12. Safety evaluation

Section 12 of a clinical study report sets out what the study reveals about a treatment's unwanted effects and how far the evidence supports conclusions about them: how much exposure the study offers, which adverse events occur across the treatment groups, which deaths, serious adverse events and other significant adverse events need individual listings and patient narratives, what the laboratory, vital-sign and other safety measurements reveal, and what the safety conclusions may claim. ICH E3 sets out what each subsection carries and where its supporting tables and listings sit, the CORE Reference manual retitles and renumbers the section, and an anticancer study adapts its order and its reporting of laboratory abnormalities.

What Section 12 asks for

Section 12 examines what the study reveals about treatment risks and how much confidence the evidence supports. ¹

The safety section is the most critical part of a clinical study report, because it presents the risks and safety of the medicinal product, which are crucial to marketing authorisation approval. ² The safety sections and datasets from across the studies eventually form the product safety summary that pharmacovigilance requires. ²

Where Section 11 establishes the evidence of benefit, Section 12 examines the other part of the clinical picture, using the safety measurements planned in Section 9 and the participant history in Section 10. ¹ Its comparisons use the control justified in 9.2, and its findings also depend on the follow-up patterns described in 10.1. ³ The adverse-event definition in Section 2 and the collection methods described in 9.5.1 are the starting point, and Section 12 applies those definitions to the observed findings. ³

Under E3, the analysis of safety-related data can be considered at three levels: exposure to treatment, patterns across groups and important individual cases. ^{4, 1} First, the extent of exposure (dose, duration, number of patients) should be examined to determine the degree to which safety can be assessed from the study. ⁴ Second, the more common adverse events (AEs), laboratory test changes and the like should be identified, classified in some reasonable way, compared across treatment groups and analysed, as appropriate, for factors that may affect their frequency, such as time dependence, demographic characteristics, dose or drug concentration. ⁴ Finally, serious adverse events (SAEs) and other significant adverse events should be identified, usually by close examination of patients who left the study prematurely because of an adverse event, whether or not it was identified as drug related, or who died. ⁴

In all tabulations and analyses, events associated with both the test drug and the control treatment should be displayed. ^{4, 5}

The full integrated report of an individual study should include the most detailed discussion of individual adverse events or laboratory abnormalities, but these should usually be re-examined in an overall safety analysis of all available data in any application. ⁶

Each subsection of Section 12 answers one everyday question:

Everyday question	Where Section 12 answers it
How many people take it, how much, and for how long?	12.1 — Extent of exposure
What health problems occur, and are they more common with the medicine?	12.2 — Adverse events
Does anyone die, become seriously ill, or need treatment changed?	12.3 — Deaths and other important events
Do blood or urine tests reveal problems?	12.4 — Laboratory evaluation
Do measurements or examinations reveal other problems?	12.5 — Vital signs and other observations
What does all this mean for using the medicine?	12.6 — Safety conclusions

³

Summaries, listings and narratives

Section 12 presents its evidence in three forms:

• Summary tables and figures: patterns across groups.
• Patient listings: detailed records, organised by participant.
• Patient narratives: clinical accounts developed in 12.3.2.

^{3, 4}

For example, a table might say that three people develop a particular problem. ³ A listing shows their measurements and dates. ³ A narrative explains the clinical sequence for one of them. ³

E3 asks for important safety data to be presented in summary figures or tables in the text of the report; where these become obtrusive because of their size or number, they go in Section 14, cross-referenced to the text, along with supportive or additional figures, tables or listings. ⁷

Section 12 does not hold every row of the underlying data. ⁸ Its cross-references link its interpretation to the supporting evidence:

Location	Material
14.3.1	Detailed AE displays
14.3.2	Listings of deaths, other SAEs, and other significant AEs
14.3.3	Patient narratives placed outside the main text
14.3.4	Abnormal laboratory-value listings
16.2.5	Relevant compliance/drug-concentration listings
16.2.7	All AEs by participant, including repeated occurrences
16.2.8	Individual laboratory results, as required

⁸

12.1 — Extent of exposure

“Exposure” means receiving the treatment. ⁹ Before interpreting the safety findings, the reader needs three things: number of people, duration, and dose. ^{9, 10}

A study of 20 people treated for two days provides a different safety picture from a study of 2,000 people treated for a year. ⁹

E3 asks for:

• Number exposed: how many participants actually receive treatment, including control treatments.
• Duration: average or median treatment duration, plus numbers treated for meaningful periods.
• Dose: typical doses and numbers receiving different dose levels.
• Dose combined with duration: for example, how many receive the most common dose, the highest dose or the maximum recommended dose for at least a month.
• Relevant subgroups: exposure by age, sex, race as described in E3, disease severity, concurrent illness, and other pertinent characteristics.

^{9, 10}

E3's examples of such periods are one day or less, two days to one week, more than one week to one month, and more than one month to six months. ¹⁰ The daily dose level counted can be each patient's maximum dose, the dose with longest exposure, or the mean daily dose. ¹⁰ Dosage may be given as the actual daily dose or on a mg/kg or mg/m² basis as appropriate, and in some cases cumulative dose might be pertinent. ¹⁰

An average alone can conceal limited longer-term experience. ⁹ For example, “mean exposure is six months” does not show whether nearly everyone receives six months or a few people receive much longer treatment. ⁹

Two concentration measures help interpret individual safety findings: maximum plasma concentration is the highest measured or estimated plasma concentration; area under the concentration–time curve (AUC) combines concentration and time into a measure of drug exposure. ^{9, 10} Where available, the concentration at the time of an event may also help to correlate drug levels with adverse events or laboratory changes in individual patients. ¹⁰

E3 assumes that the safety analysis includes everyone who enters treatment and receives at least one dose, and asks for an explanation of any departure. ^{9, 10} This set of participants is commonly called the safety population. ⁹

12.2 — Adverse events

12.2.1 — Brief summary

The brief summary is the reader’s first overview of the adverse event findings. ¹¹ It describes the overall experience, supported by the detailed tables and analyses that follow. ^{5, 11}

For example:

Adverse events occur in 45% of participants receiving the investigational treatment and 30% receiving placebo. Nausea is the most frequently reported event. ¹¹

Such figures open the discussion but do not yet establish whether events are intense, serious, persistent, or treatment-related. ¹¹

12.2.2 — Display of adverse events

E3 asks for summary tables of all adverse events occurring after study treatment starts, including events likely to reflect the underlying disease or a concomitant illness, unless the regulatory authority has agreed in advance to treat specified events as disease related. ^{12, 13} The tables should also include vital-sign and laboratory changes considered serious or other significant adverse events. ^{12, 13}

The displays should not be filtered down to the events investigators considered drug-related. ^{12, 13} That judgement is one part of the evidence, and later analyses may change the interpretation. ¹³

Treatment-emergent adverse events (TEAEs) are findings absent at baseline or worsening after treatment starts, within the analysis-defined observation window. ^{14, 13} E3 describes this concept as treatment-emergent signs and symptoms, or TESS. ^{12, 13} E3 says that in most cases it will be useful to identify TESS in the adverse event tables. ¹²

A pre-existing mild headache that worsens during treatment may therefore be treatment-emergent. ¹³ “Treatment-emergent” describes timing and change; it does not establish causation. ¹³

A simple table might look like this:

Participants with at least one occurrence	Investigational treatment, N=200	Placebo, N=200
Nausea	30 (15%)	10 (5%)
Headache	20 (10%)	18 (9%)

¹³

These tables usually count participants, not episodes. ¹³ A person with five nausea episodes contributes one participant to the nausea row and can also contribute to the headache row, so event rows cannot simply be added to give the number of people with any AE. ¹³ E3 asks for the summary tables to list each adverse event with the number of patients in each treatment group in whom it occurred and the rate of occurrence. ¹²

E3 also calls for:

• Grouping events by body system.
• Severity categories, where used.
• Causality categories, where used, while retaining all events.
• A way to identify the participants behind the detailed results.
• Complete tables in 14.3.1, plus a shorter comparison of common events in the main text.
• Separate displays by treatment cycle when helpful, such as in chemotherapy.

^{13, 12}

Severity categories might be mild, moderate and severe. ¹² Causality might be split into events considered at least possibly related to drug use and those considered not related, or follow another scheme, such as unrelated or possibly, probably or definitely related. ¹² Subsequent analyses of the study or of the overall safety database may help to distinguish between events that are, or are not, drug related. ¹²

E3's example of the complete table gives, for one treatment group, the number and percentage of patients with each event under each severity category split into related and not related, with the identification number of each patient and totals for related and not-related events. ¹² The summary table in the body of the report compares treatment and control groups without the patient identification numbers. ¹²

The 2012 questions and answers document on E3 places the example table of 12.2.2 in Section 14, not in the body of the report, and says the in-text table should compare relatively common adverse events across treatment and control groups. ¹⁵

E3’s threshold for a common event, at least 1% of the treated group, is given as an example, not a universal cutoff. ^{13, 12} Rare important events still need attention elsewhere. ¹³

Different investigators may describe the same medical problem differently. ¹³ E3 therefore asks for the investigator's original terms to be displayed and for events that probably represent the same phenomenon to be grouped, so that the true occurrence rate is not obscured. ^{13, 12} One way to group them is a standard adverse reaction or event dictionary. ^{13, 12}

MedDRA, the Medical Dictionary for Regulatory Activities, organises reported events through several levels:

• Verbatim/reported term: the investigator’s original wording.
• Preferred term (PT): a standardised medical term used in summaries.
• System organ class (SOC): a broader grouping.

¹³

The hierarchy includes further levels, and its purpose is to avoid splitting one medical phenomenon into many differently worded categories. ¹³

12.2.3 — Analysis of adverse events

The analysis asks which patterns need an explanation. ¹⁶

The basic display of adverse event rates in 12.2.2 should be used to compare rates in treatment and control groups, and combining the severity and causality categories may give a simpler side-by-side comparison. ¹⁷

E3 does not intend every adverse event to undergo rigorous statistical evaluation. ^{16, 17} When a study is small and has relatively few events, it may be sufficient to limit the analyses to a comparison of treatment and control. ¹⁷ More extensive investigation may be better suited to an integrated safety analysis, combining relevant studies. ¹⁶

Treatment groups are compared first. ^{16, 17} Then, where the study supports it, the analysis examines relationships with:

• Dose, dose adjusted for body size, dosing schedule, and cumulative dose.
• Treatment duration and drug concentration.
• Age, sex, race, and other baseline characteristics.
• Renal function and other relevant baseline characteristics.
• Efficacy outcomes.
• Time to onset and duration of events.
• Treatment cycles, where relevant.

^{16, 17}

E3 says that, if study size and design permit, it may be useful to examine the more common adverse events that seem to be drug related for such relationships, and that this is usually best done in an integrated analysis of safety. ¹⁷ A variety of further analyses may be suggested by the study results or by the pharmacology of the test drug. ¹⁷ Initial display and inspection of the data may already make it apparent that no significant relation to demographic or other baseline features is present. ¹⁷

A pattern such as more nausea at higher doses may support a treatment relationship, but it is still evidence to assess alongside alternative explanations. ¹⁶

Exposure and the timing of onset can change the interpretation of a difference between groups. ¹⁶

Unequal observation time. If one group receives treatment for a year and another for three months, the first group has more opportunity to experience events. ¹⁶ A simple percentage—often called crude incidence—does not account for that difference. ¹⁶

Time-based analyses. E3 says that under certain circumstances life-table or similar analyses may be more informative than crude adverse event rates; such analyses apply the time-to-event methods introduced in 9.7.1 to safety outcomes. ^{16, 17} Exposure-adjusted incidence rates are expressed using accumulated treatment time, such as patient-years; they are rates, not percentages of participants, and their numerator and time-at-risk definitions must be clear. ¹⁶

12.2.4 — Listing adverse events by patient

The summary table compresses the data, and the listing by patient in Appendix 16.2.7 keeps the detail, including repeated occurrences of the same event. ^{18, 19} E3 asks for the listing to be organised by investigator and treatment group and to give:

Information	What it lets the reviewer understand
Participant identifier; age, race, sex, weight, and height if relevant	Who experienced the event and relevant characteristics
Location of case report forms, if supplied	Where supporting recorded information can be found
Preferred and original event terms	Standard classification and original description
Onset date or discovery visit; event duration	When it happened and how long it lasted
Severity and seriousness	Intensity and important medical consequences
Action taken and outcome	The intervention and subsequent course
Causality assessment and its method	Whether treatment is suspected and the basis for that assessment
Timing relative to the last dose	Temporal relationship to treatment
Current or most recent treatment; dose; concentration if known	Exposure around the event
Duration of treatment	Treatment duration before the event
Concomitant treatment	Other treatments that might affect interpretation

^{19, 18}

E3's examples for the entries are mild, moderate or severe for severity; serious or non-serious for seriousness; none, dose reduced, treatment stopped or specific treatment instituted for action taken; and related or not related for causality, with how causality was determined described in the table or elsewhere. ¹⁹ The dose is given in absolute amount, mg/kg or mg/m² at the time of the event. ¹⁹

The listing connects the concomitant treatments and case report forms introduced in Section 9 to each event. ¹⁸ For reporting the outcome, E3 gives as an example the format of the Council for International Organizations of Medical Sciences (CIOMS). ^{18, 19}

Codes and abbreviations should be clearly explained, at the beginning of the listing or, preferably, on each page, so the listing can be interpreted. ^{19, 18}

12.3 — Deaths, serious events, and other significant events

Section 12.3 gives individual attention to findings whose importance may be lost in broad summaries. ^{20, 21}

Severity, causality, seriousness, action taken and outcome describe separate features of an event. ³

Serious and severe mean different things. ²¹ A serious adverse event meets criteria involving death, life threat, inpatient hospitalisation or its prolongation, persistent or significant disability/incapacity, or congenital anomaly/birth defect. ^{22, 4, 21} Medical judgement can make an event serious without those outcomes: an important medical event that may not be immediately life-threatening or result in death or hospitalisation, but that may jeopardise the participant or may require intervention to prevent serious outcomes, should generally be considered serious. ^{22, 21}

“Life-threatening” means the person is at risk of death at the time of the event, not that a more intense version could hypothetically have been fatal. ²¹ A severe headache can be non-serious; an event of moderate intensity can be serious because it requires inpatient hospitalisation. ²¹

E3’s other significant adverse events include marked laboratory abnormalities and events prompting intervention, such as stopping treatment, reducing the dose, or adding significant concomitant treatment. ^{21, 4}

In this term, “significant” concerns medical importance or consequences, not statistical significance. ²¹

12.3.1 — The three focused listings

The three listings give the same patient-level information as the listing in 12.2.4 and appear in 14.3.2. ^{23, 24}

• 12.3.1.1 — Deaths: include deaths during the study and post-treatment follow-up, plus deaths resulting from a process that began during the study. ²⁵
• 12.3.1.2 — Other SAEs: include serious events other than death, including those preceding or associated with deaths. Serious laboratory, vital-sign, and physical-examination abnormalities are included. ²⁶
• 12.3.1.3 — Other significant AEs: include marked non-serious laboratory abnormalities and non-serious events causing the specified interventions. ²⁷

²⁴

A death does not make the preceding illness disappear from the safety record. ²⁴ The reviewer needs to understand the clinical events leading to the outcome. ²⁴

Under the 2012 questions and answers document, any death should be included in 12.3.1.1, and it will also appear in 12.3.1.2 as an "other serious adverse event" with a fatal outcome. ¹⁵

12.3.2 — Patient narratives

A patient narrative, or safety narrative, tells the story of what happened in an individual case, assembling its important facts into a readable clinical sequence. ^{28, 29}

E3 asks for brief narratives for deaths, other SAEs, and selected other significant events of particular clinical importance. ^{30, 28, 31, 32} Safety narratives are a required, key part of the clinical study report. ^{29, 33} They describe deaths, serious adverse events such as fatal and life-threatening ones, and adverse events of special clinical interest, which the sponsor or regulator usually defines in advance. ^{29, 33} Depending on their number, the narratives can appear in the main text or 14.3.3. ^{30, 28}

E3 allows clearly unrelated events to be omitted from narratives or described very briefly; this does not remove the broader listing requirements. ^{30, 28} Some categories of narrative may also be waived by prior agreement with the regulatory body, for example deaths where death is a study endpoint. ³²

Regulatory agencies require a brief narrative of serious adverse events in the clinical study report, and their assessment of these events may affect drug approval and labelling. ³⁴ A high-quality narrative is essential: it describes the event accurately and makes clear how it progressed and was managed, which helps the assessment of its relevance and criticality. ³⁴

E3 sets out the elements a narrative should provide:

• Participant identifier, age, sex, and relevant general condition.
• Disease being treated and its duration, where needed.
• Relevant previous or concurrent illnesses.
• Relevant previous or concurrent medicines and their doses.
• Study treatment, dose where it varies, and treatment duration.
• The event’s nature, intensity, and clinical course.
• Timing in relation to treatment.
• Relevant laboratory findings.
• Whether and when treatment stops.
• Interventions and subsequent course.
• Post-mortem findings, where applicable.
• Investigator causality assessment and sponsor assessment, where appropriate.

^{30, 28, 35, 29}

For example, the statement “a participant develops a liver abnormality” leaves many questions unanswered. ²⁸ The narrative connects baseline tests, treatment dates, the first abnormal result, symptoms, other medicines, treatment withdrawal, and recovery. ²⁸

12.3.3 — Analysis and discussion

After the cases are presented, their implications are assessed:

• Is there a pattern across participants?
• Could an event represent a previously unsuspected important adverse effect?
• Do events lead to withdrawal, dose reduction, or additional treatment?
• Does risk appear to change with time on treatment?

^{36, 37}

For especially important SAEs, time-to-event analyses may help. ^{36, 37}

An adverse event of special interest (AESI) is an event selected for closer attention because of a particular concern about the medicine, drug class, or patient population. ³⁶ It may be serious or non-serious. ³⁶

AESI and E3’s “other significant AE” are not synonyms. ³⁶ An AESI can overlap with serious events or events causing discontinuation. ³⁶ Standardised MedDRA queries (SMQs) group related terms to investigate a medical concern that could otherwise be scattered across several categories. ³⁶

12.4 — Clinical laboratory evaluation

Blood and urine tests can reveal changes before a participant notices symptoms. ³⁸

Laboratory problems can be missed if the writer relies only on recorded AE terms, so the objective laboratory results need their own evaluation. ³⁹

Common categories include:

• Haematology: blood cells, such as haemoglobin, white blood cells, and platelets.
• Blood chemistry: substances and enzymes that help assess organs and metabolism.
• Electrolytes: substances such as sodium and potassium.
• Urinalysis: examination of urine.

³⁸

A laboratory parameter is one measured item. ³⁸

Three concepts are kept apart:

Concept	Meaning
Outside the reference range	Higher or lower than the laboratory’s stated range
Change from baseline	Different from that person’s starting value
Clinically significant abnormality	A finding judged important medically under defined criteria

³⁸

A change can be important even if the final result remains inside the reference range. ³⁸ Conversely, a small result outside the range may not be medically important. ³⁸

12.4.1 — Individual measurements and abnormal-value listings

E3 describes listings with each row representing a participant visit and columns showing relevant demographics, dose, and test results. ^{40, 41} Tests should be grouped logically and abnormal values clearly flagged. ^{40, 41}

E3 distinguishes two listings:

• All safety laboratory results: listings in 16.2.8, submitted or available as required by the authority.
• All abnormal laboratory values: a participant-level listing in 14.3.4.

^{40, 41}

In the listing of all laboratory results, patients are grouped by investigator, if there is more than one, and by treatment group. ⁴¹ Tests should be grouped as haematological tests, liver chemistries, electrolytes, urinalysis and so on, because not all of them fit in one table, and abnormal values should be identified, for example by underlining or bracketing. ⁴¹ The by-patient listing of all abnormal values in 14.3.4 should be provided for all regulatory authorities. ⁴¹

For abnormalities of particular interest, normal results before and after the finding and related tests can supply essential context. ^{40, 41}

Some minor abnormalities may be excluded from further analysis with a clear explanation. ^{40, 41} They should still be identifiable in the complete results supplied or available to authorities. ^{40, 41} E3's examples of abnormalities that can probably be defined as clinically insignificant and excluded are single, non-replicated, small abnormalities of some tests, such as uric acid or electrolytes, and occasional low values of others, such as transaminase, alkaline phosphatase or BUN. ⁴¹

E3’s example table uses older names: SGOT corresponds to AST, SGPT to ALT, and AP to alkaline phosphatase—enzymes considered in laboratory assessment. ^{40, 41}

12.4.2 — Evaluation of each laboratory parameter

E3 asks for laboratory data to be examined from three perspectives, giving reference ranges for each analysis. ^{42, 43} The evaluation needed depends in part on the results seen, and in each analysis treatment and control groups should be compared as appropriate and as far as study size allows. ^{43, 42}

12.4.2.1 — Values over time

For each parameter at each visit or other time point, the evaluation should describe:

• Mean or median.
• Range of observed values.
• Number with abnormal results.
• Number exceeding selected thresholds, with the choices explained.

^{44, 45}

ULN means upper limit of normal. ⁴⁵ If a test’s ULN is 40 units/L, a result of 80 is 2×ULN. ⁴⁵ E3's example thresholds are twice and five times the upper limit of normal. ⁴⁴ These thresholds are ways to summarise findings, not universal clinical decision rules. ⁴⁵

Graphs can show whether a group’s values rise, fall, or recover over time. ^{45, 44}

12.4.2.2 — Individual patient changes

A stable group average can conceal substantial changes in a few people. ⁴⁶ The approaches E3 offers include:

1. Shift tables: classify participants as low, normal, or high at baseline and at a later time.
2. Change-threshold tables: count changes exceeding a specified amount, even if the final result remains within the reference range.
3. Baseline-versus-treatment plots: plot each participant’s starting value against a later or most extreme value.

^{46, 47}

E3 asks for the analysis of individual patient changes to be given by treatment group. ⁴⁷

For a change-threshold table, E3's example is BUN, where a change of more than 10 mg/dL might be chosen, with the number of patients above and below it shown for one or more visits and patients usually grouped by whether their baseline BUN is normal or elevated. ⁴⁷

A fictional shift table:

Baseline → Week 8	Low	Normal	High
Low	2	3	0
Normal	1	80	10
High	0	2	2

⁴⁶

The 10 normal-to-high shifts deserve examination, but the table does not by itself establish that all ten are clinically important. ⁴⁶

In the plots, baseline goes on the horizontal axis—E3 calls it the abscissa—and the treatment value on the vertical axis, the ordinate. ^{46, 47} Points on the diagonal show no change; points above it show increases. ^{46, 47} Extreme points identify outliers, whose participant identifiers help further review. ^{46, 47} Because such a plot usually shows a single time point for a single treatment, interpreting it requires a time series of plots for the treatment and control groups. ⁴⁷

12.4.2.3 — Individual clinically significant abnormalities

Medically important changes are discussed, including:

• Applicable definitions and grading criteria.
• Discontinuations caused by laboratory findings.
• Severe graded abnormalities, whether or not serious.
• Narratives where the finding qualifies for one.
• Evidence about a treatment relationship.

^{48, 39}

The applicant defines which changes are clinically significant. ⁴⁸ The analysis of those changes, with a recapitulation of discontinuations due to laboratory measurements, should be given for each parameter. ⁴⁸ A narrative should be given, under 12.3.2 or 14.3.3, for each patient whose laboratory abnormality was considered a serious adverse event and, in certain cases, an other significant adverse event. ⁴⁸ Where toxicity grading scales such as WHO or NCI are used, changes graded as severe should be discussed regardless of seriousness. ⁴⁸

Two terms name what happens when treatment is withdrawn and restarted:

• Positive dechallenge: improvement after treatment is withdrawn.
• Positive rechallenge: recurrence after treatment is restarted.

^{39, 48}

These observations can support causality, alongside dose relationships, concentration relationships, other medicines, or resolution despite continued treatment. ^{39, 48} “Rechallenge” here describes an observed sequence; it is not an instruction to restart a medicine to test causality. ³⁹

12.5 — Vital signs, physical findings, and other safety observations

Section 12.5 covers safety evidence outside laboratory tests:

• Vital signs: for example, pulse, blood pressure, temperature, and respiratory rate.
• Physical findings: observations from examinations.
• Other safety assessments: study-relevant measurements, potentially including ECGs.

⁴⁹

E3 asks for an approach similar to laboratory evaluation: group patterns, changes over time, individual changes, and clinically meaningful abnormalities. ^{50, 49}

If a treatment effect appears, its relationships with dose, concentration, disease, demographics and concomitant treatments should be identified. ^{49, 50} The clinical relevance of the observation should also be described. ⁵⁰

Lower blood pressure may be an intended efficacy outcome in a hypertension trial, while an excessive reduction causing symptoms may also be a safety concern. ⁴⁹ E3 asks for particular attention to changes not assessed as efficacy variables and to changes considered AEs. ^{50, 49}

12.6 — Safety conclusions

The safety conclusions bring the findings together, with particular attention to:

• Deaths and SAEs.
• Events causing withdrawal.
• Events requiring dose changes or additional medication.
• Participants or groups appearing at increased risk.
• Vulnerable groups represented by small numbers.
• Implications for possible uses of the medicine.

^{51, 52}

E3 gives examples of vulnerable groups: children, pregnant women, frail older people, and people with markedly altered drug metabolism or excretion. ^{51, 52}

Metabolism is chemical processing; excretion is removal from the body. ⁵¹ Altered drug handling can affect exposure and risk. ⁵¹

A useful conclusion describes the observed findings and their limits. ⁵¹ For example:

Nausea is more frequent with the investigational treatment and leads to discontinuation in four participants. The limited number of older participants restricts conclusions about safety in that population.

⁵¹

“No events occur in this study” and “the risk does not exist” are different claims: rare effects may not appear in an individual trial, and broader safety assessment uses evidence across studies and from subsequent use. ⁵¹

The safety section in CORE Reference

The CORE Reference manual reads E3 as not entirely clear about what belongs in each subsection of Section 12, and answers with adapted section titles and more explicit instruction about their content, including that both participants and events should be summarised. ⁵³

The manual moves extent of exposure out of Section 12 into Section 10, as 10.6, because extent of exposure relates to all participants, not only those with safety data. ⁵⁴ Adverse events become 12.1, with a brief summary at 12.1.1, the most frequently reported events at 12.1.2 and categorisation of all adverse events at 12.1.3. ⁵⁵

The manual also finds E3 unclear about what belongs in its display of adverse events as against its analysis of them. ⁵⁵ The manual's categorisation subsection summarises all treatment-emergent adverse events by relationship or causality and by severity, and adds any other presentations thought relevant. ⁵⁵ This allowance gives flexibility, for example to align the report's adverse-event presentations with the EU or ClinicalTrials.gov results-posting summaries, such as by separating adverse events from serious adverse events in this subsection. ⁵⁵ The manual also removes from the main report text the description of the elements of patient listings, E3's listing of adverse events by patient. ⁵⁵

The manual advises against using significant for anything other than statistics, and so proposes renaming E3's other significant adverse events as clinically meaningful adverse events. ⁵⁶

Section 12 in an anticancer study

Order and contents of the safety evaluation

The safety evaluation of an anticancer study is one section where ICH E3 needs to be followed more flexibly. ⁵⁷ In a phase I trial of an anticancer agent, the main aim is to find the maximum tolerated dose and the recommended dose for phase II from a dose-escalation design that involves reporting dose-limiting toxicities, so the reader of the report expects to find those results first, but under E3 they sit in Section 12, after the efficacy results. ⁵⁷ The main endpoints of such a trial are all safety variables. ⁵⁷ Because the cohorts themselves are defined by the toxicity found at each dose increment, describing efficacy first forces many cross-references to the safety section to explain the rationale for the cohort distribution. ⁵⁷

A more logical, reader-friendly order is to reverse and renumber the two sections, presenting the safety evaluation as Section 11 and the efficacy evaluation as Section 12, and it is then useful to add a note at the beginning of the report along these lines: that the report has been written according to ICH E3 but Sections 11 and 12 have been reversed and renumbered in accordance with the primary and secondary objectives of this phase I trial. ⁵⁷ By 2012 this structure had been used in several reports submitted to the European Medicines Agency in marketing authorisation applications and had been accepted by the agency in validation. ⁵⁷ Electronic submission with a hyperlinked table of contents may reduce the concern about efficacy appearing first, but reversing the sections remains a more logical and reader-friendly order. ⁵⁷

A phase I oncology report also needs a section that E3 does not define, giving the results for the dose-limiting toxicities, the maximum tolerated dose and the recommended dose. ⁵⁷ A model table of contents for its safety evaluation runs:

1. Extent of exposure, broken down into cycles received, dose delays and dose reductions.
2. Maximum tolerated dose and recommended dose, organised by dose level and closing with the dose-limiting toxicities, the maximum tolerated dose and the recommended dose for phase II.
3. Adverse events, with the E3 subsections and an analysis broken out by event category, such as constitutional events and pain, or gastrointestinal events.
4. Deaths, other serious adverse events and other significant adverse events.
5. Clinical laboratory evaluation, with separate subsections for haematological abnormalities, biochemical abnormalities and individual clinically significant laboratory abnormalities.
6. Vital signs, physical findings and other safety observations.
7. Safety conclusions.

^{57, 58}

Adverse events in a fatal disease

The severity of cancer and of its treatment produces high rates of adverse events, serious adverse events and deaths. ⁵⁹ Differences between groups can become exaggerated when an effective test drug keeps patients on treatment longer, while for an established safety issue the difference in frequency may be obvious. ⁵⁹ In-text presentations should be kept manageable by leaving out the less common events that occur at similar frequencies across groups, which should still be available in the end-of-text tables. ⁵⁹ The text commonly focuses on related events, or on events of grade 3 or 4 by the National Cancer Institute criteria, rather than on overall frequencies. ⁵⁹ Whether relatedness is attributed to individual components of a combination or to the treatment in general depends on the study design, and that choice determines the best way to present the related events. ⁵⁹

Narratives

Narratives are required for clinical study reports but are often not adequately planned in advance, so criteria for writing them should be agreed with the study team as early as possible and, where necessary, explained in the report. ⁵⁹ There is no regulatory requirement to write narratives for serious adverse events or deaths clearly unrelated to the product, and unless the criteria are set in advance the writer risks being asked to produce large numbers of narratives for patients dying, entirely expectedly, of their underlying disease. ⁵⁹ One consistent policy should apply to all reports for a given product and indication. ⁵⁹

Laboratory abnormalities

Chemotherapy works by destroying rapidly growing cancer cells, but it also affects normal cells that grow rapidly, such as blood cells forming in the bone marrow and cells in the hair follicles, the mouth and the intestines. ⁵⁸ During chemotherapy a great deal hinges on the blood test results that precede each infusion: low counts can indicate serious effects including fatigue, bruising and vulnerability to infection, and can mean treatment must be postponed until values recover or the dose reduced in subsequent cycles. ⁵⁸ Laboratory findings are for that reason of extreme relevance in oncology studies, and haematological and biochemical abnormalities have to be discussed in separate subsections of the clinical laboratory evaluation. ⁵⁸

These findings are called abnormalities rather than toxicities, because patients often have asymptomatic increases or decreases in parameters such as neutrophils or transaminases. ⁵⁸ Adverse events belong in the adverse-event section, usually tabulated as the worst grade of toxicity per patient and per cycle and graded with the National Cancer Institute's Common Terminology Criteria for Adverse Events, at version 4.0 in 2012. ⁵⁸ But an adverse event that can be measured as a laboratory abnormality — neutropenia, thrombocytopenia, transaminase increases, hyperbilirubinaemia — may be under-reported if it is only listed from the adverse-event rows of the case report form, because that reporting depends entirely on the investigator's judgement and on whether the laboratory event was symptomatic. ^{58, 57} Such abnormalities are better reported from the objective laboratory results, graded with the same criteria, and should be excluded from the adverse-event tables and shown and discussed only in the clinical laboratory evaluation. ⁵⁸ The exception is the symptomatic adverse event arising from a laboratory abnormality — febrile neutropenia is the example — that results in treatment modification, discontinuation, a serious adverse event or death: that should be described in detail in the section that covers it. ⁵⁸ This oncology-specific recommendation should not be generalised into permission to omit reportable adverse events. ³⁹

Laboratory data in this population can be hard to interpret, because the underlying disease and its co-morbidities can cause wide variation in several parameters. ⁵⁹ Individual frequencies of abnormalities, and shifts by severity, are more informative than mean or median values, and any difference between groups in time on study, or any established toxicity of the trial drug, should be considered as a possible explanation for what the laboratory data show. ⁵⁹