Health disparities among lesbian, gay, and bisexual people in the CLSA

All right, so let's get started. Um, welcome everyone. I'm Sophie Hogavine. I'm the data access officer for the Canadian Longitudinal Study on Aging or the CLSA. Thanks for joining us for this CLSA webinar which is titled Health Disparities Among Lesbian, Gay and Bisexual People in the Canadian Longitudinal Study on Aging: Methodological Considerations for CLSA Data Users. Before we begin, I want to acknowledge that the CLSA National Coordinating Center and McMaster University are located on the traditional territories of the Missaga and Hoden Nations and within the land protected by the dish with one spoon wal agreement. The University of Manitoba's campuses and research spaces are located on original lands of the Anishnabag Inino Asinino Dakota Oate Da and Inuit and on the national homeland of the Red River Matei. UM recognizes that the treaties signed on these lands are a lifelong enduring relationship and we are dedicated to upholding their spirit and intent. We acknowledge the harms and mistakes of the past and the present. With this understanding, we commit to supporting indigenous excellence through active reconciliation, meaningful change, and the creation of an environment where everyone can thrive. Our collaboration with indigenous communities is grounded in respect and reciprocity. And this guides how we move forward as an institution. As attendees of this webinar, I encourage you to continue your learning following the webinar and to acknowledge the original inhabitants of the lands where we currently have the privilege to research, live, and work wherever that may be. Before we begin, uh we have a couple of housekeeping points. Everyone but the presenters will be muted throughout the webinar. If you need to change or test your audio uh during the webinar, you can click on audio settings on the bottom of the Zoom window. At the end of our presentation today, there will be a question and answer session. If you have a question for the presenter, you can post it in the Q&A box located in the bottom toolbar. So, this is different from the chat box. um it's specifically called the Q&A box and you might need to click on um the three little dots at the bottom to see more options which include that Q&A box. Um the questions will be addressed at the very end of the webinar and your questions will be visible to all attendees. If you have any technical trouble concerning the webinar, that's what the chat box is for. So please use that to communicate with our webinar team. A feedback survey will be launched at the end of the webinar and we invite you to complete it after exiting the Zoom session. This brief survey provides us with important feedback we can use to plan future CLSA webinars. All right, so that brings us to today's webinar. Again, it's titled Health Disparities Among Lesbian, Gay, and Bisexual People in the Canadian Longital Study on Aging: Method Methodological Considerations for CLSA Data Users. This is being presented by post-doal research fellow Nicole G. Hammond from the University of Manitoba. So, Nicole G. Hammond is a psychiatric epidemiologist and a post-doal fellow with research interests in health equity and determinance of mental health across the lifespan. Her work integrates the fields of psychiatric and life course epidemiology, population health equity and mediational and causal inference methods. She examines how early life adversity and social determinance shape health across the life course. She is particular particularly interested in the mechanisms by which early life adversity and social determinance affect health to isolate factors amunimal to intervention or not when the opportunity for prevention has elapsed. Her work also directly informs upstream early life stress prevention and health promotion initiatives. She's recognized as a merging scholar in 2024 with the Canadian Research Data Center Network and her post-doal research at the University of Manitoba is funded through a CIHR Canada post-doal research award. All right, Nicole, I'll turn it over to you to get set up and share your screen. >> Thank you so much, Sophie. Okay. Yes, I will get set up here. Uh, I believe I have the right screen. So, if everyone can just bear with me for a moment. Sophie, can you can confirm that you see my slide deck? >> Yeah, looks good. >> Perfect. Okay. So, thank you everyone. Um, it is an immense privilege to be able to be here today uh with you all and share some of the research that I've um conducted uh using the CLSA. Uh, as you know now, my name is Nicole Hammond. My pronouns are she, her. I would also like to say happy pride month. I hope there's good weather for the Toronto um pride parade later this month. Uh hopefully a little bit better than what's happening in Toronto at the moment. We've had a fair bit of of raining today, but it the sun is is striving to get out. Uh so fingers crossed. And as well, I would like to say uh happy National Indigenous History Month. And with that in mind, here we go. Making sure my slides are working. uh you've probably uh caught on to the fact that I've mentioned Toronto. So even though I'm conducting my post-docctoral fellowship or um participating in post-docctoral fellowship with the University of Manitoba uh the College of Community and Global Health, I'm I actually live and work in Toronto. So just making things a little bit tricky. Um so for purposes of land acknowledgement, I would also like to incorporate one for Toronto. I live and work on the traditional and treaty lands of the Missosagas of the Credit known as Adobe in Miti Sag language known as place of the alders. Uh this region is situated along the Humber River wershed and historically was a place of connection for the Anish Nav, Odinosone and Wend peoples. For my own personal journey of reconciliation, um, one aspect of my own journey has included, um, making an effort to engage to a greater extent with literary works of indigenous authors, including exposing my young children to indigenous authors. And so I took the opportunity for today's talk to share two of our uh, favorite um, books on this slide for anyone who might be interested. For those of you who have not already started your your own personal journey towards reconciliation, I challenge you to consider what that might look like for yourself andor your family. For our agenda today, I will start with a brief experience of well recounting a brief experience of my work with the CLSA over time. We'll delve into the focal paper that uh Sophie already introduced and we will throughout this talk consider some methodological considerations for CLSA data users and I will briefly wrap up at the very end. I will also pause a few times perhaps more than normal for me uh throughout the talk. My youngest daughter brought home a cold uh from daycare a few weeks ago and it's been making its way throughout our our home. So I'm you might notice that my uh voice uh will come and go a little bit so I'll take some breaks. So my journey with the CLSA it originally uh it began in the spring of 2018 when I was concluding my MSE and community uh in epidemiology and community medicine at the University of Ottawa. At that time I was hired uh with the healthy aging research program lab uh the HARP lab which is now located at the University of Ottawa um which is the originating lab of the focal paper for today's talk and this is directed by Dr. Ern Stingcomb who is a co-investigator of the Ottawa data collection site for the CLSA. So I joined the HARP lab as a data analyst in 2018. Uh the following year I began I'll use my cursor on the slide too to highlight things as we go. Um the following year I began my PhD in the school of epidemiology and public health at the University of Ottawa. And then in a little later that fall uh my first CLSA related publication came out. Uh Dr. Stinko and I examined health behaviors and social determinants of migraine in the CLSA. Since then, we've had an additional nine CLSA related publications together, not including the focal paper that was presented uh the end of last year. And it's at this moment that I should probably pause and just highlight that um I've had the tremendous opportunity to have a great deal of experience working with CLSA data. However, I'm not an expert in aging or two SLGBTQ plus um research specifically. my all of my graduate level training including my doctoral research used different data sets mostly looking at uh child and youth mental health. So I've had a tremendous opportunity to learn um a ton on these topics through my engagement with the HARP lab but most of my independent research um does not follow this specific line of inquiry. So I'm coming at today's talk from the data analyst help uh data analyst uh hat rather than perhaps my own independent lead investigator. So just a little of additional context coming up to present day I started my post-docctoral fellowship with the University of Manitoba in January and then just recently officially had my PhD recognized at the University of Ottawa. So uh big celebration there. So again I'm coming at today's talk from the perspective of a data analyst. The CLSA very clearly offers a wealth of information as it pertains to aging. And throughout my talk, I'd like to highlight some strengths and challenges um that we've encountered in terms of decision-m uh for some of the uh variables that we used in the focal paper um that I will jump right into. So this health disparities paper was published in the Canadian Journal of Public Health the end of last year. Our focus was looking at uh baseline and follow-up data collection in the CLSA. So this is actually actually an update to an original 2018 prevalence paper published by Dr. Aron Stingcom and colleagues in the uh Canadian Journal of Public Health where they demonstrated at baseline um health disparities among LGBT plus people. And so we wanted to update and extend these findings u by looking at changes in health disparities over time and whether as well there's an increased risk of health disparities. So I will break this down further as we go. But just to set up a little bit of background for this paper. It is extensively well documented that health disparities exist among sexual minority people. Most of this has been shown in cross-sectional work specifically. There is a smaller body of uh longitudinal research and an even smaller body of work looking at aging sexual minority adults. And much of the work that's available has focused on mental health trajectories with less attention given to say physical health concerns. For example, for this paper and the collection of work um stemming from the HARP lab, there have been several theoretical frameworks that have guided our approach. Some of those are listed here. Um but these theoretical frameworks have guided our understanding of how minority stress may function as a determinant of health. Altogether these uh frameworks pose it that members of minority communities are more likely to experience stigma discrimination as well as other minority stressors such as structural disadvantage. Altogether these negative experiences accumulate over time to influence health trajectories and inequities. An alternative way of conceptualizing this would be that um the experience of minority stress accelerates biological aging as a result of weathering a greater number of stressors over time. So our objective was to use the CLSA to better understand the timing of health disparities among lesbian, gay, bisexual, and other sexual minority identifying adults in the CLSA. Again, an update to that original 2018 prevalence paper and an extension of those findings. We wanted to determine whether these health disparities persist across midlife and older adulthood and whether there's a increased risk of new onset disparities over time. Why is this topic important? Well, older sexual minority adults represent a hard-to-reach population. population-based aging research often excludes sexual and gender identity measures. It really hasn't been until the last, say, decade, maybe 15 years, that we're really seeing sexual and gender identity measures incorporated in surveys, especially at say the national level. I've already noted that there's a lack of longitudinal evidence and data is often not disagregated by age groups. What I mean here is that a lot of work um would look at say adulthood, so ages 18 plus, and not separate periods of adulthood, younger adulthood, adulthood versus emerging or midlife or older adulthood. And we know that different cohorts have different life experiences and different periods of the life course or the lifespan um come with different aging related needs. Together, these challenges contribute to the reduced visibility of sexual and gender minority older adults in health re health research, which ultimately translates to um poor health services planning and resources for these individuals. On the left hand side of your slide, I'm showing you the original prevalence paper and then on the right hand side is our our update and extension. Our hypothesis was that LGBT plus adults will experience greater health disadvantage. An important point of clarification is that it's not that a sexual minority identity in and of itself leads to poor health. Rather, these health outcomes are thought to arise from the experience of minority stress associated with living within a heteronormative society. our key explanatory variable of LGBT plus. Um, I'm going to outline how we conceptualized it for this focal paper. We recognize that sexual orientation can be fluid and can change over time. Coupled with the fact that sexual orientation was captured slightly differently between baseline and first follow-up in the CLSA led us to conceptualize it as um inclusion of individuals who endorsed a sexual minority uh minority sexual orientation at one time point but not another. For example, identified as LGBT plus at baseline or first follow-up or vice versa. as well. We included individuals who identified with an LGBT plus identity at baseline and first follow-up. And we included individuals who reported that they did not identify with any of the sexual orientations listed. This was an expanded answer option included at first follow-up in the CLSA. This leads us to our first methodological consideration. when we were doing a deep dive into the data and understanding um and figuring out how figuring out how we specifically wanted to conceptualize LGBT plus identity for purposes of this work. We decided to look at sexual orientation as it was reported at baseline the top half of your slide here and then at first follow-up. And when we were looking at differences in question reporting, which I will continue to highlight over the next few slides, and look at the number of people who identified, we see interestingly that there's a greater proportion of people identifying as LGBT plus at first follow-up than there was at baseline. Um there's something I wanted to mention here. I it's specifically that these numbers are after exclusion of missing data on uh the variables that we included in our model. So, if we didn't exclude data with um people with missing data, these numbers would be a bit higher. Nonetheless, the same pattern of findings emerges with more LGBT plus people at at first follow-up. So, I just wanted to to note that for anyone on the on the call who might have personal experience looking at these variables in the CLSA and and wondering why there might be differences in sample sizes. So, we noticed that there more LGBT plus at first follow-up than at baseline, which caused us to pause and reflect for a moment and discuss in the lab because, as with any good cohort study, over time there's attrition, which I've shown here. Uh, at baseline, there's over 51,000. When you include the follow-up periods of interest for us, we retained approximately 45,000. And then further, the sample size diminishes after excluding people with missing data on our variables of interest. So with this attrition over time, as expected in any longitudinal cohort study, we're still seeing more people identify at first follow-up. So this is where in the lab we really went back to basics and and looked at question wording and we're trying to understand what might be happening, what might be what we might be capturing versus not. And so I've taken the liberty of including the exact question wording for sexual orientation uh at baseline on the top half of your slide along with the corresponding numbers I presented a few slides ago. And then on the bottom half of your slide, you'll see the question wording uh for first followup. And what we landed on as a lab was that how we ask about sexual orientation matters both to the data and to the people. I've highlighted in yellow some key details regarding the follow-up one question wording and we see greater intentionality and greater transparency in the collection of data. Specifically, we see there's a preface included with respondents um told or participants told by correctly addressing sex, gender identity, and sexual orientation, we have an opportunity to examine the potential impact on aging and health. So participants are told directly why collection of this um personal and sensitive information is important and perhaps how it might be used as well as that expanded um answer option that I noted earlier with does not identify as any of the above responses. So for these reasons that preface that intentionality that transparency perhaps that greater inclusion of of identities um we see a greater um LGB plus identification at first follow-up. And I encourage anyone specifically interested in best measurement practices um in this particular case to check out the work of Dr. AJ Loick who is really a leading uh health researcher in this area. So now we've we've made sense of our data. We have a great understanding of um of how we were conceptualizing it. We have further support for that that inclusive modeling of uh including say changing um identities over time between the baseline and first follow-up. And so at the end of the day we retained for the purposes of this focal paper approximately 1,200 uh LGBT plus people for purposes of analyses. Our stratification variable was sex. So we know that there are uh differences in health outcomes between males and females. own to account for some of these differences and align with um the broader existing literature in this field as well as our own previous work including that 2018 prevalence paper. We chose to stratify analyses by sex. Though we recognize that this is not a perfect measure for our outcomes. I've um for purposes of our talk created four higher level categories um because we captured different domains. We looked at mental health outcomes, uh, physical health outcomes, behavioral health, and healthcare utilization. Mental health outcomes were mostly physic, uh, participant report of physician diagnosis, though we did have one self-rated question, poor mental health. Physical health outcomes, we had four higher level categories, uh, cardiovascular disease, but for example, but within these categories, we modeled separate outcomes. So stroke was an indicator of cardiovascular disease that we modeled. Respiratory conditions included um say asthma for example. Neurological disorders included memory, migraine um memory problems specifically and other medical conditions included cancer among others. For behavioral health, we captured past month cigarette use, alcohol use, which was really a measure of excessive drinking, specifically whether respondents reported on average over the last year um engaging in binge drinking once a month or more. Exercise was lack thereof, so lack of exercise to increase muscle strength or increase endurance. Uh poor or disrupted sleep. and home uh renting versus home ownership. So renting being just that or an alternative housing arrangement versus home ownership as the reference, pardon me. Healthcare utilization uh captured past year use of services such as seeing a family physician, uh medical specialties such as psychiatry, uh psychology, physiootherapist or related profession, as well as whether or not respondents reported having an overnight stay in a hospital. So we had lots of of outcomes to model fortunately. Uh for our statistical analyses we conducted multivariable models. We adjusted for baseline age, household income and highest educ educational attainment of the respondent. We conducted modified plus regression with robust sandwich error variance to obtain a prevalence ratio representing study period prevalence whether the outcome occurred at baseline first or second follow-up. as well. We modeled the relative risk or the risk ratio, looking at whether or not among those free of the respective health outcome at baseline, whether they were um whether LGBT plus people were more likely to develop that health outcome over time. And I'll go over those again in just a moment um with a visual aid, which I personally find helps me a great deal. And data was excluded analysis wise. What I mean here is that after we removed people with missing data on our focal variable, our key explanatory variable, our stratification measure and confounders, we allowed each of the models to vary uh in across sample size uh for each outcome to maximize statistical power because some outcomes were less prevalent than others and we just wanted to do what we could to to maximize sample size. So here's a visual aid I mentioned. So study period prevalence was whether or not a respondent reported the outcome at baseline, first follow-up or and or second follow-up. Whereas incidence was measured as by removing anyone with the say respective health outcome of interest at baseline. Say it was cancer. We removed people who reported cancer here and then modeled new risk of cancer over time. For our key results, we found that LGBT plus older adults reported a greater proportion and incidence of most health related outcomes. In the full sample, we found that lack of exercise and sleep difficulties were highly prevalent. And for those interested, these descriptive findings are presented in tables one and two of the main text. Jumping into the um uh to the main findings by all of those four categories of outcomes. We'll start here in this table with uh mental health and physical health outcomes. They all appear on the left hand side of your slide. And then across the top we see that the results are stratified by females versus males. And we have the prevalence ratio as well as the relative risk modeled. And I will give you like a higher level summary of the most consistent findings um that we observed. So there's a very clear pattern of results with LGBT plus females and males reporting a higher prevalence and incidence of all mental health outcomes over the study period with the strongest findings observed for depression. Moving into respiratory conditions and this finding like the last one uh aligned with that original 2018 prevalence paper, we see um an elevated prevalence of respiratory conditions for both LGBT plus females and males. Though in most cases here um a lack of new onset risk of these respiratory conditions. I should take a moment to highlight because I meant to do this earlier that all statistically significant findings are bolded for ease of understanding. So if it's not bolded, it means it was not statistically significant at P less than .005. So for respiratory conditions, we see this clear pattern of an elevated prevalence ratio for both LGBT plus females and males, but lack of incidence for the these health outcomes, which to us suggests that LGBT plus people were more likely to enter the study, the CLSA, with these health disparities already present rather than developing them over the study period. For cardiovascular disease, we did see some key findings. Though there are some sex specific differences with LGBT plus females more likely to have an elevated prevalence and increased risk of angina versus for LGBT plus males. This was um the same pattern of findings was observed for stroke or cerebo vascular accident. So now we'll move into our health rellated behaviors and healthcare utilization measures for negative health behaviors. We observed uh relationships for cigarette use um elevated prevalence and incidents for LGBT plus females and only an elevated prevalence for LGBT plus males. Nonetheless, I think this overall pattern of findings, mostly consistent um pattern of findings suggests that perhaps um smoking sessation programs may wish to consider um the potential influence of minority stress. We we could not unfortunately differentiate between former smokers who reinitiated use versus first-time smokers in our relative risk or incidence models. But again on the whole I think we these findings lend support to the fact that smoking sessation programs may wish to consider minority stressors as contributing uh factors. I would also like to highlight a very positive relationship we observed for LGBT plus males with a protective effect. So to make this easier to interpret um I like to flip it. So I'll say um LGBT plus males were more likely to report engaging in physical activity to improve muscle strength and increase endurance. So that was a very positive takeaway that we we see we see here for housing. So this was renting a home versus home ownership. We see again uh uh health disparities existing for LGBT plus people more likely to report renting over the study period and for LGBT plus males more likely to report over the study period going from home ownership into a rental situation. And we're seeing lots of this on um you know in the news etc. Um the cost of living um how difficult it is to to own a home today and how this is a stressor for many individuals. But we see this health disparity um present here in this work. We also observed a strong or robust finding between uh LGBT plus identity and accessing psychology over the study uh period. It's possible that this is really an artifact of the findings I reported on the last slide that because LGBT plus people were more likely to report a mental health diagnosis um they need to access someone to be able to obtain that diagnosis. So it's unclear whether or not um accessing psychology is because of that specific need or whether LGBT plus people might be more likely to be support seeking in general. um perhaps future work can can disentangle that. And just coming back to renting very briefly, I would like to highlight that this can come into play for um retention in cohort studies over time. Maybe less of an issue today in our digital world where we're generally less paper orientated than we were a while ago. But for any survey that relies on paperbased methods to remind people to participate in a study um etc um say codes to be able to log in and access a survey. If LGBT plus people are more likely to rent and perhaps have increased housing insecurity or more likely to move, etc. Um you might be less likely or less able to retain those individuals over time if you're relying on paperbased methods. So just something to keep in mind um uh for work that relies on paperbased reminders. a higher level summary of our work um before we move into some more methodological considerations. This was the first national longitudinal study of older Canadians to offer a comprehensive examination of health outcomes of older LGBT plus people. LGBT plus participants demonstrate several differences in chronic health, health rellated behaviors, and healthcare utilization as they age. And in line with our study hypothesis, a lack of new onset risk for certain chronic conditions suggests that many health inequalities may be apparent early again perhaps um prior to study entry. Though there was that notable exception of mental health concerns with elevated prevalence and increased risk of developing new onset mental health concerns over the study period. So with any good study uh there are many limitations um but the most notable one um for us was survivor bias which was two-pronged for us and I will explain what I mean um first participants lost to follow-up were excluded over time as I showed you in those earlier um you know participant flow diagrams we started with approximately 51,000 people and then as a result of attrition we lose people and then we also exclude people with missing. As I've as I've shown you, LGBT plus people have greater health disparities and in that original 2018 problems paper and based on other work um we know that as well. So if LGBT plus participants are more likely to enter the study with in worse health, they may be more likely to die in the short term or to have premature mortality or they may be more likely to not want to participate due to poor health. So it's possible that um our estimates represent an underestimate of the true strength of the relationships. Um on the other side there is evidence originating from the United States. This Miketta and colleagues paper used health administrative data to demonstrate that LGBT people have premature mortality. The leading causes of death were cancer, respiratory disease, suicide and cardiovascular disease. Suicide is a leading cause of death among young persons. So if LGBT plus people are more likely to die by suicide and at earlier ages, it's possible that some people may have not even had the opportunity to take part in the CLSA at baseline um because they could not be surveyed. As a result, we believe that that situation would also underestimate um the true magnet magnitude of the observed relationships. Other limitations include that uh longitudinal survey retention is socially patterned. I see this in my own work using um younger cohort studies. We see that people who are more likely to remain in studies over time are more likely to be white um to come from higher socioeconomic classes, to have better health or less stress. And there's also evidence of this in other uh national surveys in the United States for example um of people with um similar ages. So it's not just us who see this. Others are are also documenting it as well. With all of this in mind, it led us to question, you know, really who we might be accurately capturing in our results. um who are we basing these conclusions on? So, we did um take the opportunity to do another data exploration and then to look at those who were retained over time versus those who were lost to see if we could again just assess the potential impact on our our findings. And we found that those lost to follow-up were more likely to be older at baseline to be males. There was a slightly greater proportion of respondents lost who identified as LGBT plus at baseline than those people who identified as heterosexual. But notably, 29.2% of respondents lost did not respond to the sexual orientation question at baseline. Unfortunately, we cannot distinguish between say survey refusals where there's just like general non-compliance across the board versus people who might not wanted might not want have um revealed their identity because um they were questioning their sexual orientation for example or they just did not see themselves represented by those um the answer options given. Remember that baseline question had less intentionality and transparency in it than the than the follow-up. So we cannot distinguish um those details but it causes us to pause and question and reflect. There are also other reasons people may not wish to reveal their sexual orientation on a survey. Um LGBT plus people face very real um real and perceived threats to their personal safety and security. These threats are ongoing. Um but I would like to highlight the fact that the entire baseline data collection period from 2011 to 2015 was basically sandwiched by two major Supreme Court decisions uh here in Canada in 2005 and in the United States in 2015 when the legal right to same-sex marriage was affirmed and in the United States this was not a unanimous decision. So there are very real reasons why people may not wish to disclose their sexual orientation in surveys. Again, this aligns with um what we're seeing elsewhere internationally, including in Australia. Uh Campbell and colleagues used two different age cohorts um from the Australian Longitudinal Study on Women's Health, and they found that older women living in Australia who identify as a sexual minority who or who did not wish to disclose their identity were more likely to exit the survey over time. So what can we do about it? This is really where I get excited and I'm starting to incorporate some um some methodological practices um that I implemented in some of my doctoral research um which was conducted with Dr. Ian Coleman at the University of Ottawa. So in the absence of survey weights that assist in maintaining sample representativeness and account for attrition because we don't have those in the CLSA, other longitudinal cohort studies um sometimes offer those. um it's possible that they might be in the works etc. I don't know. Um but for us we did not have access to those. What we can do is use what is known to as propensity scores or inverse probability of waiting. Pardon me again. Um propensity scores can be used to address selection bias specifically that those who remain in the study over time are systematically different um from those who the sample represented initially. And uh my recommendation would be to compute stabilized inverse probability of sensoring weights also known as IPCW weights. These um these propensity score approaches have been around for a long time. Um they're used in um in epidemiology often to strengthen causal inference. And then I've also included some papers here where you can see that they're talk um people are talking about using um inverse probability of waiting to and to tackle missing data and as well to handle attrition in cohort studies. So these have been around for a while. These methods um some very preliminary preliminary results of ours using propensity score weights to account for um attrition will be presented at the upcoming Alzheimer's Association International Conference in the United Kingdom. So anyone on the call who finds themselves at this conference, don't hesitate to check out our work. Um but I will briefly show you some um again very early results of ours using propensity score weights in just a short moment. Specifically, I'm interested in looking at whether the experience of early life adversity or adverse childhood experiences before the age of 18 years, such as parental separation or divorce and spanking prospectively predict um worse memory in the CLSA. And so we're again um computing these stabilized propensity score weights to account for attrition to account for differential selection over time to see if it has an impact on our findings. And I'm showing you some of the code on the slide actually from a different paper of mine. But really what you're doing at the end of the day, the basics is just you're building a logistic regression model to estimate the predicted probability of the exposure or um the sensoring measure or the like the possibility of loss to follow-up over time observed for a particular person and then using that predicted probability as a weight in your subsequent analyses. So for anyone who has experience with using population-based survey weights um offered um say you're using STA you're survey setting your design to account for unequal selection probability the complex sampling frame. So you're survey setting results you're you're waiting all of your future analyses by that weight. In contrast in this method you're developing the weights yourself to account for the potential differential uh loss to followup and then you just do the same thing. you're just waiting your future analyses using what the propensity score that you've derived. Very very short um sum higher level summary of what that looks like. For some of our preliminary anults were uh results. We're looking at delayed recall at follow-up 2 delayed memory recall captured with the uh ray auditory verbal learning test. And you see on the right hand side of your slide that once we account for attrition with these propensity score weights, we do see a slight strengthening of the relationships um becoming in at least one case becoming statistically significant. Though at the same time the magnitude of the effect or the odds ratio isn't changing substantially which likely reflects the fact that there is some selection bias happening. However, it's not in exerting undue influence of our results. Um, it is certainly possible that other results could be more um greatly impacted by the possibility of selection bias. Um, but again, we're still in early days and I I hope to present more on this at a later date. Um, another uh important piece that's come to mind is these weights are only as good as say the information that you have on people at baseline. So thinking back to that example of LGBT plus people being more likely to die by suicide, if for example, if they were not a if certain people were not even able to be included in the survey at baseline, then these weights can't account for those lost people. So um there is some other bias that could still affect your findings. Um and so we're really only accounting for people for whom we had initial data and then were lost over time. Other uses for propensity scores include balancing exposure groups on confounders. Um this is where uh observational based studies can begin to mimic the effects of like a randomized control trial in situations where perhaps you cannot randomize people to different exposure groups. Um as is my case in my post-doal research where I look at early life adversity. I cannot randomize people to to receive early adversity or not nor do I want to. Um so what we can do is in observational based studies we can make um our exposure groups equal on all confounders for for which we have measurement of mimicking a randomized control trial where there's randomization and then what this means is that your exposure groups are basically identical on all variables other than the exposure status strengthening causal inference in your conclusions. Propensity scores can also be used to address time dependent confounding which is also known as temporally induced bias. Though in that situation you have to move to more advanced methods such as using marginal structural models. We're just entering our last few slides. Some other opportunities that come to mind at least from my perspective is looking at um you know greater intersectionality with other equity dimensions of course incorporating lived and living experience from survey design um measurement um development through to analyses and interpretation of findings and I think a really promising area would be to see the CLSA linked with Canadian health administrative data. Um it would certainly be wonderful to have the opportunity to replicate um some findings that we're seeing south of the border, say for example with premature mortality. So some key takeaways from this talk. LGBT plus Canadians experience a higher proportion of many adverse health outcomes in older adulthood. Most pronounced is the persistently elevated risk of mental health concerns. Equitable healthpromoting strategies are needed to increase healthy aging. and gerontological care should be trauma-informed, inclusive, and affirming of sexual minority identities. I would like to acknowledge my collaborators on this focal paper, Dr. Stinch Arincom and Alli Grady, who is a PhD candidate in the HARP lab. as well. I've been very fortunate to have some funding along my journey which has made um my opportunities to to partner with other labs and and obtain greater um research exposure possible. And that's it. If you have any questions that we don't get to tackle in the question and answer period, happy for people to reach out to me via email or if questions come up at a later date. And uh I believe Sophie, you can correct me if I'm wrong. uh these slides will be circulated or u made available in some capacity. So I do have a list of references on the following slide which will be available. Thank you so much. >> Great. Thank you. Um that was an excellent presentation. So we are now opening it up for questions. I see we have a few already in the Q&A box. So, as a reminder, muting will remain on, but um you're welcome to enter your questions into the Q&A box um at the bottom of the Zoom window. So, first off, uh did your stratification variable of sex include nine non-binary genders? So, our stratification variable um was a biological sex assigned at birth. Um this was including male. this the only the answer options were males and females. Um and again we made this decision to um uh align with ex existing work in the area and then also for um to align with that original paper from Dr. Arncom. But thank you that's an excellent question. >> Thanks Nicole. Um next you mentioned that those who responded did not identify as any of the above. a followup were classified as LGBT plus in your sample. Can you explain the rationale? >> Um, could you repeat that one more time? Sorry, Sophie. >> Yeah. Um, you mentioned that those who responded did not identify as any of the above at followup >> were classified as LGBT plus in your sample. >> Can you explain the rationale? >> Certainly. So thank you again for this question. Um this was this decision was made after you know a great deal of reflection in the heart lab and we wanted to be as inclusive as possible and recognize that some people may not wish to um to reveal their sexual orientation and so we made that decision to include them in this umbrella term. Um certainly strengths and weaknesses to doing that and we hope that future work um will look at um um different identities in greater detail than we took the opportunity to. Um because we were really focused on a range of different health outcomes with um varying different prevalence sizes. And so we were also hoping to maximize statistical power as possible as much as possible to able to be able to um look at all of these different health outcomes. So certainly um not everything is done necessarily perfectly and we recognize as well that certain questions have take require different um perspectives and so certainly future work could uh look at some effect decomposition in greater detail than what we did. Thanks. Um the next question I might just clarify first. So um guys show uh says I'm a long-term HIV survivor since 1986. Is this being tracked? Um, my question back is, um, I'm not sure if the question is referring to the CLSA as a whole or to your studies in particular. Um, but perhaps you can comment on your study in particular. >> Certainly. So, um, HIV status was not a specific um, um, outcome for purposes of this work. And um as I preface the beginning of my talk, um I've had the immense opportunity to to work with the CLSA data for several publications. Though I'm still not an expert in it, so I'm not specifically sure whether information on that um would have been captured by the CLSA at any time point. That's certainly um something to consider. Um but I don't actually know off the top of my head. I'm I'm so sorry. >> No problem. I uh was doing a quick check myself and I don't believe we've have a question about HIV, but someone knows better, they can correct me. I'd have to look into it further. Um, but thank you for that question and and the response as well. And if I might add, Sophie, I if anything, my experience with this has shown that, you know, things change over time in terms of what questions are included versus not included. So even if it hasn't been included, there's still I would say probably a possibility that it could be included in future future cycles. >> For sure. Um um so the next one is I'm wondering how transgender people were or were not factored into the results knowing that trans people face higher levels of medical discrimination and that many of them also identify as LGBT plus. How was this addressed in the study? um namely for division between female and male. >> Another excellent question. Again, not every study is perfect and um we come at each paper from a specific lens with different objectives and really if I could stress further like our objective was to update that original prevalence paper. So we were doing a lot that was done initially and then extending some of those findings. So I would take this opportunity to note that we did not exclude transgender people. Um, so we did not exclude transgender people. I know that Dr. Arn Stingcom has looked at transgender identity um in the CLSA and um uh I believe um some estimates are that less than 100 people um at least um in the CLSA reported transgender identity. So um I think so we did not exclude people with transgender identity from these analyses. Even if we did, I think that overall, you know, I don't think the findings would have been substantively changed. But again, I think that's a wonderful future line of inquiry, something that was not a specific objective of ours to to look at for purposes of this work. >> Thanks. Um, next we have, um, so you mentioned the relationship between early childhood experiences and cognition. Can you tell us more about the findings and how you looked at childhood experience with the CLSA data set? >> Um, yes. And happy to continue this chat offline if that makes sense. But, um, we're looking at, um, several indicators of early life adversity as measured via the traditional conceptualization of ACEs or adverse childhood experiences. There are a host of different items that were asked in the CLSA at first follow-up on ACES. Some were measured before age 16. well asked of respondents to report on retrospectively um but asked participants recall whether this particular event had occurred prior to age 16 versus age 18. And so we're um doing lots of sensitivity analyses to look at different conceptualizations and teasing apart different types of adversity from others and then also looking at say a higher level combined category of whether or not adversity was experienced. And that's what you saw on um one of those um last the applied epidemiological slides that I included. Um pardon me. We just conceptualize adversity based on number of types of adversities, but our future work endeavors to tease apart how different types of adversity may be differentially related to poor health over time, including poor memory. And I think there's some other published works of other individuals using ACES in the CLSA. So, I'd encourage you to to do a Google search if you're interested in how they've also conceptualized aces. >> Nice. Um, okay. So, that's that one. And then, um, the next, so, um, there was a comment a comment in the chat and then, um, on the Q&A. So, um, this is from Carlos Rancero, who, um, says, "First off, mental health is a more predictive variable than whether or not they're transgender." I think that was in response to your, um, the earlier question. And then um he says speaking as a gay man and researcher um uh his comment is that this framing is fundamentally flawed as treating sexual minorities as a monolith and ignoring critical coariantss like socioeconomic status. Um is not just poor methodology but relies on a deficit model model that perpetually excuse me perpetually cast us as victims. Um, much like the rigid one-dimensional paradigms that Kenneth Dion confronted in his research, this approach reduces our complex, diverse lives to a single variable. Many of us resent the assumption that our sexual orientation is the sole defining factor of our existence and the academic community needs to move past these reductive stereotypes and demand rigorous multi-dimensional analysis. So, do you have any comment or response? First of all, thank you for your honesty and for sharing. Um certainly um you know I know that there are leaders in in this area who are um following a lot of best um say intersectional practice in terms of analyses and you know I agree um in terms of the level of nuance and and detail required um and this is some you know we're striving to address some of these um details in in future analyses looking at you know again the intersectionality with other equity den dimensions and then you know calculating epidemiological measures on say an additive scale which is in line with a lot of best intersectional practices and moving away from um the standard multiplicative model and you know um addressing equity by like conf um what's the word that I'm looking for um adjusting for models as adjusting for variables as confounders and and models so moving beyond that um that linear view um but anyway I just appreciate the honesty and transparency and yes, I'm there are I know that there are leaders in the field who are who are able to address some of the details that this paper um does not and was not designed to do. >> Thanks Nicole. Um so we're running out of time. I think um we can get through maybe one or two more. Um so the next one, sleep disturbance is vague. bifphasic sleep pattern maybe become more prevalent with increasing age and might be normal is like is could that be normal? >> Absolutely. Again, this is really this paper is more of a like a public health um raise awareness piece. We're looking at I don't even know off the top of my head how many different outcomes there. This is very higher level. Um certainly with um you know there could be much more specific models looking at sleep and then looking at say trajectories of sleep over time which might be a better practice to model it to not just look at sleep simply dichotomized as you know um poor or good. There are so many different factors and contextual and individual level factors that must be considered. Um, again, this paper was not designed to look at those details, but more to raise awareness of these possibilities. Um, because raising awareness contributes to um, conversations like we're having here and the opportunity for researchers like myself to be able to present these findings and contribute to advancing um, this research. >> For sure. Thank you. Um, so, um, one last question and, um, for those of you whose questions we didn't get to, Nicole has offered to, um, answer those offline. So, please, uh, feel free to reach out to her. Her email will be in the chat, um, or has already been in there. Um, so we'll just pick one more. Um so um they write thank you for your very interesting presentation. I think this is very important to ask who is missing. What can we do to address limitations of findings for traditionally underrepresented identities and experiences? What recommendations do you have for reporting these limitations when we are sharing findings from the CLSA in our publications and presentations? >> Um excellent question. I think um one approach is as we've strived to do is um by making this transparent by noting this as a limitation to talk you could use commonly used terms such as survivor bias assessing the potential implications on your finding that this underestimates the true strength of relationships. We've talked about that. Um but more broadly speaking I would say stay tuned. Uh Dr. Dr. Arn Sting has a CLSA uh sorry a CHR grant uh to look at who may be missing from the CLSA from an equity perspective and we're really hoping to do a very extensive review of um who might not have been captured to begin with and the potential implications on uh future analyses because we know that um who gets count who gets counted matters in terms of healthcare decision-m and health resource planning. So we're hoping to shed light on this going forward. So stay tuned. When possible, note this in your Olympic uh implications uh your study strengths and limitations and and we've certainly done that a few times. So if you check out some of our work, you might uh get some ideas to to borrow yourself. >> Great. Thanks so much, Nicole. Um we're going to end it there. Um thank you so much for participating in our webinar series. I um noted like you had mentioned as next steps um working with CSA data linked to health admin data. So I just wanted to pop in the chat myself actually. Um and now I messed up my whole screen. Just a second. Maybe somebody can do it. >> I'm on the edge of my seat. I don't know if there's important exciting news being revealed or >> Okay. Well, so we do have linked data available in um BC, Ontario, Nova Scotia, and New Brunswick. So researchers who are interested in accessing CLSA data linked um to health admin data in those jurisdictions um can apply for that and I put the link in the chat so you can find out more about that. Um, so anyways, I just want that's a a plug for another one of the hats that I wear is the linked data initiative. >> Thank you so much. I had no idea. So this is wonderful news myself. >> No problem. Okay, so um yeah, we'll wrap up now. Um I'd like to remind everyone that the next deadline for data access applications is July 8th, 2026. uh please visit the data access pro uh section on our website to review available data as well as additional details about the application process. Um we'd also ask um or yeah any users to keep us updated on changes that affect your data access agreements. So if you've moved institutions or your team has changed or you have a trainee that's graduated and moved on, please reach out to us at accesscla lcv.ca with any questions or updates. Um, and special requests to complete the anonymous survey when you exit this Zoom session. Um, so thank you again all of you for attending and participating in our webinar. We our CLSA webinars will return in September 2026. We hope that you have a wonderful summer and we look forward to seeing you again in the fall. And if you or a colleague is interested in presenting a CLSA webinar in the next year, please reach out to our webinar team. The recording of this webinar and the slides will be available in the coming days on the CLSA website. So, thank you all. Have a great rest of your day. >> Thanks, Nicole.