A Modified HIV Continuum of Care: A Six-Year Evaluation of a Viral Load Cascade at a Hospital-Based Clinic in Kingston, Jamaica

Abstract

Background:

To achieve the goal of HIV viral suppression, provision of medication alone is not sufficient. Concomitant frameworks to evaluate HIV care delivery programs are needed.

Methods:

This study examined the care continuum at a hospital-based HIV clinic in Kingston, Jamaica using a modified HIV continuum of care, with an increased focus on viral load indicators (viral load samples taken, results returned, and viral suppression). A statistical analysis of patient flow through the care continuum to identify gaps in program delivery was performed. Key programmatic areas for process improvement and the utility of this approach for viral load suppression interpretation were identified.

Results:

Between 2010 and 2015, more than 1600 patients had been registered for care and more than 1000 had accessed antiretroviral therapy at this location. Consistent trends in programme performance were seen from 2010–2012. Although declines in the proportion of viral load samples taken and results returned occurred because of laboratory failures in 2013, the trend of increasing numbers and proportions of virally suppressed patients continued. Statistical analysis indicated that improvements in laboratory quality (fraction of viral load samples returned with accurate test results) could increase viral load suppression among patients at the clinic by up to 17%.

Conclusions:

Refining care delivery processes can significantly improve HIV viral load suppression rates. Expanding monitoring frameworks to include all of the essential processes that affect final outcome indicators can provide valuable insight into trends of outcome indicators and programme performance.

Introduction

Monitoring HIV treatment using viral load testing is a critical component of the HIV response. Suppressed HIV viral loads provide both individual and population-level benefits: effectively treated individuals have higher quality of life and are significantly less infectious than untreated individuals.^{1, 2} Based on the Joint United Nations Programme on HIV/AIDS (UNAIDS) proposed 90-90-90 targets³, the HIV care continuum^{4, 5} is being promoted as the appropriate monitoring framework. To achieve these ambitious goals, effective care delivery programs are essential.

The Centre for HIV/AIDS Research, Education and Services (CHARES) is a hospital-based HIV treatment site at the University of West Indies Hospital in Kingston, Jamaica. An estimated 1.8% (1.5 – 2.0%) of adults aged 15 to 49 in Jamaica are HIV infected.⁶ CHARES cares for the fourth largest complement of patients in the country, accounting for approximately 8% of the HIV-infected population ever linked to care, 10% of persons on ART, and 17% of virally suppressed patients. Analysis of the HIV care continuum – from HIV diagnosis to enrolment in care, receipt of ART, adherence to ART, monitoring and viral load testing, and viral load suppression^7–13 can identify gaps in program delivery and the areas that require strengthening.^14–17 Unpacking the care continuum allows for more in-depth assessment along the cascade.^18–20

In this study, we examine the care continuum at CHARES using a modified HIV continuum of care, which includes laboratory process data, with an increased focus on viral load indicators. We perform a descriptive analysis to highlight the utility of this approach in identifying gaps and areas for improved programme delivery and viral load suppression interpretation.

Methods

We performed a six-year retrospective analysis of secondary data from CHARES incorporating routine programmatic data and data extracted from a site-level electronic database. All patients who had ever been registered with and received care at the clinic were included in the analysis. Data was analysed with the goal of identifying key areas for improvement in program delivery at this site. Following definitions in the Jamaica National HIV Guidelines for Treatment and Care²¹, we considered three traditional care cascade stages: Population Enrolled in Care, Patients on ART, and Viral Load Suppression. To these, we added three intermediate stages: Patients Engaged in Care, Viral Load Samples Collected, and Viral Load Results Returned, described in Table 1. The inclusion of Engaged in Care, recently termed Retained on ART⁵, represents patients who are on ART and actively involved in care; the difference in numbers of Patients on ART and patients Engaged in Care accounts for patients lost to followup who are on ART. We include separate steps for Viral Load Samples Collected and Viral Load Results Returned to account for laboratory quality; not all samples collected are successfully processed and returned. This framework represents a conditional cascade²² whereby for a patient to reach a particular stage all previous stages must have been completed.

Table 1.

Definitions for Viral Load Cascade Stages

Stage	Definition
Population Enrolled in Care	Aggregate number of patients testing positive, referred to CHARES* and commencing care (accounting for known deaths and transfers)
Patients on ART	All patients who satisfied national guidelines for ART commencement, were prescribed ART, and collected at least one month of medication.
Patients Engaged in Care	All patients who commenced ART and had at least 2 clinician visits for any given year
Viral Load Samples Collected	All patients who presented for blood collection at CHARES and had a venous blood sample drawn for transport to a central laboratory for viral load analysis
Viral Load Results Returned	All paper-based result sheets which were returned from the central laboratory to CHARES indicating viral load assessment (including unsuccessful tests)
Viral Load Suppression	At least one viral load test with results of < 1,000 copies/ml

^*CHARES: The Centre for HIV/AIDS Research, Education and Services

We calculated proportions using the classic cascade denominator (patients enrolled in care) and with additional denominators: for example, proportions of patients on ART. We calculated equality of proportions with corresponding p-values and confidence intervals in order to assess cascade step changes over time. We also estimated the effects of improvements in laboratory quality on viral load suppression in 2015 by adjusting the proportion of viral load samples collected to viral load results returned to match the 2014 data, and to perfect laboratory performance, while maintaining the proportion of viral load suppression to viral load results returned in 2015.

Data from 2014 and 2015 were analysed separately due to several historical programmatic-level differences. These included changes to the HIV database introduced through a comprehensive data cleaning exercise focusing on patients receiving ART. In addition, national treatment guideline changes to both the CD4 threshold for ART commencement (an increase to counts of greater than 500/mm³ as compared to 350/mm³) and the increase in the number of routine viral load tests to be performed annually (from 1 to 2), were also implemented in 2013. To illustrate the impact of these changes on the analysis, a simulated projection of the 2015 population was created with standard errors to determine how dissimilar the unadjusted data were. Also, the use of aggregate data was therefore no longer a reasonable approximation and required individual-level data analysis to avoid the effects of duplication for the three viral load indicators (Viral Load Samples Collected, Results Returned, and Suppression). With individual-level data, a comparison of the last three stages of this cascade across all six years was now appropriate and provided a more accurate reflection of clinic performance.

All statistical analysis was performed using STATA version 13 SE (StataCorp 2017). Tests for differences in proportions were reported with p-values and 95% confidence intervals, while trend analysis utilized the Cochrane-Armitage test.

Exemption from ethical approval was granted by the University of the West Indies, Mona, Ethics Committee (ECP24, 18/19) due to the analysis of secondary data performed in this study.

Results

A total of 1660 patients had been cumulatively recruited to CHARES up to the end of 2015 (Table 2). Additional frequency data for the years 2010 to 2015 are shown in Figure 1. Numbers of patients at each step of the cascade increased annually from 2010 to 2012. This trend did not continue for all stages of the care cascade in 2013 when the numbers of Viral Load Samples Collected and Viral Load Results Returned decreased. In 2015, the numbers of Viral Load Results Returned and Viral Load Suppression decreased when compared to 2014.

Table 2.

Cumulative Number of Patients in Each Viral Load Cascade Stage at CHARES, 2010–2015

Indicator	2010	2011	2012	2013	2014	2015
Population Enrolled in Care	1206	1317	1404	1472	1499	1660
Patients on ART	879	988	1081	1199	966	1043
Patients Engaged in Care	858	941	1033	1051	893	967
Viral Load Samples Collected	740	809	917	828	884	954
Viral Load Results Returned	658	722	898	626	834	817
Viral Load Suppression	319	377	445	579	713	661

Figure 1:

HIV viral load cascade at the CHARES Clinic, 2010–2015*

* All proportions are calculated using Enrolled in Care

Using the classic denominator approach as described by Gardner et al⁷, in which proportions are calculated as a function of the total number of patients enrolled in care (top half of Table 3), significant increases were seen from 2010 to 2013 for the proportion of Patients on ART (8.6%, 95% CI [.049, .122]) and the proportion of patients with Viral Load Suppression (12.8%, 95% CI [.065, .191]). Significant declines were seen in the proportion of Viral Load Samples Collected (−5.1%, 95% CI [−.099, −.001]) and Viral Load Results Returned (−12.1%, 95% CI [−.164, −.056]). No significant changes were identified in the proportion of Patients Engaged in Care (p=0.885). Between 2014 and 2015, significant declines were seen in the proportions of Viral Load Results Returned and Viral Load Suppression (−6.4%, 95% CI [−0.112, −0.016] and −7.8%, 95% CI [−0.130, −0.026], respectively).

Table 3.

Changes in Viral Load Cascade Stages Comparing 2010 vs. 2013 and 2014 vs. 2015 Data at CHARES with Different Denominators^*

			Change: 2013 vs. 2010				Change: 2015 vs. 2014
Indicator	Proportion 2010	Proportion 2013	p-value	95% CI	Proportion 2014	Proportion 2015	p-value	95% CI
Patients on ART ÷ Population Enrolled in Care	0.729	0.815	<0.001	0.049,0.122	0.644	0.628	0.457	−0.026, 0.058
Patients Engaged in Care ÷ Population Enrolled in Care	0.711	0.714	0.885	−0.037, 0.044	0.596	0.583	0.569	−0.032, 0.058
Viral Load Samples Collected ÷ Population Enrolled in Care	0.614	0.563	0.041	−0.099, −0.002	0.590	0.575	0.515	−0.060, 0.030
Viral Load Results Returned ÷ Population Enrolled in Care	0.546	0.425	<0.001	−0.175,−0.067	0.556	0.492	0.009	−0.112,−0.016
Viral Load Suppression ÷ Population Enrolled in Care	0.265	0.393	<0.001	0.065,0.191	0.476	0.398	0.004	−0.130,−0.026
Patients Engaged in Care ÷ Patients on ART	0.976	0.877§	<0.001	−0.116,−0.072	0.924	0.927	0.805	−0.029,0.021
Viral Samples Collected ÷ Patients Engaged in Care	0.862	0.788	<0.001	−0.111,−0.037	0.990	0.987	0.548	−0.007,0.013
Viral Load Results Returned ÷ Viral Load Samples Collected	0.889	0.756	<0.001	−0.174,−0.092	0.921	0.856	<0.001	−0.095, −0.035
Viral Load Suppression ÷ Patients on ART	0.363	0.483	<0.001	0.053,0.187	0.738	0.633	<0.001	−0.154,−0.056
Viral Load Suppression ÷ Patients Engaged in Care	0.372	0.551	<0.001	0.112,0.246	0.826	0.684	<0.001	−0.187,−0.097

^*Values for 2014 and 2015 are based on de-duplicated data.

For the stage “Engaged in Care” there was no significant trend over the six-year period, with the lowest numbers noted in the 2014 and 2015. For the last three stages of the cascade, there was a significant nonzero trend over the six-year period: the proportion of Viral Load Samples Collected decreased as did the proportion of Viral Load Results Returned, while the proportion of enrolled patients with Viral Load Suppressed increased. The Cochrane-Armitage test also revealed a nonlinear trend (p<0.001) for these three cascade stages over the six years.

The effects of the use of aggregate data in 2014 and 2015 are highlighted in Figure 2, which shows adjusted numbers that account for patients who had two viral load tests. When aggregate data are used, the three viral load indicators are overestimated and the drop in Viral Load Suppression between 2014 and 2015 is not identified. In the aggregate data there were significant departures from the expected numbers of Patients on ART and Engaged in Care compared to previous years.

Figure 2.

Analysis of the effects of programmatic changes and duplicate data on the modified HIV continuum of care with standard error bars for 2015

The effects of improvements in laboratory quality can be seen in Figure 3. Improvements in laboratory quality to the 2014 level shows the potential of an increase in viral load suppression by 10.2% (n=728). An increase in quality to 100% shows the potential of a 16.8% (n=772) increase in viral load suppression.

Figure 3.

Analysis of the effects of laboratory quality improvement on the modified HIV continuum of care

We next calculated proportions using alternate denominators (bottom half of Table 3). For 2013 versus 2010, all results were statistically significant. Decreases were seen in the number of Patients Engaged in Care as a fraction of Patients on ART (−9.9%, 95% CI [−.116, −.072]), Viral Load Samples Collected among Patients Engaged in Care (−7.4%, 95% CI [−.111, −.037]), and Viral Load Samples Returned as a fraction of Viral Load Samples Collected (−13.3%, 95% CI [−.174, −.092]). Increases were seen in Viral Load Suppression among Patients on ART (12.0%, 95% CI [.053, .187]) and Viral Load Suppression among Patients Engaged in Care (17.9 %, 95% CI [.112, .246]). For 2015 versus 2014, statistically significant changes – all decreases – were seen in Viral Load Samples Returned as a fraction of Viral Load Samples Collected (−6.5%, 95% CI [−.095, −.035]), Viral Load Suppression among Patients on ART (−10.5%, 95% CI [−.154, −.056]) and Viral Load Suppression among Patients Engaged in Care (−14.2 %, 95% CI [−.187, −.097]).

Discussion

Significant insight into the factors affecting viral load suppression was achieved with the modification of classic care cascade models and the inclusion of process data.²³ Our analysis shows that maintaining some of the essential steps of the care continuum and including others focusing on integral processes relevant to programmatic areas normally excluded from traditional HIV care continuums can provide valuable information for improvement. Here, we removed Total Population Infected and linked to care indicators to reflect a site-level cascade. We also created a new indicator, Engaged in Care, which responds to the global shift in treatment approach to ‘test and treat’ strategies where retention assessments for the subpopulation on ART are increasingly important. Finally, we included Viral Load Samples Collected and Viral Load Results Returned as additional stages to create this ‘Viral Load Cascade.’ Implementing these changes in evaluating the HIV care continuum at CHARES will assist in achieving significant progress towards the 90-90-90 goals.

A further beneficial feature of our approach is the inclusion of programme elements outside of the traditional treatment site. In our analysis, we incorporated the central laboratory where viral load samples are processed. This feature allows for the assessment of the impact of variations in laboratory quality on viral load suppression over time, as not all viral load samples may be correctly processed. Ensuring that all viral load samples that are submitted to the laboratory are not affected by machinery breakdown, reagent stockouts and other avoidable interruptions in service delivery (which affect the number of paper-based results returned to the treatment sites) can provide a more accurate assessment of overall HIV programmatic performance. In this analysis we showed that improvements in laboratory quality could increase viral load suppression among CHARES patients by up to 17%.

Future analyses of the HIV continuum of care can be expanded by focusing on other aspects of treatment and care that affect viral suppression. For example, the inclusion of pharmacy-level data relevant to patient adherence, such as the combination of prescribing practices, self-reported adherence, drug stockouts, and dispensation from the pharmacy, may well provide useful insights into specific actionable factors affecting treatment adherence and ultimately viral load suppression. Also, further disaggregation of care cascades by population characteristics, including age and gender, can also provide important insights into population dynamics. We did not have such data for the current analysis, but this would be a useful area for future research.

Outputs from the care continuum can serve as an entry point for constructive discussion of expanded teams to improve programmatic delivery. Once the system for collecting the required data points has been established, the indicators are easily monitored and lend themselves well to continuous quality improvement activities. As shown in this paper, individual-level data is essential for accurate data interpretation and to allow for a more in-depth understanding of the dynamic process by which patients cycle between stages throughout their treatment life course. Programme evaluation can then incorporate longitudinal methodologies²⁴ with measures supporting an “all-or-none” approach. All-or-none measurements assist in promoting a systems perspective by ensuring that upstream factors are addressed prior to achieving final outcomes²⁵ as shown here with the Viral Load Suppression measure. In addition, analysis of the effectiveness of interventions is enhanced by accounting for the correlation of data between care continuum steps.²⁶

The three additional viral load indicators that we adopted (Viral Load Samples Collected, Viral Load Samples Returned, Viral Load Suppression) identified different areas of programmatic delivery and can be used to highlight appropriate opportunities for improvement. Decreases in the proportion of Viral Load Samples Collected to Patients Engaged in Care suggest failures in access to phlebotomy services whether due to cost, availability or logistics. This, in turn, may reflect a lack of adherence to national guidelines on recommended ordering practices. Decreases in proportions of Viral Load Samples Returned to Viral Load Samples Collected highlighted inefficiencies in laboratory quality including transport and the processing of samples. The logistical challenges may be of particular importance in low- and middle-income countries where samples often have to be transported to central reference laboratories for viral load analysis. Finally, assessing proportions of Viral Load Suppression to Patients Engaged in Care can highlight issues related to ART adherence by focusing on the relevant patient population, as opposed to a denominator of total enrolled population. Also, the focus on proportions of Patients on ART to Patients Engaged in Care can draw attention to the impact of post-ART loss to followup.

Progress is needed if the 90-90-90 goals are to be achieved. Fewer than two-thirds (63%) of CHARES clinic patients received ART in 2015. Though this is higher than the estimated national average of only 43% of people living with HIV who know their infection status and receive treatment²⁷, it is well short of the 90% goal. Similarly, viral load suppression among treated patients at CHARES was 63% in 2015, indicating that further progress is needed.

Site-level indicators from the expanded HIV care continuum can highlight opportunities for improvement. A next step is to assess the likely effect of potential programs aimed at improving the case process and to determine which set of programs to invest in, for example by using methods from process improvement analysis.²⁸

Our study has several limitations, primarily related to the quality of the data and the assumptions made to facilitate comparison. First, there has been no formal assessment of the quality of the data generated from the database. Second, for the first three steps of the care continuum, only aggregate data was accessible for the years 2010 to 2013 and therefore the deduplication of this data was not possible. Neither population characteristics nor process-level data for the years 2016–2017 were available for this analysis. Although the policy change in viral load testing guidelines occurred at the beginning of 2013, we assumed that the uptake at the site level was slow and therefore duplication in viral load suppression did not affect data for that year. Finally, we assumed a constant proportion of viral suppression when analysing the effects of improved laboratory quality. Despite these limitations, we believe that our inclusion of the analyses of data projections and the effects of duplication provides an accurate reflection of the functioning of the clinic over the six-year period.

We conclude that site-level data with an increased focus on desired programme elements included in the HIV care continuum can provide useful insights into the factors affecting patient outcomes and, most importantly, furnish the data essential for tailoring and monitoring interventions that can improve the HIV care delivery process. Our approach can be applied to the identification of gaps in the care process at other treatment facilities²⁹ and for other disease types.^{22, 30} This study specifically recommends the expansion of monitoring frameworks to include all of the essential processes that affect final outcome indicators. Such improvements are critical to the identification of deficiencies in treatment and care and the overall enhancement of interventions to control the epidemic.