What Is The Right Size For A Safety Database?

Sep 10, 2012
Bart Cobert

Pharmacovigilance, Drug Safety and Regulatory Affairs Author & Expert

An often asked question.  The answer depends on whom you ask.

Ask the Chief Financial Officer (money does matter of course) and the likely answer is the smallest number of patients possible particularly before launch and particularly before the drug is on the market and bringing in money from sales.

Ask the Statistician (numbers do matter of course) and the answer will be: enough patients so that we have statistical significance and power and good confidence intervals.  And don’t forget the per protocol and intent to treat analyses. And don’t forget that we’ll lose some patients from drop-outs, protocol violators etc.  So more is better.

Ask the Clinical Research Team (they actually have to find the sites and get the patients enrolled) and the answer will be: enough so we make our timelines and budget but no more.

Ask the Health Agency and they will answer: the appropriate number of course.

Ask the Pharmacovigilance Head and the answer will be: no matter how many patients you enroll it’s not enough.

Ask the CEO just as the NDA is being completed and set to be sent to FDA who learns that there are not enough patients in one of the key trials to reduce the p value from 0.05135 to 0.04974 and her/his answer will be: &%^$#&%!

So how can one approach the issue from the safety point of view?  First we must realize that most NDA/MA programs are arranged to show efficacy to meet whatever clinical and statistical requirements are proposed by FDA and other health agencies.  Rarely if ever is the package done to look for safety as the primary goal of the (key) studies.  These studies and trials are rarely if ever powered to show statistical significance for safety issues.  To put it another way, these phase 2 & 3 trials are not designed to test specified safety hypotheses or measure or identify adverse drug reactions with any pre-specified level of sensitivity.  Most of the “statistics” done on safety data are actually just descriptive with tables and listings but no significance or power calculations done.

So when a drug is approved for marketing the safety database is often relatively small in relation to the number of patients who will be exposed after marketing.  To a large degree this is unavoidable as companies do not have the time, resources and money to study tens of thousands of patients before marketing without some income from sales.  If the barriers to market are too large, the companies will simply not study a particular drug (consistent with their fiduciary responsibility).

There are various suggestions from ICH and health agencies on numbers.  For example ICH E7 talks about studies in geriatric patients (defined, arbitrarily as over 65 years of age) and suggests, for diseases not unique to but present in the elderly, a minimum of 100 patients.  ICH suggests for drugs aimed at treatment longer than 6 months that 1500 subjects be studied in total with 300 to 600 exposed for 6 months and 100 exposed for 1 year using doses in the therapeutic range.

This is a very low number in fact.  If no cases of a particular serious adverse event (SAE) are seen in the 100 patients then all one can say, using the “rule of 3”, is that the rate of this SAE is no worse than 1/33 with a 95% confidence interval.  What this says is that with 100 patients only the most common, high frequency AEs will be picked up.

So what does this translate to in practicality?  Each NDA needs to be customized in terms of sample size for safety.  It will depend on many factors including the novelty of the drug, its class and experience with the class, the condition being treated, the duration of use, the ease and speed of enrollment, the ability to use placebo or controls (standard of care), the ability to do blinded studies etc.  Sometimes there are published guidelines from health agencies, sometimes other companies’ NDAs will give a clue (see the SBOA or EPAR), sometimes there are experts in the field who can advise and, finally, there are meetings with the health agencies to consider the size of the database and the program to be done.

Not all safety studies need be double blind placebo controlled trials, desirable though these may be.  There are situations where uncontrolled studies, adaptive design studies and other creative techniques may help better define the safety profile.

Things have changed in recent years as some agencies (FDA and EU) are approving drugs but with heavy post-marketing safety commitments that the company must do in a set length of time. Most companies (grudgingly) accept this since the alternative would be doing these studies before NDA/MA approval.

Special and individual situations exist for new technology products including biosimilars, nanotechnology, cell derived products, orphan drugs etc.  These products usually involve intense negotiations between the health authorities, the sponsor and outside experts.  A drug for chronic gout refractory to other therapy was approved a year or two ago based on (according to the package insert) two studies of about 85 patients each.  Thus, it appears the entire database was under 200 patients.

The general tendency is towards agencies being risk averse with tougher and tighter requirements.  There is often little gain and large risk for an agency to put its “neck on the line” with a shorter or smaller program.  On the other hand, the agencies really do want novel and breakthrough products to reach market quickly to benefit public health.  To put it another way, database size may also be a political issue.

What this means is that at launch the drug’s safety profile is as good as it’s ever going to be!  As more SAEs are reported and safety studies show new or different problems the safety labeling will get, in the eyes of marketing and sales, worse and worse!

So to conclude, there is no simple formula to determine the appropriate database size.  Most database size considerations are based on efficacy needs, not safety.  So the efficacy will usually be reasonably well established at the time of approval but safety is still largely unexplored.  Only in the post-marketing setting will the safety picture be fleshed out.

So how should companies approach handling their safety data for inclusion in an NDA?  The answer is partly found in…………….FDA’s Manual of Policies and Procedures!  This will be the subject of Bart’s Corner in the upcoming weeks.