financial disclosure robert nishikawa: –paid consultant for carestream and siemens –shareholder...
TRANSCRIPT
Financial Disclosure
Robert Nishikawa:– paid consultant for Carestream and Siemens
– shareholder in and receives royalties & research funding from Hologic, Inc.
• But I am here representing myself
Guidance Document on CADe
• Where the research is clear, then the document is reasonable
• However, whenever there is uncertainty, the approach is to error on the side of requiring more rather than less
• This is extremely burdensome
• This document if it goes into affect, it could kill new and innovative products in CADe
FDA is in a No-Win Situation
• If requirements are: too lenient
» non-effective products may be marketed
– Money wasted
– Patients may be harmed
too restrictive
» effective products will be delayed
– Patients’ healthcare is compromised
Guidance Document Statement
• “Test data, used once before, does not constitute independent data for testing a CADe device”
• Datasets are extremely costly and time consuming to collect
• This extremely overly burdensome
• This will stifle innovation
Training to the Test Dataset
• Difficult unless the performance on individual cases is known– Only release results of the performance on the
whole set, not individual cases
– Mitigates the example Dr. Gwise gave on overfitting
Proposed Observer Study Protocol
1. a conventional reading without the CADe device (i.e., reader alone)
2. CADe output is displayed immediately after conducting a conventional interpretation [sequential read]
3. a concurrent or simultaneous read
Proposed Observer Study Protocol
• Three reads, when only one is needed, is overly burdensome:– Costs are higher
– Much, much harder to recruit radiologists
– Time to complete study is much longer
• The increase is more than a factor of 3!
Proposed Observer Study Protocol
1. a conventional reading without the CADe device (i.e., reader alone)
2. CADe output is displayed immediately after conducting a conventional interpretation
3. a concurrent or simultaneous read
“Reading scenarios should be consistent with the intended use of the device.”
Proposed Observer Study Protocol
1. a conventional reading without the CADe device (i.e., reader alone)
2. CADe output is displayed immediately after conducting a conventional interpretation
3. a concurrent or simultaneous read
“Reading scenarios should be consistent with the intended use of the device.”
Concurrent Reading
• CADe is labeled as a second reader
• There is a possibility that radiologists will use it as a concurrent reading
• There is no clinical data to show whether this is good or bad
• It is overly burdensome to have manufactures answer a research question in their submission that goes beyond their labeling claims
Proposed Observer Study Protocol
1. a conventional reading without the CADe device (i.e., reader alone)
2. CADe output is displayed immediately after conducting a conventional interpretation
Two studies have shown that a conventional read and the unaided reading in a sequential read are comparable
Proposed Observer Study Protocol
1. CADe output is displayed immediately after conducting a conventional interpretation
Radiology: June 2009
Post-approval study
• Post-approval studies are not needed
• Clinical studies show that CADe can increase sensitivity with a comparable increase in recall rate
• CADe can increase radiologists’ sensitivity (~9%) while keeping PPV1
approximately constant (4.8% vs 4.7%)
Clinical Evaluation Methodology
• Using cancer detection rate in clinical studies that employ historical controls is not viable for measuring the effectiveness of CADe
Radiology: June 2009
Power Calculation
• Dr. Gwise power calculation for se at a fixed sp is not the same as the power required for (se,sp) true operating point