Observational Studies

Join Dr. Heather Allore, from Yale School of Medicine and Professor Basia Diug, from Monash University School of Public Health and Preventive Medicine (Melbourne, Australia), to discuss the role of observational studies in multimorbidity research. The pair also define and provide examples of several categories of observational studies, while examining the strengths and limitations of these designs.

To view a transcript click here then select the transcript tab. 

Transcript

Copy Transcript Copied

Heather Allore, PhD, MS, MA: Hello, I'm Heather Allore, Professor of Internal Medicine in the section of Geriatrics at the Yale School of Medicine, as well as Professor of Biostatistics at the Yale School of Public Health. I'm here today with Professor Basia Diug from Monash University's School of Public Health and Preventive Medicine in Melbourne, Australia. She's Head of the undergraduate courses and Head of the Medical Education Research and Quality, and she's won several awards for excellence in teaching. 

We will be discussing key points from the AGS/ AGING Learning Collaborative Clinical and Translational Learning curriculum that focuses on the science of multiple chronic conditions. As part of the research module, Dr. Diug contributed a presentation on observational study design. 

Thank you, Dr. Diug for joining me. It's exciting to have your international expertise, as I know you've [01:00] taught observational studies around the world. I wonder if you want to tell us a little about yourself.

Basia Diug, PhD: Many thanks Prof. Allore for that wonderful introduction and having me part of this series and speaking with you today. As you mentioned, I'm the Head of undergraduate courses and the Head of the Quality and Innovation in our Medical Education, Research and Quality Unit at the School of Public Health and Preventive Medicine at Monash University in Melbourne, Australia.

I am also a passionate epidemiologist and a medical educator. And I've worked in this field for over 10 years, 15 years, and I have both experience in both quantitative and qualitative research. But I think my true passion is in medical education where I have a strong focus on improving research literacy among health professionals.

And my interests lie in assessment and also developing key study designs and truly having that research literacy at the forefront of health professional education. So it's a really wonderful opportunity [02:00] to come and speak about observational study designs and why they're an important tool that we can use to really measure and evaluate and investigate important exposure such as multimorbidity.

Heather Allore, PhD, MS, MA: Fantastic. I think you've really hit the nail on the head since multimorbidity is the overarching theme of the AGS/ AGING Learning Collaborative, and many of the listeners may have already viewed the webinar and presentations by the Data Measures and Measurements Module that were really important for defining widely used definitions for multimorbidity. But why is defining this variable important in terms of choosing the correct observational study design?

Basia Diug, PhD: I think it's the key question that everyone needs to ask themselves when they start planning any research project. How do I choose the right study design? How am I defining my case [03:00] or my exposure status, or what I'm trying to investigate, particularly with a complex variable such as multimorbidity. I mean, multimorbidity- it typically refers to the presence of two or more chronic conditions, and it has a number of definitions in the literature. 

But when you start doing a, a look into the literature, these definitions actually vary quite wildly. They might include index diseases, so predefined medical conditions. It could be captured as unlimited numbers and types of medical conditions. Perhaps it's only chronic conditions, or both acute and chronic conditions, or perhaps it's being captured as only physical disease or both physical and psychiatric conditions.

However, the definition is captured will play an important role in defining your exposure and your outcome or your research investigation, your research question, and that heterogeneity and definitions and the vast scope of conditions that can be included in this multimorbidity [04:00] investigation really introduces a lot of complexities in both developing the study design and also when comparing the results of various studies.

Heather Allore, PhD, MS, MA: I think that's really helpful and it kind of really steps us into talking about different types of study designs. In your presentation, you broke these down into some broad groups. Could you maybe just walk us through some of these groups? 

Basia Diug, PhD: Be my pleasure. So when we look at study designs, there are really - you can do quantitative research and you can do qualitative research. You could also do mixed methods. So you can use tools from both. But when we break down study designs specifically, we break it down in observational studies, which follow the natural course of the disease or experimental where you are intervening. I'm gonna talk about observational because that breaks down even further into what we call descriptive studies and analytical studies.

Descriptive studies include things like case studies or case series or ecological [05:00] studies. They don't have a comparative group, so they're looking more at trends and they're very much hypothesis generating. Whereas analytical studies have two groups and depending on which study design you choose within them, which is cross-sectional, case control, or cohort study you have those three choices, really. Depending on what you choose to meet your research question and your definition, you would then apply that methodology within that category. 

Heather Allore, PhD, MS, MA: That's wonderful. And I know much like my experience with geriatric clinical researchers, and I'm sure you have this experience, often when someone is starting to explore these relationships in observational studies, they may first use a cross-sectional design.

I know you talked about this in your, your presentation, but could you share some of the purposes and strengths of cross-sectional study designs?

Basia Diug, PhD: [06:00] Absolutely. We have this thing called the pyramid of evidence. And as you go up the pyramid of evidence, the quality of the evidence you collect differs. So when you choose your study design like cross-sectional, you have to understand the strengths and limitations of that study design compared to the mode that you can capture or the way you can actually achieve the data collection you're gonna do, but also the other study designs available to you.

Cross-sectional study is very useful, and like you said, it's really good for hypothesis generating and used really broadly by epidemiologists. It's a good study design that compares two groups. It measures the presence or absence of exposure. It can measure prevalence of disease and outcome because what cross-sectional study design really does, it assesses exposure and outcome at the same point in time in individuals.

So when I mean that, I mean, it's a snapshot in time of what is happening. So this might be time interval or a [07:00] single point. And although this means that it can be hypothesis generating, one of its main limitations is that there is no direction of time. So the direction of the relationship between an exposure and an outcome cannot be determined.

My best example is to talk about it as a chicken and an egg concept. You are maybe going into a clinic and you are asking - you've defined your case and you've defined your controls, your two groups, but you're asking them the same questions at that same point in time. So you can't actually determine if the exposure preceeded, the outcome or the outcome caused the exposure.

So these are some of the main limitations there. I mean, some of the strengths of a cross-sectional study design and why it's used so well are what it captures in terms of prevalence in using existing data, but also that it's inexpensive and relatively easy to perform, which allows us to think about what other studies we can do after this.

Heather Allore, PhD, MS, MA: I think that's [08:00] very helpful because I think that hypothesis generation is very helpful, especially for early stage investigators kind of describing the natural course of multimorbidity. And you are right, it is very difficult if someone's already presenting with three or four conditions. We don't necessarily know the sequence those occurred, especially if we're using some sort of national survey or claims data. In your slide presentation, you provided an example of a cross-sectional design using some survey data. Could you talk to us a little about that? 

Basia Diug, PhD: Yes, certainly. One of the examples that I discussed was using actually a major dataset that's located in the United States. So it's called the NHANES Study, and it started in 1988 and it's called the National Health and Nutrition [09:00] Examination Survey. And the aim of this, large study and project is beyond just this cross-sectional study, but it's really to assess health and nutrition status in the United States, and the population is 57,303 non-institutionalized individuals.

The cross-sectional study takes a glance and takes a snapshot of that from the year 2013, 2014, and uses this multi-stage survey to see what's happening in that one point in time in those patients. And what they've done in this project is they've actually, their research question is, what is the link between health insurance status while patients are self-reporting multimorbdities?

So the exposure here was many because it was a massive, massive questionnaire. But when they do the cross-sectional study and their research question, they look at health insurance and multimorbidity defined as two or more chronic conditions. So they look at that [10:00] snapshot in that period. And then as a result of that, they actually find that of those participants who had had health insurance, 59% had a multimorbidity, so two or more of these indexed chronic conditions. Now they're using an existing dataset, which means, you know, data collection has been completed. They're using the timeframe that they've chosen 2013 to 2014. 

And so a lot of questions come out of this. It raises the question, does having health insurance mean you will have multiple morbidities, so that's why you go and get health insurance? Or does having multiple morbidities mean, you know, which way does that relationship go? So, or does having multiple means mean you'll go out and get health insurance? Either way, this cross-sectional study design allows us to have, look at that trend of how many, but it doesn't let us infer the direction of the relationship.

So it's important that we consider and interpret these [11:00] results and the confounding effects that would impact this relationship between health insurance and multimorbidity. So you know, what other relationships or confounders are actually affecting that? Could it be income? Education level? It's possible to propose that, you know, a higher level of education and a higher income would mean that you're more likely to afford health insurance. So it's not necessarily clear how that relationship that we're observing between health insurance and multi morbidities in the study is actually linked and, and more investigation would need to be done. 

Heather Allore, PhD, MS, MA: So I think you bring up some really important strengths to really start to explore ideas and then lead us into what is the direction of the association and that importance of temporal associations.

So that's often why after a cross-sectional study design, someone may move on to [12:00] a cohort study where they can be planning ahead and trying to get temporal associations. In your presentation, you did give us some details on analytic cohort studies. Could you give us a, a, a little bit more just a, a refresher on the purpose, strength, and weaknesses of these?

Basia Diug, PhD: Oh, definitely. So a cohort study, as you know, really great study design. It's at that top of that close, that top of the period of evidence. And I think as an epidemiologist, everybody wants to do cohort studies and case controls, but again, it's about picking that right study design. And it really builds on the cross-sectional one, because you are comparing two groups and you are comparing an exposed group and a control group, but you're starting with a healthy cohort or a healthy group of participants, and you start with the exposure and then you follow them over time to see what develops, you know, is that a disease or an outcome?

The direction of the study is always [13:00] forward in time. And the key words that I would say is that follow up period, because cohort studies unlike cross-sectional studies, which are a snapshot, are about that forward direction where you're following up people for a long time to see what happens and to capture their outcomes.

Cohort studies as a result of that are excellent at providing evidence for causation or estimating risk for developing the outcome of interest in relation to a specific exposure. So essentially, in a cohort study, investigators would conceive their study, we'd take a sample from their population, collect some baseline exposure data before any of them have actually developed the outcome of interest. They have to be a healthy participant group.

They're followed up over time and you know, we start with these disease-free participants and then we see what happens. The incidence of the outcome is then compared between those exposed and those not [14:00] exposed to the risk factor of - that we were interested in the study and it's a really good study design.

As with all study designs, there are lots of challenges with a cohort study. Despite its rigor, there are some limitations. There is, like with all study designs, there are different types of bias that impact on the study. There is also loss to follow up where, for example, participants, you're following them over years.

So there's this time period that you have to be committed to this, this study. So it takes quite a while and, and participants might not be involved for the whole duration of the study, so you might lose them. They might move, they might, you know, no longer want to be part of the study. It's not a really good study designed to look for rare outcomes because you'd need a really large population to begin with, so it's insufficient to really study rare combinations.

It's quite expensive and time consuming due to its directional characteristics, and you need to really think about the [15:00] causal inferences that you're making as with all study design. 

Heather Allore, PhD, MS, MA: One of the things that has been of of current interest are the social determinants of health, and they're really being recognized as contributors to multimorbidity and negative geriatric outcomes.

Much like your example of the cross-sectional design with health insurance status. That could be related, that could be a socioeconomic indicator. In your presentation you provided an example of a cohort study design that was looking at some social determinants of health. 

Could you provide us with some reflections on this study that had a particularly long follow up?

Basia Diug, PhD: So the study that you're referring to is called the Twenty-07 Study, and it began in 1986 and was concluded in 2007, 2008. So, as you said, a particularly long follow up. The aim of the study was to [16:00] investigate the reasons for differences in health by socioeconomic circumstances, gender, place, and where people lived, age, ethnicity, group and family type. And this was located in the UK. The study design was a prospective cohort, so they started in 1986, followed up this group over time, going forward, over 20 years.

They decided to use three cohorts, which makes this quite a complex study design because they've essentially got a group that starts at 15 years of age, 35 and 55, and they're following all three of these groups for 20 years to see where they are 20 years later. And therefore capturing a 60 year lifespan. So it's a very interesting study design, but also very complicated. 

So the best way to look at study designs and, and if the listeners have the opportunity to have a look at the additional resources provided, is to have a look at the diagram. So the direction of inquiry for cohort [17:00] study is always forward. They've taken this population, they took this sample of four and a half thousand participants. And then I've, when describing it, in my example, I've taken a component of it to really show exposure and non exposure. And they've looked at the most deprived socioeconomic level as their exposure status and then followed the participants up over time.

And to have a look at what their multimorbidity outcomes were. Did they have multimorbidities or not depending on their socioeconomic level? In this study, each participant was interviewed by a nurse and also self-reported outcomes. Interest and multicomorbidity was measured using very much a set criteria of 20 to 40 conditions.

All of this was set well and truly before the study was actually started. To ensure that they were capturing this information consistently, and they could compare it over that 20 year period. So the [18:00] results that they found were, when comparing the most and the least deprived socioeconomic groups, disadvantage was associated with an increased risk of developing multimorbidity. So they had a 1.46 odds ratio and quite a, a tight confidence interval. So it's something to to really consider. It's a really complex study and within that, all of the other demographic characteristics can also be examined, because that's one of the benefits of a cohort study. You can look at multiple exposures and see how they actually impact your outcome over time.

Heather Allore, PhD, MS, MA: You know, your example is very fitting because I think one of the things geriatricians often reflect on is someone who's 80 year old today in 2023 may look quite different than somebody who was 80, 20 years earlier, or 80, 40 years earlier. [19:00] So these age and cohort effects can really highlight differences in especially early life experiences and unmeasured factors that may be more pronounced and leading to outcomes in in later life.

Would you like to say anything about cohort effects? 

Basia Diug, PhD: I think multimorbidity is, it's, it's, it's a challenge. It's, it's a very, like you've just said our participants or our patients are changing. Their multimorbidity status will change over time. They'll change across their lifespan and we'll also link with their treatment.

I'm actually at pharmaco epidemiology or drug epidemiologist, and you see how medication schedules and how treatment changes. But similarly, multimorbidity would change. So having a cohort study that tracks over time really gives an insight and captures a quality of data that allows us to really have an [20:00] understanding, particularly when you've got it from these three cohorts from 15, 35, and 55 years really gives you a glimpse into changes across a very long period in time, which, you know, which is something that would be rather hard to capture at an individual level. And I think that speaks to the high quality and the necessity of having observational studies as one of the study designs we can choose as medical researchers to answer a research question. 

In many cases, when I teach study design. The idea that randomized control trials the gold standard is this common saying, but I think for me it's choose the right study design for your research question. And cohort studies are very rigorous, provided they are designed well or any of the observational study designs can really enhance a medical field or a research area of interest at a high level. 

Heather Allore, PhD, MS, MA: I couldn't agree with you more, and [21:00] I think your last example also leads us to understanding the importance of sampling in whether you're doing survey designs, cohort studies, that really trying to learn more about how do you sample and what is your sample representative, what is that target population. Which is all an excellent segue to follow up with another module called Inclusion Across the Lifespan: Health Equity and Vulnerable Populations, which is an excellent resource to learn more about working with these populations and designing, whether you're observational or interventional research, taking in account, really thinking about your sampling.

And just as you pointed out, Prof. Diug, about trying to avoid selection [22:00] bias. In your slide presentation, you've provided many additional examples that we don't have time in this podcast to go over. But there's resources for listeners who'd like to learn more. And since you've, you've taught so many people in so many countries, is there any advice that you could share with listeners who'd like to be designing observational study for older adults with multiple chronic conditions?

Basia Diug, PhD: My number one advice is to just talk to an epidemiologist before you start. I think if you have existing data that you'd like to use or you have, you have a research question, discussing the strengths and limitations of the observational study designs available to you. Talking to a biostatistician before you start collecting data or before you dig into it, and understanding what kind of analysis potential there is. And really understanding the strengths and limitations of [23:00] each observational study design, particularly with such a complex variable such as multimorbidity, is really necessary. These conversations will help you plan a better and choose a more, more rigorous and robust study design.

And in the end, you know, we have to respect our participants' time. We have to respect our participants contributing to our medical research and our growth and knowledge, and we want the best outcomes in our research. So these are the key considerations that are required to have the best outcomes with your medical research, questions for your participants and for the body of research and medical morbidity. 

Heather Allore, PhD, MS, MA: Thank you for joining us, discussing observational study designs with Prof. Basia. Diug from Monash University. To learn more about observational study design, see her presentation, which includes a listing of valuable resources.