Ties between event times and jump times in the Cox model

Methods for dealing with tied event times in the Cox proportional hazards model are well developed. Also, the partial likelihood provides a natural way to handle covariates that change over time. However, ties between event times and the times that discrete time-varying covariates change have not been systematically studied in the literature. In this article, we discuss the default behavior of current software and propose some simple methods for dealing with such ties. A simulation study shows that the default behavior of current software can lead to biased estimates of the coefficient of a binary time-varying covariate and that two proposed methods (Random Jitter and Equally Weighted) reduce estimation bias. The proposed methods can be easily implemented with existing software. The methods are illustrated on the well-known Stanford heart transplant data and data from a study on intimate partner violence and smoking.

Event-weighted proportional hazards modelling for recurrent gap time data

The analysis of gap times in recurrent events requires an adjustment to standard marginal models. One can perform this adjustment with a modified within-cluster resampling technique; however, this method is computationally intensive. In this paper, we describe a simple adjustment to the standard Cox proportional hazards model analysis that mimics the intent of within-cluster resampling and results in similar parameter estimates. This method essentially weights the partial likelihood contributions by the inverse of the number of gap times observed within the individual while assuming a working independence correlation matrix. We provide an example involving recurrent mammary tumours in female rats to illustrate the methods considered in this paper.

Dietary supplement use in the spinal cord injury population

STUDY DESIGN

Longitudinal, non-experimental.

OBJECTIVES

To determine the following: (1) prevalence of supplement use in a representative sample of the chronic spinal cord injury (SCI) population; (2) most frequently consumed supplements; and (3) characteristics of consistent supplement users.

SETTING

Ontario, Canada.

METHODS

A structured questionnaire was used to collect demographic information from 77 community-dwelling adults with chronic SCI (50.6% paraplegia, 81.8% male, 42.4 + or - 11.9 years, body mass index (BMI) 25.4 + or - 5.1 kg m(-2)). A standardized form was used to record dietary intake, including supplements, in the previous 24 h, at three time points (baseline, 6 months and 18 months). Logistic regression and multivariate logistic regression were used to determine which characteristic(s) was (were) associated with consistent supplement use.

RESULTS

Seventy-one percent of the sample reported using supplements at least once, with 50.6% being classified as consistent supplement users (at least twice across the three time points). The top three supplements consumed were multivitamins (25%), calcium (20%) and vitamin D (16%). Supplement use status was not associated with gender, level of injury, age, education, physical activity, BMI, smoking or alcohol intake.

CONCLUSIONS

Dietary supplement use was common in our sample of individuals with long-standing SCI, but no common characteristics distinguished users from non-users. We suggest that health practitioners be aware of the high dietary supplement use in this population so that they can probe for type, dose and frequency, as supplements may have an important influence on dietary assessment results.

Misclassification of exposure: coffee as a surrogate for caffeine intake

This study was conducted to assess the effect of exposure misclassification when coffee is used as a surrogate measure of caffeine exposure. Subjects were randomly selected from the telephone directories of four regional municipalities in southern Ontario, CANADA: Data on daily caffeine intake from foods, beverages, and medications were collected from June to November 1995 through self-administered, mailed questionnaires from 481 men and women aged 30-75 years. Although coffee was the main source of caffeine, cross-tabulations of exposure to coffee by total caffeine intake showed that assessment of coffee alone severely underestimated caffeine intake by at least one exposure level. A hypothetical 10-fold increase in risk was completely obscured when only coffee was used to estimate total caffeine intake. The results of this study suggest that measuring coffee instead of caffeine intake may contribute to a lack of positive findings in studies of coffee as a risk factor for disease occurrence, if in fact caffeine is the exposure of interest. On the other hand, measurement of coffee, tea, and cola soft drink intake in the present study appeared to approximate caffeine intake sufficiently and not affect risk estimates adversely.

Design considerations for association studies of candidate genes in families

In genetic epidemiologic studies, investigators often use generalized linear models to evaluate the relationships between a disease trait and covariates, such as one or more candidate genes or an environmental exposure. Recently, attention has turned to study designs that mandate the inclusion of family members in addition to a proband. Standard models for analysis assume independent observations, which is unlikely to be true for family data, and the usual standard errors for the regression parameter estimates may be too large or too small, depending on the distribution of the covariates within and between families. The consequences of familial correlation on the study efficiency can be measured by a design effect that is equivalent to the relative information in a sample of unrelated individuals compared to a sample of families with the same number of individuals. We examine design effects for studies in association, and illustrate how the design effect is influenced by the intra-familial distribution of covariate values such as would be expected for a candidate gene. Typical design effects for a candidate gene range between 1.1 and 2.4, depending on the size of the family and the amount of unexplained familial correlation. These values correspond to a modest 10% increase in the required sample size up to more than doubling the requirements. Design effect values are useful in study design to compare the efficiency of studies that sample families versus independent individuals and to determine sample size requirements that account for familial correlation.

Pre-natal and peri-natal exposures and risk of testicular germ-cell cancer

The present case-control study was undertaken to investigate the association between exposure to maternal hormones and risk of testicular germ-cell cancer by histologic subgroups. Cases were males, aged 16 to 59 years, diagnosed with testicular germ-cell cancer in Ontario between 1987 and 1989. Histologic review was performed on all eligible cases for the purpose of categorizing cases as seminoma or non-seminoma (the latter classified 2 ways, with and without tumors containing seminoma). Risk factor data were collected on 502 cases, 346 case mothers, 975 age-matched controls, and 522 control mothers. Exogenous hormone exposure was associated with elevated risk (OR = 4.9, 95% CI 1.7-13.9). Several additional risk factors were associated with risk of testicular cancer: bleeding and threatened miscarriage (OR = 0.6, 95% CI 0.3-1.0), maternal cigarette smoking (12+ cigarettes/day OR = 0.6, 95% CI 0. 4-1.0). pre-term birth (OR = 1.6, 95% CI 1.0-2.5), and treatment for undescended testicle (OR = 8.0, 95% CI 3.2-20.0). First births were associated with elevated risk (OR = 1.7, 95% CI 1.0-2.8) among mothers below the age of 24 years at conception. There was little evidence that risk factors differed by histologic subgroup. We found evidence that exposure to maternal hormones, particularly estrogens, is associated with testicular germ-cell cancer risk. Not only does exposure to elevated levels (exogenous hormone use, pre-term birth, and first births among young mothers) increase risk but also exposure to relatively lower levels (heavy cigarette consumption and, perhaps, bleeding and threatened miscarriage) may decrease cancer risk.

Development and validation of the Osteoporosis Risk Assessment Instrument to facilitate selection of women for bone densitometry

BACKGROUND

Although mass screening for osteoporosis is not recommended among postmenopausal women, there is no consensus on which women should undergo testing for low bone mineral density. The objective of this study was to develop and validate a clinical tool to help clinicians identify which women are at increased risk for osteoporosis and should therefore undergo further testing with bone densitometry.

METHODS

Using Ontario baseline data from the Canadian Multicentre Osteoporosis Study, we identified all cognitively normal women aged 45 years or more who had undergone testing with dual-energy x-ray absorptiometry (DXA) at both the femoral neck and the lumbar spine (L1-L4). Participants who had a previous diagnosis of osteoporosis or were taking bone active medication other than ovarian hormones were excluded. The main outcome measure was low bone mineral density (T score of 2 or more standard deviations below the mean for young Canadian women) at either the femoral neck or the lumbar spine. Logistic regression analysis and receiver operating characteristic (ROC) analysis were used to identify the simplest algorithm that would identify women at increased risk for low bone mineral density.

RESULTS

The study population comprised 1376 women, of whom 926 were allocated to the development of the tool and 450 to its validation. A simple algorithm based on age, weight and current estrogen use (yes or no) was developed. Validation of this 3-item Osteoporosis Risk Assessment Instrument (ORAI) showed that the tool had a sensitivity of 93.3% (95% confidence interval [CI] 86.3%-97.0%) and a specificity of 46.4% (95% CI 41.0%-51.8%) for selecting women with low bone mineral density. The sensitivity of the instrument for selecting women with osteoporosis was 94.4% (95% CI 83.7%-98.6%). Use of the ORAI represented a 38.7% reduction in DXA testing compared with screening all women in our study.

INTERPRETATION

The ORAI accurately identifies the vast majority of women likely to have low bone mineral density and is effective in substantially decreasing the need for all women to undergo DXA testing.

Pregnancy outcomes in females after treatment for childhood cancer

This Ontario province-wide cohort study was conducted to compare the risk of adverse pregnancy outcomes in female childhood cancer survivors who received abdominal-pelvic radiation and/or chemotherapy with alkylating agents with the risk among those who were treated by non-sterilizing alkylating agents with the risk among those who were treated by non-sterilizing surgery only. Females in Ontario, Canada, diagnosed in 1964-1988 before age 20 with a histologically confirmed malignancy and who had survived for at least 5 years, attained age 18, and were alive at the time of study, were identified through the Ontario Cancer Registry. We ascertained pregnancy outcomes by a telephone-administered questionnaire. Treatment data were abstracted from medical records for 830 subjects 18-49 years of age, the analysis comprised 340 survivors who had one or more pregnancies after treatment. There was no evidence of an increased risk of having a spontaneous abortion or an infant with a birth defect. Survivors receiving abdominal-pelvic radiation were more likely to have a low birth weight infant (odds ratio estimate [OR] = 3.64; 95% confidence interval [CI] = 1.33-9.96), a premature low birth weight infant (OR = 3.29; 95% CI = 0.97-11.1), or an infant who died in the perinatal period (OR = 2.41; 95% CI = 0.50-11.5), compared with those receiving surgery. Risks of perinatal death and having a low birth weight infant increased with dose of radiotherapy directed to the abdomen.

Evaluation of the contribution of environmental factors

For the simulated data of GAW11, the roles of two environmental factors, E1 and E2, were investigated. Logistic regression analyses measuring the association between outcome (either mild or severe disease versus no disease) and E1 and E2 exposure indicated that E1 was a risk factor for disease (either mild or severe) but that E2 was not associated with outcome. Linkage analyses were performed for strata defined by E1 and E2 exposure. A specific disease locus was identified in these stratified analyses where this locus would not have been identified with an unstratified linkage analysis. Finally, stratified generalized transmission disequilibrium test analyses yielded several false positive results.

Disease expression and class II HLA antigens in systemic lupus erythematosus

The aim of this investigation was to examine the relationship between Class II HLA antigens and disease expression in systemic lupus erythematosus (SLE). HLA-DR and DQ antigen frequency was studied serologically in 217 SLE patients followed prospectively and compared to 320 healthy controls. The relationship between HLA antigens and the presence of disease manifestations, as well as death was investigated in 117 SLE patients enrolled within the first year of their disease. A univariate analysis confirmed the association between HLA-DR3 and SLE. HLA antigen DR1, DR6, DR7, DQw1 and DQw3 were decreased in patient group compared to the controls. A logistic regression model showed a significantly negative association with HLA-DR1, DR6 and DR7, and a positive association with HLA-DR3. The reduced frequency of HLA-DQw1 and DQw3 was maintained using a logistic procedure. Cox Proportional Hazards models revealed no association between HLA-Class II antigens and death. Logistic regression models revealed no associations between central nervous system (CNS) disease nor musculoskeletal manifestations with any of the DR antigens. There was a trend towards a lower frequency of HLA-DR6 in patients with renal involvement and lower prevalence of HLA-DR1 and HLA-DR7 in patients with vasculitis.

Leukemia after irradiation for endometrial cancer in Ontario

BACKGROUND

In human studies, the risk of leukemia after ionizing radiation has been found to be increased more often than for any other cancer. It is useful to study patients with cancer treated with radiation because exposure can be measured accurately, follow-up may be long, and often a comparable and sizable nonexposed group exists. Women with endometrial cancer represent an excellent population for study because they meet these Developed Leukemia After Endometrial.

METHODS

A population-based matched case-control study, nested among all patients with endometrial cancer diagnosed in Ontario, was undertaken to describe the relationship between radiation therapy and leukemia risk. Among 13,843 subjects treated from 1964 to 1987 who survived at least 1 year, 47 confirmed cases of leukemia were identified. Four control subjects were matched to each patient based on age, calendar year of diagnosis, and length of survival free of a second neoplasm. Medical records were abstracted, and radiation dose administered to active bone marrow was determined by dosimetry.

RESULTS

An elevated risk of all leukemias other than chronic lymphocytic leukemia was observed, but only within the first 10 years after endometrial cancer treatment (odds ratio 12.0; 90% confidence interval 2.8-52.1). There was insufficient statistical evidence that risk was influenced by dose or type of radiation therapy. Nor was there any evidence that risk was influenced by age at endometrial cancer diagnosis or by calendar period at diagnosis.

CONCLUSIONS

There is an increased risk of leukemia associated with radiation therapy for patients with endometrial cancer, but only within the first 10 years after treatment.

Lifetime occupational physical activity and risk of hip fracture in women

A case-control study was conducted to examine the effects of occupational activity on the risk of hip fracture in women. Only women who worked full-time or part-time for more than 6 months and for more than 15 h/wk since the age of 16 were considered for study. Case patients were between the ages of 55 and 84 years and had a diagnosis of hip fracture in 1989 in Metropolitan Toronto (n = 331). Control subjects were a population-based random sample of women frequency-matched by 5-year age groups (n = 1002). Those who worked for 20 years or less in any type of job were not at a decreased risk of hip fracture (odds ratio (OR) = 0.96; 95% confidence interval (CI), 0.70 to 1.32) compared to those who worked for more than 20 years in a sedentary job. However, those who worked for more than 20 years in moderate- to heavy-activity jobs were strongly protected against hip fracture (OR = 0.53; 95% CI, 0.30 to 0.95). Past and recent leisure-time activity, estrogen use, obesity, having epilepsy, and a previous fracture were significant risk factors. There was no statistically significant interaction between occupational activity and leisure-time physical activity, suggesting that both types of activity are independently associated with the risk of hip fracture. This study showed that being employed for more than 20 years in a job that requires heavy activity reduces the risk of hip fracture in postmenopausal women.

Evaluation of genetic and environmental effects using GEE and APM methods

Two analytic methods were used in the Problem 2 data set. First, generalized estimating equations (GEE) modelling was developed to adjust for familial correlation in regressions evaluating candidate genes and an environmental factor. Second, the affected-pedigree-member (APM) method was used to identify chromosomal regions of interest and linkage of candidate genes with disease affection status. The GEE method identified C5 (MG1) as important for the quantitative trait Q1 and the corresponding affection status DIS, but the APM method was only suggestive. The GEE method identified C2 (MG2) as important for Q2 but only marginally important for Q1 and not important for DIS.

Risk factors for renal cell carcinoma: results of a population-based case-control study

For a case-control study of risk factors for renal cell carcinoma, a mailed questionnaire was used to collect data on 518 cases and 1,381 population-based controls in Ontario, Canada. Active cigarette smoking increased risk twofold among males (odds ratio estimate [OR] = 2.0, 95 percent confidence interval (CI) = 1.4-2.8) and females (OR = 1.9, CI = 1.3-2.6). Passive smoking appeared to increase risk somewhat among nonsmokers (males: OR = 1.6, CI = 0.5-4.7; females: OR = 1.7, CI = 0.8-3.4). A high Quetelet index (QI) was associated with a twofold increase in risk in both sexes, although this was based on reported weight at age 25 years for males (OR = 1.9, CI = 1.2-3.1) and five years prior to data collection for females (OR = 2.5, CI = 1.4-4.6). Diuretic use was associated with significantly increased risk among females, but not among males. Phenacetin use increased risk, while acetaminophen use was not associated with altered risk, although few subjects used either compound. Multiple urinary tract infections increased risk, but only significantly in females (OR = 1.9, CI = 1.2-2.9). Our data indicate the need for further exploration of passive smoking and diuretics as risk factors, as well as elucidation of mechanisms by which high lifetime QI and frequent urinary-tract infections might increase risk of this cancer.

Response of a cancer registry to reports of disease clusters

A protocol has been developed to investigate and report perceived clusters of cancer using a population-based cancer registry. The protocol comprises a series of steps which lead to assessment of the cluster's importance on the basis of three criteria: (1) statistical evidence of clustering; (2) documentation of the existence of exposure to a carcinogen; and (3) biological plausibility of the relationship between the exposure and the cancer of interest. The evaluation of these criteria results in one of three recommendations: further study, surveillance only, or no action. The protocol provides a systematic approach for investigation, makes efficient use of available cancer registry data, and responds to public concerns. The protocol is demonstrated by its application to an inquiry concerning an apparent excess of lung cancer in a small Ontario town and the possible role of radon gas exposure. The public health importance and limitations of addressing perceived disease clusters are discussed.

HLA alleles in systemic sclerosis: association with pulmonary hypertension and outcome

HLA antigen distribution was studied in 126 patients with systemic sclerosis (SSc) followed prospectively and compared to that of 325 healthy controls. The frequencies of HLA antigens DR3, DR5 and DRw52 were increased in patients with diffuse skin involvement (P = 0.02, 0.05, 0.03). The presence of DRw52 (relative risk [RR] much much greater than 1) and DRw6 (RR = 54.5) was associated with significantly increased risks of a fatal disease outcome with pulmonary hypertension (PHT). In the absence of PHT, DRw252 was inversely associated with the risk of death. These findings indicate an adverse prognosis in SSc when PHT is present in association with DRw52.

Regression modelling of HLA haplotype sharing in affected siblings

A link between the HLA system and disease susceptibility can be assessed through the observation of families containing two or more affected siblings. Departures from Mendelian inheritance of the parental haplotypes among the affected siblings are an indication of such a relationship. Other variables, such as environmental factors, may also be related to disease susceptibility. An approach to examining the degree of haplotype sharing and the effect of other variables of interest on observed sharing is presented and two examples analyzed.