Cross-classified multilevel models (CCMM) in health research: A systematic review of published empirical studies and recommendations for best practices

Best of Both Worlds: Bridging One Model for All and Group-Specific Model Approaches using Ensemble-based Subpopulation Modeling

Subpopulation models have become of increasing interest in prediction of clinical outcomes because they promise to perform better for underrepresented patient subgroups. However, the personalization benefits gained from these models tradeoff their statistical power, and can be impractical when the subpopulation's sample size is small. We hypothesize that a hierarchical model in which population information is integrated into subpopulation models would preserve the personalization benefits and offset the loss of power. In this work, we integrate ideas from ensemble modeling, personalization, and hierarchical modeling and build ensemble-based subpopulation models in which specialization relies on whole group samples. This approach significantly improves the precision of the positive class, especially for the underrepresented subgroups, with minimal cost to the recall. It consistently outperforms one model for all and one model for each subgroup approaches, especially in the presence of a high class-imbalance, for subgroups with at least 380 training samples.

Methodological insights into multilevel analysis of individual heterogeneity and discriminatory accuracy: An empirical examination of the effects of strata configurations on between-stratum variance and of fixed effects across hierarchical levels

This study aims to advance the Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) approach by addressing two key questions. First, it investigates the impact of using increasingly complex combinations of variables to create intersectional strata on between-stratum variance, measured by the variance partitioning coefficients (VPCs). Second, it examines the stability of coefficients for fixed effects across models with an increasing number of hierarchical levels. The analysis is performed using data from a survey of over 42,000 respondents on the prevalence of gender-based violence in European research organisations conducted in 2022. Results indicate that the number of intersectional strata is not significantly related to the proportion of the total variance attributable to the variance between intersectional strata in the MAIHDA approach. Moreover, the coefficients remain relatively stable and consistent across models with increasing complexity, where levels about organisations and countries are added. The analysis concludes that the MAIHDA approach can be flexibly applied for different research purposes, either to better account for structures of power and inequality; or to provide intersectionality-sensitive estimates. The findings underscore the need for researchers to clarify the specific aims of using MAIHDA, whether descriptive or inferential, and highlight the approach's versatility in addressing intersectionality within quantitative research. The study contributes to the literature by offering empirical evidence on the methodological considerations in applying the MAIHDA approach, thereby aiding in its more effective use for intersectional research.

Complementary and Alternative Medicine Use, Cancer Screening, and Medical Checkups in Japan from 2001 to 2013: A Repeated Cross-Sectional Study

Introduction: The association between complementary and alternative medicine (CAM) use and other health-related behaviors is an important issue. A previous study reported that complementary medicine use is related to greater use of cancer screening, whereas alternative medicine use is related to lower use of cancer screening. Given the sparse evidence from Japan, we aimed to examine the association of CAM use with cancer screening and medical checkups. We used a repeated cross-sectional survey of a nationally representative sample of Japanese people and conducted age-period-cohort analysis. Methods: The study population was 68,217 of 83,827 individuals observed from 2001 to 2013 who received cancer screening. CAM users were defined as individuals who received acupuncture, moxibustion, anma/massage/shiatsu, or judo therapy for their most worrisome symptom. The outcomes of interest were receiving stomach, lung, colorectal, uterine, and breast cancer screenings and medical checkups. Using crossclassified multilevel logistic regression models, we estimated odds ratios (ORs) and 95% credible intervals (CIs) for cancer screening and medical checkups. Results: For CAM users of complementary medicine, the adjusted ORs for stomach, lung, and colorectal cancer screening were 1.40 (95% CI: 1.35-1.44), 1.37 (95% CI: 1.34-1.40), and 1.52 (95% CI: 1.49-1.54), respectively. We found similar results for uterine and breast cancer screening, and medical checkups. Conclusions: Irrespective of whether they use CAM, CAM users in Japan tend to receive a variety of cancer screenings and medical checkups.

Implementation Research at NHLBI: Methodological and Design Challenges and Lessons Learned from the DECIPHeR Initiative

NHLBI funded seven projects as part of the Disparities Elimination through Coordinated Interventions to Prevent and Control Heart and Lung Disease Risk (DECIPHeR) Initiative. They were expected to collaborate with community partners to (1) employ validated theoretical or conceptual implementation research frameworks, (2) include implementation research study designs, (3) include implementation measures as primary outcomes, and (4) inform our understanding of mediators and mechanisms of action of the implementation strategy. Several projects focused on late-stage implementation strategies that optimally and sustainably delivered two or more evidence-based multilevel interventions to reduce or eliminate cardiovascular and/or pulmonary health disparities and to improve population health in high-burden communities. Projects that were successful in the three-year planning phase transitioned to a 4-year execution phase. NHLBI formed a Technical Assistance Workgroup during the planning phase to help awardees refine study aims, strengthen research designs, detail analytic plans, and to use valid sample size methods. This paper highlights methodological and study design challenges encountered during this process. Important lessons learned included (1) the need for greater emphasis on implementation outcomes, (2) the need to clearly distinguish between intervention and implementation strategies in the protocol, (3) the need to address clustering due to randomization of groups or clusters, (4) the need to address the cross-classification that results when intervention agents work across multiple units of randomization in the same arm, (5) the need to accommodate time-varying intervention effects in stepped-wedge designs, and (6) the need for data-based estimates of the parameters required for sample size estimation.

Construction of multilevel statistical models in health research: Foundations and generalities

This topic review aims to present a global vision of multilevel analysis models’ applicability to health research, explaining its theoretical, methodological, and statistical foundations. We describe the basic steps to build these models and examples of their application according to the data hierarchical structure. It ir worth noticing that before using these models, researchers must have a rationale for needing them, and a statistical evaluation accounting for the variance percentage explained by the observations grouping effect. The requirements to conduct this type of analysis depends on special conditions such as the type of variables, the number of units per level, or the type of hierarchical structure. We conclude that multilevel analysis models are a useful tool to integrate information, considering the complexity of the relationships and interactions involved in most health conditions, including the loss of independence between observation units.

Time trends in associations between neighbourhood and school characteristics and mental health problems among Dutch adolescents

We investigated trends in associations between physical and social neighbourhood and school characteristics and adolescent mental health problems between 2005 and 2017. Nationally representative, repeated cross-sectional data collected in the Netherlands among primary (N = 5,871) and secondary school students (N = 20,778) were analysed through cross-classified multilevel models. Hardly any evidence was found for over-time changes in associations between neighbourhood and school characteristics and adolescent mental health problems. Findings showed that for both groups of students, only social characteristics within the two contexts were associated with adolescent mental health problems. For secondary school students, school effects were larger than neighbourhood effects, while the opposite was true for primary school students. Specifically, primary school students residing in more socially fragmented or lower socioeconomic status (SES) neighbourhoods, and lower SES schools, reported more conduct problems and peer relationship problems. For secondary school students, only the SES of the neighbourhood and the school was associated with all four aspects of mental health problems. Remarkably, the direction of the associations between neighbourhood/ school SES and adolescent mental health problems varied across the different mental health outcomes. More research is warranted to replicate our findings.

Adolescent alcohol use: use of social network analysis and cross-classified multilevel modeling to examine peer group, school, and neighborhood-level influences

Background: Neighborhood-, school-, and peer-contexts play an important role in adolescent alcohol use behaviors. Methodological advances permit simultaneous modeling of these contexts to understand their relative and joint importance. Few empirical studies include these contexts, and studies that do typically: examine each context separately; include contexts for the sole purpose of accounting for clustering in the data; or do not disaggregate by sex.Objectives: This study takes an eco-epidemiologic approach to examine the role of socio-contextual contributions to variance in adolescent alcohol use. The primary parameters of interest are therefore variance rather than beta parameters (i.e. random rather than fixed effects). Sex-stratified models are also used to understand how each context may matter differently for male and female adolescents.Method: Data come from the National Longitudinal Study of Adolescent to Adult Health (n = 8,534 females, n = 8,102 males). We conduct social network analysis and traditional and cross-classified multilevel models (CCMM) in the full and sex-disaggregated samples.Results: In final CCMM, peer groups, schools, and neighborhoods contributed 10.5%, 10.8%, and 0.4%, respectively, to total variation in adolescent alcohol use. Results do not differ widely by gender.Conclusions: Peer groups and schools emerge as more salient contributing contexts relative to neighborhoods in adolescent alcohol use for males and females. These findings have both methodological and practical implications. Multilevel modeling can model contexts simultaneously to prevent the overestimation of variance in youth alcohol use explained by each context. Primary prevention strategies addressing youth alcohol use should focus on schools and peer networks.

Disparities in healthcare utilization by insurance status among patients with symptomatic peripheral artery disease

BACKGROUND

Peripheral artery disease (PAD) is a common circulatory disorder associated with increased hospitalizations and significant health care-related expenditures. Among patients with PAD, insurance status is an important determinant of health care utilization, treatment of disease, and treatment outcomes. However, little is known about PAD-costs differences across different insurance providers. In this study we examined possible disparities in length of stay and total charge of inpatient hospitalizations among patients with PAD by insurance type.

METHODS

We conducted a cross-sectional analysis of length of stay and total charge by insurance provider for all hospitalizations for individuals with PAD in South Carolina (2010-2018). Cross-classified multilevel modeling was applied to account for the non-nested hierarchical structure of the data, with county and hospital included as random effects. Analyses were adjusted for patient age, race/ethnicity, county, year of admission, admission type, all-patient refined diagnostic groups, and Charlson comorbidity index.

RESULTS

Among 385,018 hospitalizations for individuals with PAD in South Carolina, the median length of stay was 4 days (IQR: 5) and the median total charge of hospitalization was $43,232 (IQR: $52,405). Length of stay and total charge varied significantly by insurance provider. Medicare patients had increased length of stay (IRR = 1.08, 95 CI%: 1.07, 1.09) and higher total charges (β: 0.012, 95% CI: 0.007, 0.178) than patients with private insurance. Medicaid patients also had increased length of stay (IRR = 1.26, 95% CI: 1.24,1.28) but had lower total charges (β: -0.022, 95% CI: -0.003. -0.015) than patients with private insurance.

CONCLUSIONS

Insurance status was associated with inpatient length of stay and total charges in patients with PAD. It is essential that Medicare and Medicaid individuals with PAD receive proper management and care of their PAD, particularly in the primary care settings, to prevent hospitalizations and reduce the excess burden on these patients.

Restoring the missing person to personalized medicine and precision psychiatry

Precision psychiatry has emerged as part of the shift to personalized medicine and builds on frameworks such as the U.S. National Institute of Mental Health Research Domain Criteria (RDoC), multilevel biological "omics" data and, most recently, computational psychiatry. The shift is prompted by the realization that a one-size-fits all approach is inadequate to guide clinical care because people differ in ways that are not captured by broad diagnostic categories. One of the first steps in developing this personalized approach to treatment was the use of genetic markers to guide pharmacotherapeutics based on predictions of pharmacological response or non-response, and the potential risk of adverse drug reactions. Advances in technology have made a greater degree of specificity or precision potentially more attainable. To date, however, the search for precision has largely focused on biological parameters. Psychiatric disorders involve multi-level dynamics that require measures of phenomenological, psychological, behavioral, social structural, and cultural dimensions. This points to the need to develop more fine-grained analyses of experience, self-construal, illness narratives, interpersonal interactional dynamics, and social contexts and determinants of health. In this paper, we review the limitations of precision psychiatry arguing that it cannot reach its goal if it does not include core elements of the processes that give rise to psychopathological states, which include the agency and experience of the person. Drawing from contemporary systems biology, social epidemiology, developmental psychology, and cognitive science, we propose a cultural-ecosocial approach to integrating precision psychiatry with person-centered care.

Do Complementary and Alternative Medicine Users Also Use Conventional Medicine? A Repeated Cross-Sectional Study in Japan from 1995 to 2013

Introduction: It is recommended that users of complementary and alternative medicine (CAM) also seek conventional medical care to prevent the loss of access to appropriate medical care. However, the status of such use is unknown. The aim of this study was to examine the time trends in the proportion of CAM users who also receive conventional medical care for the same symptoms. Methods: This is a repeated cross-sectional study. Of data for 753,978 respondents to the Comprehensive Survey of Living Conditions, which was conducted seven times between 1995 and 2013, data from 17,707 individuals who used acupuncture, moxibustion, anma-massage-shiatsu, or judo therapy were analyzed. Cross-classified multilevel logistic regression models with individuals as level 1 and survey year and cohort as level 2 were used to calculate odds ratios (ORs) and 95% credible intervals (CIs) for combined use of CAM and conventional medical care. Age was entered as an individual-level variable. The period effect after 2003 was entered as a survey year-level variable because the number of eligible persons providing CAM treatments has increased since 2003. Results: Among the 17,707 CAM users, 11,567 (65.3%) were women. When age was entered as an explanatory variable, the results showed that both older men and women tended to receive conventional medical care (women, OR: 1.04, 95% CI: 1.03-1.04; men, OR: 1.03, 95% CI: 1.02-1.04). Additional examination of the possible period effect after 2003 showed a positive (although nonsignificant) association (women, OR: 1.36, 95% CI: 0.89-1.99; men, OR: 1.37, 95% CI: 0.94-1.91). Conclusions: As patient age increased, patients combined CAM use with conventional medicine. The findings also suggested that the combined use of CAM and conventional medicine has increased since 2003.

Social inequalities, length of hospital stay for chronic conditions and the mediating role of comorbidity and discharge destination: A multilevel analysis of hospital administrative data linked to the population census in Switzerland

Social factors are recognized determinants of morbidity and mortality and also have an impact on use of medical services. The objective of this study was to assess the associations of educational attainment, social and financial resources, and migration factors with length of hospital stays for chronic conditions. In addition, the study investigated the role of comorbidity and discharge destination in mediating these associations. The study made use of nationwide inpatient data that was linked with Swiss census data. The study sample included n = 141,307 records of n = 92,623 inpatients aged 25 to 84 years, hospitalized between 2010 and 2016 for a chronic condition. Cross-classified multilevel models and mediation analysis were performed. Patients with upper secondary and compulsory education stayed longer in hospital compared to those with tertiary education (β 0.24 days, 95% CI 0.14–0.33; β 0.37, 95% CI 0.27–0.47, respectively) when taking into account demographic factors, main diagnosis and clustering on patient and hospital level. However, these effects were almost fully mediated by burden of comorbidity. The effect of living alone on length of stay (β 0.60 days, 95% CI 0.50–0.70) was partially mediated by both burden of comorbidities (33%) and discharge destination (30.4%). (Semi-) private insurance was associated with prolonged stays, but an inverse effect was observed for colon and breast cancer. Allophone patients had also prolonged hospital stays (β 0.34, 95% CI 0.13–0.55). Hospital stays could be a window of opportunity to discern patients who need additional time and support to better cope with everyday life after discharge, reducing the risks of future hospital stays. However, inpatient care in Switzerland seems to take into account rather obvious individual needs due to lack of immediate support at home, but not necessarily more hidden needs of patients with low health literacy and less resources to assert their interests within the health system.

Adolescent individual, school, and neighborhood influences on young adult hypertension risk

BACKGROUND

Geographic and contextual socioeconomic risk factors in adolescence may be more strongly associated with young adult hypertension than individual-level risk factors. This study examines the association between individual, neighborhood, and school-level influences during adolescence on young adult blood pressure.

METHODS

Data were analyzed from the National Longitudinal Study of Adolescent to Adult Health (1994-1995 aged 11-18 and 2007-2008 aged 24-32). We categorized hypertension as systolic blood pressure ≥140 mm Hg and/or diastolic blood pressure ≥90 mm Hg. Secondary outcomes included continuous systolic and diastolic blood pressure. We fit a series of cross-classified multilevel models to estimate the associations between young adulthood hypertension with individual-level, school-level, and neighborhood-level factors during adolescence (i.e., fixed effects) and variance attributable to each level (i.e., random effects). Models were fit using Bayesian estimation procedures. For linear models, intra-class correlations (ICC) are reported for random effects.

RESULTS

The final sample included 13,911 participants in 128 schools and 1,917 neighborhoods. Approximately 51% (7,111) young adults were hypertensive. Individual-level characteristics-particularly older ages, Non-Hispanic Black race, Asian race, male sex, BMI, and current smoking-were associated with increased hypertension. Non-Hispanic Black (OR = 1.21; 95% CI: 1.03-1.42) and Asian (OR = 1.28; 95% CI: 1.02-1.62) students had higher odds of hypertension compared to non-Hispanic White students. At the school level, hypertension was associated with the percentage of non-Hispanic White students (OR for 10% higher = 1.06; 95% CI: 1.01-1.09). Adjusting for individual, school, and neighborhood predictors attenuated the ICC for both the school (from 1.4 null to 0.9 fully-adjusted) and neighborhood (from 0.4 to 0.3).

CONCLUSION

We find that adolescents' schools and individual-level factors influence young adult hypertension, more than neighborhoods. Unequal conditions in school environments for adolescents may increase the risk of hypertension later in life. Our findings merit further research to better understand the mechanisms through which adolescents' school environments contribute to adult hypertension and disparities in hypertension outcomes later in life.

Measurement in Intensive Longitudinal Data

Technological advances have increased the prevalence of intensive longitudinal data as well as statistical techniques appropriate for these data, such as dynamic structural equation modeling (DSEM). Intensive longitudinal designs often investigate constructs related to affect or mood and do so with multiple item scales. However, applications of intensive longitudinal methods often rely on simple sums or averages of the administered items rather than considering a proper measurement model. This paper demonstrates how to incorporate measurement models into DSEM to (1) provide more rigorous measurement of constructs used in intensive longitudinal studies and (2) assess whether scales are invariant across time and across people, which is not possible when item responses are summed or averaged. We provide an example from an ecological momentary assessment study on self-regulation in adults with binge eating disorder and walkthrough how to fit the model in Mplus and how to interpret the results.