Dataset Information

An Original Stepwise Multilevel Logistic Regression Analysis of Discriminatory Accuracy: The Case of Neighbourhoods and Health.

ABSTRACT:

Background and aim

Many multilevel logistic regression analyses of "neighbourhood and health" focus on interpreting measures of associations (e.g., odds ratio, OR). In contrast, multilevel analysis of variance is rarely considered. We propose an original stepwise analytical approach that distinguishes between "specific" (measures of association) and "general" (measures of variance) contextual effects. Performing two empirical examples we illustrate the methodology, interpret the results and discuss the implications of this kind of analysis in public health.

Methods

We analyse 43,291 individuals residing in 218 neighbourhoods in the city of Malmö, Sweden in 2006. We study two individual outcomes (psychotropic drug use and choice of private vs. public general practitioner, GP) for which the relative importance of neighbourhood as a source of individual variation differs substantially. In Step 1 of the analysis, we evaluate the OR and the area under the receiver operating characteristic (AUC) curve for individual-level covariates (i.e., age, sex and individual low income). In Step 2, we assess general contextual effects using the AUC. Finally, in Step 3 the OR for a specific neighbourhood characteristic (i.e., neighbourhood income) is interpreted jointly with the proportional change in variance (i.e., PCV) and the proportion of ORs in the opposite direction (POOR) statistics.

Results

For both outcomes, information on individual characteristics (Step 1) provide a low discriminatory accuracy (AUC = 0.616 for psychotropic drugs; = 0.600 for choosing a private GP). Accounting for neighbourhood of residence (Step 2) only improved the AUC for choosing a private GP (+0.295 units). High neighbourhood income (Step 3) was strongly associated to choosing a private GP (OR = 3.50) but the PCV was only 11% and the POOR 33%.

Conclusion

Applying an innovative stepwise multilevel analysis, we observed that, in Malmö, the neighbourhood context per se had a negligible influence on individual use of psychotropic drugs, but appears to strongly condition individual choice of a private GP. However, the latter was only modestly explained by the socioeconomic circumstances of the neighbourhoods. Our analyses are based on real data and provide useful information for understanding neighbourhood level influences in general and on individual use of psychotropic drugs and choice of GP in particular. However, our primary aim is to illustrate how to perform and interpret a multilevel analysis of individual heterogeneity in social epidemiology and public health. Our study shows that neighbourhood "effects" are not properly quantified by reporting differences between neighbourhood averages but rather by measuring the share of the individual heterogeneity that exists at the neighbourhood level.

SUBMITTER: Merlo J

PROVIDER: S-EPMC4847925 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:There is an increasing focus on structural and social determinants of inequalities in young people's mental health across different social contexts. Taking higher education as a specific social context, it is unclear whether university attendance shapes the impact of intersectional social identities and positions on young people's mental health outcomes. Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) was used to predict the odds that mental distress during adolescence, sex, socioeconomic status, sexual identity, ethnicity, and their intersections, were associated with young people's mental health outcomes at age 25, and whether this differed based on university attendance. Data from the Longitudinal Study of Young People in England cohort study were analysed with the MAIHDA approach, and the results did not reveal any evidence of multiplicative intersectional (i.e., aggravating) effects on young people's mental health outcomes. However, important main effects of social identities and positions (i.e., an additive model) were observed. The findings suggested that being female or identifying as a sexual minority increased the odds of young people experiencing mental health problems at age 25, although the odds of self-harming were half the size for sexual minorities who had attended university. Black and Asian individuals were less likely to declare a mental illness than White individuals. Young people who grew up in a more deprived area and had not attended university were more likely to experience mental health problems. These findings imply that mental health interventions for young people do not necessarily have to be designed exclusively for specific intersectional groups. Further, university attendance appears to produce better mental health outcomes for some young people, hence more investigation is needed to understand what universities do for young people, and whether this could be replicated in the wider general population.

Project description:Background Ethiopia has low malaria prevalence compared to most other malaria-endemic countries in Africa. However, malaria is still a major public health problem in the country. The binary logistic regression model has been widely used to analyse malaria indicator survey (MIS) data. However, most MIS have a hierarchical structure which may result in dependent data. Since this model assumes that conditional on the covariates the malaria statuses of individuals are independent, it ignores potential intra-cluster correlation among observations within a cluster and may generate biased analysis results and conclusions. Therefore, the aim of this study was to quantify the variation in the prevalence of malaria between sample enumeration areas (SEAs) or clusters, the effects of cluster characteristics on the prevalence of malaria using the intra-class correlation coefficient as well as to identify significant factors that affect the prevalence of malaria using the multilevel logistic regression modelling in three major regions of Ethiopia, namely Amhara, Oromia and Southern Nations, Nationalities and Peoples’ (SNNP). Methods Dataset for three regional states extracted from the 2011 Ethiopian National Malaria Indicator Surveys (EMIS) national representative samples was used in this study. It contains 9272 sample individuals selected from these regions. Various multilevel models with random sample SEA effects were applied taking into account the survey design weights. These weights are scaled to address unequal probabilities of selection within clusters. The spatial clustering of malaria prevalence was assessed applying Getis-Ord statistic to best linear unbiased prediction values of model random effects. Results About 53.82 and 28.72 per cents of the sampled households in the study regions had no mosquito net and sprayed at least once within the last 12 months, respectively. The results of this study indicate that age, gender, household had mosquito nets, the dwelling has windows, source of drinking water, the two SEA-level variables, i.e. region and median altitude, were significantly related to the prevalence of malaria. After adjusting for these seven variables, about 45% of the residual variation in the prevalence of malaria in the study regions was due to systematic differences between SEAs, while the remaining 55% was due to unmeasured differences between persons or households. The estimated MOR, i.e. the unexplained SEA heterogeneity, was 4.784. This result suggests that there is high variation between SEAs in the prevalence of malaria. In addition, the 80% interval odds ratios (IORs) related to SEA-level variables contain one suggesting that the SEA variability is large in comparison with the effect of each of the variable. Conclusions The multilevel logistic regression with random effects model used in this paper identified five individual / household and two SEA-level risk factors of malaria infection. Therefore, the public health policy makers should pay attentions to those significant factors, such as improving the availability of pure drinking water. Further, the findings of spatial clustering provide information to health policymakers to plan geographically targeted interventions to control malaria transmission.

Project description:BackgroundIn medical, social, and behavioral research we often encounter datasets with a multilevel structure and multiple correlated dependent variables. These data are frequently collected from a study population that distinguishes several subpopulations with different (i.e., heterogeneous) effects of an intervention. Despite the frequent occurrence of such data, methods to analyze them are less common and researchers often resort to either ignoring the multilevel and/or heterogeneous structure, analyzing only a single dependent variable, or a combination of these. These analysis strategies are suboptimal: Ignoring multilevel structures inflates Type I error rates, while neglecting the multivariate or heterogeneous structure masks detailed insights.MethodsTo analyze such data comprehensively, the current paper presents a novel Bayesian multilevel multivariate logistic regression model. The clustered structure of multilevel data is taken into account, such that posterior inferences can be made with accurate error rates. Further, the model shares information between different subpopulations in the estimation of average and conditional average multivariate treatment effects. To facilitate interpretation, multivariate logistic regression parameters are transformed to posterior success probabilities and differences between them.ResultsA numerical evaluation compared our framework to less comprehensive alternatives and highlighted the need to model the multilevel structure: Treatment comparisons based on the multilevel model had targeted Type I error rates, while single-level alternatives resulted in inflated Type I errors. Further, the multilevel model was more powerful than a single-level model when the number of clusters was higher. A re-analysis of the Third International Stroke Trial data illustrated how incorporating a multilevel structure, assessing treatment heterogeneity, and combining dependent variables contributed to an in-depth understanding of treatment effects. Further, we demonstrated how Bayes factors can aid in the selection of a suitable model.ConclusionThe method is useful in prediction of treatment effects and decision-making within subpopulations from multiple clusters, while taking advantage of the size of the entire study sample and while properly incorporating the uncertainty in a principled probabilistic manner using the full posterior distribution.

Project description:BACKGROUND:Dimensions of social location such as socioeconomic position or sex/gender are often associated with low response rates in epidemiological studies. We applied an intersectionality-informed approach to analyze non-response among population strata defined by combinations of multiple dimensions of social location and subjective health in a health survey in Germany. METHODS:We used data from the cross-sectional sample of the German Health Interview and Examination Survey for Adults (DEGS1) conducted between 2008 and 2011. Information about non-responders was available from a mailed non-responder questionnaire. Intersectional strata were constructed by combining all categories of age, sex/gender, marital status, and level of education in scenario 1. Subjective health was additionally used to construct intersectional strata in scenario 2. We applied multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) to calculate measures of discriminatory accuracy, proportions of non-responders among intersectional strata, as well as stratum-specific total interaction effects (intersectional effects). Markov chain Monte Carlo methods were used to estimate multilevel logistic regression models. RESULTS:Data was available for 6,534 individuals of whom 36% were non-responders. In scenario 2, we found weak discriminatory accuracy (variance partition coefficient = 3.6%) of intersectional strata, while predicted proportions of non-response ranged from 20.6% (95% credible interval (CI) 17.0%-24.9%) to 57.5% (95% CI 48.8%-66.5%) among intersectional strata. No evidence for intersectional effects was found. These results did not differ substantially between scenarios 1 and 2. CONCLUSIONS:MAIHDA revealed that proportions of non-response varied widely between intersectional strata. However, poor discriminatory accuracy of intersectional strata and no evidence for intersectional effects indicate that there is no justification to exclusively target specific intersectional strata in order to increase response, but that a combination of targeted and population-based measures might be appropriate to achieve more equal representation.

Project description:BackgroundIn light of the opioid epidemic in the United States, there is growing concern about the use of opioids in Sweden as it may lead to misuse and overuse and, in turn, severe public health problems. However, little is known about the distribution of opioid use across different demographic and socioeconomic dimensions in the Swedish general population. Therefore, we applied an intersectional Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA), to obtain an improved mapping of the risk heterogeneity of and socioeconomic inequalities in opioid prescription receipt.Methods and findingsUsing data from 6,846,106 residents in Sweden aged 18 and above, we constructed 72 intersectional strata from combinations of gender, age, income, cohabitation status, and presence or absence of psychological distress. We modelled the absolute risk (AR) of opioid prescription receipt in a series of multilevel logistic regression models distinguishing between additive and interaction effects. By means of the Variance Partitioning Coefficient (VPC) and the area under the receiver operating characteristic curve (AUC), we quantified the discriminatory accuracy (DA) of the intersectional strata for discerning those who received opioid prescriptions from those who did not. The AR of opioid prescription receipt ranged from 2.77% (95% CI 2.69-2.86) among low-income men aged 18-34, living alone, without psychological distress, to 28.25% (95% CI 27.95-28.56) among medium-income women aged 65 and older, living alone, with psychological distress. In a model that conflated both additive and interaction effects, the intersectional strata had a fair DA for discerning opioid users from non-users (VPC = 13.2%, AUC = 0.68). However, in the model that decomposed total effects into additive and interaction effects, the VPC was very low (0.42%) indicating the existence of small interaction effects for a number of the intersectional strata.ConclusionsThe intersectional MAIHDA approach aligns with the aims of precision public health, through improving the evidence base for health policy by increasing understanding of both health inequalities and individual heterogeneity. This approach is particularly relevant for socioeconomically conditioned outcomes such as opioid prescription receipt. We have identified intersections of social position within the Swedish population at greater risk for opioid prescription receipt.

Project description:Socioeconomic, ethnic and gender disparities in Chronic Obstructive Pulmonary Disease (COPD) risk are well established but no studies have applied multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) within an intersectional framework to study this outcome. We study individuals at the first level of analysis and combinations of multiple social and demographic categorizations (i.e., intersectional strata) at the second level of analysis. Here we used MAIHDA to assess to what extent individual differences in the propensity of developing COPD are at the intersectional strata level. We also used MAIHDA to determine the degree of similarity in COPD incidence of individuals in the same intersectional stratum. This leads to an improved understanding of risk heterogeneity and of the social dynamics driving socioeconomic and demographic disparities in COPD incidence. Using data from 2,445,501 residents in Sweden aged 45-65, we constructed 96 intersectional strata combining categories of age, gender, income, education, civil- and migration status. The incidences of COPD ranged from 0.02% for young, native males with high income and high education who cohabited to 0.98% for older native females with low income and low education who lived alone. We calculated the intra-class correlation coefficient (ICC) that informs on the discriminatory accuracy of the categorizations. In a model that conflated additive and interaction effects, the ICC was good (20.0%). In contrast, in a model that measured only interaction effects, the ICC was poor (1.1%) suggesting that most of the observed differences in COPD incidence across strata are due to the main effects of the categories used to construct the intersectional matrix while only a minor share of the differences are attributable to intersectional interactions. We found conclusive interaction effects. The intersectional MAIHDA approach offers improved information to guide public health policies in COPD prevention, and such policies should adopt an intersectional perspective.