Multivariable Poisson Models Research Articles

Public news announcements and quoting activity in the Euro/Dollar foreign exchange market

The effect of public news announcements on dealers’ quoting activity is analyzed with the multivariate double autoregressive conditional Poisson model. Quoting activity is measured by the frequency of price revisions in the Euro/Dollar foreign exchange market. The multivariate double autoregressive conditional Poisson model is designed for time series of count data. It is based on the double Poisson distribution, which can be both over- and underdispersed. The main findings are first a significant interaction between dealers’ quoting activity, which confirms hot potato trading. Second, news announcements have a different impact on the quoting activity of different banks. Third, impulse–response functions to news announcements show the dynamic nature of the reaction to these news releases.

Joint analysis of mixed Poisson and continuous longitudinal data with nonignorable missing values

Regression models are proposed for joint analysis of Poisson and continuous longitudinal data with nonignorable missing values under fully parametric framework. Our primary interest is to evaluate the influence of the covariates on both Poisson and continuous responses. First, we form the full likelihood with complete data using the multivariate Poisson model and conditional multivariate normal distribution and then construct an ECM algorithm to find the maximum likelihood estimates of the model parameters. Then, under the assumption that the missingness mechanisms for the two responses are independent but nonignorable, namely, dependent on both observed and missing data of the two responses, we choose the logit model for the missingness mechanisms and selection model for the full likelihood. Also, we build two implementations of the Monte Carlo EM algorithm for estimating the parameters in the model. Wald test is employed to test the significance of covariates. Finally, we present the results of the Monte Carlo simulation to evaluate the performance of the proposed methodology and an application to the interstitial cystitis data base (ICDB) cohort study. To the best of our knowledge, our model is the first parametric model for joint analysis of Poisson and continuous longitudinal data with nonignorable missing value.

Antibiotic Use in Premature Infants After Discharge From the Neonatal Intensive Care Unit

Using a retrospective cohort of premature infants, we constructed multivariable Poisson models to determine factors associated with the receipt of antibiotics during the first year after discharge, N = 891. Black race (incidence rate ratio 1.80 compared with White infants, P = .008), male gender (incidence rate ratio 1.44; P = .007), bronchopulmonary dysplasia (incidence rate ratio 1.47; P = .04), and each additional child at home (incidence rate ratio 1.21, P = .002) increased the receipt of antibiotics for any reason. Male gender and additional children at home increased the receipt of non-recommended antibiotics, while Black infants received care at facilities that prescribed more non-recommended antibiotics. Even in a high-risk population of children, factors other than the medical history and presentation of the child may alter antibiotic prescription patterns and result in variations in care.

A Multivariate Poisson Model of Consumer Choice in a Multi-Airport Region

Using the results of a unique telephone survey the frequency of consumer flights from airports in a multi-airport region are modeled using a multivariate Poisson framework, the parameters of which were estimated using a latent variable application of the expectation maximization algorithm. This offers a different perspective since other work on airport choice uses the results of airport intercept surveys that capture only a single choice per respondent, whereas the data from the phone survey is count data for the airports in the study. An airport’s own-distance had the expected negative impact on mean usage of the airport, although the cross effects were somewhat mixed. Ticket price differences between airports were not always statistically significant. Mean usage was found to be increasing in income for PHL, but was decreasing for the other airports, reflecting the increasing value of respondents’ time as their income rises. If the destination of flights is domestic (international) then the result is to increase usage of PHL, BWI and EWR (JFK). Except for JFK, if the purpose of travel is mostly pleasure then it results in more travel from JFK and less from the other three airports. The availability of a low cost carrier would result in more frequent travel.

Pairwise likelihood estimation for multivariate mixed Poisson models generated by Gamma intensities

Estimating the parameters of multivariate mixed Poisson models is an important problem in image processing applications, especially for active imaging or astronomy. The classical maximum likelihood approach cannot be used for these models since the corresponding masses cannot be expressed in a simple closed form. This paper studies a maximum pairwise likelihood approach to estimate the parameters of multivariate mixed Poisson models when the mixing distribution is a multivariate Gamma distribution. The consistency and asymptotic normality of this estimator are derived. Simulations conducted on synthetic data illustrate these results and show that the proposed estimator outperforms classical estimators based on the method of moments. An application to change detection in low-flux images is also investigated.

Serum Phosphate Levels and Risk of Infection in Incident Dialysis Patients

Hyperphosphatemia is highly prevalent in dialysis patients and may be associated with immune dysfunction. The association of serum phosphate level with infection remains largely unexamined. In an incident cohort of 1010 dialysis patients enrolled from 1995 to 1998 and treated in 80 US clinics, the association of phosphate level (low <3.5; normal 3.5 to 5.5; high >5.5 mg/dl) at baseline and during follow-up with the risk for incident inpatient and outpatient infection-related events was examined. Infectious events were identified from US Renal Data System data (mean follow-up 3.3 yr). Incidence rate ratios for all infections, sepsis, respiratory tract infections, and osteomyelitis were obtained using multivariable Poisson models, adjusting for potential confounders (age, race, gender, smoking, comorbidity, and laboratory values). Infections of any type (n = 1398) were more frequent among patients with high phosphate levels at baseline, relative to normal; this association was not changed by adjustment for parathyroid hormone level. Similarly, high versus normal baseline phosphate was associated with increased risk for sepsis and osteomyelitis but not respiratory tract infections. Associations with calcium were generally NS, and results with calcium-phosphate product mirrored the phosphate results. High phosphate levels may be associated with increased risk for infection, contributing further to the rationale for aggressive management of hyperphosphatemia in dialysis patients.

Mortality attributable to cholestatic liver disease in the United States†

In the past 2 decades, important advances have been made in the treatment of cholestatic liver diseases, including primary biliary cirrhosis (PBC) and primary sclerosing cholangitis (PSC). Whether these new therapies have had demonstrable impact on mortality on a population-wide scale has not been evaluated. This study describes the age-specific and sex-specific mortality rates from PBC and PSC in the United States between 1980 and 1998, based on the Multiple Cause of Death files. Age-specific and sex- specific mortality rates from PBC and PSC were calculated. The multivariable Poisson model was used to evaluate temporal changes in mortality rates. In 1998, the total age-adjusted and sex-adjusted PBC-related mortality rate was 0.24 per 100,000, and the age-adjusted and sex-adjusted PSC-related mortality rate was 0.23 per 100,000. During the observation period, PBC-related mortality significantly decreased over time in women younger than 65 years, and in men of all age groups, whereas in older women this number increased over time. PSC-related mortality remained essentially stable, except in men 65 years of age or older. Since the early 1980s, significant changes in mortality from PBC have occurred. The most noticeable change was an increase in the age of death, which indicates prolongation of survival. These changes may be attributable to liver transplantation or ursodeoxycholic acid. In contrast, mortality from PSC remained largely unchanged, highlighting the need for more effective therapeutic strategies.

Udder health at a Swedish research farm with both organic and conventional dairy cow management

Our aim was to compare udder health in groups of organically and conventionally managed cows, using data from a longitudinal study in a Swedish dairy-research farm. Management of the groups was identical except for feed composition and the feeding regimen. Our dataset included all lactating cows calving from 1 September 1990 to 31 August 2001 (145 organically and 151 conventionally managed cows). Udder health was assessed by the geometric average somatic-cell count (SCC) within 150 days after calving, by the number of monthly SCC tests >200,000 cells/ml within 150 days after calving and by presence of lactations with veterinary-treated cases of clinical mastitis. The effect of animal group was analysed by multivariable linear, Poisson and logistic-regression models, controlling for factors such as lactation number, breed, year, season and milk yield. The groups did not differ in any measure of udder health. We had power to rule out differences of at least 33,000 cells/ml in the geometric average somatic-cell count, an incidence rate ratio of 0.65 in the incidence of high-SCC milk-testing occasions, and an odds ratio of 0.43 in veterinary treated cases of mastitis.

Multivariate autoregressive modeling of time series count data using copulas

We introduce the Multivariate Autoregressive Conditional Double Poisson model to deal with discreteness, overdispersion and both auto and cross-correlation, arising with multivariate counts. We model counts with a double Poisson and assume that conditionally on past observations the means follow a Vector Autoregression. We resort to copulas to introduce contemporaneous correlation. We apply it to the study of sector and stock-specific news related to the comovements in the number of trades per unit of time of the most important US department stores traded on the NYSE. We show that the market leaders inside a specific sector are related to their size measured by their market capitalization.

On Models for Binomial Data with Random Numbers of Trials

A binomial outcome is a count s of the number of successes out of the total number of independent trials n=s+f, where f is a count of the failures. The n are random variables not fixed by design in many studies. Joint modeling of (s, f) can provide additional insight into the science and into the probability pi of success that cannot be directly incorporated by the logistic regression model. Observations where n= 0 are excluded from the binomial analysis yet may be important to understanding how pi is influenced by covariates. Correlation between s and f may exist and be of direct interest. We propose Bayesian multivariate Poisson models for the bivariate response (s, f), correlated through random effects. We extend our models to the analysis of longitudinal and multivariate longitudinal binomial outcomes. Our methodology was motivated by two disparate examples, one from teratology and one from an HIV tertiary intervention study.

Impact of a mass immunization campaign to cotrol an outbreak of severe respiratory infections in Nunavik, northern Canada

Objectives. In spring 2002, a mass immunization campaign using a 23-valent pneumococcal polysaccharide vaccine (23-PPV) was launched in order to control an outbreak of severe pneumonia caused by a virulent clone of serotype 1 Streptococcus pneumoniae in Nunavik, Quebec, Canada. The objective of the study was to evaluate the impact of this campaign on the incidence of invasive pneumococcal disease (IPD) and hospitalizations possibly associated with pneumococcal infections (HPAPI) in the mostly Inuit population aged 10 to 64 years.Study design. Retrospective analysis of surveillance and administrative data.Methods. Multivariate Poisson model comparing the frequency rates of selected outcomes before the outbreak, during the outbreak, and after implementation of the mass immunization program.Results. The reported incidence of serotype 1 IPD decreased markedly after the implementation of the vaccination campaign (rate ratio = 0.16; p & 0.002). The frequency of HPAPI and the mean duration of hospital stay also decreased. However, vaccine failures were documented and the HPAPI rate remained higher than in the period prior to the outbreak.Conclusions. Although 23-PPV contributed to control the outbreak, better vaccines are needed for the prevention of infections caused by serotype 1 S. pneumoniae.(Int J Circumpolar Health 2006; 65(4): 297–304).

Factors associated with preweaning mortality on commercial pig farms in England and Wales

Data from 67 pig farms with a variety of farrowing systems were used to identify factors associated with preweaning mortality in British pig herds. The median mortality reported by the...

Multivariate Poisson regression with covariance structure

In recent years the applications of multivariate Poisson models have increased, mainly because of the gradual increase in computer performance. The multivariate Poisson model used in practice is ba...

Mortality and Hospital Utilization for Hepatocellular Carcinoma in the United States

The incidence of hepatocellular carcinoma (HCC) has been increasing in the United States. Although resource-intensive treatment modalities have been increasingly applied, these patients still have poor survival. We examined 2 nationally representative databases, the Multiple Cause of Death file and the Nationwide Inpatient Sample database, to examine trends in mortality and hospital service utilization related to HCC. In both databases, a priori criteria were used to identify cases of HCC. All other available diagnostic fields were examined to characterize coexistent liver disease. Age-, sex-, and race-specific mortality from HCC was calculated, and temporal changes in mortality rates were evaluated using the multivariable Poisson model. Hospital service utilization was estimated based on length of stay, total hospitalization charges, and principal procedures. The age-, sex-, and race-specific mortality from HCC increased from 1.54 to 2.58 per 100,000 per year between 1980 and 1998. Male sex, African and Asian race, and increasing age were also associated with higher mortality. The estimated total charge for HCC hospitalizations nationwide increased from 241 million US dollars in 1988 to 509 million US dollars in 2000 after inflation adjustment. Commonly employed procedures in 2000 included angiography/embolization, resection, local ablative therapy, and liver transplantation. In the recent past, mortality and hospital service utilization related to HCC increased substantially. Closer epidemiologic surveillance to understand causation of HCC at the population level and to help implement primary and secondary prevention is urgently warranted.

A low lymphocyte percentage is a predictor of mortality and hospitalization in hemodialysis patients

Lymphocyte percentage (LYM%), an independently measured value to reflect peripheral lymphocyte count and a possible nutritional marker, may be related to clinical outcome in maintenance dialysis (MHD) patients. We examined the associations of the baseline white blood cell count (WBC) and LYM% with 12-month mortality and three measures of hospitalization in a cohort of 1,283 MHD patients from 10 outpatient DaVita dialysis clinics in Los Angeles County, as well as in a subcohort of 372 MHD patients with additional measures of inflammation, nutrition and comorbidity. Multi-variate Cox and Poisson models that included 13 co-variates including case-mix features, dialysis dose, blood hemoglobin and serum albumin were explored. Patients, aged 57.8 +/- 15.2 years, included 49% men and 49% diabetics. Baseline WBC was 7,353 +/- 2.427 per microl, and LYM% was 21.2 +/- 7.3%. LYM% had significant correlations with "malnutrition-inflammation score" and inverse correlations with serum interleukin-6. The WBC and LYM% had significant but opposite predicting values for mortality and hospitalization, indicating that a high WBC and a low LYM% were each independently associated with increased mortality. After dividing each variable into four quartiles, only the highest WBC quartile (> or = 8,500) but not the other middle two quartiles, predicted increased mortality. However, all three lower quartiles of LYM% vs. the highest quartile (based on quartile cutoffs of 16%, 20.3% and 25.5%) were significantly and progressively associated with greater risks of mortality and hospitalizations. The absolute lymphocyte count (LYM% times WBC/100) exhibited somewhat similar trends but its outcome predictability was not as strong as LYM%. A high WBC and a low LYM% are associated with significant increase in mortality and hospitalization in MHD patients. Lymphocyte percentage, compared to absolute lymphocyte count, appears to be a better nutritional and anti-inflammatory marker and a more sensitive predictor of mortality and hospitalization in MHD patients.

A multivariate Poisson mixture model for marketing applications

This paper describes a multivariate Poisson mixture model for clustering supermarket shoppers based on their purchase frequency in a set of product categories. The multivariate nature of the model accounts for cross‐selling effects between the purchases made in different product categories. However, for computational reasons, most multivariate approaches limit the covariance structure by including just one common interaction term, or by not including any covariance at all. Although this reduces the number of parameters significantly, it is often too simplistic as typically multiple interactions exist on different levels. This paper proposes a theoretically more complete variance/covariance structure of the multivariate Poisson model, based on domain knowledge or preliminary statistical analysis of significant purchase interaction effects in the data. Consequently, the model does not contain more parameters than necessary, whilst still accounting for the existing covariance in the data. Practically, retail category managers can use the model to devise customized merchandising strategies.

The Foreign Exchange Quoting Activity as an Informative Signal

We investigate the informative property of quoting activity, measured by the frequency of price revisions, in the Euro/Dollar foreign exchange market. We use the multivariate double autoregressive conditional Poisson model, designed for time series of count data. We find that dealers' quoting activity reacts to both news announcements and quoting activity of other dealers. Some dealers analyse quoting activity of others to infer useful information like their different reactions to public news announcements. Some of them increase their activity, whilst others decrease it in response to the same news. We attribute this to the heterogeneous interpretation of the news content by individual traders and to the significant influence of some dealers on others.

A low serum iron level is a predictor of poor outcome in hemodialysis patients

Background: Iron administration has been implicated as a cause of poor clinical outcome in maintenance hemodialysis (MHD) patients. However, the role of low iron levels in the clinical outcome of MHD patients is not clear. Methods: We examined the predicting value of baseline serum iron level on prospective mortality and hospitalization in a cohort of all 1,283 MHD patients from 10 DaVita dialysis facilities in Los Angeles County, CA. Results: Patients aged 57.8 ± 15.2 years included 49% men, 45% Hispanics, 25% African Americans, and 53% patients with diabetes. During the first 3 months of the cohort, 97% of patients were administered erythropoietin (EPO) and 60% were administered intravenous iron (gluconate and/or dextran) at least once. During a 12-month follow-up, mortality was significantly greater (23%) in the lowest serum iron quartile (<45.3 μg/dL [<8.1 μmol/L]) compared with other quartiles (10% to 12%). Multivariate Poisson and Cox models adjusted for demographic features, dialysis dose and vintage, serum albumin and ferritin and blood hemoglobin concentrations, and administered EPO and iron doses showed that both serum iron level and iron saturation ratio had significant, but inverse, associations with prospective mortality and hospitalization. There was a statistically significant trend toward greater rates of mortality and hospitalization with lower serum iron levels. This reverse association remained significant in a subcohort of 322 MHD patients after additional adjustments for comorbid conditions and serum C-reactive protein level to reflect inflammation. Conclusion: Low baseline serum iron indicators are associated with increased mortality and hospitalization in MHD patients independent of hemoglobin level, EPO and iron doses, indicators of nutrition and inflammation, and comorbid conditions. Clinical trials to examine the role of iron administration in improving morbidity and mortality by increasing serum iron levels in MHD patients are required.

Parathyroidectomy rates among United States dialysis patients: 1990–1999

Medical therapy for secondary hyperparathyroidism (SHPTH) has evolved considerably during the past decade. It is not known how changes in medical therapy might impact the parathyroidectomy (PTX) rate among dialysis patients. Relatively low parathyroid hormone (PTH) levels have been found among elderly dialysis patients and those with diabetes. Clinical factors associated with differing PTX rates among United States dialysis patients have not been reported. We report PTX rates in the United States from 1990 to 1999 among persons with end-stage renal disease, accounting for changes in patient characteristics. Data from the United States Renal Database were utilized. Patients insured by Medicare or Medicaid and receiving renal replacement therapy between January 1, 1990, and December 31, 1999 were considered for analysis. PTX was determined by ICD-9 procedure codes. Multivariate Poisson models were used to estimate adjusted PTX rates. The overall observed PTX rate in the study sample was 7.16 per 1000 person-years at risk. After a slight rise during the early 1990s, adjusted PTX rates declined by approximately 30% between 1995 and 1999. Adjusted PTX rates were higher among patients who were younger, female, nondiabetic, receiving peritoneal dialysis, and those with a longer cumulative duration of dialysis. PTX rates have recently decreased in the United States, independent of changes in patient characteristics. The effectiveness of medical therapy in targeting secondary hyperparathyroidism may be improving. Younger, nondiabetic patients with a longer cumulative dialysis burden are at particularly high risk for PTX.

Physical activity and ischemic stroke risk. The atherosclerosis risk in communities study.

The relationship between physical activity and stroke is inconclusive according to the 1996 US Surgeon General's Report on Physical Activity and Health. Therefore, this study examined the relationship between physical activity and ischemic stroke risk among 14 575 Atherosclerosis Risk in Communities Study participants aged 45 to 64 years free of self-reported stroke and coronary heart disease at baseline. Eligible potential stroke hospitalizations were identified from ongoing hospital surveillance and from hospitalizations reported by the cohort study participants. All strokes were validated by hospitalization records. Physical activity was measured as sport, leisure (nonsport), and work with the use of the Baecke questionnaire. Multivariable Poisson and Cox proportional hazards models were used to determine the association of differing levels of physical activity with ischemic stroke incidence. During an average of 7.2 years of follow-up, 189 incident ischemic strokes occurred. Ischemic stroke incidence rates were highest in the lowest quartile of sport, leisure, and work scores. The hazard rate ratios with 95% CIs for ischemic stroke for the highest quartile compared with the lowest quartile of activity adjusted for age, sex, race-center, education, and smoking, were sport 0.83 (0.52, 1.32), leisure 0.89 (0.57, 1.37), and work 0.69 (0.47, 1.00). Further adjustment for factors that likely were intermediate variables (hypertension, diabetes, fibrinogen, and body mass index) between physical activity and stroke attenuated the associations. Our findings suggest that physical activity was weakly associated with a reduced risk of ischemic stroke among middle-aged adults. The association may be due to links between physical activity and other risk factors or due to chance.