|| Evaluation of Triple Touch Instructor Program: An Assessment of Program Delivery
|| Stunting, Diarrheal Disease and Parasitic Infection in Peruvian Infants: Their Effect on Cognitive Test Scores in Late Childhood
||Screening for Domestic Violence in Emergency Departments: Policy vs. Practice
||Sources of Advice During Pregnancy: A Quantitative and Qualitative Study Focusing on Low-Income Women
||The Epidemiology of Prematurity at Harare Maternity Hospital , Zimbabwe : The Contribution of HIV Infection.
||5-Year Survival of Women Treated With Breast Conserving Surgery and Radiation Therapy Compared to Women Receiving a Total Mastectomy
|| Abortion Needs and Experiences of Women in Rural Uttarakhand, India
|| Objective Versus Self-Report Diary Measures of Physical Activity: A Comparison of Methods Among Middle-Aged Women
|| Characteristics and Birth Outcomes: Multiparous Women Receiving No Prenatal Care in the United States, 1992-1997
||Use of Receiver Operating Characteristic Curves in Evaluating Logistic Regression Model
||Gonadal Development in the Japanese Medaka (Oryzias latipes) as a Model for Estrogenic Potency Determination
||The Relationship of Hookworm Infection to Anemia Among Pre-School Children in Rural Malawi
|| HIV-related Needs Assessment of Upstate SC: Our Community Believes What?!?
Evaluation of Triple Touch Instructor Program: An Assessment of Program Delivery
Lauren J. Bailey, University of South Florida
Other Author: Jeannine Coreil
Triple Touch is a breast health education program developed and launched by the Florida Division of the American Cancer Society in 1994. The name Triple Touch represents the three components of a good breast health plan including breast self-examination, clinical breast examination, and mammography, all of which make up the core content of the educational curriculum. Triple Touch is a popular program in Florida . It reaches more women each year than any other breast cancer program sponsored by the Florida Division including Reach to Recovery and has the largest group of volunteers. However, Division staff and members of the Triple Touch Work Group have posed questions regarding the level of activity among instructors, and how to increase instructor involvement and improve the program for future development and implementation. The study design used to evaluate the Triple Touch instructor program consisted of a cross-sectional telephone survey. An independent research firm was hired to contact a random sample of 400 instructors from the database of 1,690 to interview during February 1999. A total of 465 instructors completed the survey. Data were analyzed using the SAS statistical software package. A combination of univariate, bivariate, and multivariate analyses were performed. Frequency distribution tables were generated to provide descriptive statistics on the data. Chi-square tables were produced to show correlation among variables. A series of logistic regression models were computed to predict whether or not instructors were active in the program and whether they planned to continue teaching the Triple Touch program.
Stunting, Diarrheal Disease and Parasitic Infection in Peruvian Infants: Their Effect on Cognitive Test Scores in Late Childhood
Douglas Berkman, Johns Hopkins University
Cognitive performance is a multi-factorial result of environmental and health related factors at different stages of life. Although stunting is known to be associated with poor cognitive outcomes, these effects have not been extensively described in well-controlled studies, particularly regarding the interactions with infectious diseases. We conducted a cross-sectional evaluation of cognitive test scores at 8-10 years of age in a cohort of 145 Peruvian children whose diarrheal and nutritional status was assessed during their first 2 years of life. Cognitive performance was measured using the Wechsler Intelligence Scale for Children-Revised (WISC-R) in combination with a socioeconomic survey and weight and height measurements. We applied sequential multivariate ANOVA to identify factors related to cognitive performance; the significance of the contribution of each variable was assessed with the F-test. We observed significant inter-tester variability and adjusted for this effect in all analyses. Children of university-educated fathers achieved WISC-R scores 11 points higher (8% of variance) than children of less educated fathers. Educational characteristics explained 23% of the variance and demonstrated independent effects on cognitive test scores. Public schooling, lower grade level, and grade retention were associated with a 6-8 point deficit in cognitive test scores. After adjusting for socioeconomic status (SES) and other significant effects, stunting for the first time between 6 and 17 months of age, stunting for at least 2 months per year, and severe stunting in the second year were associated with decreased cognitive test scores of 4.3, 4.2, and 10.0 points. Severe stunting in the second year of life explained an additional 3.6% of the variance. Three percent of stool samples tested positive for C. parvum, yet we were unable to demonstrate an association with cognitive test scores. Children experienced an average of 3 days per month of diarrhea. Each additional day of diarrhea per month in the second year corresponded to a 1.7 point decrease on the WISC-R test (p=0.008), but this association was not significant after adjusting for other factors. Children who had more than 2 Giardia episodes per hundred samples scored on average 4.1 points lower (p=0.04) than children with 2 or fewer episodes (2.3% of variance). Our final model explained 45% of the total variance and was fairly insensitive to alternative specifications.
We have demonstrated that Giardia infections, as well as the timing, severity and persistence of stunting, adversely affect cognitive performance in late childhood. These effects were significant independently of SES and other significant factors. The effect of Giardiasis may indicate residual confounding from nutritional status or could be a more sensitive indicator of environmental contamination. Our results suggest that prevention of the causes of stunting and Giardia infections in early childhood could significantly enhance the cognitive development of Peruvian children living in similar communities.
Screening for Domestic Violence in Emergency Departments: Policy vs. Practice
Kevin Delaney, University at Albany SUNY
Other Authors: Dr. Louise-Anne McNutt and Dr. Vincent Verdile
The purpose of this study is to investigate how hospital domestic violence (DV) screening and intervention policies translate into practice in emergency departments. Specifically, we quantified the number of emergency departments that have implemented universal screening for domestic violence. We also identified the status of several process measures such as: whether the hospitals have a screening question on their assessment form, whether they have domestic violence information available for patients, and the level of contact these hospitals have with community groups for victims of violence. Additionally, we assessed individual staff members perceived barriers to screening, and used their experiences to suggest methods for improving screening practices. The emergency departments of 15 hospitals from urban, suburban and rural settings in 10 contiguous counties in upstate New York were asked to participate in a cross-sectional survey. Questions related to knowledge about DV, community resources, hospital policy, attitudes and experiences with DV were included. The interviews were audiotaped and transcribed for analyses. Descriptive statistical analyses for demographic and close-ended questions included means/standard deviations or frequencies/percentages. Interviews were further analyzed using standard qualitative methods.
Nine of 15 hospitals have DV screening questions on their triage form, however only one conducts routine universal screening for domestic violence among its female patients. Staff members at the remaining hospitals only ask about DV when the nurse/physician feels the question is warranted. Staff's personal experience, comfort with asking and position appear to influence screening practice. Few staff received substantive DV intervention training, many request more contact with local community groups. Policy is beginning to translate into practice. Many hospitals have added a screening question to their triage form. Staff are aware that they can play a role in identifying and helping abused women, but many still do not ask about abuse routinely or universally. Education and interaction with community groups who can both raise awareness of abuse and actually provide services to women identified on a screen seem to be crucial to improving screening practices of staff members.
Sources of Advice During Pregnancy: A Quantitative and Qualitative Study Focusing on Low-Income Women
Caroline L. Dunn, University of Minnesota
Researchers have long noted a reliance of low-income women on nonprofessional sources of advice on pregnancy. Yet little is known today about the content of lay health advice and how friends and relatives may complement or counteract messages from health professionals on matters related to diet, drinking alcohol, and smoking. This study examines the role of female friends and relatives ('confidantes') as advice givers during pregnancy. Methods: A questionnaire was verbally administered to 225 pregnant women enrolled in the WIC program and focus group discussions were held with nine of their confidantes. Quantitative and qualitative data were collected on the types of advice confidantes give to women during pregnancy, the frequency of this advice, and how it differs from doctors' in its persuasiveness, content, and quality. Results: Data revealed important differences in the characteristics of advice provided by doctors and confidantes. Survey results indicated that pregnant women considered advice from doctors to be generally more persuasive. Focus group discussions clarified that trust in doctors related to the medical knowledge they provided on specific problems that arose. Confidantes, in contrast, were valued for their daily comfort and first-hand experience on general health-related matters. Relatives were often expected to share similar experiences in pregnancy, which added to the value of their advice. Conclusions: Established levels of trust and comfort may lead pregnant women to seek advice from their confidantes first, before approaching doctors. Reliance of low-income women on confidante advice and experience is important to consider in the development of pregnancy interventions.
The Epidemiology of Prematurity at Harare Maternity Hospital, Zimbabwe: The Contritubtion of HIV Infection
Shinga Feresu, University of Michigan
Other Authors: L. Gwanzura, S. Harlow and G. Woelk
Pre-term delivery is one of the most common causes of infant mortality and morbidity in Zimbabwe . The authors studied the contribution of HIV infection to prematurity in women delivering at Harare Maternity Hospital , the largest hospital in Zimbabwe between March 10 to June 11, 1999. This was the first study to quantify the role of HIV as a risk factor for prematurity in this country. A case-control study of 148 women who delivered pre-term infants, and 156 controls who delivered term babies had their blood tested for HIV infection. A pre-term baby was defined as a delivery after 20 weeks and before 37 weeks of pregnancy through using the Ballard method of assessing gestational age. Information on the demographic, medical and obstetrical risk factors was collected from the obstetric record and through an interview.
Out of 312, 117(37.5%) were positive of which 63(43%) cases and 52(33%) controls tested positive for HIV infection. The crude odds ratio was 1.53(95% confidence intervals 0.96 to 2.43), p-value 0.076 and did not change much after adjusting for mother's age, parity, birth weight, low birth weight, pregnancy induced hypertension and pre-labor rupture of membranes. HIV confers 53% risk of prematurity, which was borderline. The sample size, random misclassification could have masked the effect. More efforts targeted at behavior change, safe sex and premarital testing is important since HIV prevalence is high in this population. The effects of HIV on low birth weight, are now of concern and will further be explored for the presentation.
5-Year Survival of Women Treated with Breast Conserving Surgery and Radiation Therapy Compared to Women Receiving a Total Mastectomy
Aliza Fink, Boston University
Other Authors: T.L. Lash and R.A. Silliman
The National Surgical Adjuvant Breast and Bowel Project Protocol B-06 - a randomized clinical trial with five-years of follow-up - found 76% survival for breast cancer patients treated with total mastectomy compared with 86% survival for those treated with breast conserving surgery and radiation therapy (p=0.07). Our objective was to determine whether a similar pattern is observed in a clinical practice setting.
We pooled data from two observational cohort studies of breast cancer patients who were aged 45 or older with local or regional disease. The first study included 357 women diagnosed between 1984 and 1986 at eight Rhode Island hospitals. The second study included 265 women diagnosed between 1992 and 1995 at five Massachusetts hospitals. Medical records provided information on treatment and confounders. Vital status and cause of death were ascertained from the National Death Index. The data were analyzed using Cox's proportional hazards regression with propensity score adjustment for potential confounding because it simulates randomization.
72.4% of women treated with total mastectomy survived 5 years compared with 86.0% of women treated with breast conserving surgery and radiation therapy. Women treated with total mastectomy were 2-fold more likely to die during the follow-up than women treated with breast conserving surgery and radiation therapy (adjusted relative hazard =2.0; 95% CI 1.2-3.2). This analysis suggests that 5-year survival is improved among women treated with breast conserving surgery and radiation compared with those treated with total mastectomy.
Abortion Needs and Experiences of Women in Rural Uttarakhand, India
Lisa Patel, University of California at Los Angeles
Abortion related mortality is one of the major causes of death in women in India . It is quite perplexing that in a country where abortion is legal, so many women must turn to illegal and often unsafe sources for abortion services. An estimated 15-30% of maternal mortality in Uttar Pradesh is due to abortion related complications. The health situation of women in Uttarakhand, the hill districts of Uttar Pradesh, is especially poor. This qualitative research study was carried out in several villages in Almora District, located in Uttarakhand, to better understand: who seeks abortion services in theses villages and why; where village women seek abortion services and why; and what are village women's needs, experiences, knowledge, attitudes and practices with regards to abortion, abortion services and post-abortion complications. The focus groups and interviews revealed that even though a substantial number of women access abortion services in these villages, very few seem to go to providers where they can obtain safe or legal abortions. Mostly married women sought abortions in these villages. Most of the women sought abortions in response to an unwanted pregnancy to control family size or better space their pregnancies. Several women had obtained more than one abortion. After getting an abortion, most women experienced health complications including heavy bleeding, back pain, and weakness. Women use abortions as a method of family planning, suggesting an unmet need for contraception. Despite limitations, data from this study can be used by local health officials and local NGOs to design effective interventions.
Objective Versus Self-Report Diary Measures of Physical Activity: A Comparison of Methods Among Middle-Aged Women
Michael Schmidt, University of Massachusetts
Other Authors: L. Chasan-Taber, E. Puleo, A. Watras, D. Pober and P. Freedson
Physical activity may be an important modifiable risk factor for many diseases. Studies examining physical activity - disease relationships must validly measure physical activity in order to produce meaningful results. Questionnaires are often the most feasible measurement tool, their validity being ascertained by comparison with criterion standard measures of physical activity. The purpose of this study is to examine the correlation between two common criterion measures of physical activity and to compare validation estimates of a past-year activity questionnaire when each is used as the criterion standard. Fifty-eight subjects were randomly selected from a total of 215 women aged 39-65 years participating in the Alumni Health Study. Baseline and one-year measures of physical activity were obtained using a modified version of the Historical Physical Activity Survey (HPAS). Subjects also completed quarterly 7-day physical activity logs (PAL) and wore an activity monitor (CSA accelerometer) corresponding to the fall PAL time period. Pearson correlations were used to examine the relationship between criterion measures of activity and to calculate validity estimates for the past-year activity questionnaire. Correlations between PAL and CSA measures of physical activity were .21 and .39 for moderate and total activity, respectively. Validity coefficients for the questionnaire were .09 using the CSA and .20 using the PAL. Results suggest a surprisingly poor correlation between two common criterion measures of activity. It is possible that the CSA and PAL measure different aspects of physical activity or that one or both of these measures have poor validity when used with certain populations.
Characteristics and Birth Outcomes: Multiparious Women Receiving No Prenatal Care in the United States, 1992-1997
Cathy R. Taylor, University of Alabama at Birmingham
Although prenatal care is an established mechanism for identifying and managing medical, sociodemographic and behavioral risk factors impacting pregnancy outcomes, some U.S. women still receive no care prior to delivery. While they may be different from women who delay or receive inadequate care, women who receive no care may not be a homogenous group. It was hypothesized that women who receive no care comprise several groups with distinctive sociodemographic characteristics and that differences in birth outcome measures exist among the no care sub-groups. Singleton births to U.S. resident, multiparous women reporting no care were identified from 1992-1997 birth certificate data. Using clustering methodology, five clusters emerged from preliminary analysis: Cluster 1 - unmarried, adult white women living in rural areas; Cluster 2 - foreign-born, adult, Hispanic women with low education; Cluster 3 - married, adult, Black women with low education; Cluster 4 - married, adult, white women with mid-level education, and Cluster 5 - older, non-Hispanic women with high parity, birth interval greater than 4 years, living in rural areas and reporting substance abuse. Birth outcome measures varied significantly among clusters. Cluster 2 had the lowest rates of low birth weight and preterm birth, and Cluster 5 had the highest rates for these measures. These findings indicate that intervention efforts aimed at reducing the proportion of women receiving no care should be individually tailored and targeted to specific no care sub-groups with priority given to those sub-groups with the highest risk of a poor pregnancy outcome.
Use of Receiver Operating Characteristic Curves in Evaluating Logistic Regression Model
Jiping Wang, University of Pittsburgh
Other Author: Howard E. Rockette
Logistic regression is a statistical tool for identifying risk factors associated with an adverse event and for classifying patients into groups based on their prognosis. The Receiver Operator Characteristic (ROC) curve is a plot of the sensitivity (proportion of individuals predicted to have a future adverse event among those who actually have an event) versus 1-specificity (proportion of individuals predicted to have an event among those who remain event free) for the range of "cut points" that can be used to predict whether or not a patient will have an event. The Area Under the ROC Curve (AUC) is often used as an index of how well a scoring system is able to classify patients into one of the two medical alternatives (an area of 1 is perfect, 0.5 is non-informative) such as: death or survival, disease or non-disease. Recently, ROC curves have been used to evaluate adequacy of the scoring system obtained from a logistic regression model (LR-ROC). Several popular statistical software packages, such as SPSS, BMDP and STATA include ROC and LR-ROC as a procedure and this has led to the increased utilization of the use of the area under the ROC curve as a method to evaluate the predictive capability of a logistic regression model. A search of Medline for the period Jan., 1995 to July, 1999 resulted in 4292 citations for "ROC" and 133 citations for the combination of "ROC" and "logistic regression". Those articles were in a variety of biomedical and public health related fields and included the diagnosis, prognosis and treatment choice in internal medicine, surgery, obstetrics and gynecology, pediatrics, dentistry, dermatology, anesthesiology, psychiatry, radiology and oncology. ROC methods have also been used to evaluated models selecting biomarkers in human genetics, the decision for choosing the best screening test to identify high risk people from epidemiological populations and the prediction of length of stay, mortality and cost in intensive care. LR-ROC has also been used to construct or modify a standard clinical or epidemiological scoring system, such as: APACHE II, ICISS, APACHE III, POSSUM, OSF, SAPS and SAPS II. Although there is widespread use of the area under the ROC curve as an index of predictive capability of statistical models in the biomedical and public health area, there has not been much work to evaluate the procedure. Investigators using the procedure seem to implicitly assume that under the null hypothesis the AUC is 0.5 and that tests of hypothesis are at the nominal ( - level. In this paper we estimate the expected value of the area under the ROC curve when fitting a logistic regression model with M independent variables, compare the standard error (continued) based on simulations to that obtained using the standard nonparametric formula, and determine the true ( - error associated with testing the hypothesis that AUC = 0.5. We also conduct simulations to estimate these same statistical parameters where the model was fit to half of the data points and the AUC was obtained by applying the model to the remaining half of the data.
The properties of the area under the ROC curve when used as an index for evaluating the predictive capability of a regression model are investigated using computer simulation. A logistic regression models was fit to the M variables using the dichotomous variable y to indicate whether there was an event (y=1) or nonevent (y=0) for the specific set of M xi's, generated from a binormal distribution for patients with an event and patients without an event. The AUC was used to evaluate the predictive capability of the model. Formal statistical tests were conducted to determine if the AUC was significantly different from 0.5, the expected value of the AUC when there is no predicative capability.
The results of these investigations can be summarized as follows:
- For smaller sample sizes and as the number of variables in the model increases, the expected value of the area under the ROC curve can be considerably higher than by chance alone, the 0.5 usually associated with no predictive capability.
- Simulations when the number of patients with and without events is unequal indicate that this increase in expected area is even greater than the increase in expected area when the number of patients with and without events is equal.
- When the standard deviation of the area under the ROC curve was estimated empirically, the values were consistently lower than the values obtained from applying the usual nonparametric method to estimate the standard error.
- The ratios of the empirical estimates of the standard deviation to the theoretical estimate is reasonably consistent regardless of sample size but decreases as the number of variables in the model increases.
- When the hypothesis that the AUC is 0.5 is tested using the usual nonparametric method to estimate standard error and the normal approximation, the probability of rejecting is substantially higher than the nominal (- level whenever the number of factors in the model is greater than one. This is true even for large sample sizes.
- Adjusted critical regions are provided that can be used to avoid the limitations of the standard procedure, currently being used by many researchers.
Preliminary investigations indicate that the usual method of using the area under an ROC curve to evaluate the predictive capability of a logistic regression model tends to overestimate the area under the ROC curve and have an elevated Type I error when testing the predictive capability of a logistic regression model. These limitations appear to be worse for small sample sizes, as the number of variables in the model increases or when the number of patients with and without events is unequal. However, they remain substantial even for sample sizes as high as 1000. These results indicate that the algorithms being evaluated for patient diagnosis, screening, management and treatment in a large number of medical and public health areas have been overly optimistic in their presentation of the predicative capability of the model. Methods are given in this paper that provide a more realistic assessment of the predictive capability of the proposed methods.
Gonadal Development in the Japanese Medaka (Oryzias latipes) as a Model for Estrogenic Potency Determination
Mingyi Wen, Tulane University
Other Authors: A. Thiyagarajah, M.B. Anderson and Hartley
Increasing concern regarding potential estrogenic environmental contaminants highlights a need for the development of laboratory screening assays to measure estrogenic response. An in vivo screening method for natural and environmental estrogens using the Japanese medaka (Oryzias latipes) is being developed. Medaka fish exposed to 17(-estradiol (17(-E) during early life stages exhibit disruption of normal gonadal development. Some fish display an inter-sex appearance when fully mature and the proportion of females to males is higher than expected. In this study, newly hatched medaka fry with undifferentiated gonads (60/group) were exposed to sublethal concentrations of 17(-E (1.25, 2.5, and 5µg/L) for 48 hours in a water bath at 25ºC. There were two control groups: spring water and acetone carrier. After two weeks, 50% of the surviving fry were killed and processed for histopathological analysis. The remaining 50% of fry were evaluated after four months. The mortality rates ranged from 26-38% and there were no statistically significant differences among groups. After two weeks, testis-ova (hermaphroditism) were observed in all groups including the spring water control (8%). A dose-response trend for hermaphroditism was observed but it was not statistically significant (p>0.05). In the four month grow-out groups, the majority of fish developed into females, some were sexually immature, and hermaphroditism was not observed. Secondary sexual characteristics such as fin shape usually clearly identify the gender of the fish, but in our study we found that the gender of fish smaller than 2 cm in size was not easily identified. The estrogenic potency of xenoestrogens can be determined by comparing their dose response curve with 17(-E dose response.
The Relationship of Hookworm Infection to Anemia Among Pre-School Children in Rural Malawi
Thibaut Williams, Emory University
Hookworm infection has long been associated with iron deficiency and anemia among school-age children. WHO recommendations do not currently include children under five in anthelminthic campaigns in spite of the potential nutritional consequences of hookworm infection among this age group. The objective of this study is to examine the associations between hookworm infection and nutritional status among pre-school children in rural Africa . We collected stool samples, hemoglobin levels, anthropometric data and various demographic variables from a random sample of 200 children between 6 and 59 months of age living in the rural areas around Ekwendeni , Malawi . Only 1% of the children (n=2) surveyed were found to be infected with hookworm, while 90% (n=180) of the children had hemoglobin levels below the WHO cut-off for anemia for African children of this age group (Hb<110g/L). Stunting was particularly prevalent among this study sample, with 48.5% (n=97) of the children with a height-for-age Z-score less than -2. Due to its very low prevalence, hookworm infection was not found to be associated with anemia among the children in this study sample. Other variables that were associated with anemia among this age group were the age of the child (p=0.0002), wasting (p=0.0021), and higher parity of the mother (p=0.0002). The results of this study show that hookworm infection is not a significant cause of anemia among the study population. However, the very high prevalence of anemia among children under five in rural Malawi suggests that new avenues for improving the iron status of this age group need to be explored.
HIV-related Needs Assessment of Upstate SC: Our Community Believes What?!?
April Winningham, University of South Carolina