non-comparative retrospective study: sample size calculation required?


New Member

I am planning a very simple retrospective non-comparative study.
The plan is to review patient charts of patients with cancer to determine a value related to tumor biology. I will categorize the values in three categories (A,B,C). The aim of the study is to evaluate this value related to tumor biology. I simple want to know what the prevalence or proportion is of patients with a value in category A, B, C.
So for example 60% of patients have a value in A, 30% a value in B, and 10% a value in C.
This information is unknown in the literature.
I am not comparing any groups or something.
Does this type of study require a sample size calculation? If yes, could you give me a good link with more info?

Thank you for your time.


Omega Contributor
If your study is only descriptive in nature and you are not conducting any inferential statistics (statistical tests), then no you do not. Do you have any hypotheses for the proportion levels per groups, if so you may conduct one sample tests, otherwise, once again no sample size needed.

Ideally you would what the largest sample size possible and if you are just sampling, to possibly randomly select those patients that you examine. This is in order to get the most reflective same possible to the patient population. Unless you are looking at a select subgroup which warrants weighting in the sample selection.

I am located in a hospital with a regional cancer center, so if you are collecting some thing basic and would like to double the patient numbers, let me know, if not - good luck.


New Member
Thank you for your reply!

I am planning to study one type of cancer at a specific stage of the disease.
After pinpointing this timepoint, a series of lab values will be collecting for a predefined period before developing into the aforementioned disease state.
So someone really has to review the (electronic) patient charts [I am located in Belgium :)]. It will probably be a multicentre study.

I am still in the planning stage. But I still have to look how to limit a possible selection bias of the data.
I am looking at a selected subset of patients (developing in a specific cancer stage) after meeting some inclusion criteria.
I have seen studies for example allowing only 50 eligibel, consecutive patients per participating site to limit selection bias.
I am thinking about using such a strategie, and for example using the 50 most recent patients.
Last edited: