Simultaneous Treatment of Missing Data and Measurement Error in HIV Research Using Multiple Overimputation

Schomaker, Michael; Hogger, Sarah; Johnson, Leigh F.; Hoffmann, Christopher J.; Bärnighausen, Till; Heumann, Christian

doi:10.1097/ede.0000000000000334

Cited by 6 publications

(12 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Setting 3: This setting is inspired by the analysis and data in Schomaker et al [33]. We simulate X 1 ∼ logN(4.286, 1.086) and X 2 ∼ logN(10.76, 1.8086).…”

Section: Settingmentioning

confidence: 99%

Bootstrap inference when using multiple imputation

Schomaker

Heumann

2018

Statistics in Medicine

Self Cite

306

242

View full text Add to dashboard Cite

Many modern estimators require bootstrapping to calculate confidence intervals because either no analytic standard error is available or the distribution of the parameter of interest is nonsymmetric. It remains however unclear how to obtain valid bootstrap inference when dealing with multiple imputation to address missing data. We present 4 methods that are intuitively appealing, easy to implement, and combine bootstrap estimation with multiple imputation. We show that 3 of the 4 approaches yield valid inference, but that the performance of the methods varies with respect to the number of imputed data sets and the extent of missingness. Simulation studies reveal the behavior of our approaches in finite samples. A topical analysis from HIV treatment research, which determines the optimal timing of antiretroviral treatment initiation in young children, demonstrates the practical implications of the 4 methods in a sophisticated and realistic setting. This analysis suffers from missing data and uses the g-formula for inference, a method for which no standard errors are available.

show abstract

“…Setting 3: This setting is inspired by the analysis and data in Schomaker et al [33]. We simulate X 1 ∼ logN(4.286, 1.086) and X 2 ∼ logN(10.76, 1.8086).…”

Section: Settingmentioning

confidence: 99%

Bootstrap inference when using multiple imputation

Schomaker

Heumann

2018

Statistics in Medicine

Self Cite

306

242

View full text Add to dashboard Cite

show abstract

“…Practical and technical details for multiple overimputation has been described elsewhere. (19-23) Our goal is to illustrate the use of multiple overimputation to address missing data and measurement error in gestational age. Briefly, multiple overimputation addresses missing data in the same way as multiple imputation: missing values are multiply imputed based on observed covariates.…”

Section: Methodsmentioning

confidence: 99%

Multiple Overimputation to Address Missing Data and Measurement Error

et al. 2016

View full text Add to dashboard Cite

Background Investigations of the association of combination antiretroviral therapy (ART) with pregnancy outcomes often rely on routinely collected clinical data, which are prone to missing data and measurement error. Measurement error in gestational age may bias the relationship between combination ART and gestational age-based outcomes. Methods We demonstrate the use of multiple overimputation to address missing data and measurement error in gestational age. Using routinely collected clinical data from public health facilities in Lusaka, Zambia, we multiply imputed missing data and multiply overimputed observed values of gestational age. Poisson models with robust variance estimators were used to estimate risk ratios (RRs) for the associations of duration of combination ART with small for gestational age (SGA) and preterm birth. We compared results from a complete-case analysis, using multiple imputation to address missing data only, and using multiple overimputation to address missing data and measurement error. Results In the complete-case analysis, there was no evidence of an association between duration of combination ART and SGA or preterm birth. When we performed multiple overimputation, RRs for SGA moved past the null, but remained imprecise. For preterm birth, RRs for 9-32 weeks of combination ART moved away from the null as the variance due to measurement error increased. Conclusion When we used multiple overimputation to account for measurement error and missing data, we observed an increased risk of preterm birth with longer duration of combination ART. Future analyses examining associations between combination ART and pregnancy outcomes should consider using multiple overimputation to address measurement error in gestational age.

show abstract

“…The mean of this distribution is 4.17 in women (based on a review of viral load distributions in early disease in East Africa and Southern Africa [394]) and 4.36 in men (studies suggest that levels of viral load in early infection tend to be higher in men than in women [395], and the difference of 0.19 is the lower confidence limit of the difference estimated in an analysis of viral load data from across Europe [396]). The standard deviation for the normal distribution is set at the same value (0.9) for males and females, based on the reported inter-quartile range in a study of Kenyan sex workers [397], after correcting for variation due to measurement error (assumed to be 0.25 2 on the log10 scale [398]).…”

Section: Hiv Disease Progressionmentioning

confidence: 99%

“…For women we have assumed a mean of 6.54, which is 0.1 less than the average of the two studies – this slight reduction in the gender differential is incorporated as the gender difference of 0.2 estimated from the KwaZulu-Natal data is greater than that estimated in other studies in HIV-negative adults [405–407]. The standard deviation of the normal distribution from which the initial CD4 counts are sampled is set at 0.219 in men and 0.224 in women, based on the inter-quartile ranges of the CD4 distributions observed in HIV-negative adults in KwaZulu-Natal (after correcting for variation due to measurement error, assumed to be 0.25 2 on the natural log scale [398]).…”

Section: Hiv Disease Progressionmentioning

confidence: 99%

MicroCOSM: a model of social and structural drivers of HIV and interventions to reduce HIV incidence in high-risk populations in South Africa

Johnson

Kubjane

Moolla

2018

Preprint

Self Cite

View full text Add to dashboard Cite

Executive summary Background and objectives South Africa has one of the highest HIV incidence rates in the world. Although much research has focused on developing biomedical strategies to reduce HIV incidence, there has been less investment in prevention strategies that address the social drivers of HIV spread. Understanding the social determinants of HIV is closely related to understanding high-risk populations (‘key populations’), since many of the factors that place these key populations at high HIV risk are social and behavioural rather than biological. Mathematical models have an important role to play in evaluating the potential impact of new HIV prevention and treatment strategies. However, most of the mathematical modelling studies that have been published to date have evaluated biomedical HIV prevention strategies, and relatively few models have been developed to understand the role of social determinants or interventions that address these social drivers. In addition, many of the mathematical models that have been developed are relatively simple deterministic models, which are not well suited to simulating the complex causal pathways that link many of the social drivers to HIV incidence. The frequency-dependent assumption implicit in most deterministic models also leads to under-estimation of the contribution of high-risk groups to the incidence of HIV. Agent-based models (ABMs) overcome many of the limitations of deterministic models, although at the expense of greater computational burden. This study presents an ABM of HIV in South Africa, developed to characterize the key social drivers of HIV in South Africa and the groups that are at the highest risk of HIV. The objective of this report is to provide a technical description of the model and to explain how the model has been calibrated to South African data sources; future publications will assess the drivers of HIV transmission in South Africa in more detail. Methods The model is an extension of a previously-published ABM of HIV and other sexually transmitted infections (STIs) in South Africa. This model simulates a representative sample of the South African population, starting from 1985, with an initial sample size of 20 000. The population changes in size as a result of births and deaths. Each individual is assigned a date of birth, sex and race (demographic characteristics). This in turn affects the assignment of socio-economic variables. Each individual is assigned a level of educational attainment, which is dynamically updated as youth progress through school and tertiary education, with rates of progression and drop-out depending on the individual’s demographic characteristics. Each individual is also assigned to an urban or rural location, with rates of movement between urban and rural areas depending on demographic characteristics and educational attainment. The model assigns to each individual a number of healthcare access variables that determine their HIV and pregnancy risk. These include their ‘condom preference’ (a measure of the e...

show abstract

Simultaneous Treatment of Missing Data and Measurement Error in HIV Research Using Multiple Overimputation

Cited by 6 publications

References 48 publications

Bootstrap inference when using multiple imputation

Bootstrap inference when using multiple imputation

Multiple Overimputation to Address Missing Data and Measurement Error

MicroCOSM: a model of social and structural drivers of HIV and interventions to reduce HIV incidence in high-risk populations in South Africa

Contact Info

Product

Resources

About