Stata's mi commands provide powerful tools to conduct multiple imputation in the presence of ignorable missing data. In this article, I present Stata code to extend the capabilities of the mi commands to address two areas of statistical inference where results are not easily aggregated across imputed datasets. First, mi commands are restricted to covariate selection. I show how to address model fit to correctly specify a model. Second, the mi commands readily aggregate model-based standard errors. I show how standard errors can be bootstrapped for situations where model assumptions may not be met. I illustrate model specification and bootstrapping on frequency counts for the number of times that alcohol was consumed in data with missing observations from a behavioral intervention.

Mobile phone-based data collection encompasses the richness of social network research. Both individual-level and network-level measures can be recorded. For example, health-related behaviors can be reported via mobile assessment. Social interactions can be assessed by phone-log data. Yet the potential of mobile phone data collection has largely been untapped. This is especially true of egocentric studies in public health settings where mobile phones can enhance both data collection and intervention delivery, e.g. mobile users can video chat with counselors. This is due in part to privacy issues and other barriers that are more difficult to address outside of academic settings where most mobile research to date has taken place. In this article, we aim to inform a broader discussion on mobile research. In particular, benefits and challenges to mobile phone-based data collection are highlighted through our mobile phone-based pilot study that was conducted on egocentric networks of 12 gay men (n = 44 total participants). HIV-transmission and general health behaviors were reported through a mobile phone-based daily assessment that was administered through study participants' own mobile phones. Phone log information was collected from gay men with Android phones. Benefits and challenges to mobile implementation are discussed, along with the application of multi-level models to the type of longitudinal egocentric data that we collected.

Background

Relationships between alcohol, marijuana and other drug (AOD) use and contextual factors have mostly been established through retrospective self-report. Given the embeddedness of cell phones in adolescents' daily activities, cell phone-based ecological momentary assessment (CEMA) provides an opportunity to better understand AOD use in youth and how cell phones can be used to self-monitor and deliver interventions. We use CEMA to examine AOD use in Latino youth who have been especially understudied.

Methods

Twenty-eight mostly Latino youth (ages 13-18) in outpatient substance abuse treatment recorded AOD use, contextual factors, cravings, and affect through once-daily CEMA over one month periods. Random-effects logistic regression was used to compare contextual factors between periods of AOD use and non-use.

Results

The most frequent contextual factors reported during AOD use were being with close friends and "hanging out" as the primary activity. During AOD use compared to non-use, youth were more likely to be with close friends (OR=4.76; p<0.01), around users (OR=17.69; p<0.01), and at a friend's house (OR=5.97; p<0.01). Alcohol use was more frequently reported at night (63% vs 34%) and on weekends relative to other substances (64% vs 49%). Strong cravings were more frequently reported on AOD-use days (OR=7.34; p<0.01). Types of positive and negative affect were reported with similar frequencies, regardless of AOD use.

Conclusions

Reporting on social context, location, day and time of day, and cravings all show promise in developing cell phone-based interventions triggered by contextual data.

BACKGROUND

Health behavior patterns reported through daily diary data are important to understand and intervene upon at the individual level in N-of-1 trials and related study designs. There is often interest in relationships between multiple outcomes, such as stress and health behavior. However, analyses often utilize regressions that evaluate aggregate effects across individuals, and standard analyses target single outcomes.

OBJECTIVE

This paper aims to illustrate how individuals’ daily reports of stress and health behavior (time series) can be explored using visualization tools.

METHODS

Secondary analysis was conducted on 6 months of daily diary reports of stress and health behavior (physical activity and diet quality) from mostly ethnic minority mothers who pilot-tested a self-monitoring mobile health app. Time series with minimal missing data from 14 of the 44 mothers were analyzed. Correlations between stress and health behavior within each time series were reported as a preliminary step. Stress and health behavior time series patterns were visualized by plotting moving averages and time points where mean shifts in the data occurred (changepoints).

RESULTS

Median correlation was small and negative for associations of stress with physical activity (r=−.14) and diet quality (r=−.08). Moving averages and changepoints for stress and health behavior were aligned for some participants but not for others. A third subset of participants exhibited little variation in stress and health behavior reports.

CONCLUSIONS

Median correlations in this study corroborate prior findings. In addition, time series visualizations highlighted variations in stress and health behavior across individuals and time points, which are difficult to capture through correlations and regression-based summary measures.

Behavioral interventions are increasingly based on holistic approaches to health with an understanding that health-related behaviors are linked. A motivating example is provided by the Philani study, an intervention trial conducted to improve the health of South African mothers and their children. Inter-related health problems around maternal alcohol use, malnutrition, and HIV were addressed; multiple endpoints were targeted. The traditional hypothesis testing paradigm that tests significance on one primary outcome did not suffice. Past multiple endpoint studies have utilized a sign test on the number of estimated differences between treatment and control that favor the intervention. However, in order to preserve type 1 error, one must account for correlations among the outcomes. We propose an alternative approach that counts the number of significant treatment-control differences. Monte Carlo simulation is used to adjust for correlation, providing updated critical values and p values. Our method is implemented through an R package and applied to the Philani data to test the intervention's overall effect.

Background

Health behavior patterns reported through daily diary data are important to understand and intervene upon at the individual level in N-of-1 trials and related study designs. There is often interest in relationships between multiple outcomes, such as stress and health behavior. However, analyses often utilize regressions that evaluate aggregate effects across individuals, and standard analyses target single outcomes.

Objective

This paper aims to illustrate how individuals' daily reports of stress and health behavior (time series) can be explored using visualization tools.

Methods

Secondary analysis was conducted on 6 months of daily diary reports of stress and health behavior (physical activity and diet quality) from mostly ethnic minority mothers who pilot-tested a self-monitoring mobile health app. Time series with minimal missing data from 14 of the 44 mothers were analyzed. Correlations between stress and health behavior within each time series were reported as a preliminary step. Stress and health behavior time series patterns were visualized by plotting moving averages and time points where mean shifts in the data occurred (changepoints).

Results

Median correlation was small and negative for associations of stress with physical activity (r=-.14) and diet quality (r=-.08). Moving averages and changepoints for stress and health behavior were aligned for some participants but not for others. A third subset of participants exhibited little variation in stress and health behavior reports.

Conclusions

Median correlations in this study corroborate prior findings. In addition, time series visualizations highlighted variations in stress and health behavior across individuals and time points, which are difficult to capture through correlations and regression-based summary measures.

Stressful life events and daily hassles affect people living with HIV (PLWH). However, capturing stress-related events and cognitive impairment is often plagued with recall biases. Incorporating reliable information technology, such as mobile phones, can be a resourceful method for measuring health behaviors (MHB). We report findings from an MHB pilot study with 32 African American, Latino/Hispanic, and White PLWH from Los Angeles. Participants reported perceived stressors in their daily routines using a smartphone Ecological Momentary Assessment (EMA) application. Participants self-initiated in-the-moment stressful events reports for up to 6 weeks. Stressful event EMAs queried perceived stress levels (1-10 scale) and open-ended text descriptions. Qualitative analysis of participant text responses was completed using grounded thematic coding. Participants reported multiple stressors in their daily routines, impacting activities of daily living or daily functioning. Eliciting input from PLWH via EMA in real time is a novel approach for assessing and identifying sources of stress.

Stressful life events and daily hassles affect people living with HIV (PLWH). However, capturing stress-related events and cognitive impairment is often plagued with recall biases. Incorporating reliable information technology, such as mobile phones, can be a resourceful method for measuring health behaviors (MHB). We report findings from an MHB pilot study with 32 African American, Latino/Hispanic, and White PLWH from Los Angeles. Participants reported perceived stressors in their daily routines using a smartphone Ecological Momentary Assessment (EMA) application. Participants self-initiated in-the-moment stressful events reports for up to 6 weeks. Stressful event EMAs queried perceived stress levels (1-10 scale) and open-ended text descriptions. Qualitative analysis of participant text responses was completed using grounded thematic coding. Participants reported multiple stressors in their daily routines, impacting activities of daily living or daily functioning. Eliciting input from PLWH via EMA in real time is a novel approach for assessing and identifying sources of stress.

Background

Advancing technology has increased functionality and permitted more complex study designs for behavioral interventions. Investigators need to keep pace with these technological advances for electronic data capture (EDC) systems to be appropriately executed and utilized at full capacity in research settings. Mobile technology allows EDC systems to collect near real-time data from study participants, deliver intervention directly to participants' mobile devices, monitor staff activity, and facilitate near real-time decision making during study implementation.

Objective

This paper presents the infrastructure of an EDC system designed to support a multisite HIV biobehavioral intervention trial in Los Angeles and New Orleans: the Adolescent Medicine Trials Network "Comprehensive Adolescent Research & Engagement Studies" (ATN CARES). We provide an overview of how multiple EDC functions can be integrated into a single EDC system to support large-scale intervention trials.

Methods

The CARES EDC system is designed to monitor and document multiple study functions, including, screening, recruitment, retention, intervention delivery, and outcome assessment. Text messaging (short message service, SMS) and nearly all data collection are supported by the EDC system. The system functions on mobile phones, tablets, and Web browsers.

Results

ATN CARES is enrolling study participants and collecting baseline and follow-up data through the EDC system. Besides data collection, the EDC system is being used to generate multiple reports that inform recruitment planning, budgeting, intervention quality, and field staff supervision. The system is supporting both incoming and outgoing text messages (SMS) and offers high-level data security. Intervention design details are also influenced by EDC system platform capabilities and constraints. Challenges of using EDC systems are addressed through programming updates and training on how to improve data quality.

Conclusions

There are three key considerations in the development of an EDC system for an intervention trial. First, it needs to be decided whether the flexibility provided by the development of a study-specific, in-house EDC system is needed relative to the utilization of an existing commercial platform that requires less in-house programming expertise. Second, a single EDC system may not provide all functionality. ATN CARES is using a main EDC system for data collection, text messaging (SMS) interventions, and case management and a separate Web-based platform to support an online peer support intervention. Decisions need to be made regarding the functionality that is crucial for the EDC system to handle and what functionality can be handled by other systems. Third, data security is a priority but needs to be balanced with the need for flexible intervention delivery. For example, ATN CARES is delivering text messages (SMS) to study participants' mobile phones. EDC data security protocols should be developed under guidance from security experts and with formative consulting with the target study population as to their perceptions and needs.

International registered report identifier (irrid)

DERR1-10.2196/10777.

Background

Myanmar has adopted point-of-care (POC) HIV testing for its prevention of mother-to-child transmission of HIV program, and was initiated in 84 townships in 2013. This study assessed the progress of HIV testing uptake from 2012, one year prior to POC testing, to 2015, and the challenges faced by service providers during the rapid rollout of this testing strategy.

Methods

This serial cross-sectional study included 23 townships randomly selected from the 84 townships. An open-question survey was used to collect information on the challenges faced by service providers. A random effects logistic model was used for assessing the progress of HIV testing uptake among urban and rural health center groups.

Results

HIV testing uptake for antenatal care (ANC) attendees increased from 60% to 90% for rural and from 70% to 90% for urban attendees. The proportion of ANC attendees who were tested at their first visit increased from 70% to 80% for rural and from 70% to 90% for urban attendees. In addition, the proportion receiving same-day test results increased from less than 10% to 90% for both groups. Major challenges faced during the initial rollout included low health awareness among pregnant women, fear of stigma and discrimination, long travel times and costs, and increased workloads of providers in rural settings.

Conclusions and global health implications

The program should consider recruiting local volunteers to help reduce the workloads of service providers. Professional education based on need and continued mentoring and quality control schemes for HIV testing need to be in place. This decentralized strategy would be applicable to other resource-limited countries.