FDA’s Internal SOP On How To Do An NDA Safety Review Part 2

Sep 26, 2012
Bart Cobert

Pharmacovigilance, Drug Safety and Regulatory Affairs Author & Expert

We are continuing the review of one of the FDA’s critical internal SOPs:  MOPP 6010.3R: Attachment B: Clinical Safety Review of an NDA or BLA (Issued 12/14/2010; Posted 12/14/2010).

We have covered the introductory sections and now are looking at the methodology and other sections.  As in the previous part of this review, all sentences in italics represent Bart’s Corner comments and opinions unless otherwise specified.


The reviewer first needs to assemble all the documents he or she needs:

  • Common technical document (CTD) safety-related sections (i.e., module 2, sections 2.5.5, Overview of Safety, and 2.7.4, Summary of Clinical Safety),
  • The applicant’s Integrated Safety Summary (ISS)
  • AE tables from the NDA/BLA submission.  The FDA notes that “if the reviewer determines that AE tables provided by the applicant are accurate and fairly represent the data they purport to display,” these tables may be used instead of reviewer created tables. Note that FDA does not a priori accept that the sponsor’s tables are accurate or correct; the word “purport” has a negative connotation!
  • Case report forms (CRFs) for subjects who died or had serious adverse events, or who dropped out because of an AE or who dropped out for other reasons [italics from FDA]. Note that the FDA does not necessarily accept that “other reasons” are not safety issues.
  • Individual subject AR data listings, laboratory listings, and baseline listings.
  • Narrative summaries of deaths, SAEs, and other AEs causing dropouts.
  • If available, displays of individual subject safety data over time for subjects with SAEs.
  • The safety sections of the  proposed labeling.


This is a rather extensive list and is used by the reviewer to compile his or her own safety data set.  The company’s data are not necessarily accepted at face value. Make sure that the dataset is complete for FDA or they will come back and ask for more thus delaying the review process.

Reviewing Individual Cases & Causality

The reviewer will look at individual cases of death, SAEs and discontinuations to determine whether:

  1. The event was correctly coded to the MedDRA Preferred Term (PT)
  2. There is a likely explanation for the AE other than the drug
  3. There is another reason that might exclude the drug as the cause of the AE
  4. There was a rechallenge

The FDA then explains why this causality review is done:

“It is important to distinguish the processes described above from the causality analyses of drug-related events often provided by investigators and applicants in NDA/BLA submissions. The analyses of drug-related adverse events presented by applicants are usually based on assessments made by investigators at the time of an event, are highly dependent on information about the side effect profile of the drug available at the time of the clinical trial and…not by awareness of the entire safety database. Generally these analyses are not expected to provide much useful information in assessing causality and should be disregarded.”

This is an important statement!  FDA is saying here that they will disregard the company and investigator causalities since they are done at the time of the AE and usually not based on the whole database available only at the end of the trial!  Keep this in mind when you have long and painful discussions in the company on the causality of a particular case – FDA will disregard your opinion when reviewing the NDA/BLA!

The FDA also notes that for very rare SAEs even a few cases or even one case when none is expected will “represent a serious safety problem for the drug unless it provides unique efficacy or some other advantage over available treatments”.

Another notice from the FDA that very bad and very rare SAEs need to be examined very closely because it won’t take too many of them to prevent NDA approval unless the efficacy is “unique”.

Review of Safety (Section 7)

This section must include a comprehensive discussion of safety findings covering the major safety issues and critical concerns.  It should have a safety issue problem list that may lead to a Risk Evaluation and Mitigation Strategy (REMS), post-marketing studies, inclusion in the labeling etc.  It should also have pertinent negatives (e.g. no Torsades found) and concerns that don’t quite achieve inclusion in the problem list.

The FDA reviewer must then come to the following conclusions:

  • Is the available safety information adequate?
  • What are the limitations of the data?
  • Is additional information needed – both analyses on the current data and new studies?
  • How does the safety data for this drug compare to other available drugs?
  • Is a REMS needed?

Additional things the reviewer must look for:

  • Are the methods used to pool the data adequate and which data are excluded and why. There should be an explanation about why these data were excluded.
  • Problems in coding: compare the company’s codes to the investigators’ verbatim terms. In particular the “assessment should focus on the events leading to dropouts, other changes in treatment, or serious adverse events.” Look for problems of “splitting” which may understate the true incidence of a problem or “lumping” which might hide key AEs.  If problems are found, the reviewer should “recalculate rates using alternative terms.”

FDA understands that coding can hide or minimize (whether intentionally or not) safety issues.

Although the company may supply multiple and elegant analyses, the reviewer is encouraged to take the raw data and calculate his or her own tables and analyses to be sure the safety data are evaluated correctly.  One cannot assume the company is always “correct”.

Adequacy of Safety Assessments

This section assesses whether or not all tests reasonably applicable were conducted to assess the safety of the new drug: sufficient subjects, appropriate demographics, dose, duration of exposure, animal tests, ECGs/QT intervals, metabolic studies, drug-drug interactions etc.  The reviewer must comment on the “quality and completeness of the data provided” and address concerns that still need to be addressed before or after approval.

Class Effects

The reviewer must look at drugs in the same class and look at specific, known events that have been reported with these drugs and which might be expected with the new drug. Examples include: suicidal ideation, hepatotoxicity, QT prolongation, withdrawal effects, neutropenia, bleeding, muscle injury etc.

In addition to reviewing deaths, SAEs, and AEs associated with dropouts, algorithms can be created with combinations of clinical findings that may be a marker for a particular toxicity. These should be done in a blinded fashion to minimize bias.


“Causality judgments are difficult for uncommon serious events (e.g., fewer than 1/1,000) where there are, in most cases, no useful comparisons to control groups…For an event like aplastic anemia, with a background rate of perhaps 1 per million person years, finding even one case suggests a causal relationship.”

“For events that occur more frequently in the absence of drug therapy (e.g., myocardial infarction, stroke, sudden death, seizure, which could occur at rates of 0.1 to 1 percent, depending on the population), the finding of one or two cases may be difficult to interpret in the absence of a substantial controlled trial database. “

“The reviewer should be cautious about dismissing uncommon serious events that don’t seem plausibly drug-related.” (FDA’s italics). There are many examples where these rare SAEs were indeed due to the drug.

The FDA is putting the companies on notice that rare, severe and unexpected SAEs that seem unrelated to the drug and for which there is no obvious pharmacologic or physiologic explanation should still be examined carefully as a possible safety problem. Companies should thus aggressively explore and work up these rare and severe SAEs when first received.  Doing a workup months or years after the fact will be difficult if not impossible.


Obviously, all deaths must be examined.  Particular attention is paid to deaths after the patient leaves the trial if the drug has a long elimination half-life.  All deaths should be considered “without regard to investigator or applicant judgment about causality”.

A distinction should be made between what the FDA calls expected and unexpected deaths.  Expected deaths include those in trials where mortality is an endpoint and the death is likely due to the disease.  Attention should be paid to early deaths in cancer trials as patients are usually chosen who are expected to live for a while.  Such early deaths may signal drug toxicity. Unexpected deaths would include such things as aplastic anemia, progressive multifocal leukoencephalopathy, or acute hepatic necrosis and deserve “detailed individual discussion”.

The FDA again stresses that even though “fatal events may be expected in a population, the reviewer should not… readily accept the conclusion that a fatal event is caused by the underlying disease or an intercurrent illness and not the drug… consider the possibility that the event represents an as yet unsuspected adverse reaction”.

The FDA reviewer will pay particular attention to deaths and, in a certain sense, the drug is not necessarily innocent until proven guilty.  The company should be aware of this viewpoint and analyze the cases accordingly.

In our next posting we will look at how the FDA reviewer analyzes AEs.