Patient Preferences and Postapproval Safety: Keeping the Evidence Distinct

TL;DR

Patient preference research and postapproval safety research answer different questions. The former examines what patients value and which trade-offs they find acceptable; the latter evaluates adverse experiences, safety signals and risks in actual use. FDA's 2026 publications on ICH E22, ICH E2D(R1), ICH M14 and pregnancy safety make those distinctions operationally important. E22 remains a draft; the FDA E2D(R1), M14 and pregnancy-safety documents discussed here are final guidances. The separate CDRH/CBER patient-preference guidance concerns voluntary information across the medical-device product life cycle. Their scopes should not be collapsed into a single mandatory software blueprint. Guidance recommendations are nonbinding unless a separate statutory or regulatory requirement is cited. [1] [3] [5] [7] [8]

For sponsors and CROs, the common problem is preserving meaning as information moves between systems. A preference survey response is not an adverse-event report. An observation in an existing claims dataset is not automatically a new individual case safety report. A registry enrollment date does not necessarily show that pregnancy outcome was unknown at enrollment. A useful eClinical architecture keeps these distinctions visible while supporting traceability, participant protection, analysis and reporting. The implementation suggestions in this article are our synthesis; they are not claims that FDA endorses a particular EClinCloud feature or data schema.

Patient preferences: FDA published the ICH E22 draft in February 2026 and the final CDRH/CBER voluntary patient-preference guidance in March 2026. They discuss methods, representativeness, instrument development, analysis and interpretation. E22 recommends pretesting and piloting, considers quality checks, and points to E6(R3) and CDISC ODM as helpful resources. It does not require every study to implement one specific electronic survey format. Preference evidence complements clinical efficacy and safety evidence; it does not replace them. [1] [3] [9]

Postapproval case safety: FDA's March 2026 E2D(R1) guidance distinguishes spontaneous and solicited reports and addresses digital platforms, organized data collection systems (ODCS), patient support programs, market research, and noninterventional studies. Classification depends on how information was obtained and the activity's context. The minimum criteria, causality assessment and reporting timelines must be considered with regional and local requirements. The same digital channel can carry different types of information; a channel label alone is insufficient. [5] [6]

Real-world safety studies: FDA's March 2026 M14 guidance addresses planning, design, analysis and reporting of noninterventional studies using real-world data for medicine safety assessment. It emphasizes a fit-for-purpose question, feasible design, relevant and reliable data, and evaluation of bias. Its examples do not establish a universal two-diagnosis-code rule, fixed algorithm, or mandatory software implementation. Quantitative bias analysis can help assess sensitivity to systematic error; it does not guarantee removal of bias. [7]

Pregnancy safety: FDA's May 2026 final guidance discusses pharmacovigilance, pregnancy registries and complementary studies. Each contributes different evidence and has different limitations. The guidance cites approximately 3.6 million annual U.S. births and an estimate that 41.6% of U.S. pregnancies in 2019 were unintended. Those figures establish the importance of obtaining useful exposure information; they are not estimates of the risk of any particular medicine. A registry and an electronic-health-data study should be designed to complement one another, with attention to recruitment, outcome ascertainment, linkage and selection bias. [8]

The local NIH RePORTER corpus provides descriptive background: 4,876 deduplicated application IDs across 31 eClinical queries, captured August 23, 2026. Keyword matches include 430 for RWD/EHR/registry terms, 407 for eCOA/ePRO terms, 387 for safety-surveillance terms, 152 for pregnancy-related terms, 86 for broad preference/shared-decision terms, and 11 for the narrower stated-preference screen. The resulting percentages describe this query-selected corpus, not commercial adoption, grant spending or platform functionality. We do not infer an infrastructure deficit from an abstract's failure to mention a keyword. [13]

---

1. Descriptive Research Context

The NIH analysis uses unique application IDs as its unit, not unique investigators, organizations, products or clinical trials. The 31 queries yielded 5,137 raw query hits before deduplication to 4,876 IDs. The CDISC material is a separate July 31, 2026 index of 32 standards pages and 54 linked resources: 25 public downloads and 29 resources excluded because authentication was required. That coverage is useful for discovery but cannot support claims about the complete content or absence of a standard. [13]

1.1 Keyword Matches in the Selected Corpus

The lexical categories overlap and use different levels of specificity. A broad term such as registry is much more likely to appear than the name of a particular preference method. The following counts are therefore descriptive search results, not a ranked assessment of methodological maturity. [13]

1.2 The Standards Index Is a Discovery Tool

E22's footnote 6 points to ICH E6(R3) and CDISC ODM v2.0 as helpful data-capture resources. This is a recommendation to consider relevant guidance, not a requirement to implement ODM or cite it inside every survey. The limited local CDISC index does not prove that no suitable template or implementation guide exists. For an actual study, the team should inspect the applicable specification, its version, and the intended submission requirements. The conceptual records proposed later in this article are deliberately not presented as normative CDISC variables or a replacement for those specifications. [1] [9] [13]

---

2. Patient Preference Studies

2.1 Scope and Perspective

Endorsed by the ICH Assembly at Step 2 on November 19, 2025 and published by FDA on February 6, 2026 (91 FR 5484, Docket No. FDA-2026-D-0207), ICH E22 General Considerations for Patient Preference Studies proposes a harmonized framework for designing, conducting, analyzing, and submitting Patient Preference Studies (PPS) in drug and biologic development and postmarketing evaluation [1]. Complementing ICH E22 on the medical device and device-led combination product side is FDA CDRH and CBER's March 30, 2026 final guidance Incorporating Voluntary Patient Preference Information over the Total Product Life Cycle (FDA-2015-D-1580, 91 FR 15622, Document Number GUI01500006, superseding the August 24, 2016 guidance ) [3], [12].

Four distinctions are especially useful when deciding whether preference evidence is appropriate. [1] [3]

Stated and revealed preferences are different approaches. E22 focuses on stated-preference methods; it excludes revealed-preference methods that infer preferences from observed behavior. A hypothetical choice task and a prescription-refill record do not answer the same question. The study should explain why the chosen approach is suitable for its intended use. [1]

Preference information is not a PRO score. A PRO captures an aspect of health status directly from the patient. A preference study investigates relative desirability or acceptability, including trade-offs among outcomes or treatment characteristics. Both can inform development, but combining them requires a justified analytical approach rather than treating their values as interchangeable. FDA's PFDD endpoint guidance provides related context for clinical outcome assessments. [1] [3] [4]

Perspective matters. Patient, caregiver and healthcare-professional preferences are distinct. A study should state whose perspective it measures and why that population is relevant. The ICH stakeholder webinar report illustrates how patient priorities can differ from assumptions held by development teams. Caregiver input may be relevant in appropriate contexts, but it should not be silently relabeled as the patient's own preference. [1] [2] [3]

Preference evidence complements other evidence. Acceptability of a risk does not establish that the treatment is effective or determine its actual risk. The regulatory benefit-risk argument still needs clinical evidence and an explanation of how preferences inform interpretation. The voluntary device guidance discusses possible uses across development and regulatory review; it does not guarantee clearance or approval on the basis of preference findings. [1] [3] [11]

2.2 Use Across the Product Life Cycle

E22 describes uses across development and postapproval decision-making. Several are particularly relevant to an operational evidence plan. [1]

Preference research can help identify unmet needs, prioritize outcomes, inform trial design, and explain the acceptability of benefit-risk trade-offs. It may help interpret the importance of components in a multicomponent endpoint or the value patients attach to changes in an outcome. It can also inform treatment characteristics, recruitment and retention, and risk-management planning. These are potential uses, not a checklist every study must satisfy. A study designed to rank treatment convenience should not be repurposed as evidence of tolerance for a serious risk without examining whether its questions and population support that inference. [1] [4]

Moreover, in Section 2.4 (De Novo Work May Not Always be Justified, lines 80–84) and Section 2.5 (Global Applicability, lines 85–92), ICH E22 emphasizes that sponsors should first evaluate existing high-quality PPS literature before burdening patient communities with redundant de novo studies, and that a PPS conducted in one region can support regulatory decision-making in another region provided the applicant justifies cultural and healthcare-system comparability (ideally supported by local qualitative preference bridging data per Section 4.3, lines 226–229) [1].

2.3 Methods, Instruments and Quality

Method selection, instrument development and analysis should follow the research question and the intended decision. E22 and the voluntary device PPI guidance describe considerations rather than one preferred method for every setting. [1] [3]

A. Method Selection

Qualitative work can inform attributes, language and context; quantitative methods can estimate trade-offs or relative importance. The guidance discusses several methods, including the following. [1] [3]

A discrete-choice experiment asks respondents to choose among profiles whose attributes and levels vary according to a design. Best-worst scaling asks for selections of the most and least preferred items or characteristics, with the interpretation depending on the particular design. Threshold techniques investigate the point at which a respondent's choice changes as an attribute varies. Swing weighting evaluates changes from one level of an attribute to another. These approaches differ in assumptions, burden and the quantities they estimate. A short survey is not automatically less biased, and a more elaborate model is not automatically more informative. [1] [3]

The practical decision is whether the method can answer the proposed question in the intended population. The team should document what respondents are asked to compare, which attributes are held constant, how levels are varied, and what a resulting estimate means. Maximum acceptable risk or minimum acceptable benefit estimates are conditional on the attributes, ranges and analytical assumptions actually studied. They should not be advertised as context-free properties of a patient population. [1]

B. Attribute Relevance and Range

Attribute design determines the boundaries of the inference. Three considerations deserve explicit documentation. [1]

First, attributes should be relevant to patients and to the research question. Method-specific assumptions matter; E22 gives participants viewing attributes as independent as an example, not a universal statistical-independence rule for every preference method. Second, alignment with clinical endpoints is particularly important when the study will inform benefit-risk assessment. Perfect alignment may not always be possible, but the limitation should be explained. Third, the levels should cover relevant values expected in the clinical study's treatment and control groups. Extrapolation beyond the studied levels is generally discouraged. [1]

As an illustrative inference problem, a survey that presents a risk only between 1% and 8% does not directly observe choices at 14%. A later clinical result at 14% therefore requires a careful assessment of applicability; a model's ability to calculate a number does not validate extrapolation. The same concern applies when the disease stage, treatment alternatives or burden differs from the survey context. These are reasons to preserve the exact presented attributes and levels alongside the responses. [1]

C. Pretesting and Piloting

Information should be understandable, neutral and appropriate to the population. Numerical and graphical presentations can help communicate probabilities, but their suitability should be evaluated rather than assumed. Framing, order and cognitive burden can affect responses. Our implementation recommendation is to retain the exact wording and display version, test the intended devices and languages, and document the rationale for any randomized presentation. Such controls support the methodological plan; they do not guarantee that all framing or selection bias has been removed. [1] [3]

E22 describes pretesting and piloting as distinct, essential steps in instrument development. [1]

Pretesting examines comprehension, relevance, wording and format with patients and uses the findings to refine the instrument. Piloting then tests the study's operation at a smaller scale before full implementation. The two activities answer different questions: a respondent can understand a question while the electronic implementation still displays an incorrect choice set or fails to retain a response. The team should record what was tested, what changed, and why the final instrument remains suitable. Neither activity is merely a signature on a validation checklist. [1]

E6(R3)'s quality and data-governance principles are helpful context for controlled records and fit-for-purpose systems. E22's reference to E6(R3) does not erase differences between a preference study and an interventional clinical trial. The applicable safeguards, consent process and records should be determined for the actual study, with clear ownership for decisions. [1] [9]

Also notable from the February 6, 2026 ICH E22 Stakeholder Engagement Webinar (Section 2.2.2–3 of the Meeting Report) is the Sick Cells (Sickle Cell Disease) Case Study presented by A. Valentine [2]: to overcome digital-divide, literacy, and medical-mistrust barriers associated with online panels (ICH E22 Section 4.3, lines 231–234), the study deployed a community-led "train-the-trainer" model partnering with regional Community-Based Organizations (CBOs) and trained Community Health Workers (CHWs) who fielded the survey in person or via assisted sessions, provided real-time emotional trauma support, compensated CBOs, CHWs, and patients fairly, and returned a lay summary to the community [2].

D. Quality Checks and Analysis-Set Decisions

E22 provides examples of quality checks and cautions against treating any one check as definitive proof that a response should be removed. [1]

Examples include a clearly dominated choice, inconsistent information, unusually rapid completion, attribute non-attendance and potentially fraudulent responses, including synthetic participants. These are signals to investigate. A patient who consistently prioritizes one attribute may be expressing a real preference; a fast completion time may reflect familiarity rather than fraud. Rules for exclusions should be prespecified and justified, with the implications for representativeness considered. [1]

The draft recommends presenting the impact of quality-based exclusions. Where observations are removed for the primary analysis, results including those observations should also be presented to show sensitivity to the decision. An implementation should preserve the original response, the flag, the review outcome and the analysis-set derivation. Automatically deleting flagged participants would make it harder to assess whether a cleaning rule changed the substantive conclusion or disproportionately removed a patient subgroup. [1]

E. Reporting the Study

E22 recommends including the PPS in CTD Modules 2 and 5, with the full report in 5.3.5.4, Other Clinical Study Reports. It describes potential references in Module 2, including product-development rationale and benefit-risk discussion. These are reporting recommendations, not a requirement to duplicate the same content in every location. The report should connect the objective, design, population, instrument, analysis, limitations and intended regulatory use. The benefit-risk presentation should also distinguish preference estimates from the clinical evidence to which they are applied. [1] [11]

---

3. Postapproval Safety Cases

Published as final FDA guidance on March 4, 2026 (91 FR 10619, FR Doc. 2026-04252, Docket No. FDA-2024-D-0803), E2D(R1) Postapproval Safety Data: Definitions and Standards for Management and Reporting of Individual Case Safety Reports updates the 2003 ICH E2D guideline to govern modern digital platforms, organized data collection ecosystems, and real-world data studies [5], [6].

3.1 Collection Context

E2D(R1) distinguishes spontaneous from solicited reports while separately addressing other observations and regional reporting requirements. The source and circumstances of collection matter. An adverse event is not synonymous with a suspected adverse reaction, and a valid case is not automatically an expedited case. Software should preserve those distinctions so that the qualified safety team can apply the appropriate reporting rules. [5]

Organized collection: An ODCS gathers relevant product or disease-area data in a planned manner. For activities without a protocol, the guidance recommends documentation of five elements: objectives, sources, the dataset and collection period, the review method, and the process for managing identified safety information. This documentation defines the activity; it is more informative than simply tagging every app as an ODCS. [5]

Digital responsibility and context: Platforms under MAH responsibility should be screened often enough to meet applicable reporting timelines. For those platforms, day zero begins when sufficient information meeting the minimum criteria is posted. On a platform outside MAH responsibility, there is no general expectation to patrol all content. If the MAH conducts a planned data-review activity, it should document that ODCS and manage qualifying information found during the planned review. Day zero in that setting is when review identifies the event and sufficient information, not necessarily the date data were accessed. An ad hoc report encountered outside organized collection is considered in its own context. [5]

Support programs: E2D(R1) defines PSPs with attention to two-way communication and solicitation or foreseeable receipt of medical information. A stand-alone delivery service differs from a combined program that includes qualifying nurse support. When a combined program meets the PSP definition, reports from its component activities are managed as coming from the PSP. The guidance's training materials illustrate why program design should be documented before setting up routing rules. [5] [6]

Noninterventional studies: Primary collection and secondary use of existing data have different implications. Solicited reports from primary collection require appropriate causality assessment. E2D(R1) states that events identified in secondary use of data should not be reported as ICSRs unless regional or local requirements indicate otherwise. Study-level findings still require evaluation and communication, including important safety findings. The distinction does not create permission to ignore a meaningful risk. Selective safety collection should only be used where justified and applicable; E19 has its own scope and is not a blanket exemption for every registry. [5] [7] [10]

3.2 Validity, Reportability and Timing

Case validity, reporting eligibility and reporting timing should be assessed separately. The guidance provides the following distinctions. [5]

Minimum information: The usual minimum criteria concern an event or other observation, a suspect or interacting drug, an identifiable patient, and an identifiable reporter. Regional requirements determine which cases are reportable. The guidance discusses an identifiable-patient exception for intercepted medication errors that never reached a patient. This is a specific context, not a general waiver of case quality. [5]

Identifiability: A social-media handle alone, without qualifying information, is insufficient to confirm a real patient or reporter. Where permissible and feasible, follow-up should seek the information needed to assess identifiability. Anonymous reporting is not automatically invalid if the existence of a real reporter is known and other requirements are met. A workflow should retain incomplete information and follow-up attempts rather than silently discard a possible case. [5]

Causality and other observations: Spontaneous reporting implies suspected causality for reporting purposes, while solicited reports require an appropriate assessment. Other observations—such as pregnancy exposure, medication error or lack of effect—can require different handling under regional rules. A broad automated rule suppressing every solicited record labeled unrelated could therefore omit information that still needs assessment. The record should preserve the reporter's account and the MAH's separate evaluation. [5]

Initial reports, follow-up and amendments: New medically relevant follow-up has its own receipt timing and may change whether a case qualifies for expedited reporting. It does not erase the original receipt date or retrospectively extend the initial deadline. An internal quality amendment with no new external information should not receive a new day-zero date. Store the chronology, the information available at each decision, and the reason for classification changes. U.S. requirements in 21 CFR 314.80 and 600.80 should be applied with the relevant reporting category and timing, rather than assuming every incoming record has an identical 15-day deadline. [5] [12]

---

4. Real-World Safety Studies

Published alongside ICH E2D(R1) on March 4, 2026 (91 FR 10615, FR Doc. 2026-04253, Docket No. FDA-2024-D-2754), ICH M14 General Principles on Planning, Designing, Analyzing, and Reporting of Non-Interventional Studies That Utilize Real-World Data for Safety Assessment of Medicines provides recommendations for postmarketing pharmacoepidemiological safety studies [7].

4.1 A Question-Led Feasibility Assessment

M14 presents an iterative framework connecting the research question, design, data fitness, protocol, analysis and reporting. Feasibility is intended to assess whether the proposed study can produce adequate evidence. [7]

The question should specify the population, exposure or intervention, comparator, outcome, timing and setting as appropriate, and clarify whether the purpose is descriptive or causal. Design choices should then be made in that context. A target-trial framework can help make eligibility, treatment strategies and follow-up explicit for a causal question, but it is not a substitute for assessing the data actually available. [7]

M14 recommends at least two feasibility phases: an initial scan of candidate sources and a more comprehensive assessment of their suitability. Feasibility should not be used to compare exposure-outcome effects and choose the source with the most attractive result. Relevant considerations include available participants, exposure ascertainment, outcome capture, covariates, observation time and data provenance. If the data cannot support the question, the team should consider a different source, an adjusted design or additional collection rather than force an analysis from an inadequate dataset. [7]

Our operational interpretation is to keep the feasibility record separate from the final effect analysis, with a transparent record of sources considered and reasons for selection. The guidance does not mandate a particular cryptographic barrier or database product. The control should be proportionate to the risk and should make departures from the planned process visible. [7]

4.2 Measurement and Bias

M14 distinguishes the clinical construct from the operational definition used to identify it in a dataset. That distinction is central to interpreting a real-world safety estimate. [7]

A diagnosis code, medication dispensing or laboratory result is an observation produced for a particular purpose. Its meaning depends on the source and context. An operational definition should explain the code lists, timing, combinations and exclusions used to represent the clinical construct, with validation evidence where appropriate. M14 does not impose a universal requirement for two diagnosis codes or a fixed number of days between them. A more complicated algorithm can still be wrong, and a simple one may be appropriate when adequately supported. [7]

Study design should address relevant biases, including confounding, misclassification and selection. Aligning eligibility, exposure and follow-up can help avoid time-related biases. Choices between new-user and prevalent-user designs should follow the question and likely biases, not a universal template. Methods such as adjustment, validation and sensitivity analysis have assumptions that should be documented. A long list of advanced techniques does not demonstrate that those assumptions are met. [7]

Validation metrics should match the measurement problem and reference standard. Sensitivity, specificity or predictive values may be informative, but their interpretation depends on the population and design. Quantitative bias analysis can explore the likely direction and magnitude of error under specified assumptions, both when planning a study and when interpreting results. It does not create a guarantee that an adjusted estimate is unbiased. Report the parameters and their sources, what was varied, and what uncertainty remains. [7]

4.3 Data-Holder and Researcher Responsibilities

M14 assigns related but distinct quality responsibilities to data holders and researchers. Their documentation should connect the source history to the final analysis. [7]

Data-holder information can explain collection practices, curation, corrections, changes over time and known limitations. Researchers should document their extraction, selection, transformations, derivations and quality checks. A common data model can make analyses easier to implement across sources, but it does not make the original sources interchangeable. Retaining mappings and exceptions helps reviewers understand where standardization changes the interpretation or loses detail. [7]

The practical deliverable is a traceable analytical dataset and a reproducible account of how it was created. That includes the software and specification versions that actually ran, with documentation of deviations. E6(R3) offers relevant data-governance principles in its own clinical-trial scope; it should not be cited as if every noninterventional study automatically has identical requirements. M14 itself supplies the more direct framework for the RWD safety studies discussed here. [7] [9]

---

5. Pregnancy Safety: Complementary Evidence With Explicit Limitations

Published in the Federal Register on May 11, 2026 (91 FR 25579, FR Doc. 2026-09268, Docket No. FDA-2018-D-4693), FDA's 33-page final guidance Postapproval Pregnancy Safety Studies (dated May 2026, prepared by CDER's Office of New Drugs, Office of Surveillance and Epidemiology, and Office of Translational Sciences in cooperation with CBER and the Office of Women's Health) provides recommendations for evaluating drug and biologic exposures in pregnant women [8].

5.1 Regulatory Context and Exposure Timing

The pregnancy guidance's background cites approximately 3.6 million annual U.S. births and the estimate that 41.6% of pregnancies in 2019 were unintended. The latter describes pregnancy intention, not a drug-exposure rate. The guidance explains that historical exclusion from many premarketing studies can leave substantial uncertainty at approval, while recognizing the value of appropriate premarketing evidence. The operational implication is to plan how exposure and outcome information will be obtained and interpreted, rather than assuming spontaneous reports alone will establish comparative risk. [8]

Under Section 505(o)(3) of the FD&C Act (21 U.S.C. 355(o)(3), lines 243–286), FDA is authorized to require postmarketing studies or clinical trials at the time of approval (or post-approval upon learning of new safety information under Section 505(o)(3)(C)) to assess a known serious risk, evaluate a signal of serious risk, or identify an unexpected serious risk to the pregnant woman or developing fetus [8], [12]. The relevant exposure and observation windows should follow the research question and product characteristics. [8]

The relevant exposure period depends on the drug and the outcome. Information before pregnancy, during pregnancy and after delivery may be relevant, but a generic calculation such as five half-lives is not a universal enrollment or analysis rule. A protocol should justify exposure definitions and outcome windows using the product and scientific question. It should also distinguish maternal outcomes, pregnancy outcomes and infant outcomes, since their observation periods and data sources can differ. The database should preserve sufficient timing information for the planned analysis without pretending that imprecise dates are exact. [8]

FDA discusses complementary contributions from pharmacovigilance, pregnancy registries and other studies, including electronic healthcare data. These approaches need not be collapsed into one dataset or presented as a newly mandated three-part architecture. A case report may identify a concern, a registry may collect detailed prospective information, and an electronic-data study may provide a larger comparative population. Each source has limitations that the others may only partly address. [8]

5.2 Registry Design, Outcome Knowledge and Comparators

The guidance recommends prespecified definitions, ascertainment methods and analytical plans for pregnancy registries. Those choices should reflect the product and research objective. [8]

Enrollment timing: Ideally, enrollment occurs before prenatal tests provide knowledge of the outcome. FDA also recognizes that enrolling consistently before testing can be difficult and recommends including women who have had early prenatal testing. Recording what was known at enrollment helps assess selection and ascertainment bias. The system should not automatically exclude every participant with a prior test or label every later enrollment identically without considering the study definition. [8]

Outcome definitions: The guidance distinguishes live birth, miscarriage before 20 weeks, termination, and fetal death or stillbirth at or after 20 weeks, while recommending prespecified criteria for study outcomes and major congenital malformations. Consistent ascertainment and appropriate clinical expertise matter. The definition and adjudication process should be explicit, including how exposure information is handled to reduce bias. This is an evidence-quality issue, not a justification for a software-generated clinical diagnosis. [8]

Comparators and follow-up: A general-population background rate may differ from the exposed group in disease, treatment and other prognostic factors. A suitable comparator and an account of confounding are therefore important to interpreting any association. Follow-up should capture the planned outcomes, losses and reasons for missing information. A reassuring rate among participants who remain in follow-up may not represent those who were lost. Sample-size limitations should be stated, especially when examining rare specific outcomes rather than a broad composite. [8]

5.3 Complementary Electronic Healthcare Data

Complementary studies using electronic healthcare data can extend the evidence base, but their larger size does not remove limitations in exposure measurement, gestational timing, outcome ascertainment or linkage. [8]

A maternal dispensing record does not prove that the medicine was taken. Pregnancy start dates may be estimated rather than directly observed. Mother-infant linkage can fail in ways related to coverage, geography or outcomes. Restricting an analysis to live births may omit relevant non-live outcomes and introduce selection bias. These limitations should be evaluated in the protocol and analysis rather than treated as routine data-cleaning noise. [8]

Sensitivity analyses using different exposure definitions may be useful, but requiring two dispensings is not a universal correction for nonadherence and does not eliminate misclassification. Linkage rates, reasons for nonlinkage and differences between linked and unlinked records should be examined where possible. If a registry and an electronic-data study overlap, the program should consider that overlap before presenting the results as independent corroboration. A larger combined count is not necessarily a larger amount of independent evidence. [7] [8]

---

6. A Proposed Evidence Architecture for Sponsors and CROs

The following model translates the source distinctions into reviewable operational records. It is our proposed design approach, not an FDA-mandated schema or a claim about a deployed product. Actual dataset mappings, reporting codes and system controls require verification against the applicable standards, regional rules and study specifications.

6.1 Preference Instrument and Analysis Provenance

The preference-study record should connect what a participant saw to the data analyzed. A response without its attribute levels, language, instrument version and choice-set definition can be difficult to interpret later. The mapping to any exchange or submission format should be validated separately from the survey's scientific design. E22's helpful reference to ODM does not establish a universal ODM or SDTM schema for all preference tasks. [1] [13]

Our recommended acceptance test follows a synthetic response from presentation through storage, quality review, analysis-set selection and the reported estimate. Verify that a changed translation or display does not silently inherit the old version identifier. Keep quality flags distinct from decisions to exclude observations, and ensure that results including excluded observations can be reproduced when needed. Analysts should be able to explain which instrument and population support each inference. [1]

A useful handoff includes a human-readable specification as well as machine-readable data. The specification should explain attribute meanings, allowed levels, choice logic, missing-response handling and derived quantities. A technically valid export is insufficient if it changes the meaning of a choice or drops a relevant version distinction. This is where sponsor, patient-research and data-management teams need shared ownership of the evidence, rather than a late transfer of an unexplained spreadsheet. [1] [3]

6.2 Safety Intake and Reporting Decisions

The safety intake model should preserve source context and chronology before applying reporting rules. Its implementation should be validated against current E2B specifications and regional requirements. The conceptual workflow below deliberately avoids inventing data-element codes or assuming all receivers use the same implementation. [5] [10] [12]

First, record how the information arrived: spontaneous contact, an MAH-responsible platform, a planned third-party-platform review, a support program, a study or another source. Second, preserve the original account and information available at receipt. Third, assess minimum criteria, causality where relevant, seriousness, expectedness and other applicable reporting considerations. Finally, document the reporting decision, submission and subsequent information. These steps may overlap operationally, but their evidence should remain reconstructable. [5]

Test cases should include an incomplete social-media report, a combined PSP, a new follow-up that changes seriousness, and an internal amendment that adds no external information. They should also include a secondary-data study finding and an observation without an adverse event. Such cases reveal whether an overly broad routing rule incorrectly creates, suppresses or delays a report. The goal is to support timely qualified assessment, not to let a channel label or a single automated causality flag make every decision. [5] [6]

6.3 Reproducible Real-World Safety Analysis

For a real-world safety study, the implementation should make the research question, source selection and analytical derivation reviewable. [7]

Retain the feasibility criteria and the reasons candidate sources were accepted or rejected. Record the cohort definition, code lists, date windows, exclusions and transformations that produced the analyzed population. Distinguish a source-data correction from a change in the analytical definition; the two can have different implications for reproducibility.

Validation and sensitivity analyses should answer identified risks. If outcome misclassification is a concern, document the evidence supporting the operational definition and the plausible consequences of error. If unmeasured confounding is important, explain the assumptions used to assess its possible impact. Do not advertise an automated QBA module as eliminating those uncertainties. The final report should retain both the result and its practical limits, with sufficient information for another qualified analyst to understand the derivation. [7]

6.4 Pregnancy Evidence Across Sources

The pregnancy data model should preserve timing, ascertainment and linkage decisions without embedding an unvalidated clinical algorithm. [8]

For a registry, retain enrollment timing, relevant prior outcome knowledge, exposure information, planned outcomes and follow-up status. For electronic healthcare data, retain the algorithms used to identify pregnancy episodes, estimate timing, ascertain outcomes and link offspring records. In either case, distinguish observed values from derived or imputed values, and retain the uncertainty appropriate to each.

A cross-source analysis should consider duplicate inclusion, differences in comparator selection, and differences in which outcomes can be detected. It should not declare evidence independent simply because records reside in different systems. Tests should include incomplete dates, failed linkage, non-live outcomes, changed exposure information and loss to follow-up. The purpose is to reveal how those cases affect the analysis, not to force them into a convenient complete record. [7] [8]

---

7. Readiness Questions

A bounded readiness check should connect the evidence question to the operational records and the final interpretation:

1. Scope and status: Identify the applicable guidance, distinguish draft recommendations from final guidance and binding requirements, and document regional reporting rules. [1] [5] [12]
2. Meaning and provenance: Preserve instrument versions, source context, cohort definitions and timing. Verify that transformations retain the distinctions needed for the intended decision. [1] [5] [7] [8]
3. Participant perspective and protection: Check whether the represented population and collection method support the inference. Keep consent and privacy constraints visible across transfers and follow-up. [1] [3] [8]
4. Analysis and uncertainty: Reproduce exclusions, derivations, sensitivity analyses and their assumptions. Explain what the data cannot establish instead of translating every uncertainty into a software rule. [1] [7]
5. Accountability: Assign ownership for review, reporting, follow-up and change control. Test difficult cases before relying on the workflow for routine operations. [5] [6]

For EClinCloud's audience, the value of integration lies in preserving the right distinctions across the product life cycle. Preference research explains acceptability, individual-case workflows support pharmacovigilance, and well-designed observational studies evaluate safety questions in real-world use. Connecting these activities can make evidence easier to evaluate, but merging their categories or overstating guidance requirements can make it less reliable.

---

Sources

1. U.S. Food and Drug Administration (Center for Drug Evaluation and Research [CDER] and Center for Biologics Evaluation and Research [CBER]) and International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH). E22 General Considerations for Patient Preference Studies: Draft Guidance for Industry. ICH Step 2 Draft Guideline endorsed November 19, 2025; issued as FDA Draft Guidance February 2026, Docket No. FDA-2026-D-0207, 91 FR 5484 (Federal Register Document 2026-02324, published February 6, 2026), 21 pages. Official FDA PDF: https://www.fda.gov/media/190946/download; ICH Step 2 PDF: https://database.ich.org/sites/default/files/ICH_E22_Step2_draftGuideline_Assembly_Endorsed_FINAL_2025_1119.pdf; Federal Register Notice: https://www.federalregister.gov/documents/2026/02/06/2026-02324/e22-general-considerations-for-patient-preference-studies-international-council-for-harmonisation.

2. International Council for Harmonisation (ICH) E22 Expert Working Group (EWG). Webinar for Stakeholder Engagement on the New Draft Guideline on General Considerations for Patient Preference Studies (ICH E22): Meeting Report (6 February 2026) (4 pages) and ICH E22 EWG Work Plan (13 February 2026; Rapporteur Dr. Francesco Pignatti [EC, Europe], Regulatory Chair Dr. Laura Lee Johnson [FDA, United States]) (2 pages), accompanied by ICH E22 Step 2 Presentation (11 December 2025) (27 slides). Official ICH PDFs: https://database.ich.org/sites/default/files/ICH%20E22%20EWG%20Stakeholder%20Webinar%20Meeting%20Report_2026_0707.pdf and https://database.ich.org/sites/default/files/ICH52_E22_EWG_WorkPlan_2026_0318.pdf.

3. U.S. Food and Drug Administration (Center for Devices and Radiological Health [CDRH] and CBER). Incorporating Voluntary Patient Preference Information over the Total Product Life Cycle: Guidance for Industry, Food and Drug Administration Staff, and Other Interested Parties (Final Guidance, Document Number GUI01500006, issued March 30, 2026, superseding the August 24, 2016 guidance; draft issued September 6, 2024 at 89 FR 72856). Docket No. FDA-2015-D-1580, 91 FR 15622 (Federal Register Document 2026-06063, published March 30, 2026), 38 pages. Official PDF: https://www.fda.gov/media/92593/download; Federal Register Notice: https://www.federalregister.gov/documents/2026/03/30/2026-06063/incorporating-voluntary-patient-preference-information-over-the-total-product-life-cycle-guidance.

4. U.S. Food and Drug Administration (CDER, CBER, and CDRH). Patient-Focused Drug Development: Incorporating Clinical Outcome Assessments Into Endpoints for Regulatory Decision-Making: Guidance for Industry, Food and Drug Administration Staff, and Other Stakeholders (PFDD Guidance 4, Draft Guidance). Docket No. FDA-2023-D-0026, April 2023, 48 pages. Official PDF: https://www.fda.gov/media/166830/download.

5. U.S. Food and Drug Administration (CDER and CBER) and International Council for Harmonisation (ICH). E2D(R1) Postapproval Safety Data: Definitions and Standards for Management and Reporting of Individual Case Safety Reports: Guidance for Industry (Final Guidance). issued as FDA Final Guidance March 2026, Docket No. FDA-2024-D-0803, 91 FR 10619 (Federal Register Document 2026-04252, published March 4, 2026; draft published March 14, 2024 at 89 FR 18644), 28 pages. Official FDA PDF: https://www.fda.gov/media/176977/download; ICH Step 4 PDF: https://database.ich.org/sites/default/files/ICH_E2D%28R1%29_Step4_FinalGuideline_2025_0819.pdf; Federal Register Notice: https://www.federalregister.gov/documents/2026/03/04/2026-04252/e2dr1-postapproval-safety-data-definitions-and-standards-for-management-and-reporting-of-individual.

6. International Council for Harmonisation (ICH) E2D(R1) Expert Working Group. ICH E2D(R1) Training Material: Post Approval Safety Data: Definitions and Standards for Management and Reporting of Individual Case Safety Reports (November 2025; file posted December 2025) (63 slides) and ICH E2D(R1) Step 4 Presentation (4 September 2025) (28 slides). Official ICH PDFs: https://database.ich.org/sites/default/files/ICH_E2D%28R1%29_TrainingMaterial_2025_1211.pdf and https://database.ich.org/sites/default/files/ICH%20E2D%28R1%29_Step%204_Presentation_2025_0904.pdf.

7. U.S. Food and Drug Administration (CDER and CBER) and International Council for Harmonisation (ICH). M14 General Principles on Planning, Designing, Analyzing, and Reporting of Non-Interventional Studies That Utilize Real-World Data for Safety Assessment of Medicines: Guidance for Industry (Final Guidance). issued as FDA Final Guidance March 2026, Docket No. FDA-2024-D-2754, 91 FR 10615 (Federal Register Document 2026-04253, published March 4, 2026; draft published July 5, 2024 at 89 FR 55613), 47 pages. Official FDA PDF: https://www.fda.gov/media/179795/download; Federal Register Notice: https://www.federalregister.gov/documents/2026/03/04/2026-04253/m14-general-principles-on-planning-designing-analyzing-and-reporting-of-non-interventional-studies.

8. U.S. Food and Drug Administration (CDER Office of New Drugs, Office of Surveillance and Epidemiology, and Office of Translational Sciences; CBER; and Office of Women's Health). Postapproval Pregnancy Safety Studies: Guidance for Industry (Final Guidance). Docket No. FDA-2018-D-4693, 91 FR 25579 (Federal Register Document 2026-09268, published May 11, 2026; guidance dated May 2026), 33 pages. Official PDF: https://www.fda.gov/media/124746/download; Federal Register Notice: https://www.federalregister.gov/documents/2026/05/11/2026-09268/postapproval-pregnancy-safety-studies-guidance-for-industry-availability.

9. International Council for Harmonisation (ICH) and U.S. Food and Drug Administration (CDER and CBER). ICH E6(R3) Guideline for Good Clinical Practice (Principles and Annex 1) (Step 4 Final Document, January 6, 2025). Official ICH PDF: https://database.ich.org/sites/default/files/ICH_E6%28R3%29_Step4_FinalGuideline_2025_0106.pdf.

10. International Council for Harmonisation (ICH) and U.S. Food and Drug Administration (CDER and CBER). ICH E19 A Selective Approach to Safety Data Collection in Specific Late-Stage Pre-Approval or Post-Approval Clinical Trials: Guidance for Industry (December 2022) and ICH E2B(R3) Electronic Transmission of Individual Case Safety Reports (ICSRs) Implementation Guide — Data Elements and Message Specification (October 2026, Revision 2). Official FDA PDFs: https://www.fda.gov/media/163670/download and https://www.fda.gov/media/81904/download.

11. International Council for Harmonisation (ICH). ICH M4E(R2) Common Technical Document (CTD) — Efficacy: Revision of M4E Guideline on Enhancing the Format and Structure of Benefit-Risk Information in ICH (Sections 2.5.1, 2.5.6, 2.5.6.5, and 5.3.5.4, June 2016). Official ICH PDF: https://database.ich.org/sites/default/files/M4E_R2__Guideline.pdf.

12. U.S. Statutory and Code of Federal Regulations Mandates for Patient Experience Data, Postmarketing Safety Reporting, and Pregnancy Safety Labeling: Federal Food, Drug, and Cosmetic Act Section 569C (21 U.S.C. 360bbb-8c, Patient Experience Data), Section 505(o)(3) (21 U.S.C. 355(o)(3), Postmarketing Studies and Clinical Trials), 21 CFR 314.80 (Postmarketing reporting of adverse drug experiences), 21 CFR 600.80 (Postmarketing reporting of adverse experiences for licensed biological products), 21 CFR 201.57(c)(9)(i)(A) (Pregnancy Exposure Registry contact information in prescription drug labeling), 21 CFR 860.7(c)(1) (Valid scientific evidence for medical devices), and 21 CFR Part 50 Subpart D (21 CFR 50.50–50.56, Additional Safeguards for Children in Clinical Investigations). Official Portal: https://www.ecfr.gov/current/title-21.

13. National Institutes of Health (NIH) Office of Extramural Research, NIH RePORTER Extramural eClinical Research Project Corpus (nih_reporter_eclinical, Snapshot August 23, 2026; 4,876 deduplicated project applications across 31 eClinical queries), and Clinical Data Interchange Standards Consortium (CDISC), Foundational and Therapeutic Area Standards Repository (cdisc_standards, Snapshot July 31, 2026; 32 standards pages including ODM v2.0, QRS, CDASH, SDTM, ADaM and 54 specification artifacts). Primary Portals: https://reporter.nih.gov/ and https://www.cdisc.org/standards.