Response The concept of secondary base

Study Designs for Extending Causal Inferences From a Randomized Trial to a Target Population

In this article, we examine study designs for extending (generalizing or transporting) causal inferences from a randomized trial to a target population. Specifically, we consider nested trial designs, where randomized individuals are nested within a sample from the target population, and nonnested trial designs, including composite data-set designs, where observations from a randomized trial are combined with those from a separately obtained sample of nonrandomized individuals from the target population. We show that the counterfactual quantities that can be identified in each study design depend on what is known about the probability of sampling nonrandomized individuals. For each study design, we examine identification of counterfactual outcome means via the g-formula and inverse probability weighting. Last, we explore the implications of the sampling properties underlying the designs for the identification and estimation of the probability of trial participation.

Hospital outpatients are satisfactory for case-control studies on cancer and diet in China: a comparison of population versus hospital controls

BACKGROUND

To investigate the internal validity of a food-frequency questionnaire (FFQ) developed for use in Chinese women and to compare habitual dietary intakes between population and hospital controls measured by the FFQ.

MATERIALS AND METHODS

A quantitative FFQ and a short food habit questionnaire (SFHQ) were developed and adapted for cancer and nutritional studies. Habitual dietary intakes were assessed in 814 Chinese women aged 18-81 years (407 outpatients and 407 population controls) by face-to-face interview using the FFQ in Shenyang, Northeast China in 2009-2010. The Goldberg formula (ratio of energy intake to basal metabolic rate, EI/BMR) was used to assess the validity of the FFQ. Correlation analyses compared the SFHQ variables with those of the quantitative FFQ. Differences in dietary intakes between hospital and population controls were investigated. Odds ratios (ORs) and 95% confidence intervals (CIs) were obtained using conditional logistic regression analyses.

RESULTS

The partial correlation coefficients were moderate to high (0.42 to 0.80; all p<0.05) for preserved food intake, fat consumption and tea drinking variables between the SFHQ and the FFQ. The average EI/BMR was 1.93 with 88.5% of subjects exceeding the Goldberg cut-off value of 1.35. Hospital controls were comparable to population controls in consumption of 17 measured food groups and mean daily intakes of energy and selected nutrients.

CONCLUSIONS

The FFQ had reasonable validity to measure habitual dietary intakes of Chinese women. Hospital outpatients provide a satisfactory control group for food consumption and intakes of energy and nutrients measured by the FFQ in a Chinese hospital setting.

Population versus hospital controls for case-control studies on cancers in Chinese hospitals

BACKGROUND

Correct control selection is crucial to the internal validity of case-control studies. Little information exists on differences between population and hospital controls in case-control studies on cancers in Chinese hospital setting.

METHODS

We conducted three parallel case-control studies on leukemia, breast and colorectal cancers in China between 2009 and 2010, using population and hospital controls to separately match 540 incident cases by age, gender and residency at a 1:1 ratio. Demographic and lifestyle factors were measured using a validated questionnaire in face-to-face interview. Odds ratios (ORs) and 95% confidence intervals (CIs) were obtained using conditional logistic regression analyses.

RESULTS

The two control groups had closely similar exposure distributions of 15 out of 16 factors, with the only exception being that hospital controls were less likely to have a BMI ≥ 25 (OR = 0.71, 95% CI: 0.54, 0.93). For exposure of green tea drinking, the adjusted ORs (95% CIs) comparing green tealeaves intake ≥ 1000 grams annually with non-drinkers were 0.51 (0.31, 0.83) and 0.21 (0.27, 0.74) for three cancers combined, 0.06 (0.01, 0.61) and 0.07 (0.01, 0.47) for breast cancer, 0.52 (0.29, 0.94) and 0.45 (0.25, 0.82) for colorectal cancer, 0.65 (0.08, 5.63) and 0.57 (0.07, 4.79) for leukemia using hospital and population controls respectively.

CONCLUSIONS

The study found that hospital controls were comparable with population controls for most demographic characteristics and lifestyle factors measured, but there was a slight difference between the two control groups. Hospital outpatients provide a satisfactory control group in hospital-based case-control study in the Chinese hospital setting.

Design issues for drug epidemiology

Despite the difficulties involved in designing drug epidemiology studies, these studies are invaluable for investigating the unexpected adverse effects of drugs. The aim of this paper is to discuss various aspects of study design, particularly those issues that are not easily found in either textbooks or review papers. We have also compared and contrasted drug epidemiology with the randomized controlled trial (RCT) wherever possible. Drug epidemiology is especially useful in the many situations where the RCT is not suitable, or even possible. The study base has to be defined before the appropriate cohort of subjects is assembled. If all of the cases are identified, then a referent sample of controls may be assembled by random sampling of the study base. If all of the cases cannot be assembled, a hypothetical secondary base may need to be created. Preferably, only new-users of the drug should be included, and the risk-ratio will be different for acute users and chronic users. Studies will usually only be possible when researching the unintended effects of drugs. It is difficult to study efficacy because of confounding by indication. In occasional circumstances it may be possible to study efficacy (examples are given). Discussion of the dangers of designing with generalisability in mind is provided. Additionally, the similarities in study design between drug epidemiology and the RCT are discussed in detail, as well as the design-characteristics that cannot be shared between the two methods.

Risk factors for injury to women from domestic violence

BACKGROUND

Domestic violence is the most common cause of nonfatal injury to women in the United States. To identify risk factors for such injuries, we examined the socioeconomic and behavioral characteristics of women who were victims of domestic violence and the men who injured them.

METHODS

We conducted a case-control study at eight large, university-affiliated emergency departments. The 256 intentionally injured women had acute injuries resulting from a physical assault by a male partner. The 659 controls were women treated for other conditions in the emergency department. Information was collected with a standardized questionnaire; no information was obtained directly from the male partners.

RESULTS

The 256 intentionally injured women had a total of 434 contusions and abrasions, 89 lacerations, and 41 fractures and dislocations. In a multivariate analysis, the characteristics of the partners that were most closely associated with an increased risk of inflicting injury as a result of domestic violence were alcohol abuse (adjusted relative risk, 3.6; 95 percent confidence interval, 2.2 to 5.9); drug use (adjusted relative risk, 3.5; 95 percent confidence interval, 2.0 to 6.4); intermittent employment (adjusted relative risk, 3.1; 95 percent confidence interval, 1.1 to 8.8); recent unemployment (adjusted relative risk, 2.7; 95 percent confidence interval, 1.2 to 6.5); having less than a high-school-graduate's education (adjusted relative risk, 2.5; 95 percent confidence interval, 1.4 to 4.4); and being a former husband, estranged husband, or former boyfriend (adjusted relative risk, 3.5; 95 percent confidence interval, 1.5 to 8.3).

CONCLUSIONS

Women at greatest risk for injury from domestic violence include those with male partners who abuse alcohol or use drugs, are unemployed or intermittently employed, have less than a high-school-graduate's education, and are former husbands, estranged husbands, or former boyfriends of the women.

Design issues in cohort studies

Some basic features of cohort studies are reviewed along with a brief discussion of relative risk estimation procedures. This background, and a discussion of factors influencing cohort study power, provides a context to examine various cohort study design choices, including the choice of study population, the selection of cohort size and follow-up duration, cohort ascertainment and subsampling choices, and decisions concerning validation and reliability substudies. It is noted that confounding issues, and especially issues related to measurement error in the assessment of primary exposure and confounding variables, may have a major influence on the precision and reliability of cohort study analyses. A brief discussion is provided of the role of cohort studies in relation to other observational and experimental research strategies.

[Proportional mortality studies: criteria for the selection of participant groups]

Proportional mortality designs are used widespread in occupational epidemiology. In this review those biases which can affect them, mainly the healthy worker bias, are discussed. Several options for their analysis and the assumptions to be accomplished for validity are reviewed: proportionate mortality analysis, standardized mortality ratio, and odds ratio. The inclusion of dead participants in a research exhibits several drawbacks. Starting out from the analysis of this sort of designs, the first criterium to select diseases is similar to case-control studies: the reference group must not include diseases related with the exposure under study. Analyzing the relationship between mortality and incidence rates, criteria to select diseases to be investigated by proportional mortality studies are offered. These designs yield a valid inference when the disease is rare and irreversible. If the exposure shortens duration of disease, a toward-the-null bias is introduced. The direction of bias is variable under other circumstances, although it shows a trend to be negative.