Potential Cost-effectiveness of HIV Viral Load Sample Collection and Testing Methods in Malawi

Background HIV viral load (VL) monitoring informs antiretroviral therapy failure and helps to guide regimen changes. Typically, VL monitoring is performed using dried blood spot (DBS) samples transported and tested in a centralized laboratory. Novel sample collection technologies based on dried plasma stored on a plasma separation card (PSC) have become available. The cost-effectiveness of these different testing approaches to monitor VL is uncertain, especially in resource-limited settings. The objective of this study is to evaluate the potential cost-effectiveness of HIV VL testing approaches with PSC samples compared to DBS samples in Malawi. Methods We developed a decision-tree model to evaluate the cost-effectiveness of two different sample collection and testing methods— DBS and PSC samples transported and tested at central laboratories. The analysis used data from the published literature and was performed from the Malawi Ministry of Health perspective. We estimated costs of sample collection, transportation, and testing. The primary clinical outcome was test accuracy (proportion of patients correctly classified with or without treatment failure). Sensitivity analysis was performed to assess the robustness of results. Results The estimated test accuracy for a DBS testing approach was 87.5% compared to 97.4% for an approach with PSC. The estimated total cost per patient of a DBS testing approach was $19.39 compared to $17.73 for a PSC approach. Based on this, a PSC-based testing approach “dominates” a DBS-based testing approach (i.e., lower cost and higher accuracy). Conclusion The base-case analysis shows that a testing approach using PSC sample is less costly and more accurate (correctly classifies more patients with or without treatment failure) than with a DBS approach. Our study suggests that a PSC testing approach is likely an optimal strategy for routine HIV VL monitoring in Malawi. However, given the limited data regarding sample viability, additional real-world data are needed to validate the results.


Abstract
Background HIV viral load (VL) monitoring informs antiretroviral therapy failure and helps to guide regimen changes. Typically, VL monitoring is performed using dried blood spot (DBS) samples transported and tested in a centralized laboratory. Novel sample collection technologies based on dried plasma stored on a plasma separation card (PSC) have become available. The cost-effectiveness of these different testing approaches to monitor VL is uncertain, especially in resource-limited settings. The objective of this study is to evaluate the potential cost-effectiveness of HIV VL testing approaches with PSC samples compared to DBS samples in Malawi. Methods We developed a decision-tree model to evaluate the cost-effectiveness of two different sample collection and testing methods-DBS and PSC samples transported and tested at central laboratories. The analysis used data from the published literature and was performed from the Malawi Ministry of Health perspective. We estimated costs of sample collection, transportation, and testing. The primary clinical outcome was test accuracy (proportion of patients correctly classified with or without treatment failure). Sensitivity analysis was performed to assess the robustness of results. Results The estimated test accuracy for a DBS testing approach was 87.5% compared to 97.4% for an approach with PSC. The estimated total cost per patient of a DBS testing approach was $19.39 compared to $17.73 for a PSC approach. Based on this, a PSC-based testing approach "dominates" a DBS-based testing approach (i.e., lower cost and higher accuracy). Conclusion The base-case analysis shows that a testing approach using PSC sample is less costly and more accurate (correctly classifies more patients with or without treatment failure) than with a DBS approach. Our study suggests that a PSC testing approach is likely an optimal strategy for routine HIV VL monitoring in Malawi.
However, given the limited data regarding sample viability, additional real-world data are needed to validate the results. HIV VL monitoring generally involves the collection of a blood specimen, which is then stored and transported to a centralized laboratory for testing. In remote and/or or resource-limited settings, access to VL testing is challenging due to limited healthcare infrastructure and sample transport constraints associated with whole blood or plasma.
Dried blood spot (DBS) specimens are commonly collected for HIV VL testing. Compared to whole blood or plasma, DBS samples are easier to collect from capillary finger pricks and remain more stable over time and temperature variations. However, DBS testing yields lower testing accuracy and is associated with greater probability of both upward and downward VL misclassification compared to plasma testing because of the smaller sample volume and excessive cellular genetic material contained in whole blood [3,4]. The WHO recommends use of plasma specimens for determining virological failure at a threshold of 1,000 copies/mL [5]. To overcome the test performance limitations of DBS, novel sample 4 collection technologies based on dried plasma spots have become available. However, the cost-effectiveness of these different testing approaches for routine VL monitoring is uncertain, especially in resource-limited settings. The objective of this study is to evaluate the potential cost-effectiveness of HIV VL testing approaches using dried plasma spot samples stored on a plasma separation card (PSC) compared to DBS samples in Malawi.

Analytic Overview
A cost-effectiveness model (decision tree structure) (Figure 1) was developed using Microsoft ® Excel to project the costs and test accuracy of HIV VL testing approaches using PSC compared to DBS samples. The population assumed in the base-case analysis was HIV-positive patients on ART, undergoing treatment and follow-up at outlying clinics in Malawi. The reference case (the average analytic entity) was a single sample of blood collected from one of these patients for routine VL testing. The analysis was conducted from the healthcare payer perspective, assumed to be the Malawi Ministry of Health. The time horizon of analysis was from sample collection to completion of sample testing, assumed to be a maximum of thirty days. This time horizon was appropriate to capture the costs and diagnostic outcomes of using different testing approaches for routine VL testing.

Comparators
The comparators evaluated in the analysis included: (1) capillary blood collected and transported on a DBS card to a central laboratory and tested using the Abbott m2000 RealTime System (Abbott Laboratories, Abbott Park, Illinois) ("DBS"); (2) capillary blood collected and transported on the cobas ® Plasma Separation Card to a central laboratory and tested using the cobas ® 6800 System (Roche Molecular Systems, Pleasanton, California) ("PSC").

5
The cobas ® Plasma Separation Card is a novel sample collection device that separates the plasma component from a whole blood spot (minimum 140µL volume) resulting in a dried plasma spot. The cobas ® Plasma Separation Card has an integrated stabilizer that maintains sample integrity between 18-45⁰C and 85% humidity for up to 21 days [6].

Decision-Tree Model
In the decision-tree model, a reference case sample could be viable or non-viable (sample optimality) after transport and upon arrival at a central laboratory. Upon testing, the reference case blood sample may have a viral load of 1,000 copies per ml or more (positive VL test) or less than 1,000 copies per ml (negative VL test). The sensitivity and specificity of the testing platforms were used to estimate the true-positive, false-negative, true-negative, and false-positive rates. We made the following simplifying assumptions: (1) non-viable samples are discarded and are by definition inaccurate, (2) sub-optimal sample collection, storage, and transportation exert an effect on VL result accuracy via sensitivity and specificity, and (3) no human or other sources of error occurred in the sample preparation and testing process at the central laboratory. The primary outcome in the analysis was test accuracy, defined as the proportion of patients correctly classified with or without treatment failure. This outcome evaluates the combined effectiveness of the sample collection approach and testing platform.

Clinical Inputs
The estimated prevalence of ART failure in Malawi was obtained from the published literature. The test performance estimates used in the base-case analysis were derived from a head-to-head evaluation of HIV VL testing with the cobas ® Plasma Separation Card on the cobas ® 8800 System compared with DBS on the Abbott m2000 RealTime System using the WHO ART failure threshold of 1,000 copies/mL [7]. The cobas ® 6800 and 8800 6 Systems have demonstrated equivalent test performance differing only in throughput [8].
We were not able to identify a data source to estimate the impact of transportation on sample viability upon arrival at a central laboratory (i.e., the proportion of samples optimal vs. not optimal for testing). Due to this data limitation, we assumed that all samples were viable in the base-case analysis and evaluated the impact of this parameter on results in sensitivity analysis. The clinical estimates used in the base-case analysis are described in Table 1.  Table 1.

Sensitivity Analysis
Univariate sensitivity analysis was conducted to assess which parameters had the greatest influence on the results. All parameters were varied across plausible ranges, either 95% confidence intervals when available or +/-20% when unavailable. We conducted probabilistic sensitivity analyses (PSA) by creating probability distributions for all parameters using beta distributions for probabilities and log-normal distributions for costs. We used base-case estimates as the mean, and standard errors were estimated assuming that ranges represented 95% confidence intervals with the range equal to four times the standard error. The PSA was conducted using 1,000 runs of Monte Carlo simulation. The parameter ranges assessed in sensitivity analyses are presented in Table   1.

Base-case Analysis
The base-case results are shown in Table 2. The estimated test accuracy for a DBS testing approach and PSC testing approach were 87.5% and 97.4%, respectively. The cost per patient sample of a DBS testing approach was $19.39, while the cost per patient sample of a PSC testing approach was $17.73. Based on this, a PSC testing approach "dominates" a DBS testing approach as it is associated with lower costs and higher test accuracy.

Sensitivity analysis
Given that a PSC testing approach dominates a DBS testing approach, we performed univariate sensitivity analysis on incremental accuracy and incremental costs separately.
The univariate sensitivity analysis comparing the PSC testing approach with the DBS testing approach showed that the incremental accuracy was most affected by the proportion of samples optimal for testing (sample viability) and the test performance (sensitivity and specificity) of the test platforms (Figure 2a). The incremental cost was most affected by the cost of test consumables in the laboratory for both platforms ( Figure   2b).

8
PSA results are presented in Figure 3 as a cost-effectiveness scatter plot which shows the parameter uncertainty surrounding the incremental accuracy and incremental cost projected in the base-case analysis. As can be seen, all 1,000 simulations suggest that a PSC testing approach would have higher accuracy with lower cost relative to a DBS testing approach, and would be considered the economically dominant strategy across the parameter ranges described in Table 1.

Discussion And Conclusion
We developed a decision-analytic model to evaluate the costs and test accuracy of two different HIV VL sample collection and testing approaches in Malawi, a low-income country with high HIV prevalence. From a Malawi Ministry of Health perspective, our results suggest that the current practice of DBS sample transported to a central laboratory and tested on the Abbott m2000 RealTime System is not optimal and is projected to lead to approximately 10% higher misclassification rate than a testing approach using dried plasma spots collected on the cobas ® Plasma Separation Card, transported to a central laboratory and tested on the cobas ® 6800/8800 System. Additionally, the DBS testing approach was estimated to cost $1.66 more per patient sample compared to the PSC testing approach. Although our study did not attempt to model the downstream impact of VL misclassification on costs and health outcomes, it is expected that false-positive results could amplify the cost impact of inaccurate testing through unnecessary switching of virally suppressed patients to more expensive second-line therapy.
For this analysis, we faced real-world data limitations for the key parameter of sample viability. We could not find adequate information about the proportion of HIV VL samples adversely affected by transportation conditions. As a result, our best analytic option was to assume all samples are viable upon arrival to a central laboratory (100% sample viability). In sensitivity analysis, the sample viability parameter was shown to be impactful on results. Therefore, our base-case estimate of DBS test accuracy may be an overestimate if, in real-world use, the DBS card has less protective effect on sample integrity over varying environmental conditions than previously assumed.
The Malawi Ministry of Health is committed to scaling up routine HIV VL testing services for patients on ART [3]. The question at hand is how best to scale-up these services to efficiently meet the increasing short-, medium-, and long-term demand for routine VL monitoring. Our analysis suggests that a testing approach utilizing the cobas ® Plasma

Consent for publication: Not applicable
Availability of data and material: Data sharing is not applicable as no datasets were generated or analyzed in the study. Model inputs were derived from publicly available sources and were referenced accordingly.  Figure 1 Decision tree comparing HIV viral load testing approaches in Malawi Cost-effectiveness Scatter Plot