Experiments with Infinite-Horizon, Policy-Gradient Estimation

In this paper, we present algorithms that perform gradient ascent of the average reward in a partially observable Markov decision process (POMDP). These algorithms are based on GPOMDP, an algorithm introduced in a companion paper (Baxter & Bartlett, this volume), which computes biased estimates of the performance gradient in POMDPs. The algorithm's chief advantages are that it uses only one free parameter beta, which has a natural interpretation in terms of bias-variance trade-off, it requires no knowledge of the underlying state, and it can be applied to infinite state, control and observation spaces. We show how the gradient estimates produced by GPOMDP can be used to perform gradient ascent, both with a traditional stochastic-gradient algorithm, and with an algorithm based on conjugate-gradients that utilizes gradient information to bracket maxima in line searches. Experimental results are presented illustrating both the theoretical results of (Baxter & Bartlett, this volume) on a toy problem, and practical aspects of the algorithms on a number of more realistic problems.

Nonequivalent nuclear location of immunoglobulin alleles in B lymphocytes

Individual B lymphocytes normally express immunoglobulin (Ig) proteins derived from single Ig heavy chain (H) and light chain (L) alleles. Allelic exclusion ensures monoallelic expression of Ig genes by each B cell to maintain single receptor specificity. Here we provide evidence that at later stages of B cell development, additional mechanisms may contribute to prioritizing expression of single IgH and IgL alleles. Fluorescent in situ hybridization analysis of primary splenic B cells isolated from normal and genetically manipulated mice showed that endogenous IgH, kappa and lambda alleles localized to different subnuclear environments after activation and had differential expression patterns. However, this differential recruitment and expression of Ig alleles was not typically seen among transformed B cell lines. These data raise the possibility that epigenetic factors help maintain the monoallelic expression of Ig.

A case-control study of physical activity and non-insulin dependent diabetes mellitus (NIDDM). the San Luis Valley Diabetes Study

PURPOSE

The purpose of this study was to examine the association between physical activity and non-insulin dependent diabetes mellitus (NIDDM).

METHODS

We conducted a population-based case-control study in Hispanic and non-Hispanic white men and women, ages 20-74. A total of 167 cases with NIDDM and 1100 controls with normal glucose tolerance were included. All subjects completed an oral glucose tolerance test.

RESULTS

Persons with recently diagnosed NIDDM reported significantly lower levels of physical activity than control subjects. For total metabolic units, the odds ratio for subjects in the highest tertile compared to those in the lowest tertile was 0.60 (95% confidence interval (CI) = 0.37-0.98) after adjusting for age, sex, ethnicity, and family history of diabetes. The adjusted odds ratio for persons reporting high levels of vigorous activity (at least three times per week for 20 minutes) was also less than 1, but was not statistically significant (odds ratio (OR) = 0.73, 95% CI = 0.47-1.14). Similar adjusted odds ratios were observed for high versus low levels of self-assessed work activity (OR = 0.50, 95% CI = 0.34-0.74) and leisure time physical activity (OR = 0.62, 95% CI = 0.44-0.90). Further adjustments for body mass index attenuated the strength of the association between physical activity and NIDDM. This is consistent with the hypothesis that obesity is one consequence of physical inactivity that puts individuals at increased risk for NIDDM. The association of physical activity and NIDDM was stronger in Hispanic than in non-Hispanic white subjects, although this difference was not statistically significant.

CONCLUSIONS

High levels of physical activity are associated with lower odds of NIDDM and this relationship may be stronger in Hispanic subjects.

Physician referral patterns to a breast cancer support program

PURPOSE

The purpose of this study was to describe the differences between physicians who did and did not refer their patients with breast cancer to the American Cancer Society's Reach to Recovery program, as well as to determine the factors associated with these physician referral patterns.

DESCRIPTION OF STUDY

This descriptive cross-sectional study included 54 physicians who had and 23 physicians who had not referred their patients with breast cancer to the Mile High Unit Reach to Recovery program in 1999. Participating physicians completed a questionnaire about their knowledge, behaviors, and beliefs regarding Reach to Recovery, the characteristics of the patients they referred, and their own demographic and practice characteristics.

RESULTS

General surgeons were more likely than all other specialties to refer patients to the Reach to Recovery program. The program is a part of the American Cancer Society. Feedback from patients about their experiences with Reach to Recovery and the proportion of patient time spent caring for patients with breast cancer were all positively associated with referral.

CLINICAL IMPLICATIONS

These findings show that those providing support programs, such as Reach to Recovery, need to use targeted strategies to ensure that physicians are aware of the services they provide and that patients relay their experiences back to their referring physician. Because the value of social support programs for women with breast cancer has been established, physicians can become valuable partners in helping more women gain access to these programs.

Depressive symptoms in Hispanic and non-Hispanic White rural elderly: the San Luis Valley Health and Aging Study

Literature on depression in rural and Hispanic elderly adults is sparse. This report describes the prevalence of depressive symptoms in 1,151 community-dwelling, Hispanic and non-Hispanic White participants in the San Luis Valley Health and Aging Study, conducted in rural Colorado during 1993-1995. The prevalence and odds ratios of high depressive symptoms, defined as a Center for Epidemiologic Studies Depression Scale score of > or = 16, were calculated. The crude prevalence of high depressive symptoms was 11.4% (95% confidence interval: 9.6, 13.6). Female gender, chronic diseases, dissatisfaction with social support, living alone, and lower income and education were associated with depressive symptoms. There were no ethnic differences in the men. The age-adjusted odds ratio of depressive symptoms in Hispanic women compared with that of non-Hispanic White women was 2.11 (95% confidence interval: 1.32, 3.38). After adjustment for multiple sociodemographic and health risk factors, the odds ratio in Hispanic women was 2.12 (95% confidence interval: 1.19, 3.80). Higher depressive symptoms in Hispanic women varied by acculturation level. The odds ratio in the high acculturation stratum was 1.56 (95% confidence interval: 0.75, 3.27) and in the low acculturation stratum was 2.51 (95% confidence interval: 1.11, 5.70). A lower acculturation level may increase the risk for depression in older Hispanic women.

Effect of iron(III) chitosan intake on the reduction of serum phosphorus in rats

Because of the widespread use of aluminium- and calcium-containing phosphate binders for the control of hyperphosphataemia in patients with end-stage renal failure, an iron(III) chitosan complex was synthesised and fed to rats to measure its effect on serum phosphorus and calcium, intestinal phosphate binding and phosphate absorption. Thirty-six Wistar rats were randomly selected and distributed into a baseline group (n = 6), a control group (n = 8 (days 0-15), n = 8 (days 16-30)) and a treatment group (n = 8 (days 0-15), n = 8 (days 16-30)). The control groups ingested AIN-76 diet mix with a 1% w/w fibre content; however, the treatment groups had the fibre content completely substituted with iron(III) chitosan. The mean weights of the treated rats were slightly lower from 15 days (not significant); but overall, rat growth was not stunted in the treatment groups. The serum phosphorus levels of the treated group (n = 8) were significantly reduced after 15 days (P = 0.004; control: 5.7+/-0.9 mg dL(-1); treatment: 4.4+/-0.5 mg dL(-1); 95% CI of difference: 0.5-2.2) and 30 days (P = 0.002; control: 5.5+/-0.9 mg dL(-1); treatment = 4.1+/-06 mg dL(-1); 95% CI of difference: 0.6-2.3) as compared with the respective control group. The serum calcium-phosphorus product was 62.0+/-12.1 mg2 dL(-2) for the control and 45.1+/-6.6 mg2 dL(-2) for the treatment group after 30 days (P = 0.004). The serum iron concentration of the treatment group did not differ from the baseline value after 15 and 30 days, but the treatment group was significantly higher than the control group (P<0.05) after 30 days. The faeces phosphorus levels (mg day(-1)) were higher (P<0.01) and its iron content was much higher (P<0.01) for the treated group. The urine phosphorus (mg kg(-1)) was not significantly reduced for the treated group, but the mean was consistently less. The kidney and liver weights of both groups were similar, but the phosphorus content of the kidney (mg (g kidney)(-1)) was higher for the treated group after 30 days (P = 0.041; control, 4.2+/-1.2 mg g(-1) vs treatment, 5.6+/-1.4 mg g(-1). Because iron(III) chitosan had a high phosphorus-binding capacity of 308 (mg P) per gram of Fe3+ for both the in-vitro (pH 7.5) and in-vivo studies, which is greater than nearly all commonly used phosphate binders, and a small net phosphorus absorption difference of 3.7 mg day(-1), it is an efficient phosphate binder for lowering serum phosphate levels without increasing serum calcium levels.

The relation between baseline HIV drug resistance and response to antiretroviral therapy: re-analysis of retrospective and prospective studies using a standardized data analysis plan

To assess the relation between resistance to antiretroviral drugs for treatment of HIV-1 infection and virological response to therapy, results from 12 different studies were re-analysed according to a standard data analysis plan. These studies included nine clinical trials and three observational cohorts. The primary end-point in our analyses was virological failure by week 24. Baseline factors that were investigated as predictors of virological failure were plasma HIV-1 RNA, the number and type of new antiretroviral drugs in the regimen, and viral susceptibility to the drugs in the regimen, determined by genotyping or phenotyping methods. These analyses confirmed the importance of both genotypic and phenotypic drug resistance as predictors of virological failure, whether these factors were analysed separately or adjusted for other baseline confounding factors. In most of the re-analysed studies, the odds of virological failure were reduced by about twofold for each additional drug in the regimen to which the patient's virus was sensitive by genotyping methods, and by about two- to threefold for each additional drug that was sensitive by phenotyping.

A Model of Inductive Bias Learning

A major problem in machine learning is that of inductive bias: how to choose a learner's hypothesis space so that it is large enough to contain a solution to the problem being learnt, yet small enough to ensure reliable generalization from reasonably-sized training sets. Typically such bias is supplied by hand through the skill and insights of experts. In this paper a model for automatically learning bias is investigated. The central assumption of the model is that the learner is embedded within an environment of related learning tasks. Within such an environment the learner can sample from multiple tasks, and hence it can search for a hypothesis space that contains good solutions to many of the problems in the environment. Under certain restrictions on the set of all hypothesis spaces available to the learner, we show that a hypothesis space that performs well on a sufficiently large number of training tasks will also perform well when learning novel tasks in the same environment. Explicit bounds are also derived demonstrating that learning multiple tasks within an environment of related tasks can potentially give much better generalization than learning a single task.

A new class of selective and potent inhibitors of neuronal nitric oxide synthase

The synthesis and SAR of a series of 6-(4-(substituted)phenyl)-2-aminopyridines as inhibitors of nitric oxide synthase are described. Compound 3a from this series shows potent and selective inhibition of the human nNOS isoform, with pharmacokinetics sufficient to provide in vivo inhibition of nNOS activity.

Cognitive functioning and impairment among rural elderly Hispanics and non-Hispanic whites as assessed by the Mini-Mental State Examination

We conducted a cross-sectional study to determine the distribution of cognitive functioning as measured by the Mini-Mental State Examination (MMSE) among a sample of Hispanic and non-Hispanic White (NHW) residents from two counties in rural, southern Colorado. Residents aged 60 years and older (N = 1,360) were administered the full MMSE, a sociodemographic and medical interview. Protocols were developed to administer the MMSE equitably in both ethnic groups. Younger Hispanics tended to be categorized as severely impaired more than similarly aged NHWs (OR at age 70 = 4.14), however, older Hispanics and NHWs performed similarly after adjusting for education and gender (OR at age 90 = 1.00). The use of a modified MMSE scale that removed the ethnic bias demonstrated that NHWs and Hispanics had similar levels of severe impairment after full adjustment (OR = 0.93). Given the widespread use of the MMSE, these findings indicate the need for further validation of this instrument.

Methods and prevalence of ADL limitations in Hispanic and non-Hispanic white subjects in rural Colorado: the San Luis Valley Health and Aging Study

PURPOSE

The Hispanic population in the United States is the fastest growing minority group, yet there is little understanding of the disability patterns that occur as this population ages. We conducted a cross-sectional study to define the prevalence of limitations of activities of daily living (ADL) and measures of observed function.

METHODS

We censussed two rural counties in southern Colorado and selected a stratified sample of both Hispanic and non-Hispanic white (NHW) residents; 81.6% completed the protocol.

RESULTS

Among the 1250 subjects aged 65 years and older, Hispanic elderly living in the community had greater ADL disability than NHW subjects, both for any difficulty (p = 0.006), and for needing assistance (p = 0.002). Hispanic persons were less likely to reside in nursing homes (3.4%) compared with NHW persons (9.3%). Hispanic elderly had excess prevalence of dependent ADL tasks (needs assistance or unable to do), (age, gender-adjusted odds ratio = 1.39, 95% CI = 1.01-1.92) in community dwelling and nursing home residents combined. There was no Hispanic excess of less severe difficulty compared with NHW persons, and there was a similar prevalence of limitation on observed functional tasks (timed walk, stooping, rising from a chair) in both groups.

CONCLUSIONS

There was a modest Hispanic excess of reported dependent ADL limitation, and no excess of observed functional difficulties. Hispanics enter older age with much less income and education, yet they do not have a marked excess prevalence of limitations in activities of daily living when compared with NHW persons living in the same area.

Indicators of nutritional risk in a rural elderly Hispanic and non-Hispanic white population: San Luis Valley Health and Aging Study

OBJECTIVE

To describe the prevalence of nutritional risk factors among elderly residents in a rural Hispanic and non-Hispanic white population.

DESIGN

A geographically based survey of community-dwelling elderly adults.

SUBJECTS/SETTING

From July 1993 to July 1995, all Hispanic persons older than 65 years and an age-stratified, random sample of 69% of non-Hispanic white persons, from 2 Colorado counties, were invited to participate in a study of functional impairment and disability (81% responded). After exclusion of 184 respondents who required a surrogate respondent and 8 with missing diet data, the study consisted of 1,006 subjects. Interviews included questions similar to the Nutrition screening Initiative checklist, a 21-item food frequency questionnaire, and anthropometric measures.

STATISTICAL ANALYSES PERFORMED

Gender- and ethnicity-specific, age-adjusted prevalence for each risk factor was estimated by use of logistic regression.

RESULTS

Hispanic participants were more likely than non-Hispanic whites to report inadequate intake of vegetables, problems with teeth or dentures that limited the kinds and amounts of food eaten, difficulty preparing meals, and lack of money needed to buy food. Hispanic women reported nutritional risk factors more often than Hispanic men, although anthropometric markers indicated that Hispanic men may be at higher risk of nutritional deficiency.

APPLICATIONS/CONCLUSIONS

Hispanic men and women had a higher prevalence of nutritional risk factors than non-Hispanic whites. Intervention programs targeting rural, elderly Hispanics should aim to ensure that basic nutrition needs (access to food, help preparing meals, and adequate dental care) are being met. Community programs to increase activity levels and consumption of nutrient-dense foods are recommended.

Dynamic repositioning of genes in the nucleus of lymphocytes preparing for cell division

We show that several transcriptionally inactive genes localize to centromeric heterochromatin in the nucleus of cycling but not quiescent (noncycling) primary B lymphocytes. In quiescent cells, centromeric repositioning of inactive loci was induced after mitogenic stimulation. A dynamic repositioning of selected genes was also observed in developing T cells. Rag and TdT loci were shown to relocate to centromeric domains following heritable gene silencing in primary CD4+8+ thymocytes, but not in a phenotypically similar cell line in which silencing occurred but was not heritable. Collectively, these data indicate that the spatial organization of genes in cycling and noncycling lymphocytes is different and that locus repositioning may be a feature of heritable gene silencing.

The association of non-insulin-dependent diabetes mellitus with perceived quality of life in a biethnic population: the San Luis Valley Diabetes Study

OBJECTIVES

This study evaluated the association between quality of life and non-insulin-dependent diabetes mellitus (NIDDM) status, and whether this association differs between Hispanics and non-Hispanic Whites.

METHODS

Between 1986 and 1989, cross-sectional data on perceived quality of life (PQOL) were collected from 223 persons with NIDDM and 753 non-diabetic subjects.

RESULTS

After adjustment, persons with NIDDM rated their PQOL significantly lower than did control subjects. The relationship of diabetes and PQOL did not differ by ethnicity. The number of complications of diabetes was not associated with lower PQOL scores.

CONCLUSIONS

Control and treatment strategies should reflect an understanding of the impact that diabetes has on social functioning, leisure activities, and physical and mental health.

Social network factors associated with perceived quality of life. The San Luis Valley Health and Aging Study

Demographic and social network factors associations with perceived quality of life (PQOL) were examined in a sample of rural Hispanic and non-Hispanic White (NHW) elderly. A 20-item PQOL scale measured participants' satisfaction with multiple facets of their lives. Age, gender, marital status, and living alone were not associated with this outcome. Hispanic persons with higher incomes had consistently higher ratings of quality of life compared to NHWs. More education was associated with a higher PQOL rating only among NHWs. More close friends and relatives, visiting in homes, and participation in outside activities were all independently related to PQOL. Interaction tests showed that the impact of social networks was stronger in NHWs. These findings suggest that network size and contact are important social factors that can improve quality of life for both ethnic groups. However, differences may need to be examined when judging the magnitude of these associations and when planning interventions.

Higher instrumental activities of daily living disability in Hispanics compared with non-Hispanic whites in rural Colorado. The San Luis Valley Health and Aging Study

This study examined Hispanic versus non-Hispanic white patterns of needing assistance with instrumental activities of daily living (IADL). The authors interviewed 798 Hispanic and 614 non-Hispanic white residents of rural Colorado, who were aged 60 years and older between 1993 and 1995. Seventy-five participants were nursing home residents at the time of the interview. Community-dwelling Hispanics were 1.6 times as likely as non-Hispanic whites to need assistance with at least one IADL task (95% confidence interval 1.25-2.13). A larger proportion of disabled non-Hispanic whites were in nursing homes but, after including nursing home residents, Hispanics remained significantly more likely to need assistance on at least one IADL task (odds ratio = 1.49, 95% confidence interval 1.16-1.93). Hispanics were also more likely to have difficulty on observed performance tasks. The Hispanic excess was not removed by adjusting for chronic disease, reported difficulty walking, or income. English language proficiency adjustment lowered the Hispanic excess, but adjusting for years of education or Mini-Mental State Examination scores more completely removed the ethnic differences. Higher education was protective for both Hispanic and non-Hispanic white elderly. Efforts to further investigate what facets or correlates of education are operating may offer useful insights into limiting IADL difficulties in future cohorts.

Executive cognitive abilities and functional status among community-dwelling older persons in the San Luis Valley Health and Aging Study

OBJECTIVES

The purpose of this study was to evaluate the contribution of the executive cognitive functions to self-reported and observed performance of activities of daily living and instrumental activities of daily living.

DESIGN

These data were collected as part of a cross-sectional survey of community-dwelling older persons, using statistical sampling, in a two-county area of southern Colorado.

SETTING

Participants were interviewed and administered measures of general cognition, depression, executive functioning, and performance of self-care and instrumental activities, either in their homes or at the study clinic.

PARTICIPANTS

A total of 1158 community-dwelling persons between the ages of 60 and 99 participated. Of these, 657 were female, 501 were male, 637 were Hispanic, and 521 were non-Hispanic whites.

MEASUREMENTS

Subjects were administered the Mini-Mental State Exam (MMSE), a measure of executive functioning (the Behavioral Dyscontrol Scale), the Center for Epidemiologic Studies Depression scale (CES-D), and the Structured Assessment of Independent Living Skills (SAILS, a measure of observed performance of daily tasks). They also were interviewed regarding their reported ADL and IADL status using measures from the Longitudinal Study on Aging.

RESULTS

Both general mental status and executive functioning demonstrated statistically significant univariate associations with all seven functional status measures (both self-report and observed performance). In a series of ordinary least squares regression models, executive functioning was a predictor for self-reported ADLs and observed performance of complex IADL tasks such as managing money and medications. Mental status did not predict self-reported functioning but was a predictor of observed performance. Depression was a significant variable for self-report measures but not for observed performance. Executive functioning and general mental status demonstrated some degree of independence from one another.

CONCLUSION

Executive functioning is an important determinant of functional status for both self-reported and observed ADLs and IADLs and should be assessed routinely when evaluating the mental status and functional abilities of older people. These results replicate and extend previous research on executive functioning among older persons by examining these factors in a large bi-ethnic community sample.

A randomized, controlled trial of octreotide in the management of patients with acute pancreatitis

CONCLUSION

In this study, octreotide was not found to be of benefit in the treatment of acute pancreatitis.

BACKGROUND

Somatostatin and its long-acting analog octreotide have both been proposed for the treatment of patients with acute pancreatitis, but neither agent has been adequately assessed in patients with severe attacks.

METHODS

Fifty-eight patients with moderate or severe acute pancreatitis who were admitted to hospitals within the west of Scotland over an 18-mo period were randomized to receive octreotide, 40 micrograms/h, by continuous i.v. infusion for 5 d, or placebo in addition to standard supportive therapy.

RESULTS

There was no significant difference in the incidence of complications (54% octreotide group and 40% placebo group) or mortality (octreotide group 18%; placebo group 20%).

Self-rated health among Hispanic vs non-Hispanic white adults: the San Luis Valley Health and Aging Study

OBJECTIVES

This study investigated whether objective health indicators explained lower self-rated health among Hispanics compared with non-Hispanic Whites. It also considered socioeconomic and cultural explanations.

METHODS

Health ratings of 429 Hispanics and 583 non-Hispanic Whites aged 20 through 74 were analyzed with logistic regression.

RESULTS

Illness indicators were found to be strongly correlated with self-rated health in both ethnic groups, but after such markers were controlled for, Hispanics remained 3.6 times more likely to report fair or poor health (95% confidence interval = 2.4, 5.3). Adjustment for socioeconomic factors accounted for a portion of Hispanics' lower health rating, but the strongest explanatory factor was acculturation.

CONCLUSIONS

Because of cultural and economic influences on definitions of health, ethnic differences in self-assessed health may not accurately reflected patterns resulting from objective health measurements.