© 2022 MJH Life Sciences and AJMC – Managed Care News, Research, and Expert Insights. All rights reserved.
© 2022 MJH Life Sciences™ and Clinical Care Targeted Communications, LLC. All rights reserved.
Tom Robinson, vice president of global access at JDRF, talks about the predictive modeling exercise used to develop and refine the Type 1 Diabetes (T1D) Index.
Collaborating with experts from different groups, countries, and academic disciplines helped formulate refinements to the Type 1 Diabetes (T1D) Index, said Tom Robinson, vice president of global access at JDRF.
Transcript
What research went into developing the T1D Index, especially to create more specific estimates?
From the start, we knew that we had to deal with the challenge of scarce data. We don't have registries in every country, we're not gonna get them anytime soon, although that's what we need to get to eventually. So we decided to engage in a predictive modeling exercise and we got some really good data scientists to help us with that. What they steered us towards is an iterative and collaborative approach. So, iterative in the sense that we built the first version of the model in a week, and the guys who did it, Gabriel Gregory, he did not sleep. But at the end of the week, we had our first estimate and it was not terrible, it was actually a pretty decent start. But then we said, "okay, well, how can we improve it?" And that's where the collaborative came in.
We started bringing in experts from different groups, different countries, different academic disciplines, and they'd come in and they'd challenge us in 1 of 2 ways. One would be to say, "I don't like your methods, I don't think they're valid, or you should try different approach." Or they'd say, "hey, the numbers don't seem right in this respect. There's too many old people, or there's not enough people in this setting, or you're overestimating how many people might be in Central Africa."
By holding that model up to reality, we could see, oh actually, you know what we need to do? We have a problem with regionality in Canada, or our mortality model is too consistent across all ages—we need to make sure it flexes by age more—so we started making these methodological refinements. That was the process we went through and we were probably in that phase for about 18 months, 12 to 18 months before we started to see the numbers become very consistent. Then we spent another 12 months continuing to iterate and seeing that those changes didn't move the numbers very much. It was at that point we said, okay, this is pretty solid now, it's time to start thinking about publishing.