Naiji Lu

On fitting generalized linear mixed‐effects models for binary responses using different statistical packages

The generalized linear mixed-effects model (GLMM) is a popular paradigm to extend models for cross-sectional data to a longitudinal setting. When applied to modeling binary responses, different software packages and even different procedures within a package may give quite different results. In this report, we describe the statistical approaches that underlie these different procedures and discuss their strengths and weaknesses when applied to fit correlated binary responses. We then illustrate these considerations by applying these procedures implemented in some popular software packages to simulated and real study data. Our simulation results indicate a lack of reliability for most of the procedures considered, which carries significant implications for applying such popular software packages in practice.

Log transformation: application and interpretation in biomedical research

The log transformation has been widely used in biomedical research to deal with the skewed data. However, in the medical publications, we have found many misuses and misinterpretations of analysis based on log-transformed data. In this paper, we list some common scenarios of misuse and misinterpretation of log transformation in biomedical applications. We also provide both theoretical and practical justifications to support our viewpoints.

Outcomes of subsyndromal depression in older primary care patients.

OBJECTIVES Most older persons in primary care suffering clinically significant depressive symptoms do not meet criteria for major or minor depression. The authors tested the hypothesis that patients with subsyndromal depression (SSD) would have poorer psychiatric, medical, and functional outcomes at follow-up than nondepressed patients but not as poor as those with minor or major depression. The authors also explored the relative outcomes of three definitions of SSD to determine their relative prognostic value. DESIGN Prospective observational cohort study. SETTING Primary care practices in Monroe County, NY. PARTICIPANTS Four hundred eighty-one primary care patients aged 65 years and older who completed research assessments at intake and at least 1 year of follow-up evaluation. MEASUREMENTS Depression diagnoses and three definitions of SSD were determined by the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, and the 24-item Hamilton Depression Rating Scale. Other validated measures assessed anxiety, cognition, medical burden, and functional status. RESULTS Patients with SSD had poorer 1-year lagged outcomes than nondepressed subjects in terms of psychiatric symptoms and functional status, often not significantly different than major or minor depression. Two of the SSD definitions identified subjects with poorer psychiatric and functional outcomes than the third SSD definition. CONCLUSIONS Clinicians should be vigilant in caring for patients with SSD, monitoring for persistent, or worsening depressive symptoms including suicidality, anxiety, cognitive impairment, and functional decline. Researchers may use particular SSD definitions to identify individuals at higher risk of poor outcomes, to better understand the relationships of SSD to functional disability, and to test innovative preventive and therapeutic interventions.

Mental disorder comorbidity and suicide among 2.96 million men receiving care in the Veterans Health Administration health system.

Comorbid mental disorders are common among suicide decedents. It is unclear if mental disorders in combination confer additive risk for suicide, in other words, if risk associated with two disorders is approximately the sum of the risk conferred by each disorder considered separately, or if there are departures from additivity such that the combined risk is less (i.e., subadditive) or more than additive (i.e., synergistic). Using a retrospective cohort design, all male Department of Veterans Affairs, Veterans Health Administration (VHA) service users who utilized VHA services in fiscal year (FY) 1999 and were alive at the start or FY 2000 (N = 2,962,810) were analyzed. Individuals were followed until death or the end of FY 2006. Using the VHA National Patient Care Database, diagnoses of mental disorders in FY 1999 were grouped into six categories (e.g., posttraumatic stress disorder). In proportional hazards models, 2-way interactions between disorders were used to examine departures from additive risk. There were 7,426 suicide deaths in the study period. Two-way interaction tests were nearly all statistically significant, indicating departures from additivity, and the results of these tests were consistent with subadditive risk. Sensitivity analyses examining the first year of follow-up showed similar results. Subadditive risk may be explained by factors that serve to lower the increased risk associated with a comorbid diagnosis, which may include common underlying causes of mental disorders, difficulties of differential diagnosis, the nature of etiological relationships between mental disorders, and intensive clinical care and monitoring of patients with comorbidity.

Posttraumatic stress disorder and suicide in 5.9 million individuals receiving care in the veterans health administration health system

BACKGROUND Post-traumatic stress disorder (PTSD) confers risk for suicidal ideation and suicide attempts but a link with suicide is not yet established. Prior analyses of users of the Veterans health administration (VHA) Health System suggest that other mental disorders strongly influence the association between PTSD and suicide in this population. We examined the association between PTSD and suicide in VHA users, with a focus on the influence of other mental disorders. METHODS Data were based on linkage of VA National Patient Care Database records and the Centers for Disease Control and Prevention׳s National Death Index, with data from fiscal year 2007-2008. Analyses were based on multivariate logistic regression and structural equation models. RESULTS Among users of VHA services studied (N=5,913,648), 0.6% (N=3620) died by suicide, including 423 who had had been diagnosed with PTSD. In unadjusted analysis, PTSD was associated with increased risk for suicide, with odds ratio, OR (95% confidence interval, 95% CI)=1.34 (1.21, 1.48). Similar results were obtained after adjustment for demographic variables and veteran characteristics. After adjustment for multiple other mental disorder diagnoses, PTSD was associated with decreased risk for suicide, OR (95% CI)=0.77 (0.69, 0.86). Major depressive disorder (MDD) had the largest influence on the association between PTSD and suicide. LIMITATIONS The analyses were cross-sectional. VHA users were studied, with unclear relevance to other populations. CONCLUSION The findings suggest the importance of identifying and treating comorbid MDD and other mental disorders in VHA users diagnosed with PTSD in suicide prevention efforts.

Prevalence and natural course of late-life depression in China primary care: A population based study from an urban community

BACKGROUND Primary care is the most promising venue for the management of late-life depression in China. The current study was designed to establish the prevalence of major depressive disorder among older adults in primary care, and to examine the correlates, and the natural course of late-life depression over a year. METHODS A sample of 1275 adults aged over 60 years was recruited from a primary care clinic in urban China for screening with PHQ-9, and 262 participants stratified by PHQ-9 score were interviewed to collect the presence of major depressive disorder (MDD), the availability of social support, and physical health and functional status. Participants were followed up for 12 months at 3-month intervals. RESULTS The estimated prevalence of MDD was 11.3% with the SCID interview. Increasing age, female gender, and lower educational level, living alone, low support from family, high medical illness burden, and impairment of daily function were significantly associated with MDD in later life. Less than 1% of these patients received treatments. More than 60% of patients with MDD at baseline remained depressed throughout the 12 month follow-up period; and only 3 patients had been treated during the 12-month follow-up. LIMITATIONS The correlates of late-life depression observed here may not necessarily serve as risk factors guiding the development of future prevention strategies. DISCUSSION In an urban Chinese primary care setting, late-life depression was found to be a common condition. Few patients with MDD received treatment for their condition, and the majority remained depressed over the following year.

On the Impact of Parametric Assumptions and Robust Alternatives for Longitudinal Data Analysis

Models for longitudinal data are employed in a wide range of behavioral, biomedical, psychosocial, and health-care-related research. One popular model for continuous response is the linear mixed-effects model (LMM). Although simulations by recent studies show that LMM provides reliable estimates under departures from the normality assumption for complete data, the invariable occurrence of missing data in practical studies renders such robustness results less useful when applied to real study data. In this paper, we show by simulated studies that in the presence of missing data estimates of the fixed effect of LMM are biased under departures from normality. We discuss two robust alternatives, the weighted generalized estimating equations (WGEE) and the augmented WGEE (AWGEE), and compare their performances with LMM using real as well as simulated data. Our simulation results show that both WGEE and AWGEE provide valid inference for skewed non-normal data when missing data follows the missing at random, the most popular missing data mechanism for real study data.

The effect of interpersonal psychotherapy for depression on insomnia symptoms in a cohort of women with sexual abuse histories

Insomnia frequently occurs with trauma exposure and depression, but can ameliorate with improvements in depression. Insomnia was assessed by the insomnia subscale of the Hamilton Rating Scale for Depression in 106 women with childhood sexual abuse (CSA) and major depression receiving interpersonal psychotherapy (IPT) in an uncontrolled pilot (n = 36) and an immediately subsequent randomized controlled trial (n = 70) comparing IPT to treatment as usual. Depression improved in each study and in both treatment conditions; insomnia had smaller, nonsignificant improvements. Overall, 95 women (90%) endorsed insomnia on the Structured Clinical Interview for DSM-IV at baseline and, of those, 90% endorsed insomnia following treatment. Despite improvements in depression, insomnia persists for most women with CSA.

Depression care management for adults older than 60 years in primary care clinics in urban China: a cluster-randomised trial

BACKGROUND Chinas national health policy classifies depression as a chronic disease that should be managed in primary care settings. In some high-income countries use of chronic disease management principles and primary care-based collaborative-care models have improved outcomes for late-life depression; however, this approach has not yet been tested in China. We aimed to assess whether use of a collaborative-care depression care management (DCM) intervention could improve outcomes for Chinese adults with depression aged 60 years and older. METHODS Between Jan 17, 2011, [corrected] and Nov 30, 2013, we did a cluster-randomised trial in patients from primary care centre clinics in Shangcheng district of Hangzhou city in eastern China. We randomly assigned (1:1) clinics to either DCM (involving training for physicians in use of treatment guidelines, training for primary care nurses to function as care managers, and consultation with psychiatrists as support) or to give enhanced care as usual to all eligible patients aged 60 years and older with major depressive disorder. Clinics were chosen randomly for inclusion from all primary care clinics in the district by computer algorithm and then randomly allocated depression care interventions remotely by computer algorithm. Physicians, study personnel, and patients were not masked to clinic assignment. Our primary outcome was difference in Hamilton Depression Rating Scale (HAMD) score using data for clusters at baseline and 3, 6, and 12 month follow-up in a mixed-effects model of the intention-to-treat population. We originally aimed to analyse outcomes at 24 months, however the difference between groups at 12 months was large and funding was insufficient to continue to 24 months, therefore we decided to end the trial at 12 months. This trial is registered with ClinicalTrials.gov, number NCT01287494. FINDINGS Of 34 primary care clinics in Shangcheng district, 16 were randomly chosen. We randomly assigned eight clinics to the DCM intervention (164 patients enrolled) and eight primary care clinics to enhanced care as usual (162 patients). There were no major differences in baseline demographic and clinical variables between the groups of patients for each intervention. Over the 12 months, patients in clinics assigned to DCM had a significantly greater reduction in HAMD score than did those in practices assigned to enhanced care as usual (estimated between group difference -6·5 [95% CI -7·1 to -5·9]; Cohens d 0·8 [95% CI 0·8-0·9]; p<0·0001). The intercluster correlation for change in HAMD total score was 0·07 (95% CI 0·06-0·08). There were no study-related adverse events in either group. INTERPRETATION Clinical outcomes of Chinese adults older than 60 years who had major depression were improved when their primary care clinic used DCM. Primary care-based collaborative management of depression is promising to address this pressing public health need in China. FUNDING National Institutes of Health, Program for New Century Excellent Talents in Universities of China, Ministry of Education, China.

Extending the Mann–Whitney–Wilcoxon rank sum test to longitudinal regression analysis

Outliers are commonly observed in psychosocial research, generally resulting in biased estimates when comparing group differences using popular mean-based models such as the analysis of variance model. Rank-based methods such as the popular Mann–Whitney–Wilcoxon (MWW) rank sum test are more effective to address such outliers. However, available methods for inference are limited to cross-sectional data and cannot be applied to longitudinal studies under missing data. In this paper, we propose a generalized MWW test for comparing multiple groups with covariates within a longitudinal data setting, by utilizing the functional response models. Inference is based on a class of U-statistics-based weighted generalized estimating equations, providing consistent and asymptotically normal estimates not only under complete but missing data as well. The proposed approach is illustrated with both real and simulated study data.