Statistical and Substantive Inferences in Public Health: Issues in the Application of Multilevel Models

Efficient analysis of Q-level nested hierarchical general linear models given ignorable missing data

This article extends single-level missing data methods to efficient estimation of a Q-level nested hierarchical general linear model given ignorable missing data with a general missing pattern at any of the Q levels. The key idea is to reexpress a desired hierarchical model as the joint distribution of all variables including the outcome that are subject to missingness, conditional on all of the covariates that are completely observed and to estimate the joint model under normal theory. The unconstrained joint model, however, identifies extraneous parameters that are not of interest in subsequent analysis of the hierarchical model and that rapidly multiply as the number of levels, the number of variables subject to missingness, and the number of random coefficients grow. Therefore, the joint model may be extremely high dimensional and difficult to estimate well unless constraints are imposed to avoid the proliferation of extraneous covariance components at each level. Furthermore, the over-identified hierarchical model may produce considerably biased inferences. The challenge is to represent the constraints within the framework of the Q-level model in a way that is uniform without regard to Q; in a way that facilitates efficient computation for any number of Q levels; and also in a way that produces unbiased and efficient analysis of the hierarchical model. Our approach yields Q-step recursive estimation and imputation procedures whose qth-step computation involves only level-q data given higher-level computation components. We illustrate the approach with a study of the growth in body mass index analyzing a national sample of elementary school children.

Associations of neighborhood and family factors with trajectories of physical and social aggression during adolescence

Adolescents develop within multiple contexts that synergistically influence their behavior and health. To understand the simultaneous influence of neighborhood and family contexts on adolescents, this study examined relationships of neighborhood socioeconomic disadvantage, neighborhood social disorganization, family conflict, parent-child bonding and parental control with trajectories of physical and social aggression. The sample included 5,118 adolescents between ages 11 and 18 (50% female, 52% Caucasian) living in predominantly rural areas. Multilevel growth curve models showed an interaction between neighborhood disadvantage, family conflict and gender on the physical aggression trajectories. The interaction suggested more rapid processes of both increase in and desistance from physical aggression over time for boys with high neighborhood disadvantage and high family conflict, as well as a higher starting point, more gradual increase and slower process of desistance over time for girls in similar neighborhood and family contexts. Less parent-child bonding and less parental control also were associated with higher initial levels of physical aggression. For social aggression, an interaction between family conflict and gender showed girls with high family conflict had the highest initial levels of social aggression, with a more gradual increase over time for these girls compared to their male counterparts in high-conflict families or their female counterparts in low-conflict families. Less parent-child bonding was associated with higher initial levels and a faster increase over time of social aggression, and less parental control was associated with higher initial levels of social aggression. The findings suggest early family-based interventions may help prevent perpetration of both physical and social aggression during adolescence.

Assessing the psychometric and ecometric properties of neighborhood scales in developing countries: Saúde em Beagá Study, Belo Horizonte, Brazil, 2008-2009

Although specific measurement instruments are necessary to better understand the relationship between features of neighborhoods and health, very few studies have developed instruments to measure neighborhood features in developing countries. The objective of the study was to develop valid and reliable measures of neighborhood context useful in a Latin American urban context, assess their psychometric and ecometric properties, and examine individual and neighborhood-level predictors of these measures. We analyzed data from a multistage household survey (2008-2009) conducted in Belo Horizonte City by the Observatory for Urban Health. One adult in each household was selected to answer a questionnaire that included scales to measure neighborhood domains. Census tracts were used to proxy neighborhoods. Internal consistency was evaluated by Cronbach's alpha, and multilevel models were used to estimate ecometric properties and to estimate associations of neighborhood measures with socioeconomic indicators. The final sample comprised 4048 survey respondents representing 149 census tracts. We assessed ten neighborhood environment dimensions: public services, aesthetic quality, walking environment, safety, violence, social cohesion, neighborhood participation, neighborhood physical disorder, neighborhood social disorder, and neighborhood problems. Cronbach's alpha coefficients ranged from 0.53 to 0.83; intraneighborhood correlations ranged from 0.02 to 0.53, and neighborhood reliability varied from 0.76 to 0.99. Most scales were associated with individual and neighborhood socioeconomic predictors. Questionnaires can be used to reliably measure neighborhood contexts in developing countries.

Sleep quality and its psychological correlates among university students in Ethiopia: a cross-sectional study

BACKGROUND

Sleep is an important physiological process for humans. University students in most resource limited countries often report poor sleep quality due to changing social opportunities and increasing academic demands. However, sleep quality among university students has not been studied in Ethiopia. Thus, this study assessed sleep quality and its demographic and psychological correlates among university students.

METHODS

A cross-sectional survey was conducted in two universities in Ethiopia. Multistage sampling procedures were used to enroll 2,817 students into the study. A self-administered structured questionnaire including the Pittsburgh Sleep Quality Index (PSQI), the Depression Anxiety Stress Scale-21, the Perceived Stress Scale (PSS) and selected modules of the World Health Organization STEPS instrument was used for the study. This research included 2,551 students. Frequency, median, mean with standard deviation and 95% confidence interval were used to characterize sleep quality and other variables. Analysis of variance and binary logistic regression procedures were also used.

RESULT

The prevalence of poor sleep quality (total PSQI score > 5) was 55.8% (1,424). Female students (adjusted odds ratio (AOR) 1.23; 95% CI: 1.00, 1.57), second year (AOR 2.91; 95% CI: 2.1, 4.02) and third year students (AOR 2.25; 95% CI 1.62, 3.12) had statistically significant higher odds of poor sleep quality. Perceived stress level and symptoms of depression and anxiety were strongly associated with sleep quality.

CONCLUSION

A substantial proportion of university students are affected by poor sleep quality. If our results are confirmed in prospective studies, health promotion and educational programs for students should emphasize the importance of sleep and mental health.

Clarifying the use of aggregated exposures in multilevel models: self-included vs. self-excluded measures

BACKGROUND

Multilevel analyses are ideally suited to assess the effects of ecological (higher level) and individual (lower level) exposure variables simultaneously. In applying such analyses to measures of ecologies in epidemiological studies, individual variables are usually aggregated into the higher level unit. Typically, the aggregated measure includes responses of every individual belonging to that group (i.e. it constitutes a self-included measure). More recently, researchers have developed an aggregate measure which excludes the response of the individual to whom the aggregate measure is linked (i.e. a self-excluded measure). In this study, we clarify the substantive and technical properties of these two measures when they are used as exposures in multilevel models.

METHODS

Although the differences between the two aggregated measures are mathematically subtle, distinguishing between them is important in terms of the specific scientific questions to be addressed. We then show how these measures can be used in two distinct types of multilevel models-self-included model and self-excluded model-and interpret the parameters in each model by imposing hypothetical interventions. The concept is tested on empirical data of workplace social capital and employees' systolic blood pressure.

RESULTS

Researchers assume group-level interventions when using a self-included model, and individual-level interventions when using a self-excluded model. Analytical re-parameterizations of these two models highlight their differences in parameter interpretation. Cluster-mean centered self-included models enable researchers to decompose the collective effect into its within- and between-group components. The benefit of cluster-mean centering procedure is further discussed in terms of hypothetical interventions.

CONCLUSIONS

When investigating the potential roles of aggregated variables, researchers should carefully explore which type of model-self-included or self-excluded-is suitable for a given situation, particularly when group sizes are relatively small.

Smoking in context: a multilevel analysis of 49,088 communities in Canada

BACKGROUND

The extent to which the prevalence of smoking in Canada varies across geographic areas independently of individual characteristics has not been quantified.

PURPOSE

To estimate the extent and potential sources of geographic variation in smoking among communities, health regions, and provinces/territories in Canada.

METHODS

Data are from the Canadian Community Health Surveys conducted between 2001 and 2008 (n=461,709). Current cigarette smoking among adults (aged ≥18 years) was the primary outcome. Individual-level markers of SES were education, household income, and occupation. Contextual variables potentially related to smoking considered were provincial cigarette taxes, workplace smoking bans, and collective family norms discouraging smoking in communities. A multilevel logistic regression analysis was conducted to model variation in smoking at the geographic scale of communities, health regions, and provinces.

RESULTS

Overall, the contribution of geography as a percentage of the total variation in smoking was 8.4%, with 2.4% attributable to provinces, 1.2% attributable to health regions, and 4.8% attributable to communities after adjusting for age, gender and survey period. In models that accounted for socioeconomic and demographic characteristics in addition to age and gender, the contribution of geography to the total variation in smoking was attenuated to 4.1%; with 2.0% at the province level, 0.4% at the health region level, and 1.7% at the community level. Within provinces/territories, the community variation in smoking ranged from 2.4% in Prince Edward Island to 9.1% in British Columbia. Nationally, 71% of community and 21% of provincial differences in smoking were explained by individual, socioeconomic, and demographic factors alone; the inclusion of contextual covariates explained an additional 27% of the variation among communities. Collective family norms discouraging smoking in a community was the strongest contextual predictor of individual smoking; provincial cigarette taxes and workplace bans were only modestly related to individual smoking behavior.

CONCLUSIONS

Geographic variation in smoking remained after accounting for individual, socioeconomic, and demographic characteristics, suggesting the importance of place, at the level of provinces and communities in Canada. Remaining community variation in smoking was largely attenuated after accounting for collective family norms discouraging smoking. Area-level influences such as the social and/or environmental conditions of provinces and communities may be important sources of variation in smoking and therefore need to be considered if rates of smoking are to be modified.

Accounting for context in studies of health inequalities: a review and comparison of analytic approaches

BACKGROUND

A common epidemiologic objective is to evaluate the contribution of residential context to individual-level disparities by race or socioeconomic position.

PURPOSE

We reviewed analytic strategies to account for the total (observed and unobserved factors) contribution of environmental context to health inequalities, including conventional fixed effects (FE) and hybrid FE implemented within a random effects (RE) or a marginal model.

METHODS

To illustrate results and limitations of the various analytic approaches of accounting for the total contextual component of health disparities, we used data on births nested within neighborhoods as an applied example of evaluating neighborhood confounding of racial disparities in gestational age at birth, including both a continuous and a binary outcome.

RESULTS

Ordinary and RE models provided disparity estimates that can be substantially biased in the presence of neighborhood confounding. Both FE and hybrid FE models can account for cluster level confounding and provide disparity estimates unconfounded by neighborhood, with the latter having greater flexibility in allowing estimation of neighborhood-level effects and intercept/slope variability when implemented in a RE specification.

CONCLUSIONS

Given the range of models that can be implemented in a hybrid approach and the frequent goal of accounting for contextual confounding, this approach should be used more often.

Bringing the individual back to small-area variation studies: a multilevel analysis of all-cause mortality in Andalusia, Spain

We performed a multilevel analysis (including individuals, households, census tracts, municipalities and provinces) on a 10% sample (N=230,978) from the Longitudinal Database of the Andalusian Population (LDAP). We aimed to investigate place effects on 8-year individual mortality risk. Moreover, besides calculating association (yielding odds ratios, ORs) between area socio-economic circumstances and individual risk, we wanted to estimate variance and clustering using the variance partition coefficient (VPC). We explicitly proclaim the relevance of considering general contextual effects (i.e. the degree to which the context, as a whole, affects individual variance in mortality risk) under at least two circumstances. The first of these concerns the interpretation of specific contextual effects (i.e. the association between a particular area characteristic and individual risk) obtained from multilevel regression analyses. The second involves the interpretation of geographical variance obtained from classic ecological spatial analyses. The so-called "ecological fallacy" apart, the lack of individual-level information renders geographical variance unrelated to the total individual variation and, therefore, difficult to interpret. Finally, we stress the importance of considering the familial household in multilevel analyses. We observed an association between percentage of people with a low educational level in the census tract and individual mortality risk (OR, highest v. lowest quintile=1.14; 95% confidence interval, CI 1.08-1.20). However, only a minor proportion of the total individual variance in the probability of dying was at the municipality (M) and census tract (CT) levels (VPC(M)=0.2% and VPC(CT)=0.3%). Conversely, the household (H) level appeared much more relevant (VPC(H)=18.6%) than the administrative geographical areas. Without considering general contextual effects, both multilevel analyses of specific contextual effects and ecological studies of small-area variation may provide a misleading picture that overstates the role of administrative areas as contextual determinants of individual differences in mortality.

Shared decision making in prostate-specific antigen testing: the effect of a mailed patient flyer prior to an annual exam

INTRODUCTION AND AIMS

Professional societies recommend that the decision to screen for prostate cancer involves a shared discussion between patient and provider. Many men are tested without this discussion. Prostate cancer screening decision aids increase patient knowledge and participation in prostate-specific antigen (PSA) testing decisions under ideal circumstances but are often resource intensive and elaborate. There is a need for evaluation of interventions that are low cost, low literacy, and practical for widespread distribution. The authors evaluated the effect of a mailed low-literacy informational patient flyer about the PSA test on measures of shared decision making.

METHODS

A pragmatic randomized controlled trial comparing the mailed flyer versus usual care was conducted among 303 men aged 50 to 74 years who were scheduled for annual health maintenance exams in 2 general internal medicine clinics (University of Colorado and University of Colorado Hospital). Charts were reviewed after the visits for documentation of PSA screening discussions and PSA testing rates. Follow-up patient surveys assessed include perceived participation in PSA screening decisions, knowledge of the PSA test, and flyer acceptability.

RESULTS

Rates of chart-documented PSA discussions were low with no difference between the flyer and control groups (17.7% and 13.6%, respectively; P = .28). Rates of PSA testing were also similar in both groups (62.5% vs 58.5%; P = .48). Rates of patient-reported PSA discussions were higher than the documented rates but also without differences between the groups (71.8% vs 62.3%; P = .22). The intervention had no effect in the PSA knowledge scores (3.5/5 vs 3.3/5, P = .60). Patients found the flyer to be highly acceptable.

CONCLUSIONS

A mailed low-literacy informational flyer was well received by patients but had no effect on rates of PSA discussions, PSA testing, or patient knowledge of prostate cancer screening.

Project-based Housing First for chronically homeless individuals with alcohol problems: within-subjects analyses of 2-year alcohol trajectories

OBJECTIVES

Two-year alcohol use trajectories were documented among residents in a project-based Housing First program. Project-based Housing First provides immediate, low-barrier, nonabstinence-based, permanent supportive housing to chronically homeless individuals within a single housing project. The study aim was to address concerns that nonabstinence-based housing may enable alcohol use.

METHODS

A 2-year, within-subjects analysis was conducted among 95 chronically homeless individuals with alcohol problems who were allocated to project-based Housing First. Alcohol variables were assessed through self-report. Data on intervention exposure were extracted from agency records.

RESULTS

Multilevel growth models indicated significant within-subjects decreases across alcohol use outcomes over the study period. Intervention exposure, represented by months spent in housing, consistently predicted additional decreases in alcohol use outcomes.

CONCLUSIONS

Findings did not support the enabling hypothesis. Although the project-based Housing First program did not require abstinence or treatment attendance, participants decreased their alcohol use and alcohol-related problems as a function of time and intervention exposure.

Expanding the scope of risk assessment: methods of studying differential vulnerability and susceptibility

Several methodological issues have been identified in analysis of epidemiological data to better assess the distributional effects of exposures and hypotheses about effect modification. We discuss the hierarchical mixed model and some more complex methods. Methods of capturing inequality are a second dimension of risk assessment, and simulation studies are important because plausible choices for air pollution effects and effect modifiers could result in extremely high risks in a small subset of the population. Future epidemiological studies should explore contextual and individual-level factors that might modify these relationships. The Environmental Protection Agency should make this a standard part of their risk assessments whenever the necessary information is available.

The effect of dialysis chains on mortality among patients receiving hemodialysis

OBJECTIVE

To examine the association between dialysis facility chain affiliation and patient mortality.

STUDY SETTING

Medicare dialysis population.

STUDY DESIGN

Data from the United States Renal Data System (USRDS) were used to identify 3,601 free-standing dialysis facilities and 34,914 Medicare patients' incidence to end-stage renal disease (ESRD) in 2004. Mixed-effect regression models were used to estimate patient mortality by dialysis facility chain and profit status during the 2-year follow-up.

DATA COLLECTION

USRDS data were matched with facility, cost, and census data.

PRINCIPAL FINDINGS

Of the five largest dialysis chains, the lowest mortality risk was observed among patients dialyzed at nonprofit (NP) Chain 5 facilities. Compared with Chain 5, hazard ratios were 19 percent higher (95 percent CI 1.06-1.34) and 24 percent higher (95 percent CI 1.10-1.40) for patients dialyzed at for-profit (FP) Chain 1 and Chain 2 facilities, respectively. In addition, patients at FP facilities had a 13 percent higher risk of mortality than those in NP facilities (95 percent CI 1.06-1.22).

CONCLUSIONS

Large chain affiliation is an independent risk factor for ESRD mortality in the United States. Given the movement toward further consolidation of large FP chains, reasons behind the increase in mortality require scrutiny.

Neighborhood socioeconomic status, depression, and health status in the Look AHEAD (Action for Health in Diabetes) study

Background

Depression and diminished health status are common in adults with diabetes, but few studies have investigated associations with socio-economic environment. The objective of this manuscript was to evaluate the relationship between neighborhood-level SES and health status and depression.

Methods

Individual-level data on 1010 participants at baseline in Look AHEAD (Action for Health in Diabetes), a trial of long-term weight loss among adults with type 2 diabetes, were linked to neighborhood-level SES (% living below poverty) from the 2000 US Census (tracts). Dependent variables included depression (Beck Inventory), and health status (Medical Outcomes Study (SF-36) scale). Multi-level regression models were used to account simultaneously for individual-level age, sex, race, education, personal yearly income and neighborhood-level SES.

Results

Overall, the % living in poverty in the participants' neighborhoods varied, mean = 11% (range 0-67%). Compared to their counterparts in the lowest tertile of neighborhood poverty (least poverty), those in the highest tertile (most poverty) had significantly lower scores on the role-limitations(physical), role limitations(emotional), physical functioning, social functioning, mental health, and vitality sub-scales of the SF-36 scale. When evaluating SF-36 composite scores, those living in neighborhoods with more poverty had significantly lower scores on the physical health (β-coefficient [β] = -1.90 units, 95% CI: -3.40,-0.039), mental health (β = -2.92 units, -4.31,-1.53) and global health (β = -2.77 units, -4.21,-1.33) composite scores.

Conclusion

In this selected group of weight loss trial participants, lower neighborhood SES was significantly associated with poorer health status. Whether these associations might influence response to the Look AHEAD weight loss intervention requires further investigation.

The Happy Life Club™ study protocol: a cluster randomised controlled trial of a type 2 diabetes health coach intervention

BACKGROUND

The Happy Life Club™ is an intervention that utilises health coaches trained in behavioural change and motivational interviewing techniques to assist with the management of type 2 diabetes mellitus (T2DM) in primary care settings in China. Health coaches will support participants to improve modifiable risk factors and adhere to effective self-management treatments associated with T2DM.

METHODS/DESIGN

A cluster randomised controlled trial involving 22 Community Health Centres (CHCs) in Fengtai District of Beijing, China. CHCs will be randomised into a control or intervention group, facilitating recruitment of at least 1320 individual participants with T2DM into the study. Participants in the intervention group will receive a combination of both telephone and face-to-face health coaching over 18 months, in addition to usual care received by the control group. Health coaching will be performed by CHC doctors and nurses certified in coach-assisted chronic disease management. Outcomes will be assessed at baseline and again at 6, 12 and 18 months by means of a clinical health check and self-administered questionnaire. The primary outcome measure is HbA1c level. Secondary outcomes include metabolic, physiological and psychological variables.

DISCUSSION

This cluster RCT has been developed to suit the Chinese health care system and will contribute to the evidence base for the management of patients with T2DM. With a strong focus on self-management and health coach support, the study has the potential to be adapted to other chronic diseases, as well as other regions of China.

TRIAL REGISTRATION

Current Controlled Trials ISRCTN01010526.

Nutritional status of children in India: household socio-economic condition as the contextual determinant

BACKGROUND

Despite recent achievement in economic progress in India, the fruit of development has failed to secure a better nutritional status among all children of the country. Growing evidence suggest there exists a socio-economic gradient of childhood malnutrition in India. The present paper is an attempt to measure the extent of socio-economic inequality in chronic childhood malnutrition across major states of India and to realize the role of household socio-economic status (SES) as the contextual determinant of nutritional status of children.

METHODS

Using National Family Health Survey-3 data, an attempt is made to estimate socio-economic inequality in childhood stunting at the state level through Concentration Index (CI). Multi-level models; random-coefficient and random-slope are employed to study the impact of SES on long-term nutritional status among children, keeping in view the hierarchical nature of data.

MAIN FINDINGS

Across the states, a disproportionate burden of stunting is observed among the children from poor SES, more so in urban areas. The state having lower prevalence of chronic childhood malnutrition shows much higher burden among the poor. Though a negative correlation (r = -0.603, p < .001) is established between Net State Domestic Product (NSDP) and CI values for stunting; the development indicator is not always linearly correlated with intra-state inequality in malnutrition prevalence. Results from multi-level models however show children from highest SES quintile posses 50 percent better nutritional status than those from the poorest quintile.

CONCLUSION

In spite of the declining trend of chronic childhood malnutrition in India, the concerns remain for its disproportionate burden on the poor. The socio-economic gradient of long-term nutritional status among children needs special focus, more so in the states where chronic malnutrition among children apparently demonstrates a lower prevalence. The paper calls for state specific policies which are designed and implemented on a priority basis, keeping in view the nature of inequality in childhood malnutrition in the country and its differential characteristics across the states.

Neighborhood and weight-related health behaviors in the Look AHEAD (Action for Health in Diabetes) study

Background

Previous studies have shown that neighborhood factors are associated with obesity, but few studies have evaluated the association with weight control behaviors. This study aims to conduct a multi-level analysis to examine the relationship between neighborhood SES and weight-related health behaviors.

Methods

In this ancillary study to Look AHEAD (Action for Health in Diabetes) a trial of long-term weight loss among individuals with type 2 diabetes, individual-level data on 1219 participants from 4 clinic sites at baseline were linked to neighborhood-level data at the tract level from the 2000 US Census and other databases. Neighborhood variables included SES (% living below the federal poverty level) and the availability of food stores, convenience stores, and restaurants. Dependent variables included BMI, eating patterns, weight control behaviors and resource use related to food and physical activity. Multi-level models were used to account for individual-level SES and potential confounders.

Results

The availability of restaurants was related to several eating and weight control behaviors. Compared to their counterparts in neighborhoods with fewer restaurants, participants in neighborhoods with more restaurants were more likely to eat breakfast (prevalence Ratio [PR] 1.29 95% CI: 1.01-1.62) and lunch (PR = 1.19, 1.04-1.36) at non-fast food restaurants. They were less likely to be attempting weight loss (OR = 0.93, 0.89-0.97) but more likely to engage in weight control behaviors for food and physical activity, respectively, than those who lived in neighborhoods with fewer restaurants. In contrast, neighborhood SES had little association with weight control behaviors.

Conclusion

In this selected group of weight loss trial participants, restaurant availability was associated with some weight control practices, but neighborhood SES was not. Future studies should give attention to other populations and to evaluating various aspects of the physical and social environment with weight control practices.

Changes in the health burden of a national sample of children with asthma

This paper focused on the extent to which factors that are modifiable by health policies or provider recommendations influenced the level and changes in the burden of childhood asthma. Demographic factors, access to health care services, and asthma control activities were posited to potentially influence the level and changes in health burden of children with asthma. The Medical Expenditure Panel Survey data from 1996-1999 on 3-11 year old U.S. children with asthma (N=784) were used. The findings of multilevel models of perceived burden indicated unfavorable trajectories among those families who had public health insurance. Asthma control activities were associated with favorable trajectories of both perceived and objectively measured burden. These findings emphasized the significance of asthma control and access to high quality and stable health care services as health policy targets.

Predictors of mortality in HIV-associated hospitalizations in Portugal: a hierarchical survival model

BACKGROUND

The beneficial effects of highly active antiretroviral therapy, increasing survival and the prevention of AIDS defining illness development are well established. However, the annual Portuguese hospital mortality is still higher than expected. It is crucial to understand the hospitalization behaviour to better allocate resources. This study investigates the predictors of mortality in HIV associated hospitalizations in Portugal through a hierarchical survival model.

METHODS

The study population consists of 12,078 adult discharges from patients with HIV infection diagnosis attended at Portuguese hospitals from 2005-2007 that were registered on the diagnosis-related groups' database.We used discharge and hospital level variables to develop a hierarchical model. The discharge level variables were: age, gender, type of admission, type of diagnoses-related group, related HIV complication, the region of the patient's residence, the number of diagnoses and procedures, the Euclidean distance from hospital to the centroid of the patient's ward, and if patient lived in the hospital's catchment area. The hospital characteristics include size and hospital classification according to the National Health System. Kaplan-Meier plots were used to examine differences in survival curves. Cox proportional hazard models with frailty were applied to identify independent predictors of hospital mortality and to calculate hazard ratios (HR).

RESULTS

The Cox proportional model with frailty showed that male gender, older patient, great number of diagnoses and pneumonia increased the hazard of HIV related hospital mortality. On the other hand tuberculosis was associated with a reduced risk of death. Central hospital discharge also presents less risk of mortality.The frailty variance was small but statistically significant, indicating hazard ratio heterogeneity among hospitals that varied between 0.67 and 1.34, and resulted in two hospitals with HR different from the average risk.

CONCLUSION

The frailty model suggests that there are unmeasured factors affecting mortality in HIV associated hospitalizations. Consequently, for healthcare policy purposes, hospitals should not all be treated in an equal manner.

A review of HIV/AIDS system-level interventions

The escalating HIV/AIDS epidemic worldwide demands that on-going prevention efforts be strengthened, disseminated, and scaled-up. System-level interventions refer to programs aiming to improve the functioning of an agency as well as the delivery of its services to the community. System-level interventions are a promising approach to HIV/AIDS prevention because they focus on (a) improving the agency's ability to adopt evidence-based HIV prevention and care programs; (b) develop and establish policies and procedures that maximize the sustainability of on-going prevention and care efforts; and (c) improve decision-making processes such as incorporating the needs of communities into their tailored services. We reviewed studies focusing on system-level interventions by searching multiple electronic abstracting indices, including PsycInfo, PubMed, and ProQuest. Twenty-three studies out of 624 peer-reviewed studies (published from January 1985 to February 2007) met study criteria. Most of the studies focused on strengthening agency infrastructure, while other studies included collaborative partnerships and technical assistance programs. Our findings suggest that system-level interventions are promising in strengthening HIV/AIDS prevention and treatment efforts. Based on our findings, we propose recommendations for future work in developing and evaluating system-level interventions.

Effect of utilization policies for fluoroquinolones: a pilot study in nova scotia hospitals

BACKGROUND

Antimicrobial resistance results in increased morbidity, mortality, and costs to the health care system. Evidence suggests an association between the use of antimicrobials in hospitals and the development of antimicrobial resistance. Fluoroquinolones constitute one group of antimicrobials that are effective against a variety of bacterial infections, yet they may be subject to misuse. Many hospitals in Nova Scotia have implemented policies to improve antimicrobial prescribing, but the impact of these policies on utilization is unknown.

OBJECTIVES

To evaluate the use of fluoroquinolones in Nova Scotia hospitals using the World Health Organization's Anatomical Therapeutic Chemical classification system with defined daily doses (ATC/DDD) and to examine the influence of hospital policies for utilization of fluoroquinolones in community-acquired pneumonia.

METHODS

During the study period (April 1, 1997, to March 31, 2003), fluoroquinolones were administered at 31 of the 37 hospitals in Nova Scotia's 9 district health authorities. Hospital administrative data, hospital characteristics, and pharmaceutical purchasing data related to use of these drugs were aggregated using the ATC/DDD methodology for the fiscal years 1997/1998 to 2002/2003. District pharmacy directors were surveyed to obtain information about district and individual hospital antibiotic policies. Descriptive statistics were calculated, and univariable regression and multilevel analyses were performed.

RESULTS

Mean overall fluoroquinolone use increased over the study period, from 47.2 DDD/1000 bed-days per year in fiscal year 1997/1998 to 163.8 DDD/1000 bed-days per year in fiscal year 2002/2003 (p < 0.001). Multilevel analysis showed that utilization policies aimed at appropriate prescribing did not affect the use of fluoroquinolones.

CONCLUSION

This study revealed that drug purchasing, hospital administrative, and diagnostic data could be combined to compare the utilization of fluoroquinolones among different hospitals and district health authorities. Utilization policies had little effect on the amount, type, or route of fluoroquinolone use.

Neighborhood effects on birthweight: an exploration of psychosocial and behavioral pathways in Baltimore, 1995--1996

Neighborhood characteristics have been proposed to influence birth outcomes through psychosocial and behavioral pathways, yet empirical evidence is lacking. Using data from an urban, low-income sample, this study examined the impact of the neighborhood environment on birthweight and evaluated mediation by psychosocial and behavioral factors. The sample included 726 women who delivered a live birth at Johns Hopkins Hospital in Baltimore, Maryland, USA between 1995 and 1996. Census-tract data were used to create a principal component index of neighborhood risk based on racial and economic stratification (% Black, % poverty), social disorder (violent crime rate), and physical deterioration (% boarded-up housing) (alpha=0.82). Information on sociodemographic, psychosocial, and behavioral factors was gathered from a postpartum interview and medical records. Random intercept multilevel models were used to estimate neighborhood effects and assess potential mediation. Controlling for sociodemographic characteristics, a standard deviation increase in neighborhood risk conferred a 76g birthweight decrement. This represents an approximate 300g difference between the best and worst neighborhoods. Although stress (daily hassles), perceived locus-of-control, and social support were related to birthweight, their adjustment reduced the neighborhood coefficient by only 12%. In contrast, the neighborhood effect was reduced by an additional 30% and was no longer statistically significant after adjustment for the behavioral factors of smoking, drug use, and delayed prenatal care. These findings suggest that neighborhood factors may influence birthweight by shaping maternal behavioral risks. Thus, neighborhood level interventions should be considered to address multiple maternal and infant health risks. Future studies should examine more direct measures of neighborhood stress, such as perceived neighborhood disorder, and evaluate alternative mechanisms by which neighborhood factors influence behavior (e.g., social norms and access to goods and services).

A new tool for epidemiology: the usefulness of dynamic-agent models in understanding place effects on health

A major focus of recent work on the spatial patterning of health has been the study of how features of residential environments or neighborhoods may affect health. Place effects on health emerge from complex interdependent processes in which individuals interact with each other and their environment and in which both individuals and environments adapt and change over time. Traditional epidemiologic study designs and statistical regression approaches are unable to examine these dynamic processes. These limitations have constrained the types of questions asked, the answers received, and the hypotheses and theoretical explanations that are developed. Agent-based models and other systems-dynamics models may help to address some of these challenges. Agent-based models are computer representations of systems consisting of heterogeneous microentities that can interact and change/adapt over time in response to other agents and features of the environment. Using these models, one can observe how macroscale dynamics emerge from microscale interactions and adaptations. A number of challenges and limitations exist for agent-based modeling. Nevertheless, use of these dynamic models may complement traditional epidemiologic analyses and yield additional insights into the processes involved and the interventions that may be most useful.

Association of birth outcomes and maternal, school, and neighborhood characteristics with subsequent numeracy achievement

This study investigated the relation between birth characteristics and numeracy attainment at age 8 years. Using a multilevel approach, the authors analyzed all non-Aboriginal singletons born in Western Australia who attended government schools and participated in a Western Australia-wide numeracy test in grade 3 between 1999 and 2005. Appropriateness of intrauterine growth was expressed as the proportion of optimal growth parameters for gestational duration, infant sex, and maternal height and parity, which was derived from a total population of births without risk factors for growth restriction. After the authors controlled for sociodemographic factors, term birth and proportion of optimal head circumference at birth were associated with higher numeracy scores. Increasing proportion of optimal birth length and being firstborn were associated with relatively higher numeracy scores among children born to mothers residing in the most educationally deprived area. The relative advantage of being born first was also higher for children born to single mothers. In contrast, higher Apgar scores and greater proportion of optimal birth weight were associated with a lower relative advantage for children born to single mothers. In summary, term birth and increased growth in head circumference and length are key birth characteristics associated with higher numeracy scores, especially among disadvantaged children.

Toward the next generation of research into small area effects on health: a synthesis of multilevel investigations published since July 1998

To map out area effects on health research, this study had the following aims: (1) to inventory multilevel investigations of area effects on self rated health, cardiovascular diseases and risk factors, and mortality among adults; (2) to describe and critically discuss methodological approaches employed and results observed; and (3) to formulate selected recommendations for advancing the study of area effects on health. Overall, 86 studies were inventoried. Although several innovative methodological approaches and analytical designs were found, small areas are most often operationalised using administrative and statistical spatial units. Most studies used indicators of area socioeconomic status derived from censuses, and few provided information on the validity and reliability of measures of exposures. A consistent finding was that a significant portion of the variation in health is associated with area context independently of individual characteristics. Area effects on health, although significant in most studies, often depend on the health outcome studied, the measure of area exposure used, and the spatial scale at which associations are examined.

A simulation study of sample size for multilevel logistic regression models

Background

Many studies conducted in health and social sciences collect individual level data as outcome measures. Usually, such data have a hierarchical structure, with patients clustered within physicians, and physicians clustered within practices. Large survey data, including national surveys, have a hierarchical or clustered structure; respondents are naturally clustered in geographical units (e.g., health regions) and may be grouped into smaller units. Outcomes of interest in many fields not only reflect continuous measures, but also binary outcomes such as depression, presence or absence of a disease, and self-reported general health. In the framework of multilevel studies an important problem is calculating an adequate sample size that generates unbiased and accurate estimates.

Methods

In this paper simulation studies are used to assess the effect of varying sample size at both the individual and group level on the accuracy of the estimates of the parameters and variance components of multilevel logistic regression models. In addition, the influence of prevalence of the outcome and the intra-class correlation coefficient (ICC) is examined.

Results

The results show that the estimates of the fixed effect parameters are unbiased for 100 groups with group size of 50 or higher. The estimates of the variance covariance components are slightly biased even with 100 groups and group size of 50. The biases for both fixed and random effects are severe for group size of 5. The standard errors for fixed effect parameters are unbiased while for variance covariance components are underestimated. Results suggest that low prevalent events require larger sample sizes with at least a minimum of 100 groups and 50 individuals per group.

Conclusion

We recommend using a minimum group size of 50 with at least 50 groups to produce valid estimates for multi-level logistic regression models. Group size should be adjusted under conditions where the prevalence of events is low such that the expected number of events in each group should be greater than one.

Use of hierarchical models to evaluate performance of cardiac surgery centres in the Italian CABG outcome study

Background

Hierarchical modelling represents a statistical method used to analyze nested data, as those concerning patients afferent to different hospitals. Aim of this paper is to build a hierarchical regression model using data from the "Italian CABG outcome study" in order to evaluate the amount of differences in adjusted mortality rates attributable to differences between centres.

Methods

The study population consists of all adult patients undergoing an isolated CABG between 2002–2004 in the 64 participating cardiac surgery centres.

A risk adjustment model was developed using a classical single-level regression. In the multilevel approach, the variable "clinical-centre" was employed as a group-level identifier. The intraclass correlation coefficient was used to estimate the proportion of variability in mortality between groups. Group-level residuals were adopted to evaluate the effect of clinical centre on mortality and to compare hospitals performance. Spearman correlation coefficient of ranks (ρ) was used to compare results from classical and hierarchical model.

Results

The study population was made of 34,310 subjects (mortality rate = 2.61%; range 0.33–7.63). The multilevel model estimated that 10.1% of total variability in mortality was explained by differences between centres. The analysis of group-level residuals highlighted 3 centres (VS 8 in the classical methodology) with estimated mortality rates lower than the mean and 11 centres (VS 7) with rates significantly higher. Results from the two methodologies were comparable (ρ = 0.99).

Conclusion

Despite known individual risk-factors were accounted for in the single-level model, the high variability explained by the variable "clinical-centre" states its importance in predicting 30-day mortality after CABG.

Neighborhoods and health: where are we and were do we go from here?

In recent years there has been an explosion of interest in neighborhood health effects. Most existing work has relied on secondary data analyses and has used administrative areas and aggregate census data to characterize neighborhoods. Important questions remain regarding whether the associations reported by these studies reflect causal processes. This paper reviews the major limitations of existing work and discusses areas for future development including: (1) definition and measurement of area or ecologic attributes; (2) consideration of spatial scale; (3) cumulative exposures and lagged effects; (4) the complementary nature of observational, quasi-experimental, and experimental evidence. As is usually the case with complex research questions, consensus regarding the presence and magnitude of neighborhood health effects will emerge from the work of multiple disciplines, often with diverse methodological approaches, each with its strengths and its limitations. Partnership across disciplines, as well as among health researchers, communities, urban planners, and policy experts will be key.

Neighborhood socioeconomic deprivation and residential instability: effects on incidence of ischemic heart disease and survival after myocardial infarction

BACKGROUND

Previous literature has shown that neighborhood socioeconomic position influences the risk of ischemic heart disease, but little is known about the mechanisms linking the residential context to ischemic heart disease incidence and mortality. We examined whether neighborhood socioeconomic position and neighborhood residential stability (as a determinant of social interaction patterns) have an influence on ischemic heart disease risk. Moreover, we investigated whether dissimilar contextual influences operate at different stages of the disease process, ie, on incidence, 1-day case-fatality, and long-term survival after acute myocardial infarction (MI).

METHODS

Using a large 27-year longitudinal cohort (baseline: 1 January 1996) defined in the Scania region, Sweden, we estimated multilevel survival models adjusted for individual sociodemographic factors and previous diseases of the persons.

RESULTS

After adjustment, multilevel survival models indicated that the incidence of ischemic heart disease increased with neighborhood socioeconomic deprivation but was only weakly associated with neighborhood residential instability (for high vs low residential instability, hazard ratio = 1.2; 95% credible interval = 1.0-1.4). Conversely, beyond effects of individual and contextual socioeconomic circumstances and distance to the hospital, we saw a markedly higher 1-day case-fatality (4.9; 1.8-15) and shorter survival time after MI among individuals still alive 28 days after MI (4.3; 1.2-17) in neighborhoods with a high versus low residential instability.

CONCLUSIONS

Effects of residential instability on post-MI survival may be mediated by the lower availability of social support in residentially unstable neighborhoods, suggesting a new class of intermediate processes that should be taken into account when investigating contextual influences on ischemic heart disease. Moreover, dissimilar contextual effects may operate at various stages of the disease process (ie, on incidence, case-fatality, and survival after MI).

Dynamic wait-listed designs for randomized trials: new designs for prevention of youth suicide

BACKGROUND

The traditional wait-listed design, where half are randomly assigned to receive the intervention early and half are randomly assigned to receive it later, is often acceptable to communities who would not be comfortable with a no-treatment group. As such this traditional wait-listed design provides an excellent opportunity to evaluate short-term impact of an intervention. We introduce a new class of wait-listed designs for conducting randomized experiments where all subjects receive the intervention, and the timing of the intervention is randomly assigned. We use the term "dynamic wait-listed designs" to describe this new class.

PURPOSE

This paper examines a new class of statistical designs where random assignment to intervention condition occurs at multiple times in a trial. As an extension of a traditional wait-listed design, this dynamic design allows all subjects to receive the intervention at a random time. Motivated by our search for increased statistical power in an ongoing school-based trial that is testing a program of gatekeeper training to identify suicidal youth and refer them to treatment, this new design class is especially useful when the primary outcome is a count or rate of occurrence, such as suicidal behavior, whose rate can fluctuate over time due to uncontrolled factors.

METHODS

Statistical power is computed for various dynamic wait-listed designs under conditions where the underlying rate of occurrence is allowed to vary nonsystematically. We also present as an example a large ongoing trial to evaluate a gatekeeper training suicide prevention program in 32 schools which we initially began as a classic randomized wait-listed design. The primary outcome of interest in this study is the count of the number of children who are identified by the school system as having suicidal thoughts or behaviors who are then validated as being suicidal by mental health professionals in the community.

RESULTS

A general result shows that dynamic wait-listed designs always have higher statistical power over a traditional wait-listed design. This power increase can be substantial. Efficiency gains of 33% are easy to obtain for situations where the intervention has a small effect and the variation in rate across time is quite high. When the rate variation for an outcome is very low or the intervention effect is large, efficiency gains approach 100%. A small increase in the number of times where random assignment occurs from 2 - for the standard wait-listed design, to say 4 can provide a large reduction in variance. Efficiency gains can also be high when converting standard wait-listed design to a dynamic one half-way into the study.

LIMITATIONS

As with all wait-listed designs, dynamic wait-listed designs can only be used to evaluate short-term impact. Since all subjects eventually receive the intervention, no comparison can be made after the end of the random assignment period. The statistical power benefits are primarily limited to outcomes that can be treated as count or time to event data.

CONCLUSIONS

A dynamic design randomly assigns units - either individuals or groups - to start the intervention at varying times during the course of the study. This design is useful in testing interventions that screen for new or existing cases, as well as testing the scalability of interventions as they are disseminated or expanded system wide. They can improve on the traditional wait-listed design both in terms of statistical power and robustness in the presence of exogenous factors. This paper demonstrates that such designs yield smaller standard errors and can achieve higher statistical power than that of a standard wait-listed design. Just as important, dynamic designs can also help reduce the logistical challenges of implementing an intervention on a wide scale. When the intervention requires that significant training resources be allocated throughout the study, the dynamic wait-listed design is likely to increase the rate of training and lead to a higher level of program implementation.

Urban neighborhoods, chronic stress, gender and depression

Using multilevel analysis we find that residents of "stressed" neighborhoods have higher levels of depression than residents of less "stressed" neighborhoods. Data for individuals are from two cycles of the Canadian Community Health Survey, a national probability sample of 56,428 adults living in 25 Census Metropolitan Areas in Canada, with linked information about the respondents' census tracts. Depression is measured with the Center for Epidemiologic Studies-Depression Scale Short Form and is based on a cutoff of 4+ symptoms. Factor analysis of census tract characteristics identified two measures of neighborhood chronic stress--residential mobility and material deprivation--and two measures of population structure--ethnic diversity and dependency. After adjustment for individual-level gender, age, education, marital and visible minority status and neighborhood-level ethnic diversity and dependency, a significant contextual effect of neighborhood chronic stress survives. As such, the daily stress of living in a neighborhood where residential mobility and material deprivation prevail is associated with depression. Since gender frames access to personal and social resources, we explored the possibility that women might be more reactive to chronic stressors manifested in higher risk of depression. However, we did not find random variation in depression by gender across neighborhoods.

Effect of area poverty rate on cancer screening across US communities

STUDY OBJECTIVE

To analyse the contextual effect of area poverty rate on never having been screened for breast, cervical, and colorectal cancer by (1) describing the extent of the variation in screening behaviours among 98 US metropolitan areas; (2) determining if the variation in lack of screening can be explained by differences in the characteristics of the persons who resided in these areas; and (3) determining if living in a metropolitan area with a higher poverty rate increased the likelihood of never having been screened for cancer over and above individual characteristics.

DESIGN

Cross sectional survey using data from the 2002 Behavioral Risk Factor Surveillance System. Multilevel logistic regression included both individual level factors as well as area poverty rate.

SETTING

Ninety eight areas across the USA.

PARTICIPANTS

Over 118 000 persons residing in 98 areas; a sample aimed at estimating 48.3% of the US population age 18 or older.

MAIN RESULTS

After adjustment for individual level factors, increasing area level poverty rate (per 5%) remained associated with never having had a mammogram (odds ratio (OR) = 1.28, 95% confidence interval (CI): 1.03 to 1.37); clinical breast examination (OR = 1.28, 95% CI: 1.11 to 1.48), colonoscopy/sigmoidoscopy (OR = 1.10, 95% CI: 1.01 to 1.19), and a faecal occult blood test (OR = 1.19, 95% CI: 1.12 to 1.27). Poverty rate was not independently associated with never having had a Pap smear (OR = 1.12; 95% CI: 0.90 to 1.41). The size of the variance among metropolitan or micropolitan statistical areas (MMSAs) varied by type of screening test, with intraclass correlation coefficients ranging from 4.9% (never having had a Pap smear) to 1.2% (never having had a colonoscopy/sigmoidoscopy).

CONCLUSIONS

Area poverty rate was independently associated with never having been screened for breast and colorectal cancer, but not cervical cancer. The size of the variance among MMSAs was modest at best.

Levels of analysis for the study of environmental health disparities

Reducing racial/ethnic and socioeconomic environmental health disparities requires a comprehensive multilevel conceptual and quantitative approach that recognizes the various levels through which environmental health disparities are produced and perpetuated. We propose a conceptual framework that incorporates the micro level, contained within the local level, which in turn is contained within the macro level. We discuss the utility of multilevel techniques to examine environmental level (both physical and social) and individual-level factors to appropriately quantify and improve our understanding of environmental health disparities. We discuss the reasoning and the methodological approach behind multilevel modeling, including differentiating between individual and contextual influences on individual outcomes. Next we address the questions and principles that guide the choice of levels or geographic units in multilevel studies. Finally, we address the ways in which different data sources can be combined to produce suitable data for multilevel analyses. We provide some examples of how such data sources can be linked to create multilevel data structures, and offer suggestions to facilitate the integration of multilevel techniques in environmental health disparities research and monitoring.

Comparative ageing research: a flourishing field in need of theoretical cultivation

Comparative ageing research is a flourishing field, partly because of European funding in this area. Comparing different societies and cultures seems especially fruitful for the analysis of societal and cultural factors in development over the life course. From a nomothetic perspective, the aim of comparisons is the search for similarities and communalities in different societies and cultures; from an idiographic perspective, researchers are looking for societal and cultural specificity and distinctiveness. However, the potentials of comparative ageing research are not fully realized for the time being. In many cases, there is little theorizing as to whether there should be differences (or similarities) in ageing processes across countries, societies, or cultures. This paper discusses theoretical aims and ambitions of comparative ageing research in general. Comparative theories are sketched which could serve as a basis for comparative ageing research, and ageing theories are discussed which could be modified to be used in comparative research. The rationale of comparative ageing research is described and illustrated through empirical examples. Epistemological and methodological pitfalls (problems of conceptual, operational, functional, and measurement equivalence) are a substantial obstacle to comparative ageing research. Hence, merits and limitations of comparative designs and sampling procedures are considered.

Multilevel modeling was a convenient alternative to common regression designs in longitudinal suicide research

BACKGROUND AND OBJECTIVE

The prospective investigation of repetitive nonfatal suicidal behavior is associated with two methodological problems. Due to the commonly used definitions of nonfatal suicidal behavior, clinical samples usually consist of patients with a considerable between-person variability. Second, repeated nonfatal suicidal episodes of the same subjects are likely to be correlated. We examined three regression techniques to comparatively evaluate their efficiency in addressing the given methodological problems.

STUDY DESIGN AND SETTING

Repeated episodes of nonfatal suicidal behavior were assessed in two independent patient samples during a 2-year follow-up period. The first regression design modeled repetitive nonfatal suicidal behavior as a summary measure. The second regression model treated repeated episodes of the same subject as independent events. The third regression model represented a hierarchical linear model.

RESULTS

The estimated mean effects of the first model were likely to be nonrepresentative for a considerable part of the study subjects. The second regression design overemphasized the impact of the predictor variables. The hierarchical linear model most appropriately accounted for the heterogeneity of the samples and the correlated data structure.

CONCLUSION

The nonhierarchical regression designs did not provide appropriate statistical models for the prospective investigation of repetitive nonfatal suicidal behavior. Multilevel modeling provides a convenient alternative.

Social capital and health: does egalitarianism matter? A literature review

The aim of the paper is to critically review the notion of social capital and review empirical literature on the association between social capital and health across countries. The methodology used for the review includes a systematic search on electronic databases for peer-reviewed published literature. We categorize studies according to level of analysis (single and multilevel) and examine whether studies reveal a significant health impact of individual and area level social capital. We compare the study conclusions according to the country's degrees of economic egalitarianism. Regardless of study design, our findings indicate that a positive association (fixed effect) exists between social capital and better health irrespective of countries degree of egalitarianism. However, we find that the between-area variance (random effect) in health tends to be lower in more egalitarian countries than in less egalitarian countries. Our tentative conclusion is that an association between social capital and health at the individual level is robust with respect to the degree of egalitarianism within a country. Area level or contextual social capital may be less salient in egalitarian countries in explaining health differences across places.

Defining urban and rural areas in U.S. epidemiologic studies

Among epidemiologists, there has been increasing interest in the characteristics of communities that influence health. In the United States, the rural health disparity has been a recent focus of attention and made a priority for improvement. While many standardized definitions of urban and rural exist and are used by social scientists and demographers, they are found in sources unfamiliar to health researchers and have largely not been used in public health studies. This paper briefly reviews some available definitions of urban and rural for American geographic subunits and their respective strengths and weaknesses. For example, some definitions are better suited than others for capturing access to health care services. The authors applied different definitions to breast cancer incidence rates to show how urban/rural rate ratio comparisons would vary by choice of definition and found that dichotomous definitions may fail to capture variability in very rural areas. Further study of the utility of these measures in health studies is warranted.

Anxiolytic–hypnotic drug use associated with trust, social participation, and the miniaturization of community: A multilevel analysis

The concept of social capital has gained wide interest in public health research in recent years. However, we suggest a concept that was introduced and developed by Fukuyama, named "miniaturization of community", as an alternative to that of social capital. The concept of miniaturization of community emphasizes that a high level of social participation can be accompanied by a low level of trust, both at the individual and at the community level, which may in turn result in social disorder and lack of social cohesion. When society becomes more disordered, people may tend to feel more insecure and anxious. Use of anxiolytic-hypnotic drugs (AHDs) could under such circumstances be a coping strategy. In this study, we first wanted to investigate whether the contextual component of the miniaturization of community concept (i.e. area high social participation and low trust) is associated with individual AHD use, over and above individual characteristics. Secondly, we aimed to study whether people living in the same municipality share a similar probability of AHD use, after adjusting for individual characteristics, and if so, how large this contextual phenomenon is. We used data on 20,319 women and 17,850 men aged 18-79 years from 58 municipalities in six regions in central Sweden, who participated in the Life & Health year 2000 postal survey. We applied multilevel logistic regression analysis with individuals at the first level and areas at the second level. Our results suggest that living in an area with a high level of miniaturization of community seems to be associated with individual AHD use, beyond people's individual characteristics including their own level of social participation and trust. The concept of miniaturization of community may be an extension of the classic concept of social capital and may increase our understanding of contextual effects on health.

A brief conceptual tutorial on multilevel analysis in social epidemiology: investigating contextual phenomena in different groups of people

STUDY OBJECTIVE

(1) To provide a didactic and conceptual (rather than mathematical) link between multilevel regression analysis (MLRA) and social epidemiological concepts. (2) To develop an epidemiological vision of MLRA focused on measures of health variation and clustering of individual health status within areas, which is useful to operationalise the notion of "contextual phenomenon". The paper shows how to investigate (1) whether there is clustering within neighbourhoods, (2) to which extent neighbourhood level differences are explained by the individual composition of the neighbourhoods, (3) whether the contextual phenomenon differs in magnitude for different groups of people, and whether neighbourhood context modifies individual level associations, and (4) whether variations in health status are dependent on individual level characteristics.

DESIGN AND PARTICIPANTS

Simulated data are used on systolic blood pressure (SBP), age, body mass index (BMI), and antihypertensive medication (AHM) ascribed to 25 000 subjects in 39 neighbourhoods of an imaginary city. Rather than assessing neighbourhood variables, the paper concentrated on SBP variance between individuals and neighbourhoods as a function of individual BMI.

RESULTS

The variance partition coefficient (VPC) showed that clustering of SBP within neighbourhoods was greater for people with a higher BMI. The composition of the neighbourhoods with respect to age, AHM use, and BMI explained about one fourth of the neighbourhood differences in SBP. Neighbourhood context modified the individual level association between BMI and SBP. Individual level differences in SBP within neighbourhoods were larger for people with a higher BMI.

CONCLUSIONS

Statistical measures of multilevel variations can effectively quantify contextual effects in different groups of people, which is a relevant issue for understanding health inequalities.

Mechanisms by which antibiotics promote dissemination of resistant pneumococci in human populations

Mechanisms by which antimicrobials contribute to dissemination of pneumococcal resistance are incompletely characterized. A serial cross-sectional study of nasopharyngeal pneumococcal carriage in healthy, home-living children <or=6 years of age was conducted in four rural communities-two in Utah (1998-2003) and two in Idaho (2002-2003). Prevalence odds ratios for carriage of resistant pneumococci (OR(res)) and of susceptible pneumococci (OR(sus)) were estimated. Dynamic transmission models were developed to facilitate a mechanistic interpretation of OR(res) and OR(sus) and to compare the population impact of distinct antimicrobial classes. A total of 5,667 cultures were obtained; 25% of the cultures were positive, and 29% of isolates exhibited reduced susceptibility to penicillin. The adjusted OR(res) for recent individual and sibling cephalosporin use was 2.2 (95% confidence interval: 1.4, 3.4) and 1.8 (95% confidence interval: 1.0, 3.3), respectively. Neither individual nor sibling penicillin use was associated with increased OR(res). Rather, recent use of penicillins was associated with decreased carriage of susceptible pneumococci (OR(sus) = 0.2, 95% confidence interval: 0.1, 0.3). In simulations, both types of effects promoted dissemination of resistant pneumococci at the population level. Findings show that oral cephalosporins enhance the risk of acquiring resistant pneumococci. Penicillins accelerate clearance of susceptible strains. The effect of penicillins in increasing resistance is shared equally by treated and untreated members of the population.

A brief conceptual tutorial of multilevel analysis in social epidemiology: linking the statistical concept of clustering to the idea of contextual phenomenon

STUDY OBJECTIVE

This didactical essay is directed to readers disposed to approach multilevel regression analysis (MLRA) in a more conceptual than mathematical way. However, it specifically develops an epidemiological vision on multilevel analysis with particular emphasis on measures of health variation (for example, intraclass correlation). Such measures have been underused in the literature as compared with more traditional measures of association (for example, regression coefficients) in the investigation of contextual determinants of health. A link is provided, which will be comprehensible to epidemiologists, between MLRA and social epidemiological concepts, particularly between the statistical idea of clustering and the concept of contextual phenomenon.

DESIGN AND PARTICIPANTS

The study uses an example based on hypothetical data on systolic blood pressure (SBP) from 25,000 people living in 39 neighbourhoods. As the focus is on the empty MLRA model, the study does not use any independent variable but focuses mainly on SBP variance between people and between neighbourhoods.

RESULTS

The intraclass correlation (ICC = 0.08) informed of an appreciable clustering of individual SBP within the neighbourhoods, showing that 8% of the total individual differences in SBP occurred at the neighbourhood level and might be attributable to contextual neighbourhood factors or to the different composition of neighbourhoods.

CONCLUSIONS

The statistical idea of clustering emerges as appropriate for quantifying "contextual phenomena" that is of central relevance in social epidemiology. Both concepts convey that people from the same neighbourhood are more similar to each other than to people from different neighbourhoods with respect to the health outcome variable.

Neighbourhood characteristics, individual level socioeconomic factors, and depressive symptoms in young adults: the CARDIA study

STUDY OBJECTIVE

To investigate the relation between neighbourhood socioeconomic and ethnic characteristics with depressive symptoms in a population based sample.

DESIGN

Cross sectional data from the CARDIA study, including the Center for Epidemiological Studies depression scale score (CES-D). Neighbourhoods were 1990 US census blocks of 1000 people; six census variables reflecting wealth/income, education, and occupation investigated separately and as a summary score; neighbourhood racial composition (percentage white and black) and individual level income and education were also examined.

SETTING

Participants recruited in 1985/86 from community lists in Birmingham, AL; Chicago, IL; Minneapolis MN; from a health plan in Oakland, CA.

PARTICIPANTS

3437 adults aged 28-40 years in 1995/96: 24% white men, 27% white women, 20% black men, 29% black women.

MAIN RESULTS

For each race-sex group, CES-D was inversely related to neighbourhood score and individual income and education. Associations of neighbourhood score with CES-D became weak and inconsistent after adjusting for individual level factors; personal income remained strongly and inversely associated with CES-D. Age adjusted mean differences (standard errors) in CES-D between the lowest and highest income categories were 3.41 (0.62) for white men, 4.57 (0.64) for white women, 5.80 (0.87) for black men, and 5.74 (0.83) for black women. For both black and white participants, CES-D was associated negatively with percentage of white people and positively with percentage of black people in their census block, before, but not after, adjustment for individual and neighbourhood socioeconomic variables.

CONCLUSIONS

Neither neighbourhood socioeconomic characteristics nor ethnic density were consistently related to depressive symptoms once individual socioeconomic characteristics were taken into account.

The family contribution to health status: a population-level estimate

PURPOSE

Clinical studies have shown strong family influences on individual health, but the aggregate importance of family effects for population health is unknown. Our objective was to estimate, at a population level, the variance in individual health status attributable to the family.

METHODS

Secondary data were used from the Community Tracking Study, a stratified random sample of the US population. Hierarchical linear modeling was used to estimate the individual and family components of health status. The setting was 60 US communities, which account for approximately one half of the population. Participants were US residents aged 18 years and older who shared a household with family members in the study (N = 35,055). Main outcome measures were the Short Form-12 (SF-12) self-reported physical and mental subscales.

RESULTS

Depending on the family configuration, 4.5% to 26.1% of the variance in individual health status was derived from the family. The proportion was highest for older married persons. The family effect on health status was generally similar for physical and mental health. Including age, family income, and health insurance status in the regression equations moderately reduced the family variance component.

CONCLUSIONS

At a population level, the family contribution to individual health status is measurable and substantial. The shared characteristics of income and health insurance account for only a modest portion of the effect. Health policy and interventions should place more emphasis on the family's role in health.

Mortality After Cardiac Bypass Surgery

BACKGROUND

Risk-adjusted outcome rates frequently are used to make inferences about hospital quality of care. We calculated risk-adjusted mortality rates in veterans undergoing isolated coronary artery bypass surgery (CABS) from administrative data and from chart-based clinical data and compared the assessment of hospital high and low outlier status for mortality that results from these 2 data sources.

STUDY POPULATION

We studied veterans who underwent CABS in 43 VA hospitals between October 1, 1993, and March 30, 1996 (n=15,288).

METHODS

To evaluate administrative data, we entered 6 groups of International Classification of Diseases (ICD)-9-CM codes for comorbid diagnoses from the VA Patient Treatment File (PTF) into a logistic regression model predicting postoperative mortality. We also evaluated counts of comorbid ICD-9-CM codes within each group, along with 3 common principal diagnoses, weekend admission or surgery, major procedures associated with CABS, and demographic variables. Data from the VA Continuous Improvement in Cardiac Surgery Program (CICSP) were used to create a separate clinical model predicting postoperative mortality. For each hospital, an observed-to-expected (O/E) ratio of mortality was calculated from (1) the PTF model and (2) the CICSP model. We defined outlier status as an O/E ratio outside of 1.0 (based on the hospital's 90% confidence interval). To improve the statistical and predictive power of the PTF model, selected clinical variables from CICSP were added to it and outlier status reassessed.

RESULTS

Significant predictors of postoperative mortality in the PTF model included 1 group of comorbid ICD-9-CM codes, intraortic balloon pump insertion before CABS, angioplasty on the day of or before CABS, weekend surgery, and a principal diagnosis of other forms of ischemic heart disease. The model's c-index was 0.698. As expected, the CICSP model's predictive power was significantly greater than that of the administrative model (c=0.761). The addition of just 2 CICSP variables to the PTF model improved its predictive power (c=0.741). This model identified 5 of 6 high mortality outliers identified by the CICSP model. Additional CICSP variables were statistically significant predictors but did not improve the assessment of high outlier status.

CONCLUSIONS

Models using administrative data to predict postoperative mortality can be improved with the addition of a very small number of clinical variables. Limited clinical improvements of administrative data may make it suitable for use in quality improvement efforts.

Associations between individual and contextual factors and smoking in 13,293 Mexican students

OBJECTIVE

Factors correlated with cigarette smoking in young people have yet to be documented in most developing countries. This study assesses the correlates of smoking in Mexican young people.

METHODS

School-based, cross-sectional study in the central Mexican state of Morelos during the 1998-1999 school year of 13,293 public school students aged 11 to 24 years. Multinomial logistic regression models were constructed with smoking as the dependent variable.

RESULTS

Regular smoking (one or more cigarettes daily) prevalence was 13.1% (95% confidence interval [CI]=12.2-13.9) in males, and 6.1% (95% CI=5.6-6.6) in females. Frequent alcohol intoxication was strongly associated with regular smoking (females, odds ratio [OR]=68.5, 95% CI=37.6-125.2; males, OR=34.5, 95% CI=22.6-52.7). Regular smoking was associated with illegal drug use and smoking by both parents in females, and with illegal drug use in males (males, OR=4.9, 95% CI=3.7-6.5). Also associated with tobacco smoking were high socioeconomic status, low academic achievement, illegal drug use by peers, marijuana use by parents, and depression in adolescents.

CONCLUSIONS

This study documents a strong correlation between tobacco smoking and other health risk behaviors, especially alcohol and drug abuse. In young women especially, the risk of tobacco use increased with alcohol abuse and higher socioeconomic status. School-based interventions are needed that focus on preventing smoking and also take into account other unhealthy behaviors.