Using machine learning and qualitative interviews to design a five-question survey module for women's agency

Open-ended interview questions elicit rich information about people's lives, but in large-scale surveys, social scientists often need to measure complex concepts using only a few close-ended questions. We propose a new method to design a short survey measure for such cases by combining mixed-methods data collection and machine learning. We identify the best survey questions based on how well they predict a benchmark measure of the concept derived from qualitative interviews. We apply the method to create a survey module and index for women's agency. We measure agency for 209 married women in Haryana, India, first, through a semi-structured interview and, second, through a large set of close-ended questions. We use qualitative coding methods to score each woman's agency based on the interview, which we use as a benchmark measure of agency. To determine the close-ended questions most predictive of the benchmark, we apply statistical algorithms that build on LASSO and random forest but constrain how many variables are selected for the model (five in our case). The resulting five-question index is as strongly correlated with the coded qualitative interview as is an index that uses all of the candidate questions. This approach of selecting survey questions based on their statistical correspondence to coded qualitative interviews could be used to design short survey modules for many other latent constructs.

Prioritizing and Funding Nepal's Multisector Nutrition Plan

BACKGROUND

Nepal has a long tradition of designing good multisectoral nutrition policy. However, success of policy implementation has varied. More evidence on how to successfully carry out multisector nutrition policy is needed.

OBJECTIVE

We tracked the influence of Nepal's multisectoral nutrition plan (MSNP) on the process of priority setting and budgeting from 2014 to 2016.

METHODS

This study used a mixed-method longitudinal design to track qualitative and budgetary changes related to MSNP processes nationally as well as in 3 districts. Qualitative changes in each study area were assessed through interviews, observation, news content, and meeting notes. Changes in allocations and expenditures were calculated based on budget documents, work plans, and validation interviews.

RESULTS

Improved understanding of the MSNP was documented nationally and in study districts but not in VDCs. Human resources, ownership, bottom-up planning, coordination, advocacy, and sustainable structures all emerged as important factors within the enabling environment. Evidence suggests the MSNP influenced improvements in the last 3 factors. We also found notable increases in activities and financing for nutrition-allocations increased steadily between FY 2013-2014 and FY 2015-2016, and 28% of total nutrition allocations in the final year came from new or expanded MSNP-affiliated activities. Data from 3 districts highlight challenges linking local planning and budgeting to central-level structures.

CONCLUSIONS

The MSNP appears to have strengthened the nutrition system in Nepal and increased priority and funding for nutrition. Next steps include strengthening linkages to the districts and below. Other countries can learn from the MSNP's success in increasing investment for nutrition.

Non-barrier contraceptive use and relation to condom use behaviour by partner type among female sex workers in Andhra Pradesh, India

OBJECTIVE

The study assessed non-barrier contraceptive use among female sex workers (FSW) in Andhra Pradesh, India and relation to inconsistent condom use among commercial and non-commercial male sexual partners.

METHODS

FSW at least 18 years of age (n=2338) were recruited through respondent-driven sampling for an HIV risk survey. Analysis was restricted to women of childbearing age (n=2197). Crude and adjusted logistic regression models were used to assess non-barrier contraceptive use and relation to inconsistent condom use with husbands or regular male partners (i.e. non-clients), regular clients and occasional clients.

RESULTS

Non-barrier methods of contraception included contraceptive pills (3.8%) and sterilisation (68.4%). In logistic regression models adjusted for relevant demographics, FSW using contraceptive pills were more likely to report inconsistent condom use with a regular client (past week) [adjusted odds ratio (AOR) 2.2, 95% confidence interval (CI) 1.2-4.0] and with an occasional client (past week) (AOR 2.6, 95% CI 1.6-5.3), as well as accepting more money for sex without a condom (past 30 days) (AOR 2.5, 95% CI 1.5-4.3). No significant associations were found between pill use and inconsistent condom use among women's non-client partners, potentially related to small sample sizes within these subgroups. Reporting sterilisation, which was more common among FSW who were older in age, was not associated with inconsistent condom use with client or non-client sexual partners.

CONCLUSIONS

Findings document potential unmet need for modern, spacing contraceptives (i.e. pill, intrauterine device), but also indicate the importance for family planning services, particularly those promoting modern contraceptive methods to be provided alongside HIV prevention among FSW in Andhra Pradesh, India.

Client-perpetrated and husband-perpetrated violence among female sex workers in Andhra Pradesh, India: HIV/STI risk across personal and work contexts

Objectives

This study examines violence experienced in work and personal contexts and relation to HIV risk factors in these contexts among female sex workers (FSW) in Andhra Pradesh, India.

Methods

FSW at least 18 years of age (n=2335) were recruited through three rounds of respondent-driven sampling between 2006 and 2010 for a survey on HIV risk. Using crude and adjusted logistic regression models, any sexual/physical violence (last 6 months) perpetrated by clients and husbands were separately assessed in association with accepting more money for sex without a condom (last 30 days), consistent condom use with clients and husbands (last 30 days), and sexually transmitted infection (STI) symptoms (last 6 months).

Results

The mean age among participants was 32, 22% reported being currently married, and 22% and 21% reported physical/sexual violence by clients and husbands, respectively. In adjusted logistic regression models, FSW who experienced client violence were more likely to report accepting more money for unprotected sex trades (adjusted OR (AOR)=1.7; 95% CI 1.4 to 2.2), less likely to report consistent condom use with clients (AOR=0.6; 95% CI 0.5 to 0.7) and more likely to report STI symptoms (AOR=3.5; 95% CI 2.6 to 4.6). Women who reported husband violence were more likely to report accepting more money for unprotected sex trades (AOR=2.1; 95% CI 1.2 to 3.7), less likely to report consistent condom use with clients (AOR=0.5; 95% CI 0.3 to 0.8) and more likely to report STI symptoms (AOR=2.6; 95% CI 1.6 to 4.1).

Conclusions

Among FSW, experiences of violence in work and personal contexts are associated with sexual HIV risk behaviours with clients as well as STI symptoms.

Factors Influencing the Delivery of HIV-Related Services to Severely Mentally Ill Individuals: The Provider's Perspective

BACKGROUND

Individuals with severe mental illnesses (SMI) are disproportionately vulnerable to HIV infection but are not consistently engaged in HIV-related services.

OBJECTIVE

To understand factors influencing implementation of HIV-related services to individuals with SMI, we conducted a series of focus groups with multidisciplinary clinicians and staff serving individuals with SMI in outpatient, emergency, acute inpatient, and chronic inpatient levels of care.

METHOD

Six focus groups with 30 participants were conducted, audiotaped, and transcribed. Our qualitative analysis drew on Grounded Theory. Using NVivo Version 9, coding was conducted by the first and senior authors; interrater reliability was verified by running Coding Comparison queries.

RESULTS

The providers' narratives highlighted (1) patient-related factors, (2) stigma, and (3) administrative factors as themes particularly relevant to the delivery of HIV-related services to individuals with SMI. The reported relevance of these factors ranged across levels of care, from creating multiple barriers in the outpatient care to relatively seamless and effective delivery of full continuum of HIV-related services in the chronic inpatient environment, where adequate structural support is provided.

CONCLUSION

Providers' narratives suggest that effective delivery of HIV-related services for individuals with SMI requires sustained structural support that is coordinated across levels of psychiatric care and tailored to individual patient's needs. The narratives also suggest that such support is currently not available.

Migration/mobility and risk factors for HIV among female sex workers in Andhra Pradesh, India: implications for HIV prevention

We examined the relation between high mobility/migration (sex work in three or more villages/towns within the past year) and HIV risk factors among a sample of female sex workers (FSWs) in Andhra Pradesh, India. We recruited FSWs aged ≥18 years (n = 673) through respondent-driven sampling for a survey on HIV risk. Adjusted logistic and linear regression models assessed high mobility in relation to sexual and physical victimization, sexually transmitted infection (STI) symptoms and treatment, condom use and negotiation, number and/or types of sex trades, number of clients and number of days worked. Twelve percent (n = 82) of FSWs were highly mobile; those with high mobility were more likely to report recent HIV risk factors: sexual violence (adjusted odds ratio [AOR] = 5.2; 95% confidence interval [CI]: 3.0-8.9), physical violence (AOR = 1.7; 95% CI: 1.1-2.7), unprotected sex for more money (AOR = 1.7; 95% CI: 1.1-3.0), at least one STI symptom (AOR = 1.9; 95% CI: 1.1-3.1), a greater number of vaginal sex trades (β = 3.9, P = 0.003), a greater number of clients (β = 2.5, P = 0.02) and anal sex with clients (AOR = 2.4; 95% CI: 1.4-4.1). Findings from this study underscore the violence and HIV-related vulnerability faced by mobile/migrant FSWs and highlight the need to inform and tailor related prevention strategies.

Trends in condom use among female sex workers in Andhra Pradesh, India: the impact of a community mobilisation intervention

BACKGROUND

Community mobilisation interventions for HIV prevention among female sex workers (FSWs) aim to organise FSWs for collective action and challenge the structures of power that underlie HIV risk. Assessing intervention impact is challenging because the importance of direct individual exposure to intervention components may decrease over time as change occurs at social-normative, policy and other structural levels. In this paper, the authors examine changes over time in consistent condom use among FSWs in Rajahmundry, Andhra Pradesh, the location of a long-standing community mobilisation intervention.

METHODS

The authors analyse cross-sectional data collected among FSWs at three time points (n=2276) using respondent-driven sampling. Multiple logistic regression was used to assess the association of programme exposure with consistent condom use and whether this association varied over time.

RESULTS

The proportion of FSWs having no exposure or only receptive exposure to the intervention decreased over time, while active utilisation increased from 19.4% in 2006 to 48.5% in 2009-2010. Consistent condom use with clients also increased from 56.3% in 2006 to 75.3% in 2009-2010. Multivariate analysis showed that age, age at start of sex work, venue, living conditions and programme exposure were significantly associated with condom use. The positive association between programme exposure and consistent condom use did not vary significantly over time.

CONCLUSIONS

Findings indicate improvements in HIV risk reduction behaviour among FSWs and suggest that the intervention has substantial reach in the FSW population. The intervention's strategies may be contributing to population-level HIV risk reduction among FSWs.

The role of housing in determining HIV risk among female sex workers in Andhra Pradesh, India: considering women's life contexts

Recent research on HIV prevention, regardless of the population, has increasingly recognized the relevance of contextual factors in determining HIV risk. Investigating such factors among female sex workers (FSW) is especially relevant in the South Indian state of Andhra Pradesh, where HIV rates are among the highest across Indian states and where HIV has largely affected FSW. Stable housing is a particular contextual challenge experienced by female sex workers in this region (as well as elsewhere); however, local studies have not examined the impact of this issue on HIV risk. In this paper, we examine residential instability, defined as a high frequency of reported evictions, among FSW and relation to experiences of violence (as a factor increasing risk for HIV) and sexual risk factors for HIV. Women were recruited through respondent-driven sampling for a survey on HIV risk. Using logistic regression models, we assessed: (1) residential instability and association with HIV sexual risk variables (including unprotected sex, reported STIs, and recent physical and sexual victimization) and (2) whether the association between residential instability and reported STI (as an indicator of HIV risk) was attenuated by individual risk behaviors and violence. In adjusted logistic regression models, FSW who reported residential instability were more likely to report: sexual violence, physical violence, accepting more money for unprotected sex, and a recent STI symptom. Violence associated with residential instability contributed to reported STIs; however, residential instability remained significantly associated with STIs beyond the influence of both violence and unprotected sex with clients. Findings highlight the interrelation among residential instability, violence, and HIV risk. Residential instability appears to be associated with women's HIV risk, above and beyond its association with individual risky sexual behaviors.

The context of economic insecurity and its relation to violence and risk factors for HIV among female sex workers in Andhra Pradesh, India

OBJECTIVE

We examined the context of economic insecurity and debt among female sex workers (FSWs), how this context varies among FSWs, and its association with experiences of violence and sexual risk factors for human immunodeficiency virus (HIV).

METHODS

We recruited FSWs aged > or =18 years (n = 673) through respondent-driven sampling for a survey on HIV risk in this region. Using logistic regression models (adjusted for partner status, education, financial support, and literacy), we assessed the relation between debt and sexual and physical victimization as well as sexual risk. We also conducted qualitative interviews with a subsample of the survey participants and examined these for related themes.

RESULTS

In adjusted logistic regression models, FSWs who reported debt were more likely to report the following: recent physical violence (adjusted odds ratio [AOR] = 2.4, 95% confidence interval [CI] 1.5, 3.9), unprotected sex with occasional clients in the past week (AOR = 2.3, 95% CI 1.2, 4.3), anal sex with clients in the past 30 days (AOR = 2.0, 95% CI 1.1, 3.9), and at least one sexually transmitted infection symptom in the past six months (AOR = 1.6, 95% CI 1.1, 2.4). FSWs with debt were more likely to report current husbands or other male partners, and less likely to report condom use with these partners, further increasing their sexual risk. Qualitative data elaborated on these findings.

CONCLUSION

Findings indicate the violence- and HIV-related vulnerability of FSWs who report debt and further highlight how male partners may contribute to the debt and economic insecurity of FSWs.

Informing women about their breast cancer risks: truth and consequences

We assessed the extent to which informing women about their risk for breast cancer affected their perceived 10-year and lifetime risks for getting breast cancer, their emotional reactions toward getting breast cancer, and their intentions to get mammograms. In a pre- to posttest design, 121 women were given their 10-year risk of getting breast cancer with or without being compared with women their age and race at lowest risk. Women's perceptions of their 10-year risks became more congruent (i.e., more accurate) with their actual risk. Participants were more accurate when they received their own risk without being compared with women at lowest risk. Women who received only their own risk estimate reported being at lower risk than other women. Overall, women reported that obtaining their 10-year risk estimate either did not affect or increased their intentions to get mammograms. These results suggest that giving women their individual risk of getting breast cancer improves accuracy while also enhancing their feelings that they are at lower risk than other women. Counter to many theories of health behavior, reducing women's perceived risk of breast cancer did not lower their intentions to get mammograms. Implications for the communication of breast cancer risk are discussed.

Testing different formats for communicating colorectal cancer risk

This study assessed the extent to which different formats of informing men and women age 50 and over of the risks of colorectal cancer (CRC) affected their perceptions of their absolute and comparative (self versus other) 10-year and lifetime risks; emotional reactions about getting CRC; and screening intentions. Forty-four men and 78 women received information about the absolute lifetime risk of getting CRC. In addition, participants either did or did not receive information about (1) lifetime risk of getting CRC compared with other cancers, and (2) risk factors for CRC (age and polyps). Participants who received risk factors information were more likely to increase their perceived absolute 10-year and lifetime risks of getting CRC compared with participants who did not receive risk factors information. In addition, participants who received risk factors information were more likely to believe age was related to getting CRC and felt at greater risk for having polyps compared with participants who did not receive this information. None of the experimental conditions affected how worried, anxious, and fearful participants felt about getting CRC, nor did they affect screening intentions. Independent of experimental condition, participants tended to increase their intentions to get screened for CRC in the next year or two. Intention to be screened was more pronounced among participants who had been screened via a fecal occult blood test (FOBT) or sigmoidoscopy (SIG). Implications for the design of interventions involving the communication of CRC risks are discussed.