Discussion: View Thread

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Andy Hi!

I would recommend pooling answers to these questions from the clinicians on the study. In my experience, clinically relevant metrics are more important than the optimal. And most experienced clinicians have some kind of rule(s) of thumb about these kinds of things which you can analyze using a competing risks framework.

Cheers,

AB

------------------------------
Adam Batten
Lead Statistician & President
AB EVERGREEN ANALYTICS LLC

Original Message:
Sent: 05-05-2024 10:05
From: Thomas Sandry
Subject: Optimal age of surveillance

------------------------------
Thomas Sandry

Original Message:
Sent: 05-04-2024 22:30
From: Mart Andrew Maravillas
Subject: Optimal age of surveillance

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

------------------------------
Mart Andrew Maravillas
Biostatistician/Data Scientist
University Hospitals Cleveland Medical Center
------------------------------

Andy Hi!

I would recommend pooling answers to these questions from the clinicians on the study. In my experience, clinically relevant metrics are more important than the optimal. And most experienced clinicians have some kind of rule(s) of thumb about these kinds of things which you can analyze using a competing risks framework.

Cheers,

AB

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Hi Andy,

I am a member of the Statistical Consulting Section with the clinical background that should be helpful for answering your questions. Two days ago, I already sent you a detailed answer, but it never showed up on this forum. I understand that I am not the only one who had such problems on the Consulting forum. My apologies for asking you a question before you can get my answer to your questions, but could you kindly confirm whether you received this message ? I am attaching my e-mail to facilitate communication in case the misconfigured path of my posts persists.

Thank you and best regards, 

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Dear Gerhart, 

As to a post not appearing. Please contact Rick Peterson at ASA (rick at amstat.org). Rick does not make the fix. he can forward to the appropriate technical support people. And sometimes its a relatively simple matter of clearing your browser cache. Confirming, yes, from time to time posts do not appear - and at least one time it was due to a software bug. While I was section chair, I had that problem and discovered a bug. For a brief period I had the  section chair elect post on my behalf. Typically you should see your post in the community and you should also get an email copy. -chris

Hi Andy,

I am a member of the Statistical Consulting Section with the clinical background that should be helpful for answering your questions. Two days ago, I already sent you a detailed answer, but it never showed up on this forum. I understand that I am not the only one who had such problems on the Consulting forum. My apologies for asking you a question before you can get my answer to your questions, but could you kindly confirm whether you received this message ? I am attaching my e-mail to facilitate communication in case the misconfigured path of my posts persists.

Thank you and best regards, 

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Hi Andy,

Encouraged by Chris Barker's response, I am sending you the updated version 2 of my post.

Before the statistics discussion, let me bring up a few clinical considerations that are important for the stratification of the patients:

(1) How old are the patients at time of detection of the cyst – wide age range or narrowly defined age range? Is mostly one gender affected? Are children among the patients? Is there a wide scatter of observational intervals?

(2) What do you know about the location and the size of the cyst at detection time? Are you dealing with truly one cyst per patient, or may the patients have several cysts, with one large cyst being selected and scored as representative for the patient? What do you know about the underlying pathology for each patient? Do you have enough patients to stratify by disease?

(3) Is the detection of the cyst in the majority of patients incidental (no specific disease symptoms have triggered a search for a cyst), or is cyst detection in a substantial number of patients the consequence of a targeted clinical work-up (specific clinical symptoms suggesting the existence of a cyst existed)? In the latter case, how much information from radiologists, pathologists and other clinical disciplines is available? Do you have enough patients to use this information for stratification by covariates?

The answers to these questions give you a finely grained picture of the data heterogeneity, and with that, a first estimate for the number of necessary covariates and thus patient subgroups requiring distinct, non-identical decision points for stopping vs continuing monitoring (your question #1).

To address the cancer angle in your analysis, initial monitoring intervals need to be narrow enough for clinically suspect patients to capture a rapidly progressing disease process (change of cyst size over a period of weeks rather than months, together with emergence of new clinical symptoms that define additional covariates). Depending on the clinically assessed risk of the cyst representing a pre-cancerous lesion, an initial monitoring frequency every 3-6 weeks is accepted for many cancer forms

(your question #2).

Once the cyst has been found to be stable over an observation period of 6 months, further monitoring every 6 months for 2 years may be sufficient. Current monitoring guidelines published by the societies for different clinical specialties vary by organ environment and properties of the cyst (preferably established by biopsy, including fluid collection and cytology).

Now to the statistical analysis. You have already received a good number of excellent ideas how to proceed. Here are my 2 cents: the fact that many of the patients in your database die without ever developing cancer proposes exploring logistic regression models for longitudinal data with a focus on "cure" models. As motivation and introduction, the most recent review on this type of models is by Patilea and van Keilegom: A general approach for cure models in survival analysis. Annals of Statistics 2020 48(4), 2323-2346. See also Amico and van Keilegom: Cure models in survival analysis. Annual Reviews of Statistics and its Applications 2018 5:311-342. Now, conventional survival models assume that covariates do not vary over time. If you find that some of your critical covariates do vary over time, and that accounting for this variation is linked to model quality, you may benefit from a more advanced model - like a partitioned generalized method of moments (GMM) model - that allows the influence of each covariate on the response variable over time to vary individually (see Irimata, Broach and Wilson: Partitioned GMM logistic regression models for longitudinal data. Statistics in Medicine 2019 38(12), 2171-2183). Partitioned GMM models have been successfully used with Medicare data.

Hope this helps – best of luck!

------------------------------
Gerhart Graupner MD ScD
Graupner Consulting Services

Original Message:
Sent: 05-06-2024 05:35
From: Chris Barker
Subject: Optimal age of surveillance

Dear Gerhart, 

As to a post not appearing. Please contact Rick Peterson at ASA (rick at amstat.org). Rick does not make the fix. he can forward to the appropriate technical support people. And sometimes its a relatively simple matter of clearing your browser cache. Confirming, yes, from time to time posts do not appear - and at least one time it was due to a software bug. While I was section chair, I had that problem and discovered a bug. For a brief period I had the  section chair elect post on my behalf. Typically you should see your post in the community and you should also get an email copy. -chris

Hi Andy,

I am a member of the Statistical Consulting Section with the clinical background that should be helpful for answering your questions. Two days ago, I already sent you a detailed answer, but it never showed up on this forum. I understand that I am not the only one who had such problems on the Consulting forum. My apologies for asking you a question before you can get my answer to your questions, but could you kindly confirm whether you received this message ? I am attaching my e-mail to facilitate communication in case the misconfigured path of my posts persists.

Thank you and best regards, 

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy

Thank you everyone for your overwhelming support. 

I wonder how I can address your questions and thank each one of you through this platform. I responded to Dr. Sandry, but it looks my response does not appear below his message. 

Best regards, 
Andy

Hi Andy,

Encouraged by Chris Barker's response, I am sending you the updated version 2 of my post.

Before the statistics discussion, let me bring up a few clinical considerations that are important for the stratification of the patients:

(1) How old are the patients at time of detection of the cyst – wide age range or narrowly defined age range? Is mostly one gender affected? Are children among the patients? Is there a wide scatter of observational intervals?

(2) What do you know about the location and the size of the cyst at detection time? Are you dealing with truly one cyst per patient, or may the patients have several cysts, with one large cyst being selected and scored as representative for the patient? What do you know about the underlying pathology for each patient? Do you have enough patients to stratify by disease?

(3) Is the detection of the cyst in the majority of patients incidental (no specific disease symptoms have triggered a search for a cyst), or is cyst detection in a substantial number of patients the consequence of a targeted clinical work-up (specific clinical symptoms suggesting the existence of a cyst existed)? In the latter case, how much information from radiologists, pathologists and other clinical disciplines is available? Do you have enough patients to use this information for stratification by covariates?

The answers to these questions give you a finely grained picture of the data heterogeneity, and with that, a first estimate for the number of necessary covariates and thus patient subgroups requiring distinct, non-identical decision points for stopping vs continuing monitoring (your question #1).

To address the cancer angle in your analysis, initial monitoring intervals need to be narrow enough for clinically suspect patients to capture a rapidly progressing disease process (change of cyst size over a period of weeks rather than months, together with emergence of new clinical symptoms that define additional covariates). Depending on the clinically assessed risk of the cyst representing a pre-cancerous lesion, an initial monitoring frequency every 3-6 weeks is accepted for many cancer forms

(your question #2).

Once the cyst has been found to be stable over an observation period of 6 months, further monitoring every 6 months for 2 years may be sufficient. Current monitoring guidelines published by the societies for different clinical specialties vary by organ environment and properties of the cyst (preferably established by biopsy, including fluid collection and cytology).

Now to the statistical analysis. You have already received a good number of excellent ideas how to proceed. Here are my 2 cents: the fact that many of the patients in your database die without ever developing cancer proposes exploring logistic regression models for longitudinal data with a focus on "cure" models. As motivation and introduction, the most recent review on this type of models is by Patilea and van Keilegom: A general approach for cure models in survival analysis. Annals of Statistics 2020 48(4), 2323-2346. See also Amico and van Keilegom: Cure models in survival analysis. Annual Reviews of Statistics and its Applications 2018 5:311-342. Now, conventional survival models assume that covariates do not vary over time. If you find that some of your critical covariates do vary over time, and that accounting for this variation is linked to model quality, you may benefit from a more advanced model - like a partitioned generalized method of moments (GMM) model - that allows the influence of each covariate on the response variable over time to vary individually (see Irimata, Broach and Wilson: Partitioned GMM logistic regression models for longitudinal data. Statistics in Medicine 2019 38(12), 2171-2183). Partitioned GMM models have been successfully used with Medicare data.

Hope this helps – best of luck!

------------------------------
Gerhart Graupner MD ScD
Graupner Consulting Services

Original Message:
Sent: 05-06-2024 05:35
From: Chris Barker
Subject: Optimal age of surveillance

Dear Gerhart, 

As to a post not appearing. Please contact Rick Peterson at ASA (rick at amstat.org). Rick does not make the fix. he can forward to the appropriate technical support people. And sometimes its a relatively simple matter of clearing your browser cache. Confirming, yes, from time to time posts do not appear - and at least one time it was due to a software bug. While I was section chair, I had that problem and discovered a bug. For a brief period I had the  section chair elect post on my behalf. Typically you should see your post in the community and you should also get an email copy. -chris

Hi Andy,

I am a member of the Statistical Consulting Section with the clinical background that should be helpful for answering your questions. Two days ago, I already sent you a detailed answer, but it never showed up on this forum. I understand that I am not the only one who had such problems on the Consulting forum. My apologies for asking you a question before you can get my answer to your questions, but could you kindly confirm whether you received this message ? I am attaching my e-mail to facilitate communication in case the misconfigured path of my posts persists.

Thank you and best regards, 

Hello friends! 

I have read several posts on this discussion board and they are really helpful. So, I dare to post here that I might get some pieces of advice. 

I am currently involved in a research as the statistician. In the study are patients who were identified to have a cyst. Due to the concern of this cyst may develop into a cancer, many of these patients were under surveillance. It required time, effort, and money to do this surveillance. One of the aims of the study is to determine the optimal frequency of the surveillance and/or optimal age of patients to start and/or end the surveillance. Essentially, we are answering the questions: 

1. At what age should we start/stop surveying patients?
2. How many times should we do check-ups/tests to ascertain that the cyst will no longer be a concern? 

You see, some patients with cysts under surveillance may have lived long and died eventually due to old age, outpacing the development of cancer. I wonder if there is a method/s that I can use to answer the research questions. 

Your thoughts will be greatly appreciated. Thank you very much. 

Best regards, 
Andy