Poisson Log-linear Regression Models Research Articles

Poisson Log-Linear Regression Model for Rural Signalized Intersection

Safety performance of rural signalized intersections is critical for identifying high-risk sites and predicting the hazardousness. This paper aims to develop a predictive model that will describe the safety of rural signalized intersections based on various input variables. Data are examined from 124 rural signalized intersections over three states, and Poisson log-linear regression model is presented, which connected traffic number and the average traffic volumes, geometric characteristics and signalization characteristics variables together. The model and associated data analysis reveal that average daily traffic, media width, speed limit, degree of horizontal curvature and left-turn lane are the factors that have greatest overall effect on safety. The results show that the Poisson log-linear regression model is able to describe the rural signalized intersection safety accurately. Using this model, effective countermeasures can be applied for improving traffic safety.

Meteorological conditions and incidence of Legionnaires' disease in Glasgow, Scotland: application of statistical modelling

This study investigated the relationships between Legionnaires' disease (LD) incidence and weather in Glasgow, UK, by using advanced statistical methods. Using daily meteorological data and 78 LD cases with known exact date of onset, we fitted a series of Poisson log-linear regression models with explanatory variables for air temperature, relative humidity, wind speed and year, and sine-cosine terms for within-year seasonal variation. Our initial model showed an association between LD incidence and 2-day lagged humidity (positive, P = 0·0236) and wind speed (negative, P = 0·033). However, after adjusting for year-by-year and seasonal variation in cases there were no significant associations with weather. We also used normal linear models to assess the importance of short-term, unseasonable weather values. The most significant association was between LD incidence and air temperature residual lagged by 1 day prior to onset (P = 0·0014). The contextual role of unseasonably high air temperatures is worthy of further investigation. Our methods and results have further advanced understanding of the role which weather plays in risk of LD infection.

Seasonality, incidence and prognosis in atrial fibrillation and stroke in Denmark and New Zealand

ObjectivesThere are relatively few large studies of seasonal variation in the occurrence of stroke in patients with atrial fibrillation (AF). We investigated the seasonal variation in incidence rates of hospitalisation...

Age- and Sex-Specific Trends in the Incidence of Hospitalized Acute Coronary Syndromes in Western Australia

BACKGROUND- The incidence of myocardial infarction has declined during the past 4 decades in many populations. However, there are limited population data measuring trends in acute coronary syndromes (ACS). We therefore examined temporal trends in the incidence of hospitalized ACS by age and sex in a population-based cohort. METHODS AND RESULTS- The Western Australian Data Linkage System, a repository of linked administrative health data, was used to identify 29 421 incident ACS hospitalizations between 1996 and 2007. Poisson log-linear regression models were used to calculate incidence rate changes. Age-standardized incidence rates of ACS declined annually in men by 1.7% (95% confidence interval [CI], -2.1 to -1.3) and in women by 1.6% (95% CI, -2.1 to -1.0). These declining rates were underpinned by annual reductions in the incidence of unstable angina (men, -3.0%; 95% CI, -3.7 to -2.4; women, -2.5; 95% CI, -3.3 to -1.7), whereas annual changes in myocardial infarction incidence were less (men, -1.0%; 95% CI, -1.5 to -0.5; women, -0.8%; 95% CI, -1.6 to 0). However, the overall trends masked age group differences, with ACS incidence increasing annually in 35- to 54-year-old women (2.3%; 95% CI, 1.0 to 3.8), predominantly driven by increasing incidence of myocardial infarction. CONCLUSIONS- The age-standardized incidence of ACS decreased significantly in Western Australia from 1996 to 2007. However, an increase in ACS incidence in women ages 35 to 54 years is troubling and warrants further investigation.

A hierarchical model for real-time monitoring of variation in risk of non-specific gastrointestinal infections

The AEGISS (Ascertainment and Enhancement of Disease Surveillance and Statistics) project uses spatio-temporal statistical methods to identify anomalies in the incidence of gastrointestinal infections in the UK. The focus of this paper is the modelling of temporal variation in incidence using data from the Southampton area in southern England. We identified and fitted a hierarchical stochastic model for the time series of daily incident cases to enable probabilistic prediction of temporal variation in risk, and demonstrated the resulting gains in predictive accuracy by comparison with a conventional analysis based on an over-dispersed Poisson log-linear regression model. We used Bayesian methods of inference in order to incorporate parameter uncertainty in our predictive inference of risk. Incorporation of our model in the overall spatio-temporal model, will contribute to the accurate and timely prediction of unusually high food-poisoning incidence, and thus to the identification and prevention of future outbreaks.

Socioeconomic and Environmental Determinants of Hospitalizations for Childhood and Adult Asthma in Hawaii

PP-30-168 Background/Aims: Asthma has been estimated to affect 17 million people or over 5% of the total population, with 1.9 million emergency room visits, and 466,000 hospital admissions annually in the United States. Approximately 9.7% of children and 6.9% of adults have been diagnosed with asthma in Hawaii. Understanding the disproportionate increase and remarkable disparities in asthma among low socioeconomic status indigenous and ethnic minority population groups may provide insight into the roots of the asthma epidemics and more effective disease management. Methods: The data on annual asthma hospitalizations for children and adults (ICD-9-CM 493.0–493.9), at sub-county school administrative district level (n = 22), in 2000–2005, was collected from Hawaii Health Information Corporation, and covered the counties of Honolulu, Hawaii, and Maui. The information on socioeconomic status, included socio-behavioral, economic, housing, and environmental characteristics was obtained from Hawaii Behavioral Risk factors Surveillance System and Healthy Hawaii Survey Programs, conducted in 2005. Principal component analysis techniques were employed to reduce the total number of significant area socioeconomic status indicators (n = 18), and to develop the composite area socioeconomic deprivation index for each administrative district as a single predictor of asthma hospital admissions in children and adults. Poisson log-linear regression models were used to estimate the influence of each selected socio-behavioral, economic, housing, and environmental factors on children and adult asthma hospitalization rates. The GIS ArcView 9.2 program was utilized for geospatial data analysis and result representation. Results: The study results revealed that the number of inactive, adult smokers, overweight and obese people, as well as fast food and other unhealthy food entities in the area were significant predictors of asthma hospitalization rates for both children and adults. Developed composite area socioeconomic deprivation index was a significant predictor of children and adult hospitalization rates for asthma. Conclusion: The study results support previous empirical findings that severe exacerbations and hospital admissions for asthma are distributed unequally within specific geographical areas and among specific ethnic minority population groups. The socioeconomic deprivation index could be used to explain social, economic, cultural, and environmental disparities in health as a powerful tool to describe social inequalities and also to identify high risk geographical areas and population groups in order to develop more effective disease prevention and health promotion strategies for asthma.

Time Trend and Age-Period-Cohort Effects on Acute Myocardial Infarction Mortality in Korean Adults from 1988 to 2007

We examined time trend and age-period-cohort effects on acute myocardial infarction (AMI) mortality in Korean adults from 1988 to 2007. Annual AMI mortality data and population statistics from 1988 to 2007 were obtained from the STATISTICS KOREA website. Age adjusted mortality for four 5-yr calendar periods (1988-1992 to 2003-2007) was calculated by direct standardization using the Year 2000 WHO world standard population. A log-linear Poisson regression model was used to estimate age, period, and cohort effects on AMI mortality. In both genders, age-adjusted AMI mortality increased from period one (1988-1992) to period three (1998-2002) but decreased in period four (2003-2007). An exponential age effect was noted in both genders. The rate ratio of the cohort effect increased up to the 1943 birth cohort and decreased gradually thereafter, and the rate ratio of the period effect increased up to period three (1998-2002) and decreased thereafter. Our results suggest that AMI mortality in Korean adults has decreased since the period 1998-2002 and age, period, and cohort effects have influenced on AMI mortality.

Suicide Rates and Risk Factors among Korean Cancer Patients, 1993-2005

As the number of cancer survivors increases, suicide risk approaches that of the general population. We therefore investigated suicide rates and risk factors among Korean cancer patients. We observed 816,295 cancer patients for 3,007,294 person-years from 1993 to 2005 through a nationwide cancer registry. We calculated their sex- and age-standardized mortality ratios (SMR) and studied suicide risk factors using rate ratios (RR) based on a log-linear Poisson regression model. Compared with the Korean general population, the suicide rate among cancer patients was high [SMR, 2.00; 95% confidence interval (95% CI), 1.91-2.08]. The rates were highest in the year following the cancer diagnosis (SMR, 3.45; 95% CI, 3.19-3.73) and were still elevated 5 years later (SMR, 1.23; 95% CI, 1.12-1.36). The clinical groups at highest risk were male pancreas cancer patients (SMR, 6.01; 95% CI, 4.33-8.33) and female lung cancer patients (SMR, 3.55; 95% CI, 2.55-4.94). The sociodemographic groups at highest risk were those who had no spouse versus those who were married (RR, 1.50; 95% CI, 1.35-1.68), those who were not employed versus those who were (RR, 1.39; 95% CI, 1.26-1.54), and those who did not have high school education versus those who had (RR, 1.52; 95% CI, 1.30-1.79). Korean cancer patients are at increased risk of suicide. Both clinical and sociodemographic factors play a role. There is a need for social support and suicide prevention strategies for cancer survivors in Korea.

Association of highly active antiretroviral therapy coverage, population viral load, and yearly new HIV diagnoses in British Columbia, Canada: a population-based study

Results of cohort studies and mathematical models have suggested that increased coverage with highly active antiretroviral therapy (HAART) could reduce HIV transmission. We aimed to estimate the association between plasma HIV-1 viral load, HAART coverage, and number of new cases of HIV in the population of a Canadian province. We undertook a population-based study of HAART coverage and HIV transmission in British Columbia, Canada. Data for number of HIV tests done and new HIV diagnoses were obtained from the British Columbia Centre for Disease Control. Data for viral load, CD4 cell count, and HAART use were extracted from the British Columbia Centre for Excellence in HIV/AIDS population-based registries. We modelled trends of new HIV-positive tests and number of individuals on HAART using generalised additive models. Poisson log-linear regression models were used to estimate the association between new HIV diagnoses and viral load, year, and number of individuals on HAART. Between 1996 and 2009, the number of individuals actively receiving HAART increased from 837 to 5413 (547% increase; p=0.002), and the number of new HIV diagnoses fell from 702 to 338 per year (52% decrease; p=0.001). The overall correlation between number of individuals on HAART and number of individuals newly testing positive for HIV per year was -0.89 (p<0.0001). For every 100 additional individuals on HAART, the number of new HIV cases decreased by a factor of 0.97 (95% CI 0.96-0.98), and per 1 log(10) decrease in viral load, the number of new HIV cases decreased by a factor of 0.86 (0.75-0.98). We have shown a strong population-level association between increasing HAART coverage, decreased viral load, and decreased number of new HIV diagnoses per year. Our results support the proposed secondary benefit of HAART used within existing medical guidelines to reduce HIV transmission. Ministry of Health Services and Ministry of Healthy Living and Sport, Province of British Columbia; US National Institute on Drug Abuse; US National Institutes of Health; Canadian Institutes of Health Research.

Methods for Monitoring Trends in the Incidence of Foodborne Diseases: Foodborne Diseases Active Surveillance Network 1996–2008

Foodborne illnesses are a substantial health burden in the United States. The Foodborne Diseases Active Surveillance Network (FoodNet) is the principal foodborne disease component of the Centers for Disease Control and Prevention's Emerging Infections Program. FoodNet is a collaborative project among Centers for Disease Control and Prevention, Emerging Infections Program sites, the U.S. Department of Agriculture, and the U.S. Food and Drug Administration. One of FoodNet's main objectives is to monitor changes in the incidence of selected foodborne pathogens. In 1996, FoodNet began active, population-based surveillance for laboratory-diagnosed cases of Campylobacter, Listeria, Salmonella, Shiga toxin-producing E. coli O157, Shigella, Vibrio, and Yersinia infection. Surveillance for cases of Cryptosporidium and Cyclospora infection was added in 1997 and surveillance for non-O157 Shiga toxin-producing E. coli was added in 2000. From 1997 to 2008, the FoodNet surveillance population increased, primarily through the addition of new sites. The increase in the number of FoodNet sites and the size of the population under surveillance as well as the variation in the incidence of infections among sites posed challenges in the selection of the most appropriate method to monitor changes in incidence. To account for variation introduced by changes in population size, a main-effects, log-linear Poisson (negative binomial) regression model was adopted to estimate the magnitude of changes in the incidence of pathogens by comparing current year incidence to reference periods. The article explains how FoodNet uses the negative binomial model to examine changes in incidence over time, describes the reference periods used, explains the graphics used to display results, and discusses future directions in the analysis of trends over time.

Reduction in the rate of methicillin-resistant Staphylococcus aureus acquisition in surgical wards by rapid screening for colonization: a prospective, cross-over study

AbstractIdentification of patients colonized with methicillin-resistant Staphylococcus aureus (MRSA) and subsequent isolation and decolonization is pivotal to the control of cross infection in hospitals. The aim of this study was to establish if early identification of colonized patients using rapid methods alone reduces transmission. A prospective, cluster, two-period cross-over design was used. Seven surgical wards at a large hospital were allocated to two groups, and for the first 8 months four wards used rapid MRSA screening and three wards used a standard culture method. The groups were reversed for the second 8 months. Regardless of the method of detection, all patients were screened for nasal carriage on admission and then every 4 days. MRSA control measures remained constant. Results were analysed using a log linear Poisson regression model. A total of 12 682/13 952 patient ward episodes (PWE) were included in the study. Admission screening identified 453 (3.6%) MRSA-positive patient ward episodes, with a further 268 (2.2%) acquiring MRSA. After adjusting for other variables, rapid screening was shown to statistically reduce MRSA acquisition, with patients being 1.49 times (p 0.007) more likely to acquire MRSA in wards where they were screened using the culture method. Screening of surgical patients using rapid testing resulted in a statistically significant reduction in MRSA acquisition. This result was achieved in a routine surgical service with high bed occupancy and low availability of isolation rooms, making it applicable to the majority of health-care systems worldwide.

Description of cervical cancer mortality in Belgium using Bayesian age-period-cohort models

ObjectiveTo correct cervical cancer mortality rates for death cause certification problems in Belgium and to describe the corrected trends (1954-1997) using Bayesian models.MethodCervical cancer (cervix uteri (CVX), corpus uteri (CRP), not otherwise specified (NOS) uterus cancer and other very rare uterus cancer (OTH) mortality data were extracted from the WHO mortality database together with population data for Belgium and the Netherlands.Different ICD (International Classification of Diseases) were used over time for death cause certification. In the Netherlands, the proportion of not-otherwise specified uterine cancer deaths was small over large periods and therefore internal reallocation could be used to estimate the corrected rates cervical cancer mortality. In Belgium, the proportion of improperly defined uterus deaths was high. Therefore, the age-specific proportions of uterus cancer deaths that are probably of cervical origin for the Netherlands was applied to Belgian uterus cancer deaths to estimate the corrected number of cervix cancer deaths (corCVX).A Bayesian loglinear Poisson-regression model was performed to disentangle the separate effects of age, period and cohort.ResultsThe corrected age standardized mortality rate (ASMR) decreased regularly from 9.2/100 000 in the mid 1950s to 2.5/100,000 in the late 1990s. Inclusion of age, period and cohort into the models were required to obtain an adequate fit. Cervical cancer mortality increases with age, declines over calendar period and varied irregularly by cohort.ConclusionMortality increased with ageing and declined over time in most age-groups, but varied irregularly by birth cohort. In global, with some discrete exceptions, mortality decreased for successive generations up to the cohorts born in the 1930s. This decline stopped for cohorts born in the 1940s and thereafter. For the youngest cohorts, even a tendency of increasing risk of dying from cervical cancer could be observed, reflecting increased exposure to risk factors. The fact that this increase was limited for the youngest cohorts could be explained as an effect of screening.Bayesian modeling provided similar results compared to previously used classical Poisson models. However, Bayesian models are more robust for estimating rates when data are sparse (youngest age groups, most recent cohorts) and can be used to for predicting future trends.

A negative binomial model for time series of counts

We study generalized linear models for time series of counts, where serial dependence is introduced through a dependent latent process in the link function. Conditional on the covariates and the latent process, the observation is modelled by a negative binomial distribution. To estimate the regression coefficients, we maximize the pseudolikelihood that is based on a generalized linear model with the latent process suppressed. We show the consistency and asymptotic normality of the generalized linear model estimator when the latent process is a stationary strongly mixing process. We extend the asymptotic results to generalized linear models for time series, where the observation variable, conditional on covariates and a latent process, is assumed to have a distribution from a one-parameter exponential family. Thus, we unify in a common framework the results for Poisson log-linear regression models of Davis et al. (2000), negative binomial logit regression models and other similarly specified generalized linear models.

Support points of locally optimal designs for nonlinear models with two parameters

We propose a new approach for identifying the support points of a locally optimal design when the model is a nonlinear model. In contrast to the commonly used geometric approach, we use an approach based on algebraic tools. Considerations are restricted to models with two parameters, and the general results are applied to often used special cases, including logistic, probit, double exponential and double reciprocal models for binary data, a loglinear Poisson regression model for count data, and the Michaelis-Menten model. The approach, which is also of value for multi-stage experiments, works both with constrained and unconstrained design regions and is relatively easy to implement.

MTHFR polymorphism, dietary folate intake and breast cancer in young women: a case control study in Rio de Janeiro, Brazil

We have studied the effect of genetic polymorphisms in the DNA repair genes hOGG1, XRCC1, XRCC3, ERCC2 and the MTHFR gene in the folate metabolism on the frequencies of cells with chromosomal aberrations (CA), chromosome-type aberrations (CSA), chromatid-type aberrations (CTA), chromatid breaks (CTB) and chromatid gaps (CTG) scored in peripheral blood lymphocytes from 651 Norwegian subjects of Caucasian descendant. DNA was extracted from fixed cell suspensions. The log-linear Poisson regression model was used for the combined data which included age, smoking, occupational exposure and genotype for 449 subjects.Our results suggest that individuals carrying the hOGG1 326Cys or the XRCC1 399Gln allele have an increased risk of chromosomal damage, while individuals carrying the XRCC1 194Trp or the ERCC2 751Gln allele have a reduced risk regardless of smoking habits and age.Individuals carrying the XRCC1 280His allele had an increased risk of CSA which was only apparent in non-smokers. This was independent of age.A protective effect of the XRCC3 241Met allele was only found in the older age group in non-smokers for CA, CSA and CTA, and in smokers for CSA. In the youngest age group, the opposite effect was found, with an increased risk for CA, CTA and CTG in smokers. Carrying the MTHFR 222Val allele gave an increased risk for chromosome and chromatid-type aberrations for both non-smokers and smokers, especially for individuals in the older age group, and with variable results in the youngest age group. The variables included in the different regression models accounted, however, for only 4–10% of the variation. The frequency ratio for CTG was significantly higher than for CTA and CTB for only 7 of the 43 comparisons performed. Some of the gap frequencies diverge from the trend in the CA, CSA, CTA and CTB results.

Risks of breast, endometrial, and ovarian cancers after twin births

The concentrations of endogenous hormones differ between women with twin and singleton births, with a possible influence on the risk of cancer. We used the nationwide Swedish Family-Cancer Database, including 30,409 women with a twin birth, to examine the subsequent risks of breast, endometrial, and ovarian cancers. Relative risks (RRs) were calculated in a log-linear Poisson regression model of person-years as offset. Cancer data were retrieved from the Swedish Cancer Registry; a total of 1010, 210, and 174 women were diagnosed with breast, endometrial, and ovarian cancers respectively, after a twin birth. A significant decrease in the risk of breast cancer was noted among women with a twin birth compared with women with a singleton birth (RR 0.85, 95% confidence interval (CI) 0.74-0.98). The protective effects were observed throughout the intervals after last pregnancy and they were strongest shortly after the last pregnancy in women who delivered a twin birth before 30 years of age. Twin birth did not change the risk of endometrial cancer (1.08, 95% CI 0.79-1.47) but the RR was increased for women with the number of pregnancies > or =4 (1.39, 95% CI 1.11-1.76). The RR for ovarian cancer was 0.95 (95% CI 0.79-1.15). Our study showed that twin births significantly reduced the subsequent risk of breast cancer. However, the associations of twin births with endometrial and ovarian cancers were not substantial.

Prostate cancer mortality predictions for Ireland up to 2015

The objective of this analysis was to predict average world age-standardized mortality rates per person-years (100,000 person-years) and numbers of prostate cancer deaths in Ireland for the years 2005, 2010 and 2015. Poisson linear and log-linear regression models of Irish prostate cancer mortality data for the years 1950-2002 were used to predict trends for the whole population, for men aged 0-64 and 50-74 years. The prostate cancer world age-standardized mortality rate in 2015 is predicted to remain unchanged from the average recorded in 1998-2002, while, because of population growth, the number of deaths is predicted to increase. In persons under 65 years of age, the world age-standardized mortality rate is expected to increase, but the number of deaths in this age group is expected to nearly double between 2002 and 2015. Similarly, the world age-standardized mortality rate for men aged 50-74 years is predicted to rise with the number of deaths in this age group expected to increase sharply. The historical evidence predicts a small increase of age-adjusted prostate cancer mortality rates in Ireland and only in the age groups of 0-64 and 50-74 years in the next 10 years, along with a continuing marked increase in number of deaths due to demographic change.

Application of Poisson regression analysis for queueing system of the demand for medical and health service

This paper suggests that log-linear analysis can be used to predict the expected number of patients in a queueing system. A log-linear Poisson regression model has been developed to analyse the time series of count data. In finding a Poisson regression model, parameters are estimated and goodness-of-fit is utilized to carefully extract the best model to fit the count data. The marginal effect is the basis function which can be used in the Poisson regression analysis. As a result, it allows us to arrive at better predictions of welfare health services and rehabilitation decision making. Health care utilization is a clinical marker of the well being of patients. One of the important end points of the demand for medical and health services in a queueing system is to identify factors associated with increased health care utilization; particularly those factors related to welfare health services. In this paper, we propose a Poisson model in which the number of patients in a queueing system is modeled by Poisson regression analysis. To this end, it is of great importance to assess the relationship between welfare health services and average waiting time in a queueing system. For this purpose, data were obtained from a survey conducted by the Samutpakran Hospital in Thailand for the year 2005. This is a difficult task for several reasons. First, even in the case of constant demand levels over the day, statistical fluctuations in individual patient waiting times and the variability in the time needed by a provider to service patients can create long delays even when overall average steady state capacity is greater than average demand. Second, the magnitude of delays is a log-linear function of the demand for medical and health services level, and are thus impossible to predict without the use of a queueing model (1). Herbert C. Heien, William A. Baumanm and Mezbahur Rahman have studied the categorical data analysis. In general, these models are termed Rate Models or Possion Regression Models. The Possion Regression Models can easily be connected to the standard Possion Models. Then, the Possion Models are directly related to the Exponential Models by making conversation of rate per unit interval with the waiting time until the first occurrence (2). To explore the relationship between the average waiting time and the number of welfare patient visits, one could consider a log-linear model. The average waiting time could be included in the model as a linear predictor. Log-linear models are among the most frequently used models for count data Possion regression analysis, but they offer only limited flexbility in accommodating the maginal effect of the independent varible (3).

Influence of DNA repair gene polymorphisms of hOGG1, XRCC1, XRCC3, ERCC2 and the folate metabolism gene MTHFR on chromosomal aberration frequencies

We have studied the effect of genetic polymorphisms in the DNA repair genes hOGG1, XRCC1, XRCC3, ERCC2 and the MTHFR gene in the folate metabolism on the frequencies of cells with chromosomal aberrations (CA), chromosome-type aberrations (CSA), chromatid-type aberrations (CTA), chromatid breaks (CTB) and chromatid gaps (CTG) scored in peripheral blood lymphocytes from 651 Norwegian subjects of Caucasian descendant. DNA was extracted from fixed cell suspensions. The log-linear Poisson regression model was used for the combined data which included age, smoking, occupational exposure and genotype for 449 subjects. Our results suggest that individuals carrying the hOGG1 326Cys or the XRCC1 399Gln allele have an increased risk of chromosomal damage, while individuals carrying the XRCC1 194Trp or the ERCC2 751Gln allele have a reduced risk regardless of smoking habits and age. Individuals carrying the XRCC1 280His allele had an increased risk of CSA which was only apparent in non-smokers. This was independent of age. A protective effect of the XRCC3 241Met allele was only found in the older age group in non-smokers for CA, CSA and CTA, and in smokers for CSA. In the youngest age group, the opposite effect was found, with an increased risk for CA, CTA and CTG in smokers. Carrying the MTHFR 222Val allele gave an increased risk for chromosome and chromatid-type aberrations for both non-smokers and smokers, especially for individuals in the older age group, and with variable results in the youngest age group. The variables included in the different regression models accounted, however, for only 4–10% of the variation. The frequency ratio for CTG was significantly higher than for CTA and CTB for only 7 of the 43 comparisons performed. Some of the gap frequencies diverge from the trend in the CA, CSA, CTA and CTB results.

Immunosuppression, Hepatitis C Infection, and Acute Renal Failure in HIV-Infected Patients

Low CD4 cell counts predict HIV-related morbidity and mortality and may be associated with acute renal failure (ARF). To estimate the effect of CD4 cell count on the incidence rate (IR) of ARF in ambulatory HIV-infected patients with access to highly active antiretroviral therapy. Observational clinical cohort of HIV-infected patients recruited from a university-based infectious diseases clinic, between 2000 and 2002, and followed up until December 31, 2002. Poisson log-linear regression models were used to calculate ARF IRs, IR differences, and IR ratios. The mean age of the 705 study participants was 40 years, two thirds were male, and 61% were African American. Incidence rates of ARF were higher at lower CD4 cell counts and among patients who were coinfected with hepatitis C. Patients with hepatitis C coinfection who also had low CD4 cell counts had the highest adjusted IR of ARF. Immunosuppression and hepatitis C virus coinfection are associated with increased IRs of ARF in ambulatory HIV-1-infected patients.