@@marinstatlectures Great presentation. There are two typos - time 16 and 27 , for survival (1/4 -> 3/4 and 1/3 -> 2/3). The main idea is very good presented. Thanks.
Great video- I think you change tactic at the 16 level of table when you multiply .428 by 1/4 - I think it should be multiplied by 3/4 as that is 1-Haz. This would be most consistent with what you did in the first five rows of the table (0-15) as well. It also aligns with the product (ie. .428 X 3/4= .321).
This video saved my life. I was really lost in my course and my teacher didn't help. It helped me finally get past a huge roadblock that I was freaking out over for several days. Thank you so much.
This video a gem, (just ignore couple of typos, and actually if you catch them means you are paying attention, and not to worry since the final result is correct)
Hi, great video. What does 5-year survival actually mean? Do we need all the patients data follow up time > 5 year? Or if the patiens live >5 years after treatment, we call them alive irrespective of relapse? Patiens follow up time < 5 year and censored would be "censored"? How does it differ from vital status (dead or alive) for survival analysis? Would you kindly make a video on this ? Thanks!
Haha, copied from pre recorded script, but the end result is correct, cos for months 16 and 27, he copied the wrong fraction (should be 1-haz and not haz). I was initially stumped, then discovered it was just a typo. 🤣😂🤣😂
What are the numbers that usually written under the KM plot? And can we know the actual number of patients experienced the event from the graph?? How??
Thanks so much for these videos! I'm a little confused with 1 - HAZ and S(t). Is 1 - HAZ the probability of not dying at time t, given that you're alive up until time t? And, is S(t) the probability of surviving beyond time t? If so, then the probability of surviving beyond time t [S(t)] is the probability of surviving at t-1 (e.g., .917) multiplied by the probability of not dying at time t (1 - HAZ), given that you're alive until time t (rather than the probability of not dying beyond time t)?
Thanks for the video and your great explanation! I saw one mistake in survival time that happened on time 16 months. I think you mistakenly multiplied hazard to survival time of the previous time instead of 1- haz to survival time. It should have been 0.428*(3/4). This made survival time 27,30, and 32 wrong.
If you have thousands of observations, is it easier to perform this by hand instead of inputting thousands of rows of data in an excel spreadsheet to compute in R?
It seems like I always see these survivorship curves dropping. Would it be possible for a survivorship curve to increase? Say there is a high mortality rate up front, but later in the survey there are no more deaths... Would the curve drop down at first but then step back up, since the survivors are not dying?
The curve is cumulative, so you can’t see that. E.g. it is not possible that at time t=2 70% of people are still alive, but at time t=3 80% of people would still be alive, as that would require people coming back to life. What you’re thinking of is the hazard (the rate of decrease for the curve). The hazard (rate of death/decrease) may be high at first, but then get lower as time passes. In some videos in this series we compare and contrast different survival models. The KM model allows the hazard to fluctuate. The Exponential model assumes a constant hazard. There are models that allow the hazard to change increase (or decrease) proportionally with time, and the Cox model which allows the hazard to fluctuate over time. There are pros/cons to each, and i discuss some these over the series of videos
It is the same as the life table method if you let the intervals become infinitesimally small for the life table. Or, you can think of it as like the life table method if you start each of the intervals for the life table wherever a death occurs. That’s why the KM method is also known as the product-limit method
If you are "right censored" i.e. you still have a bunch of people surviving at the end of your observation period, I assume you just put them as 0 at the last date you have observed?
At 09:02
Survival Function will be multiplied by .428 with 3/4 instead of .428 with 1/4.
Thanks for catching that typo! Fortunately the survival value of 0.321 is correct :)
@@marinstatlectures
Great presentation.
There are two typos - time 16 and 27 , for survival (1/4 -> 3/4 and 1/3 -> 2/3).
The main idea is very good presented.
Thanks.
Hi, Marin. At months 15 and 16 the survival fx you have calculated them with the razzard of dying and not surviving. Thank a lot for you videos
Good catch.
Even for month 27 and 30 also
Great video- I think you change tactic at the 16 level of table when you multiply .428 by 1/4 - I think it should be multiplied by 3/4 as that is 1-Haz. This would be most consistent with what you did in the first five rows of the table (0-15) as well. It also aligns with the product (ie. .428 X 3/4= .321).
Yes, that was a typo...no way to change it now :) but the final answer is correct
@@marinstatlectures you can pin the comment and/or put an annotation in the video with RUclips's video editing tools :)
This video saved my life. I was really lost in my course and my teacher didn't help. It helped me finally get past a huge roadblock that I was freaking out over for several days. Thank you so much.
This video a gem, (just ignore couple of typos, and actually if you catch them means you are paying attention, and not to worry since the final result is correct)
You're a great teacher. Thanks for the content.
Excellent lecture. Searched a lot to understand.Finally understood how to use the censored data.
So clear and easy to understand! Thank you for posting this video!!!
You’re welcome ☺️
Thank you so much. I wish my university would teach us this knowledge.
thank you, from the bottom of my heart
Thanks for simple and clear explanation marin!
You are the man. You make it look easy! Thanks a lot
Amazing explanation!!!
Hi, great video. What does 5-year survival actually mean? Do we need all the patients data follow up time > 5 year? Or if the patiens live >5 years after treatment, we call them alive irrespective of relapse? Patiens follow up time < 5 year and censored would be "censored"? How does it differ from vital status (dead or alive) for survival analysis?
Would you kindly make a video on this ? Thanks!
Thank you very much, I've got the idea
This really helped me, thank you so much!
You’re welcome
extremely helpful! Amazing.
Thanks prof🎉
Great tutorial, thanks
The best!
Thankyou very much for this video
Very useful, thank you very much.
Can you do a video on progression free survival and overall survival? Do you recommend any books on survival analysis with r? Thanks a bunch!
thank you! You are a great teacher :)
You’re welcome
I think there are some typos on whiteboard (see column S(t) ): it is not 0.428 (1/4) but 0.428 (3/4) and it is not 0.321 (1/3) but 0.321 (2/3)
Haha, copied from pre recorded script, but the end result is correct, cos for months 16 and 27, he copied the wrong fraction (should be 1-haz and not haz).
I was initially stumped, then discovered it was just a typo. 🤣😂🤣😂
did you calculate the survival time correct at 16 and 27 months? You multiplied the probability by hz than 1-hz.
Those were typos, and should be multiplied by (1-HAZ), although the final answer for the survival is correct
@@marinstatlectures Thank you, your videos are fantastic
Magnificent
What are the numbers that usually written under the KM plot? And can we know the actual number of patients experienced the event from the graph?? How??
on the 16 its supposed to be 0.107 at the end cause we are multiplying 0.428 x 1\4
Thanks a lot
Thanks so much for these videos! I'm a little confused with 1 - HAZ and S(t). Is 1 - HAZ the probability of not dying at time t, given that you're alive up until time t? And, is S(t) the probability of surviving beyond time t? If so, then the probability of surviving beyond time t [S(t)] is the probability of surviving at t-1 (e.g., .917) multiplied by the probability of not dying at time t (1 - HAZ), given that you're alive until time t (rather than the probability of not dying beyond time t)?
Thanks for the video and your great explanation! I saw one mistake in survival time that happened on time 16 months. I think you mistakenly multiplied hazard to survival time of the previous time instead of 1- haz to survival time. It should have been 0.428*(3/4). This made survival time 27,30, and 32 wrong.
i want to use Kaplan Meier in information technology field please suggest me any topic..
Great video can I ask a question please many thanks
If you have thousands of observations, is it easier to perform this by hand instead of inputting thousands of rows of data in an excel spreadsheet to compute in R?
No, software is easier, and also greatly reduces the chance of errors... a computer won’t make a calculation error, while us humans will at times
It seems like I always see these survivorship curves dropping. Would it be possible for a survivorship curve to increase? Say there is a high mortality rate up front, but later in the survey there are no more deaths... Would the curve drop down at first but then step back up, since the survivors are not dying?
The curve is cumulative, so you can’t see that. E.g. it is not possible that at time t=2 70% of people are still alive, but at time t=3 80% of people would still be alive, as that would require people coming back to life.
What you’re thinking of is the hazard (the rate of decrease for the curve). The hazard (rate of death/decrease) may be high at first, but then get lower as time passes.
In some videos in this series we compare and contrast different survival models. The KM model allows the hazard to fluctuate. The Exponential model assumes a constant hazard. There are models that allow the hazard to change increase (or decrease) proportionally with time, and the Cox model which allows the hazard to fluctuate over time. There are pros/cons to each, and i discuss some these over the series of videos
@@marinstatlectures That makes sense. Thank you so much for these videos!!
You’re welcome
Is it different from life table?
It is the same as the life table method if you let the intervals become infinitesimally small for the life table. Or, you can think of it as like the life table method if you start each of the intervals for the life table wherever a death occurs.
That’s why the KM method is also known as the product-limit method
ياخي أحبك
I could not follow the sound...
If you are "right censored" i.e. you still have a bunch of people surviving at the end of your observation period, I assume you just put them as 0 at the last date you have observed?