The Role of Remote Diagnostics to Better Assess Uroflow Variability: Insights From Combining at Home Uroflows and Frequency Volume Charts From 19,868 Voids Using a Novel, Hand Held, Cellular Embedded Device.

OBJECTIVE: To examine how representative 24-hour data collection is of the overall patient experience utilizing a home uroflow device in men with benign prostatic hyperplasia (BPH). MATERIALS AND METHODS: Home uroflow data were collected with the iO Urology CarePath device from men at a single urology clinic and retrospectively analyzed. Void characteristics were summarized for data collected over 24 hours compared to data collected over several days or weeks with the device (overall study excludes the 24-hour data). A linear mixed-effects model was used to evaluate differences in average Qmax from voids collected during a single 24-hour period compared to the overall study. RESULTS: A total of 486 men (mean age 67.4) used the device resulting in 465 included in the analysis with a total of 15,521 voids in the overall study and 4347 voids in the single-day analysis. Average Qmax from the model was 11.2 mL/s (95% CI: 10.80, 11.65) and 11.2 mL/s (95% CI: 10.81, 11.64) in the 24-hour and overall study groups, respectively. Both groups had similar between (20.3% vs 20.4%) and within-subject (12.0% vs 11.9%) variance. The difference in Qmax was not significantly different (P = .970). A subgroup analysis including voids >150 mL yielded similar results. CONCLUSION: Data show variability in voiding parameters, but similar average Qmax collected in 24 hours as compared to several days/weeks. The combination of home uroflowmetry with frequency-volume chart information can provide objective data and a unique perspective on uroflow variability (UFV) for clinician review to support the development of individualized patient treatment plans.

Diabetes and peripheral arterial disease in men: trends in prevalence, mortality, and effect of concomitant coronary disease.

BACKGROUND: Recent data on trends in diabetes mellitus (DM) prevalence and long-term effect on mortality in peripheral arterial disease (PAD) subjects is lacking. METHODS: All subjects discharged from any VA medical center between October 1990 to September 1997 with an International Classification of Diseases (ICD)-9 code for PAD and DM in the discharge summary were retrospectively identified. Demographic data were extracted from the database. Mortality data were obtained from the Beneficiary Information and Resource Locator. Outcome measures were age specific DM prevalence over time, and short-term and long-term mortality. RESULTS: Of 33, 629 patients with PAD, 9474 (29%) had DM. Diabetes mellitus subjects were less likely to be white and had more comorbidities. Mean length of hospital stay was greater for DM (22.3 d vs 18.7 days, P < 0.001). Mortality was higher for DM at 180 days (9.8% vs 8.4%, P < 0.001), 1 year (16.4% vs 13.7%, P < 0.001), and continues to increase at 8 years of follow-up. Logistic regression analysis showed no interaction between DM and coronary artery disease (CAD). CONCLUSIONS: Diabetes mellitus increases all-cause mortality in subjects with PAD starting at 6 months post-discharge and continues to be higher even at 8 years of follow-up. There was a lack of interaction of DM and CAD on mortality in this cohort of subjects with PAD.

Prevalence rates for dementia and Alzheimer's disease in African Americans: 1992 versus 2001.

BACKGROUND: This study compares age-specific and overall prevalence rates for dementia and Alzheimer's disease (AD) in two nonoverlapping, population-based cohorts of elderly African Americans in Indianapolis in 2001 and 1992. METHODS: We used a two-stage design. The first stage involves the Community Screening Interview for Dementia (CSI-D). The CSI-D scores are grouped into good, intermediate, and poor performance before selection for clinical assessment. Diagnoses were performed using standard criteria in a consensus diagnosis conference; clinicians were blind to performance groups. In 1992, interviewers visited randomly sampled addresses to enroll self-identified African Americans aged > or =65 years. Of 2582 eligible, 2212 enrolled (9.6% refused, and 4.7% were too sick). In 2001, Medicare rolls were used for African Americans aged >70 years. Of 4260 eligible, 1892 (44%) enrolled, 1999 (47%) refused, and the remainder did not participate for other reasons. RESULTS: The overall age-adjusted prevalence rate for dementia at age > or =70 years in 2001 was 7.45% (95 confidence interval [CI], 4.27-10.64), and in the 1992 cohort, this prevalence rate was 6.75% (95% CI, 5.77-7.74). The overall age-adjusted prevalence rate at age > or =70 years for AD in the 2001 cohort was 6.77% (95% CI, 3.65-9.90), and for the 1992 cohort, it was 5.47% (95% CI, 4.51-6.42). Rates for dementia and AD were not significantly different in the two cohorts (dementia, P = .3534; AD, P = .2649). CONCLUSIONS: We found no differences in the prevalence rates of dementia and AD between 1992 and 2001, despite significant differences in medical history and medical treatment within these population-based cohorts of African American elderly.

Selenium level is associated with apoE epsilon4 in rural elderly Chinese.

OBJECTIVE: Se is an essential trace element in human nutrition associated with antioxidant activity. Previous studies on predictors of toenail Se or serum Se have mostly concentrated on demographic factors such as age and gender. The present paper examines the association between apoE genotype and Se levels in nail samples in a rural elderly Chinese cohort. DESIGN: Two thousand Chinese aged 65 years and over from four counties in China were enrolled in a cohort to study the association of Se with cognitive decline. Nail samples were collected from each participant and analysed for Se levels. Dietary Se intake was estimated from an FFQ using Se contents measured in food items collected from each village. Blood samples on filter cards were collected and analysed for apoE genotype. Mixed-effect models were constructed with nail Se level as the dependent variable and each village as the random effect, which controlled for the potential confounding effect from correlation in Se measures obtained from participants residing in the same village. RESULTS: In this elderly Chinese cohort, carriers of the apoE epsilon4 allele had significantly lower Se levels measured in nail samples than non-carriers after adjusting for other significant covariates and controlling for estimated dietary Se intake. There was no significant difference between the two genotypes on estimated Se dietary intake (P = 0.6451). CONCLUSIONS: Future studies are needed to examine the mechanism underlying the association between the apoE epsilon4 allele and Se levels, including the role of oxidative stress and that of reduced lipid metabolism in the apoE epsilon4 carriers.

Long term postischemic stroke mortality in diabetes: a veteran cohort analysis.

BACKGROUND AND PURPOSE: Recent data on stroke mortality in diabetics in the United States is lacking. We investigated trends in diabetes prevalence and stroke morality among diabetics in a large veteran cohort. METHODS: The Patient Treatment File was used to identify all patients discharged from any Veterans hospital between October 1990 and September 1997 with a diagnosis of ischemic stroke (ICD-9-CM codes 434, 436) listed as primary diagnosis. Demographic, morbidity, and mortality data were recorded. Chi-square tests were used to examine differences between diabetics and nondiabetics, and t tests were used for continuous variables. Cox proportional hazards regression was used to examine the effects of diabetes (DM) on the survival times controlling for multiple covariates. RESULTS: Of 48 733 ischemic stroke patients identified, 98% were male and 13 925 (25%) had DM. Mean age was similar between DM and non-DM (67.2 versus 67.5, P=NS). Prevalence of DM among stroke subjects increased from 25% to 31%. Charlson index >2 was much higher in DM (68.2% versus 47.9%, P<0.001). Mortality at 60 days and 1 year was similar in both groups (2.9 versus 2.7%, P=NS; 12.6 versus 13.1, P=NS). Kaplan-Meier survival plot showed that DM had shorter long term survival time (log-rank, P<0.001). Multivariate Cox proportional hazards regression showed a higher risk of death for diabetics (HR=1.15, 95% CI 1.11 to 1.19, P<0.001). CONCLUSIONS: Despite greater comorbidity, postacute ischemic stroke mortality at 60 days and 1 year is not different between subjects with and without DM. Long term mortality after stroke is much lower among DM than that reported in older studies.

Trace element levels and cognitive function in rural elderly Chinese.

BACKGROUND: Trace elements are involved in metabolic processes and oxidation-reduction reactions in the central nervous system and could have a possible effect on cognitive function. The relationship between trace elements measured in individual biological samples and cognitive function in an elderly population had not been investigated extensively. METHODS: The participant population is part of a large cohort study of 2000 rural elderly Chinese persons. Six cognitive assessment tests were used to evaluate cognitive function in this population, and a composite score was created to represent global cognitive function. Trace element levels of aluminum, calcium, cadmium, copper, iron, lead, and zinc were analyzed in plasma samples of 188 individuals who were randomly selected and consented to donating fasting blood. Analysis of covariance models were used to assess the association between each trace element and the composite cognitive score adjusting for demographics, medical history of chronic diseases, and the apolipoprotein E (APOE) genotype. RESULTS: Three trace elements-calcium, cadmium, and copper-were found to be significantly related to the composite cognitive score. Increasing plasma calcium level was associated with higher cognitive score (p <.0001). Increasing cadmium and copper, in contrast, were significantly associated with lower composite score (p =.0044 and p =.0121, respectively). Other trace elements did not show significant association with the composite cognitive score. CONCLUSIONS: Our results suggest that calcium, cadmium, and copper may be associated with cognitive function in the elderly population.

Factorial invariance found in survey instrument measuring arthritis-related health beliefs among African-Americans and Whites.

OBJECTIVE: Reliable and valid instruments for measuring health beliefs related to arthritis and total joint arthroplasty (TJA) are lacking. The present study examined the factorial invariance of an Arthritis-related Health Belief Instrument (AHBI) that measures four constructs: severity of arthritis, susceptibility of arthritis to get worse, and perceived benefits and barriers of TJA. STUDY DESIGN AND SETTING: In an urban primary care clinic setting, African-American (AA) and White (W) patients (n=664) with at least a moderately severe degree of osteoarthritis completed the AHBI. For the cross-sectional data on the first 120 enrolled participants, exploratory factor analysis was performed to derive the initial factor structures. For the remaining 544 participants, single-group and multi-group confirmatory factor analyses (CFA) were performed to assess factorial invariance. RESULTS: The CFA analyses revealed that a four-factor model was invariant across race. CONCLUSION: The findings of invariance across race support the use of the AHBI to measure ethnic group differences in health beliefs as they relate to arthritis and TJA. Future analyses should continue to assess other aspects of reliability and validity of the AHBI.

A Solution to Separation and Multicollinearity in Multiple Logistic Regression.

In dementia screening tests, item selection for shortening an existing screening test can be achieved using multiple logistic regression. However, maximum likelihood estimates for such logistic regression models often experience serious bias or even non-existence because of separation and multicollinearity problems resulting from a large number of highly correlated items. Firth (1993, Biometrika, 80(1), 27-38) proposed a penalized likelihood estimator for generalized linear models and it was shown to reduce bias and the non-existence problems. The ridge regression has been used in logistic regression to stabilize the estimates in cases of multicollinearity. However, neither solves the problems for each other. In this paper, we propose a double penalized maximum likelihood estimator combining Firth's penalized likelihood equation with a ridge parameter. We present a simulation study evaluating the empirical performance of the double penalized likelihood estimator in small to moderate sample sizes. We demonstrate the proposed approach using a current screening data from a community-based dementia study.

Impact of pain on the outcomes of depression treatment: results from the RESPECT trial.

OBJECTIVE: Pain is prevalent in patients with depression. The purpose of this study was to determine the impact of pain on depression treatment outcomes. METHODS: Data was analyzed from a randomized controlled trial comparing a collaborative care intervention to usual care for the treatment of depression in 60 primary care practices. A total of 405 patients with either current major depressive disorder or dysthymia were enrolled, and assessed at baseline, 3, and 6 months. Main measures included the 20-item Hopkins Symptom Check List (HSCL-20) depression score, and the SF-36 pain interference score. RESULTS: Pain severe enough to produce at least moderate interference with daily activities was present in 42% of depressed patients at baseline. Pain outcomes did not differ between intervention and control groups but improved similarly in both over time. However, pain was still at least moderately severe in 32% of patients at 6 months. Both baseline pain and the amount of pain improvement over time were associated with depression remission and response rates. In a multivariate model controlling for age, gender, and medical co-morbidity, depression severity increased with higher pain interference and decreased with the passage of time (p<.0001 for both). There was also a significant pain by time by treatment group interaction (p=.027). The beneficial effects of collaborative care on depression outcome persisted (p=.049) even after controlling for pain interference, time, covariates, and interaction effects. CONCLUSIONS: Pain has a strong negative impact on the response of depression to treatment. Recognizing and optimizing the management of comorbid pain that commonly coexists with depression may be important in enhancing depression response and remission rates.

Renal disease in an antiretroviral-naïve HIV-infected outpatient population in Western Kenya.

BACKGROUND: Several commonly used antiretrovirals (ARVs) require dose adjustments to prevent toxicities in the presence of renal insufficiency. Because no prospective studies of the prevalence or risk factors for kidney disease in stable outpatient human immunodeficiency virus (HIV)-infected indigenous African populations have been published to date, it is not known if already scarce resources should be allocated to detect renal dysfunction, in those without risk factors for kidney disease, prior to initiation of increasingly available antiretrovirals in developing countries. METHODS: A cross-sectional study to determine the prevalence of and risk factors for renal disease in a cohort of medically stable, HIV-infected, antiretroviral-naïve adults, without diabetes or hypertension, presenting to an HIV clinic in western Kenya. RESULTS: Of 373 patients with complete data, renal insufficiency (CrCl <60 ml/min) was identified in 43 (11.5%) [18 (4.8%) had a CrCl <50 ml/min]. Despite high correlation coefficients between the three renal function estimating equations used, when compared to creatinine clearance as calculated by Cockcroft-Gault, lower rates of moderate to severe renal insufficiency were identified by the Modification of Diet in Renal Disease equations. Proteinuria, defined as a urine dipstick protein of equal to or greater than 1+, was detected in only 23 subjects (6.2%). CONCLUSIONS: Renal insufficiency is not uncommon, even in stable patients without diabetes or hypertension. Conversely, proteinuria was unexpectedly infrequent in this population. Utilizing resources to assess renal function prior to initiation of antiretrovirals in order to identify those likely to benefit from dosage adjustment is justified.

Stroke mortality and race: does access to care influence outcomes?

BACKGROUND: Stroke is the third leading cause of death in the United States. We investigated racial differences in death after hospital discharge for ischemic stroke in a large cohort of Veterans Health Affairs (VHA) stroke patients. We hypothesized that having access to VA care would ameliorate the excess stroke mortality rates in African-Americans (AA) reported in non-VA studies. METHODS: Hospital administrative data were used to identify all patients discharged from any VA hospital between October 1990 and September 1997 with a primary discharge diagnosis of ischemic stroke (ICD-9-CM codes 434 and 436). We obtained demographic data and clinical data recorded during the index hospitalization and after discharge, including deaths, from VA clinical and administrative databases. The Charlson comorbidity index was constructed for each patient from the index admission's discharge diagnoses. Patients were followed through 1998. RESULTS: Of 55,094 VHA stroke patients discharged after ischemic strokes, 34,579 (63%) were white and 11,530 (21%) were AA. Charlson index was similar between the groups. One-year mortality rate was significantly higher for whites: Adjusting for demographic and clinical differences, being white remained predictive of higher mortality rates (multivariable hazard ratio, 1.06; 95% CI, 1.02 to 1.10). From Kaplan-Meier estimates, the probability that whites would survive for 1 year was 0.86 compared with 0.87 for AA. CONCLUSIONS: Despite having similar severity of illness and adjusting for other clinical differences, mortality rate was marginally lower in AA after being discharged from VA hospitals after ischemic strokes. This is contrary to prior reports from non-VA hospitals and suggests the possibility of access to care playing a role in stroke deaths.

Selenium level and cognitive function in rural elderly Chinese.

Selenium is a trace element associated with antioxidant activity and is considered to be a protective agent against free radicals through enhanced enzyme activity. Studies on selenium and cognitive function or Alzheimer's disease have yielded inconsistent results. A cross-sectional survey of 2,000 rural Chinese aged 65 years or older from two provinces in the People's Republic of China was conducted from December 2003 to May 2005 by use of the Community Screening Instrument for Dementia, the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) Word List Learning Test, the Indiana University Story Recall Test, the Animal Fluency Test, and the Indiana University Token Test. Over 70% of the study participants have lived in the same village since birth. Nail samples were collected and analyzed for selenium contents. Analysis-of-covariance models were used to estimate the association between quintile selenium levels measured in nail samples and cognitive test scores, with adjustment for other covariates. Lower selenium levels measured in nail samples were significantly associated with lower cognitive scores (p < 0.0087 for all tests) except the Animal Fluency Test (p = 0.4378). A dose-response effect of selenium quintiles was also seen for those significant associations. Results in this geographically stable cohort support the hypothesis that a lifelong low selenium level is associated with lower cognitive function.

Declining stroke burden among white and black male veteran subjects.

The prevalence of hypertension and diabetes is increasing. We hypothesized that this could contribute to increasing burden of stroke-the third leading cause of mortality-and investigated national trends in stroke burden among blacks and whites. From October 1990 to October 1997, 55,094 veterans were admitted with diagnosis of ischemic stroke (International Classification of Diseases [ICD] code 434 or 436) at any veterans hospital in the country were included in the study. We extracted demographic data from Veterans' Administration administrative databases. Discharge ICD-9 codes were used to assess stroke risk factors. Patients were classified as white or black, and prevalence of stroke was calculated and stratified according to age. Overall annual incidence rate fell from 2.7 per 1,000 in 1991 to 1.67 per 1,000 in 1997. Annual incidence number of stroke declined steadily in both the white and black groups from 1991 to 1997. Incidence numbers fell for all ages with greater decrease in younger age groups. Our data indicate that the burden of stroke among male veterans is decreasing. This appears to be true for both blacks and whites and across all age groups.

Primary care physician treatment of low HDL: Rational approach or Pandora's Box?

BACKGROUND: Guidelines for treating high low-density lipoproteins are clear, whereas guidelines for treating low high-density lipoproteins (HDL) are less so. Physicians approach to treating low HDL cholesterol is not known. OBJECTIVE: To determine primary care physicians approach to managing low HDL. METHODS: Three-thousand, nine-hundred and nineteen surveys were mailed to all primary care physicians in the State of Indiana, asking questions regarding demographics, case studies to assess the provider's approach to managing low HDL, and direct questions regarding management of HDL levels and general lipid knowledge questions. RESULTS: Seven-hundred and eighty-one surveys were returned, for a response rate of 19.9%. Fifty-eight percent of participants would initiate HDL-raising therapy after achieving the appropriate low-density lipoprotein cholesterol goal. The approaches used to raise HDL included lifestyle changes (diet, exercise, smoking cessation) (85%), niacin (83%), fibric acid derivative (61%), and alcohol (31%). Reasons inhibiting initiating therapy for raising HDL included concern over side effects (56%), perceived lack of effectiveness of currently available drugs (24%), lack of clear guidelines (22%), and lack of evidenced-based trials demonstrating benefit of raising HDL (14%). For men, 40% of physicians use 40 mg/dL as a cutoff for initiating HDL-raising therapy, while 25% using a cutoff of 35 mg/dL. For women, 24% use a cutoff of 50 mg/dL for initiating HDL-raising therapy, while 12% use 45 mg/dL as a cutoff. CONCLUSIONS: The majority of primary care physicians in the State of Indiana treated low HDL with appropriate approaches, although use of alcohol to raise HDL raises concerns.

Youth With Epilepsy: Development of a Model of Children's Attitudes Toward Their Condition.

A model of children's attitudes toward their epilepsy was tested in 173 children (9-14 years) with epilepsy and their parents. Predictor variables tested were child characteristics, family mastery, child worry, child self-efficacy for seizure management, child psychosocial care needs, and seizure variables. Data were analyzed using structural equation modeling, leading to a revised model in which less child worry, greater family mastery, and greater child seizure self-efficacy were directly related to more child positive attitudes. Discussion focuses on potential targets for psychosocial interventions aimed at improving attitudes toward epilepsy.

Validation analysis of informant's ratings of cognitive function in African Americans and Nigerians.

OBJECTIVES: To examine informant validity using the Community Screening Interview for Dementia (CSI 'D') both cross-sectionally and longitudinally in two very different cultures and to explore the effects of informants and study participants' characteristics on the validity of informants' reports. METHODS: Elderly African Americans age 65 years and older residing in Indianapolis, USA and elderly Yoruba Nigerians age 65 years and older residing in Ibadan, Nigeria were assessed on cognitive functioning using the CSI 'D' at baseline (1992-1993) and five-year follow-up (1997-1998). At baseline, the informant validity in both samples was evaluated against participants' cognitive tests using Pearson correlation and regular regression models. At follow-up, informants ratings on cognitive decline were assessed against participants' cognitive decline scores from baseline to follow-up using biserial correlation and logistic regressions. RESULTS: At baseline, informants' reports on cognitive functioning significantly correlated with cognitive scores in both samples (Indianapolis:r = -0.43, p < 0.001; Ibadan:r = -0.47, p < 0.001). The participant-informant relationships significantly affected the informants' reports in the two samples with different patterns (p = 0.005 for Indianapolis and p < 0.001 for Ibadan) at a given level of cognitive functioning. African Americans spouses reported more cognitive problems, while siblings reported more problems for the Yoruba Nigerians. At follow-up, informants' ratings on cognitive decline significantly correlated with the cognitive decline scores (Indianapolis r = 0.38, p < 0.001; Ibadan r = 0.32, p < 0.001). The characteristics of study participants and informants had little impact on the informants' ratings on cognitive decline. CONCLUSIONS: Informant reports are valid in assessing the cognitive functioning of study participants both cross-sectionally and longitudinally in two very different cultures, languages and environments.

Neuropsychological predictors of academic underachievement in pediatric epilepsy: moderating roles of demographic, seizure, and psychosocial variables.

PURPOSE: Academic underachievement is common in pediatric epilepsy. Attempts to identify seizure and psychosocial risk factors for underachievement have yielded inconsistent findings, raising the possibility that seizure and psychosocial variables play a complex role in combination with other variables such as neuropsychological functioning. This study cross-validated a neuropsychological measurement model for childhood epilepsy, examined the relation between neuropsychological functioning and academic achievement, and tested the degree to which demographic, seizure, and psychosocial variables moderate that relation. METHODS: Children with chronic epilepsy (N = 173; ages 8 to 15 years; 49% girls; 91% white/non-Hispanic; 79% one seizure type; 79% taking one medication; 69% with active seizures) completed a comprehensive neuropsychological battery. Children diagnosed with mental retardation were excluded. RESULTS: Structural equation modeling identified a three-factor measurement model of neuropsychological function: Verbal/Memory/Executive (VME), Rapid Naming/Working Memory (RN/WM), and Psychomotor (PM). VME and RN/WM were strongly related to reading, math, and writing; PM predicted writing only. Family environment moderated the impact of neuropsychological deficits on writing (p < or = 0.01) and possibly for reading (p = 0.05); neuropsychological deficits had a smaller impact on achievement for children in supportive/organized homes compared with children in unsupportive/disorganized homes. CONCLUSIONS: These findings lend partial support for our theoretical model showing direct effects of neuropsychological function on achievement and the moderating role of family factors. This study suggests that a subgroup of children with epilepsy (those who have not only neuropsychological deficits but also disorganized/unsupportive home environments) are particularly at risk for adverse academic outcomes. Implications for intervention are discussed.

Measuring stigma in children with epilepsy and their parents: instrument development and testing.

PURPOSE: The goal of this work is to describe psychometric properties of two scales measuring perceived stigma in children with epilepsy and their parents. METHODS: Data were collected for the parent scale in two samples: parents of 173 children with epilepsy and of 224 children with new-onset seizures. The child scale was tested in the chronic sample. Content validity, internal consistency reliability, and construct validity were tested. RESULTS: Both scales had strong internal consistency reliability and construct validity. Higher scores were associated with greater seizure severity scores. In the parent scale, lower scores were associated with more positive mood, less worry, and more family leisure activities. In the child scale, higher scores were correlated with more negative attitude, greater worry, poorer self-concept, and more depression symptoms. CONCLUSIONS: Both scales were found to have strong psychometric properties. They are short, and items are easy to understand. These scales have potential for use in research and in the clinical setting to measure stigma.

Selecting pre-screening items for early intervention trials of dementia--a case study.

Our goal was to review and extend statistical methods for discriminating between normal subjects and those with dementia or cognitive impairment. We compared six different methods to one constructed by expert opinion, in their brevity and predictive power. The methods include logistic regression and neural networks, with standard and least absolute shrinkage and selection operator (LASSO) variable selection, as well as decision trees with and without boosting. These methods were applied to the baseline data of a subgroup of subjects in a dementia study, using their screening interview items to predict their clinical diagnosis of normal or non-normal (cognitively impaired or demented). The derived models were then validated on a different subgroup of subjects in the same study who had the screening and clinical diagnosis two to five years later. Performance of different models was compared based on their sensitivity and specificity in the validation sample. Generally, the six statistical methods performed slightly to moderately better than the expert-opinion model. Neural networks generally performed better than the logistic and decision tree models. LASSO improved the performance of logistic and neural network models, but it eliminated few input variables in the neural network. The single decision tree performed at least as well as the standard logistic model, and with fewer items, making it an attractive pre-screening tool. Using the boosting option for decision trees did not substantially improve the performance. We recommend that for each situation, different methods of classification should be attempted to obtain optimal results for a given purpose.

A latent variable model approach for assembling and scoring screening tests for dementia.

In dementia studies, the diagnosis of dementia often relies on results of screening tests aimed at measuring various dimensions of cognitive functions. The current practice of scoring a screening test involves simply summing the correct responses from each item. However, this method may be imprecise and inefficient in the predictive power of the score for dementia. We propose a latent variable model approach for the scoring and item selection of such tests. We model the item responses to be random variables based on latent variables. We also model the disease outcomes to be a function of the latent variables. Maximum likelihood estimates are obtained by maximizing the joint likelihood functions of disease and the item responses over a specified distribution function for the latent variables. Variances of model parameters are estimated using a nonparametric bootstrap method. We illustrate the approach using a screening test for dementia from a community-based study.