1
$\begingroup$

I have data on medical procedures completed at hospitals in major U.S hospitals. Each medical procedure is assigned a code, for example: Kidney Transplant is X6571.

DATA

I define the success criteria and create a 'good' and 'bad' binary outcome whether. I have other variables such as wait time before surgery, other medical complication, patient age, frequency of surgeries that I control for.

For each medical procedure surgery code, I'd like to obtain a probability of success score. What models / techniques would be appropriate for this problem? I am open to suggestions beyond regression models.

$\endgroup$
2
  • $\begingroup$ Why do you ask beyond regression models? Is a logistic regression model not good enough? $\endgroup$ Nov 30 '20 at 6:38
  • $\begingroup$ It's an option, however, I want to explore other methods as well and compare accuracy. $\endgroup$
    – kms
    Nov 30 '20 at 6:39
1
$\begingroup$

This sound like a typical classification problem. So, beyond logistic regression, which is actually a probability modelling technique under the assumption of normally distributed classes, there is a great choice of machine learning methods: (k) nearest neighbour(s) classifier, neural networks (multi-layer perceptrons), support vector machines, random forests, and many more. There is a link on Wikipedia, and this sister site offers a list of algorithms available in Python.

$\endgroup$
0
$\begingroup$

Since each record of a surgery is either "good" or "bad", it can be considered as a Bernoulli distribution with the probability of $p$ for "good" and $1-p$ for "bad". Based on the surgery records for each surgery code type, you may calculate a confidence interval (CI) of $p$ corresponding to a specified confidence level(CL, e.g. CL=95%) to indicate the rate of the surgery successfulness.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.