Does Prism do logistic regression or proportional hazards regression?

Logistic regression is available as an analysis beginning in Prism 8.3. However, proportional hazards regression is currently not available in Prism.

Logistic regression and proportional hazards regression (for survival analysis also called Cox proportional hazards regression or simply Cox regression) are related - but distinctly different - techniques.

Logistic regression in Prism

Prism provides the ability to perform logistic regression using either a single predictor (X) variable (called simple logistic regression) or using many predictor variables (called multiple logistic regression). Our curve fitting guide provides details on all of the concepts and math involved in logistic regression, and includes guided walkthroughs for performing both simple logistic regression and multiple logistic regression within Prism.

Prism can also generate survival curves using the Kaplan-Meier method, and can compare survival curves generated this way (comparisons based on one categorical variable such as "male vs. female"). However, if you wanted to adjust for additional variables, you would need to utilize proportional hazards regression, currently not offered by Prism.

Logistic regression compared to proportional hazards regression

Logistic regression and proportional hazards regression often seem to be similar methods, and sometimes it can be difficult to know which model you should choose. However, without getting into the math involved with each of these models, it's normally possible to make this decision based on the experimental data that you've collected and the questions that you're wanting to answer.

Logistic regression: the data involved here provide information on whether an event occurred or not within a set period of time, and must be binary (yes/no, alive/dead, pass/fail, etc.)*

Proportional hazards regression: the data involved here provide information on the amount time passed to an event occurring (or time to censored event), and are continuous (can be any positive value)*

Understanding this difference between these two techniques will help you choose which is most appropriate to use when analyzing your own data. Let's consider a few different situations in which you're interested in the survival of mice after treatment with an experimental drug.

• Situation 1: you've administered the drug at various concentrations to groups of mice, and you only record if the mouse dies (event occurs) or survives (event does not occur). In this case, you could use logistic regression
• Situation 2: you've administered drug to one group of mice, but not to another, and you record for each mouse the amount of time until death. In some cases, the event may not occur at all within the period of time of observation, and that's ok (this requires the concept of "censoring" data). In this case, you would want to perform survival analysis, and could use the Kaplan-Meier method to generate survival curves. If you wanted to include additional explanatory variables to this mode (such as mouse age), you would need to use a proportional hazards regression model
• Situation 3: you've recorded for each mouse the time until death, but for the purposes of your study a mouse that died after two minutes is the same as a mouse that died after two days. Although this is unlikely, you could perform logistic regression on this data

---------------

*More technically, logistic regression is concerned with the odds of an event occurring, while proportional hazards regression is concerned with the hazard rate of the event. Even MORE technically, logistic regression models the logarithm of the odds, while proportional hazards regression models the logarithm of the hazard rate.