My first post on this site. I have a dataset with data about cancer patients. I would like to see how their survival is compared to a "normal population". I also have a table with information about age, sex and risk of dying during one year. This table covers all years from 1980 to 2016. So for example a male 64 year old has during 2011 a risk of dying of 0.23 % and for 2012 that is 0.28 %. This table is based on mortality data in the whole country and is collected from the government. I also notice that for example a 53 year old woman has a risk of dying in 2016 of 0.156 % but if she would have been 54 years in 2010, the risk would have been 0.173 %. I was thinking about the following procedure to create at reference population:
For every cancer patient, I would collect age at diagnosis, which year they were diagnosed with cancer and Sex. Then I create a list of 2000 subjects and for every year since the year of diagnosis of cancer and to 2016, I calculate how many of the 2000 subjects that would die according to the reference table I have. I then randomise the day in the year they die. The next year I do the same for the remaining until I have followed them to the year 2016. How long this will be depends on when the patient with the cancer was diagnosed with cancer.
If I do so for every single patient with cancer I have in my database, I will then have two groups. One with cancer patients and one created group with the same death rate as the whole population in the country. I can then plot these two groups in a Kaplan Meier plot to visualize.
My concern is that in my dataset with cancer patients all patients have received the same treatment. When this treatment was introduced many years ago, only a few young and healthy patients received the treatment. During the years since, more older patients have been treated. What this does is that the patients with longest follow up time (longest because they were diagnosed earliest, I mean most years ago) are young patients. I can not have a 80 year old patient with follow up time of 10 years because no one got the treatment 10 years ago and are therfore not in my cancer patient database.
I do not know if this somehow could impact my reference population and the interpretation of a possible difference (or no difference) between survival for cancer patients and "normal population"
Would it be better to do in some other way? I do really want to be able to plot it in a Kaplan Meier plot and not use relative survival functions.
Hoping for help