A comparison of confounding adjustment methods with an application to early life determinants of childhood obesity.

We implemented six confounding adjustment methods: (1) covariate-adjusted regression, (2) propensity score (PS) regression, (3) PS stratification, (4) PS matching with two calipers, (5) inverse probability weighting and (6) doubly robust estimation to examine the associations between the body mass index (BMI) z-score at 3 years and two separate dichotomous exposure measures: exclusive breastfeeding v. formula only (n=437) and cesarean section v. vaginal delivery (n=1236). Data were drawn from a prospective pre-birth cohort study, Project Viva. The goal is to demonstrate the necessity and usefulness, and approaches for multiple confounding adjustment methods to analyze observational data. Unadjusted (univariate) and covariate-adjusted linear regression associations of breastfeeding with BMI z-score were -0.33 (95% CI -0.53, -0.13) and -0.24 (-0.46, -0.02), respectively. The other approaches resulted in smaller n (204-276) because of poor overlap of covariates, but CIs were of similar width except for inverse probability weighting (75% wider) and PS matching with a wider caliper (76% wider). Point estimates ranged widely, however, from -0.01 to -0.38. For cesarean section, because of better covariate overlap, the covariate-adjusted regression estimate (0.20) was remarkably robust to all adjustment methods, and the widths of the 95% CIs differed less than in the breastfeeding example. Choice of covariate adjustment method can matter. Lack of overlap in covariate structure between exposed and unexposed participants in observational studies can lead to erroneous covariate-adjusted estimates and confidence intervals. We recommend inspecting covariate overlap and using multiple confounding adjustment methods. Similar results bring reassurance. Contradictory results suggest issues with either the data or the analytic method.

Comparison of interactive voice response, patient mailing, and mailed registry to encourage screening for osteoporosis: a randomized controlled trial.

UNLABELLED: Guidelines recommend screening for osteoporosis with bone mineral density (BMD) testing in menopausal women, particularly those with additional risk factors for fracture. Many eligible women remain unscreened. This randomized study demonstrates that a single outreach interactive voice response phone call improves rates of BMD screening among high-risk women age 50-64. INTRODUCTION: Osteoporotic fractures are a major cause of disability and mortality. Guidelines recommend screening with BMD for menopausal women, particularly those with additional risk factors for fracture. However, many women remain unscreened. We examined whether telephonic interactive voice response (IVR) or patient mailing could increase rates of BMD testing in high risk, menopausal women. METHODS: We studied 4,685 women age 50-64 years within a not-for-profit health plan in the United States. All women had risk factors for developing osteoporosis and no prior BMD testing or treatment for osteoporosis. Patients were randomly allocated to usual care, usual care plus IVR, or usual care plus mailed educational materials. To avoid contamination, patients within a single primary care physician practice were randomized to receive the same intervention. The primary endpoint was BMD testing at 12 months. Secondary outcomes included BMD testing at 6 months and medication use at 12 months. RESULTS: Mean age was 57 years. Baseline demographic and clinical characteristics were similar across the three study groups. In adjusted analyses, the incidence of BMD screening was 24.6% in the IVR group compared with 18.6% in the usual care group (P < 0.001). There was no difference between the patient mailing group and the usual care group (P = 0.3). CONCLUSIONS: In this large community-based randomized trial of high risk, menopausal women age 50-64, IVR, but not patient mailing, improved rates of BMD screening. IVR remains a viable strategy to incorporate in population screening interventions.

Do community-level predictors of pneumococcal carriage continue to play a role in the conjugate vaccine era?

This paper examined whether previously identified community-level factors (high proportion of crowded households and/or persons below the poverty level) remained associated with childhood pneumococcal carriage in the heptavalent pneumococcal conjugate vaccine (PCV7) era. Using logistic regression, individual factors were used to develop base models to which community-level factors were added to evaluate impact on pneumococcal carriage within two paediatric study cohorts from Massachusetts (urban Boston, outside Boston). Six years after introduction of universal childhood PCV7 vaccination, we found no consistent evidence that census tract characteristics (e.g. population size and density, age and race distribution, percent participating in group childcare, parental education, percent lacking in-unit plumbing, poverty, and community stability) affected odds of pneumococcal carriage when added to individual predictors (e.g. younger age, current respiratory tract infections, and attendance in group childcare). How community-level factors influence pneumococcal carriage continues to change in the era of increasing immunization coverage.

Maternal experiences of racial discrimination and child weight status in the first 3 years of life.

Among US racial/ethnic minority women, we examined associations between maternal experiences of racial discrimination and child growth in the first 3 years of life. We analyzed data from Project Viva, a pre-birth cohort study. We restricted analyses to 539 mother-infant pairs; 294 were Black, 127 Hispanic, 110 Asian and 8 from additional racial/ethnic groups. During pregnancy, mothers completed the Experiences of Discrimination survey that measured lifetime experiences of racial discrimination in diverse domains. We categorized responses as 0, 1-2 or ≥3 domains. Main outcomes were birth weight for gestational age z-score; weight for age (WFA) z-score at 6 months of age; and at 3 years of age, body mass index (BMI) z-score. In multivariable analyses, we adjusted for maternal race/ethnicity, nativity, education, age, pre-pregnancy BMI, household income and child sex and age. Among this cohort of mostly (58.2%) US-born and economically non-impoverished mothers, 33% reported 0 domains of discrimination, 33% reported discrimination in 1-2 domains and 35% reported discrimination in ≥3 domains. Compared with children whose mothers reported no discrimination, those whose mothers reported ≥3 domains had lower birth weight for gestational age z-score (ß -0.25; 95% CI: -0.45, -0.04), lower 6 month WFA z-score (ß -0.34; 95% CI: -0.65, -0.03) and lower 3-year BMI z-score (ß -0.33; 95% CI: -0.66, 0.00). In conclusion, we found that among this cohort of US racial/ethnic minority women, mothers' report of experiencing lifetime discrimination in ⩾ 3 domains was associated with lower fetal growth, weight at 6 months and 3-year BMI among their offspring.

Using encounters versus episodes in syndromic surveillance.

BACKGROUND: Automated electronic medical records may be useful for syndromic surveillance to quickly detect infectious disease outbreaks. Some syndromic surveillance systems include every encounter in the analysis, whereas others exclude individuals' repeat encounters within the same syndrome occurring within a short period of time, with the rationale that these represent follow-up visits rather than new episodes of illness. METHODS: We evaluate the effect of keeping all encounters as compared with removing repeat encounters. Using the prospective space-time permutation scan statistic, we performed daily analyses on all encounters versus on episodes defined as encounters new within 2, 6 or 12 weeks. Data were taken from a Massachusetts Health Maintenance Organization (HMO) for the calendar year 1999 for four different syndromes. RESULTS: We found extensive disagreement in the number of signals detected: 70, 68, 21 and 15 signals when using all encounters versus 15-20, 3, 4-5 and 0 signals when using only new episodes for lower respiratory, lower gastrointestinal, upper gastrointestinal and neurologic syndromes, respectively. CONCLUSION: Using all encounters in syndromic surveillance may not only create too many signals but may also miss some signals by masking the anomalies generated by actual episodes. However, it is also possible to miss signals when using episodes.

Elevated midpregnancy corticotropin-releasing hormone is associated with prenatal, but not postpartum, maternal depression.

CONTEXT: Elevated hypothalamic CRH has been implicated in melancholic major depression in nonpregnant individuals, but the role of placental CRH in maternal prenatal and postpartum depression is largely unexplored. OBJECTIVE: The objective of the study was to examine the association of maternal midpregnancy plasma CRH levels with prenatal and postpartum depression. PARTICIPANTS: The study included 800 participants in Project Viva, a pregnancy and childhood cohort. METHODS: CRH levels were analyzed from blood samples obtained at mean 27.9 wk gestation (+/- 1.3 sd; range 24.6-37.4 wk) and were normalized on the logarithmic scale. Depression was assessed with the Edinburgh Postpartum Depression Scale (range 0-30 points) in midpregnancy and at 6 months postpartum. We used logistic regression to estimate the odds of scoring 13 or more points on the Edinburgh Postpartum Depression Scale as indicative of major or minor depression. RESULTS: Seventy (8.8%) and 46 (7.5%) women had prenatal and postpartum depression symptoms, respectively. Mean log CRH was 4.93 (+/- 0.62 sd). After adjusting for confounders, an sd increase in log CRH was associated with nearly 50% higher odds of prenatal depression symptoms (odds ratio 1.48, 95% confidence interval 1.14-1.93). Higher CRH levels during pregnancy were unassociated with greater risk of postpartum depressive symptoms. In fact, there was a suggestion that prenatal CRH levels might be inversely associated with risk of postpartum depressive symptoms (odds ratio 0.82, 95% confidence interval 0.58-1.15). CONCLUSIONS: Elevated placental CRH levels in midpregnancy are positively associated with risk of prenatal depression symptoms but not postpartum depression symptoms.

Comparison of syndromic surveillance and a sentinel provider system in detecting an influenza outbreak--Denver, Colorado, 2003.

INTRODUCTION: Syndromic surveillance systems can be useful in detecting naturally occurring illness. OBJECTIVES: Syndromic surveillance performance was assessed to identify an early and severe influenza A outbreak in Denver in 2003. METHODS: During October 1, 2003-January 31, 2004, syndromic surveillance signals generated for detecting clusters of influenza-like illness (ILI) were compared with ILI activity identified through a sentinel provider system and with reports of laboratory-confirmed influenza. The syndromic surveillance and sentinel provider systems identified ILI activity based on ambulatory-care visits to Kaiser Permanente Colorado. The syndromic surveillance system counted a visit as ILI if the provider recorded any in a list of 30 respiratory diagnoses plus fever. The sentinel provider system required the provider to select "influenza" or "ILI." RESULTS: Laboratory-confirmed influenza cases, syndromic surveillance ILI episodes, and sentinel provider reports of patient visits for ILI all increased substantially during the week ending November 8, 2003. A greater absolute increase in syndromic surveillance episodes was observed than in sentinel provider reports, suggesting that sentinel clinicians failed to code certain cases of influenza. During the week ending December 6, when reports of laboratory-confirmed cases peaked, the number of sentinel provider reports exceeded the number of syndromic surveillance episodes, possibly because clinicians diagnosed influenza without documenting fever. CONCLUSION: Syndromic surveillance performed as well as the sentinel provider system, particularly when clinicians were advised to be alert to influenza, suggesting that syndromic surveillance can be useful for detecting clusters of respiratory illness in various settings.

Ambulatory-care diagnoses as potential indicators of outbreaks of gastrointestinal illness--Minnesota.

INTRODUCTION: Syndromic surveillance's capability to augment existing surveillance for community-acquired gastrointestinal disease is unknown. OBJECTIVE: The objective of this study was to evaluate the capability of a syndromic surveillance system to detect outbreaks of gastrointestinal disease. METHODS: A retrospective analysis was conducted comparing ambulatory care data from a health plan with a set of 110 gastrointestinal-disease outbreaks identified by the Minnesota Department of Health during 2001-02. Unusual clusters of illness (i.e., signals) in the health-plan data were identified by analyzing daily counts of gastrointestinal illness using an adjusted space-time scan statistic. Concordance was defined as < or =5 km between outbreak and signal and the signal occurring within 1 week of the outbreak. RESULTS: During 104 weeks, the number of signals was roughly what would have been expected by chance, suggesting that the modeling did a good job of estimating the expected counts of illness and that false alarms would not have occurred much more often than the number predicted at the various thresholds. During the same period, the health department identified 110 eligible gastrointestinal outbreaks. Apparent associations of the three statistically most unusual concordant signals with outbreaks of viral or bacterial gastrointestinal illness were ruled out by the health department on the basis of detailed knowledge of the circumstances and low numbers of affected persons seeking medical care. CONCLUSION: No previously known gastrointestinal outbreaks were identified by this surveillance system. However, relatively few recognized outbreaks resulted in patients seeking medical care, and the sensitivity of this system to detect outbreaks of real significance to public health remains to be determined. Prospective evaluation probably will be required to understand the usefulness of syndromic surveillance systems to enhance existing disease surveillance.

A model-adjusted space-time scan statistic with an application to syndromic surveillance.

The space-time scan statistic is often used to identify incident disease clusters. We introduce a method to adjust for naturally occurring temporal trends or geographical patterns in illness. The space-time scan statistic was applied to reports of lower respiratory complaints in a large group practice. We compared its performance with unadjusted populations from: (1) the census, (2) group-practice membership counts, and on adjustments incorporating (3) day of week, month, and holidays; and (4) additionally, local history of illness. Using a nominal false detection rate of 5%, incident clusters during 1 year were identified on 26, 22, 4 and 2% of days for the four populations respectively. We show that it is important to account for naturally occurring temporal and geographic trends when using the space-time scan statistic for surveillance. The large number of days with clusters renders the census and membership approaches impractical for public health surveillance. The proposed adjustment allows practical surveillance.

National Bioterrorism Syndromic Surveillance Demonstration Program.

The National Bioterrorism Syndromic Surveillance Demonstration Program identifies new cases of illness from electronic ambulatory patient records. Its goals are to use data from health plans and practice groups to detect localized outbreaks and to facilitate rapid public health follow-up. Data are extracted nightly on patient encounters occurring during the previous 24 hours. Visits or calls with diagnostic codes corresponding to syndromes of interest are counted; repeat encounters are excluded. Daily counts of syndromes by zip code are sent to a central data repository, where they are statistically analyzed for unusual clustering by using a model-adjusted SaTScan approach. The results and raw data are displayed on a restricted website. Patient-level information stays at the originating health-care organization unless required by public health authorities. If a cluster surpasses a threshold of statistical aberration chosen by the corresponding public health department, an electronic alert can be sent to that department. The health department might then call a clinical responder, who has electronic access to records of cases contributing to clusters. The system is flexible, allowing for changes in participating organizations, syndrome definitions, and alert thresholds. It is transparent to clinicians and has been accepted by the health-care organizations that provide the data. The system's data are usable by local and national health agencies. Its software is compatible with commonly used systems and software and is mostly open-source. Ongoing activities include evaluating the system's ability to detect naturally occurring outbreaks and simulated terrorism events, automating and testing alerts and response capability, and evaluating alternative data sources.

Using automated medical records for rapid identification of illness syndromes (syndromic surveillance): the example of lower respiratory infection.

BACKGROUND: Gaps in disease surveillance capacity, particularly for emerging infections and bioterrorist attack, highlight a need for efficient, real time identification of diseases. METHODS: We studied automated records from 1996 through 1999 of approximately 250,000 health plan members in greater Boston. RESULTS: We identified 152,435 lower respiratory infection illness visits, comprising 106,670 episodes during 1,143,208 person-years. Three diagnoses, cough (ICD9CM 786.2), pneumonia not otherwise specified (ICD9CM 486) and acute bronchitis (ICD9CM 466.0) accounted for 91% of these visits, with expected age and sex distributions. Variation of weekly occurrences corresponded closely to national pneumonia and influenza mortality data. There was substantial variation in geographic location of the cases. CONCLUSION: This information complements existing surveillance programs by assessing the large majority of episodes of illness for which no etiologic agents are identified. Additional advantages include: a) sensitivity, uniformity and efficiency, since detection of events does not depend on clinicians' to actively report diagnoses, b) timeliness, the data are available within a day of the clinical event; and c) ease of integration into automated surveillance systems. These features facilitate early detection of conditions of public health importance, including regularly occurring events like seasonal respiratory illness, as well as unusual occurrences, such as a bioterrorist attack that first manifests as respiratory symptoms. These methods should also be applicable to other infectious and non-infectious conditions. Knowledge of disease patterns in real time may also help clinicians to manage patients, and assist health plan administrators in allocating resources efficiently.

ACTIVE: a cognitive intervention trial to promote independence in older adults.

The Advanced Cognitive Training for Independent and Vital Elderly (ACTIVE) trial is a randomized, controlled, single-masked trial designed to determine whether cognitive training interventions (memory, reasoning, and speed of information processing), which have previously been found to be successful at improving mental abilities under laboratory or small-scale field conditions, can affect cognitively based measures of daily functioning. Enrollment began during 1998; 2-year follow-up will be completed by January 2002. Primary outcomes focus on measures of cognitively demanding everyday functioning, including financial management, food preparation, medication use, and driving. Secondary outcomes include health-related quality of life, mobility, and health-service utilization. Trial participants (n = 2832) are aged 65 and over, and at entry into the trial, did not have significant cognitive, physical, or functional decline. Because of its size and the carefully developed rigor, ACTIVE may serve as a guide for future behavioral medicine trials of this nature.

Adaptive double data entry: a probabilistic tool for choosing which forms to reenter.

Several recent articles have supported differing opinions about the value and cost of double data entry in specific clinical trials. The cost of the reentry, combined with the low error rate typical in controlled clinical trials suggests to some that single data entry may be sufficient, with the cost of reentry allocated to more productive quality assurance tools. In this article, an alternative approach to limiting costs and maintaining the quality of entered data is offered. The technique is a formal, adaptive method for choosing a subset of forms to be reentered. The idea behind the approach is to decide whether a given form should be reentered on a form-by-form basis for each data entry person, using an estimated probability that the form contains an error as a guideline. The method automatically adapts to each data entry person and to temporal changes in accuracy within data entry person. The estimated probability is based on a lagged set of the most recently double-data-entered forms. A simple simulation shows that much of the reentry can be avoided while still detecting many of the errors. A real data example demonstrates that the procedure can be effective in practice as well. Control Clin Trials 2001;22:2-12

Does bicycling contribute to the risk of erectile dysfunction? Results from the Massachusetts Male Aging Study (MMAS).

An association between bicycling and erectile dysfunction (ED) has been described previously, but there are limited data examining this association in a random population of men. Such data would incorporate bicyclists with varied types of riding and other factors. Data from the Massachusetts Male Aging Study (MMAS) were utilized to examine the association between bicycling and ED. Logistic regression was used to test for an association, controlling for age, energy expenditure, smoking, depression and chronic illness. Bicycling less than 3 h per week was not associated with ED and may be somewhat protective. Bicycling 3 h or more per week may be associated with ED. Data revealed that there may be a reduced probability of ED in those who ride less than 3 h per week and ED may be more likely in bikers who ride more than 3 h per week. More population-based research is needed to better define this relationship.

Socioeconomic factors and incidence of erectile dysfunction: findings of the longitudinal Massachussetts Male Aging Study.

Despite the well-documented relationship of socioeconomic factors (SEF) to various health problems, the relationship of SEF to erectile dysfunction (ED) is not well understood. As such, the goals of this paper are: (1) to determine whether incident ED is more likely to occur among men with low SEF; and (2) to determine whether incident ED varies by SEF after taking into consideration other well-established ED risk factors that are also associated with SEF such as smoking, diabetes, and high blood pressure. We used data from 797 participants in the longitudinal population-based Massachusetts Male Aging Study (baseline 1987-1989, follow-up 1995-1997) who were free of ED at baseline and had complete data on ED and all risk factors. ED was determined by a self-administered questionnaire and its relationship to SEF was assessed using logistic regression. We first analyzed the age-adjusted relationship of education, income, and occupation to incidence of ED. The results show that men with low education (O.R. = 1.46, 95% C.I. = 1.02-2.08) or men in blue-collar occupations (O.R. = 1.68, 95% C.I. = 1.16-2.43) are significantly more likely to develop ED. For the multivariate model, due to multicollinearity among education, income, and occupation, we ran three separate models. After taking into consideration all the other risk factors--age, lifestyle and medical conditions--the effect of occupation remained significant. Men who worked in blue-collar occupations were one and a half times more likely to develop ED compared to men in white-collar occupations (O.R. = 1.55, 95% C.I. = 1.06-2.28).

Erectile dysfunction and coronary risk factors: prospective results from the Massachusetts male aging study.

BACKGROUND: Erectile dysfunction (ED), a wide spread and troublesome condition among middle-aged men, is partly vascular in origin. In the Massachusetts Male Aging Study, a random-sample cohort study, we investigated the relationship between baseline risk factors for coronary heart disease and subsequent ED, on the premise that subclinical arterial insufficiency might be manifested as ED. METHODS: Men ages 40-70, selected from state census lists, were interviewed in 1987-1989 and reinterviewed in 1995-1997. Data were collected and blood was drawn in participants' homes. ED was assessed from responses to a privately self-administered questionnaire. Analysis was restricted to 513 men with no ED at baseline and no diabetes, heart disease, or related medications at either time. RESULTS: Cigarette smoking at baseline almost doubled the likelihood of moderate or complete ED at followup (24% vs. 14%, adjusted for age and covariates, P = 0.01). Cigar smoking and passive exposure to cigarette smoke also significantly predicted incident ED, as did overweight (body-mass index > or =28 kg/m(2)) and a composite coronary risk score. Weaker prospective associations were seen for hypertension and dietary intake of cholesterol and unsaturated fat. CONCLUSIONS: Erectile dysfunction and coronary heart disease share some behaviorally modifiable determinants in men who, like our sample, are free of manifest ED or predisposing illness. Open questions include whether modification of coronary risk factors can prevent ED and whether ED may serve as a sentinel event for coronary disease.

A new surrogate variable for erectile dysfunction status in the Massachusetts male aging study.

Erectile dysfunction (ED) is the subject of a vast clinical literature, but little information has been gathered from random samples of the general public. The Massachusetts Male Aging Study (MMAS) addressed this important aspect of men's health. The MMAS was conducted in two waves, with baseline data collection in 1987-1989 and follow-up in 1995-1997. Subsequent to the baseline MMAS survey, a consensus developed that subjective measures are optimal for defining ED. Unfortunately, the baseline questionnaire did not ask subjects directly about their erectile functioning. Thus, we previously assigned the MMAS subjects a degree of impotence at baseline using a series of related questions, employing a discriminant formula constructed from a separate sample of urology clinic patients. At follow-up the men classified themselves directly in addition to answering the original series of related questions. In the present article, we report the results of a new discriminant function, based on the MMAS men at follow-up. We also compare the two methods and discuss our reasons for preferring the internally calibrated method.

Incidence of erectile dysfunction in men 40 to 69 years old: longitudinal results from the Massachusetts male aging study.

PURPOSE: We estimated the incidence of erectile dysfunction in men 40 to 69 years old at study entry during an average 8.8-year followup, and determined how risk varied with age, socioeconomic status and medical conditions. MATERIALS AND METHODS: Data from a randomly sampled population based longitudinal study of Massachusetts men were analyzed. A total of 1,709 men completed the baseline interview during 1987 to 1989 and 1,156 survivors completed followup from 1995 to 1997. The analysis sample consisted of 847 men without erectile dysfunction at baseline and with complete followup information. Erectile dysfunction was assessed by discriminant analysis of 13 questions from a self-administered sexual function questionnaire and a single global self-rating question. RESULTS: The crude incidence rate for erectile dysfunction was 25.9 cases per 1,000 man-years (95% confidence interval [CI] 22.5 to 29.9). The annual incidence rate increased with each decade of age and was 12.4 cases per 1,000 man-years (95% CI 9.0 to 16.9), 29.8 (24.0 to 37.0) and 46.4 (36.9 to 58.4) for men 40 to 49, 50 to 59 and 60 to 69 years old, respectively. The age adjusted risk of erectile dysfunction was higher for men with lower education, diabetes, heart disease and hypertension. Population projections for men 40 to 69 years old suggest that 17,781 new cases of erectile dysfunction in Massachusetts and 617,715 in the United States (white males only) are expected annually. CONCLUSIONS: Although prevalence estimates and cross-sectional correlates of erectile dysfunction have recently been established, incidence estimates were lacking. Incidence is necessary to assess risk, and plan treatment and prevention strategies. The risk of erectile dysfunction was about 26 cases per 1,000 men annually, and increased with age, lower education, diabetes, heart disease and hypertension.

Prostate cancer: how much do we know and how do we know it?

We use data from the Massachusetts Male Aging Study to approximate the total current knowledge about prostate cancer and to evaluate the relative contributions of various risk factors. The sum total of current knowledge is assessed using the area under the receiver operating characteristic curve (AUROC) and by the Hosmer-Lemeshow statistic in a logistic regression model that includes 30 risk factors identified in the literature that are available in the data set. Relative contributions are measured using the adjusted generalized R2 (AR2). To measure relative contributions, we group risk factors with similar etiology and then remove groups and compare the AR2 attained without each group to the AR2 with all variables included. The overall model fits adequately. The A UROC is 0.788, relatively far from its default value of 0.5, and the Hosmer-Lemeshow statistic has a p value of 0.926. We conclude that, while some knowledge about prostate cancer has been accumulated, there are still more risk factors yet unsuspected. The relative importance analysis shows that immutable factors (i.e. age, genotype) contribute 42%, dietary factors 30%, other lifestyle factors 15%, and endocrinological factors 11%. We recommend that age and other unchanging factors should be the primary focus of screening and risk evaluation and that future interventions should center on nutritional behavior.

A semi-parametric Bayesian approach to generalized linear mixed models.

The linear mixed effects model with normal errors is a popular model for the analysis of repeated measures and longitudinal data. The generalized linear model is useful for data that have non-normal errors but where the errors are uncorrelated. A descendant of these two models generates a model for correlated data with non-normal errors, called the generalized linear mixed model (GLMM). Frequentist attempts to fit these models generally rely on approximate results and inference relies on asymptotic assumptions. Recent advances in computing technology have made Bayesian approaches to this class of models computationally feasible. Markov chain Monte Carlo methods can be used to obtain 'exact' inference for these models, as demonstrated by Zeger and Karim. In the linear or generalized linear mixed model, the random effects are typically taken to have a fully parametric distribution, such as the normal distribution. In this paper, we extend the GLMM by allowing the random effects to have a non-parametric prior distribution. We do this using a Dirichlet process prior for the general distribution of the random effects. The approach easily extends to more general population models. We perform computations for the models using the Gibbs sampler.