In the next series of segments, were going to be looking at things that might be happening in the middle of a study after you launch while you're collecting data. First we'll be looking at data changes. Mid-study data changes. After that, in this session, we'll talk about interim reporting requirements. And then finally we'll end up in this particular session with information about data, data and safety monitoring plans and data and safety monitoring boards. In the next several segments, but not covered in this one, we'll talk about data quality and audi-, auditing. So, the first principle that we've talked about many, many times is, you put adequate planning time into your study on the front end. You get everything just right. And it's really going to help. You can, you can make sure that you're making less mistakes, that you're collecting data on all patients that that basically you've thought everything through. And so putting that time in at the very front of a study before you see patient number one, is absolutely the thing to do. However, things happen. So I don't think I've ever worked with a, with a study that's longer than several weeks that, that we didn't have some sort of mid-study data correction needed. So, despite the best planning efforts, mid study changes are required. They, they do sometimes happen. And it could be from several, several different angles. Or for several different reasons. First one is, maybe, new science changes the research plan. So maybe you start a study. You're, you're six months into a pain study, where you're looking at, Individual's threshold, or their response to a particular drug, and how they are impacted by pain. And then a new, brand new study comes out that says, hey, hair color really correlates well with pain level sensitivity. So that's this new confounding factor that really didn't exist, that you didn't. Really have an opportunity to think about before this study came out. So, you know, you need to collect that. So, you're basically left with the option of not collecting and then losing information going forward, or changing it and adding that particular variable. In this particular case, adding hair color, assuming you could go back and re-contact, volunteers that are participating in your study. You might be able to retrospectively go get, go back and get the hair color of the individual that you're studying. At least the hair root color doesn't change very often. And so, so that's something that's not time-sensitive versus, you know, if you, if you had to go back and get a seated blood pressure measurement. one you know there's no way that you can get that at that particular time, at that particular time in the study after the fact. So it's sometimes can be expensive sometimes it's, it's absolutely impossible if you can't find old study volunteers to get this information. But, but it's not necessarily catastrophic and, and, and this is type of data, data addition that you might have to consider in the middle of a study. Another one might be something, you know, beyond your, your, your, your, best efforts. Beyond, beyond all planning and beyond all, So, sort of, natural, natural events, something catastrophic happens, like your lab burns down and you have to start with different analysis equipment. Or perhaps you are doing MRI studies and the, the university that you're working in or the setting that you're working in. They decide that they're going to change magnets. Which is a huge investment, a huge expense. And they're probably not going to worry about your particular research study, when, when they're making those, those specific decisions. So things happen, particularly if you're, if you're dealing with With, with studies that are going on for some time. So, so this particular case might, might include some versioning, maybe information about data that were collected before a particular event and, and data that were collected after a particular event. And certainly you'd want to document any sort of changes in the, in the historical audit logs, as well as the standard operating procedures for the study. Another one, especially if you've got long-running longitudinal studies on the neighborhood of years or even decades, in working with maybe a longitudinal study, looking at the impact of technology, on, on adolescent or pediatric behavior. There are things that we can't think about now, that might be happening in the technology world in two years, that really, they haven't been invented, they haven't even been thought of yet, so, you know, from a data perspective, we also, we always have to be flexible, you know, even in the mid-study mode, even when we've done all of the preparation just right up front. we have to be open to the, to the fact that things do change. And so, from a data perspective, again, think about versioning, label data with the, you know, the version of the, of the instruments that, that you're using. And think, think proactively about how adding this particular data element or this measurement is going to impact the the statistical analysis of the study downstream. Because, at the very least, you're going to have a subset of individuals, that you never ask those sorts of questions of. It's okay and, you know, really, you can you can overcome this and you need to overcome it all the time in these long longitudinal studies. But, just be mindful of it. So mid-study data changes are stress worthy. You know they, they happen. they're not something to be taken lightly, and, and when they happen, you really should as a data management profess, professional as an informatics consultant on, on studies, you really need to take them serious. But, but you really need to aim for low-stress rather than, than, than high-stress, things that are absolutely catastrophic, for your study. So I asked some of my favorite experts. These, these are people on my team here at Vanderbilt, that deal day in and day out with research studies across all kinds of diverse study domains, sorry, scientific domains. You know what are the, what are the common things that, that we see most often in studies when, when there are. mid-study changes required or necessary and, and what do you, what do you recommend in, in terms of that practical advice to those research teams? What, what are you really sort of warning them against? And I've got maybe five or six of those that, that I'll present out in the next several slides. The, the first one is, be mindful about recoding data. So I've got a question where, you know, maybe the question around this data element would be what is your favorite ice cream? And, and. we've been collecting for some time, chocolate and vanilla. We've been following our coding rules. And, and we've got one, coded as chocolate and two as vanilla. We've even got a, a good data dictionary. Or, a code book there that, that. Where we understand that. we've seen research teams. You, you know, that are naive about sort of the impact about making changes. We've seen from time to time people that, people that wanted to do something like insert strawberry. And let's, let's insert strawberry and let's code that one now as two. And, and code vanilla as three. This could be one of those catastrophic messing up all of your data, or at least all of the data that have been collected previously. Unless you go back, [COUGH] excuse me. Unless you go back and, and you make sure that any previous two entries that are in your data elements or data base. they, they are automatically recoded to three to sort of take this into account. But even there, you've got audit trails in the background. Where, where you've got, computer logs that, that say on this particular day. This particular person, added this particular element. It can be a real mess. And so. I always just recommend not recoding data. not, not reusing codes that have previously been used for the same, questions. Because it's, it's, it's really, It's a disaster in the making. So, be mindful about deleting data. So even if you remove a question. You may not, and, and in my opinion. You should not be allowed, by study protocol, to remove data previously collected. You may not think it's important. But, but if it was collected it's important and so you need to make sure that, you know, you may, you may take that off of the case report form if, if the study design calls for it, or if the study planning committee calls for it, but, but I would say you never actually delete those data elements. You, you keep them that way if, if something comes up later where you need those or if you need to provide some sort of evidence about the, the decisions that were made. You've got those and you can bring them forward. It's just the right thing to do. So be mindful about renaming data fields. And so let's go back to our simple example before. I've got that field around, around what's your favorite ice cream. It's, it's called ice underscore cream. If, if I went back and I renamed that data field to something else, especially back in, in the, in the guts of your data system you, you may actually be, be doing essentially the same thing as deleting data because it's you know, you've got all of those audit trail logs again. That sort of point to the fact that, you know? This, that variable called ice cream was changed. Or it was added here at this particular time. By renaming, fields that, that you've used previously, you run into lots of problems. And so just, just don't do it. And then finally, be mindful about changing the meaning of, data fields. And this, this is one that, you know? It's pretty obvious what, what can happen from a, from a danger standpoint. But, but we see it sometimes. And. We, we as a data coordinating center you know we're always on the lookout for these sorts of things. So here I've got a field that's called ice cream it's a nominal coded variable and the label on that was previously what is your favorite ice cream flavor? Field choices 1, 2 and 3 for chocolate, vanilla and strawberry. If I came in, in the middle of a study and said you know really I want to ask I meant to ask what is your least favorite ice cream flavor. So it's kind of. You know, obviously, you've got data that are already collected. You actually just change the meaning of those data 180 degrees, and so it really could invalidate old collected data. Better in these particular cases to to sort of archive a data element and then start collecting it fresh using, using a different name. Be mindful about adding data. And we talked a little bit about this earlier, but you know, if I've been asking this, this ice cream question, and I've got one is chocolate, two is vanilla. If, if I go in, and I don't invalidate the, the label rule that we just talked about, and I don't invali, don't violate the coding rule that we talked about earlier, but, but instead add that. That strawberry as my number three coded choice should be okay, because we haven't used that coding before. It's not going to interfere with data that are already collected. There's not going to be any surprises there when we get ready to analyze it. But, again, remember, always, that, that you know, since you're doing this in the middle of a study, if you've got five patients or a 100 patients that have already taken that particular survey. then, then at the very least, you know, the, the, from 101 on, they saw that additional option, from, from the people that have already collected, you've already collected data on. You know, it's not really the same question if the answers are, are are different. So they never actually had the choice to choose strawberry, and so maybe they chose the second best choice. Silly example, but, but, you know I think it's pretty obvious what, what can happen there. And if you're not careful, if you don't keep, import, keep records of, of when additions were made, then it can be really, really tricky when you get ready to analyze it. And again, these longitudinal studies, sometimes that can be many, many years downstream. And so, just be really careful and know that. Any type of permutation that you do on the data, it can really have lasting impacts on the analysis side.