which, and neither can an inspector.

A Null Flavor Policy in Four Rules

  • Document, per required data element, which null flavors are allowed and what each one means in your process.
  • Do not let the safety database default to NI on fields where ASKU, NASK, or MSK would be accurate. Make the processor choose.
  • Use MSK only where a privacy rule applies, and record which rule. MSK tells the receiver that information exists.
  • Report null flavor rates by element and by case source each month. A sudden rise in one element usually points to a form change or a configuration change, not to reporters.

MedDRA Versioning at the Message Level

Our earlier article on reference data governance covered how MedDRA version drift affects analysis across time. Submission quality has a narrower and more mechanical MedDRA problem: the version recorded inside each message.

One Version per ICSR

The ICH implementation guide requires the MedDRA version to be identified each time a MedDRA term is populated, and says the same MedDRA version should be used for all MedDRA-coded elements within a single ICSR. Different ICSRs in the same batch can use different versions7. FDA’s regional guide states the same rule6.

That rule is simple until a follow-up arrives. The initial case was coded in the previous version. The follow-up adds a new event, coded by the processor in the current version. Unless the system recodes the earlier terms or the processor codes the new event in the older version, the outgoing message carries two versions. Some safety databases handle this automatically when the version upgrade runs. Others only recode on save, which means a case nobody has opened since the upgrade will go out with a mix the first time someone adds a follow-up. The fields that most often get missed are not the reaction terms but the secondary ones: indication, medical history, test names, and the sender’s diagnosis, each of which carries its own version field in R37.

The Switchover Date

The EU ICSR implementation guide sets out the convention used across ICH regions: a new MedDRA release becomes the reporting version on the first Monday of the second month after it is released, with the MSSO recommending a switchover at midnight GMT between Sunday and Monday10. It also warns that a sender who moves to the new version before the receiving party has switched should expect that “the transmitted ICSRs will be rejected.”10

For post-switch quality, the switchover creates a predictable spike. Cases that were in process over the weekend of the change, partner exchanges in flight, and literature cases coded in bulk the week before can all go out in the wrong version. A team that knows the date can watch for it. A team that does not will find out from the rejection rate.

Practical step: Put the MedDRA switchover dates on the same calendar as your submission deadlines, and add a rule to your acknowledgment monitoring that flags any rejection referencing a MedDRA version in the two weeks either side of the change. These rejections are easy to fix, but each one delays the report it belongs to.

Duplicates, Follow-Ups, and Nullifications Under R3 Identifiers

Our September article on duplicate detection rules covered how to match records without flooding reviewers. This section is about something different: the duplicates that the submission process itself creates, and the identifiers that let a receiver tell a follow-up from a new case.

The Identifiers That Must Not Change

R3 separates two identifiers. The sender’s safety report unique identifier (C.1.1) is your report number. The worldwide unique case identification number (C.1.8.1) identifies the case across every organization that has handled it. The ICH guide says that for an amendment, the sender should use the same C.1.1 and C.1.8 values previously submitted, and that if a nullified report later needs to be submitted again, it should go out with new identifiers7. The guide also tells receivers to accept the worldwide identifier as is and not apply business validation rules to it, because the elements it is built from, such as a country code, can change over time7.

The consequence is that a receiver has very little protection against a sender that changes the worldwide identifier by mistake. If a case is re-created in a migration, merged and split in the safety database, or re-keyed by a service provider, and the C.1.8.1 value changes, AEMS will see a new case. No validation rule catches it. Only your own reconciliation does.

Linked Cases and Other Identifiers

GVP Module VI Addendum I, the EU guideline on duplicate management, points to two R3 sections that help receivers. Filling the “other case identifiers” section (C.1.9.1) with every reference number by which the case is known helps receivers detect and confirm duplicates. Filling the “linked reports” section (C.1.10.r) with cases that share a common element but are distinct helps receivers confirm that cases are not duplicates of each other11. The same guideline says that when duplicates are merged, the worldwide identifier of the case chosen as the master should be kept, and the subordinate case numbers should not be reused but recorded under other case identifiers11.

After the switch, check both sections on a sample of cases. Converted cases are especially likely to have C.1.9.1 incomplete, for the reason described in the mapping section: R2 allowed two external case numbers in identifier fields, and conversion kept only one in C.1.8.18.

Resubmission Is Where New Duplicates Come From

FDA’s regional guide gives exact instructions for failed submissions. If the second acknowledgment does not arrive within 24 hours of the first, resubmit the original without changing the batch identifier. If the acknowledgment reports a failure for a single ICSR, resubmit the corrected ICSR with a new unique batch identifier. If some ICSRs in a batch failed, separate them, correct them, and resubmit only the failed ones in a new batch. FDA’s example is a batch of 50 where 15 failed: correct and resubmit those 15 only, and do not include any that processed successfully6.

Every one of those instructions exists because the opposite behavior creates duplicates or confusion. Re-sending a whole batch because part of it failed submits the successful reports twice. Changing the batch identifier on a resend after a missing acknowledgment makes it look like a different transmission. In the first weeks after a switch, when rejection rates are higher and teams are under pressure to clear the queue, these are exactly the shortcuts people take.

Why This Matters More Now

FDA has published work on how large the duplicate problem in its own data is. A 2025 study by FDA and Johns Hopkins researchers describes a deduplication pipeline applied to all 29 million FAERS reports, which identified about 5 million reports as duplicates and now runs on historic and daily incoming reports at FDA14. An earlier FDA study found that a probabilistic matching algorithm identified 77 percent of known duplicate pairs in VAERS but only 13 percent in FAERS, at 100 percent precision for FAERS12. A study using VigiBase found suspected duplicate rates of about 2.5 percent overall and higher rates for literature reports and reports with a fatal outcome13.

The point for a sponsor is not the exact rate. It is that regulators deduplicate after the fact using the same fields you control: identifiers, patient details, dates, products, and events. Clean identifiers and complete C.1.9.1 and C.1.10.r sections make your reports easier to group correctly. Inconsistent ones make your product look like it has more cases than it does, or make two different patients look like one.

Nullifications and Amendments

R3 provides a specific data element (C.1.11) for nullifying a report that should be considered void, or for amending one after internal review, for example when event terms, seriousness, or causality have been corrected7. Use the amendment flag for corrections and the nullification flag only when the whole case is void. A common post-switch mistake is to nullify and resubmit a corrected case as new, which, under the ICH rule, requires new identifiers and breaks the case history that receivers rely on7. The EU duplicate guideline also has a section on sending nullifications when duplicates are merged11, which is worth reading even for US-focused teams because it shows the reasoning a regulator expects.

When the Narrative and the Structured Fields Disagree

Every ICSR must include a narrative (H.1), which the ICH guide describes as a focused, factual, and clear description of the case, including the reporter’s own words or short phrases7. The structured fields carry the same information in coded form. Signal detection uses the structured fields. Medical reviewers read the narrative. When they disagree, the two audiences come to different conclusions from the same report.

What R3 Changes About This

The EU ICSR implementation guide gives a direct explanation of why European regulators made R3 mandatory in EudraVigilance from June 30, 2022. When R2 reports were converted to the ISO ICSR format on receipt, “important information is often not available in dedicated data elements and may only be found in case narratives.” The guide continues: “This reduces data quality as well as search and data analysis capabilities.”10 R3 has more structured fields so that information can move out of the narrative and into places a query can find it.

That improvement only happens if case processing uses the new fields. If processors continue to type test results, medical history, and dechallenge outcomes into the narrative and leave the structured fields empty or set to a null flavor, the R3 report contains the same information in the same place as the R2 report did, and the promised gain in data quality does not arrive.

The EU guide also says case narratives should not be provided as file attachments, and notes that the 100,000 character narrative field, together with the repeatable case summary element, means the narrative has no practical limit10. FDA’s regional guide requires attachments to be embedded in the message using base 64 encoding, sets naming rules for the attachment file name, and rejects files whose media type does not match the extension6. A narrative pasted into a PDF attachment instead of H.1 is invisible to anyone searching structured data and may be rejected on a naming technicality.

The Mismatches That Matter

Not every difference between narrative and structured data is a defect. The narrative is supposed to include context that has no field. The mismatches worth monitoring are the ones where a fact appears in both places with different values:

  • Outcome. The narrative says the patient died; the outcome field for the event says recovering or unknown.
  • Seriousness. The narrative describes a hospitalization; no event carries the hospitalization criterion. After conversion, the reverse also happens, as described in the mapping section.
  • Dates. Onset and stop dates in the narrative do not match the structured event dates, which changes time-to-onset analysis.
  • Suspect products. The narrative names a concomitant drug as the reporter’s suspect, but the drug role field says concomitant.
  • Events. The narrative describes an event that was never coded, which means it will never appear in a MedDRA-based query.

The literature supports paying attention to structured completeness. The Uppsala Monitoring Centre’s vigiGrade completeness score measures how much clinically relevant information a report carries in structured format, such as time to onset, age, sex, indication, outcome, and dose. In its original study, only about 13 percent of reports in VigiBase scored as well documented, and the median completeness was 0.4115. FDA’s 2017 duplicate study found that narrative text did not improve automated duplicate classification, but described it as essential for the manual review that follows12. Both findings point the same way: structured fields drive automated analysis, and narratives drive human judgment, so the two need to agree.

Automated check

Rule-Based Consistency Tests

Flag cases where the narrative contains death, hospitalization, or life-threatening language and no event carries the matching criterion or outcome. Simple keyword rules catch the most serious mismatches.

Sample review

Monthly Medical Review Sample

A safety physician reviews a small random sample of submitted ICSRs side by side: narrative, coded events, seriousness, outcome, and causality. Record each mismatch type, not only a pass or fail.

Process fix

Structured First, Narrative Second

Train processors to enter data in structured fields before writing the narrative, and to write the narrative from the structured data plus context. This reverses the R2-era habit.

Configuration

Attachment Rules in the Database

Block narratives from being sent as attachments and validate attachment file names and media types before transmission, not after a rejection.

Rejected Submissions and Acknowledgments: Closing the Loop

An ICSR is not reported because it was sent. It is reported when the receiver accepts it. The acknowledgment messages are the only evidence of that, and in many organizations they are the least governed records in the safety system.

The Acknowledgments FDA Sends

FDA’s regional guide describes two acknowledgments. The first (ACK1) is the gateway’s message delivery notification. It confirms receipt, gives the official FDA receipt date, and in FDA’s words “does not imply FDA acceptance of the submission.” The second (ACK2) reports whether the system processed the submission and accepted or rejected all or part of it, with codes for any validation errors6. The regional guide also says that some business rules cause rejection and others produce a warning but still result in acceptance, and that the list of each is in FDA’s core and regional data elements and business rules document6.

The ICH guide defines the codes inside the acknowledgment message. At the batch level, AA means the message was processed with no further action needed; AE means an error was found and some ICSRs need action; AR means a parsing error, no data were extracted, and the whole transmission must be resent7. The EU guide describes the ICSR-level codes used by EudraVigilance: CA (commit accept, loaded successfully) and CR (commit reject, a fatal error prevented loading and the ICSR needs to be corrected and resent)10. The EU guide also notes that EudraVigilance extends the error and warning comment field beyond the ICH length so it can give senders more detail10.

SignalWhere it comes fromWhat it tells youWhat to do
ACK1 receivedFDA ESG6Received; official receipt date set; not yet acceptedRecord receipt date against the case; start the 24-hour clock for ACK2
No ACK1 within 24 hoursFDA regional guide6Possible gateway problemCheck ESG status page before resending
No ACK2 within 24 hours of ACK1FDA regional guide6Processing not confirmedResend the original with the same batch identifier
ACK2 rejectionFDA regional guide6One or more ICSRs failed validationCorrect and resend only failed ICSRs with a new batch identifier
Accepted with warningFDA business rules6Loaded, but a rule was not fully metTrend by rule; treat repeats as a data quality defect
AR at batch levelICH implementation guide7Parsing error, nothing extractedFix the message structure and resend the whole transmission

Warnings Are the Early Signal

Most teams track rejections because a rejection stops a report. Fewer track warnings, because a warning does not. But a warning means the report was accepted with a business rule not fully met, and the same configuration or process gap that produces a warning today is often one small change away from producing a rejection. Warnings are also the closest thing you have to free feedback from the regulator’s own rule engine on your data. Trend them by rule and by case source. A warning that appears on every literature case, for example, is a specific fix to a specific intake path.

The Clock Does Not Stop for a Rejection

Expedited reporting timelines are measured from when the company received the information, not from when a corrected file was accepted. FDA’s regional guide ties the official receipt date to ACK16, and a rejected submission still needs a corrected resubmission. Every rejection therefore eats into the reporting window for that case. In the weeks after a format switch, a rejection rate that would be a minor nuisance on routine reports can push expedited reports late. Late-report analysis should include a field for whether a rejection contributed, so that the root cause is recorded as a submission quality defect rather than as a case processing delay.

Acknowledgments Are Records

Acknowledgments are the evidence that each report was received and accepted. They belong in the safety system’s records, linked to the case and the specific transmission, and they should be retained and retrievable the same way the ICSR is. The EU guide tells senders to monitor for missing acknowledgments after two business days and act on them10. FDA’s 24-hour rules are tighter6. Either way, the question an inspector asks is simple: for this case, show me what you sent, when, what came back, and what you did about it. If answering that requires someone to search an email inbox or a gateway log by hand, the process is not controlled.

What good looks like: Every outbound ICSR has a transmission record with its batch identifier, the C.1.1 and C.1.8.1 values, the ACK1 timestamp and receipt date, the ACK2 result, any error or warning text, and a link to any resubmission. A daily report lists every ICSR without a final acceptance, sorted by reporting deadline.

Monitoring Submission Quality: A Working Scorecard

None of the problems above require new technology to find. They require a short list of measures, an owner for each, and a regular review that looks at the transmission layer as its own thing. The scorecard below is a starting point. The measures are deliberately simple so they can be produced from the safety database and gateway records without a new project. Set your own thresholds from your first month of post-switch data, rather than borrowing numbers from someone else’s operation.

MeasureWhat it showsSource
First-pass acceptance rateShare of ICSRs accepted on first transmission, by report type and case sourceACK2 results
Warnings by business ruleWhich rules your data most often fail without rejectionACK2 warning text, mapped to the FDA business rules document6
Open transmissionsICSRs without final acceptance, by days to deadlineTransmission records
Rejection-related late reportsExpedited reports that went late with a rejection in their historyLate-report log
Resubmission integrityResends that included already accepted ICSRs, or changed batch identifiers incorrectlyBatch history compared with the FDA rules6
Identifier stabilityCases whose C.1.8.1 changed between versionsCase version history
Null flavor profileNI, ASKU, NASK, MSK, and UNK rates by element and sourceOutbound XML
MedDRA version consistencyICSRs with more than one version, or the wrong version near a switchoverOutbound XML
Narrative and structured agreementMismatch types found in the monthly medical sample and by rule-based checksReview log
Converted case defectsFindings from targeted checks on legacy cases converted from R2Targeted queries

A 90-Day Post-Switch Review

For a company that completed its switch on or shortly before October 1, the first quarter after go-live is the time to look for defects while the team still remembers the configuration decisions. The plan below is one way to sequence that work.

1

Rebuild the Acknowledgment Record (Weeks 1 to 3)

Confirm that every ICSR sent since the switch has a linked ACK1 and ACK2 in the safety system. List any without final acceptance. Check every resubmission against FDA’s batch identifier rules. This is the foundation for everything else.

2

Profile the Outbound XML (Weeks 3 to 6)

Extract null flavor rates, MedDRA version counts per ICSR, attachment counts and types, and narrative lengths from the submitted messages themselves, not from the database screens. The XML is what FDA received.

3

Test the Converted Cases (Weeks 4 to 8)

Run the targeted queries from the mapping section: identical seriousness across events, multi-suspect drug matrices, dropped external case numbers. Have a medical reviewer confirm a sample from each population and decide whether amendments are needed.

4

Start the Medical Consistency Sample (Weeks 6 to 10)

Begin a monthly side-by-side review of narratives and structured fields. Record mismatch types. Add simple rule-based checks for the most serious mismatches, such as fatal outcomes not reflected in the coded data.

5

Set Ownership and Thresholds (Weeks 10 to 13)

Assign an owner for each scorecard measure. Set thresholds from your own first quarter of data. Agree what triggers a deviation or CAPA (corrective and preventive action) and what is handled as routine correction. Add the scorecard to the regular safety governance review.

Ownership: The Part That Usually Gets Missed

Submission quality falls between teams. Case processing owns the case. IT or the vendor owns the gateway and the safety database configuration. Regulatory operations may own the relationship with FDA. Partners and service providers own their own transmissions. When a warning appears on every literature case, it is not obvious whose problem it is, so it stays.

The fix is to name one owner for the transmission layer as a whole, usually within safety operations, with clear inputs from the others. That owner reads the scorecard, assigns the fixes, and reports to safety governance. Where a service provider sends on your behalf, the contract and the safety agreement should require the provider to give you the acknowledgment records and the data to produce the scorecard. A provider that can tell you how many reports it sent but not how many were accepted on first pass is giving you half the picture.

Do Not Forget the Next Change

The R3 specification continues to change. FDA now accepts the new C.5.4 values for patient support programs, market research programs, and organized data collection from digital platforms, following ICH’s December 2025 information paper on alignment with E2D(R1)29. The information paper explains that the new values let companies and regulators analyze reports from these solicited sources separately9. That only works if your case processors classify the source correctly and your database maps the value correctly. It is a small change, but it is a good first test of whether the post-switch monitoring above catches a new defect before it becomes a pattern. FDA’s posted copy of the ICH implementation guide now carries an October 2026 revision date7, so confirm that your configuration was built against the current documents.

Conclusion

The October 1 deadline was a format requirement, and most companies met it. The quality of what goes out under the new format is a separate question that the deadline did not answer. R3 gives regulators more structure, finer distinctions about missing data, event-level seriousness, and cleaner case identifiers. Each of those is an improvement only if the data in those fields are right, and each one is also a new place where a conversion rule, a configuration default, or a hurried resubmission can make the data wrong while the validator stays green. With AEMS publishing in real time, those errors are visible sooner than they used to be.

The work is not complicated. Treat acknowledgments as records. Profile the XML you send, not just the screens you see. Check converted cases where the mapping rules are known to change meaning. Make null flavors a choice, not a default. Review narratives against structured fields on a schedule. And give the transmission layer one owner. Sakara Digital works with pharma and biotech safety and quality teams on exactly this kind of post-implementation data quality review. If you have completed your E2B(R3) switch and want an independent view of what your submissions look like from the receiving end, we are happy to have that conversation.

For Further Reading