Bridging the gap between different statistical approaches: An integrated framework for modelling

This paper proposes a template for modelling complex datasets that integrates traditional statistical modelling approaches with more recent advances in statistics and modelling through an exploratory framework. Our approach builds on the well-known and long standing traditional idea of 'good practice in statistics' by establishing a comprehensive framework for modelling that focuses on exploration, prediction, interpretation and reliability assessment, a relatively new idea that allows individual assessment of predictions. The integrated framework we present comprises two stages. The first involves the use of exploratory methods to help visually understand the data and identify a parsimonious set of explanatory variables. The second encompasses a two step modelling process, where the use of non-parametric methods such as decision trees and generalized additive models are promoted to identify important variables and their modelling relationship with the response before a final predictive model is considered. We focus on fitting the predictive model using parametric, non-parametric and Bayesian approaches. This paper is motivated by a medical problem where interest focuses on developing a risk stratification system for morbidity of 1,710 cardiac patients given a suite of demographic, clinical and preoperative variables. Although the methods we use are applied specifically to this case study, these methods can be applied across any field, irrespective of the type of response.

Partitioned Bremer support and multiple trees

Partitioned Bremer support (PBS) is a valuable means of assessing congruence in combined data sets, but some aspects require clarification. When more than one equally parsimonious tree is found during the constrained search for trees lacking the node of interest, averaging PBS for each data set across these trees can conceal conflict, and PBS should ideally be examined for each constrained tree. Similarly, when multiple most parsimonious trees (MPTs) are generated during analysis of the combined data, PBS is usually calculated on the consensus tree. However, extra information can be obtained if PBS is calculated on each of the MPTs or even suboptimal trees. (C) 2002 The Willi Hennig Society. Published by Elsevier Science (USA). All rights reserved.

Arrhythmias and mortality after myocardial infarction in diabetic patients - Relationship to diabetes treatment

OBJECTIVE- To assess the relationship between clinical course after acute myocardial infarction (AMI) and diabetes treatment. RESEARCH DESIGN AND METHODS- Retrospective analysis of data from all patients aged 25-64 years admitted to hospitals in Perth, Australia, between 1985 and 1993 with AMI diagnosed according to the International Classification of Diseases (9th revision) criteria was conducted. Short- (28-day) and long-term survival and complications in diabetic and nondiabetic patients were compared. For diabetic patients, 28-day survival, dysrhythmias, heart block, and pulmonary edema were treated as outcomes, and factors related to each were assessed using multiple logistic regression. Diabetes treatment was added to the model to assess its significance. Long-term survival was compared by means of a Cox proportional hazards model. RESULTS- Of 5,715 patients, 745 (12.9%) were diabetic. Mortality at 28 days was 12.0 and 28.1% for nondiabetic and diabetic patients, respectively (P < 0.001); there were no significant drug effects in the diabetic group. Ventricular fibrillation in diabetic patients taking glibenclamide (11.8%) was similar to that of nondiabetic patients (11.0%) but was lower than that for those patients taking either gliclazide (18.0%; 0.1 > P > 0.05) or insulin (22.8%; P < 0.05). There were no other treatment-related differences in acute complications. Long-term survival in diabetic patients was reduced in those taking digitalis and/or diuretics but type of diabetes treatment at discharge had no significant association with outcome. CONCLUSlONS- These results do not suggest that ischemic heart disease should influence the choice of diabetes treatment regimen in general or of sulfonylurea drug in particular.

Epidermal iontophoresis: II. Application of the ionic mobility-pore model to the transport of local anesthetics

Purpose, An in vitro study was carried out to determine the iontophoretic permeability of local anesthetics through human epidermis. The relationship between physicochemical structure and the permeability of these solutes was then examined using an ionic mobility-pore model developed to define quantitative relationships. Methods. The iontophoretic permeability of both ester-type anesthetics (procaine, butacaine, tetracaine) and amide-type anesthetics (prilocaine, mepivacaine, lidocaine, bupivacaine, etidocaine, cinchocaine) were determined through excised human epidermis over 2 hrs using a constant d.c. current and Ag/AgCl electrodes. Individual ion mobilities were determined from conductivity measurements in aqueous solutions. Multiple stepwise regression was applied to interrelate the iontophoretic permeability of the solutes with their physical properties to examine the appropriateness of the ionic mobility-pore model and to determine the best predictor of iontophoretic permeability of the local anesthetics. Results. The logarithm of the iontophoretic permeability coefficient (log PCj,iont) for local anesthetics was directly related to the log ionic mobility and MW for the free volume form of the model when other conditions are held constant. Multiple linear regressions confirmed that log PCj,iont was best defined by ionic mobility (and its determinants: conductivity, pK(a) and MW) and MW. Conclusions. Our results suggest that of the properties studied, the best predictors of iontophoretic transport of local anesthetics are ionic mobility (or pK(a)) and molecular size. These predictions are consistent with the ionic mobility pore model determined by the mobility of ions in the aqueous solution, the total current, epidermal permselectivity and other factors as defined by the model.

Do simple prudent health behaviours protect men from myocardial infarction?

Background We tested whether behaviours such as discarding obvious fat on meat, cessation of smoking, avoidance of passive smoking, habitual use of reduced fat milk, prudent consumption of alcohol and regular but moderate physical exercise are associated with a reduction of cardiovascular risk. Methods This was a population-based case-control study done in Perth, Western Australia. The cases (n = 336) were men aged 27-64 years with a first-ever acute myocardial infarction (AMI) during the period 1992-1993, and who survived at least 28 days. The controls (n = 735) were participants in a population-based survey of cardiovascular risk factors conducted during May-November 1994. Both groups completed the same questionnaire and the data were analysed with multiple logistic regression using backward elimination technique. Results Among men aged 27-64 years simple measures such as participation in non-vigorous exercise (odds ratio [OR] = 0.5, 95% CI : 0.4-0.7), and avoidance of added salt (OR = 0.6, 95% CI : 0.4-0.9) are associated with significant and Important protection from AMI. Conclusion After 25 years of falling mortality in Australia, lifestyles can still be significantly improved to reduce heart disease even further.

Who fails to complete tuberculosis treatment? Temporal trends and risk factors for treatment interruption in a community-based directly observed therapy programme in a rural district of South Africa

SETTING: Hlabisa Tuberculosis Programme, Hlabisa, South Africa. OBJECTIVE: To determine trends in and risk factors for interruption of tuberculosis treatment. METHODS: Data were extracted from the control programme database starting in 1991. Temporal trends in treatment interruption are described; independent risk factors for treatment interruption were determined with a multiple logistic regression model, and Kaplan-Meier survival curves for treatment interruption were constructed for patients treated in 1994-1995. RESULTS: Overall 629 of 3610 surviving patients (17%) failed to complete treatment; this proportion increased from 11% (n = 79) in 1991/1992 to 22% (n = 201) in 1996. Independent risk factors for treatment interruption were diagnosis between 1994-1996 compared with 1991-1393 (odds ratio [OR] 1.9, 95% confidence interval [CT] 1.6-2.4); human immunodeficiency virus (HIV) positivity compared with HIV negativity (OR 1.8, 95% CI 1.4-2.4); supervised by village clinic compared with community health worker (OR 1.9, 95% CI 1.4-2.6); and male versus female sex (OR 1.3, 95% CI 1.1-1.6). Few patients interrupted treatment during the first 2 weeks, and the treatment interruption rate thereafter was constant at 1% per 14 days. CONCLUSIONS: Frequency of treatment interruption from this programme has increased recently. The strongest risk factor was year of diagnosis, perhaps reflecting the impact of an increased caseload on programme performance. Ensuring adherence to therapy in communities with a high level of migration remains a challenge even within community-based directly observed therapy programmes.

Beyond ovulation: Oral contraceptives and epithelial ovarian cancer

In a case-control study in three Australian states that included 794 women with epithelial ovarian cancer and 853 community controls for whom we had adequate contraceptive and reproductive histories, Re examined the effects of oral contraceptive use after controlling for estimated number of ovulatory cycles. Other covariates included in the multiple logistic regression analysis were parity, smoking, and history of pelvic surgery. The protective effect of duration of oral contraceptive use appeared to be multiplicative, with a 7% decrease in relative risk per year [95% confidence interval (CI) = 4-9%], persisting beyond 15 years of exposure. Use for up to 1 year may have a greater effect than predicted (odds ratio = 0.57; 95% CI = 0.40-0.82), whereas use before the first pregnancy may be additionally beneficial (odds ratio = 0.95; 95% CI = 0.87-1.03, adjusted for overall duration of use). Better control for ovulatory life might attenuate these estimates somewhat. There was little evidence of waning protection with time since last exposure or of extra benefit with early commencement of oral contraceptive use. We found no convincing evidence of effect modification in any factor examined or differences in effect among the three main histologic cancer types or between borderline and malignant tumors. Oral contraceptives may act by both suppressing ovulation and altering the tumor-promoting milieu.

High HIV incidence and prevalence among young women in rural South Africa: Developing a cohort for intervention trials

Objective: To measure prevalence and model incidence of HIV infection. Setting: 2013 consecutive pregnant women attending public sector antenatal clinics in 1997 in Hlabisa health district, South Africa. Historical seroprevalence data, 1992-1995. Methods: Serum remaining from syphilis testing was tested anonymously for antibodies to HIV to determine seroprevalence. Two models, allowing for differential mortality between HIV-positive and HIV-negative people, were used. The first used serial seroprevalence data to estimate trends in annual incidence. The second, a maximum likelihood model, took account of changing force of infection and age-dependent risk of infection, to estimate age-specific HIV incidence in 1997. Multiple logistic regression provided adjusted odds ratios (OR) for risk factors for prevalent HIV infection. Results: Estimated annual HIV incidence increased from 4% in 1992/1993 to 10% in 1996/1997. In 1997, highest age-specific incidence was 16% among women aged between 20 and 24 years. in 1997, overall prevalence was 26% (95% confidence interval [CI], 24%-28%) and at 34% was highest among women aged between 20 and 24 years. Young age (<30 years; odds ratio [OR], 2.1; p = .001), unmarried status (OR 2.2; p = .001) and living in less remote parts of the district (OR 1.5; p = .002) were associated with HIV prevalence in univariate analysis. Associations were less strong in multivariate analysis. Partner's migration status was not associated with HIV infection. Substantial heterogeneity of HIV prevalence by clinic was observed (range 17%-31%; test for trend, p = .001). Conclusions: This community is experiencing an explosive HIV epidemic. Young, single women in the more developed parts of the district would form an appropriate cohort to test, and benefit from, interventions such as vaginal microbicides and HIV vaccines.

Variation in cervical cancer screening by region, socio-economic, migrant and Indigenous status in women in New South Wales

The purpose of this study was to estimate the extent of association of cervical screening in NSW women with socio-economic status (SES), rurality, and proportions of non-English speaking background (NESB) and Indigenous status. Data on women who had at least one Pap test over two years (January 1998-December 1999) were obtained from the NSW Pap test Register. Each local government area (LGA) was allocated to categories of population proportions of NESB and Indigenous status, a rurality classification based on population density and remoteness, and to an SES quintile. The odds ratios (OR) of having a Pap test were estimated and confounding adjusted by multiple logistic regression analysis. Implied Pap test rates in urban NESB and in rural Indigenous women were estimated from the modelled estimates. The adjusted OR for a Pap test in large rural centres (1.14) was significantly higher than those for metropolitan or capital city residents (0.9 and 1.0 respectively). Adjusted OR for a Pap test in other rural centres (0.73) and other remote areas (0.64) were significantly lower than those for metropolitan or capital city residents. In urban populations the lowest OR were in areas with both low SES and high proportion of NESB. The lowest OR for Pap screening in rural populations occurred in the most remote areas with the highest proportion of Indigenous women. For urban NESB women the biennial Pap test rate was estimated as 50%, and for rural Indigenous women 29%, compared with the NSW average of 59%.

Weight loss during prolonged lactation in rural Bangladeshi mothers

To determine the duration of lactation which is associated with weight loss in rural Bangladeshi mothers and also to determine the relationship with consumption patterns of principal food items, a cross-sectional study was carried out among 791 lactating rural Bangladeshi mothers aged 18-40 years. Results were compared with 333 non-pregnant and non-lactating mothers of a similar age group. The duration of lactation was up to 60 months. The mean difference in body-weight and body mass index (BMI) of lactating mothers who breastfed their children up to 24 months was significantly lower compared to non-lactating mothers of the same age group, but no differences were observed for those who breastfed beyond 24 months. The frequency of consumption of principal food items was comparable between the non-lactating and the lactating mothers who breastfed beyond 24 months. Results of multiple linear regression analysis showed that body-weight of mothers was negatively correlated with 1-12 month(s) and 13-24 months of lactation after controlling for height, education, and food consumption (slope -1.04, p < 0.05 and slope -1.23, p < 0.05 respectively). Height and consumption of meat and milk were significantly positively correlated with body-weight (slope 0.53, p < 0.001; slope 1.44, p < 0.001; and slope 0.75, p < 0.05 respectively). The study concluded that Bangladeshi women who breastfed up to 24 months were of lower weight than non-lactating mothers, most likely due to the effect of lactation. These mothers were not taking any additional foods during their lactating period. Based on the findings of the study, it is recommended that mothers consume additional energy-rich foods during the first 24 months of lactation to prevent weight loss.

Breastfeeding, menopause, and epithelial ovarian cancer

No previous study has examined the modifying effect of menopausal status on the association between lactation and ovarian cancer risk. We recruited 824 epithelial ovarian cancer cases and 855 community controls in three Australian states, collecting reproductive and lactation histories by means of a contraceptive calendar and pregnancy and breastfeeding record. We report results in women with at least one liveborn infant for unsupplemented breastfeeding, in line with a biological model linking suppression of ovulation to reduction in ovarian cancer risk. We derived odds ratios from multiple logistic regression models including number of liveborn children, age, age at first or last birth, and other potential confounders, overall and by menopausal status. Estimates of relative risk of ovarian cancer per month of full lactation were 0.99 [95% confidence interval(CI) = 0.97-1.00] overall and 1.00 (95% CI = 0.99-1.01) and 0.98 (95% CI = 0.95-1.01) among post- and premenopausal women, respectively. We tailored a lactation variable to the incessant ovulation hypothesis by progressively discounting breastfeeding the longer after birth it occurred, finding odds ratios similar to those for the unmodified duration variable. We found no association of note among postmenopausal women. Breastfeeding seems to be somewhat protective against ovarian cancer, but only before menopause.

Tubal sterilisation, hysterectomy and decreased risk of ovarian cancer

We have examined the effect of tubal sterilisation and hysterectomy on risk of ovarian cancer in a large case-control study in eastern Australia involving 824 women aged 18-79 years, diagnosed with epithelial ovarian cancer between 1990 and 1993, and 855 controls randomly selected from the electoral roll. Relative risks for ovarian cancer were estimated using multiple categorical regression to adjust for age, parity, oral contraceptive use and other risk factors. Tubal sterilisation was associated with a 39% reduction in risk of ovarian cancer (RR 0.61, 95% Cl 0.46-0.85) and hysterectomy with a 36% reduction (RR 0.64, 95% Cl 0.48-0.85). Risk remained low 25 years after surgery and was reduced irrespective of sterilisation technique, and estimates were similar among various types of epithelial ovarian cancer. The greatest reduction (74%) was observed among women with primary peritoneal tumours. Pelvic infection and use of vaginal sprays or contraceptive foams were not related to ovarian cancer, while use of talc in the perineal region slightly but significantly increased risk among women with patent fallopian tubes. Reportedly heavy or painful menses, perhaps associated with retrograde flow, were associated with ovarian cancer, and reduction in risk of disease after hysterectomy was greatest among women who had heavy periods. Our findings support the theory that contaminants from the vagina, such as talc, and from the uterus, such as endometrium, gain access to the peritoneal cavity through patent fallopian tubes and may enhance the malignant transformation of ovarian surface epithelium. Surgical tubal occlusion may reduce the risk of ovarian cancer by preventing the access of such agents. (C) 1997 Wiley-Liss, Inc.

Determinants of hepato- and spleno-megaly in Hunan, China: cross-sectional survey data from areas endemic for schistosomiasis

In order to understand the determinants of schistosome-related hepato- and spleno-megaly better, 14 002 subjects aged 3-60 years (59% male; mean age =32 years) were randomly selected from 43 villages, all in Hunan province, China, where schistosomiasis caused by Schistosoma japonicum is endemic. The abdomen of each subject was examined along the mid-sternal (MSL) and mid-clavicular lines, for evidence of current hepato- and/or spleno-megaly, and a questionnaire was used to collect information on the medical history of each individual. Current infections with S. japonicum were detected by stool examination. Almost all (99.8%) of the subjects were ethnically Han by descent and most (77%) were engaged in farming. Although schistosomiasis appeared common (42% of the subjects claiming to have had the disease), only 45% of the subjects said they had received anti-schistosomiasis drugs. Overall, 1982 (14%) of the subjects had S. japonicum infections (as revealed by miracidium-hatching tests and/or Katon Katz smears) when examined and 22% had palpable hepatomegaly (i.e. enlargement of at least 3 cm along the MSL), although only 2.5% had any form of detectable splenomegaly (i.e. a Hackett's grade of at least 1). Multiple logistic regression revealed that male subjects, fishermen, farmers, subjects aged greater than or equal to 25 years, subjects with a history of schistosomiasis, and subjects who had had bloody stools in the previous 2 weeks were all at relatively high risk of hepato- and/or spleno-megaly. In areas moderately endemic for Schistosoma japonicum, occupational exposure and disease history appear to be good predictors of current disease status among older residents. These results reconfirm those reported earlier in the same region.

Self-reported calcium intake and bone mineral content in children and adolescents

Objective: We examined the relationship between self-reported calcium (Cal intake and bone mineral content (BMC) in children and adolescents. We hypothesized that an expression of Ca adjusted for energy intake (El), i.e., Ca density, would be a better predictor of BMC than unadjusted Ca because of underreporting of EI. Methods: Data were obtained on dietary intakes (repeated 24-hour recalls) and BMC (by DEXA) in a cross-section of 227 children aged 8 to 17 years. Bivariate and multivariate analyses were used to examine die relationship between Ca, Ca density, and the dependent variables total body BMC and lumbar spine BMC. Covariates included were height, weight, bone area, maturity age, activity score and El. Results: Reported El compared to estimated basal metabolic rate suggested underreporting of El. Total body and lumbar spine BMC were significantly associated with El, but not Ca or Ca density, in bivariate analyses. After controlling for size and maturity, multiple linear regression analysis revealed unadjusted Ca to be a predictor of BMC in males in the total body (p = 0.08) and lumbar spine (p = 0.01). Unadjusted Ca was not a predictor of BMC at either site in females. Ca density was not a better predictor of BMC at either site in males or females. Conclusions: The relationship observed in male adolescents in this study between Ca intake and BMC is similar to that seen in clinical trials. Ca density did not enable us to see a relationship between Ca intake and BMC in females, which may reflect systematic reporting errors or that diet is not a limiting factor in this group of healthy adolescents.

Sun exposure and interaction with family history in risk of melanoma, Queensland, Australia

Sun exposure is the main environmental risk factor for melanoma, but the timing of exposure during life that confers increased risk is controversial. Here we provide the first report of the association between lifetime and age-specific cumulative ultraviolet exposure and cutaneous melanoma in Queensland, Australia, an area of high solar radiation, and examine the association separately for families at high, intermediate and low familial melanoma risk. Subjects were a population-based sample of melanoma cases diagnosed and registered in Queensland between 1982 and 1990 and their relatives. The analysis included 1,263 cases and relatives with confirmed cutaneous melanoma and 3,111 first-degree relatives without melanoma as controls. Data an lifetime residence and sun exposure, family history and other melanoma risk factors were collected by a mailed questionnaire. Using conditional multiple logistic regression with stratification by family, cumulative sun exposure in childhood and in adulthood after age 20 was significantly associated with melanoma, with estimated relative risks of 1.15 per 5,000 minimal erythemal doses (MEDs) from age 5 to 12 years, and 1.52 per 5 MEDs/day from age 20. There was no association with sun exposure in families at high familial melanoma risk. History of nonmelanoma skin cancer (relative risk [RR] = 1.26) and multiple sunburns (RR = 1.31) were significant risk factors. These findings indicate that sun exposure in childhood and in adulthood are important determinants of melanoma but not in those rare families with high melanoma susceptibility, in which genetic factors are likely to be more important. (C) 2002 Wiley-Liss, Inc.