I planned to analyze the data of "Hospitalization rates and outcome of invasive bacterial vaccine-preventable diseases". The outcome variable is hospitalization and exposure variable is clinical presentation (on diagnosis). Since the data constitute those with invasive bacterial vaccine-preventable diseases and without invasive bacterial vaccine-preventable diseases. Thus I am confused which regression model is appropriate.
not sure exactly what you are looking at.e'g .
If invasiveness is the only clinical presentation factor or whether there are others.
if it is invasiveness only just do a difference in mean hospitalisation rates between invasive and non invasive cases
if there are other factors, include these in a multiple regression of hospitalisation rate on invasiveness (= or - ) and the other factors as binary, ordinal or quantitative variables
Dear Lemma Regassa thank you for your interesting technical question. As an Inorganic chemist I'm far from being an evert in your filed of research. Thus all I can do is suggest a few relevant articles. For example, please have a look at the following potentially useful paper:
Hospitalization rates and outcome of invasive bacterial vaccine-preventable diseases in Tuscany: A historical cohort study of the 2000-2016 period
Article Hospitalization rates and outcome of invasive bacterial vacc...
As you can see, this article is freely available as public full text. Also recommended is the following RG link:
The burden of vaccine-preventable invasive bacterial infections and pneumonia in children admitted to hospital in urban Nepal
Article The burden of vaccine-preventable invasive bacterial infecti...
thank you! Dear James Leigh my question is I want to 1. hospitalization rate (of those with IBD) and 2. predictors of Hospital outcome
as the disease rare in a hospital and logistic regression (and/or prevalence rate) over estimate rates what will be best model to use?
The Tuscany study cited above is an excellent model if you have access to similar case and population data via registers etc over a large enough region and time period.
Another simpler approach could be to assemble patients and to do a multiple regression of length of stay in hospital or death/recovery versus clinical and demographic predictors. This would not give hospitalization rates.
If you have the counts for your outcome variable, I would consider a Poisson or Negative Binomial regression depending on whether you have overdispersion in your data (equal mean & variance). If you’re considering predictors at multiple levels (clinic and individual), need to account for clustering (by hospital/ward/physician/etc), or think there may be autocorrelation if there are any repeated measures, you could build a generalized linear mixed effects model, otherwise just a simple regression should do!
If your outcome is just hospitalization or not (not considering the rate), then I would just use a binomial distribution and build logistic regression (still can decide between a mixed effects or not). If your disease is rare then how well logistic regression performs is going to be dependent on your sample size and the number of cases you have. There are corrections you can apply if you’re concerned about bias due to sample size with LR.
Thank You Brittany Lauren Morgan Bustamante for your well described suggestions and direction. I want to know what type of correction could I use (since outcome is rare)?
As I understand, firstly, you want to show the prevalence or incidence to present the frequency of disease. Finally, you will try to reveal the association among predictors and outcome. So, I think you can do binary logistic regression.
- Badges
- Science topic
- Similar topics
- Journal Impact Factor