February 21, 2011

Letter Encouraging Rejection of Florbetapir F18 Injection (Amyvid)

Janet Woodcock, M.D.
Director
Center for Drug Evaluation and Research
Food and Drug Administration
Department of Health and Human Services
WO51/Room 6133
10903 New Hampshire Avenue
Silver Spring, MD 20993-0002

Dear Dr. Woodcock:

These comments from Public Citizen Health Research Group are being submitted in response to Avid Pharmaceuticals’ New Drug Application (NDA), # 202-008, for florbetapir F18 injection (Amyvid^TM), which was considered by the Food and Drug Administration’s (FDA’s) Peripheral and Central Nervous System Drugs Advisory Committee on January 20, 2011.

(1) We strongly oppose FDA approval of Avid Pharmaceuticals’ NDA for florbetapir F18 injection.

(2) Given the significant inter-reader variability that was seen in the single phase 3 clinical trial evaluating the performance of florbetapir positron emission tomographic (PET) imaging, such PET scans would have little clinical utility in the evaluation of patients presenting with cognitive deficits or early dementia and suspected of having Alzheimer’s disease (AD).

Avid Pharmaceuticals’ proposed indication for florbetapir is:

Florbetapir F 18 Injection is a diagnostic radiopharmaceutical indicated for Positron Emission Tomography (PET) imaging of β-amyloid aggregates in the brain. A negative florbetapir-PET scan is clinically useful in ruling out the presence of pathologically significant levels of β-amyloid in the brain.[1]

The single phase 3 clinical trial evaluating florbetapir-PET (study A07) enrolled two-cohorts: a young, healthy, cognitively intact cohort (“specificity cohort”) and an end-of-life cohort (“autopsy cohort”).[2] The primary efficacy endpoint for the autopsy cohort was the correlation between (a) the amyloid burden in the brain as measured by florbetapir-PET performed within one year prior to death, using a pre-specified semi-quantitative visual rating scale (0-4), and (b) cortical amyloid burden by pathology postmortem using quantitative immunohistochemistry (IHC).[3] The published measurement of the amyloid burden for this correlation was the median rating of three expert independent readers who underwent extensive training, a process that the authors acknowledge “is not likely to be replicated in clinical settings.”[4] Therefore, data on the inter-reader variability for the three independent readers for study A07 is crucial to evaluating the potential clinical utility of florbetapir-PET.

The FDA statistical reviewer’s analysis of the data regarding florbetapir-PET scoring by the individual readers for the 35 subjects in study A07 who underwent autopsy revealed substantial inconsistency among the readers (see table below).[5] This analysis assumed that a semi-quantitative score of 0 or 1 was a negative test and a score of 2 to 4 was a positive test, consistent with the Avid Pharmaceuticals’ approach.

Reader	Sensitivity (%, 95% CI)	Specificity (%, 95% CI)
1	90 (69, 97)	100 (82, 100)
2	55 (28, 79)	100 (86, 100)
3	85 (64, 95)	80 (55, 93)
Median read of all 3 readers	85 (62, 95)	100 (82, 100)

The FDA statistical reviewer also noted the following:

The kappa statistic for the reader agreement evaluation (for the total 147 subjects with 0-4 read) is low to moderate for the autopsy cohort study (0.14 for reader 1 and 2, 0.33 for reader 1 and 3[,] and 0.32 for reader 2 and 3 using simple kappa; 0.54 for reader 1 and 2, 0.7 for reader 1 and 3, 0.68 for reader 2 and 3 using weighted kappa).[6]

The FDA statistical reviewer concluded that:

[T]hese data do not produce evidence of clinical usefulness of this detection since its performance characteristics (sensitivity and specificity) show considerable inconsistency among the readers for the patients from various end-of-life populations. It is not clear if this reader-to-reader variability will increase or decrease in the intended patient population. The specificity results, although consistent across readers, are obtained from the population of young healthy volunteers and again it is not clear if these results will be upheld in the intended patient population. Moreover, the sponsor has proposed using binary, qualitative read of Florbetapir F 18 images which has been applied only to 14 patients in whom pathological standard of truth was available. This sample size is too small to assess the clinical usefulness of the proposed qualitative read.[7] [emphasis added]

The primary FDA medical reviewer’s analysis of the data regarding florbetapir-PET scoring by the individual readers for the 35 subjects in study A07 who underwent autopsy demonstrated that in 10 of 35 subjects (29 percent) there was at least one reader whose rating of global amyloid burden on the florbetapir-PET images differed from that of the other two readers by at least 2 points on the 5-point rating scale. Moreover, in 8 of 20 autopsy subjects (40 percent) who had positive IHC for amyloid, there was at least one reader whose scoring of the florbetapir-PET scans differed from that of the other two readers by 2 or 3 points on the 5-point scale (0-4).[8]

The primary FDA medical reviewer concluded the following regarding the analysis of the autopsy cohort results:

Although the primary endpoint of correlation between median PET read on a semi-quantitative 5-point scale (0-4) and amyloid burden by IHC was met, several issues–including (a) high inter-reader variability, (b) small sample size of 29, (c) undetermined clinical meaning of post-hoc thresholds, (d) wide range of true amyloid burden for some PET ratings, and (e) absence of the population of intended use in the enrolled subjects–cast doubt on the validity, reproducibility, and clinical utility of Amyvid.[9] [emphasis added]

Likewise, the secondary FDA medical reviewer, after conducting a similar analysis, concluded, “This reviewer supports the opinion of the primary clinical reviewer that the efficacy data in this NDA fails to provide convincing evidence to support the efficacy of Amyvid PET for imaging β-amyloid aggregates in the brain.”[10]

On January 20, 2011, the Peripheral and Central Nervous System Drugs Advisory Committee voted to recommend approval for florbetapir-PET scans, provided that (a) the sponsor implemented a training program for physicians that demonstrates accurate diagnosis in the autopsy population; and (b) the sponsor demonstrates reliable consistency in the reading of florbetapir-PET scans in the patient population of the intended application.

Regarding the committee’s first condition of approval, given the alarmingly high inter-reader variability seen in study A07, where rigorous training was provided for the three experienced physicians chosen to read the scans for study A07, it is highly unlikely that an effective training program can be deployed in real-world settings to ensure that the thousands of physicians whose practice includes reading PET scans will achieve consistently reliable interpretation of florbetapir-PET scans in the intended patient population. Regarding the committee’s second condition of approval, conducting another clinical study in the more appropriate intended population that involves comparison of florbetapir-PET scan results to the gold standard test (i.e., measurement of cortical amyloid in the brain by IHC) would take many years to complete because one would need to wait until study subjects died and were available for autopsy.

In conclusion, study A07 implemented florbetapir-PET in a rigorously controlled setting with well-trained readers, using patient populations at two extremes of the spectra for both age and health. However, despite these optimal conditions, the test yielded disparate results when looking at the analysis of the individual readers. If widely deployed in real-world settings, with more variability in reader training and skill and in the patient population for which florbetapir-PET presumably is intended, the performance of the test will be only substantially worse.  For these reasons, FDA, based on current evidence, should not approve florbetapir for the evaluation of patients suspected of having Alzheimer’s disease. It is unlikely that better training and a new study in the more appropriate patient population are realistic hurdles to overcome.

Sincerely,

Michael A. Carome, M.D.
Deputy Director

Sidney M. Wolfe, M.D.
Director
Health Research Group

cc: Dr. Margaret A. Hamburg, Commissioner, FDA
Dr. Russell G. Katz, Director, Division of Neurology Products, Office of New Drugs, Center for Drug Evaluation and Research, FDA