Assessing the known or intended effects of a drug using non-experimental epidemiologic designs is often infeasible because of the absence of accurate data on a major confounder, the severity of the disease treated by this drug. To circumvent this problem of confounding by indication, I propose the case-time-control design, which does not require a measure of this confounder. Instead, the design uses subjects from a conventional case-control design as their own controls and thus requires that exposure be measurable at two or more points in time. I present a logistic model to estimate relative risks under this design and illustrate the method with data from a case-control study of 129 cases of fatal or near-fatal asthma and 655 controls. The exposure of interest was quantity of use of inhaled beta-agonists, drugs prescribed for the treatment of asthma. I found that the "best" estimate of relative risk for high vs low beta-agonist use using the conventional case-control approach is 3.1 [95% confidence interval (CI) = 1.8-5.4], which inherently includes the confounding effect of unmeasured severity. The corresponding estimate of drug effect using the proposed case-time-control approach is 1.2 (95% CI = 0.5-3.0), which excludes the confounding effect of unmeasured severity. This example indicates that the class of beta-agonists may not play the leading role attributed to it in the risk of fatal or near-fatal asthma, as had been previously suspected, except perhaps at excessive doses, as indicated by the dose-response analyses.
The case-time-control design is a strategy that was developed to tackle the problem of confounding by indication in the nonexperimental assessment of intended or known effects of drugs. By using subjects as their own controls, the case-time-control design, under an explicitly defined model, eliminates the biasing effect of unmeasured confounding factors in the situation where exposure varies over time. The correct application of this design is based on a specific model that contains inherent assumptions and imposes certain conditions for the approach to be valid. In a recent article, Greenland questioned the validity of the case-time-control design by presenting several "counterexamples." In this paper, we review the assumptions inherent to the validity of the case-time-control model. We show that the presumed counterexamples are not what they are claimed to be, simply because they do not conform to the logistic model explicitly underlying the case-time-control approach. These examples are shown to arise from an alternative model that includes a confounder by period interaction, a term expressly avoided in the case-time-control model. When the data from these examples are modified to satisfy the correct model, the resulting case-time-control estimates of the treatment odds ratio are exactly 2, the true treatment effect. We clarify the necessity of this assumption in the context of matching in epidemiology. We also discuss briefly the assumptions of conditional independence and carryover effects.
A cohort of 1,904 vegetarians and persons leading a health-conscious life-style in the Federal Republic of Germany was identified in 1978. After a follow-up of 11 years, mortality from all causes was reduced by one-half compared with the general population [the standardized mortality ratio (SMR) was 0.44 for men, 0.53 for women]. Among the 858 men, 111 deaths were observed, with 255 expected; among the 1,046 women, 114 deaths were observed, with 215 expected. The lowest mortality was found for cardiovascular diseases (SMR = 0.39 for men, 0.46 for women); in particular, for ischemic heart diseases, mortality was reduced to one-third of that expected. Cancer mortality was reduced by one-half in men (SMR = 0.48), but only by one-quarter in women (SMR = 0.74). The deficit in cancer deaths was mainly observed for lung cancer and gastrointestinal cancers in males and for gastrointestinal cancers in females. Deaths from diseases of the respiratory and digestive systems were also reduced by about 50%. An excess of deaths occurred only for anemia. When the strict and the moderate vegetarians were analyzed separately, the strongest differential was found for ischemic heart diseases, which were much less frequent among strict vegetarians for both sexes. Some nondietary factors, such as higher socioeconomic status, virtual absence of smoking, and lower body mass index, may also have contributed to the lower mortality of the study participants.
We assessed the impact of smoking cessation on subsequent death rates among a cohort of 51,343 men and 66,751 women in California enrolled in late 1959 in the original American Cancer Society (ACS) Cancer Prevention Study (CPS I) and followed for 38 years. We compared the age-adjusted death rate, expressed as deaths per 1,000 person-years, among all subjects who smoked cigarettes in 1959 but who had largely quit as of 1997 with the death rate among never smokers over a 38-year period. The all causes death rate for males decreased from 20.67 during 1960-1969 to 18.68 during 1960-1997 for smokers and decreased from 10.51 to 9.46 for never smokers. The lung cancer death rate for males increased from 1.558 to 1.728 for smokers and increased from 0.127 to 0.133 for never smokers. The all causes death rate for females increased from 9.54 to 10.14 for smokers and decreased from 6.95 to 6.44 for never smokers. The lung cancer death rate for females increased greatly from 0.208 to 0.806 for smokers and increased from 0.094 to 0.116 for never smokers. These results indicate there has been no important decline in either the absolute or relative death rates from all causes and lung cancer for cigarette smokers as a whole compared with never smokers in this large cohort, in spite of a substantial degree of smoking cessation. While cessation clearly reduces the mortality risk among long-term former smokers, the population impact of cessation appears to be less than currently believed.
We hypothesized that stress induced by the terrorist attacks of September 11, 2001 might shorten pregnancy. To test this hypothesis, we compared gestational duration and risk of preterm delivery among women who were pregnant on September 11 with women who had delivered before that date.
We conducted a matched cohort study among pregnant women enrolled in the Boston-based cohort study Project Viva between 1999 and 2001. Each of 606 participants, pregnant on September 11, 2001, was matched to 1 or 2 participants who delivered before that date.
Compared with women who delivered before September 11th, women who were pregnant on September 11th had mean gestation length that was 0.13 weeks longer (95% confidence interval = -0.05 to 0.30) and an odds ratio for preterm delivery before 37 weeks' gestation of 0.60 (0.36 to 0.98). Only women exposed in the first trimester had longer gestation.
Contrary to expectation, Boston-area women who were pregnant on September 11th had a lower risk of preterm delivery than women who delivered before that date. Although the interpretation of this finding is difficult, it is clear that the acute psychologic stress documented nationwide after the terrorist attacks did not increase the risk of preterm delivery in this population at some distance from the attacks.
We examined the study design features and data collection methods from 13 case-control studies of colorectal cancer and diet, which had been previously combined and analyzed, to determine whether they influenced the results obtained from a pooled analysis. We assessed the methods used in each study, estimated a quality score, and used random effects models to re-estimate the pooled odds ratio for the association between dietary fiber and colorectal cancer for these data. Key features of the methods used in each study and the quality score were examined in random effects models to determine whether the heterogeneity found between study-specific risk estimates could be explained by these variables. The odds ratio for dietary fiber and colorectal cancer was 0.46 (95% confidence interval = 0.34-0.64) for the 13 case-control studies as estimated with a random effects model. Two factors, whether the diet questionnaire had been validated before use in the case-control study and whether qualitative data on dietary habits and cooking methods had been incorporated into the nutrient estimation, explained some of the heterogeneity found between studies. Risk estimates for dietary fiber and colorectal cancer were closer to the null for the studies that had these two characteristics. Quality score did not explain any between-study heterogeneity. Random effects models, which included fixed effects covariates, explained some between-study heterogeneity in these data and would be useful for future pooled analyses.
Recently, an association was reported between prenatal and postnatal exposure to cell phones and neurobehavioral problems in children at the age of 7 years.
A birth cohort was established in Sabadell, Spain between 2004 and 2006. Mothers completed questions about cell phone use in week 32 of the pregnancy (n = 587). Neurodevelopment of their children was tested at age 14 months using the Bayley Scales of Infant Development (n = 530).
We observed only small differences in neurodevelopment scores between the offspring of cell phone users and nonusers. Those of users had higher mental development scores and lower psychomotor development scores, which may be due to unmeasured confounding. There was no trend with amount of cell phone use within users.
This study gives little evidence for an adverse effect of maternal cell phone use during pregnancy on the early neurodevelopment of offspring.
The t(14;18) translocation is a common somatic mutation in non-Hodgkin's lymphoma (NHL) that is associated with bcl-2 activation and inhibition of apoptosis. We hypothesized that some risk factors might act specifically along t(14;18)-dependent pathways, leading to stronger associations with t(14;18)-positive than t(14;18)-negative non-Hodgkin's lymphoma. Archival biopsies from 182 non-Hodgkin's lymphoma cases included in a case-control study of men in Iowa and Minnesota (the Factors Affecting Rural Men, or FARM study) were assayed for t(14;18) using polymerase chain reaction amplification; 68 (37%) were t(14;18)-positive. We estimated adjusted odds ratios (OR) and 95% confidence intervals (CI) for various agricultural risk factors and t(14;18)-positive and -negative cases of non-Hodgkin's lymphoma, based on polytomous logistic regression models fit using the expectation-maximization (EM) algorithm. T(14;18)-positive non-Hodgkin's lymphoma was associated with farming (OR 1.4, 95% CI = 0.9-2.3), dieldrin (OR 3.7, 95% CI = 1.9-7.0), toxaphene (OR 3.0, 95% CI = 1.5-6.1), lindane (OR 2.3, 95% CI = 1.3-3.9), atrazine (OR 1.7, 95% CI = 1.0-2.8), and fungicides (OR 1.8, 95% CI = 0.9-3.6), in marked contrast to null or negative associations for the same self-reported exposures and t(14;18)-negative non-Hodgkin's lymphoma. Causal relations between agricultural exposures and t(14;18)-positive non-Hodgkin's lymphoma are plausible, but associations should be confirmed in a larger study. Results suggest that non-Hodgkin's lymphoma classification based on the t(14;18) translocation is of value in etiologic research.
Persistent organic pollutants may affect the immune and respiratory systems, but available evidence is based on small study populations. We studied the association between prenatal exposure to dichlorodiphenyldichloroethylene (DDE) and polychlorinated biphenyl 153 (PCB 153) and children's respiratory health in European birth cohorts.
We included 4608 mothers and children enrolled in 10 birth cohort studies from 7 European countries. Outcomes were parent-reported bronchitis and wheeze in the first 4 years of life. For each cohort, we performed Poisson regression analyses, modeling occurrences of the outcomes on the estimates of cord-serum concentrations of PCB 153 and DDE as continuous variables (per doubling exposure) and as cohort-specific tertiles. Summary estimates were obtained through random-effects meta-analyses.
The risk of bronchitis or wheeze (combined variable) assessed before 18 months of age increased with increasing DDE exposure (relative risk [RR] per doubling exposure = 1.03 [95% confidence interval = 1.00-1.07]). When these outcomes were analyzed separately, associations appeared stronger for bronchitis. We also found an association between increasing PCB 153 exposure and bronchitis in this period (RR per doubling exposure = 1.06 [1.01-1.12]) but not between PCB 153 and wheeze. No associations were found between either DDE or PCB 153 and ever-wheeze assessed after 18 months. Inclusion of both compounds in the models attenuated risk estimates for PCB 153 tertiles of exposure, whereas DDE associations were more robust.
This large meta-analysis suggests that prenatal DDE exposure may be associated with respiratory health symptoms in young children (below 18 months), whereas prenatal PCB 153 levels were not associated with such symptoms.
To study the combined effect of smoking and human papillomavirus (HPV) type 16 infection in high-grade cervical intraepithelial neoplasia, we analyzed data from a Norwegian population-based case-control study including 90 patients and 216 controls, 20-44 years of age. We assessed HPV-16 status both by polymerase chain reaction detecting virus DNA and by enzyme-linked immunosorbent assay detecting antibodies against virus capsid. Smoking was associated with cervical intraepithelial neoplasia grade II-III in HPV-16-positive individuals. Using the jointly unexposed (HPV-16 DNA-negative never-smokers) as the reference group, we determined the risk of cervical intraepithelial neoplasia grade II-III in HPV-16 DNA-positive never-smokers and HPV-16 DNA-positive ever-smokers (odds ratio = 15.7; 95% confidence limits = 3.2, 76.5, and odds ratio = 65.9; 95% confidence limits = 22.3, 194.3, respectively). The estimated proportion of cases among HPV-16-positive smokers that is attributable to the interaction between the two causes is 74%, based on HPV-16 DNA positivity.
The presence in serum of antibodies to viral antigens is generally considered a well-defined marker of past infection or vaccination. However, analyses of serological data that use a cut-off value to classify individuals as seropositive are prone to misclassification bias, in particular when studying infections with a weak serological response, such as the sexually transmitted human papillomavirus (HPV).
We analyzed the serological concentrations of HPV type 16 (HPV16) antibodies in the general Dutch population in 2006-2007, before the introduction of mass vaccination against HPV. We used a 2-component mixture model to represent persons who were seronegative or seropositive for HPV16. Component densities were assumed to be log-normally distributed, with parameters possibly dependent on sex. The age-dependent mixing proportions were smoothed using penalized splines to obtain a flexible seroprevalence profile.
Our results suggest that HPV16 seropositivity is associated with higher antibody concentrations in women as compared with men. Seroprevalence shows an increase starting from adolescence in men and women alike, coinciding with the age of sexual debut. Seroprevalence stabilizes in men around age 40, whereas it has a decreasing trend from age 50 onwards in women. Analyses that rely on a cut-off value to classify persons as seropositive yield substantially different seroprevalence profiles, leading to a qualitatively different interpretation of HPV16 infection dynamics.
Our results provide a benchmark for examining the effect of HPV16 vaccination in future serological surveys. Our method may prove useful for estimating seroprevalence of other infections with a weak serological response.