Beyond Social Desirability Bias: Investigating Inconsistencies in Self-Reported HIV Testing and Treatment Behaviors Among HIV-Positive Adults in North West Province, South Africa

Abstract

This mixed-methods study used qualitative interviews to explore discrepancies between self-reported HIV care and treatment-related behaviors and the presence of antiretroviral medications (ARVs) in a population-based survey in South Africa. ARV analytes were identified among 18% of those reporting HIV-negative status and 18% of those reporting not being on ART. Among participants reporting diagnosis over a year prior, 19% reported multiple HIV tests in the past year. Qualitative results indicated that participant misunderstandings about their care and treatment played a substantial role in reporting inaccuracies. Participants conflated the term HIV test with CD4 and viral load testing, and confusion with terminology was compounded by recall difficulties. Data entry errors likely also played a role. Frequent discrepancies between biomarkers and self-reported data were more likely due to poor understanding of care and treatment and biomedical terminology than intentional misreporting. Results indicate a need for improving patient-provider communication, in addition to incorporating objective measures of treatment and care behaviors such as ARV analytes, to reduce inaccuracies.

Introduction

The effects of antiretroviral treatment (ART) in preventing HIV transmission have led to a call for universal access to treatment, and a heightened focus on meeting the UNAIDS targets of 90–90–90—or ensuring that 90% of people living with HIV know their status, 90% of people diagnosed with HIV receive sustained treatment, and 90% of people on treatment achieve viral suppression [1]. In many low-resource and high prevalence settings, including a number of African countries, a large part of national monitoring and evaluation of programs focused on reaching UNAIDS goals relies on simple and inexpensive self-reported measures of HIV testing, treatment, and adherence to identify gaps in care and measure progress towards national targets [2, 3]. However, studies have found poor agreement between medical record data and self-reported HIV treatment regimens, care visits, CD4 counts and viral loads [4–7], as well as poor correlation between self-reported adherence and viral suppression [8–11]. The extant research calls into question the utility of self-reported data as well as possible reasons for wide discrepancies.

Inaccuracies in self-reported data are often attributed to social desirability bias—over reporting socially desirable behaviors and underreporting socially undesirable ones—stemming from a need for social approval and desire to avoid embarrassment [12]. Several studies have supported the contribution of social desirability bias to inaccuracies in self-reported HIV risk and treatment behaviors [13–17]. Most commonly, efforts to mitigate bias in self-reported behaviors have focused on self-administered surveys to increase privacy, and have been found to enhance the validity of self-reported sexual behaviors [18]. However, the effects of self-interviewing modalities on improving self-reports of sensitive HIV-related behaviors varies substantially by outcome, question time frame, and region [19], suggesting additional factors are at play. In some cases, self-administered techniques show greater discrepancies than face-to-face interviews [20, 21]. This highlights the need to understand the range of factors contributing to inconsistencies, to in turn develop effective strategies to improve the accuracy of self-report.

Challenges with recall of adherence and healthcare utilization may also contribute to inaccuracies in self-reports. A review of studies comparing objectively measured ART adherence with self-report found poor correlation; respondents tended to use estimation rather than enumeration of doses taken, and responses may have reflected intention as opposed to action [22]. In a study of HIV care utilization in an Australian sample, there was substantial imprecision in self-reported dates of HIV diagnosis and date and results of most recent CD4 counts compared with medical records, but the direction of error was not systematic [23]. The broader literature on healthcare utilization suggests that accuracy tends to decline with longer recall periods, particularly for frequent behaviors, though longer periods are more accurate for infrequent, salient events [24].

Research using techniques such as cognitive interviewing or other forms of qualitative research suggests additional factors beyond social desirability and recall bias could also be contributing to inconsistencies in self-reported HIV treatment and service utilization, though few studies exist. Cognitive interviews on self-reported HIV care in the U.S. found that respondents reported near perfect ART adherence on self-administered surveys, but their doctors differed in the extent to which they had described how medications should be taken, and respondents had difficulties recalling those instructions and their list of medications [25]. A study of self-reported male circumcision in Zambia and Swaziland found inconsistencies in reported circumcision status were due more to misunderstandings in how male circumcision was defined by participants than to intentional misreporting [21].

Few studies have investigated what causes misreporting of HIV service utilization and adherence behaviors in sub-Saharan Africa. One exception has been innovative work by HIV pre-exposure prophylaxis (PrEP) trial investigators, who conducted qualitative research demonstrating that when participants were presented with evidence of non-adherence after the trials, most explained they had over-reported use of the product out of fear their participation would be terminated [26–29].

While evidence from sub-Saharan Africa suggests 90% of individuals retained in care achieve virological suppression [30], in a 2014 population-based survey in South Africa we documented that only 52% of respondents reporting near perfect ART adherence had attained viral suppression [8], similar to the large discrepancies observed in the PrEP trials. The large gap between reported ART initiation and adherence and viral suppression in our study could imply misreporting due to social desirability bias, recall bias, and/or poor understanding of the survey questions. Low viral load despite reported adherence could also be due to medication resistance, however, discrepancies between objective measures of ART adherence and self-reports are common [31], and are unlikely fully explained by drug resistance.

In order to understand the sources of discrepancies between reported behaviors and viral load, we conducted additional laboratory testing, specifically ARV drug exposure analysis, on stored samples to assess agreement between reported adherence and presence of ARV. We also instituted a qualitative study during a second population-based survey (second round of data collection) in 2016 to explore the chief causes of discrepancies and identify strategies to improve reporting accuracy.

Methods

Study Setting

Quantitative data were collected through two independent surveys in 2014 and 2016 in Lekwa-Teemane and Greater Taung sub-districts, within Dr. Ruth Segomotsi Mompati (RSM) District of North West Province, Republic of South Africa. Qualitative data were collected in 2017 based on findings in the 2016 sample only. RSM is comprised of both rural and peri-urban areas. The study area includes approximately 230,000 people, the majority of whom speak Setswana. North West Province has the fourth highest HIV prevalence in South Africa, with an estimated prevalence of 20.3% in the adult population 15–49 years [32]. Lekwa Teemane was the site of a United States Centers for Disease Control and Prevention comprehensive HIV prevention initiative, Re Mogo Pholong (RMP), the evaluation of which gave rise to the two surveys [8].

Data Collection

We employed two serial cross-sectional surveys using multi-stage cluster sampling, with 23 enumeration areas selected proportionate to size in each sub-district based on 2011 census data (the most recent national census). The same process was applied to both waves. The first survey was conducted between January and April 2014; the follow-up was conducted between August and November 2016. In each enumeration area up to 46 inhabited dwelling units were randomly selected for inclusion in the sample and one adult (18–49 years) was randomly selected per dwelling unit for participation in the bio-behavioral survey. The sample size was estimated based on the number necessary for the program evaluation to have 80% power to detect an increase in annual HIV testing from 10 to 16%, we estimated a design effect of 1.20 associated with the correlation of observations within clusters, assuming an intra-class correlation of 0.014.

An intensive training course prepared fieldworkers, community health workers and supervisors prior to 2014 and 2016 data collection. The course entailed seven training days on topics including HIV, STI, and TB-related health information; research ethics and informed consent; discussing sensitive topics; study protocols for data collection; dried blood spot collection and rapid HIV antibody testing. Following in-classroom training, field staff role-played procedures prior to beginning data collection.

When conducting the survey, trained fieldworkers assessed eligibility criteria, which included being 18–49 years, residing in the sampled home, and able to provide consent (i.e. not apparently intoxicated or cognitively impaired). After obtaining consent, fieldworkers administered a survey by computer-assisted personal interviewing in a private location at the participant’s home in the participant’s language of choice (English, Xhosa, Afrikaans, or Setswana). Survey questions included demographics, health care utilization, HIV testing and care history, ART initiation, and medication adherence. The fieldworkers then referred consenting participants for rapid HIV antibody testing and dried blood spot (DBS) sample collection, which was performed by trained community health workers [8]. Participants testing HIV-positive, indeterminate, or who preferred not to undergo an HIV rapid test but consented to DBS provision for laboratory HIV diagnosis were asked to provide finger-prick blood for DBS. Full details of sampling and survey and biological data collection have been described elsewhere [8, 33]. A total of 91.9% (n = 2028) of eligible participants consented to the survey, and 77.1% of the analytic sample consented to HIV testing through rapid test or DBS (n = 1555; n = 5 HIV test results were inconclusive).

Due to a large number of inconsistencies in reported ART adherence and viral load findings in 2014 [8], a consent for permission to contact participants for future interviews was included during the survey in 2016 for those reporting being currently on ART and who also consented to DBS collection. All fifteen consenting HIV-positive participants with discrepancies between ARV analytes and self-reported ART uptake were then contacted in April 2017. Of these, 10 were located and consented to participate in qualitative interviews in the same month. The interviews consisted of two sections—cognitive and in-depth interviewing. The cognitive interview section focused on assessing participant understanding of the survey questions, relevant to their particular discrepancies. Interviewers read each survey question to participants, and following their response, asked them to explain what the question was asking in their own words, and how they arrived at their answer. Participants were probed to expand upon or clarify their explanations. The in-depth interview section focused on exploring the discrepancy seen in the participant’s survey results, experiences participating in research and engaging in care, community norms around testing and treatment, and community stigma. Using standardized interview guides, interviews were conducted in English or Setswana, depending on participant preference, by two trained, local field staff, one male and one female. Interviews were audio-recorded, transcribed verbatim, and translated into English for coding and analysis. A bilingual field staff member read a sample of translated transcripts which were analyzed by English-speaking investigators, to ensure the quality of translation. When any aspect of the translated transcript was unclear to the English-speaking investigators, the bilingual field staff members reviewed the original transcript and clarified the translation to the English-speaking investigators.

The protocol was approved by the Committee for Human Research at the University of California, San Francisco; the Human Subjects Division at University of Washington; the Human Sciences Research Council Research Ethics Committee in South Africa; the Policy, Planning, Research, Monitoring and Evaluation Committee for the North West Provincial Department of Health; and the CDC’s Center for Global Health, Human Research Protection Office.

Measures

Biomarkers

ARV Drug Exposure

Participants who tested positive for HIV by rapid antibody testing in the field or laboratory and consented to provide DBS were tested for ARV drug exposure. Exposure was determined by a validated qualitative LC MS/MS method for the determination of the presence or absence of various antiretroviral analytes against cutoff samples analyzed at a known concentration (0.02 μg/ml). Analytes were assessed for Efavirenz, Lopinavir, and Nevirapine in order to include both first and second line regimens. Because Tenofovir and Emtricitabine are only given in combination with Efavirenz, these were not included [34].

Survey Items

All participants were asked about HIV testing in the last year, prior to questions regarding HIV status. Participants were grouped by the number of reported HIV tests in the past year using three categories: zero, one, and two or more. All participants reporting knowledge of positive status were asked about HIV care frequency in the following intervals: once every 3 months or more often, once every 4–6 months, once every 7–12 months, or less than once a year. They were asked how often they receive CD4 testing: 6 months or more frequently, every seven to 12 months, or less than once a year. Participants reporting ART initiation were asked how many days they did not take their ART in the past seven and past 30 days. We grouped seven-day adherence responses into zero days missed, 1–6 days missed, and seven (all) days missed. For 30-day adherence, responses were grouped into 0 day missed, 1–29 days missed, and 30 days missed.

Analysis

In the results that follow, we report on inconsistencies in two domains. First, we discuss cases where participants whose self-reported HIV status, treatment initiation, or adherence were discrepant with analyte results. Second, we explore reported recent HIV testing among those who had also reported being diagnosed with HIV more than a year ago. Using qualitative data, we examine potential causes of these discrepancies. The quantitative data presented is from the surveys conducted in 2014 and 2016, and the analyte testing using stored DBS. The quotes are from the qualitative interviews conducted in 2017, approximately 6 months after the second survey round.

Of the 2015 total participants surveyed, this analysis was restricted to the 385 (19%) participants who tested HIV-positive and/or reported a positive HIV status; this is an unweighted proportion so does not represent an estimate of HIV prevalence [8]. Only those who provided DBS (n = 317) were included in the analysis of discrepancies between self-report measures and ART exposure. We computed descriptive statistics on reported HIV status and testing and treatment behaviors and compared reported behaviors to the presence of ARVs in DBS to identify discrepancies. We also explored whether demographic variables were associated with discrepancies using logistic regression of discrepant reports (self-reported HIV status vs. ARV analyte results, self-reported treatment adherence vs. ARV analyte results, and self-reported time since HIV diagnosis vs. self-reported number of HIV tests in the past year) on gender, age, and education. Given the significant association between gender and nearly every measure of discrepancy, results are stratified by gender.

Qualitative analysis was done using targeted coding based on a priori categories determined by the primary focus on the interviews. These included: (1) participant perceptions of discrepancies; (2) participant understanding of questions; and (3) explanations of discrepancies. Axial coding was then used to further synthesize the larger categories [35]. The first and second author conducted initial coding and interpretation of the coded transcripts, which was then discussed and verified with the two South African, qualitative interviewers. Qualitative analysis was conducted using Dedoose [36].

Following qualitative data analysis, quantitative data were revisited to explore themes that had emerged during interviews. In particular, based on qualitative evidence of misunderstandings of “HIV test,” we computed additional descriptive statistics to assess the frequency of reports of HIV testing occurring post-HIV diagnosis. Quantitative analyses were performed using Stata version 14 (StataCorp LP, College Station, TX).

Results

Survey participants included in the quantitative analyses (n = 385; Table 1) had a mean age of 35 years (IQR: 29–41), and the majority were female (68%, vs. 32% male). Most respondents reported education levels of ‘primary or less’ (31%) or ‘some secondary’ (42%), most were unemployed (69%), and 31% were married. Seventy-one percent reported that they were HIV-positive (reported knowing their positive status). Twelve percent of those who were positive did not consent to providing DBS and could not be tested for ARV exposure; 52% had positive ARV analyte results, and 36% had no trace of ARV analytes. The 10 qualitative interview participants (Table 1) were similar in age to the overall population of HIV-positive residents in the survey (mean: 34.5, IQR: 29–40), had a more even gender split (50% male and 50% female), and a larger proportion were married (50%) and had a primary education or less (50%). Sixty percent reported in the survey that they were HIV positive. All had consented to providing DBS; half were positive and half negative for ARV analytes.

Table 1.

Demographic and laboratory indicators among HIV-positive survey participants from a population-based survey in North West Province, South Africa

Measure	N (%) or mean (IQR)
	Quantitative sample (n = 385)	Qualitative sample (n = 10)
Sex
Female	262 (68)	5 (50)
Male	123 (32)	5 (50)
Age	35 (29–41)	35 (29–40)
Education
Primary or less	120 (31)	5 (50)
Some secondary	160 (42)	3 (30)
Completed secondary	95 (25)	2 (20)
College	10 (3)	0
Employment
Full time	45 (12)	3 (30)
Part time	33 (9)	1 (10)
Unemployed	265 (69)	6 (60)
Other (e.g. self-employed, student)	42 (11)	0
Marital status
Married/living with partner	121 (31)	5 (50)
In a relationship/single	235 (61)	4 (40)
Separated/divorced	15 (4)	1 (10)
Widowed	14 (4)	0
Self-reported HIV-positive
Yes	274 (71)	6 (60)
No	111 (29)	4 (40)
Tested HIV-positive via rapid test or PCR*
Yes	340 (98)	10 (100)
No	6 (2)	0 (0)
Indeterminate	1 (< 1)	0 (0)
ARV analytes**
Positive	199 (52)	5 (50)
Negative	137 (36)	5 (50)
No DBS consent	47 (12)	0

^*Among those consenting to rapid test or DBS

^**Two participants provided DBS but had no analytes result

Logistic regression of discrepancies on gender, age, and education (completed secondary school vs. less than secondary) indicated that gender predicted discrepancy across nearly every measure, while age and education were non-significant. Among those reporting ART days missed of the past seven, the odds of discrepancies with ARV analytes (i.e., reported missing 0 day and testing negative for ARV analytes, or reporting missing all days and testing positive for ARV analytes) were higher among men (OR 4.94, 95% CI 1.63, 14.96, p = 0.005). Men had 1.92 times higher odds of reporting they were diagnosed with HIV more than 1 year ago, yet had two or more HIV tests in the past year (95% CI 0.95, 3.88, p = 0.068), versus reporting fewer than two HIV tests in the past year or an HIV diagnosis within the past year. Men also had higher odds of reporting that they were not on ART while testing positive for ARV analytes, vs. self-report ART status that coincided with ARV analyte results (OR 6.38, 95% CI 1.12, 36.20, p = 0.036). Though male gender was most often associated with discrepancy, among those reporting an HIV negative status, men had a 0.34 times lower odds of testing positive for analytes (95% CI 0.12, 1.13, p = 0.081).

Discrepancies Between Reported HIV Status, ARV Use and Analyte Testing

Comparisons of ARV analyte results and reported HIV status and ART intake revealed several discrepancies (Table 2; grey shading used to highlight discrepancies). ARV analytes were identified in 18% (26% of women and 11% of men) reporting an HIV-negative or unknown status. ARV analytes were also found in 18% of people (10% of women and 27% of men) who reported that they were HIV-positive but that they had never initiated ART. Overall, 7% of those reporting perfect adherence in the past 7 days, and 8% of those reporting perfect adherence in the past 30 days had negative analyte results. However, a substantial proportion of the few reporting they missed all doses in question also had positive analyte results: 45% for the 7-day measure and 67% for the 30-day measure. Qualitative findings described below revealed several potential explanations for these inconsistent findings, chiefly poor understanding of care and medications, and data entry errors.

Table 2.

ARV analytes versus self-reported HIV status and treatment among HIV-positive survey participants from a population-based survey in North West Province, South Africa

Measure	ARV analyte exposure
	Negative N (%)			Positive N (%)
	Total (n = 122)	Female (n = 62)	Male (n = 60)	Total (n = 195)	Female (n = 146)	Male (n = 49)
HIV status
Reported negative or never tested	75 (82)	35 (74)	40 (89)	17 (18)	12 (26)	5 (11)
Reported positive	47 (21)	27 (17)	20 (31)	178 (79)	134 (83)	44 (69)
Are you on antiretroviral treatment or ART now, or were you ever on ART?*
No	28 (82)	17 (90)	11 (73)	6 (18)	2 (10)	4 (27)
Yes	19 (10)	10 (7)	9 (18)	172 (90)	132 (93)	40 (82)
In the last week, that is the last 7 days, how many days did you NOT take your ART?**
0	12 (7)	5 (4)	7 (16)	157 (93)	121 (96)	36 (84)
1–6	1 (10)	1 (11)	0 (0)	9 (90)	8 (89)	1 (100)
7	6 (55)	4 (57)	2 (50)	5 (45)	3 (43)	2 (50)
In the past 30 days, how many days did you NOT take your ART?**
0	12 (8)	4 (3)	8 (23)	140 (92)	113 (97)	27 (77)
1–29	1 (5)	1 (8)	0 (0)	19 (95)	12 (92)	7 (100)
30	6 (33)	5 (42)	1 (17)	12 (67)	7 (58)	5 (83)

Shading highlights discrepant results. Percentages total to 100 for each row across “total” columns, “male” columns, and “female” columns

^*Includes those reporting HIV-positive status

^**Includes those reporting HIV-status and currently on ART

Poor Understanding of Care and Medications

Narratives in qualitative data suggested that some participants’ responses may be shaped by an incomplete understanding of their diagnoses and the medications they were prescribed. Half (n = 5) of our participants described some level of inadequate patient-provider communication or comprehension of information provided by clinicians or test counselors. For example, being tested for HIV by mobile services and then sent to a different facility to receive test results or treatment, created delays or misunderstandings about diagnoses for some participants. One man who had reported that he was HIV-negative but tested positive for analytes explained that he had been tested by mobile testing staff, but never received the results. He later received treatment from a clinic provider who simply told him that the medications “will help my immune system to pick up.” For this participant, confirmation that he was HIV positive came after he found out that his brother, who is also HIV-positive, had been prescribed the same pills. He told the field staff from our study that he was HIV negative because he wanted to be retested by an independent party: “I wanted to find out if these people from the mobile are telling me the truth, if what you tell me will be the same or not.” (Participant #1: Male, age 31)

Another participant was unsure about what medications she was taking, and therefore reported that she was not on ART, despite her positive analyte results. She explained that the clinic prescribed her “white pills,” and maintained that she was not on ART, though she was not certain of what she was taking:

[Clinics] give you ARVs when they see you are weak. If they see you are strong, they give you the white pills and they tell you to take two. When you are weak they give you the ARVs to give you strength… They gave the whole packet so you take two once a day… They told me it is for when you are not that weak. (Participant #2: Female, age 27)

This participant went on to report that she was eager to initiate ART, but the poor communication and interactions she experienced at public clinics led her to consider private hospitals where she anticipated better explanations (“…they take your blood test and explain everything to you”) and more responsive care.

Three participants shared experiences with receiving HIV care medications from the clinic without adequate explanation about the content and purpose of the prescription, or how the medication works. Participants believed it was common for people on ART to have a rudimentary understanding of their treatment regimen:

Interviewer: Do you think people who take the HIV treatment do know about the treatment they are taking?

Participant: Those who take it? I think they just take it because they got that they are infected. Even myself I know that it is the HIV treatment but I do not know how, I just heard that I am infected, then I started taking it but not really understanding how, you see? And sometimes, I do not know if it is because of the period [time], but when I take these pills I look at their numbers and realize that at this period [time] I found these certain pills and if it is another period [time], maybe I find different ones, but they tell me that it is one and the same thing. So I just accepted, as long as I am able to take my treatment and see myself survive, I do not have a problem.

Interviewer: Have they not sat with you at the clinic and told you that the pill you are taking is for what or rather how it works on your body?

Participant: No, they never explained that to me. (Participant #3: Female, age 35)

Of the two participants we tracked who reported perfect ART adherence and had negative analyte results, both maintained that they took their treatment daily (Participant #4: male, age 40; Participant #5: male, age 36). However, one also shared that he takes herbal supplements he purchases from the pharmacy to “clean the blood” and “clean my stomach” each morning, and had not told his HIV care provider. The effects of these supplements may explain his negative analyte results, or the participant may in fact not be taking ARVs and have reported perfect adherence for a different medication from the pharmacy he mistook to be ARVs.

We also found evidence that discrepancies can be entirely unintentional, and in some cases caused by scenarios unlikely to be anticipated by researchers. In one example, a woman reported that she was HIV negative, but her analyte results were positive. She explained that she was unaware of her HIV status until the antibody testing during fieldwork; she had been taking her sister’s ART in solidarity, in hopes of encouraging her sister’s adherence.

Interviewer: How did you come about taking her [participant’s sister] pills for the HIV virus?

Participant: She refused to take them, so I wanted to show her that they are just pills, and she is going to get healed. (Participant #6: Female, age 40)

Data Entry Errors

During qualitative interviews, a few (n = 4) participants stated explicitly that they remembered participating in the survey and that they had not answered the questions as documented. For example, when asked about the discrepancy between her reporting an HIV-negative status and her positive analyte test result, one participant responded:

It means [the fieldworker] is the one who is at fault, who made a mistake, because I told her that I am positive because I take my treatment. I take the HIV treatment, and she said to me, “Then if you are taking the treatment we will bring machines so we take your CD4 count.” So that means it is her who did not write it down well, she made a mistake, but I could also see that this person is in [a hurry]. (Participant #3: Female, age 35)

Responses like these indicate that that some proportion of cases with apparent misreporting of HIV status or treatment behaviors may have been data entry errors by fieldworkers at the time of the survey.

Reported HIV Testing Among Those Diagnosed More than a Year Ago

The findings of the quantitative survey revealed that a high proportion of participants (19% total; 16% of females and 29% of males) who reported that they were diagnosed with HIV more than a year ago also reported being tested for HIV two or more times within the last year (Table 3). Further, more frequent self-reported HIV care corresponded with higher reported frequencies of recent HIV tests, pointing to potential conflation of HIV antibody testing and care-related testing (e.g. CD4 and viral load). For example, 25% of those with a CD4 test within the past year (22% of females and 36% of males) also reported testing for HIV two or more times in the same period. Twenty percent of those reporting HIV care visits every 3 months or more, and 29% of those reporting HIV care every 4–6 months, reported ‘being tested for HIV’ two or more times in the past year. These proportions were particularly high among men: 30 and 40% for the same measures, respectively. The same pattern was found for measures of CD4 testing. Among those reporting that they receive a CD4 test every 0–6 months also, 28% overall reported two or more HIV antibody tests in the past year, with the proportion, among men, a striking 46%. In comparison, just 7% of those reporting CD4 test frequencies of less than once per year reported HIV testing two or more times in the past 12 months. Qualitative findings offer some possible explanations for these findings.

Table 3.

Self-reported HIV testing frequency versus self-reported HIV status and care among HIV-positive survey participants from a population-based survey in North West Province, South Africa

Measure	Times tested for HIV in past 12 months
	0 (n = 151)			1 (n = 57)			2 + (n = 66)
	Total N (%)	Female N (%)	Male N (%)	Total N (%)	Female N (%)	Male N (%)	Total N (%)	Female N (%)	Male N (%)
Time since diagnosis*
12 months or less	7 (12)	7 (16)	0 (0)	27 (47)	16 (37)	11 (79)	24 (41)	21 (48)	3 (21)
More than 12 months	144 (67)	111 (69)	33 (59)	30 (14)	23 (14)	7 (13)	42 (19)	26 (16)	16 (29)
Among those diagnosed > 1 year ago
Last post-diagnosis CD4 test*
12 months or less	62 (60)	50 (61)	12 (55)	16 (15)	14 (17)	2 (9)	26 (25)	18 (22)	8 (36)
More than 12 months	37 (80)	32 (89)	5 (50)	5 (11)	2 (6)	3 (30)	4 (9)	2 (6)	2 (20)
Frequency of HIV care*
> Every 3 months	117 (68)	91 (73)	26 (57)	20 (12)	14 (11)	6 (13)	34 (20)	20 (16)	14 (30)
Every 4–6 months	9 (53)	6 (50)	3 (60)	3 (18)	3 (25)	0 (0)	5 (29)	3 (25)	2 (40)
Every 7–12 months	6 (67)	6 (67)	0 (0)	2 (22)	2 (22)	0 (0)	1 (11)	1 (11)	0 (0)
< Once a year	5 (56)	5 (63)	0 (0)	2 (22)	1 (13)	1 (100)	2 (22)	2 (25)	0 (0)
CD4 test frequency*
0–6 months	65 (60)	53 (65)	12 (46)	13 (12)	11 (13)	2 (8)	30 (28)	18 (22)	12 (46)
7–12 months	32 (82)	27 (82)	5 (83)	2 (5)	2 (6)	0 (0)	5 (13)	4 (12)	1 (17)
< Once a year	20 (69)	17 (81)	3 (38)	7 (24)	3 (14)	4 (50)	2 (7)	1 (5)	1 (13)

Shading highlights discrepant results. Percentages total to 100 for each row across “total” columns, “male” columns, and “female” columns

^*Among those reporting HIV-positive status.

Confusion with Terminology

The narratives provided by a few participants (n = 3) suggest that confusion with terminology may be one explanation for these findings. For example, participants appeared to interpret being tested for HIV, or antibody testing, as synonymous with regular CD4 and viral load testing to monitor treatment.

Interviewer: So for the past 12 months how many times have you tested for HIV?

Participant: Some time during next week I must go do the blood test. And I do it again every four months.

Interviewer: Aren’t you talking about a test which they do to check if you are taking the treatment correctly? I am talking about a test they do to see if you have HIV.

Participant: The last time was when they came to the farm.

Interviewer: Since you did it last year September?

Participant: Yes.

Interviewer: Meaning in the past 12 months you only tested for HIV once?

Participant: Yes. (Participant #7: Male, age 42)

For another participant, a similar exchange occurred:

Interviewer: From the past 12 months, now we are talking about months, how many times have you tested for the HIV Virus?

Participant: How many times? No! I was just taking the treatment only; I am going to be tested in August.

Interviewer: What are you going to be tested for in August?

Participant: They are going to take my blood, so they will take that blood to the laboratory and check CD4 count, how my antibodies are coming about.

Interviewer: Okay, so my question is that, I am talking about the HIV test, you are already on the treatment, right? You know your status? I want the number of times that you did HIV tests, how many times is it?

Participant: I do not remember, it has been long, it is old stuff. (Participant #3: Female, age 35)

Both participants had reported testing for HIV within the past 12 months during the household survey. The responses above indicated that the term “HIV test” or simply “test” was used for both HIV antibody testing and CD4 counts, and the ascribed meaning depended upon whether the participant reported an HIV positive status. In one example, after a participant had reported testing positive for HIV infection, the interviewer asked how many times she was tested in the past year. She responded, “I have never been tested,” suggesting that the meaning of the word may have shifted, for her, to CD4 or viral load testing once she was diagnosed.

Though testing terminology appeared to be a primary source of misreporting, disparate understandings of terms for HIV treatment were a source of confusion for participants as well. For example, one participant’s response pointed to a perceived difference between “ARVs” and “ART,” and confusion with evolving terms used in clinical care:

Interviewer: How would it have made you feel had you told them you are on ART?

Participant: I still don’t understand the term ART.

Interviewer: ART refers to HIV treatment.

Participant: So it’s no longer ARV, it’s ART? (Participant #8: Female, age 23)

These participant responses suggested that in some cases, the terminology that researchers and clinicians use may be unfamiliar, unclear, or have different meanings for different participants.

Trouble with Temporal Questions

During cognitive interviews, half of participants (n = 6) expressed difficulty with questions that referenced specific time frames. As one woman explained:

Honestly all these questions you asked me about three days, two months, seven months, it is a bit hard you see…I do not quite remember them! (Participant #3: Female, age 35)

Another participant, when asked when he was diagnosed with HIV, responded:

My file is still at the clinic and my appointments are at home…it is all written there… I think it could be 5 years or 3 years back. (Participant #7: Male, age 42)

Without his paperwork from the clinic, it was difficult for him to recall when his appointments were or when he was last tested. However, for some participants, referencing a significant event in their lives assisted them in remembering time frames. For one woman, remembering when she tested positive was aided by recalling her pregnancy:

Participant: …I don’t remember the date but I think I was already 3 months pregnant by then… The baby is 9 months old now.

Interviewer: So when we add 9 months and the remaining 6 months of pregnancy…so it is between 1-2 years ago?

Participant: Yes! (Participant #3: Female, age 35)

These responses indicated that some proportion of the inconsistent responses around timing of testing and diagnosis may have been due to the difficulty of remembering question time frames accurately, but for some participants, questions regarding specific time frames could be reframed to reference life events and circumstances.

Discussion

We set out to understand the large discrepancy between reported consistent adherence to ART and extremely low levels of viral suppression in a population-based survey in rural South Africa by comparing ARV analyte testing to reported HIV status and behaviors among HIV-positive respondents, and then exploring discrepant responses in a qualitative sub-sample. This mixed-methods study points to several key findings about both the quantity and causes of discrepancies in self-reported HIV-status, testing, and treatment behaviors. While discrepancies typically have been ascribed to social desirability bias, we did not find evidence of purposeful misreporting, but instead found that most discrepancies were explained by inadequate understanding of HIV antibody test results and treatments prescribed, confusion about terminology like “HIV test,” difficulties recalling referenced time frames, and data entry error. Given the focus on the UNAIDS 90–90–90 goals [1] for increasing HIV status awareness, treatment, and viral suppression, these findings have critical implications for both improving status and treatment awareness and reducing misclassification in studies seeking to characterize progress and evaluate HIV programs.

In clinical settings, there is a clear need for improving patient-provider communication. Our findings indicated that a number of participants lacked a clear understanding of their diagnoses, prescribed medications, and potential interactions. This problem may be exacerbated by a lack of connection and coordination between service providers, and by patients’ hesitancy to ask more about their treatment and/or willingness to take medications without question, as long as they are surviving. The participant who reported not being on ART, but referred to taking two white pills, for example, could be referring to Bactrim prophylaxis for pneumocystis, which is prescribed for those with CD4 < 200 and often is a set of two large white pills. This kind of misinformation and miscommunication about prognosis and treatment is dangerous, as is poor specification of HIV positivity when one is referred to other services. These findings are consistent with those from cognitive interviews on self-reported HIV care in the U.S. In the U.S.-based study, respondents reported near perfect ART adherence on self-administered surveys, but their doctors differed in the extent to which they had described how medications should be taken, and respondents had difficulties recalling those instructions and their list of medications [25].

Of those reporting that they were HIV-negative, nearly 20% tested positive for ARV analytes. In another study based in South Africa, 41% of participants reporting an HIV-negative status and testing positive for HIV antibodies, also had analyte positive results [37]. The authors suggested that some intentional misreporting of HIV status could not necessarily be separated from issues with patient-provider communication or patient understanding. As exemplified in our findings, participants may feel hesitant to report status if they are uncertain of their diagnosis and instead opt to confirm test results and resolve confusion from interactions with healthcare providers. Others may unintentionally misreport for reasons that the researcher cannot predict, as in our case of a participant who took ARVs when she was HIV negative as a form of social support when her sister was diagnosed with HIV. While the absence of PrEP in the study region at the time of data collection rules out PrEP as an explanation for HIV-negative participants testing positive for ARV analytes, future studies or those in regions where PrEP is available may find increasing discrepancies of this nature.

Findings also pointed to fundamental misunderstandings of commonly used HIV testing terminology. The term “HIV test” has been used for decades, and researchers and clinicians may assume its meaning is widely understood as testing to determine HIV status. Our findings indicated this assumption must be questioned, as there may be a perception that all tests are broadly looking for the presence of HIV or immune response, whether for the purposes of diagnosis or to evaluate treatment effectiveness. Increases in access to CD4 and viral load testing may have generated confusion with how these tests operate distinctly from HIV antibody tests, and evolving understandings of the ‘testing’ term could be producing inaccurate estimates of testing patterns in population research. Our participants’ narratives indicated that the meaning of testing might be differentiated by where it was located on a person’s trajectory of diagnosis and care, with testing pre- versus post diagnosis carrying different meanings. To avoid misunderstanding and misclassification in research, programmatic, and care settings, testing terms could be accompanied with explanations regarding whether the test in question is for HIV diagnosis on someone whose status is unknown, or CD4 or viral load testing for someone who knows they are HIV positive and are being monitored in care.

Discrepancies were more frequent among men on most measures. Studies of discrepancies between self-reported adherence and indirect measures or biomarkers have not typically examined associations with socio-demographic factors [38], while studies of self-reported health utilization have found either no associations with inaccuracies, or older age as a predictor of underreporting [24]. Our findings suggest a particular need for care and treatment education among men to reduce discrepant reporting, perhaps related to their lower engagement in healthcare services or discomfort discussing diagnoses with the largely female provider base.

Though cognitive interviewing revealed that participants were able to recall past-week ART adherence, they struggled with questions about diagnoses, testing, and care visits that occurred within longer time frames. Recall challenges are corroborated by previous studies of self-reported HIV diagnosis, care utilization, and treatment adherence [22, 23]. Studies of other self-reported healthcare utilization measures have similarly found poor adherence to question time frames, and increasing difficulty with recall as the frequency of visits increases [24, 39]. We found that jogging participants’ memories using life events such as pregnancies, which may correspond with HIV service utilization, aided recall. Using landmark events such as birthdays, or public holidays to mark points during the recall period has been found to improve accuracy in other studies of self-reported healthcare utilization [22, 24].

Data entry error appeared to be a factor in discrepancies as well, with several participants recorded as HIV-negative in survey data insisting this was captured inaccurately. Given automated skip patterns for those recorded as negative status, no data were collected on these participants’ care and treatment behaviors during the survey. Including second confirmation questions to read to participants prior to skips may reduce these forms of unintentional error (e.g. “You reported the results of your last HIV test were negative. Is that correct?”). Similarly, in response to the question, “In the last seven days (also asked for 30 days), how many days did you NOT take your ART?”, a large proportion of those reporting “seven days” ultimately tested positive for ARV analytes. These participants may have misheard the “NOT” portion of the question, and intended to report they took ART all 7 days. Confirmation questions may help to catch these forms of errors as well, and reduce discrepancies. We also advocate for frequent, unscheduled field observations by study supervisors during data collection, and comparing metrics across field workers such as survey durations, frequency of lab data collected, and frequency of responses that lead to sections skipped.

Our findings represent a departure from other adherence studies that have identified social desirability bias as a significant source of misclassification [27–29], where social desirability may have a greater impact in reporting for clinical trial participants concerned about termination or who have ongoing study visits. For example, PrEP trial investigators demonstrated that when participants were presented with evidence of non-adherence after the trials, most explained they had over-reported out of fear their participation would be terminated [26–29]. Though we cannot rule out social desirability bias given our small qualitative sample size, our findings more closely reflect studies such as Hewett et al. [21], who found that misclassification of male circumcision status was primarily a result of differing understandings of the term, and Wilson et al. [25], who identified challenges among participants on ART with recalling time periods, doctors’ instructions, and medications. Other researchers have also hypothesized misunderstandings in the sub-Saharan African context when investigating treatment-related inconsistencies. A population-based survey in Kenya found viral suppression in 30% of people who reported they were HIV positive but not on ART; the authors suspected that participants misunderstood questions about ART [9].

Differentiating sources of misreporting is critical, because the strategies to address them are quite different. As Hewett et al. [21] found, self-interviewing techniques, while successful in reducing social desirability bias by increasing privacy [18, 40], can exacerbate misreporting due to misunderstanding of terms. Authors reduced inaccuracies by employing illustrations and in-person interviews that provided the opportunity to discuss the meaning of male circumcision. In a research context, increased accuracy of HIV treatment information may be achieved by accompanying survey items with pictures of ARVs for participants to compare with their own medications.

In 2016, the National Department of Health in South Africa initiated a universal test and treat policy, along with differentiated care for stable patients to minimize health facility visits and decongest clinics [41]. Currently, adherent patients may receive faster service through spaced and fast lane appointments, and the option to collect medications in community settings. Resulting longer periods between visits with providers may further limit patients’ knowledge or literacy about their health and treatments. Patient-provider communication is a challenge world-wide, and this study setting is no exception. Care must be taken to monitor the quality of post-test counseling and follow-up, while ongoing care services should seek to ensure that clear and complete information is provided at each visit, and continually re-assess patient understanding. As one example, a two-day workshop on patient-centered communication for clinicians in Tanzania successfully increased non-adherence reporting by patients on ART [42]. With differentiated care models slated to be scaled up in South Africa’s National Strategic Plan for HIV, TB, and STIs for 2017–2022 [41], there may be a need to consider assessing patient understanding prior to recommending reduced clinic visits, or to ensure that patient education is provided at medication collection points.

Limitations

While this study offers the strengths of biomarker comparisons with self-report, and qualitative follow-up interviews to understand the causes of discrepancies, the number of qualitative interviews was limited. Secondly, quantitative examination of discrepancies between reported HIV testing frequencies and time of diagnosis were only explored after qualitative data collection provided some insights into these misunderstandings. While the issue arose in interviews, it was not built into the selection criteria for qualitative interviews or into the interview guide, which would have allowed for a larger qualitative sample size and more systematic qualitative exploration of the discrepancy.

Conclusion

We found inconsistencies between self-reported responses and biological data on HIV infection and treatment, but little evidence of social desirability bias in a non-clinical setting. A substantial proportion of inconsistencies seem unintentional and largely based on poor understanding of treatment regimens and HIV monitoring. The majority of discrepancies may be addressed by enhancing patient-provider communication in HIV care settings to ensure patient understanding of their diagnosis, care and monitoring, and treatment regimens. Further inquiry into the quality of information shared with patients and the language used to provide information is warranted. Additionally, finding alternate ways to assess temporal questions in survey design and ensuring clear terminology around medical services should mitigate inconsistencies, as might additional checks and balances within a survey’s branching patterns.