Unannounced phone-based pill counts for monitoring antiretroviral medication adherence in South Africa

Abstract

Background

Unannounced phone-based pill counts (UPC) are an objective measure of medication adherence that may be used in resource limited settings. The current study reports the feasibility and validity of UPC for monitoring antiretroviral therapy (ART) adherence among people living with HIV in South Africa. People living with HIV (N = 434) in an economically impoverished township and receiving ART for at least 3-months completed: two UPC in a one-month period; measures of clinic and medication experiences; and provided blood samples for HIV viral load and CD4 testing. Analyses compared two methods for managing values of over-dosing (> 100%), specifically censoring values to 100% (> 100% = 100%) vs. subtracting over-dosing from two months of perfect adherence (200% - > 100% value).

Results

Findings showed that two UPC calls were successfully completed with 91% of participants in a one-month period. The average number of call attempts needed to reach participants was 2.4. Results showed that lower UPC adherence was significantly associated with male gender, alcohol use, higher HIV viral loads, lower CD4 cell counts, running out of ART, and intentionally not taking ART. Comparisons of methods for adjusting over-dosing found subtraction yielding a better representation of the data than censoring.

Conclusions

UPC were demonstrated feasible and valid with patients receiving ART in a resource limited setting and offers a viable method for objectively measuring ART adherence in these settings.

Keywords:

• Adherence
• adherence measurement
• HIV treatment
• medication monitoring
• objective measurement
• clinical trials

Introduction

Oral administration of antiretroviral therapy (ART) remains the cornerstone of HIV treatment and prevention in low and middle income countries [1,2] and ART requires sustained optimal adherence to effectively suppress HIV replication [3–5]. Pill counts are among the most widely used objective measure of ART adherence [6–8]. Most common are pill counts conducted in clinics and other facilities, which requires patients to bring all of their medications to the setting, leaving the pill count vulnerable to participant manipulation and other sources of error [9]. To resolve the limitations of facility-based pill counts, Bangsberg et al. [10] developed procedures for conducting unannounced pill counts in patient’s homes. Bangsberg et al. demonstrated that unannounced home-based pill counts are reliable and valid, with monthly adherence correlating with electronic medication monitoring [10–14] as well as HIV viral load [10,15,16]. Unannounced home-based pill counts have subsequently been demonstrated valid in a wide-range of contexts, including low-income and resource-limited countries [17–19].

Unannounced home-based pill counts, however, are burdensome to participants and are costly in terms of staff resources. While home visits are feasible in a concentrated area, such as residential clusters like housing developments or rural villages, in most settings significant amounts of time will be spent getting to participants’ homes and may require multiple visit attempts [20]. Home visits also carry added ethical concerns regarding participant privacy and assuring staff safety.

A potential solution to the challenges of unannounced home visits is to conduct pill counts remotely, specifically over the phone [21]. Researchers have therefore adapted Bangsberg et al.’s procedures for home-based pill counts for phone administration. Unannounced phone-based pill counts (UPC) have been used with such populations as recently released incarcerated persons [22], children and adolescents with perinatally acquired HIV infection [23,24], people residing in urban [25] and rural areas [26], and in a variety of clinical trials [27–29]. In validation studies, UPC have demonstrated nearly perfect concordance with contemporaneous unannounced home-based pill counts [15,30], significant associations with HIV viral loads in adults [16,20] and youth [23,24], with no evidence for manipulation or assessment reactivity [16].

To date, however, there are few reports of using UPC outside of the US. In Lesotho and Ethiopia, Hirsch-Moverman et al. [31] tested the implementation of UPC to monitor ART and TB treatment adherence in clinics. Hirsch-Moverman et al. reported that rates for reaching participants ranged from 20% to 34% of attempted calls completed, with 71% of participants reached at least once. The mean number of calls attempted to reach participants was 1.9, with similar completion rates for men and women. The researchers concluded that UPC were not feasible for clinical implementation, but did not report adherence values or their validity. Other researchers have reported challenges when conducting UPC in sub-Saharan Africa, including limited access to phones and finding private areas of the home to conduct pill counts [32].

The lack of positive results for using UPC in resource-limited settings stands juxtaposed to studies conducted in the US. Here we report findings from a study of UPC to assess ART adherence in Cape Town, South Africa. The aim of this study was to test the feasibility and validity of UPC for research in a resource-limited setting.

Methods

Participants and setting

The public health clinics participating in the current research were located in a resource-limited urban community in Cape Town, South Africa. Participants were recruited from January 2021 through March 2022 onsite at two clinics located within 2.5 kilometers of each other. We enrolled only patients who were age 18 or older, had access to a phone and were receiving ART for at least the previous 3-months to assure sufficient time for viral suppression. Participants were recruited while waiting to receive their ART in the clinic dispensaries at all times and days of the week. All research staff were native Xhosa speakers and fluent in English, the two languages of this community.

Procedures

Following clinic recruitment, participants were scheduled for an enrollment appointment at a research site located in the same community as the clinics. The enrollment appointment commenced with written informed consent and was followed by participants completing measures and receiving training in the UPC procedures (see below). In addition, participants provided blood samples drawn by a phlebotomist and tested for HIV viral load, CD4 cell counts, and serum creatinine by a laboratory service contracted for the study. Participants had access to the lab results in written form. Participants were compensated for their time and effort, receiving US$9.50 for the enrollment appointment and each of the two UPC calls, and US$12.50 for going to the lab to provide a blood sample. All study protocols were approved by ethics committees at the University of Connecticut and South African Medical Research Council. The study was registered with clinicaltrials.gov (NCT01359280). The authors declare no conflicts of interest.

Unannounced pill counts for ART adherence

We used the same UPC protocol previously reported [15,16,18,19,24,28,29], which was adapted from unannounced home-based pill counts [10]. The UPC protocol has been validated against home-based pill counts and viral load in the US [15,20], but has not yet been reported outside of the US. Participants were called by a trained phone assessor at an unscheduled time to count their pills within 3 days of the intake session and were called again to count their pills between 28 and 32 days later. There was no limit on the number of attempted calls to reach participants, with phone assessors calling daily within a 2-week window centered on 30 days since the first pill count. For each call, assessors verified that participants had their medications available and were in a private place to count their pills. The phone assessor asked participants to bring all medications in their home to a flat surface, including closed bottles, containers, loose pills, and pills kept in their pockets, purses, bags etc. If more than one medication was taken, participants sorted the medications into clusters. The participant then reported the dates and quantities dispensed from labels on each pill bottle or dispensing bag. Participants were then asked to count their pills aloud and reported the number to the phone assessor and repeated the count to ensure accuracy.

Calculating adherence

Adherence was calculated for all ART medications counted. Adherence was defined by the difference between pills counted at the two pill counts divided by the pills prescribed, taking into account the number of pills dispensed, pills lost, gained, and taken that day. Stopped or changed medications were adjusted for the number of days between the previous pill count and the stop date. Medication refill information, specifically dispensed dates and quantities were used to verify the accuracy of medications dispensed over the course of the pill counts. We also recorded the prescribed number of doses taken per day and the number of pills taken per dose.

To calculate adherence, a ratio was taken of the number of pills dispensed plus the number of previously counted pills less the number of pills counted at the second assessment, divided by the number of pills prescribed to be taken each day multiplied by the number of days since the previous pill count. Adherence was calculated for each ART medication and for participants taking more than one medication the values were averaged.

Adjusting for over-dosing

Previous studies have shown that patients who do not take all of their prescribed doses of ART in a given month may ‘over-dose’ in subsequent months because of the extra pills they have on hand. Previous research has adjusted for over-dosing by censoring the data such that any value greater than 100% is recoded as 100% [33]. Clinically, in one month of over-dosing patients have taken more than enough ART for clinical benefit, albeit at the risk increased toxicity [34]. Censoring values greater than 100% assumes that over time there will be periods of under-dosing. Censoring values > 100% to equal 100% in a single time point, however, may over-estimate adherence. In the current study, we compared censored over-dosing (> 100% = 100%) to an alternative approach by subtracting the adherence value from 200%, the maximum possible adherence over two months. For example, an adherence value of 125% is recoded using the formula (200% − 125%) = 75%. Conceptually, this approach assumes 75% of medications were taken in the previous month because an additional 25% were on hand. We compared the censoring and subtraction approaches for dealing with over-dosing.

Biologically determined health status

Participants provided whole blood samples for HIV viral load, CD4 cell count, and serum creatinine testing. HIV viral load was determined by reverse transcription-PCR testing with sensitivity of 40 RNA copies/mL. We defined detectable and undetectable viral loads using 100 copies/mL, a threshold that nearly eliminates most errors caused by viral load blips or assay variability [20]. CD4 cell counts (cells/mL) were determined by flow cytometry. We also conducted creatinine testing with elevated creatinine defined as 65 or greater for men and 52 or greater for women.

Additional study measures

Demographic and health characteristics

Participants reported their gender, age, marital status, history of TB treatment, the month and year they tested HIV positive, when they initiated ART and completed the Alcohol Use Disorders Identification Test [(AUDIT [35] with the consumption subscale, AUDIT-C [36–38]].

Clinic and ART experiences

As part of the second UPC call, participants were asked about their most recent clinic experiences, including interactions with nurses. We also asked about experiences with their medications over the previous month including running out of medications and storing medications so others would not see them. Responses were dichotomous as to whether or not the experiences occurred.

Intentional non-adherence

During the second UPC call, participants were asked whether they had missed their medications in the previous month and if so, whether they had skipped taking their medications because of social situations, side-effects, feeling their body was too sensitive, they had been drinking alcohol, or they did not have food to take their medications. Responses were dichotomous as to whether or not they had engaged in the action.

ART management

As part of the UPC call, the assessor asked participants about the management of their medications over the previous month. ART management included whether the participant used a pill box, carried their medications with them out of their home, shared, borrowed, lost or threw away medications.

Data analyses

We conducted descriptive analyses to examine the distributions of pills counted and the associations of adherence with demographic and health characteristics. Descriptive statistics for pill counts included data from all 482 participants who completed the first UPC call. Analyses for adherence values and groups defined by adherence required participants who completed both UPC calls (n = 434). Comparisons between groups (e.g. participants on 1^st and 2^nd line ART, gender), used t-tests for continuous variables with Cohen’s d for effect size and contingency table chi-square (X²) tests for categorical variables with Cramer’s v for effect size. All statistical tests defined significance as p < .05.

Results

A total of 995 patients were approached for the study; 8 were ineligible because they were under age 18, 45 had not yet been on ART for at least 3-months, 104 did not have access to a phone, and 125 declined participation. Among 713 patients who agreed to participate, 488 (68%) attended the enrollment appointment and enrolled in the study. The majority of participants were women (n = 362, 75%) and nearly all participants identified as Black (n = 478, 99%) and were not married (n 417, 87%). The mean age for the sample was 37.2 years (SD = 9.4). Seventy percent (n = 336) had attained less than grade 12 education. Forty percent (n = 194) had CD4 counts less than 500 cells/mL and 3% (n = 17) had elevated blood serum creatinine. One third of the sample (n = 157, 33%) had a history of being treated for TB and 5% (n = 25) were currently being treated for TB. On average, it had been 6.8 years (SD = 5.7) since testing positive and 6.3 years (SD = 4.8) since initiating ART. The majority of participants were receiving first line ART (n = 455, 94%), with the remaining participants taking second line ART (n = 27, 6%).

Reaching participants for pill counts

We attempted to execute two UPC for each participant over the 28 to 32 day period. Among the 482 participants, we made 1,170 call attempts for the first pill count (mean = 2.4, SD= 1.8, median = 2, range = 1 to 14) and 1,092 call attempts for the second pill count (mean = 2.4, SD = 1.8, median = 2, range = 1 to 14). We were able to contact all participants for the first call and were able to contact 434 (90%) for the second call.

There were no differences between men and women for the number of first call attempts (men, mean = 2.6, SD = 2.0; women, mean = 2.3, SD = 1.7), t = 1.2, p > .10, and second call attempts (men, mean = 2.7, SD = 2.1; women, mean = 2.3, SD = 1.6), t = 1.2, p > .10. In addition, there were no differences in call attempts needed to reach participants for the first call between those who drank alcohol (mean = 2.5, SD = 1.8) and those who did not drink alcohol (mean = 2.2, SD = 1.7), t = 1.4, p > .10. However, participants who drank alcohol required more call attempts to reach them for their second call (mean =2.6, SD = 2.0) than participants who did not drink alcohol (mean = 2.2, SD = 1.2), t = 2.4, p < .01.

Managing medications

Among the 482 participants completing at least one pill count, 5% (n = 32) were unable to count their pills on their own and required assistance from a friend or family member, of which 19 completed both pill counts. Fourteen percent of participants (n = 63) used a pill box device to organize their medications. Most participants reported carrying medications with them on weekends and when traveling (n = 301, 69%), and few (n = 22, 5%) carried medications with them daily, such as to work or school. It was uncommon for participants to share their ART with others (n = 42, 9%), borrow ART from others (n = 35, 8%), lose their ART (n = 35, 8%) or remove medications from their bottles for a reason other than to use a pill box (n = 21, 4%). However, nearly one in five participants (n = 83, 19%) reported throwing away excess medications.

Pill counts and calculating adherence

Across the two pill counts we counted a total of 49,939 pills, representing a mean of 103 pills (SD = 88.7) per participant across the two pill counts. The minimum and maximum number of pills counted for first line ART were 15 and 791 (median = 77), and the minimum and maximum number of pills counted for second line ART were 88 and 1007 (median = 235, see Table 1).

Table 1. Descriptive characteristics for pills counted to calculate ART adherence for all participants taking first and second line ART.

	1^st Line ART (N = 455)		2^nd Line ART (N = 27)
Variable	N	%	N	%	X^2	v
ART Doses/day
1 Dose	451	99	12	44	201.2**	0.64
2 Doses	4	1	15	56
ART pills/dose
1 Pill	451	99	19	70	86.7**	0.42
2 Pills	4	1	8	30
	M	SD	M	SD	t	d
Pills counted	92.6	60.8	288.3	208.3	4.8**	1.27

Note: ART = antiretroviral therapy.

**p < .01; d = Cohen’s d; v = Cramer’s v.

Adjusting for over-dosing

A total of 14% of participants (n = 69) had adherence values over 100%, with a range of 101% to 157%. The means, standard deviations, medians and percent of participants below 70%, 80% and 90% adherence are shown in Table 2. Censoring over-dosing (> 100% = 100%) and subtracting over-dosing (200% - > 100% value) both adjusted adherence to lower than original distributions. Censoring did not alter values for participants with adherence less than 100%. In contrast, subtraction reduced adherence for the over-dosing participants to less than 100%. Specifically, 6 participants who had over 70% adherence converted to below 70% adherence after subtraction, as was the case for 11 participants at 80% cut-off and 28 participants at 90% cut-off. To avoid inflated adherence values from censoring the data, all further analyses use the subtraction adjusted values for over-dosing.

Table 2. Adherence distributions for unadjusted, censored and subtracted adjustments for over-adherence.

	Unadjusted Adherence		Censored Adherence (>100% = 100%)		Subtracted Adherence (>100% value – 200%)
	N	%	N	%	N	%
< 70%	42	9.7	42	9.7	48	11
< 80%	81	18.6	81	18.6	92	21.1
< 90%	160	36.8	160	36.8	188	43.2
Mean	91.2		89.3		87.5
SD	17.2		14.8		14.8
Median	96.4		96.4		92.8

Internal validity

We observed lower ART adherence among men than women and participants who reported current alcohol use had lower adherence than participants not using alcohol (see Table 3). Participants on 2^nd line ART also had significantly lower adherence than those on 1^st line ART. In addition, participants with detectable HIV viral loads had significantly lower adherence than those with undetectable viral loads. Compared to participants with less than 90% adherence, those with 90% or greater had lower log value viral loads. In addition, participants with higher adherence had higher CD4 cell counts. There were no differences in age between the higher and lower adherence groups (see Table 4).

Table 3. Differences in adherence by gender, alcohol use, ART regimen, and viral load.

	% ART Adherence
Characteristic	M	SD	t	d
Gender
Men	84.7	16.1	2.2*	.24
Women	88.4	14.2
Alcohol use
Currently drinking alcohol	86.1	15.8	2.3*	.23
Not currently drinking alcohol	89.5	12.8
ART regimen
1^st Line	87.9	14.7	2.7**	.56
2^nd Line	79.7	14.8
Viral load
Undetectable (≤ 100 copies)	88.0	14.6	1.9*	.26
Detectable (> 100 copies)	84.1	15.3

Note: ART = antiretroviral therapy.

*p < .05.

**p < .01; d = Cohen’s d.

Table 4. Differences in age and health characteristics between lower and higher adherence groups.

	Lower adherence (< 90%) (n = 188)		Higher adherence (> 90%) (N = 246)
Characteristic	M	SD	M	SD	t	d
Age	37.8	9.6	37.1	9.3	0.8	.08
Log Viral Load	0.69	1.3	0.45	1.1	2.0*	.19
CD4 cell count	573.0	308.0	645.7	363.4	2.2*	.21
Alcohol use (AUDIT-c) score	2.6	2.7	1.6	2.4	3.8	.37
Call attempts for first pill count	2.5	1.8	2.2	1.5	1.9	.18
Call attempts for second pill count	2.3	1.5	2.4	1.7	0.4	.04
	N	%	N	%	X^2	v
History of TB	70	37	94	38	5.4	.11
CD4 < 500 cells/mL	85	45	2.2	1.5	2.2	.07
Elevated serum creatinine	12	6	5	2	5.4	.11

Note: ART = antiretroviral therapy; TB = tuberculosis.

*p < .05.

**p < .01; d = Cohen’s d; v = Cramer’s v.

Table 5 shows the associations of adherence with clinic experiences and intentional non-adherence. Participants with less than 90% adherence did not differ from those with 90% or greater adherence on experiences with nurses asking them about side-effects or missed ART during their most recent clinic visit. However, participants with lower adherence were more likely to have run out of ART during the month between pill counts. With respect to intentional non-adherence, participants with lower adherence were significantly more likely to report skipping their ART in social situations, because of side-effects, perceived sensitivity, and alcohol use. The lower and higher adherence groups did not differ for having skipped their ART because they did not have food.

Table 5. Differences in clinic experiences and medication management between lower and higher adherence groups.

	Lower adherence (<90%) (n = 188)		Higher adherence (>90%) (N = 246)
Clinic and ART Experiences	n	%	n	%	X^2	v
Healthcare worker discussed side- effects at last clinic visit	66	35	69	27	2.5	.07
Healthcare worker asked about missing ART at last clinic visit	71	37	75	30	2.5	.07
Ran out of ART^a	12	6	5	2	5.3*	.11
Stored ART out of bottles so others would not know they are ART^a	17	9	24	6	0.1	.04
Intentional non-adherence^a
Skipped ART when with other people	14	7	8	3	3.8*	.09
Skipped ART when others were nearby	37	19	16	6	17.2**	.19
Skipped because of side-effects	19	10	7	2	9.9**	.15
Skipped because body is sensitive	16	8	8	3	5.6*	.11
Skipped ART when drinking alcohol	45	35	25	22	4.9*	.14
Skipped ART because did not have food to take with pill	17	9	12	5	2.9	.08
ART management
Uses pill box	47	25	16	6	29.6**	.26
Carries medications
Never carries medications	40	21	72	29	4.3	.10
Carries daily; work or school	8	4	14	5
Carries periodically; weekends and when traveling	140	75	161	65
ART Management
Shares medications with others	22	11	20	8	1.5	.06
Borrows medications from others	19	4	16	3	1.9	.06
Has lost medications	18	9	17	6	1.0	.04
Throws away excess medications	36	19	47	19	0.1	.01
Removes medications from bottles	8	4	13	5	0.2	.02

Note: ^a past month.

*p < .05.

**p < .01; d = Cohen’s d; v = Cramer’s v.

Discussion

The current study demonstrated the feasibility and validity of UPC for research conducted outside of the US. Only 10% of patients approached did not have access to a phone and 90% of participants completed two UPC calls within one month. The number of call attempts required to reach participants was similar to that in US studies [15,20] and barriers to conducting UPC were uncommon, such as taking medications outside the home and sharing medications with others. Adherence values appeared valid, with lower adherence associated with higher HIV viral loads, lower CD4 cell counts, 2^nd line ART, current alcohol use, and male gender. In addition, adherence was associated with having run out of medications and multiple indicators of intentional non-adherence. As reported in other research conducted with people living in poverty [10,15] and in resource limited settings [31], numeracy and literacy challenges did not appear to interfere with UPC. Our findings are also the first to test alternative approaches to managing the effects of over-dosing on adherence data, with subtracting over-dosing performing better than censoring values in a single assessment.

Our results are far more optimistic for using UPC in sub-Saharan Africa than previous studies [32]. We found that 90% of patients had access to a phone and participants did not report concerns about having a private place for counting pills. Our study demonstrated high completion rates that are needed for any meaningful study of ART adherence. Hirsch-Moverman et al. [31] reported an average of 1.9 call attempts needed to reach participants, a rate that is similar to our study and in the US [20]. At odds with our results, less than 30% of participants in Hirsch-Moverman et al.’s study completed the two UPC calls necessary to calculate adherence. However, Hirsch-Moverman et al. treated UPC as a potential means of monitoring ART in clinical practice and did not compensate participants for assessments. In addition, differences in settings may account for low completion rates, with Hirsch-Moverman et al. reporting that the most common reasons for unsuccessful UPC calls included participants not having a phone, network problems, phones being switched off, and phones not answered. These issues were uncommon in our study and may become less common throughout sub-Saharan Africa as access to telecommunications continues to increase.

While our findings support using UPC in resource limited settings, several limitations to our study should be considered. Our research was conducted in Cape Town, South Africa, the most resourced city in the most resourced country in southern Africa. Our sample was also constrained to patients living in a single Cape Town township that cannot be generalized to other communities. In addition, any interest in implementing UPC outside of a research setting without participant compensation may not have sufficient completion rates [39]. Finally, although UPC adherence demonstrated evidence for validity, we cannot rule out the possibility that some participants may have attempted to manipulate their pill counts, albeit an exceedingly difficult task in mental calculation. With these limitations in mind, we believe that our results support the use of UPC for monitoring ART adherence in research conducted in southern Africa.

Unannounced phone-based pill counts offer a viable option for monitoring ART adherence in clinical trials as well as clinical care. While HIV viral load is often considered the ‘gold standard’ for ART adherence, monitoring viral load rather than actual adherence is an after-the-fact strategy. Behavioral measures of adherence, such as UPC, are more akin to an early detection strategy that affords intervention opportunities to prevent viral rebound and sustain viral suppression. Pill counts can be performed by trained non-medical staff, including lay counselors and community health workers. In addition to monitoring adherence, UPC can be conducted along with brief phone-delivered counseling to assist patients in problem-solving barriers to adherence. In addition, the pill count procedure itself can serve as an educational tool regarding storing, organizing, and self-monitoring medications. It is therefore recommended that UPC be used as an objective measure of ART adherence in clinical research and implemented in care settings as an enhanced adherence service for patients at-risk for non-adherence.

Conclusions

Objective measures of medication adherence are a necessary element of research. While self-report measures of adherence have improved [40,41], they remain limited by over-estimating adherence and relying on patients to recall events they have forgotten [42,43]. In addition, although long-term injectables may ultimately replace oral ART, it is likely that injectables will not be universally available to patients in resource limited settings for some time and not all patients will opt-in for injectable ART. Thus, the need for objective measures of oral ART adherence is likely to persist and UPC offers a viable option for monitoring medication adherence across diverse settings.

Authors’ contributions

SK and CM conceptualized the study design, conducted data analyses and contributed to the writing of the paper. EB, BS and MK oversaw scientific execution of the study, contributed to the conceptualization of the study design, and contributed to the writing of the paper. All authors read and approved the final manuscript.

Acknowledgments

The authors appreciate the involvement of the City of Cape Town clinics that participated in this research.

Disclosure statement

The authors declare no competing interests.

Data availability statement

Data are retained by the authors and available upon reasonable request.

Additional information

Funding

This work was supported by the National Institute of Mental Health (NIMH) Grant R01MH19913.