Pharmacy and self-report adherence measures to predict virological outcomes for patients on free antiretroviral therapy in Tamil Nadu, India

Abstract

Over 480,000 individuals receive free antiretroviral therapy (ART) in India yet data associating ART adherence with HIV viral load for populations exclusively receiving free ART are not available. Additionally estimates of adherence using pharmacy data on ART pick-up are not available for any population in India. After 12-months ART we found self-reported estimates of adherence were not associated with HIV viral load. Individuals with < 100% adherence using pharmacy data predicted HIV viral load, and estimates combining pharmacy data and self-report were also predictive. Pharmacy adherence measures proved a feasible method to estimate adherence in India and appear more predictive of virological outcomes than self-report. Predictive adherence measures identified in this study warrant further investigation in populations receiving free ART in India to allow for identification of individuals at risk of virological failure and in need of adherence support.

Background

It is estimated that 2.4 million people are living with HIV in India with over 480,000 people receiving free National AIDS Control Organization (NACO) funded antiretroviral therapy (ART) (1) However, ART coverage remains a challenge with somewhere between 23 – 55% of eligible patients receiving ART during 2009 (2). Furthermore a smaller group of individuals estimated to be somewhere between 6 – 25% of the population receiving ART (3) pay for these medications in the private system. Rapid scale-up of free government funded ART has occurred since 2004 yet it is recognized that nearly 20% of patients are presenting at a very late stage (CD4 count < 50) with an increased risk of mortality. NACO has responded by decentralizing ART services to the district and sub-district level in an attempt to close gaps in public health infrastructure between HIV testing and treatment programs, and non-HIV related health services (4).

Achieving optimal adherence to ART is critical to prevent treatment failure, HIV related mortality, emergence of HIV drug resistance, and preserve the efficacy of available ART (5-8). High levels of adherence have been reported in meta-analyses of studies performed in sub-Saharan Africa and understanding this success has become a focus of investigation (9, 10). Maintaining maximal adherence is particularly important in countries such as India where switching to second line ART is more costly, complex, and restricts future treatment options (11). In addition patients in India suspected of treatment failure on first-line ART are only recommended to commence second-line ART once good adherence has been ensured (12).

Multiple methods to estimate adherence to ART are available and they include: self report, pharmacy adherence measures, electronic pill container caps (MEMS caps), measuring antiretroviral drug concentrations and web-enabled pill boxes (13) MEMS caps are considered the gold standard for estimating adherence by many authors but their use is largely confined to research settings in a similar manner to measuring drug levels or web based systems that record the opening of pill boxes (13, 14). Pharmacy adherence measures (PAMs) estimate ART adherence using pill pick-up data that is routinely recorded at pharmacies dispensing ART according to the prescription of a medical practitioner. This contrasts with the most widely utilized “over the counter” practice for delivering medication in India where pick-up data is often not available. PAMs predict virological and other clinical outcomes in high and low-middle income countries (15) (LMICs) and have been adopted by the World Health Organisation (WHO) as a standard for estimating population level adherence (16, 17). Interestingly most LMIC data originates from sub-Saharan Africa (15), with many prominent studies performed in the private health sector (7, 18, 19). Until now, no studies from India have reported PAMs and their association with virological outcomes which is notable considering the potential advantages of PAMs over self-report adherence measures (13, 15). Furthermore, only 2 cohorts in India have documented adherence to ART in association with virological outcomes with both cohorts assessing adherence by self-report. Importantly neither cohort reported on populations exclusively receiving free ART (20-23). Shah reports on patients in the private system who paid out-of-pocket for ART (21), while a Bangalore cohort document self-reported adherence predicting virological outcomes for individuals receiving free ART, or paying for ART in the private system (20, 22-24). The Bangalore cohort also documented more treatment interruptions (23) and virological failures (22) in patients paying for ART but did not document associations between adherence and virological outcomes for individuals only receiving free ART. This is an important distinction as at least 75% of individuals now receive free NACO funded ART (1, 3, 25) and ART cost has been repeatedly reported as a barrier to ART adherence in India (22, 23, 26, 27). Therefore relationships between adherence and viral load may be different from what is currently reported for most individuals receiving ART in India. In addition to ART cost other barriers to adherence have also been reported in India including: stigma, ART side effects, depression and co-morbid medical conditions (21-23, 26-29).

Therefore, our objective was to determine associations between ART adherence and HIV viral load for individuals receiving free ART within the public sector in India using both self-reported and pharmacy measures of adherence.

Methods

Population

The study was conducted at a NACO sponsored ART Clinic at Christian Medical College (ACTFID), Vellore, Tamil Nadu and is one of 5 sites providing free government sponsored ART in the Vellore district (population 3.5 million) The clinic is one of many public-private partnership sites in India where non-governmental and private organizations collaborate with NACO to provide free clinical care and ART services via the NACO program (12). Patients attended monthly for medical review and picked-up ART from a pharmacy staffed by a dedicated pharmacist within the clinic.. Patients did not require specific appointment times to attend the clinic which was open 6 days a week and all routine pathology including testing for CD4 T-cell counts was performed at a laboratory approximately 10 minutes walk from the clinic. At the time of the study 500 people were receiving ART and the clinic was staffed by two doctors, a social worker, pharmacist and clerical staff. The clinic was also able to manage some other medical conditions such as intercurrent respiratory or skin infections. Patients requiring hospital admission or other specialist medical care were referred to inpatient services or other outpatient clinics within Christian Medical College.

Design

The study was a retrospective cohort of consecutive adults initiating ART and followed for 12-months. Patients were recruited from October 26, 2009, until October 10, 2010 and eligible for inclusion if initiating first line ART (12). Patients transferred in from other sites or re-initiating ART after a treatment interruption were excluded. Self-reported adherence (30, 31) and HIV viral load were determined for patients remaining in care 12-months after ART start.

Procedures

230 consecutive initiators of ART were identified during a routine clinic visit after 11-15 months of ART. Patients were considered lost to follow-up (LTFU) at 12-months if they had not attended or picked up ART within 90 days of their last missed appointment. All baseline clinical and demographic data was abstracted from clinical records and ART dispensing data from pharmacy records.

Standardized self-report adherence measures asked about adherence since; initiating ART, or the preceding 30-days.(30) An additional 30-day self-report measure was the visual analog scale (VAS) where patients indicated on a line marked from 0% to 100% the point that best corresponded to the percentage of pills taken (31). Adherence questions were originally written in English, translated into Tamil or Telugu and independently back-translated. Questionnaires were administered in local languages by trained staff experienced in HIV counseling and treatment. ART adherence was also estimated using the medication possession ratio (MPR). This was calculated by dividing the days of ART dispensed by the period of time from ART start to the day of recruitment.. All patients completing 12-months ART provided written informed consent and the study was approved by the institutional review boards of Christian Medical College, Tufts University Health Sciences and Monash University.

Laboratory testing

The HIV viral load test was performed at the same time as the routine assessment of CD4 T-cell counts (FACSCount) after 12-months of ART. HIV viral load was assessed by the Artus HIV-1 RT-PCR (Qiagen) with a detectable viral load defined as greater than 200 copies/mL based on viral load blips rarely being above 200 copies/mL (32)

Analysis

Baseline characteristics and dichotomous adherence estimates after 12-months ART were compared to 12-month viral load using χ², Fisher’s exact, Student’s t-test, and Wilcoxon rank-sum tests as appropriate. Odds ratios of a detectable viral load after 12-months ART were also calculated for the estimates of adherence. The 30-day self-report question was dichotomized around excellent (highest adherence category) versus less than excellent adherence, the self-report question for the entire period receiving ART was dichotomized around those reporting never having missed versus ever having missed ART and the VAS was dichotomized around 95% adherence. Dichotomous MPR estimates were created with different thresholds to define low adherence (<95%, <100%). To establish if MPR accuracy could be improved we combined the most predictive MPR measure with the 30-day and 12-month self-report questions. Individuals with low pharmacy adherence and less than excellent adherence in last 30-days, or ever reported missing ART were considered to have low adherence for this variable. Overall accuracy of adherence estimates was also assessed by calculating the area under receiver operating characteristic curves (AUROCs) and 95% confidence intervals (CIs) for continuous (MPR) or ordinal variables (self-report). 95% CIs of the AUROC that did not cross 0.5 indicated a statistically significant association. All analyses were conducted using SAS v9.2 (SAS Institute, Cary, NC).

Results

Baseline demographics

Baseline characteristics of 230 patients included: 65% male, 41% WHO clinical stage IV, active tuberculosis in 27%, and median CD4 141 T-cells micro/L (Table 1). After 12-months: 77% (n=177) were on ART of which 98% (n=174) undertook HIV viral load testing, 10% died, 8% transferred out, 5% were LTFU and no patients switched to second line therapy. Median CD4 T-cell count after 6 months was 309 cells/microL and after 12 months was 410 cells/microL which were both significant increases from baseline (p<.001) and 80% (n=140) of patients on treatment at 12-months had HIV viral load <200 copies/mL. There were no significant differences in baseline characteristics when stratified by viral load although a trend (p=0.08) for virological suppression was present in married individuals (Table 1).

Table I.

Baseline characteristic stratified by detectable viral load

Baseline characteristic		Total (n=174)	Viral load < 200 copies/mL (n=140)	Viral load > 200 copies/mL (n=34)	p- value
Age		38.3 ± 8.7	38.5 ± 8.7	37.3 ± 9.3	0.5
Gender	Male	105 (60.7)	84 (60.4)	21 (61.8)	0.9
Transmission Risk Factor	Heterosexual	136 (87.7)	108 (87.1)	28 (90.3)	0.5
	MSM	2 (1.3)	2 (1.6)	0
	Other	13 (8.4)	10 (8.0)	3 (9.7)
Education level	Non-literate	31 (19.4)	25 (19.5)	6 (18.8)	0.8
	Primary School	43 (26.9)	34 (26.6)	9 (28.1)
	Secondary School	68 (42.5)	56 (43.8)	12 (37.5)
	College	18 (11.3)	13 (10.2)	5 (15.6)
Employed		106 (66.7)	84 (65.6)	22 (71.0)	0.5
Marital Status	Single / Separated / Partner Died	50 (29.0)	36 (25.9)	14 (41.2)	0.08
	Married	123 (71.1)	103 (74.1)	20 (58.8)
Previous ARV exposure		5 (3.0)	3 (2.3)	2 (6.3)	0.2
Baseline WHO clinical stage	I/II	74 (42.5)	58 (41.4)	16 (47.1)	0.7
	III	38 (21.8)	30 (21.4)	8 (23.5)
	IV	62 (35.6)	52 (37.1)	10 (29.4)
Receiving TB treatment		40 (23.4)	34 (24.6)	6 (18.2)	0.4
ART regimen	D4T/3TC/NVP	78 (44.8)	59 (42.1)	19 (55.9)	0.5
	AZT/3TC/NVP	58 (33.3)	49 (35.0)	9 (26.5)
	D4T/3TC + EFV	27 (15.5)	22 (15.7)	5 (14.7)
	AZT/3TC + EFV	11 (6.3)	10 (7.1)	1 (2.9)
CD4 (cells/microL)		146 (77-202)	142 (73-201)	159 (81-219)	0.5
HepBsAg positive		9 (6.0)	6 (5.0)	3 (10.0)	0.4

NOTE: MSM, men who have sex with men; ARV, antiretroviral; PMTCT, prevention of mother to child transmission; ART, antiretroviral therapy; WHO, World Health Organization; TB, tuberculosis; D4T, stavudine; 3TC, lamivudine; NVP, nevirapine; AZT, zidovudine; EFV, efavirenz

Adherence measures

Table 2 demonstrates associations between adherence estimates after 12-months ART and HIV viral load. All estimates of adherence solely using self-report were not associated with the virological outcome (p>.4). Furthermore AUROCs for self-report estimates demonstrated no association including: 30-day self-report 0.52 (95% CI: 0.42 – 0.61), last time missed ART 0.55 (95% CI: 0.45 – 0.65) and 30-day VAS 0.54 (95% CI: 0.44 – 0.63). The 12-month MPR with a 95% threshold was not associated with the virological outcome (OR 1.7, p=.2) but there was a significant association with the 100% threshold (OR 2.6, p=.01), although a greater number of individuals were considered to have low adherence with the 100% threshold (48.9%) compared to the 95% adherence threshold (16.7%). The MPR AUROC was 0.61 (95% CI: 0.50 – 0.72) demonstrating a statistical association albeit on the borderline of significance The variable that combined the 12-month MPR of 100% threshold with the 2 self-report questions was associated with HIV viral load (OR 2.1, p=.05) and less patients were considered to have low adherence (35.6%) compared to the MPR with 100% threshold not combined with self-report.

Table II.

Adherence measures and CD4 change after 12-months ART predicting viral load (n=174)

Adherence measure or CD4 criteria		Total	Viral load > 200 copies	Viral load < 200 copies	Odds Ratio	P value
30 day Self-report (5 point Likert item)	< Excellent	130 (76.5)	25 (73.5)	105 (77.2)	0.8	0.7
	Excellent	40 (23.5)	9 (26.5)	31 (22.8)
Self-report – Last time missed	> Never	57 (33.5)	13 (38.2)	44 (32.3)	1.3	0.5
	Never	113 (66.5)	21 (61.8)	92 (67.7)
30 day Visual analog scale	≤ 95%	50 (29.4)	12 (35.3)	38 (27.9)	1.4	0.4
	> 95%	120 (70.6)	22 (64.7)	98 (72.1)
12 Month MPR (Days ART / Whole time receiving ART)	< 95%	29 (16.7)	8 (23.5)	21 (15.0)	1.7	0.2
	≥ 95%	145 (83.3)	26 (76.5)	119 (85.0)
	< 100%	85 (48.9)	23 (67.7)	62 (44.3)	2.6	0.01
	≥ 100%	89 (51.1)	11 (32.3)	78 (55.7)
Combined Self-report and MPR (12 Month MPR < 100% + suboptimal adherence on either of 2 self-report measuresa)	Low adherence	62 (35.6)	17 (50.0)	45 (32.1)	2.1	0.05
	High adherence	112 (64.4)	17 (50.0)	95 (67.9)
NACO immunological criteria for treatment failureb	Positive	18 (10.9)	3 (8.8)	15 (11.5)	0.7	1.0
	Negative	147 (89.1)	31 (91.2)	116 (88.5)

Values represent n (% with that characteristic)

Characteristics compared by Chi-squared test or Fisher’s exact test if expected cell frequencies ≤ 5

^a< Excellent adherence in last 30 days, or ever reported missing ART

^bMinimum requirement baseline and 6 month CD4. Positive criteria; 6 or 12 month CD4 < 100, or 12 month CD4 50% lower than 6 month CD4, or 6 or 12 month CD4 < baseline CD4

NOTE: ART, antiretroviral therapy; MPR medication possession ratio; NACO, India national AIDS control organization

Discussion

This is the first report from India describing pharmacy adherence measures for individuals receiving ART and the first report from India documenting associations between any measure of adherence and HIV viral load for a population that has exclusively received free ART.

Importantly, and different from studies including patients who paid for ART (20, 21), we did not observe self-reported adherence predicting virological outcomes. A potential explanation is the increased likelihood of a social desirability bias (33, 34) leading to underreporting of missed doses in programs where patients receive free care compared to patients who pay for ART. Inaccurate and more socially desirable responses by individuals receiving free care may fail to detect associations between ART adherence and virological outcomes. Furthermore, objective assessments of adherence using pharmacy data were more closely associated with virological outcomes, with the 100% threshold variable significantly associated with viral load. This is notable as the 100% threshold establishes if individuals were in possession of ART for the entire 12-month period since initiation. By definition individuals with less than 100% pharmacy adherence did not have enough ART to take medication as prescribed for these first 12-months.

The 95% threshold of adherence is the most widely cited threshold to maximise virological suppression based on data from Paterson in treatment experienced patients receiving unboosted protease inhibitor based ART (35). Furthermore, attaining individual adherence above 95% is cited by NACO as one of the key goals of the national ART program (12). Subsequent studies have reported higher and lower thresholds predicting virological outcomes for populations receiving non-nucleoside reverse transcriptase inhibitor (NNRTI) based regimens in high-income and LMICs (19, 36). Therefore alternative thresholds to identify groups at risk for poor virological outcomes warrant consideration. Findings in this study suggest an MPR threshold of 100% may be more useful for defining individuals at risk of poor virological outcomes in this population. However, this threshold classifies approximately half the study population as having low adherence. The ability to target this patient group for viral load testing or adherence intervention may depend on available resources. Therefore selection of optimal adherence measures for different settings may be influenced by costs of subsequent interventions for patients with low adherence.

Combining a PAM with questions measuring self-reported adherence to more accurately identify a subpopulation at risk of a detectable viral load, in this study was above 200 copies/mL, is an innovative technique. This resulted in approximately one third of individuals defined as having low adherence yet this group was still significantly associated with the virological outcome. This finding suggests that combining different adherence measures should be further examined in populations receiving free ART in India. Replicating this technique in different settings may reinforce findings from this study and potentially identify alternate methods to accurately identify sub-populations at risk for virological failure or that require adherence support. In addition further research is necessary to identify risk factors for low adherence and virological failure for people receiving free ART in India. Barriers such as the stigma of HIV, medication side effects and depression have already been identified in studies where patients paid for ART and cost was the most commonly reported barrier (21-23, 26-29). Identifying barriers to adherence and targeting interventions to these factors is an essential step to improve virological outcomes for individuals receiving free ART in India, in addition to identifying the best methods to estimate adherence.

Finally, immunological criteria recommended to define treatment failure in India performed poorly for predicting viral load greater than 200 copies/mL with only 9% of subjects with detectable viral load satisfying CD4 change criteria. This is consistent with other LMIC data concerning the limited ability of CD4 criteria to detect virological failure (18, 37, 38) and supports efforts to identify non-virological factors that accurately identify individuals with virological failure, including assessments of ART adherence. Failure to correctly identify individuals failing virologically that continue NNRTI containing regimens, leads to accumulation of HIV drug resistance mutations, decreased efficacy of the current regimen, potential reduction in the activity of future regimens, immunological progression and increased risk of clinical deterioration. Furthermore, individuals who satisfy CD4 change criteria but remain virologically suppressed results in unnecessary switching to expensive second line ART. Despite the limited availability of testing for HIV drug resistance in India, surveys performed on patients initiating ART in 2007 and 2008 reported 8-9% of individuals initiating ART had drug resistance detected after 12-months ART (39). These data highlight the need for accurate measures to identify individuals at risk of failing ART that can limit the development of HIV drug resistance.

Pharmacy adherence measures were established using routinely collected data in the pharmacy register. This register is essential element to establish the volume of ART stock by documenting the amount of ART dispensed, hence there is an emphasis on accurate recording of data to ensure continuous antiretroviral supply. In practical terms estimating the MPR requires a clinic staff member to tally up the days of ART dispensed and divide that by the number of days since the patient initiated ART. This adherence estimate can be easily updated at subsequent ART pick-ups and integrated into the work flow of the clinic. Furthermore, if dispensing data is recorded electronically there is the potential for pharmacy databases to automatically generate the MPR based on the dates of ART pick-up and amount of ART dispensed.

Limitations of this study include the generalisability to other people in India receiving free antiretrovirals in different settings. However, considering the paucity of data examining adherence measures and virological outcomes for those on free ART in India the findings of this study still merit consideration in alternate settings. In addition MPR estimates in this study did not account for remnant pills which may have lead to estimates of adherence with different characteristics for predicting viral load. However a recent systematic review did not find evidence that adherence estimates that included counting remaining pills were superior to MPR for predicting virological outcomes (15).Finally the findings of this study were limited by a relatively low sample size to detect significant association between the measures of adherence and virological outcomes.

Conclusions

Pharmacy adherence measures such as the medication possession ratio are a feasible method to assess adherence within the public health model of care in India and appear more predictive of virological outcomes that commonly employed self-reported assessments of adherence. Combining the MPR with self-reported adherence is an innovative technique to further define at risk populations in this setting and warrants further investigation. As viral load testing is not currently required for monitoring ART in India and immunological criteria performed poorly for predicting HIV viral load, adherence measures such as the ones identified in this study should be further investigated to identify individuals at risk of virological failure and in need of increased adherence support.