Fertility and Sterility On Air - Seminal Article: Julia DiTosto and Sunni Mumford

Join us for a special episode of F&S On Air: an interview discussing the seminal article, "Target trial emulation of preconception serum vitamin D status on fertility outcomes: a couples-based approach." Host Micah Hill interviews authors Julia DiTosto and Sunni Mumford.

Read the article:  https://www.fertstert.org/article/S0015-0282(24)01963-0/fulltext

View Fertility and Sterility at https://www.fertstert.org/

Welcome to Fertility and Sterility On Air, the podcast where you can stay current on the latest global research in the field of reproductive medicine. This podcast brings you an overview of this month's journal, in-depth discussion with authors, and other special features. F&S On Air is brought to you by Fertility and Sterility family of journals in conjunction with the American Society for Reproductive Medicine, and is hosted by Dr. Kurt Barnhart, Editor-in-Chief, Dr. Eve Feinberg, Editorial Editor, Dr. Micah Hill, Media Editor, and Dr. Pietro Bortoletto, Interactive Associate-in-Chief.

Welcome everyone to a special edition of Fertility and Sterility On Air. I'm Micah Hill, the Media Editor, and we have a behind-the-scenes, one-on-one discussion with two authors on a recent F&S article. So we have Sunni Mumford, who's one of our editors in Fertility and Sterility at the University of Pennsylvania.

Good morning, Sunni. Good to see you. Good morning.

We're excited to be here. And we have Julia DiTosto. She's a PhD candidate who's also at Penn and who was the lead author on this paper.

Julia, it's great to see you today. Great to see you too. Thanks for having us.

All right. So let's dive right in. So your study is titled Target Trial Emulation of Preconception Serum Vitamin D Status on Fertility Outcomes, a Couples-Based Approach.

So let's just start with that title, Target Trial Emulation. At a big level, so we're talking about using observational data to try to recreate a clinical trial. So explain to me, like big picture, how does that even work? How do we get to that point? Well, I think these methods are really exciting because when we think about all the randomized trials that we would want to do, oftentimes it's impossible.

We would need the trials to be really big. They would take a long time. Sometimes it may not even be ethical to randomize for what we want to evaluate.

And so these causal inference methods, it's really a way of thinking about setting up the study and the question like a randomized trial so we can have results that are more interpretable and allow us to use the rich observational data that we have thinking about it in a randomized trial setting. Sunni, from a big perspective, you and Enrique still do a lot of RCTs. When you were at the NIH, you did a lot of RCTs.

What was the typical budget you're talking about for the example of the FAST trial, the EAGER trial? How much money did it take to do those trials? Yeah, the FAST trial was about $15 to $20 million. So can't really be done on a typical R01 budget. So the time and cost of these trials really is prohibitive for a lot of questions.

Julia, I'm a clinician mostly. I'm a novice statistician. I hang around Sunni and Enrique, people like that, to try to get smarter about this.

So help me conceptualize this. So when I do an observational study, there's all these methods we go through to try to account for the biases that may be in that data. Sometimes we're doing DAGs like you guys did and trying to find where that confounding may be.

We might do things like propensity score matching or other types of matching or stratify our data. So to get those confounders to be accounted for that way. Or we might do regression models or GE models to try to account for the confounding sort of on the back end.

So conceptually, how does the approach you took in a trial like this, where you're emulating a target trial, sort of differ from maybe traditional approaches of how we look at our observational data to account for that confounding? Yeah, that's a great question. So a lot of those methods can actually be used in a target trial emulation. So this idea is mainly a framework where you're trying to emulate a randomized controlled trial, specifically using the observational data.

It doesn't necessarily mean that you're only using specific methods, like you're only using propensity score methods or something in the emulation, but rather you're using those tools that we already know how to do, like propensity score methods or matching or stratifying, but you're using those methods in the context that you would to try and make your observational data as closely mimic as possible as the target trial of interest. And by doing this, you're able to kind of identify different sources of biases that you might not initially see when just doing a traditional observational study. So it kind of helps to point out those things really clearly.

And another thing that we often do in observational data that's very different than RCTs is in RCTs, the alignment of the start of follow-up, the assessment of eligibility criteria, and the assignment of the treatment strategies, it all happens on the same day. Like people are approached in the clinic, they're recruited, and then that day they're randomized. And so everything's like very clearly on the same day.

But in observational data, you only have information on the days that you see, whether it be like electronic health records and it's the days they're in the clinic or things like that, or it's claims data and it's only the days they're billed. And so things get a little more difficult with aligning those three. And in using the target trial emulation framework, you're able to really clearly align those three time periods.

And so that helps you kind of avoid some of the biases that we might see traditionally. So at least at a conceptual level, I think I have a basic framework of what we're trying to do with target trial emulation. So let's talk specifically through what you guys did and sort of learn through the methods of how you practically did that.

So tell me, first of all, what was the purpose or the aim of your study? What were you looking to look at? Yeah, so we know for many reasons that vitamin D may be critical for reproductive health. There are receptors and reproductive organs and it's directly involved in key processes and in conception. But a lot of previous literature on this topic, specifically looking at preconception vitamin D and fertility outcomes have shown mixed results for a number of reasons that we think.

So in this study, we aim to use a target trial emulation framework to assess the associations between preconception vitamin D among a large cohort of couples seeking infertility treatment. So not just like males and females, but specifically examining the individual and the joint association of vitamin D, as well as other biomarkers, specifically among people who were seeking infertility treatment across four US centers. And we also investigated the potential effect modification by BMI because BMI is closely linked to chronic inflammation and we know that vitamin D has anti-inflammatory properties.

So just seeing if there was any treatment effect heterogeneity by BMI when looking at vitamin D and fertility outcomes. You specifically make the point that this is not the same as emulating a trial where you randomize patients to vitamin D supplementation. Can you just explain why it's different than that? Yeah, of course.

So one of the key considerations when doing a target trial emulation is you're not necessarily emulating the ideal target trial. So you're not, it's not the ideal RCT that you would do in all circumstances, but rather it's the trial that you would be able to do. So in this case, we only have data on vitamin D levels overall.

We don't know if people are taking vitamin D supplementation. We don't know, for example, their sun exposure during the day, we do have data on physical activity and nutrition, but we don't actually know if the vitamin D levels that people have at certain times are because of supplementation or because of lifestyle behavior variables. So when extrapolating these data into a trial, it wouldn't necessarily correspond to a trial of supplementation, but rather a trial that would be more akin to like behavioral modification trial where people are randomized to a number of different things.

And that overall is changing their vitamin D levels. So that's kind of a limitation of the data and a specific consideration of the method. So from a conceptual standpoint, you're essentially randomizing patients to a vitamin D level.

That's what you're trying to do with this trial. Great. So we talked about sort of the framework of this and how it uses methods we're maybe familiar with from some of our approaches data.

Talk me through the practical steps of how you then went through those processes of emulating a target trial with this observational data set that you had. So this is what Julie was talking about. You actually go through the steps of thinking about what this trial would look like.

So, and there's a table that goes through this in the paper where we look at the eligibility criteria. We define what the intervention would be. We define the outcomes, the follow-up time, and we think about what that ideal trial would look like.

And the tricky part here is that we're not actually able to randomize, right? So we're emulating the randomization. So it's important to have information on as many potential confounders as possible so that we can adjust for those factors to emulate that randomization. So with randomization, you're getting balance on everything, all the characteristics except for the intervention.

And so here we're trying to adjust for as many confounders as we can to make sure that the differences are here just between vitamin D levels. But it really is just going through, okay, we want to look at females 18 to 45. Are there certain health conditions we would want to exclude? When do we want to start follow-up? And really thinking through what that trial would look like.

And then mapping it with the data that we have. So making any exclusions that we need with the cohort, making sure the timing is aligned. And then we apply the usual data analysis methods that we have, depending on if it's repeated measures or a binary outcome, that part is consistent with what we know.

It's really the setting up the question that is the different part of the framework. And Sunni, does the evidence suggest that we get closer to what RCTs show as far as effects estimates or compared to observational data, traditional ways of approaching these questions? Yeah, so there have been several studies that have compared randomized trials that have been done with this approach using observational data, and they do come quite close. So they are comparable.

Okay, so Julia, I think we understand the framework of this study and how you did it. What did you guys find? What are the big take-home points from the actual trial that you emulated? Yeah, we found that women with sufficient vitamin D status had a much higher likelihood, specifically a 28% higher likelihood of having a live birth compared to those who had deficient levels of vitamin D. And we also saw really strong associations when both partners and the couples had higher vitamin D status compared to if both partners had deficient vitamin D status. Interestingly, we found some differences by BMI.

So we found the strongest associations by BMI, and we think this might be because, as I mentioned earlier, BMI is associated with chronic inflammation. And so when there's a lot of chronic inflammation in the body, the ability for vitamin D to act in the way that it needs to act for conception might be hindered a bit more. So that could explain why we saw stronger associations among those with normal BMI.

We did not find any associations between preconception vitamin D among the male partners and semen quality parameters, and this suggests that maybe the effect of vitamin D on fertility outcomes is not likely to be mediated through semen quality. And interestingly, we also did not find any associations between preconception vitamin D and pregnancy loss. Great, thank you.

Very interesting data. Basically, I think this study confirmed what many of us already thought about vitamin D and outcomes. In general, when you look across literature, not just in reproductive medicine, but lower vitamin D is associated with a host of worse outcomes from a medical perspective.

Do you have any gestalt as to whether that's something that we can target and treat medically, or is it just a marker of general overall health and other things that are going on that maybe are associated with worse outcomes? From an epi perspective, do you have any gestalt as to which direction that likely is? Yeah, so I've thought a lot about, should we do a vitamin D supplement trial next, given the strong links that we see between the vitamin D levels and reproductive outcomes. But a lot of the supplement trials that we've seen tend to be null. And I think the vitamin D levels are a really good marker.

But that the supplements are not the only thing that we need to be thinking about. Because the vitamin D levels, you need pretty high levels of supplementation to move vitamin D levels. And maybe we need to be spending some time in the sun, you know, joint a joint intervention of supplementation plus some time in the sun, because that's a much more efficient way to raise vitamin D levels.

So with vitamin D, I feel like it's a little more complicated than just taking a supplement. But the things that we know to do to raise vitamin D do seem to be important in it. Yeah, I think we're always looking for that pill that can fix something, but it seems to be a little bit more.

And I think vitamin D is probably what you say a marker. Okay, no, I appreciate that perspective. And as you say, most of the trials, at least in reproductive medicine with these supplements have unfortunately been null as we're all looking for the cheap magic bullet that can help improve outcomes.

That's a low cost intervention. Give me an idea of your typical approach that you would have done for the same study just using traditional observational data techniques versus emulating a trial. How much extra work is this? Is it half 50% more time, double the time, four times more effort? How much more work is this? And the reason I ask this is so, Sunni, one of your former postdocs, Kerry Flanagan works with me every day.

And so we're talking about trying to start taking this approach for a lot of our observational data. But we have a lot of fellows, 11 of them. And so that's a lot of efforts.

Just how much work, how much time is it to get to this level of accuracy with this observational data? I don't think it takes that much more time because we're thinking about these questions and it's kind of like drawing the DAG. It takes some time to think through the setting, what confounders are there. But I think it's really important in setting up the question in a way that we have really translatable findings.

I feel like that extra effort in putting in the time to think about what the trial would look like, how to set up the models, really pays off on the other end. And it's not too much more, I don't think. Maybe Julia can speak more to that too.

This was my first time applying this method. And I think what made it a little easier for being my first time is that we were using data from an RCT. So the data collected in this cohort was from the FAST trial, as Sunni mentioned, which was a supplement trial of folic acid and zinc among males and couples seeking infertility treatment.

So we were really lucky in that we already had extremely detailed and reliable capturing of outcome assessment. We also had a huge number of important variables collected that might be confounders, including physical activity and nutrition that often are not included in some of these observational databases. So we were able to leverage this extremely rich data set, which I think more clearly allowed us to apply this method, especially for someone who's doing it for the first time.

And in some of our other work right now, we are applying this method in more traditional, like real-world evidence areas, like using claims data or electronic health records. And I think some of the considerations are the same, but perhaps there's a little bit more legwork in the front half when you're working with some of those data sets that weren't initially created for research purposes. That makes a lot of sense.

And so that you just led me, Julie, into what was going to be my final question. So for someone like me and Kerry, if we're dipping our toes into the water on taking this approach, where's the ideal question or data set to first apply this to? And I think you sort of hinted at a lot of those answers with that. But if you have any advice for fellows and researchers that maybe want to try to do this with their data sets that they have, what would it be? I think with anything, you just take it step by step.

So it really has helped me in trying to think about this by going through that table. Like what are the eligibility criteria? What do we want these interventions to look like? And just really taking it step by step. It's one of those things that you break it down and it's doable.

And there are these approaches that help do that by walking through the tables that are out there. Love that answer. Like many things that seem complex when you first start thinking about them, when you break them down at the methods of each step, which you guys did beautifully in this paper, it all makes sense to me.

I'm still trying to wrap my head around the big picture, but I feel a little bit smarter after hearing you two explain it to me. So I think we need to have back for a full journal club to talk about this because it's such an important methodology that I think for us as readers to interpret our literature is going to be really important. So thank you, Julia.

Thank you, Sunni, for sharing your expertise with us today. Thank you for having us. This concludes our episode of Fertility and Sterility On Air, brought to you by the Fertility and Sterility Family of Journals in conjunction with the American Society for Reproductive Medicine.

This podcast was developed by Fertility and Sterility and the American Society for Reproductive Medicine as an educational resource in service to its members and other practicing clinicians. While the podcast reflects the views of the authors and the hosts, it is not intended to be the only approved standard of practice or to direct an exclusive course of treatment. The opinions expressed are those of the discussants and do not reflect Fertility and Sterility or the American Society for Reproductive Medicine.