comparing*sensitivity of a test

[Abdulwahab Sessolo]

Hello to all. Can someone explain how you can use STATA to compare performance of the test in 1 population with different characters eg comparing if sensitivity between diabetics who have history of TB Vs Diabetic who have never been sick of TB. is significantly different.

[Clyde Schechter]

You don't show example data, nor do you say what the test attempts to diagnose. So I'll just give you an outliine of the approach.

First -keep- only those observations for people who have the disease that the test is being used to diagnose, and who are diabetic.

Then use -tabulate- to create a cross-tab of test result and the TB variable. Use the -chi2- option to get a chi square test out of that.

[Abdulwahab Sessolo]

thanks for the response. To clarify let use this example. I have a test which diagnoses TB and i use it to test its accuracy of TB diagnosis in patients with and without diabetes separately. In the Diabetics Sensitivity is 60% While in the non diabetics its 55% . My question is how do i determine that this difference in sensitivity is statistically significant using STATA

[Clyde Schechter]

You cannot do this with just summary percentages. You need to have a data set of individuals who have been tested. The data set must contain the following variables:

1. A 0/1 indicator of diabetes. Let's call it diabetes.
2. A 0/1 indicator of tubercluosis. Let's call it tb.
3. A 0/1 indicator of the test result (0 = negative, 1 = positive). Let's call it test_result

keep if tb == 1
tab test_result diabetes, col chi2

That said, do bear in mind that the American Statistical Association now recommend abandoning the concept of statistical significance. See https://www.tandfonline.com/doi/full...5.2019.1583913 for a full explanation, or https://www.nature.com/articles/d41586-019-00857-9 for a brief one. Your situation sounds like one in which the concept of statistical significance is especially inappropriate. You should be comparing these tests in a decision analytic framework, which also means that the specificity of both tests also needs to be taken into account, as does the actual prevalence of tuberculosis in the population for which you are deciding which test is preferred. And the decision should be made on the expected utility of each test. Uncertainty in that should be quantified by doing a sensitivity analysis about underlying assumptions and uncertainty of data inputs. Statistical significance never had any relevance here, even before the American Statistical Association's new position.

[Abdulwahab Sessolo]

Thanks alot sir for the comprehensive explanation.