I came accross this article about AMD (age-related macular degeneration, a kind of eye diseases) recently. The article fitted a Cox PH model to predict risk of disease advancing to advanced stage. The authors have made a risk calculator based on this article.


As current risk calculator is in English only. I would like to create a new calculator in my language so our doctors can utilize it. From the article I have all the beta coefficient.

enter image description here

enter image description here

Still I cannot find any way to predict risk of disease progression in the next few years like how they did in their calculator.

I checked out Coxph package from R, Sklearn survival analysis, Tensorflow survival analysis but all of them require original data. Is it possible to predict without original data? I would appreciate if anyone can give me a hint.


1 Answer 1


There are both small-world, model-specific, and large-world, application-specific issues here.

Small-world issue

A strength and a weakness of Cox models as used to develop this calculator is that the baseline survival curve doesn't need to be specified to get the hazard ratios among conditions. It is possible to regenerate a baseline hazard for a reference condition if you have the original data, as shown for example on this page. That's presumably what the authors did to get their risk predictions. Unless you have the original data (or something close enough to estimate a reliable baseline hazard) you can't do that.

To solve this small-world problem you could ask the authors for the underlying data or, perhaps simpler still, ask them for permission to use their code in a web page that translates their calculator from English. Or you could offer a translation that they could present on their web site.

Large-world issue

Even if one could do the above, I'd be very reluctant to implement it for prognostication in practice outside the population on which the model was based. Both the baseline hazards and the hazard ratios might be expected to differ substantially depending on region-specific genetic backgrounds, diet, access to health care in general and to optometry/ophthalmology in particular, etc. There is no assurance that a calculator built on US-focused AREDS studies would work well in other populations. If the society in which you want to use the calculator is so removed in those ways from where the calculator was developed that you need to translate the calculator, that doesn't seem to be a good risk.

That said, it would be interesting to see how well that calculator actually works in a prospective study in a new population. That's the ultimate test of reliability for a calculator like this.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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