Integrated summaries of safety and efficacy

A marketing application summarises a development programme's efficacy and safety data across its studies at more than one level, from the Common Technical Document's clinical summaries in Module 2 to the integrated summaries of efficacy and safety that the FDA requires in Module 5. What these summaries can show depends on how the programme's data were captured, which studies were pooled and how the pool was analysed, and writing them is the work of a team of writers alongside data management and biostatistics.

Levels of clinical summary in the dossier

A clinical development programme is designed to collect the information pertinent to evaluating a product's benefit-to-risk ratio, and at later stages, when preparing for submission, the task is to understand the picture that arises from all the available data. ¹

ICH M4E(R2) describes three levels of detail for presenting clinical efficacy and safety data: the clinical overview, including the overviews of efficacy and safety in Modules 2.5.4 and 2.5.5; the summary of clinical efficacy and the summary of clinical safety in Modules 2.7.3 and 2.7.4; and the detailed reports of the individual clinical studies, written to ICH E3, in Module 5.3.5. ² The summary of clinical efficacy and the summary of clinical safety are required across the ICH regions, and they give a detailed factual summary of all the data relevant to efficacy and safety in the intended patient population and may include a summary of results from integrated databases. ^{2, 3}

The two Module 2 summaries identify, organise and condense efficacy and safety data from across the development programme and give conclusions on those data, so they are a highly condensed version of the case for approval. ³ Applications to the FDA must also contain two summary documents that the other ICH regions do not require: the integrated summary of efficacy, also called the integrated summary of effectiveness, and the integrated summary of safety, both in Module 5.3.5.3. ^{2, 3} They report analyses of data from more than one study, describing and presenting the results from integrated efficacy and safety databases. ²

Unlike the integrated summaries, the Module 2 summaries do not require data to be integrated across studies. ⁴ All studies can be summarised, but not all can be integrated, and disparate study designs precluded pooling of data in many development programmes. ⁴

Pooled and study-by-study presentation

Analyses of integrated efficacy and safety databases are an FDA requirement, and they are very useful in evaluating the data gathered in multiple clinical studies. ¹ Their usefulness depends on the quality of the studies and of the methods that gathered the data, on a scientifically sound strategy for pooling and analysing the data, and on adequate reporting of the results. ¹

A programme's clinical safety and efficacy data can be presented in two main ways. ¹ Separate studies set side by side allow specific types of data to be compared between studies. ⁵ Pooled data treat a group of studies as a single dataset, an integrated database used as one large study, which gives the benefit of a large sample size and can help in understanding the "big picture" as well as in addressing specific clinical issues with the product. ⁵

The integrated summary of effectiveness aims to provide insights beyond those seen in the individual trials. ⁶ The individual studies are presented to demonstrate the claimed effects, and a statistical combination of their results may then be considered where it is applicable and appropriate. ⁶

Studies are generally pooled for two main purposes: to achieve greater power and precision, and to assess the drug's effect in demographic or clinical subpopulations that have too few subjects in any individual study to support meaningful conclusions. ^{6, 7} Many pooled analyses, however, are exploratory, designed to probe the data for trends across studies, for example in disease-specific subgroups. ⁷

Pooling data for Module 2.7.4

The submission strategy can pose its own challenges for Module 2.7.4, and a key decision of its safety strategy is whether to pool data across studies, present each study individually, or take a hybrid approach in which only some data are pooled. ⁸ The final decision should not be the writer's, but the writer can still contribute to the discussion. ⁸ The team should always consider ease of understanding for the reviewer, which helps the reviewer reach an opinion on the product's safety. ⁸

A product studied in several indications may have a development programme with diverse study designs and dosing regimens. ⁸ The focus should be on the pivotal studies in the proposed indication, so that the reviewer can easily see how data from the supportive studies are relevant to the submission. ⁸ Safety data could be pooled by regimen, for example, but it may be more appropriate to pool data from indications closely related to the proposed one, or from all supportive studies for comparison with the pivotal ones. ⁸ Whatever the pooling approach, it should be presented clearly and consistently throughout the module. ⁸

A data strategy for pooling

For valid and informative results on a drug's safety and efficacy, studies and their data capture and management strategies should be designed with integration in mind. ⁹ Early involvement of data management and biostatistics professionals can help produce valuable integrated summaries, through study designs and data management strategies geared towards pooling data from multiple studies. ¹ A data strategy plan across all the studies, as consistent as possible in how data are captured and cleaned, allows a streamlined analysis and reduces the need for retrospective review and processing. ⁷

When starting a phase 2a study, many sponsors are uncertain whether their product will be eligible for a marketing application, and some are not yet adequately funded. ⁷ So instead of a long-term, programme-wide data strategy, they choose "minimum essential" data capture and management, using cheap, less reliable and inconsistent methods such as paper case report forms or spreadsheets. ⁷ The FDA requires specific database design standards for database submissions, such as CDISC's Study Data Tabulation Model. ⁷ When planning phase 2b studies, companies are more likely to choose electronic data capture, but through lack of awareness or of resources the standards the FDA requires are not always taken into account. ⁷

For pooling and cross-study analyses, the data have to be in a consistent format; for example, all adverse event data should be coded in the same version of MedDRA. ⁷ Phase 3 studies are usually designed with data management experts with submission in mind, and an adequate electronic data capture system with CDISC-compliant capture is likely to be chosen. ⁷ When a phase 3 programme includes more than one study, the structure of the studies, the treatment duration, the visit schedules and the collection of safety and efficacy variables are all planned consistently, which allows data capture and cleaning to be standardised. ⁷

Legacy data should be processed to the same standard. ⁷ Legacy data captured on paper are entered into an electronic system retrospectively, with minimal resources for cleaning the data and resolving queries, and electronic legacy data captured in a format that is not CDISC-compliant must be converted. ⁷ This retrospective processing adds costs and time that can be minimised by implementing a data strategy for integration and submission early in development. ⁷

The statistical strategy for integration

Statistical considerations are essential in forming the data integration strategy, and the analytical strategy should include: ¹⁰

• the objectives for integration
• the regulatory guidelines
• which studies to pool
• the outcomes and time points
• the statistical methods

^{10, 11}

Guidelines that can be used in establishing a strategy for the integrated summaries or for pooling data across studies are: ⁶

• "Summarising the Clinical Database" in ICH E9
• the FDA's 2018 draft guidance, Meta-Analyses of Randomised Controlled Clinical Trials to Evaluate the Safety of Human Drugs or Biological Products
• the FDA's 2015 guidance, Integrated Summary of Effectiveness

⁶

The draft meta-analysis guidance focuses primarily on meta-analyses with predefined hypotheses designed to confirm a suspected safety risk, rather than on exploratory meta-analyses. ⁷ It discusses in detail and with scientific rigour the important considerations in pooling study data for regulatory purposes, and it is a valuable resource even when there is no formal predefined safety hypothesis to check. ⁷

Choosing the studies to pool

The first step in deciding which data to pool is to list all the clinical studies with their critical design characteristics and their roles in the development programme. ¹¹ Any study included in pooled data is still evaluated and discussed thoroughly as an individual study. ¹¹

Trials should be selected for pooling with careful attention to trial design and data quality. ⁹ The aim of data integration is a valid description of expected safety and efficacy in the target population, and its usefulness depends on individual trials with high-quality data, so the first principle is quality over quantity. ⁶ For efficacy integration, the principle means limiting the candidates to individual studies that are considered "adequate and well-controlled" and serve as the basis for efficacy claims. ⁶ For safety integration, findings from a limited set of trials, selected with careful attention to trial and data quality and to the product's intended use, can give a more informative view of product safety than a broader set that includes trials of poor quality. ⁶

The candidate studies are likely to differ, so whether they can be meaningfully pooled should be assessed against: ⁶

• the population, including demographic or clinical factors such as age range and disease severity
• the exact indication evaluated, from the study objectives and the inclusion and exclusion criteria
• relevant trial design factors, such as the control group, if any, the treatment and follow-up duration, the allocation ratio and the endpoints collected

⁶

The variability tolerated in each factor may differ between the integrated datasets for efficacy and for safety. ⁶ For instance, analysing adverse events may be appropriate in an integrated dataset that includes subjects with different disease severities, while efficacy conclusions may be harder to draw from such a varied dataset. ⁶

A single integrated database can be planned if all the candidate studies are similar enough for integration. ⁶ Otherwise, several data pools can be proposed for groups of similar studies, such as studies with a certain active control or with long-term follow-up. ⁶ Healthy-volunteer studies should not be included in the integrated datasets for either efficacy or safety, because they assess a population distinctly different from the target patients and are commonly much shorter. ⁶ They are analysed separately and may be pooled as a distinct safety cohort. ⁶ Statistical methods can adjust for important differences between the patients and studies that form the integrated database. ⁶

Endpoints and time points

When efficacy data are combined, the focus should be on the prespecified primary endpoints of the confirmatory phase 3 studies. ¹² Where important outcomes are common to all the studies, even when their primary endpoints differ, analysing those outcomes can give an important assessment of consistency. ¹² One example in the FDA's guidance on the integrated summary of effectiveness is a series of studies that assessed an important variable at several time points, where results at a common time point can be shown even though the time point of the primary analysis differed between studies. ¹²

Unlike efficacy endpoints, safety endpoints are generally standard, usually adverse events, laboratory data and vital signs, so pooling all the safety outcomes can be expected. ¹² Their collection time points may still not be uniform, and studies may differ in treatment duration and follow-up. ¹² Common time points shared by all the studies, together with statistical handling of the differences in follow-up duration, may help integrate these measures across studies. ¹²

Analysing the pooled data

Selected studies should be combined with appropriate statistical methods, with care over naive pooling of the data. ⁹

Efficacy analysis mainly estimates a pooled treatment effect, comparing treatment and control groups with appropriate statistical models. ¹³ The regression model used for each endpoint in the individual confirmatory study can be applied to the integrated database, further stratified by study. ¹³ The pooled treatment effect is accompanied by an assessment of its homogeneity across studies, contrasting the study-specific effects and testing the study-by-treatment interaction, and a forest plot of the individual study results and the pooled effect is often provided. ¹³ The integrated database is thus treated as a single large trial, while the stratification accounts for variability between studies and keeps the randomisation within each study. ¹³

The statistical analyses of the common safety outcomes are descriptive, but they need special attention. ^{13, 14} Individual clinical study reports generally analyse adverse events as crude percentages, the number of patients with the event divided by the number of patients treated. ¹⁴ Pooling the safety database naively, as one large study with crude percentages for each treatment group, may give a biased result when the trials use different randomisation allocations. ¹⁴

The FDA's draft meta-analysis guidance illustrates this bias, termed Simpson's paradox. ¹⁴ In its example, the risk of a specific safety event was identical in the treatment and control groups of every trial, so analysis of the individual studies would show no concern for the product. ¹⁴ The risk was not the same across the trials, however: one trial had a higher risk in both its treatment and control groups and used a 4:1 allocation of treatment to control, so simple pooling enriched the treatment group with high-risk patients and gave a biased overall result of increased risk. ¹⁴

Statistical solutions to this bias use stratification by study and weighting. ¹⁴ Contrasting the results of a pooled analysis with those of each study can help show whether a bias has occurred, which supports presenting the individual study results as well. ¹⁴ Pooling the integrated database in these ways assumes that subject-level data are available; where only trial-level summary measures exist, meta-analytic statistical methods are available. ¹⁴

Writing the summaries

A summary document must be succinct and simple, yet not simplistic, as well as encompassing and complex. ¹⁵ It gives concise descriptions of the key data in the application and overarching messages about them, and it must be wide enough in scope and detailed enough to account for the primary, secondary and exploratory findings of the major studies in the programme. ¹⁵

The writer's part

Medical writers take the data and all the background material and turn them into a coherent narrative of what is currently known about the product's efficacy and safety. ¹⁶ So, like data management and biostatistics professionals, they should be involved early in planning for submission, to acquire the knowledge and understanding needed to report the data accurately and meaningfully, to get acquainted with the pivotal points of the product information and to devise a project management plan. ^{16, 1, 9} The writer, whether in-house or outsourced, should be familiar with the programme's clinical studies and understand their distinct characteristics as well as the shared ones that make them eligible for pooling. ¹⁶ It is also highly important to understand the evaluation methods each study used and how each result contributes to the overall claim. ¹⁶

Preparation for writing the submission documents should be based on three classes of documents: ¹⁶

1. sponsor-submitted documents, including the clinical study reports of earlier studies in the programme and any submission documents from previous programmes for the same product
2. documented communications with regulatory authorities
3. templates and guidance provided to sponsors and reviewers

¹⁶

Together, these materials can help the writer understand the company message, how knowledge about the product has evolved over time, and the agreements reached with the different regulators relevant to a global submission. ¹⁶ The presentation of both efficacy and safety should focus on the indication sought in the target patient population, taking care to include the types of information the regulator is specifically interested in. ¹⁶

Length and order of writing

The documents in Module 2.7 and in Module 5.3.5.3 generally follow the same outline. ¹⁶ The Module 2.7 summaries, however, are limited in length, up to about 200 pages, and cannot go into the same level of detail as the Module 5.3.5.3 documents, which can be thousands of pages long. ¹⁶ So it may be advisable to write the integrated summaries of safety and efficacy first and then summarise their most important information, taking care not to "cherry-pick" favourable results. ¹⁶ The volume can be reduced by giving fewer methodological details of the pooling and integration and by cross-referring amply from the summaries to the corresponding sections of the Module 5.3.5.3 documents and to source tables and listings. ¹⁶

The writing team

The summary documents and the clinical study reports of the pivotal studies the application relies on are likely to be written concurrently. ¹⁶ So a submission should be written by a team of writers who keep in very frequent communication to keep the messages and focus consistent across documents. ¹⁶ The team should plan for several cycles of cross-review of the submission documents. ¹⁶ It may be reasonable to assign one leader for the efficacy documents, the integrated summary of efficacy and the summary of clinical efficacy, and one for the safety documents, the summary of clinical safety and the integrated summary of safety. ¹⁶

Writers' tasks on each summary document

The writers surveyed

In a survey run from 11 February to 1 March 2015, 188 writers believed to have written at least one of the four summary documents within the previous two years were invited to describe the tasks involved in creating them: the summary of clinical efficacy, the summary of clinical safety, the integrated summary of effectiveness and the integrated summary of safety. ^{17, 18, 19} A respondent qualified as a lead writer of a document by having determined its regulatory content and written at least 50% of it, or as a substantive contributor by having written at least 30% of it with the lead writer. ²⁰

Of the 68 respondents who qualified, 73.5% (50/68) had five or more years' experience writing summary documents, 67.6% (46/68) were full- or part-time employees of a pharmaceutical or biotechnology company or contract research organisation, 29.4% (20/68) were contract writers, and 79.4% (54/68) lived in the US. ^{20, 21, 4} They had submitted drug applications in the US (95.6%, 65/68), the EU (82.4%, 56/68), Japan (27.9%, 19/68) and other regions (33.8%, 23/68), shares that do not sum to 100% because some had submitted in more than one region. ^{20, 21}

Of the 68, 54 (79.4%) had been a lead writer or substantive contributor to a summary of clinical efficacy, 57 (83.8%) to a summary of clinical safety, 32 (47.1%) to an integrated summary of effectiveness and 42 (61.8%) to an integrated summary of safety. ^{22, 23} Fewer had written the integrated summaries, as only US submissions require them. ²⁴ Respondents answered only about the documents they had written, and only those who had submitted an application to the FDA answered about the integrated summaries. ²²

Ways of producing content

Respondents used four methods to create content for the summary documents: ²⁵

1. writing it de novo, as new text or in-text tables for new or previously nonexistent information
2. repurposing it, by identifying and modifying content from an outside source and then integrating it into the summary document
3. a mix of de novo and repurposed content
4. receiving it from another team member or an outside source

²⁵

The mix of de novo and repurposed content was the method used most. ^{23, 24} Repurposing is a strategy used to save time and ensure accuracy where content has to be created that is similar to content elsewhere, in other sections of the submission or in sources outside it. ²⁴ It requires identifying the relevant content, including data, and adapting it to the other document, and beyond copy and paste it requires placing the content thoughtfully in the summary document, so it can be just as challenging as creating content de novo. ²⁴

The summary of clinical efficacy

For Section 2.7.3.1, the background and overview, 94.4% (51/54) of respondents used a mix of de novo and repurposed text. ²⁶ Section 2.7.3.2, the summary of results of individual studies, requires narrative descriptions of the programme's studies, a narrative being a textual summary of one clinical study. ²⁶ For studies that had investigated both efficacy and safety, 57.4% (31/54) described the study methods and summarised the efficacy results only, while 22.2% (12/54) summarised both the efficacy and the safety results. ^{4, 27} For a study whose primary focus was safety, 18.5% (10/54) described its safety methods and summarised its safety results in this section. ^{4, 27}

How content for Section 2.7.3.3, the comparison and results across studies, was created depended on whether the efficacy data were pooled. ⁴ Where they were pooled, 46.3% (25/54) took part in decisions about analysing the data, and any text based on pooled efficacy tables must be written de novo. ^{4, 23, 24} Where they were not pooled, 64.8% (35/54) repurposed efficacy data from individual source documents such as the clinical study reports, and 31.5% (17/54) received efficacy data from another team member. ⁴

Section 2.7.3.4, the analysis of clinical information relevant to dosing recommendations, and Section 2.7.3.5, persistence of efficacy or tolerance effects, must be written de novo, because the information they require does not exist in the source documents, for the most part the clinical study reports. ⁴ For Section 2.7.3.4, 53.7% (29/54) were primarily responsible for writing it and 37.0% (20/54) received the text from another team member. ^{4, 28, 29} For Section 2.7.3.5, the shares were 57.4% (31/54) and 35.2% (19/54). ^{4, 28, 29}

For the optional summary of published literature, 27.8% (15/54) searched the literature and summarised it de novo, 27.8% (15/54) received a summary from another team member and changed it substantively, 25.9% (14/54) received one and incorporated it with few or no changes, and 11.1% (6/54) cross-referred to a stand-alone summary of the published literature. ^{4, 28, 29} Finally, 77.8% (42/54) closed the document with a de novo high-level summary of its efficacy data. ^{28, 29}

The summary of clinical safety

As with the efficacy summary, writers of the summary of clinical safety primarily used a mix of de novo and repurposed content. ³⁰ Although the guideline does not mandate an introductory section, 78.9% (45/57) wrote an introduction repurposed from Section 2.7.3.1 of the efficacy summary, and 14.0% (8/57) wrote one de novo. ³⁰ For Section 2.7.4.1.1, the overall safety evaluation plan and narratives of safety studies, 54.4% (31/57) were part of the team that determined the plan. ^{30, 23, 24} Most (73.7%, 42/57) wrote narratives summarising the study methods and the safety results only, while 22.8% (13/57) used the electronic capacities of the CTD to cross-refer to the study narratives in Section 2.7.3.2 of the efficacy summary. ³⁰

Section 2.7.4.5, safety in special populations, requires de novo text, and 77.2% (44/57) were primarily responsible for writing it. ³⁰ The content of Section 2.7.4.6, postmarketing data, usually relies heavily on data collected by a pharmacovigilance group, and 56.1% (32/57) received it from that or another responsible group. ³⁰ For the optional section on published literature, no one method predominated, and 89.5% (51/57) closed the document with a de novo high-level summary of its safety data. ³⁰

The integrated summary of effectiveness

Of the 32 who had worked on an integrated summary of effectiveness, 78.1% (25/32) had worked on it and the efficacy summary simultaneously: 65.6% (21/32) as lead writer or substantive contributor to both, and 12.5% (4/32) in that role for one while contributing in a minor way to the other. ^{31, 32} Where the respondent wrote one of the two documents and another writer wrote the other, 78.1% (25/32) developed a plan with that writer to keep the documents aligned. ^{31, 32, 24, 23} Of the 32, 59.4% (19/32) repurposed content from the integrated summary for the efficacy summary, and 84.4% (27/32) were primarily responsible for writing the integrated summary's discussion and conclusions section, which requires de novo writing. ^{31, 32}

The integrated summary of safety

Of the 42 who had worked on an integrated summary of safety, 78.5% (33/42) had worked on it and the safety summary simultaneously: 64.3% (27/42) as lead writer or substantive contributor to both, and 14.3% (6/42) in that role for one while contributing in a minor way to the other. ³³ No formal guidance exists for the integrated summary of safety, unlike the efficacy one, and 57.1% (24/42) were given a template for it, while 42.9% (18/42) developed one de novo from the guidelines in 21 CFR 314.50(vi). ³³ Where another writer wrote the companion document, 83.3% (35/42) developed a plan with that writer to keep the documents aligned. ^{33, 24, 23}

The regulation at 21 CFR 314.50(vi) requires a section on pertinent animal data in the integrated summary of safety. ³³ Of the 42, 47.6% (20/42) included the section and the same share used the electronic capacities of the CTD to cross-refer to another document holding the information, such as Section 2.7.2, the summary of clinical pharmacology, Section 2.4, the nonclinical overview, or Section 2.6, the nonclinical written and tabulated summaries, while 4.8% (2/42) did not include it. ³³ In addition, 57.1% (24/42) repurposed content from the integrated summary for the safety summary, and 76.2% (32/42) were primarily responsible for writing the integrated summary's discussion and conclusions section, which requires de novo writing. ^{33, 24}

State of the study reports when writing starts

Working from finalised clinical study reports is a best practice in medical writing for creating scientifically accurate and regulatory-compliant clinical summaries. ²⁴ It is intended to reduce the inaccuracies, discrepancies and unclear messaging that can occur when the summaries and the study reports are written simultaneously. ²⁴ For most respondents, time constraints and application deadlines often clashed with that practice. ²⁴

Of the 68, 13.2% (9/68) started a summary document from final, approved study reports, and 85.3% (58/68) began before the reports in the programme had been finalised. ^{34, 35} By the state of the reports, 38.2% (26/68) started from near-final reports with stable but unapproved data, 29.4% (20/68) from reports with draft data, 11.8% (8/68) from a shell with the methods section completed and placeholders for the results, and 5.9% (4/68) before the report had been started. ³⁵ They then had to keep the summary consistent with the study reports as these evolved to approval, and aligned with a companion summary that another writer was writing at the same time. ^{24, 36}

Timelines for completing summary documents are often set on the assumption of project managers and upper-level executives that the documents are mostly a mechanical copy and paste from finalised study reports. ²⁴ It follows from that assumption that a summary document, which optimally presents concise, relevant data from the whole programme and key messages on their clinical implications, including an assessment of the drug's benefits and risks, can be written, reviewed and approved quickly. ²⁴

Errors in final tables and listings

Most respondents found errors in the final tables and listings for each of the four summary documents, and every error had the potential to compromise the accurate reporting of study-specific variables. ³⁷ For the summary of clinical efficacy, 75.9% found errors in the number of efficacy parameters reported, 79.6% in patient disposition and 85.2% in other variables such as demographics and primary efficacy measurements. ²⁴ For the integrated summary of effectiveness, the share was 93.8% for each of the three. ²⁴ For the summary of clinical safety, 84.2% found errors in the number of safety parameters reported and 89.5% in other safety parameters such as adverse events, and for the integrated summary of safety the share was 88.1% for each. ²⁴

The errors belie the assumption of project managers and upper-level executives that tables and listings identified as final are flawless; often they are not. ²⁴ Any error in them compromises the scientific accuracy of the application and undermines the reviewers' trust, so writers who identify and correct such errors contribute to the integrity of the submission. ^{24, 38, 36}

The writer's contribution

The results suggest that writers make substantive contributions to clinical summary documents and strategic contributions to the drug submission process. ^{38, 36} Writers create de novo text or intelligently repurpose content for the critical sections of each document: the introductory sections, the sections that present and summarise data, and the discussion and conclusions. ^{38, 36} Such activities belie the notion that a summary document is the result of a simplistic copy and paste. ^{38, 36}

Oncology programmes

An oncology submission dossier often has several interim study reports in addition to the final clinical study report, and multiple data cuts over time, sometimes different across studies, which can be tricky to explain to the reader. ³⁹

During an oncology programme it is not uncommon to have several dose modifications as the investigators manage adverse events and home in on the optimal dose regimen, and early studies can have different dosing regimens from later ones. ⁴⁰ Treatment groups can then be very fragmented, which makes the data very difficult to interpret, particularly in a pooled dataset, because they cannot easily be compared across doses. ⁴⁰ Changes in dosing can also leave the proposed dose with less exposure time than earlier doses. ⁴⁰ These problems affect the interpretation of both efficacy and safety and need careful consideration when planning how to present the data in the dossier. ⁴⁰

Because different cancers affect different organs, there is often little consistency in the adverse events of special interest collected in different studies, which makes summarising them across studies in Module 2.7.4 a challenge. ⁴¹ The choice is between finding a consistent grouping across studies in different cancer types and presenting only the pivotal trial's adverse events of special interest. ⁴¹ In oncology these events are driven by the risk factors of the underlying cancer type, and a large dossier has to find a way to bring very diverse safety data together. ⁴¹ This should be thought about as early as possible when the team begins planning the dossier, and it certainly needs discussing in the statistical analysis plan for the safety summaries. ^{41, 40}

Where submission in other regions, such as Japan, is likely, colleagues from those regions should be consulted while the statistical analysis plans for the efficacy and safety summaries are developed, so that all analyses are planned as their local agencies require or expect. ⁴¹ Otherwise a dossier thought fit for global submission can turn out to need additional analyses run and incorporated into its files. ⁴¹