b***8
发帖数: 60 | 1 In biomarker studies, the ultimate goal is to screen dozens, or
hundreds, or potentially hundreds of thousands of potential predictors
using explanatory methods to understand the existing data and then
ultimately identify a factor or set of factors that could be
used in the prediction of new observations with high accuracy. While
the concepts of "association testing" and "classification" are not
unrelated in this scenario, they are inherently separate objectives
requiring the implementation of different approaches.
Consider the trial design of XXX, a phase 3 randomized, double-blind,
placebo-controlled, multi-center trial to study efficacy of MAGIC DRUG
in schizophrenia patients that were followed for 6 weeks. The primary
endpoint of interest is response / non-response as defined by a pre-
specified threshold on the response scale at 6 weeks. In
XXX, there were 400 subjects randomized 3:1 to MAGIC DRUG and placebo
with treatment response rates of 30% and 5%, respectively. For a
predictive model to have clinical utility and marketable advantage,
the client requires at least 40% of the patient population be
classified as responders to MAGIC DRUG with a response rate ≥ 60%.
Consider you have the following baseline demographic and clinical
characteristics that are relatively easy to obtain for each patient
compared to obtaining biomarker data: baseline value on the continuous
response scale, age, gender, age of onset of disease, and country of
origin. Then, consider you have 12 potential biomarkers, each have
data captured on the continuous scale between 0 and 100.
Questions
(1) In the set up for your analysis, describe the difference between
association testing and classification and set client expectations for
the study at hand by outlining the relationship between these two
concepts in easy-to-understand terms or graphics.
(2) Provide the client with a better understanding of how the
available independent variables explain inter-individual variability
in response. Your response should include a description of the
statistical methodology you intend to use to accomplish this
objective.
(3) Provide the client with a strategy for how you would approach
evaluating the available data with the ultimate goal of developing and
optimizing a classifier (or final predictive modeling framework) that
can be carried forward into a future study for confirmation. Your
response should include a description of the statistical methodology
you intend to use to accomplish this objective and how you intend to
evaluate the performance of your classifier. | c*****1
发帖数: 131 | 2 好长，有么有摘要？
【在 b***8 的大作中提到】
: In biomarker studies, the ultimate goal is to screen dozens, or
: hundreds, or potentially hundreds of thousands of potential predictors
: using explanatory methods to understand the existing data and then
: ultimately identify a factor or set of factors that could be
: used in the prediction of new observations with high accuracy. While
: the concepts of "association testing" and "classification" are not
: unrelated in this scenario, they are inherently separate objectives
: requiring the implementation of different approaches.
: Consider the trial design of XXX, a phase 3 randomized, double-blind,
: placebo-controlled, multi-center trial to study efficacy of MAGIC DRUG
| k*******a
发帖数: 772 | 3 等牛人来解答，好好学习一下
【在 b***8 的大作中提到】
: In biomarker studies, the ultimate goal is to screen dozens, or
: hundreds, or potentially hundreds of thousands of potential predictors
: using explanatory methods to understand the existing data and then
: ultimately identify a factor or set of factors that could be
: used in the prediction of new observations with high accuracy. While
: the concepts of "association testing" and "classification" are not
: unrelated in this scenario, they are inherently separate objectives
: requiring the implementation of different approaches.
: Consider the trial design of XXX, a phase 3 randomized, double-blind,
: placebo-controlled, multi-center trial to study efficacy of MAGIC DRUG
| f******d
发帖数: 537 | 4 面试题 or take-home final and you're looking for answers?
The questions are very relevant, as this is what pharma/biotech companies
have been after all these years....
【在 b***8 的大作中提到】
: In biomarker studies, the ultimate goal is to screen dozens, or
: hundreds, or potentially hundreds of thousands of potential predictors
: using explanatory methods to understand the existing data and then
: ultimately identify a factor or set of factors that could be
: used in the prediction of new observations with high accuracy. While
: the concepts of "association testing" and "classification" are not
: unrelated in this scenario, they are inherently separate objectives
: requiring the implementation of different approaches.
: Consider the trial design of XXX, a phase 3 randomized, double-blind,
: placebo-controlled, multi-center trial to study efficacy of MAGIC DRUG
| b***8
发帖数: 60 | 5 if this is a take-home exam, how can it be so relevant to real
industry?
they are coming from a industry interview from a colleague and it was
said to be an open problem. Post here just for those who are interested
in industry to have a feeling of what is going on there and solicit
more discussion on this topic
companies
【在 f******d 的大作中提到】
: 面试题 or take-home final and you're looking for answers?
: The questions are very relevant, as this is what pharma/biotech companies
: have been after all these years....
| f******d
发帖数: 537 | 6 Well, academics are doing these too, not only in industry. Anyway, putting
this aside, it's rare to see this kind of rather comprehensive question
during an interview. Were these asked by the interviewer or it was a paper
exam he had to take during the interview?
【在 b***8 的大作中提到】
: if this is a take-home exam, how can it be so relevant to real
: industry?
: they are coming from a industry interview from a colleague and it was
: said to be an open problem. Post here just for those who are interested
: in industry to have a feeling of what is going on there and solicit
: more discussion on this topic
:
: companies
|
|
