Advancing Research on Chronic Conditions in Women (2024)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

3 Methodological Considerations for Studies of Chronic Conditions in Women Scientific evidence on the frequency and distribution, risk and protective factors, biomarkers, prognostic factors, and interventions for chronic conditions in women is based on a limited number of study designs and methodologies. Interpreting the evidence presented in this report and, more broadly, in the literature, assumes a familiarity with methods and principles of fields such as statistics, demography, and epidemiology. This chapter highlights some of the important methodological considerations when interpreting, evaluating, and applying the evidence and gaps in research on chronic conditions in women. What is known is informed by the data generated, collected, and analyzed. Furthermore, positionality influences how research questions are posed, and how individuals see and experience the world, and many have multiple, intersecting identities (Wilson et al., 2022). The data apply to specific subsets of conditions that are understood enough to be diagnosed and prioritized enough to be studied, with varying degrees of misclassification and uncertainty. These data are analyzed using statistical methods and contribute to a body of scientific and clinical literature and evidence that may inform care and policy. However, naive analyses that fail to reflect a clear causal hypothesis or account for known biases may lead to misinterpretation. Automated analyses, such as with artificial intelligence applications, may inadvertently propagate bias. This chapter discusses three types of study designs. The first part includes several considerations relevant to population and non-population based studies and descriptive epidemiology. The second part addresses methodological issues in studies of risk factors, associations, and causal inference from observational studies and the third part addresses issues related to experimental studies, primarily clinical trials of medical interventions. These considerations are relevant whether studying biologic samples collected from individuals, therapeutic effectiveness and side effects, or behavioral modifications and social factors. This chapter focuses primarily on more conventional quantitative and statistical approaches, although many points on bias relate to other types of study. Some methodologic considerations, such as the definition of the disease, study participants’ level of involvement, and representativeness of the studied population, apply to all three types of studies. PREPUBLICATION COPY: UNCORRECTED PROOFS 

2 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN POPULATION AND NON-POPULATION BASED STUDIES Population-Based Studies Population-based studies include everyone in a defined, targeted population, called “complete enumeration,” or the investigators apply statistical sampling strategies to generate a sample that represents the target population. Statistical sampling techniques requires the population to be defined and enumerated. The most common definition is geographic—every woman in Nebraska, for example—but that also requires setting a specific period. If the size of the target population is large, selecting a probability sample may be necessary to reduce the cost and time of the study. Using random selection when creating a probability sample allows for extrapolating study results back to the target population using statistical inference methods. The sampling can be simple—random sampling from a list of residents—or more complex, in the case of stratified random sampling, which establishes subgroups, or strata, based on the subgroup members’ shared attributes or characteristics such as income or educational attainment. Population-Based Data Sources or Studies National Health and Nutrition Examination Survey Since the 1960s, National Health and Nutrition Examination Survey (NHANES) has provided prevalence data on various chronic diseases, conditions, and risk factors for the U.S. population based on a survey of a representative sample (CDC, 2020a; HHS, 2013). NHANES oversamples some populations to improve estimation in frequently underrepresented groups, including several racial and ethnic groups, and provides survey weights to account for this oversampling when analyzing the data. In addition to survey self-reports, NHANES uses laboratory test results and medication reviews to confirm some diagnoses (CDC, 2020a). The 2017–2020 NHANES report includes data on cardiometabolic conditions such as heart disease, diabetes, and stroke, but included only a few female-specific and gynecologic conditions, such as dysmenorrhea/abnormal menses and pelvic inflammatory disease. Among autoimmune diseases, which afflict women more than men, NHANES only reports on rheumatoid arthritis (RA). For musculoskeletal and nervous system conditions, it only includes data on osteoarthritis and osteoporosis. For brain and neurocognitive conditions, it only includes data on depressive disorders (CDC, 2020b). National Health Interview Survey Since the late 1950s, National Health Interview Survey (NHIS) contacts approximately 30,000 persons each year to obtain information on core health measures such as chronic conditions, functioning and disability, health insurance, health care, health-related behaviors, and demographics. This information is used to monitor U.S. health trends (NCHS, 2023). The survey included three cardiometabolic conditions (heart disease, diabetes, and stroke), one female- specific condition (menopausal symptoms), two autoimmune conditions (systemic lupus erythematosus (SLE) and RA, four musculoskeletal and nervous system conditions (osteoarthritis, chronic pain, myologic encephalomyelitis/chronic fatigue syndrome (ME/CFS) and fibromyalgia), three brain and neurocognitive conditions (depressive disorders, migraine/headache, and Alzheimer’s disease/dementia), and one infectious disease, human immunodeficiency virus (HIV). PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 3 Behavioral Risk Factor Surveillance System Since 1984, Behavioral Risk Factor Surveillance System (BRFSS) has collected data on health behaviors, including preventive behaviors, and chronic conditions, from over 400,000 U.S. adults each year (CDC, 2014) via telephone surveys in each state, the District of Columbia, and all U.S. Of the chronic conditions considered in this report, the 2022 BRFSS reported on several (CDC, 2023a). The survey included three cardiometabolic conditions (heart disease, diabetes, stroke), no female-specific and gynecologic conditions, two autoimmune conditions (SLE and RA), three musculoskeletal and nervous system conditions (osteoporosis, osteoarthritis, fibromyalgia), one brain and neurocognitive condition (depressive disorders), and one infectious disease (HIV). Health Care Claims Data Health care claims provide another source of data on chronic conditions in women and may include demographic data, such as race and ethnicity, age, geographic information, and outcomes based on diagnoses or procedure codes. The Centers for Medicare and Medicaid Chronic Conditions Data Warehouse includes data from Medicare claims representing a subset of the population eligible for Medicare (Chronic Conditions Data Warehouse, 2023). Limitations or caveats of claims data, in general, are that they may be influenced by a provider’s coding knowledge and behavior and fail to capture the experience of individuals who rarely or never seek care. A similar database including persons younger than 65 is not available for the entire United States because of the lack of a national health system. National or Regional Medical Databases Publicly controlled medical databases or systems are common in countries such as Canada, Denmark, and Taiwan with a centralized medical system informed by health care claims and population-based registries (e.g., Swedish Cancer Register) (Socialstyrelsen, 2024). Although data from countries universal-access with national registries of health care interactions may address some of the limitations of non-population-based samples, the design and analysis of these data must still carefully consider sources of bias. Further complicating interpretation and generalizability, societies with universal access to health care may also provide other critical social support that can influence risk factors and social (and structural) determinants of health (Marmot, 2013). Publicly controlled medical databases are not available at the national level in the United States (except for Medicare). However, some health maintenance organizations and other health care institutions have developed a medical-records-linkage system for patients affiliated with their plan. Unfortunately, often the patients a plan covers do not represent all of a geographically defined population. The Kaiser Permanente plan in California and Oregon, and the Rochester Epidemiology Project medical records-linkage system in 27 counties of Minnesota and Wisconsin are two examples of systems targeting a defined region but covering only a sunset of the total population (Kaiser Permanente, 2024; Rocca et al., 2018). The Rochester Epidemiology Project records-linkage system covering Olmsted County, Minnesota can be considered population-based because the entire population is included, and multiple care institutions share their data. These regional U.S. databases were used to measure the prevalence of some of the select chronic conditions reported in Chapter 4 (e.g., fibromyalgia, and multiple sclerosis). PREPUBLICATION COPY: UNCORRECTED PROOFS 

4 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN Non-Population-Based Studies Some studies are based on convenience samples recruited via spontaneous referral to a care facility, advertisement, internet websites, or other methods that cannot be replicated. Despite the methodological limitations of such selected samples, study sponsors have invested heavily in establishing large, non-population-based longitudinal because it is feasible; in some studies, the sampling may accommodate including racial and ethnic groups. Selecting the individuals in a convenience sample may introduce unwanted distortions in the distribution of demographic variables, presence of risk and protective factors, or presence of comorbid conditions. For example, a 2012 study compared magnetic resonance imaging (MRI) measurements in the brain of subjects in a convenience sample with those in a population-based sample (Whitwell, 2012). The rates of decline in hippocampal volume in cognitively normal subjects and subjects with amnestic mild cognitive impairment differed significantly in the two samples. In some cases, as in this study, investigators can use statistical techniques to account for these differences, but this requires knowing about these discrepancies (Whitwell, 2012). Non-Population-Based Databases or Studies These types of data can be used in a range of study designs including but not limited to cohort and case-control studies (see section on Observational Studies). The specific methodologic concerns related to those designs and to how data are sampled are discussed elsewhere in this chapter. It is critical for researchers to understand the source of their data, and the potential consequences to the internal validity and generalizability of studies. Some cohorts from these types of data provide detailed, longitudinal details of the participants that may be less plausible to accrue in another setting. Investigators should consider how the respective strengths of the data (e.g., sample size, availability of biospecimens, detailed follow-up, or even convenience) weigh against the potential limitations to the generalizability. The Nurses’ Health Studies The Nurses’ Health Studies (NHSs) I is a longitudinal cohort study that began in 1976 and focused on contraception methods, heart disease, cancer, and smoking (NHS, 2016). In 1976, nurses aged 30-55 at enrollment from 11 states participated, with some 121,700 women completing the survey. Since the original cohort, the research team established two subsequent cohorts. NHS II, begun in 1989, focused on the effects of long-term use of oral contraceptives, physical activity, and diet in nurses aged 25-42 at enrollment in 14 states (about 116,430 women). The ongoing NHS III cohort, established in 2010, initially included female nurses aged 19-46 at enrollment and males of the same age starting in 2015. Diet, lifestyle behaviors, environment, and hormone-related effects on women’s health are areas of focus. The Women’s Health Initiative The Women’s Health Initiative (WHI), funded by the National Heart, Lung, Blood Institute, operated from the early 1990s to 2005 (WHI, 2021). Focus areas included cardiovascular disease (CVD), cancers, and fractures related to osteoporosis. Since 2005, WHI has continued through extensions to collect updated health data on the participants and enrolled 93,500 women in 2010. WHI recruited women from regional center sites located in Buffalo, Winston-Salem, Columbus, Stanford, and Seattle (WHI, 2021). The WHI Observational Study, PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 5 which enrolled more than 93,000 women, was established to explore the predictors and the natural history of several important causes of morbidity and mortality in postmenopausal women (Langer et al., 2003). The sample was also intended to serve as an observational control for the WHI clinical trials. The Study of Women’s Health Across the Nation The Study Women’s Across the Nation (SWAN) was established as a “multicenter, multiethnic, longitudinal study to characterize the physiologic and psychological changes during the menopausal transition and study their effect on health and risk factors for age-related diseases.” In 1996 and 1997, the study enrolled more than 3,300 women aged 42–52 from seven research centers located in Ann Arbor, Boston, Chicago, Alameda and Contra Costa County, Los Angeles, Jersey City, and Pittsburgh (Santoro and Sutton-Tyrrell, 2011; SWAN, 2024). Disease Registries and Cohorts According to the Department of Health and Human Services, a registry is “an organized system for the collection, storage, retrieval, analysis, and dissemination of information on individual persons who have either a particular disease, a condition (e.g., a risk factor) that predisposes to the occurrence of a health-related event, or prior exposure to substances (or circumstances) known or suspected to cause adverse health effects” (AHRQ, 2020; HHS, 2022). In the United States, the Surveillance, Epidemiology, and End Results (SEER) Program is a surveillance-based registry collecting data on every cancer case from 22 geographic areas covering nearly 50 percent of the U.S. population (NIH, 2024b). Registry data can be used to estimate disease prevalence, incidence, and trends over time; identify high-risk groups; assess service delivery and use, characterize the types of patient services providers deliver; and conduct research (AHRQ, 2020). In addition to programs such as SEER, several disease-specific registries and cohorts exist. Some registries are population-based, but some are not and include volunteers. NIH has compiled a list of some national-level disease registries (NIH, 2024a). Those relevant to women’s health mentioned in the current NIH report include the following: • Alzheimer’s Prevention Registry (Banner Alzheimer's Institute, 2024) • Autoimmune Registry (Autoimmune Registry, 2023), and • Lupus Family Registry and Repository (ORDRCC, 2023) (Fryer et al., 2024) (Arbuckle et al., 2003) STUDIES OF FREQUENCY AND DISTRIBUTION (DISEASE) Impact, or disease burden, is an important concept for comparing trends of various diseases and conditions in the population and tracking and prioritizing implementing and allocating interventions, resources, and policies for chronic disease (Porta, 2014). In addition, conditions may have differential effects on different groups of women, such as racial and ethnic groups (see Chapter 4). PREPUBLICATION COPY: UNCORRECTED PROOFS 

6 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN Defining Chronic Conditions and Study Populations Diagnostic Criteria Before measuring impact or studying associations and interventions, it is essential to define a disease’s diagnostic criteria. Clinicians use diagnostic criteria to diagnose individuals accurately and objectively. In contrast, classification criteria assist in identifying individuals with a condition by excluding those without it (Aggarwal et al., 2015). For example, both osteoporosis and Type 2 diabetes have standard diagnostic criteria to identify individuals with the disease, whereas conditions such as RA and migraine use criteria based on exclusion (Aletaha et al., 2010; Angus-Leppan, 2013). In addition, conditions such as pain rely on subjective reporting of symptoms. The heterogeneous aspects of experiencing pain contribute to challenges in developing measures of impact, such as prevalence or in designing observational or experimental studies. Diagnostic criteria may change over time as research uncovers new information about a condition. The criteria may become more inclusive, increasing the number of people diagnosed, or more restrictive, causing the impact of disease to decrease. In these situations, the true biological impact of the disease has not changed, just the way it is measured. For example, in 1987, the Diagnostic and Statistical Manual of Mental Disorders, Third Revised Edition (DSM- III-R) defined dementia as a syndrome requiring impairment in both short-term and long-term memory, impairment in at least one additional cognitive domain, and interference with usual social activities or relationships (APA, 1987). In 2013, the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) abandoned the term “dementia” and replaced it with “major neurocognitive disorder.” The new criteria required evidence of decline from a previous level of performance in at least one cognitive domain, with the decline based on the concern of the individual, of an informant, or by a clinician; impairment in cognitive performance documented by neuropsychological testing or by a qualified clinical assessment; and a cognitive deficit that interferes with independence in everyday activities (APA, 2013). The criteria have become more inclusive—a decline in one cognitive domain is now sufficient for the diagnosis— with the result that the number of people diagnosed with major cognitive disorder, formerly dementia, increased. Study Design and Data Sources Research on the impact of chronic conditions in women uses several study designs and data sources. Experimental studies always involve active patient participation. By contrast, observational studies may be classified as (1) studies that leverage pre-existing data sources such as diagnoses in electronic health records (EHR), and (2) active, which directly collect new data from research participants. Both pre-existing data and de novo collected data have unique strengths and weaknesses. Some studies combine elements from both pre-existing data sources and active methodologies. In studies of pre-existing data, researchers do not ask the participants any questions or collect any biological measure or sample. Often, the research team does not need to inform the participants of a study using their data if they have provided a general permission for the use of their data for research, a government institution manages the data, or the data are in the public domain. In active studies, the research team uses letter, phone, or in-person contact to invite people to participate in a study and sign an informed consent if they agree. Participants may be PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 7 asked to be interviewed, undergo a physical examination or imaging study, provide blood or other tissue materials such as saliva or a skin biopsy, use a wearable device to collect information over time, or partake in an intervention study, such as taking a drug candidate or a placebo. Representativeness and Generalizability Representativeness of a sample corresponds to how well it represents the distribution of the population that was sampled, while generalizability considers how well the findings from the sample may generalize to other similar populations. Studies can be population-based and non- population-based. In the former, the study sample may include everyone in or be representative of a defined population, the latter may include persons identified via a non-reproducible process. The inferences that can be made from the study sample to the general population—their generalizability—depend on the level of representativeness which is different from inclusivity or diversity. For example, a study of the entire population of Nebraska may be population-based but limited by the low percentage of persons from certain racial or ethnic minorities living there. The results of the study apply to the state of Nebraska and possibly other U.S. populations with similar racial and ethnic composition but may not generalize to other U.S. populations. Confusing representativeness with inclusivity should be avoided when judging the quality of a study or the generalizability of results across populations (St Saver et al., 2012). Measures of Disease Occurrence and Impact The most commonly used measures of disease occurrence and impact are prevalence, incidence or risk, and mortality. As mentioned previously, prior to quantifying impact, an accurate definition of the disease/condition is needed, which could be based on diagnostic criteria, rule out symptomology, or self-report. However, measures of occurrence may not be sufficient to reflect the impact of a disease in terms of its effect on the life of a person, which is why studies often include other measures of impact, such as disability-adjusted life years (DALYs). Studies may also measure the economic or societal impact of chronic conditions. Prevalence represents the number of persons a disease affects at one point in time, and incidence is the risk of developing disease during a given time window. Prevalence is commonly expressed as a percentage of the population, although for rare conditions it may be reported as number of cases per 10,000 or 100,000 persons. Incidence rates reflect the average number of new cases per person-years 1 or other unit of person-time. Incidence can be measured as the cumulative incidence over time and expressed as the percentage of new cases in a population, or as an average density of new cases over time using incidence rates (Porta, 2014). Distinguishing newly diagnosed or incident disease from existing or prevalent disease is critical, and prevalence and incidence are not interchangeable. Women with a chronic condition were at one time incident cases and survived long enough to be identified as living with the condition. Factors associated with living with a condition may include having a milder version of that condition, access to better care, or other factors related to survival and can be different from factors associated with developing the condition (Nussinovitch and Shoenfeld, 2012; Temkin et al., 2023). 1 Person-time estimates the time that individuals are at risk of experiencing the studied outcome. PREPUBLICATION COPY: UNCORRECTED PROOFS 

8 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN Mortality refers to the risk of dying with a disease during a given time window. It is similar to incidence and can be measured as cumulative risk of death over time or number of deaths per person-years (mortality rate) (Porta, 2014). Mortality does not give a complete picture of disease impact displayed by individuals in different populations, which is why DALYs are used. The World Health Organization (WHO) defines one DALY as representing “the loss of the equivalent of 1 year of full health. DALYs for a disease or health condition are the sum of the years of life lost to due to premature mortality and the years lived with a disability due to prevalent cases of the disease or health condition in a population” (WHO, 2023). Disability-Adjusted Life Years DALYs are calculated using a disability weight—which reflects the severity of health state from disease or injury, with 0 equivalent to full health and 1 equivalent to death (Liu et al., 2023)—multiplied by chronological age to reflect the impact of the disability. U.S. DALYs can be obtained from the Institute for Health Metrics and Evaluation (IHME) Global Burden of Disease (IHME, 2023). The 2023 National Institutes of Health (NIH) Framework is used to quantify the impact of chronic conditions (Temkin et al., 2023) and have also been used to identify gender disparity in NIH’s disease-specific funding (Mirin, 2021). DALYs do not include economic burden and are not specific measures of disability or function (Grosse et al., 2009). Quality Adjusted Life Years Quality Adjusted Life Years (QALY) are an index measure of disease impact that combines both morbidity and mortality (Howren, 2013). It is an outcome measurement of health over time related to a disease or condition under study and the measure of quality of life is multiplied by the time spent in that health state to obtain QALYs. There are several different instruments available that measure QALYs and should consider the population being studied. Economic Burden The economic cost of diseases includes two components, direct costs such as hospitalization, testing, office visits, pharmaceuticals, medical devices, skilled nursing facilities, and rehabilitation, and indirect costs, including lost labor, early mortality, effects on employees, nonmedical expenses, and caregiver expenses. Administrative data from sources such as Medicare or insurance databases, can provide the direct costs. For example, the direct cost of medical care in the United States was estimated at $4.3 trillion annually in 2021 (Hartman et al., 2024). For some diseases, the indirect cost may be as much or more than the direct cost (Weintraub, 2023). However, it is difficult to measure the indirect costs of diseases, and better metrics are needed to capture these, including, but not limited to, caregiving which may disproportionately fall on women. Challenges with Capturing Disease Occurrence and Impact As mentioned in the previous section, many of the chronic disease surveillance systems such as NHANES, NHIS, and BRFSS neither capture nor track female-specific and gynecological conditions. Prevalence estimates for these arise mainly from population-based studies and samples of patients from EHRs and health care delivery systems from certain geographic locations. Chapter 4 discusses the impact of chronic conditions for select conditions. PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 9 Prevalence Analysis of different data sources can yield different estimates of prevalence for chronic conditions. For example, the Center for Medicare and Medicaid Services claims data from 2018 yields an estimate of 10.8 percent for osteoporosis (CMS, 2023), versus 19.6 percent from NHANES (Sarafrazi et al., 2021) based on confirmation by bone mineral density measurement. For Type 2 diabetes, NHANES 2017–2020 reports an age-adjusted prevalence rate of 12.4 per 100,000 women with diabetes based on self-reported diagnosis and undiagnosed cases identified from laboratory results (includes diagnosed (self-report) and undiagnosed (based on lab results)) (CDC, 2023b). Medicare fee-for service claims data found that 25.5 percent of women, most aged 65 and older, had diabetes (CMS, 2023). Two longitudinal studies provided additional information on diabetes prevalence. In the SWAN 2 cohort, about 5 percent of participants had diabetes at baseline and 12 percent had diabetes at follow-up in 2021 (Reeves et al., 2021), but NHS reported that 7,401 women out of the 121,700 followed 1976–1996 had developed diabetes (Al-Delaimy et al., 2001). The accuracy of measures of impact vary based on how a condition is diagnosed. For example, HIV diagnosis is based on results from antigen/antibody tests, with mandated reporting of positive cases yielding robust calculation of prevalence. However, for other conditions with challenging diagnoses, such as vulvodynia with no known etiology, the data to calculate prevalence are incomplete. Prevalence by Race and Ethnicity Despite some data to support that racial and ethnic differences exist for some chronic conditions, reporting prevalence for these groups is inconsistent. In addition, groupings can vary by study, in part resulting from using “other” as a group. For example, a study of Medicare data, including both Medicare Advantage encounter data and claims from traditional Medicare, reported the age-adjusted prevalence of dementia, including Alzheimer’s Disease (AD) was higher in Black (10.9 percent), Hispanic (10.0 percent), and American Indian and Alaska Native (9.7 percent) compared to White individuals (7.7 percent). However, diagnosed dementia was slightly lower among Asian individuals (7.2 percent). It was not reported separately for men and women by race and ethnicity (Haye et al., 2023). In the Chicago Health and Aging Study, a representative survey of four Chicago neighborhoods, the age-adjusted prevalence of AD and related dementias was 11.3 percent overall, 10.0 percent in non-Hispanic White individuals, 14.0 percent among Hispanic and Latino individuals, and 18.6 percent among non-Hispanic Black individuals. Prevalence was not reported separately for men and women by race and ethnicity (Rajan et al., 2021). The Centers for Disease Control and Prevention (CDC) has supported multiple registries of individuals with SLE, and a meta-analysis of 2018 data from four sites--Michigan, Georgia, New York, and California—estimated an overall age-standardized prevalence of nearly 73 per 100,000. However, the analysis found significant differences by sex and race, with an approximate prevalence of 231 per 100,000 among Black females, 85 per 100,000 for White females, and 27 per 100,000 for Black males (Izmirly et al., 2021). Many population-based studies and national surveillance systems are still not representative of the entire U.S. population and do not include women from various racial and ethnic groups. Furthermore, certain racial and ethnic groups are studied less, with less data 2 The SWAN cohort included registered nurses aged 30–55. PREPUBLICATION COPY: UNCORRECTED PROOFS 

10 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN available that disaggregates certain subpopulations. Black, Hispanic/Latina, American Indian and Alaska Native, Native Hawaiian and Pacific Islander, and Asian populations are heterogeneous and comprise many groups. Therefore, prevalence rates for these populations may vary substantially, but a lack of data disaggregation can mask these variations. Time Trends in Disease Impact Over time, some diseases become more or less frequent or disappear completely. For example, data from several Western countries, including the United States, shows the incidence of dementia has declined over time, and some studies have reported that the prevalence has also declined (Hudomiet et al., 2022; Rocca, 2017). However, the impact of dementia in terms of the absolute number of people affected is growing as the U.S. population ages, with a dramatic increase projected for the coming decades. The effect of demographic change may be more powerful than the reduced risk indicated by the decline in incidence and prevalence (Rajan et al., 2021; Rocca, 2017). One group has projected that dementia will affect 6.07 million in 2020 but 13.85 million in 2060 as the “baby boomer” 3 generation ages and the increase will be greater in women than men (Rajan et al., 2021). A review of over 40 years of data from Olmsted County, Minnesota found that the prevalence of SLE has increased fourfold, from 30.6 per 100,000 in 1985 in males and females combined to approximately 97.4 per 100,000 in 2015 (Duarte-García et al., 2022). Given the variability by age, sex, race, and ethnicity, overall prevalence estimates may not accurately reflect the true population heterogeneity. Studies have reported increasing trends in the incidence of SLE, although measuring incidence is difficult because of diagnostic challenges. The Spectrum of Disease and Impact Some diseases manifest as a spectrum that includes preclinical, prodromal, and clinically manifest stages. In the preclinical stage, one or several biomarkers of the disease are present, but no clinical signs or symptoms. The prodromal stage has some signs and symptoms but does not represent the full disease. The clinical stage involves signs and symptoms fulfilling the diagnostic criteria for the disease. Mild cognitive impairment, an intermediate cognitive state between normal cognitive aging and dementia, is an example of a prodromal stage. Although it interferes less with activities of daily living than dementia, its prevalence is higher than that of dementia. For example, in the Chicago Health and Aging Study, a representative survey of four Chicago neighborhoods, the prevalence in persons aged 65 and older was 22.7 percent for mild cognitive impairment compared to 11.3 percent for dementia: 21.1 percent in non-Hispanic White individuals, 25.9 percent among Hispanic and Latina/o individuals, and 32.0 percent among non- Hispanic Black individuals (Rajan et al., 2021). In the Mayo Clinic Study of Aging, a population-based study in Olmsted County, MN, the prevalence of mild cognitive impairment was 19.0 percent in men compared to 14.1 percent in women (Petersen et al., 2010). 3 The baby boomer generation represents persons born between 1946–1964. PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 11 OBSERVATIONAL STUDIES The most common designs for observational studies assessing associations, and reporting measures of association such as relative risks, odds ratios, and risk differences, include cross- sectional, cohort, and case-control designs. The conclusions drawn from each design may differ. A cross-sectional study tends to collect exposure and outcome information at the same time. This is a common design for survey-based studies that frequently yields estimates of prevalence. In cross-sectional studies, researchers identify the chronic disease as they collect data on potential risk factor exposures. In a cohort study, researchers identify groups of women who either have or do not have a characteristic and follow both groups over time to study the incidence of a new condition or outcomes such as death or nursing home placement. Cohort studies are particularly useful for studying relatively common outcomes, or for life course studies, how biological, social, and behavioral factors across the life or generations, influences health and disease. In case-control studies, researchers identify women with or without a given disease or condition and with previous exposures to risk factors. These studies are generally less expensive and can be conducted more rapidly than cohort studies. Researchers compare the odds of an exposure in women with a disease to that with those without it and then calculate odds ratios as the measure of association between exposure and outcome. Cases and controls may be matched by age to reduce confounding by age, but matching on potential confounding factors in case- control studies requires that matching factors be accounted for in the analysis as well (Rothman et al., 2008). The use of different groups of controls, either women free of the study disease, or having a different condition, may also limit the comparability of results across studies. Combining Results across Studies In general, for any risk or protective hypothesis, research teams will conduct several cross-sectional, cohort, and case-control studies. Often, the results are inconsistent because of study design and analytic differences, making it difficult to combine the existing evidence into a simple answer to the causal question. Therefore, interpreting evidence commonly involves judgment. For example, one hypothesis holds that women who undergo bilateral oophorectomy before reaching spontaneous menopause to prevent developing ovarian cancer may have a higher long-term risk of developing Parkinson’s disease. As of 2023, research teams have been testing this association in six case-control and five cohort studies. Seven studies supported the association, but four did not. Some discrepancies may be explained by differences in study design, including the potential confounding by indication. However, this inconsistency was attributed primarily to the age at the time of bilateral oophorectomy, rather than to a difference in study design (Pesce et al., 2022; Rocca et al., 2022). Social Determinants of Health Research As discussed in Chapter 2, Social Determinants of Health (SDOH) are conditions relevant to the environment that impact health. As these can influence individual health outcomes and affect large groups of people—often underserved and vulnerable populations—it is critical to evaluate and consider them when designing and conducting a research study. PREPUBLICATION COPY: UNCORRECTED PROOFS 

12 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN Multiple disciplines, including social epidemiology, health services research, and policy work in this space. Some research specifies race and ethnicity as SDOH, but many scholars underscore that it is not the race that matters but rather the experience with and exposure to systemic racism and discrimination associated with race and ethnicity (Paradies et al., 2015; Williams et al., 2019; Williams et al., 1997). To study the effects of discrimination, researchers may consider a number of measures, including the validated Everyday Discrimination Scale (Williams et al., 1997). Additional measures in SDOH research include the Social Vulnerability Index, or various deprivation indexes that use area-level data—by census tract or zip code, for example—to estimate vulnerability across U.S. regions. The Social Vulnerability Index ranks over a dozen individual variables according to socioeconomic status, housing composition and disability, race and ethnicity, English language proficiency, and housing and transportation (Flanagan et al., 2011). Consideration on Sampling and Selection All research, whether complex longitudinal studies of one million women, phone-based surveys, or small biomarker studies, is based on a particular study population. Understanding chronic conditions in women requires that all the individuals in a sample participate in the study or have data available to be included in the analyses. Longitudinal studies, whether survey-based or with in-person follow-up, there is a natural attrition of the study population. For example, women who remained for 15 years in a cohort follow-up may not represent the full population in the study when the cohort was formed. They have survived and continued participating, so they may be younger and healthier than the original cohort. Similarly, when a study uses random digit dialing to estimate prevalence, women with chronic conditions may be less likely to be employed full-time and thus more likely to be at home and available to answer which may overestimate the prevalence of conditions associated with disability. The potential selection bias, lack of representativeness, and limited generalizability apply to the questions participants answered on surveys, medical records or data available for extraction, or even to the samples available for biomarker or imaging studies. Given the protracted and delayed time to diagnosis for many chronic conditions, the timing of collecting biological samples is also important. Samples from newly diagnosed patients may not reflect features and characteristics of incident disease. In conditions such as endometriosis, where diagnosis may occur an average 7 years after symptom onset (Fryer et al., 2024; Nnoaham et al., 2012), samples may represent the natural history of untreated, undiagnosed disease and not the characteristics of new onset endometriosis. This contrasts with conditions where screening and general awareness may lead to earlier diagnosis. The composition of a study’s population also requires considering the setting of associational studies, such as examining environmental exposures or other risk factors. Study populations that do not reflect the composition of the population at risk may find associations that do not generalize to other groups. It is also important to account for the time in the disease course or diagnostic workup at which individuals are included in a study. When studying risk factors for a chronic disease, the time for assessing them should coincide with a period when it would be plausible for that factor to influence disease risk or onset. For example, patients with SLE have relevant autoantibodies and other measures up to 5 years before they are diagnosed PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 13 (Arbuckle et al., 2003). If risk factors are measured only at or near to the time of diagnosis, the patients may already be affected by the disease at the time of exposure. Considerations Regarding Measurement Many data sources provide little information on behavioral factors of interest, such as smoking, physical activity, sedentary behavior, or diet. In addition, they often fail to capture biomarker or serology data. Some methods exist to analyze data given these unknowns, but quantifying their role in chronic conditions remains a challenge. In addition to incomplete data on potential risk factors, many data sources may not capture specific chronic conditions because of lack of established coding, diagnostic criteria, or specific research questions. When possible, adding write-in options to data collection instruments may facilitate including these potentially overlooked, and hence underreported conditions. Similarly, one cannot report on or measure something that has no metric, and a metric, even if shown to be valid in men, may not be valid nor reliable in women. Measurement is also complicated by other factors. For example, the experience of pain is subjective; one person’s reported score of 5 on a visual analog scale might be a 2 for another person or a 7 for someone else. Although some data sources such as administrative data or electronic health records opportunistically capture the real-world experience of patients moving through the health care system, the type of insurance, type of system, and incentives for coding influence whose data are captured and what part of their care is visible. The codes may capture the diagnoses assigned by the provider, but similar codes may reflect how the encounter was billed. Area-level measures may serve as proxies for individual-level data, such as using air pollution monitor reports to estimate exposure to particulate matter. This can lead to misclassification or measurement error because the assumption that the area-level measure applies to an individual may be incorrect. Other area-level measures, such as the Social Vulnerability Index, assign a value to a community, but the estimation is about the area in which the person is embedded. Validity How accurately data reflect the truth influences validity. Common measures of validity of a diagnostic test or diagnostic code include sensitivity, specificity, and positive and negative predictive values. 4 Researchers can also apply these measures of validity for how data from self- reports compare to a gold standard or reference standard for a variety of variables or factors. Consequences of poor validity are misclassification and measurement errors, which are sources of bias that can affect assessments of exposure to risk factors, outcomes, or other factors, such as confounders. How measurement error biases results is not always predictable, as it depends on whether the error is random or systematic. For example, random noise in a measure is frequently described as non-differential misclassification or non-differential measurement error. In general, investigators assume that random noise would lead to a conservative bias that 4 Sensitivity is frequently described as the true positive rate, or the proportion of persons who have the disease who are correctly identified. Specificity is the true negative rate, or the proportion of persons who do not have the disease who are correctly identified. Positive predictive value is the proportion of persons who are accurately identified as positive by the test. Negative predictive value is the proportion of persons who are accurately identified as not having the disease by the test. PREPUBLICATION COPY: UNCORRECTED PROOFS 

14 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN would attenuate an association, but that may not always be true (Rothman et al., 2008). Systemic or differential measurement error, on the other hand, can bias results in any direction. Considerations Regarding Analyses Confounding, Effect Modification, and Mediation When studying the association between an independent variable x and a dependent variable y, it is important to recognize that they do not exist in isolation but rather are within a complex environmental, social, and historical context. For instance, focusing solely on one event or lifestyle habit as a potential risk factor can lead to erroneous conclusions because numerous other variables may be associated with or modify the variables or association under study; these additional variables may be categorized as confounders, effect modifiers, and mediation variables (Hernán and Robins, 2020; Szklo, 2018). To accurately acknowledge the role of these variables, understanding the biology of the mechanisms being investigated is crucial. Without a causal model, analyses may obscure or distort the association being studied (Lipsky and Greenland, 2022). Age and sex are common confounders. Comparing a group of individuals with dementia to a group without dementia without matching by age may yield numerous spurious associations with medical events preceding the onset of dementia. For example, if the median age of cases is 75 and the median age of controls is 60, the cases will have a higher positive history of hypertension, hyperlipidemia, head trauma, and myocardial infarction. Similarly, without matching by sex, dementia may appear to be associated with sex-specific events and habits. If 70 percent of the cases and 40 percent of the controls are women, dementia will be associated with many of the reproductive risk factors that occur only in women, such as the use of oral contraceptives. Effect modification occurs when an association between some exposure and outcome differs by another factor, such as sex or age. Examining sex-specific effects is an example of evaluating for effect modification (Hernán and Robins, 2020). If a new drug is safe and effective in a group of patients with a disease, stratifying the analysis by sex may uncover important heterogeneity, such as the drug is more effective in women than men. A drug may be beneficial in one group and have no effect in another. In this case, treating the other sex will expose them to side effects without any benefit. Finally, a drug may have beneficial effects in one sex and be harmful in the other. In this case, lack of stratified analyses will cause harm. Age, race and ethnicity, are also common effect modifiers. For example, in one study examining the risk of comorbid chronic conditions such as CVD associated with SLE, authors stratified by race (Falasinnu et al., 2019). Additionally, one overlooked potential modifier is the reproductive hormonal state of the woman, such as menstrual cycle regularity, stage of menopausal transition, or use of hormone therapy or contraceptives, as it may impact exposures and outcomes. One complication is that effect modifiers can also be confounders. Adjusting for an effect modifier, as for a confounding variable, would not uncover this important heterogeneity, and may lead to uninterpretable results. However, approaches such as stratified analyses simultaneously account for confounding and examine effect modification. One or more mediator variables are part of the causal chain connecting two variables, often exposure and outcome, and should not be confused with confounders (Lee et al., 2019). For example, in a study of the association between hypertension and stroke, myocardial infarction is a mediation variable, not a confounder. Hypertension is a risk factor for myocardial infarction, PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 15 and myocardial infarction is a risk factor for stroke. Assuming that is the only chain of causality between hypertension and stroke, adjusting for stroke would completely conceal the true causal association. Using endometriosis as an example, a “mediator” is an exposure or characteristic/biomarker that occurs after or concurrent with endometriosis but before the lifelong health outcome, where endometriosis can be assumed to influence the mediator and not vice-versa (Farland et al., 2020). In other words, mediators shed light on the pathways that are causing the greater risk of lifelong health outcomes among those with endometriosis (Harris et al., 2022). Analytic Approaches In addition to descriptive analyses such as estimating prevalence or risk, many studies using statistical approaches such as regression modeling to estimate associations between multiple exposures or predictors and outcomes. The majority of studies on chronic disease in women focus on one outcome at a time, and even when a study considers multiple outcomes, they tend to be studied individually or collapsed into a composite outcome, such as adverse pregnancy outcomes such as severe maternal morbidity. Although composite outcomes may be appropriate from some research questions, the overlap and co-occurrence of multiple chronic conditions (see Chapter 8) may make multivariate regression or structural equation modeling a better strategy (Beran and Violato, 2010). Despite the common use of various methods to mitigate potential bias from confounding, concern remains for unmeasured or residual confounding resulting from unavailable data or unknown factors. Similarly, measurement is imperfect and can lead to bias resulting from misclassification or measurement error. Quantitative bias analysis methods can incorporate information about the uncertainty, potential degree of bias, and validity of measures to provide some correction or adjustment to the analyses (Lash et al., 2009). When the outcomes studied are rare or the study population is small, data may be sparse. Without a sufficiently large sample and enough outcomes, estimates will be imprecise, and models may not converge. This may require investigators to present unadjusted or minimally adjusted regression results, leaving concerns of bias resulting from residual or uncontrolled confounding. It may be necessary to collapse multiple categories into larger ones, but this may obscure results if there is heterogeneity across those categories. When studying chronic disease in women, researchers use numerous other approaches discussed in more detail throughout this report to collect and analyze data, such as community- based participatory research, structured interviews, focus groups, and implementation science (see Chapter 9). Analytic approaches in disparities and health equity research frequently employ multilevel modeling approaches that consider not only the role of individual-level factors but the role and influences of structural-, social-, and community-level factors. These approaches also account for intersectionality and have been found to have numerous advantages over conventional statistical methods (Evans et al., 2018). Looking through a different lens, systems biology approaches support identifying significant cellular- and molecular-level contributors by analyzing data acquired through high- throughput “-omics technologies”. 5 Systems biology approaches integrate this information into dynamic models that predict and explain mechanisms of pathology. Systems biology has two approaches: bottom up and top down. Both are central to assembling information from all levels 5 Omics technologies include genome, epigenome and proteome information. PREPUBLICATION COPY: UNCORRECTED PROOFS 

16 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN of biological pathways involved in physiological processes. A bottom-up approach encompasses draft reconstruction, manual curation, network reconstruction through mathematical methods, and validation of these models through literature analysis, whereas a top-down approach encompasses metabolic network reconstructions using -omics data generated using modern high- throughput genomic techniques with appropriate statistical and bioinformatics methodologies. Two analytic techniques are primarily used: logic-based mechanistic machine learning and deterministic mathematical models. Although much of this methodology focused on understanding etiology and risks and identifying modifiable factors to improve health, some research emphasizes prediction. This is largely beyond the scope of this chapter, but it is a growing area of research. In machine learning and artificial intelligence approaches, researchers may train models on data generated in a system with embedded discrimination and inequity that can be inadvertently perpetuated in the models. As health systems explore how to embed and leverage large language models and other artificial intelligence in research and health care, it is critical to evaluate the representativeness and bias of the data from which these models are trained. EXPERIMENTAL STUDIES OF INTERVENTIONS Lack of Inclusion Historically, researchers excluded women from long-term clinical trials because of the possibility of becoming pregnant during the study or the complexity of the hormonal milieu and menstrual cycle. In 1977, a Food and Drug Administration (FDA) policy recommended excluding women of childbearing potential from Phase I and early Phase II drug trials, even if they used contraception, were single, or had a husband with a vasectomy (FDA and U.S. Department of Health, 1997). However, that exclusion clinical trials has led to a shortage of data on how drugs affect women. Between 1989 and 1993, NIH policy required including women in clinical research, and in 1993, Congress wrote the NIH inclusion policy into federal law through a section in the NIH Revitalization Act of 1993. 6 One of the key features is its requirement that trials with women and individuals from racially and ethnically minoritized populations should be designed and conducted to enable analyzing whether the variables being studied affect women and minorities differently. 7 Lack of Sex-Specific Analyses In clinical trials that include women, researchers should stratify the analyses by sex given that the efficacy or the safety of a drug or other medical intervention may vary by sex. For example, in July 2023, FDA approved the first drug to slow progression of Alzheimer’s disease (FDA, 2023). However, some concern arose that the drug’s therapeutic effect may be weaker in women compared to men (van Dyck et al., 2022). If that is confirmed, using the drug in women may not reach an acceptable efficacy-safety balance. 6 P.L. 103–43 S.1—National Institutes of Health Revitalization Act of 1993 Subtitle B—Clinical Research Equity Regarding Women and Minorities 7 https://orwh.od.nih.gov/sites/orwh/files/docs/NIH-Revitalization-Act-1993.pdf (accessed April 2, 2024). PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 17 Practically, having adequate statistical power to test for efficacy or safety in women and men separately requires approximately twice as many participants, raising the cost of conducting the trial. In addition, if the drug is efficacious in only one sex, the market for it would be cut in half. To ensure that these important sex-specific analyses are being performed, it is important to understand if, and why, women are under-enrolled in trials. Later in the chapter, the need for further stratified analyses highlights the importance not only of sex-specific analyses but also age, race, and ethnicity, which is relevant in experimental and observational studies. SUMMARY Various methodological considerations should be accounted for and include changing diagnostic criteria for chronic conditions, which would have a direct influence on measures of impact. Population-based data sources and studies capture several chronic conditions but not others, face challenges with capturing various metrics, and are based on various types of reporting (self-report, physician report, etc.). Healthcare claims data may be limited due to dependency on provider coding knowledge and behavior, which does not capture individual experiences, or in some cases, the experiences of subsets of the population (e.g. those on Medicare and at least 65 years old). Many challenges with measuring disease occurrence and impact exist, even when data are available due to the lack of detailed longitudinal data over the entire life course. Support for long-term population-based studies of women is needed. Preliminary findings on risk and protective factors obtained from convenience samples should be validated in population-based samples. Convenience samples of women may include more severe cases, cases with unusual clinical manifestations or earlier age of onset, cases with different comorbidities, or cases with different sociodemographic characteristics. Results of clinical trials and observational studies should be reported separately for men and women. Therefore, power calculations should reflect the need for analyses stratified by sex. Lack of inclusion of women from long-term clinical trials is an issue, and there is a lack of sex-specific analysis for clinical trials. PREPUBLICATION COPY: UNCORRECTED PROOFS 

18 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN REFERENCES AHRQ (Agency for Healthcare Research and Quality). 2020. Chapter 1: Patient registries. In Registries for Evaluating Patient Outcomes: A User’s Guide, 4th ed., edited by R. E. Gliklich, M. B. Leavy, and N. A. Dreyer. Rockville, MD: Agency for Healthcare Research and Quality. Aggarwal, R., S. Ringold, D. Khanna, T. Neogi, S. R. Johnson, A. Miller, H. I. Brunner, R. Ogawa, D. `Felson, A. Ogdie, D. Aletaha, and B. M. Feldman. 2015. Distinctions between diagnostic and classification criteria? Arthritis Care Research 67(7):891–897. Al-Delaimy, W. K., W. C. Willett, J. E. Manson, F. E. Speizer, and F. B. Hu. 2001. Smoking and mortality among women with type 2 diabetes: The nurses’ health study cohort. Diabetes Care 24(12):2043– 2048. Aletaha, D., T. Neogi, A. J. Silman, J. Funovits, D. T. Felson, C. O. Bingham, N. S. Birnbaum, G. R. Burmester, V. P. Bykerk, M. D. Cohen, B. Combe, K. H. Costenbader, M. Dougados, P. Emery, G. Ferraccioli, J. M. W. Hazes, K. Hobbs, T. W. J. Huizinga, A. Kavanaugh, J. Kay, T. K. Kvien, T. Laing, P. Mease, H. A. Ménard, L. W. Moreland, R. L. Naden, T. Pincus, J. S. Smolen, E. Stanislawska‐Biernat, D. Symmons, P. P. Tak, K. S. Upchurch, J. Vencovský, F. Wolfe, and G. Hawker. 2010. 2010 rheumatoid arthritis classification criteria: An American College of Rheumatology/European League Against Rheumatism Collaborative Initiative. Arthritis & Rheumatism 62(9):2569–2581. Angus-Leppan, H. 2013. Migraine: Mimics, borderlands and chameleons. Practical Neurology 13(5):308– 318. APA (American Psychiatric Association). 1987. Diagnostic and statistical manual of mental disorders: DSM-III-R. Washington DC: American Psychiatric Publishing, Inc. APA. 2013. Diagnostic and statistical manual of mental disorders: DSM-5, 5th ed. Arlington, VA: American Psychiatric Publishing, Inc. Arbuckle, M. R., M. T. McClain, M. V. Rubertone, R. H. Scofield, G. J. Dennis, J. A. James, and J. B. Harley. 2003. Development of autoantibodies before the clinical onset of systemic lupus erythematosus. New England Journal of Medicine 349(16):1526–1533. Autoimmune Registry. 2023. Patient-Powered Research for All Autoimmune Diseases. https://www.autoimmuneregistry.org/ (accessed December 7, 2023). Banner Alzheimer’s Institute. 2024. Alzheimer’s Prevention Registry. https://www.endalznow.org/ (accessed December 7, 2023). Beran, T. N., and C. Violato. 2010. Structural equation modeling in medical research: A primer. BMC Research Notes 3(1):267. CDC (Centers for Disease Control and Prevention). 2014. About BRFSS. https://www.cdc.gov/brfss/about/index.htm (accessed December 7, 2023). CDC. 2020a. National Health and Nutrition Examination Survey (NHANES). https://health.gov/healthypeople/objectives-and-data/data-sources-and-methods/data- sources/national-health-and-nutrition-examination-survey-nhanes (accessed December 7, 2023). CDC. 2020b. 2017-March 2020 pre-pandemic examination variable list. https://wwwn.cdc.gov/nchs/nhanes/Search/DataPage.aspx?Component=Examination&Cycle=201 7-2020 (accessed June 18, 2024). CDC. 2023a. LlCP 2022 codebook report: Overall version data weighted with _llcpwt: Behavioral risk factor surveillance system. https://www.cdc.gov/brfss/annual_data/annual_2022.html (accessed June 18, 2024). CDC. 2023b. National diabetes statistics report: Appendix. https://www.cdc.gov/diabetes/php/data- research/appendix.html#cdc_report_pub_study_section_2-table-2 (accessed June 18,2024). Chronic Conditions Data Warehouse. 2023. Welcome to the Chronic Conditions Data Warehouse. https://www2.ccwdata.org/web/guest/home/ (accessed December 7, 2023). PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 19 CMS. 2023. Chronic Conditions: Table: Prevalence State Level: All Beneficiaries by Sex and Age 2018. https://www.cms.gov/data-research/statistics-trends-and-reports/chronic-conditions/chronic- conditions (accessed March 4, 2024). Duarte-García, A., M. Hocaoglu, M. Valenzuela-Almada, S. A. Osei-Onomah, J. Y. Dabit, A. Sanchez- Rodriguez, S. Q. Duong, R. E. Giblon, H. E. Langenfeld, G. S. Alarcón, C. G. Helmick, and C. S. Crowson. 2022. Rising incidence and prevalence of systemic lupus erythematosus: A population- based study over four decades. Annals of the Rheumatic Diseases 81(9): 1260–1266. Evans, C. R., D. R. Williams, J. P. Onnela, and S. V. Subramanian. 2018. A multilevel approach to modeling health inequalities at the intersection of multiple social identities. Social Science & Medicine 203:64–73. Falasinnu, T., Y. Chaichian, J. Li, S. Chung, B. E. Waitzfelder, S. P. Fortmann, L. Palaniappan, and J. F. Simard. 2019. Does SLE widen or narrow race/ethnic disparities in the risk of five co-morbid conditions? Evidence from a community-based outpatient care system. Lupus 28(14):1619–1627. FDA (U.S. Food and Drug Administration). 2023. FDA converts novel Alzheimer's disease treatment to traditional approval: Action follows confirmatory trial to verify clinical benefit. https://www.fda.gov/news-events/press-announcements/fda-converts-novel-alzheimers-disease- treatment-traditional-approval (accessed December 7, 2023). FDA. 1997. General considerations for the clinical evaluation of drugs. https://www.fda.gov/regulatory- information/search-fda-guidance-documents/general-considerations-clinical-evaluation-drugs (accessed December 7, 2024). Flanagan, B. E., E. W. Gregory, E. J. Hallisey, J. L. Heitgerd, and B. Lewis. 2011. A social vulnerability index for disaster management. Journal of Homeland Security and Emergency Management 8(1). Fryer, J., A. J. Mason-Jones, and A. Woodward. 2024 (unpublished). Understanding diagnostic delay for endometriosis: A scoping review. MedRxiv. Advance online publication. https://doi.org/10.1101/2024.01.08.24300988 (accessed June 18, 2024). Grosse, S. D., D. J. Lollar, V. A. Campbell, and M. Chamie. 2009. Disability and disability-adjusted life `years: Not the same. Public Health Reports 124(2):197–202. Harris, H. R., K. M. N. Korkes, T. Li, M. Kvaskoff, E. Cho, L. F. Carvalho, A. A. Qureshi, M. Abrao, and S. A. Missmer. 2022. Endometriosis, psoriasis, and psoriatic arthritis: A prospective cohort study. American Journal of Epidemiology 191(6):1050–1060. Hartman, M., A. Martin, L. Whittle, A. Catlin, and The National Health Expenditure Accounts Team. 2024. National health care spending in 2022: Growth similar to prepandemic rates. Health Affairs 43(1):6–17. Haye, S., J. Thunell, G. Joyce, P. Ferido, B. Tysinger, M. Jacobson, and J. Zissimopoulos. 2023. Estimates of diagnosed dementia prevalence and incidence among diverse beneficiaries in traditional Medicare and Medicare Advantage. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring 15(3). Hernán, M., and J. Robins. 2020. Causal inference: What if. Boca & Raton: Chapman & Hall/CRC. HHS (Department of Health and Human Services). 2013. National Health and Nutrition Examination Survey: Plan and operations, 1999–2010. Hyattsville, MD: National Center for Health Statistics. HHS. 2022. Frequently asked questions about medical and public health registries: What is a registry? https://ncvhs.hhs.gov/wp-content/uploads/2022/11/FAQ-on-Public-Health-Registries.pdf (accessed June 18, 2024). Howren, M. B. 2013. Quality-adjusted life years (QALYS). In Encyclopedia of Behavioral Medicine, edited by M. D. Gellman and J. R. Turner. New York, NY: Springer New York. Pp. 1605–1606. Hudomiet, P., M. D. Hurd, and S. Rohwedder. 2022. Trends in inequalities in the prevalence of dementia in the United States. Proceedings of the National Academy of Sciences 119(46). IHME. 2023. Global Health Data Exchange. https://ghdx.healthdata.org/ (accessed November 29, 2023). Izmirly, P. M., H. Parton, L. Wang, W. J. McCune, S. S. Lim, C. Drenkard, E. D. Ferucci, M. Dall’Era, C. Gordon, C. G. Helmick, and E. C. Somers. 2021. Prevalence of systemic lupus erythematosus in PREPUBLICATION COPY: UNCORRECTED PROOFS 

20 ADVANCING RESEARCH ON CHRONIC CONDITIONS IN WOMEN the United States: Estimates from a meta‐analysis of the Centers for Disease Control and Prevention national lupus registries. Arthritis & Rheumatology 73(6):991–996. Kaiser Permanente. 2024. Fast Facts. https://about.kaiserpermanente.org/who-we-are/fast-facts (accessed June 18, 2024). Langer, R. D., E. White, C. E. Lewis, J. M. Kotchen, S. L. Hendrix, and M. Trevisan. 2003. The Women’s Health Initiative Observational study: Baseline characteristics of participants and reliability of baseline measures. Annals of Epidemiology 13(9, Supplement):S107–S121. Lash, T. L., M. P. Fox, and A. K. Fink. 2009. Applying quantitative bias analysis to epidemiologic data. Statistics for Biology and Health. New York, NY: Springer. Lee, H., R. D. Herbert, and J. H. McAuley. 2019. Mediation analysis. JAMA 321(7):697–698. Lipsky, A. M., and S. Greenland. 2022. Causal directed acyclic graphs. JAMA 327(11):1083–1084. Liu, X., Y. Guo, F. Wang, Y. Yu, Y. Yan, H. Wen, F. Shi, Y. Wang, X. Wang, H. Shen, S. Li, Y. Gong, S. Ke, W. Zhang, Q. Jin, G. Zhang, Y. Wu, M. Zhou, and C. Yu. 2023. Disability weight measurement for the severity of different diseases in Wuhan, China. Population Health Metrics 21(1). Marmot, M. 2013. Universal health coverage and social determinants of health. The Lancet 382(9900):1227–1228. Mirin, A. A. 2021. Gender disparity in the funding of diseases by the U.S. National Institutes of Health. Journal of Women’s Health 30(7):956–963. NCHS (National Center for Health Statistics). 2023. National Health Interview Survey. https://www.cdc.gov/nchs/nhis/about_nhis.htm (accessed June 18, 2023). NIH (National Institutes of Health). 2024a. List of Registries. https://www.nih.gov/health-information/nih- clinical-research-trials-you/list-registries (accessed June 18, 2024). NIH. 2024b. Surveillance, Epidemiology, and End Results (SEER): Surveillance Research Program. https://surveillance.cancer.gov/blog/seer-50-anniversary.html (accessed June 18, 2024). Nnoaham, K. E., P. Webster, J. Kumbang, S. H. Kennedy, and K. T. Zondervan. 2012. Is early age at menarche a risk factor for endometriosis? A systematic review and meta-analysis of case-control studies. Fertility and Sterility 98(3):702–712.e706. NHS (Nurses’ Health Study). 2016. History. https://nurseshealthstudy.org/about-nhs/history (accessed December 6, 2023). Nussinovitch, U., and Y. Shoenfeld. 2012. The role of gender and organ specific autoimmunity. Autoimmunity Reviews 11(6):A377–A385. ORDRCC (Oklahoma Rheumatic Disease Research Core Center). 2023. ORDRCC. https://ordrcc.omrf.org/ (accessed December 7, 2023). Paradies, Y., J. Ben, N. Denson, A. Elias, N. Priest, A. Pieterse, A. Gupta, M. Kelaher, and G. Gee. 2015. Racism as a determinant of health: A systematic review and meta-analysis. PLOS ONE 10(9):e0138511. Pesce, G., F. Artaud, E. Roze, I. Degaey, B. Portugal, T. T. H. Nguyen, A. Fournier, M.-C. Boutron-Ruault, G. Severi, A. Elbaz, and M. Canonico. 2022. Reproductive characteristics, use of exogenous hormones and Parkinson disease in women from the E3N study. Brain 146(6):2535–2546. Petersen, R. C., R. O. Roberts, D. S. Knopman, Y. E. Geda, R. H. Cha, V. S. Pankratz, B. F. Boeve, E. G. Tangalos, R. J. Ivnik, and W. A. Rocca. 2010. Prevalence of mild cognitive impairment is higher in men. Neurology 75(10):889–897. Porta, M. 2014. A dictionary of epidemiology. New York, NY: Oxford University Press. Rajan, K. B., J. Weuve, L. L. Barnes, E. A. McAninch, R. S. Wilson, and D. A. Evans. 2021. Population estimate of people with clinical Alzheimer’s disease and mild cognitive impairment in the United States (2020–2060). Alzheimer’s and Dementia 17(12):1966–1975. Reeves, A. N., M. R. Elliott, M. M. Brooks, C. A. Karvonen-Gutierrez, I. Bondarenko, M. M. Hood, and S. D. Harlow. 2021. Symptom clusters predict risk of metabolic-syndrome and diabetes in midlife: The study of women's health across the nation. Annals of Epidemiology 58:48–55. PREPUBLICATION COPY: UNCORRECTED PROOFS 

METHODOLOGICAL CONSIDERATIONS 21 Rocca, W. A. 2017. Time, sex, gender, history, and dementia. Alzheimer Disease & Associated Disorders 31(1):76–79. Rocca, W. A., B. R. Grossardt, S. M. Brue, C. M. Bock-Goodner, A. M. Chamberlain, P. M. Wilson, L. J. Finney Rutten, and J. L. St. Sauver. 2018. Data resource profile: Expansion of the Rochester Epidemiology Project medical records-linkage system (E-REP). International Journal of Epidemiology 47(2):368–368j. Rocca, W. A., C. Y. Smith, L. Gazzuola Rocca, R. Savica, and M. M. Mielke. 2022. Association of premenopausal bilateral oophorectomy with Parkinsonism and Parkinson disease. JAMA Network Open 5(10):e2238663. Rothman, K. J., S. Greenland, and T. Lash. 2008. Modern epidemiology, 3rd ed. Philadelphia, PA: Lippincott Williams and Wilkins. Santoro, N., and K. Sutton-Tyrrell. 2011. The SWAN song: Study of Women’s Health Across the Nation’s recurring themes. Obstetrics and Gynecology Clinics of North America 38(3):417–423. Sarafrazi, N., E. A. Wambogo, and J. A. Shepherd. 2021. Osteoporosis or low bone mass in older adults: United States, 2017-2018. NCHS Data Brief (405):1–8. https://www.cdc.gov/nchs/data/databriefs/db405-H.pdf (accessed June 18, 2024). Socialstyrelsen. 2024. National cancer register. https://www.socialstyrelsen.se/en/statistics-and- data/registers/national-cancer-register/ (accessed April 17, 2024). St. Sauver, J. L., B. R. Grossardt, C. L. Leibson, B. P. Yawn, L. J. Melton, III, and W. A. Rocca. 2012. Generalizability of epidemiological findings and public health decisions: An illustration from the Rochester Epidemiology Project. Mayo Clinic Proceedings 87(2):151–160. Szklo, M. 2018. Epidemiology: Beyond the basics, 4th ed. Burlington, MA: Jones & Bartlett Learning. SWAN. 2024. About SWAN. https://www.swanstudy.org/about/about- swan/#:~:text=The%20research%20centers%20are%20located%20in%20the%20following,of%2 0Medicine%29%2C%20and%20Pittsburgh%2C%20PA%20%28University%20of%20Pittsburgh %29 (accessed December 7, 2023). Temkin, S. M., E. Barr, H. Moore, J. P. Caviston, J. G. Regensteiner, and J. A. Clayton. 2023. Chronic conditions in women: The development of a National Institutes of Health framework. BMC Women’s Health 23(1):162. van Dyck, C. H., C. J. Swanson, P. Aisen, R. J. Bateman, C. Chen, M. Gee, M. Kanekiyo, D. Li, L. Reyderman, S. Cohen, L. Froelich, S. Katayama, M. Sabbagh, B. Vellas, D. Watson, S. Dhadda, M. Irizarry, L. D. Kramer, and T. Iwatsubo. 2022. Lecanemab in early Alzheimer’s disease. New England Journal of Medicine 388(1):9–21. Weintraub, W. S. 2023. The economic burden of illness. JAMA Network Open 6(3):e232663. WHI. 2021. About WHI. https://www.whi.org/about-whi (accessed December 6, 2023). Whitwell, J. L. 2012. Comparison of imaging biomarkers in the Alzheimer Disease Neuroimaging Initiative and the Mayo Clinic Study of Aging. Archives of Neurology 69(5):614. WHO (World Health Organization). 2023. Disability-Adjusted Life Years (DALYs). https://www.who.int/data/gho/indicator-metadata-registry/imr-details/158 (accessed December 21, 2023). Williams, D. R., Y. Yan, J. S. Jackson, and N. B. Anderson. 1997. Racial differences in physical and mental health: Socio-economic status, stress and discrimination. Journal of Health Psychology 2(3):335– 351. Williams, D. R., J. A. Lawrence, and B. A. Davis. 2019. Racism and health: Evidence and needed research. Annual Review of Public Health 40:105–125. Wilson, C., G. Janes, and J. Williams. 2022. Identity, positionality and reflexivity: Relevance and application to research paramedics. British Paramedic Journal 7(2):43–49. PREPUBLICATION COPY: UNCORRECTED PROOFS