Secondary Logo

Journal Logo

Supplement Article

Is Interview Length Associated With Blood Test Participation? Evidence From Three Population-Based HIV Impact Assessment Surveys Conducted From 2016 to 2017

Bray, Rachel MSa; Palma, Anton M. PhDa,b; Philip, Neena M. PhDa; Brown, Kristin MPHc; Levin, Bruce PhDd; Thompson, John L. P. (Seamus) PhDd; Ginindza, Choice MSc, BState; Mulenga, Lloyd B. PhDf

Author Information
JAIDS Journal of Acquired Immune Deficiency Syndromes: August 1, 2021 - Volume 87 - Issue - p S57-S66
doi: 10.1097/QAI.0000000000002705

Abstract

BACKGROUND

Population-based health surveys are essential tools to assess global progress and impact of national health programs.1–3 Biomarker data, such as those obtained from a blood draw, are increasingly of primary interest; thus, given the high cost of large-scale biomarker collection, obtaining high response rates is critical to ensure that findings are a representative of the target population. However, these surveys tend to have low rates of participation in blood testing. For example, Demographic and Health Surveys (DHS) observe blood test response rates typically ranging from 70% to 85%, compared with household response rates above 95% and interview response rates between 80% and 95%.4 Maximizing blood test participation remains an area of utmost importance.

Surveys with a biomarker component often involve an interview before testing, during which a survey questionnaire is administered. There is evidence that the length of the interview may influence willingness to participate in biomarker testing. Dillman5 and Deutskens6 observed associations between long questionnaires and higher rates of midsurvey drop-off (leading to partially complete surveys) and item-level nonresponse. Lopez and Walsh found in a survey of multiple persons within a household, the length of the first person's interview predicted nonresponse for subsequent household members.7 Sharp and Frankel8 found that respondents who underwent longer interviews were less willing to participate in future surveys. Reducing interview length through shorter questionnaire design has the potential to improve blood test participation, at the expense of lost opportunities to collect potentially useful questionnaire data.

An individual's choice to participate in a survey and subsequent blood testing is influenced by a number of factors, including cultural context, individual beliefs, and the perceived benefits and burden of participation. In an analysis of blood test participation in the DHS for 14 countries, individuals who were wealthier, more educated, and living in urban areas were less likely to participate in HIV testing; and although the authors hypothesized otherwise, HIV test participation was no different for chronically ill individuals.4 Food insecurity within the household may be another reason that an individual chooses not to participate in the blood test.

In HIV biomarker surveys, a participant's knowledge of their HIV status may also influence participation in blood testing. Reniers and Eaton9 and Larmange et al 10 found that people with prior knowledge of their HIV-positive status were less likely to participate in surveys. Reniers and Eaton9 showed that this resulted in a negative bias in HIV prevalence estimates. Individuals who already know that they are HIV-positive have lower incentive to participate in a survey that includes HIV testing. On the other hand, if HIV testing includes additional biomarkers such as CD4 and viral load testing, these additional tests may incentivize participation among HIV-positive persons. Among those who do not know their HIV status, participating in the interview, which likely includes HIV-related questions on knowledge, attitudes, and service uptake, may instead increase their interest in HIV testing to learn about their status. This could be attributed to a learning hypothesis, wherein individuals who believe they are HIV-negative or do not know their status would have increased personal interest in the survey topic and greater incentive to complete the interview and receive HIV testing, a concept described by some authors as salience.11 We hypothesized that this would attenuate effects of long interviews on blood test participation, compared with persons previously diagnosed with HIV.

There is no evidence, to the best of our knowledge, of the impact of interview length on blood test participation in the context of HIV biomarker surveys conducted in low-resource settings.12–14 This analysis aimed to explore this relationship using data from 3 Population-based HIV Impact Assessments (PHIAs) conducted in Zambia,15 Eswatini16 (formerly Swaziland), and Lesotho17 in 2016–2017. Consent for the blood test was obtained before the interview in Eswatini and after the interview in Zambia and Lesotho. Therefore, these surveys presented a unique opportunity to test whether interview length was associated with the likelihood of participating in blood testing. We hypothesized a negative association between interview length and blood test participation in Zambia and Lesotho. In Eswatini, where respondents gave blood test consent before the interview taking place, we hypothesized no relationship between interview length and blood test participation.

METHODS

Data Source

The PHIAs are nationally representative surveys designed to assess the reach and impact of HIV programs in PEPFAR-supported countries. Each PHIA survey consisted of a household interview in which the household roster was constructed, followed by an individual interview for adults aged 15 and older in the randomly selected household. The interview was followed by collection of venous blood from consenting participants for HIV testing. HIV rapid testing, in line with national rapid test algorithms, was conducted in the household with immediate return of test results. Individuals who tested HIV-seropositive were offered referral for HIV care and treatment services, as well as point-of-care CD4 testing. Individual interviews, counseling, testing, and return of test results were all conducted privately, within the home, or around the home. Laboratory-based HIV viral load testing was conducted on transported specimens with results returned to a health facility of the respondent's choice for consultation with a health care provider.

The interview consisted of a core set of questions consistent across countries, plus additional questions tailored to the country setting. The questionnaire consisted of several modules: demographics, reproductive history, children, HIV testing, care and treatment history, tuberculosis, and other HIV-related risk factors (male circumcision, sexual activity, HIV knowledge, gender norms, and violence). Selected modules were administered to a subset of respondents, based on random selection and/or demographic criteria. Interviewers trained in Good Clinical Practice18 and survey procedures administered the survey through an electronic tablet. The surveys were approved by the Institutional Review Boards at Columbia University Irving Medical Center, the Centers for Disease Control and Prevention, and the Local Ethics Boards in Eswatini (National Health Research and Review Board), Lesotho (Lesotho Ministry of Health Research and Ethics Committee), and Zambia (Zambia Tropical Diseases Research Center Institutional Review Board).

Sample

The sample included participants from 3 PHIAs conducted in 2016–2017: the Zambia PHIA (ZAMPHIA 2016), Swaziland HIV Incidence Measurement Survey 2 (SHIMS2) conducted in Eswatini, and Lesotho PHIA (LEPHIA 2016–2017). We included all adults with reported age of 15–59 years. Participants who did not complete the first question of the tuberculosis module (ie, the last module consistently administered across all survey countries) were excluded from the analysis because biomarker data were not collected from participants with incomplete interviews (Fig. 1 for eligibility flowchart). In Eswatini and Zambia over 99% and in Lesotho around 90% of participants who started the interview reached the tuberculosis module.

FIGURE 1.
FIGURE 1.:
Eligibility criteria for inclusion in questionnaire length and interview duration analyses, ZIMPHIA 2016, SHIMS2 2016–2017, and LePHIA 2016–2017. 1Adult respondents were eligible for questionnaire length analysis if they had complete data on blood test participation and all demographic variables and did not drop out of the survey early, as defined by having a valid response to the first question of the last required module (tuberculosis module). 2Eligibility for interview duration analysis excluded participants with interviews longer than 120 minutes and those who on average answered more than 20 questions per minute.

Measures

We quantified interview length in 2 ways: (1) “questionnaire length”, the number of questions answered by the participant in the individual interview and (2) “interview duration”, the difference in minutes between the time of initial consent to the individual interview and the time of completing the individual interview, which were captured automatically by the tablet. Questions answered in the household interview and the time spent on the household interview were not included in the interview length measures. Participants whose interviews were longer than 120 minutes or who answered more than 20 questions per minute on average were determined to have invalid timestamps and were excluded for analyses on interview duration. We defined blood test participation as having consented to and provided a blood sample, regardless of whether HIV test results were determinate. In Eswatini, individuals were given the opportunity to change their consent after the interview, but few respondents changed their consent status (0.05% of those who initially consented subsequently withdrew and 0.36% of those who initially refused subsequently consented).

We considered additional variables that could influence interview length and blood test participation as potential covariates. These included demographic and other characteristics that influence survey participation such as age, gender, urban/rural residence, marital status, education, household wealth quintile, food insecurity within the household, and employment in the past 12 months. Wealth quintile is relative to other individuals within the country. Food insecurity in the household is defined as households where the household respondent indicated in the past 4 weeks there was ever no food to eat of any kind in their household because of lack of resources to get food. A household head indicator was also included to account for the effort household respondents had expended in answering the household survey, before their individual interview. Survey language captures information about the survey administration and serves as a proxy for ethnicity. The analysis also included factors that determine eligibility to receive certain questionnaire modules, including self-reported HIV status, whether an individual reported ever having sex, household size, and the number of children for which the respondent provided information in the survey. Self-reported HIV status was comprised of 3 categories: self-reported HIV-positive, self-reported HIV-negative, and unknown HIV status (individuals who never tested or did not know their result). Participants who refused to answer questions about their HIV testing history were excluded because we were unable to determine their self-reported HIV status. Finally, among individuals self-reporting HIV-positive and HIV-negative, we examined recent testing as whether they were tested in the 12 months before the survey, more than 12 months before the survey, or had unknown date information. For individuals with unknown self-reported HIV status, we also examined whether they were ever tested.

Analysis

We conducted descriptive analyses of all analytic variables, stratified by country. Bivariate associations between blood test participation with questionnaire length, interview duration, and selected demographics were described using the χ2 tests of association. We assessed the effects of questionnaire length and interview duration on blood test participation using logistic regression to estimate odds ratios (ORs) and 95% confidence intervals (CI), adjusted for all other covariates. Covariates were selected using backward elimination with a P-value cut-off of P < 0.10. To maintain consistency between countries, covariates that were significant in any country were retained in all models. As a secondary analysis, we explored whether effect estimates varied by self-reported HIV status. An additional covariate of whether the individual had ever tested was considered for inclusion in the unknown status model, whereas an indicator of whether the respondent tested in the past 12 months was included in the self-reported HIV-positive and HIV-negative models. The analyses did not use survey weights because the current research questions were focused on survey operations and thus did not require inferring to the national population.

In Zambia and Lesotho, the blood consent was obtained after the interview, and what happened during the interview process could influence whether an individual consented to the blood draw. In Eswatini, consent for blood testing was obtained directly after interview consent and before the interview took place, so any observed effects on blood test participation could not be due to the interview process. As such, Eswatini is included to serve as a test of the assumption of no unmeasured confounding.

RESULTS

The household response rate was 89.1% in Zambia, 84.9% in Eswatini, and 93.2% in Lesotho. The interview response rate among adults age 15–59 was 82.8% in Zambia, 90.4% in Eswatini, and 91.9% in Lesotho.16,19,20

The total eligible sample consisted of 44,153 adults aged 15–59 years, n = 20,940 in Zambia, n = 10,218 in Eswatini, and n = 12,995 in Lesotho. The combined sample was 42.1% men, 37.6% urban, and had a median age of 29 (range 15–59, IQR 21–40), with 49.3% of the sample married or living with a partner as if married, 85.6% having ever had sex, 25.6% in food insecure households, 35.6% having worked in the past 12 months, and 38.5% reporting for at least one child during the interview. The median household size was 5 (range 1–29, IQR 3–7), and 37.2% of individuals responded to the household interview as household head. Of respondents reporting on at least one child, the median number of reported children was 3 (range 1–18, IQR 2–4). Overall, 15.6% of respondents self-reported HIV-positive, and 98.8% of respondents who self-reported HIV-positive tested as HIV-seropositive across the 3 countries. The median interview duration was 16 minutes (range 2–117, IQR 11–24), and the median questionnaire length was 77 questions (range 21–258, IQR 57–100). Blood test participation rates were 90.3% in Zambia, 93.8% in Eswatini, and 91.0% in Lesotho (Table 1).

TABLE 1. - Sample Characteristics by Country, ZAMPHIA 2016, SHIMS2 2016–2017, and LePHIA 2016–2017
Characteristic ZAMPHIA, 2016 (n = 20,940) SHIMS2, 2016–2017 (n = 10,218) LePHIA, 2016–2017 (n = 12,995)
Residence
 Urban, n (%) 9142 (43.7) 2347 (23) 5126 (39.4)
 Peri-urban, n (%) 895 (6.9)
Male, n (%) 8815 (42.1) 4363 (42.7) 5405 (41.6)
Age, median (IQR) 29 (21–39) 29 (21–39) 30 (22–40)
Marital status
 Never married, n (%) 7044 (33.6) 5580 (54.6) 4851 (37.3)
 Married or living together, n (%) 11,728 (56.0) 3792 (37.1) 6234 (48.0)
 Divorced, separated, or widowed, n (%) 2168 (10.4) 846 (8.3) 1910 (14.7)
Self-reported HIV status
 HIV-positive, n (%) 1727 (8.2) 2495 (24.4) 2670 (20.5)
 HIV-negative, n (%) 13,641 (65.1) 6387 (62.5) 8653 (66.6)
 Unknown HIV status, n (%) 5572 (26.6) 1336 (13.1) 1672 (12.9)
Ever had sex, n (%) 18,153 (86.7) 8250 (80.7) 11,379 (87.6)
Household size, median (IQR) 6 (4–8) 5 (3–7) 4 (3–6)
Worked in the past 12 months, n (%) 6749 (32.2) 4233 (41.4) 4754 (36.6)
Reported on at least one child during the interview, n (%) 9329 (44.6) 3056 (29.9) 4608 (35.5)
Household head, n (%) 6803 (32.5) 3767 (36.9) 5869 (45.2)
Food insecurity in household, n (%) 3503 (16.7) 3625 (35.5) 4184 (32.2)
Wealth quintile
 Lowest, n (%) 3317 (15.8) 2264 (22.2) 2515 (19.4)
 Second, n (%) 3869 (18.5) 2145 (21.0) 2561 (19.7)
 Middle, n (%) 4252 (20.3) 2376 (23.3) 2574 (19.8)
 Fourth, n (%) 4497 (21.5) 1689 (16.5) 2629 (20.2)
 Highest, n (%) 5005 (23.9) 1744 (17.1) 2716 (20.9)
Education
 No education, n (%) 1093 (5.2) 382 (3.7) 621 (4.8)
 Primary, n (%) 9061 (43.3) 2812 (27.5) 5260 (40.5)
 Secondary, n (%) 9131 (43.6) 3104 (30.4) 5725 (44.1)
 More than secondary, n (%) 1655 (7.9) 3920 (38.4) 1389 (10.7)
Language
 English, n (%) 3433 (16.4) 312 (3.1) 345 (2.7)
 Sesotho, n (%) 12,650 (97.3)
 Siswati, n (%) 9906 (96.9)
 Bemba, n (%) 8913 (42.6)
 Nyanja, n (%) 4188 (20.0)
 Lozi, n (%) 1008 (4.8)
 Tonga, n (%) 1831 (8.7)
 Lunda, n (%) 582 (2.8)
 Luvale, n (%) 415 (2.0)
 Kaonde, n (%) 497 (2.4)
 Others, n (%) 73 (0.3)
Interview length
 Questionnaire length, median (IQR) 85 (63–115) 62 (49–80) 79 (63–97)
 Interview duration in minutes, median (IQR) 19 (12–27) 12 (8–18) 16 (11–23)
Participated in blood testing 18,918 (90.3) 9580 (93.8) 11,823 (91.0)

In unadjusted analyses, there was no significant association between blood test participation and interview duration in Zambia (χ2 = 8.7, P-value 0.27), Lesotho (χ2 = 9.9, P-value 0.19), or Eswatini (χ2 = 7.6, P-value 0.36). Blood test participation was associated with the questionnaire length in all 3 countries, with blood test participation increasing with the questionnaire length (Table 2).

TABLE 2. - Blood Test Participation Rates by Interview Length and Selected Demographics, ZAMPHIA 2016, SHIMS2 2016–2017, and LePHIA 2016–2017
ZAMPHIA, 2016 SHIMS2, 2016-2017 LePHIA, 2016-2017
Blood Test Participation (%) N Blood Test Participation (%) N Blood Test Participation (%) N
Interview duration
 Less than 10 min 89.9 3349 94.3 3322 91.8 2541
 10–20 89.9 7728 93 4780 91 5832
 20–30 90.6 5703 94.7 1515 90.5 3100
 30–40 90.5 2444 93.8 320 90.3 968
 40–50 92.3 911 95.1 82 92.6 258
 50–60 91.4 360 93.5 31 90.3 72
 60+ 91.5 259 96.7 30 88.6 70
 Missing 92.5 186 94.9 138 87 154
χ2 (P-value) 8.7 (0.27) 9.9 (0.19) 7.6 (0.36)
Questionnaire length
 Less than 40 questions 89.7 1832 95.2 1816 90.8 663
 40–59 88.8 2977 92.6 2701 89.7 2172
 60–79 88.7 4237 92.6 3144 88.9 3774
 80–99 90.5 4153 94.2 1702 91.4 3568
 100–119 91.8 3122 97.5 600 93.7 1978
 120–139 91.5 2557 98.0 203 95.4 655
 140–159 92.8 1402 100.0 37 97.5 159
 160+ 92.3 660 100.0 15 96.2 26
χ2 (P-value) 46.3 (<0.01) 44.5 (<0.01) 67.6 (<0.01)
Self-reported HIV status
 HIV-positive 97.5 1727 97.8 2495 96.7 2670
 HIV-negative 89.4 13,641 92.6 6387 89.8 8653
 Unknown status 90.4 5572 91.5 1336 87.7 1672
χ2 (P-value) 113.8 (<0.01) 94.3 (<0.01) 141.7 (<0.01)
Gender
 Male 89.5 8815 92.0 4363 89.2 5405
 Female 90.9 12,125 95.0 5855 92.3 7590
χ2 (P-value) 11.6 (<0.01) 38.0 (<0.01) 36.7 (<0.01)

Across all 3 countries, self-reported HIV-positive individuals had significantly higher blood test participation rates (97.3%) compared with those who self-reported HIV-negative or had unknown status (90.2%; t = 28.39, P-value < 0.01). Women were more likely to participate in blood testing than men (92.3% vs 90.0%, t = 8.04, P-value < 0.01). Women and individuals who reported as HIV-positive had longer questionnaires and interview duration (see Supplemental Table 1, Supplemental Digital Content, http://links.lww.com/QAI/B668).

Results from multivariable logistic regression models are presented in Table 3. Effects for marital status, household food insecurity, and urban/rural residence were not included because they were not significant in any of the models. After controlling for the demographic covariates, the odds of participating in biomarker testing increased by 10% for every additional 20 questions answered by participants in Zambia (adjusted OR, 95% CI: 1.10, 1.05 to 1.16). By contrast, longer interview duration was associated with decreased participation in biomarker testing in Lesotho (OR per additional 5 minutes, 95% CI: 0.94, 0.89 to 0.99) and Eswatini (OR, 95% CI: 0.94, 0.91 to 0.97). Self-reported HIV status had a strong effect on blood test participation after controlling for either questionnaire length or interview duration and other demographic factors, with self-reported HIV-negative individuals and those with unknown status being much less likely to consent, compared with those who self-reported HIV-positive.

TABLE 3. - Adjusted Odds Ratios (95% CI) of Blood Test Participation by Interview Length, ZAMPHIA 2016, SHIMS2 2016–2017, and LePHIA 2016–2017
Interview Duration Questionnaire Length
ZAMPHIA 2016 SHIMS2 2016–2017 LePHIA 2016–2017 ZAMPHIA 2016 SHIMS2 2016–2017 LePHIA 2016–2017
Interview duration
 Per 5 additional minutes 1.00 (0.98–1.03) 0.94 (0.89–0.99) 0.94 (0.91–0.97)
Questionnaire length
 Per 20 additional questions 1.10 (1.05–1.16) 1.01 (0.88–1.16) 1.07 (0.98–1.17)
Gender
 Male vs female 0.84 (0.75–0.94) 0.75 (0.62–0.90) 0.8 (0.70–0.91) 0.89 (0.79–0.99) 0.74 (0.61–0.89) 0.82 (0.72–0.94)
Age
 25–34 vs 15-24 0.78 (0.68–0.9) 0.64 (0.5–0.82) 0.71 (0.59–0.85) 0.75 (0.65–0.85) 0.62 (0.49–0.8) 0.68 (0.57–0.82)
 35–44 vs 15-24 0.73 (0.63–0.85) 0.70 (0.52–0.93) 0.69 (0.56–0.85) 0.71 (0.61–0.83) 0.67 (0.50–0.90) 0.67 (0.55–0.83)
 45–59 vs 15-24 0.88 (0.75–1.04) 0.74 (0.55–1.00) 0.90 (0.72–1.13) 0.9 (0.76–1.06) 0.74 (0.55–1.00) 0.92 (0.74–1.15)
Self-reported HIV status
 HIV-negative vs HIV-positive 0.21 (0.15–0.29) 0.30 (0.22–0.41) 0.30 (0.24–0.38) 0.24 (0.18–0.33) 0.32 (0.23–0.43) 0.35 (0.27–0.44)
 Unknown vs HIV-positive 0.26 (0.18–0.36) 0.20 (0.14–0.29) 0.21 (0.16–0.28) 0.31 (0.22–0.43) 0.21 (0.14–0.31) 0.25 (0.18–0.33)
Ever had sex
 Ever vs never 1.68 (1.42–1.99) 0.77 (0.57–1.04) 1.20 (0.96–1.49) 1.45 (1.21–1.74) 0.71 (0.51–1.00) 1.01 (0.80–1.28)
Household size
 3–5 vs 1-2 1.02 (0.85–1.23) 1.20 (0.96–1.50) 1.09 (0.93–1.28) 1.02 (0.85–1.21) 1.17 (0.94–1.46) 1.09 (0.93–1.28)
 6+ vs 1-2 1.31 (1.09–1.57) 1.35 (1.07–1.70) 1.28 (1.06–1.55) 1.28 (1.06–1.53) 1.32 (1.05–1.66) 1.28 (1.06–1.54)
Worked in the past 12 mo
 Worked vs did not 1.02 (0.92–1.14) 0.84 (0.7–1.01) 0.82 (0.71–0.94) 1.01 (0.91–1.13) 0.83 (0.69–0.99) 0.79 (0.68–0.9)
Reported on at least one child during the interview
 Reported on 1+ vs reported on 0 children 0.97 (0.85–1.09) 1.20 (0.94–1.52) 1.21 (1.03–1.42) 0.81 (0.70–0.94) 1.13 (0.86–1.51) 1.10 (0.91–1.32)
Wealth quintile
 Second vs lowest 1.13 (0.97–1.32) 0.70 (0.50–0.98) 0.97 (0.75–1.25) 1.12 (0.96–1.31) 0.70 (0.50–0.98) 0.99 (0.77–1.28)
 Middle vs lowest 1.36 (1.15–1.60) 0.65 (0.47–0.90) 0.80 (0.62–1.03) 1.37 (1.16–1.61) 0.64 (0.46–0.88) 0.81 (0.63–1.04)
 Fourth vs lowest 1.21 (1.01–1.45) 0.55 (0.39–0.76) 0.74 (0.57–0.95) 1.24 (1.03–1.48) 0.55 (0.4–0.77) 0.73 (0.57–0.93)
 Highest vs lowest 1.19 (0.97–1.46) 0.38 (0.28–0.53) 0.61 (0.47–0.79) 1.23 (1.00–1.50) 0.38 (0.28–0.53) 0.61 (0.47–0.79)
Education
 Primary vs no education 1.69 (1.4–2.05) 1.26 (0.66–2.44) 1.23 (0.88–1.71) 1.66 (1.37–2.01) 1.25 (0.65–2.40) 1.21 (0.87–1.67)
 Secondary vs no education 1.82 (1.48–2.24) 0.80 (0.42–1.51) 0.99 (0.70–1.39) 1.79 (1.46–2.20) 0.78 (0.41–1.49) 0.97 (0.70–1.36)
 More than secondary vs no education 1.20 (0.92–1.57) 0.47 (0.25–0.89) 0.60 (0.42–0.86) 1.19 (0.91–1.55) 0.46 (0.24–0.86) 0.59 (0.41–0.84)
Language
 Sesotho vs English 0.88 (0.62–1.26) 0.92 (0.65–1.30)
 Siswati vs English 1.78 (1.29–2.48) 1.80 (1.3–2.49)
 Bemba Vs English 1.03 (0.87–1.23) 1.05 (0.88–1.24)
 Nyanja vs English 1.16 (0.94–1.42) 1.17 (0.95–1.43)
 Lozi vs English 0.84 (0.55–1.27) 0.85 (0.56–1.29)
 Tonga Vs English 1.23 (0.90–1.69) 1.25 (0.91–1.71)
 Lunda vs English 1.56 (1.03–2.37) 1.64 (1.09–2.49)
 Luvale vs English 1.08 (0.71–1.62) 1.13 (0.75–1.71)
 Kaonde vs English 1.05 (0.70–1.57) 1.09 (0.73–1.63)
 Others vs English 0.64 (0.31–1.32) 0.57 (0.28–1.16)
Also adjusted for province (Zambia), region (Eswatini), and district (Lesotho), not shown.
Statistically significant effects are presented in bold text.

Odds ratios for interview duration, questionnaire length, and HIV testing characteristics from the multivariable logistic regression models stratified by self-reported HIV status are presented in Table 4. Owing to low variability, the models for self-reported HIV-positive individuals in all 3 countries could not support the inclusion of province, region, district, or sexual activity as covariates; the model for HIV-positive individuals in Zambia could not support the inclusion of wealth quintile or language. Effects for household food insecurity and marital status were not significant in any of the models stratified by self-reported HIV status.

TABLE 4. - Adjusted Odds Ratios (95% CI) of Blood Test Participation by Interview Length and HIV Testing Characteristics, Stratified by Self-Reported HIV Status and Controlling for Additional Demographic Characteristics, ZAMPHIA 2016, SHIMS2 2016–2017, and LePHIA 2016–2017
Interview Duration Questionnaire Length
ZAMPHIA 2016 SHIMS2 2016–2017 LePHIA 2016–2017 ZAMPHIA 2016 SHIMS2 2016–2017 LePHIA 2016–2017
Self-reported HIV-positive*
 Interview duration
  Per 5 additional minutes 1.05 (0.92–1.19) 0.98 (0.84–1.14) 0.93 (0.85–1.01)
 Questionnaire length
  Per 20 additional questions 1.14 (0.86–1.5) 1.05 (0.71–1.53) 1.12 (0.87–1.44)
 Recent HIV testing
  Tested more than 12 months ago vs tested in the past 12 months 1.01 (0.5–2.04) 2.08 (1.17–3.68) 1.95 (1.24–3.07) 0.97 (0.48–1.96) 2.16 (1.23–3.81) 2.03 (1.3–3.16)
  Unknown date of HIV test vs tested in the past 12 months 1.06 (0.28–4.03) 0.87 (0.20–3.88) 1.31 (0.39–4.40) 1.01 (0.26–3.88) 0.90 (0.20–4.00) 1.51 (0.45–5.08)
Self-reported HIV-negative
 Interview duration
  Per 5 additional minutes 0.99 (0.97–1.02) 0.95 (0.89–1.01) 0.96 (0.92–1.00)
 Questionnaire length
  Per 20 additional questions 1.06 (1.01–1.11) 1.06 (0.93–1.21) 1.1 (1.01–1.20)
 Recent HIV testing
  Tested more than 12 months ago vs tested in the past 12 months 1.5 (1.33–1.69) 0.82 (0.66–1.02) 0.94 (0.79–1.12) 1.52 (1.35–1.72) 0.83 (0.67–1.04) 0.96 (0.81–1.13)
  Unknown date of HIV test vs tested in the past 12 months 1.07 (0.82–1.4) 0.92 (0.55–1.53) 0.84 (0.57–1.23) 1.08 (0.83–1.4) 0.96 (0.58–1.6) 0.86 (0.59–1.26)
Unknown self-reported HIV status
 Interview duration
  Per 5 additional minutes 1.02 (0.96–1.07) 0.89 (0.77–1.02) 0.84 (0.75–0.93)
 Questionnaire length
  Per 20 additional questions 1.16 (1.02–1.31) 0.95 (0.64–1.41) 0.97 (0.75–1.24)
 HIV testing
  Ever tested for HIV vs never testing or missing 1.1 (0.72–1.69) 1.31 (0.54–3.21) 2.29 (1.05–4.99) 1.04 (0.68–1.58) 1.37 (0.56–3.35) 2.06 (0.95–4.46)
Statistically significant effects are presented in bold text.
*Model for self-reported positive individuals also adjusted for urban/rural residence, gender, age, number of children on which the individual reported, education, and language (Eswatini and Lesotho).
Model for self-reported negative individuals also adjusted for gender, age, whether the individual ever had sex, household size, employment status, wealth quintile, education, language, and province (Zambia), region (Eswatini), and district (Lesotho).
Model for individuals with unknown status also adjusted for gender, age, whether the individual ever had sex, number of children on which the individual reported, wealth quintile, education, language, and province (Zambia), region (Eswatini), and district (Lesotho).

After controlling for other demographics and whether individuals had tested recently, we observed no significant effect of interview duration on blood test participation in Lesotho, Eswatini, or Zambia for self-reported HIV-positive and HIV-negative individuals. Among those with unknown HIV status, blood test participation decreased with longer interview duration in Lesotho (OR 95% CI for each additional 5 minutes: 0.84, 0.75 to 0.93). Effects for interview duration were not significant for individuals with unknown HIV status in Zambia or Eswatini.

There was a small positive effect of questionnaire length on blood test participation for those with self-reported HIV-negative status in Zambia (OR 95% CI: 1.06, 1.01 to 1.11) and Lesotho (OR 95% CI: 1.10, 1.01 to 1.20) and in Zambia for individuals with unknown HIV status (OR 95% CI: 1.16, 1.02 to 1.31). There was no significant effect of questionnaire length on blood test participation for individuals with self-reported HIV-negative status in Eswatini and Lesotho or unknown HIV status in Eswatini. For self-reported HIV-positive individuals, the effect of questionnaire length on blood test consent was not significant in any of the 3 countries.

The effect of recent HIV testing varied by country and self-reported HIV status, with HIV-positive individuals in Lesotho and Eswatini who tested more than 12 months before the survey having higher odds of blood test participation compared with those who tested recently. Self-reported HIV-negative individuals in Zambia who tested more than 12 months before the survey had higher odds of blood test participation than those who tested more recently. Finally, among those in Lesotho with unknown status, the odds of blood test participation were higher among those who had ever tested, compared with those who had never tested or were missing a prior HIV testing status.

DISCUSSION

This analysis aimed to assess the effect of interview duration and questionnaire length on blood test participation in 3 population-based HIV surveys conducted in Zambia, Eswatini, and Lesotho. Blood test participation rates among those participating in the individual interview were generally over 90% in the PHIAs, similar to those observed in other biomarker surveys in these countries. For example, DHS has been conducting a population-based HIV testing since 2001, consisting of collecting blood spots on a filter paper from a finger prick, tested in a central laboratory. Refusal rates for HIV testing in the latest DHS were 2.0% for men and 1.5% for women in Lesotho, 6.7% for men and 5.4% for women in Zambia, and 16.6% for men and 9.5% for women in Eswatini.21–23

Our findings suggest that self-reported HIV status was an independent predictor of blood test participation. Although previous studies found that people with prior knowledge of their HIV-positive status were less likely to participate in HIV biomarker surveys,9,10 we found that self-reported HIV-positive persons were more likely to participate compared with self-reported HIV-negative persons and those with unknown HIV status. In PHIA, and in contrast to most prior HIV surveys, CD4 enumeration and HIV viral load test results were made available to participants who tested HIV-seropositive. These additional tests may have encouraged previously diagnosed people to participate. Although there may be some concern about the misreporting of HIV status, nearly all respondents reporting in the PHIAs as previously diagnosed tested as HIV-seropositive at 98.8%.

Contrary to our hypothesis, increases in the questionnaire length were positively associated with blood test participation in Zambia, a finding that persisted after adjusting for potential covariates. After stratifying by self-reported HIV status, this effect was found to be driven by those with self-reported HIV-negative and unknown HIV status in Zambia. After stratifying by self-reported HIV status, this result was replicated in HIV-negative individuals in Lesotho. For each additional 20 questions, the odds of participation increased by 1.06 (95% CI: 1.01 to 1.19) in Zambia and 1.10 (95% CI: 1.01 to 1.20) in Lesotho for HIV-negative individuals and 1.16 for those in Zambia with unknown status (95% CI: 1.02 to 1.31). These findings support a learning hypothesis that respondents with self-reported negative and unknown HIV status may gain renewed interest in learning about their HIV status as a result of taking the survey and answering questions about health and HIV testing.

Eswatini was included to serve as a test of the assumption of no unmeasured confounding. Because consent for blood testing was obtained before the interview in Eswatini, any observed associations could not be due to effects of the interview process on blood test participation. After stratifying by self-reported HIV status and including effects for whether individuals ever tested in the unknown status model and whether they tested recently in the HIV-positive and HIV-negative models, there were no significant effects of questionnaire length or interview duration on blood test participation in Eswatini.

There were increased odds of blood test participation associated with testing more than 12 months before the survey compared with testing more recently in individuals who self-reported HIV-positive in both Lesotho and Eswatini and in individuals who self-reported HIV-negative in Zambia. The differential impact of recent testing on blood test participation by self-reported status suggests that the country context influences the reasons for participation by different subgroups.

This analysis had several limitations. First, we conceptualized interview length to represent the degree of effort required to complete the survey, but the selected interview length variables are imperfect measures of the underlying construct. The effects of interview length on blood test participation differed by whether questionnaire length in the number of questions or interview duration in minutes was used, suggesting that each measure captures different aspects of interview participation. Questionnaire length partly reflects differential eligibility to receive additional follow-up questions (eg, participants who reported being previously tested for HIV were asked to recall the date of their last test). We controlled for sex, household size, and number of children to adjust for key eligibility criteria for certain modules, but there may be additional unaccounted sources of variability. Meanwhile, interview duration measures how long the interview actually took, and longer interview duration may reflect interruptions or time in which the interviewer was clarifying the questions for the respondent. We conducted sensitivity analyses with interviewer random effects to account for interviewer variability and found that results did not appreciably change (data not shown). Although this analysis used the same covariates in the models across countries, there was variation between countries in the bivariate associations between blood test participation and additional demographics (see Supplemental Table 2, Supplemental Digital Content, http://links.lww.com/QAI/B668).

The absence of significant results in Eswatini suggests that the models here sufficiently captured factors related to both interview duration and blood test participation. Finally, we were unable to assess the relationship between interview length and blood participation in the 10% of respondents in Lesotho who began the interview without reaching the tuberculosis module. Because these respondents ended the interview before the blood test consent questions, we are unable to remark on how they would have responded.

Most effect estimates were not statistically significant, which is partly because of limited statistical power as a result of high overall blood test participation rates. In fact, most nonparticipation was at the household level or individual level, and eligibility to be included in this sample was conditional on having completed the individual interview. These findings therefore represent a self-selected subsample of individuals who completed the interview in each country and are therefore not generalizable to the entire target population. Furthermore, this analysis is specific to Zambia, Lesotho, and Eswatin,i and one should not assume these findings hold in lower-prevalence settings or in other African countries. Finally, variability in findings across stratified analyses by known HIV status across countries also suggests that the motivations underlying blood test participation may be context-specific.

Our findings provide inconclusive evidence regarding the overall effect of interview length on blood test participation, and the overall magnitude of effects was small. The conceptualization of interview length in the number of questions versus interview duration is important in studies of interview length, as is the country context in studies of blood test participation. Further studies to explore factors contributing to nonparticipation and the potential cost-benefits of decreasing interview burden on participants are necessary.

REFERENCES

1. Brown K, Williams DB, Kinchen S, et al. Status of HIV epidemic control among adolescent girls and young women aged 15–24 years—seven African countries, 2015–2017. MMWR Morb Mortal Wkly Rep. 2018;67:29.
2. Saito S, Chung H, Mahy M, et al. Pediatric HIV treatment gaps in 7 East and Southern African Countries: examination of modeled, survey, and routine program data. J Acquir Immune Defic Syndr. 2018;78:S134–S141.
3. Justman J, Hoos D, Kalton G, et al. Real progress in the HIV epidemic: PHIA findings from Zimbabwe, Malawi and Zambia. Paper presented at: Conference on Retroviruses and Opportunistic Infections2017.
4. Mishra V, Hong R, Khan S, et al. Evaluating HIV Estimates from National Population-Based Surveys for Bias Resulting from Non-response. Calverton. Maryland: Macro International; 2008.
5. Dillman DA, Sinclair MD, Clark JR. Effects of questionnaire length, respondent-friendly design, and a difficult question on response rates for occupant-addressed census mail surveys. Public Opin Q. 1993;57:289–304.
6. Deutskens E, de Ruyter K, Wetzels M, et al. Response rate and response quality of internet-based surveys: an experimental study. Marketing Lett. 2004;15:21–36.
7. Perez-Lopez D, Walsh R. Differing Person Nonresponse Rate by Interview Length and Respondent Characteristics: Results from the 2011 SIPP-EHC Field Test. Washington, DC: U.S. Census Bureau; 2011.
8. Sharp LM, Frankel J. Respondent burden: a test of some common assumptions. Public Opin Q. 1983;47:36–53.
9. Reniers G, Eaton J. Refusal bias in HIV prevalence estimates from nationally representative seroprevalence surveys. AIDS. 2009;23:621–629.
10. Larmarange J, Mossong J, Bärnighausen T, et al. Participation dynamics in population-based longitudinal HIV surveillance in rural South Africa. PLoS One. 2015;10:e0123345.
11. Groves RM, Singer E, Corning A. Leverage-saliency theory of survey participation: description and an illustration. Public Opin Q. 2000;64:299–308.
12. Bogen K. The Effect of Questionnaire Length on Response Rates - A Review of the Literature. Washington, DC: U.S. Census Bureau; 1996.
13. Lynn P. Longer interviews may not affect subsequent survey participation propensity. Natl Health Stat Report. 2014;78:500–509.
14. Johnson K, Grant M, Khan S, et al. Fieldwork-Related Factors and Data Quality in the Demographic and Health Surveys Pro. Maryland: ICF Macro; 2009.
15. Ministry of Health [Zambia]. Centers for Disease control and prevention (CDC), ICAP at Columbia university. Zambia population-based HIV impact assessment (ZAMPHIA) 2016 [dataset]. ZAMPHIA2016adultind.sas7bdat. Ministry of Health [Lusaka, Zambia], CDC (Atlanta, GA), ICAP at Columbia University, New York, NY. ICAP at Columbia University; 2016.
16. Government of the Kingdom of Eswatini, Centers for Disease Control and Prevention (CDC). ICAP at Columbia University. Eswatini Population-Based HIV Impact Assessment (SHIMS 2) 2016-2017 Final Report. New York, NY: ICAP at Columbia University; 2016.
17. Ministry of health [Lesotho], Centers for Disease control and prevention (CDC), ICAP at Columbia university. Lesotho population-based HIV impact assessment (LePHIA) 2016-2017 [dataset]. LePHIA2016adultind.sas7bdat. In: Ministry of Health [Maseru, Lesotho], CDC (Atlanta, GA), ICAP at Columbia University (New York, NY), eds: ICAP at Columbia University; 2016.
18. Dixon JR. The international conference ON harmonization good clinical practice guideline. Qual Assur. 1999;6:65–74.
19. Ministry of health [Lesotho], Centers for Disease control and prevention (CDC), ICAP at Columbia university. Lesotho population-based HIV impact assessment (LePHIA) 2016-2017 final report. In: Ministry of Health [Maseru, Lesotho], CDC Atlanta, GA, ICAP at Columbia University (New York, NY), ICAP at Columbia University; 2016.
20. Ministry of Health [Zambia]. Centers for Disease control and prevention (CDC), ICAP at Columbia university. Zambia population-based HIV impact assessment (ZAMPHIA) 2016 final report. Ministry of Health [Zambia]Lusaka, Zambia; Atlanta, GA; New York, NY, CDC, ICAP at Columbia University. Lusaka, Zambia; Atlanta, GA; New York, NY: ICAP at Columbia University; 2016.
21. Central Statistical Office [Swaziland], Macro International. Swaziland Demographic and Health Survey 2006-07. Mbabane, Swaziland: Central Statistical Office and Macro International;2008.
22. Central statistical office [Zambia], Ministry of health (MOH) [Zambia], ICF International. Zambia Demographic and Health Survey 2013-14. Rockville, MD; Lusaka, Zambia: Central Statistical Office, Ministry of Health, and ICF International;2014.
23. Ministry of Health [Lesotho]. ICF International. Lesotho Demographic and Health Survey 2014. Maseru, Lesotho; Rockville, MD: Ministry of Health and ICF International; 2016.
Keywords:

HIV; Southern Africa; consent; interview length; HIV disclosure

Supplemental Digital Content

Copyright © 2021 Wolters Kluwer Health, Inc. All rights reserved.