Dear colleagues,
I am performing an observational study on a cohort of n=51 patients with a relatively rare condition. I do appreciate this is not a massive n, this is due to the rarity of the disease and time available, and for now this is more of an audit-style report on the overall management of these patients. Nevertheless, I'd like to perform some simple statistics on the cohort to determine if there are relatively more male v female, smokers/drinkers/etc in my sample than in the general population.
My questions are:
1. Does it make sense to source general population % from national statistical data to then use as controls to perform a Fisher's of Chi-squared test? Or is my sample too small?
2. How do I test groups with multiple variables for example smoker, non-smoker, former smoker
Also, if I want to determine whether some features like age of BMI range are significantly more common internally, what is the best approach? What I have done so far, for example for age, is transformed the sample from a list of patients with individual ages to a frequency distribution. I then divided the distribution in age ranges and used a t-test to check whether one age range had higher frequency than the other.
I'm not sure if these approaches make sense. Any advice would be appreciated.
Simone
For the categorical measures (gender, drinker), you could use a chi-square goodness-of-fit test, and compare your counts to the proportions in the national population. With the caveat that I wrote it, there is an example here that compares the race and ethnicity of students in a class to the proportions in the general population: https://rcompanion.org/handbook/H_03.html .
I'm not sure about quantities like BMI and age. If they are categorized into ranges in the national data, you might mimic this in your data. But I think I would treat the ranges as ordinal data, and so maybe use a one-sample sign test to compare the median age category of your sample to the median age category of the national data. This is useful to ask the question if, e.g., your sample is overall older than the national median... But you might also treat the age categories as nominal categories and use the chi-square goodness-of-fit approach. This is useful to identify if e.g. your sample is relatively higher in the 18 - 45 years category.
Since you seem to be looking for risk factors for a rare disease I would suggest a a logistic regression of the form as is used in the attached paper. Since the disease is rare in the general population I would suggest Firth regression as the method of estimation not the one used in the attached paper. Thus you would need to add a general population cohort to your dataset to provide the non
disease class. HTH Best wishes, David Booth
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire