Jessie K. Edwards

The parametric g-formula for time-to-event data: intuition and a worked example.

Background: The parametric g-formula can be used to estimate the effect of a policy, intervention, or treatment. Unlike standard regression approaches, the parametric g-formula can be used to adjust for time-varying confounders that are affected by prior exposures. To date, there are few published examples in which the method has been applied. Methods: We provide a simple introduction to the parametric g-formula and illustrate its application in an analysis of a small cohort study of bone marrow transplant patients in which the effect of treatment on mortality is subject to time-varying confounding. Results: Standard regression adjustment yields a biased estimate of the effect of treatment on mortality relative to the estimate obtained by the g-formula. Conclusions: The g-formula allows estimation of a relevant parameter for public health officials: the change in the hazard of mortality under a hypothetical intervention, such as reduction of exposure to a harmful agent or introduction of a beneficial new treatment. We present a simple approach to implement the parametric g-formula that is sufficiently general to allow easy adaptation to many settings of public health relevance.

A comparison of respondent-driven and venue-based sampling of female sex workers in Liuzhou, China

Objectives To compare two methods for sampling female sex workers (FSWs) for bio-behavioural surveillance. We compared the populations of sex workers recruited by the venue-based Priorities for Local AIDS Control Efforts (PLACE) method and a concurrently implemented network-based sampling method, respondent-driven sampling (RDS), in Liuzhou, China. Methods For the PLACE protocol, all female workers at a stratified random sample of venues identified as places where people meet new sexual partners were interviewed and tested for syphilis. Female workers who reported sex work in the past 4 weeks were categorised as FSWs. RDS used peer recruitment and chain referral to obtain a sample of FSWs. Data were collected between October 2009 and January 2010. We compared the socio-demographic characteristics and the percentage with a positive syphilis test of FSWs recruited by PLACE and RDS. Results The prevalence of a positive syphilis test was 24% among FSWs recruited by PLACE and 8.5% among those recruited by RDS and tested (prevalence ratio 3.3; 95% CI 1.5 to 7.2). Socio-demographic characteristics (age, residence and monthly income) also varied by sampling method. PLACE recruited fewer FSWs than RDS (161 vs 583), was more labour-intensive and had difficulty gaining access to some venues. RDS was more likely to recruit from areas near the RDS office and from large low prevalence entertainment venues. Conclusions Surveillance protocols using different sampling methods can obtain different estimates of prevalence and population characteristics. Venue-based and network-based methods each have strengths and limitations reflecting differences in design and assumptions. We recommend that more research be conducted on measuring bias in bio-behavioural surveillance.

Evolving methods for inference in the presence of healthy worker survivor bias

Healthy worker survivor bias may occur in occupational studies due to the tendency for unhealthy individuals to leave work earlier, and consequently accrue less exposure, compared with their healthier counterparts. If occupational data are not analyzed using appropriate methods, this bias can result in attenuation or even reversal of the estimated effects of exposures on health outcomes. Recent advances in computing power, coupled with state-of-the-art statistical methods, have greatly increased the ability of analysts to control healthy worker survivor bias. However, these methods have not been widely adopted by occupational epidemiologists. We update the seminal review by Arrighi and Hertz-Picciotto (Epidemiology.1994; 5: 186-196) of the sources and methods to control healthy worker survivor bias. In our update, we discuss methodologic advances since the publication of that review, notably with a consideration of how directed acyclic graphs can inform the choice of appropriate analytic methods. We summarize and discuss methods for addressing this bias, including recent work applying g-methods to account for employment status as a time-varying covariate affected by prior exposure. In the presence of healthy worker survivor bias, g-methods have advantages for estimating less biased parameters that have direct policy implications and are clearly communicated to decision-makers.

Accounting for Misclassified Outcomes in Binary Regression Models Using Multiple Imputation With Internal Validation Data

Outcome misclassification is widespread in epidemiology, but methods to account for it are rarely used. We describe the use of multiple imputation to reduce bias when validation data are available for a subgroup of study participants. This approach is illustrated using data from 308 participants in the multicenter Herpetic Eye Disease Study between 1992 and 1998 (48% female; 85% white; median age, 49 years). The odds ratio comparing the acyclovir group with the placebo group on the gold-standard outcome (physician-diagnosed herpes simplex virus recurrence) was 0.62 (95% confidence interval (CI): 0.35, 1.09). We masked ourselves to physician diagnosis except for a 30% validation subgroup used to compare methods. Multiple imputation (odds ratio (OR) = 0.60; 95% CI: 0.24, 1.51) was compared with naive analysis using self-reported outcomes (OR = 0.90; 95% CI: 0.47, 1.73), analysis restricted to the validation subgroup (OR = 0.57; 95% CI: 0.20, 1.59), and direct maximum likelihood (OR = 0.62; 95% CI: 0.26, 1.53). In simulations, multiple imputation and direct maximum likelihood had greater statistical power than did analysis restricted to the validation subgroup, yet all 3 provided unbiased estimates of the odds ratio. The multiple-imputation approach was extended to estimate risk ratios using log-binomial regression. Multiple imputation has advantages regarding flexibility and ease of implementation for epidemiologists familiar with missing data methods.

Age at Entry Into Care, Timing of Antiretroviral Therapy Initiation, and 10-Year Mortality Among HIV-Seropositive Adults in the United States

BACKGROUND The goal of targeted antiretroviral therapy initiation is to minimize disease progression among patients with human immunodeficiency virus while minimizing the therapeutic burden on these patients. We examine whether the effect of delaying therapy initiation from 500 cells/mm(3) to 350 or 200 cells/mm(3) is modified by age at entry into care. METHODS We used the parametric g-formula to compare 10-year mortality under 3 CD4 cell count thresholds for therapy initiation among 3532 patients who entered care at 1 of 8 sites in the United States between 1998 and 2013. Results are reported separately for patients 18 to 34, 35 to 44, and 45 to 65 years of age at study entry. RESULTS In the observed data, 10-year mortality was 13% (165 deaths). Mortality increased from 11% under therapy initiation at 500 cells/mm(3) to 12% at 350 cells/mm(3) (risk difference [RD]: 0.87; 95% confidence interval [CI]: .56, 2.17) and to 14% at 200 cells/mm(3) (RD: 2.71; 95% CI: 1.79, 5.38). The effect of delaying therapy became greater with age: RDs comparing the 350-cells/mm(3) threshold with the 500-cells/mm(3) threshold ranged from -0.03 (95% CI: -0.15, 1.76) for patients 18 to 34 years of age to 0.99 (95% CI: -.27, 1.98) for patients 35 to 44 and to 2.30 (95% CI: 1.29, 5.42) for patients 45 to 65. CONCLUSIONS Delaying therapy increased 10-year mortality in the full cohort. Subgroup analysis highlights that patients entering care at older ages may be more vulnerable to the consequences of delayed ART initiation than younger patients.

Causal impact: Epidemiological approaches for a public health of consequence

In the article, the authors discuss issues in health care in the U.S. as of mid-2016, particularly the epidemiological approaches for a public health of consequence. Also examined are such topics as the causal impact framework, the components of said framework like the population intervention effects and internal validity, as well as the need to focus on the causes of health and disease in public health research.

All your data are always missing: incorporating bias due to measurement error into the potential outcomes framework

Epidemiologists often use the potential outcomes framework to cast causal inference as a missing data problem. Here, we demonstrate how bias due to measurement error can be described in terms of potential outcomes and considered in concert with bias from other sources. In addition, we illustrate how acknowledging the uncertainty that arises due to measurement error increases the amount of missing information in causal inference. We use a simple example to show that estimating the average treatment effect requires the investigator to perform a series of hidden imputations based on strong assumptions.

Understanding the High Prevalence of HIV and Other Sexually Transmitted Infections among Socio-Economically Vulnerable Men Who Have Sex with Men in Jamaica

Objectives This study estimates HIV prevalence among men who have sex with men (MSM) in Jamaica and explores social determinants of HIV infection among MSM. Design An island-wide cross-sectional survey of MSM recruited by peer referral and outreach was conducted in 2011. A structured questionnaire was administered and HIV/STI tests done. We compared three groups: MSM who accepted cash for sex within the past 3 months (MSM SW), MSM who did not accept cash for sex (MSM non-SW), and MSM with adverse life events (ever raped, jailed, homeless, victim of violence or low literacy). Results HIV prevalence among 449 MSM was 31.4%, MSM SW 41.1%, MSM with adverse life events 38.5%, 17 transgender MSM (52.9%), and MSM non-SW without adverse events 21.0%. HIV prevalence increased with age and number of adverse life events (test for trend P < 0.001), as did STI prevalence (P = 0.03). HIV incidence was 6.7 cases/100 person-years (95% CI: 3.74, 12.19). HIV prevalence was highest among MSM reporting high-risk sex; MSM SW who had been raped (65.0%), had a STI (61.2%) and who self identified as female (55.6%). Significant risk factors for HIV infection common to all 3 subgroups were participation in both receptive and insertive anal intercourse, high-risk sex, and history of a STI. Perception of no or little risk, always using a condom, and being bisexual were protective. Conclusion HIV prevalence was high among MSM SW and MSM with adverse life events. Given the characteristics of the sample, HIV prevalence among MSM in Jamaica is probably in the range of 20%. The study illustrates the importance of social vulnerability in driving the HIV epidemic. Programs to empower young MSM, reduce social vulnerability and other structural barriers including stigma and discrimination against MSM are critical to reduce HIV transmission.

Occupational radon exposure and lung cancer mortality: estimating intervention effects using the parametric g-formula.

Background: Traditional regression analysis techniques used to estimate associations between occupational radon exposure and lung cancer focus on estimating the effect of cumulative radon exposure on lung cancer. In contrast, public health interventions are typically based on regulating radon concentration rather than workers’ cumulative exposure. Estimating the effect of cumulative occupational exposure on lung cancer may be difficult in situations vulnerable to the healthy worker survivor bias. Methods: Workers in the Colorado Plateau Uranium Miners cohort (n = 4,134) entered the study between 1950 and 1964 and were followed for lung cancer mortality through 2005. We use the parametric g-formula to compare the observed lung cancer mortality to the potential lung cancer mortality had each of 3 policies to limit monthly radon exposure been in place throughout follow-up. Results: There were 617 lung cancer deaths over 135,275 person-years of follow-up. With no intervention on radon exposure, estimated lung cancer mortality by age 90 was 16%. Lung cancer mortality was reduced for all interventions considered, and larger reductions in lung cancer mortality were seen for interventions with lower monthly radon exposure limits. The most stringent guideline, the Mine Safety and Health Administration standard of 0.33 working-level months, reduced lung cancer mortality from 16% to 10% (risk ratio = 0.67 [95% confidence interval = 0.61 to 0.73]). Conclusions: This work illustrates the utility of the parametric g-formula for estimating the effects of policies regarding occupational exposures, particularly in situations vulnerable to the healthy worker survivor bias.

Generalizing Study Results: A Potential Outcomes Perspective

Great care is taken in epidemiologic studies to ensure the internal validity of causal effect estimates; however, external validity has received considerably less attention. When the study sample is not a random sample of the target population, the sample average treatment effect, even if internally valid, cannot usually be expected to equal the average treatment effect in the target population. The utility of an effect estimate for planning purposes and decision making will depend on the degree of departure from the true causal effect in the target population due to problems with both internal and external validity. Herein, we review concepts from recent literature on generalizability, one facet of external validity, using the potential outcomes framework. Identification conditions sufficient for external validity closely parallel identification conditions for internal validity, namely conditional exchangeability; positivity; the same distributions of the versions of treatment; no interference; and no measurement error. We also require correct model specification. Under these conditions, we discuss how a version of direct standardization (the g-formula, adjustment formula, or transport formula) or inverse probability weighting can be used to generalize a causal effect from a study sample to a well-defined target population, and demonstrate their application in an illustrative example.