Cost-effectiveness of cryptococcal antigen screening at CD4 counts of 101–200 cells/µL in Botswana

Background: Cryptococcal antigen (CrAg) screening in individuals with advanced HIV reduces cryptococcal meningitis (CM) cases and deaths. The World Health Organization recently recommended increasing screening thresholds from CD4 ≤100 cells/µL to ≤200 cells/µL. CrAg screening at CD4 ≤100 cells/µL is cost-effective; however, the cost-effectiveness of screening patients with CD4 101–200 cells/µL requires evaluation. Methods: Using a decision analytic model with Botswana-specific cost and clinical estimates, we evaluated CrAg screening and treatment among individuals with CD4 counts of 101–200 cells/µL. We estimated the number of CM cases and deaths nationally and treatment costs without screening. For screening we modeled the number of CrAg tests performed, number of CrAg-positive patients identified, proportion started on pre-emptive fluconazole, CM cases and deaths. Screening and treatment costs were estimated and cost per death averted or disability-adjusted life year (DALY) saved compared with no screening. Results: Without screening, we estimated 142 CM cases and 85 deaths annually among individuals with CD4 101–200 cells/µL, with treatment costs of $368,982. With CrAg screening, an estimated 33,036 CrAg tests are performed, and 48 deaths avoided (1,017 DALYs saved). While CrAg screening costs an additional $155,601, overall treatment costs fall by $39,600 (preemptive and hospital-based CM treatment), yielding a net increase of $116,001. Compared to no screening, high coverage of CrAg screening and pre-emptive treatment for CrAg-positive individuals in this population avoids one death for $2440 and $114 per DALY saved. In sensitivity analyses assuming a higher proportion of antiretroviral therapy (ART)-naïve patients (75% versus 15%), cost per death averted was $1472; $69 per DALY saved. Conclusions: CrAg screening for individuals with CD4 101–200 cells/µL was estimated to have a modest impact, involve additional costs, and be less cost-effective than screening populations with CD4 counts ≤100 cells/µL. Additional CrAg screening costs must be considered against other health system priorities.


Introduction
Botswana had an estimated adult HIV prevalence of over 20% in 2018, with approximately 350,000 adults living with HIV 1 . This includes a sizable population with advanced HIV disease (CD4 ≤200 cells/µL) who are at an increased risk of opportunistic infections such as cryptococcal meningitis (CM) 2 . Reflex cryptococcal antigen (CrAg) screening with targeted fluconazole treatment for the prevention of CM was adopted in national HIV guidelines in 2016 at a CD4 count threshold of ≤100 cells/µL 3 . We previously found screening at this threshold to be highly cost-effective (either cost-neutral or cost-saving across different model assumptions) and likely to prevent a significant proportion of CM cases and deaths 4 .
In 2018, the World Health Organization (WHO) conditionally recommended increasing the CD4 count threshold for CrAg screening from ≤100 cells/µL to ≤200 cells/µL for the prevention of CM 5 . Patients with CD4 counts of 101-200 cells/µL are also relatively immunocompromised and at risk for CM 6 , but prevalence of CrAg positivity in this population, estimated at 2.0% (95% confidence interval (CI): 1.2-2.7%; 21 studies) 7 is substantially lower than prevalence among patients with CD4 ≤100 cells/µL. The impact and cost-effectiveness of increasing the CrAg screening CD4 count threshold have not been systematically evaluated, and a better understanding of the potential impact (in terms of CM cases and deaths avoided), screening program resource needs, and cost effectiveness will inform countries as they consider changes to national screening guidelines.
Using data and estimates from Botswana in patients with a CD4 count of 101-200 cells/µL, the objective of this analysis is to expand our CrAg screening models to include those with a CD4 count of 101-200 cells/µL, with an aim of informing policy regarding CrAg screening for patients with higher CD4 counts. As in our previous analysis 4 , we evaluated CrAg screening among patients who are antiretroviral therapy (ART)-naïve (those targeted for pre-emptive treatment in guidelines) as well as ART-experienced patients found to be CrAg-positive through reflex CrAg screening. This ART-experienced population re-engaging in care and treatment now makes up about half of those with incident CM [8][9][10] in the region and are likely to derive clinical benefit from pre-emptive fluconazole treatment for the prevention of CM.

Overview
We used a decision analytic model to evaluate the number of patients receiving CD4 testing in Botswana who are at risk of cryptococcal meningitis and (1) develop CM without CrAg screening and (2) with national reflex CrAg screening adoption, as previously described 4 , but in this analysis focused on those with a CD4 count of 101-200 cells/µL. The model estimates number of CM cases, CM-related deaths and disability-adjusted life years (DALYs) lost, and associated costs of CM management in the absence of screening ( Figure 1A). This is compared with the estimated number of CM cases, CM-related deaths and DALYs lost, and associated costs of screening and pre-emptive therapy as well as costs of CM management for incident cases occurring despite implementation of screening ( Figure 1B).
For these models, CD4 count distribution data were obtained from the Botswana-Harvard HIV Reference Laboratory 11 , and local CrAg prevalence and titre data used to predict risk for progression to CM in the CD4 101-200 cells/µL population. Local data were obtained from a 2018-2019 CrAg screening cohort of patients with advanced HIV disease in Gaborone, which included over 900 patients with a CD4 count of 101-200 cells/µL who received reflex CrAg screening and were followed for up to 6 months for incident CM and mortality 12 . In our model, based on local estimates we assume that 650,000 CD4 tests are performed annually for the adult HIV-positive population of 350,000 (around two tests per patient) 11 .

Screening module
The screening module (see Figure S1 in extended data 13 ), adapted from our previous model 4 , estimates the proportion of patients who receive CD4 testing with a CD4 101-200 cells/µL, how many of these patients receive reflex CrAg screening, the proportion who are CrAg-positive, and the proportion previously initiated on ART, i.e. "ART-experienced" (see Figure S1 in extended data and key parameter assumptions in Table 1). From country data 11 , 5.35% of all CD4 tests performed in greater Gaborone have a CD4 T-cell count between 101 and 200 cells/µL (Table 1). Only a small proportion (15%) of patients with a CD4 101-200 cells/µL were ART-naïve in 2018-2019. Patients were considered ART-experienced if they had a prior history of HIV viral load testing documented in the national electronic medical record, as viral load testing is exclusively performed after initiation of ART as per national guidelines 3 . In the absence of prior documented viral load testing, a patient was assumed to be ART-naïve.
Based on data from the prospective 2018-2019 CrAg screening cohort 12 , among screened outpatients in the 101-200 cells/µL

Amendments from Version 1
The following is a summary of changes made from the previous version of the manuscript:

Introduction:
The introduction was reorganized to first describe advanced HIV and cryptococcal antigen (CrAg) screening guidelines in Botswana then discuss updated World Health Organization recommendations for screening at a higher CD4 T-cell count threshold.

Methods:
In the first paragraph, a general overview of the CrAg screening model was added along with a new figure.
Methods/Results: A sensitivity analysis (SA4) was added considering differing costs of cryptococcal disease treatment in sub-Saharan Africa using published estimates; updated Excel models have been added to the online Extended Data. CD4 T-cell count range, CrAg prevalence was estimated at 3.1%, 35% of whom had a history of treated CM; thus 2.0% of screened outpatients with a CD4 count of 101-200 cells/µL are estimated to be incident CrAg positives (no history of prior CM) and the target population for pre-emptive fluconazole treatment.
We used serum CrAg titre data to stratify the risk of CrAgpositive patients progressing to CM 14 , with a titre >1:160 corresponding with a high risk for incident cryptococcal disease. Approximately 20% of CrAg-positive outpatients with a CD4 101-200 cells/µL had a high CrAg titre, compared to 59% among those with lower CD4 counts of ≤100 cells/µL 4 . For our CD4 101-200 cells/µL models, we assume that patients who screen CrAg-positive and return to clinic are started on preemptive fluconazole therapy and none receive a diagnostic lumbar puncture to evaluate for central nervous system infection, given the lower distribution of CrAg titres in the CD4 101-200 cells/µL population compared to ≤100 cells/µL and frequent lumbar puncture refusal in routine-care settings 15 .
Our base model assumes that 5% of patients with CD4 101-200 cells/µL do not receive CrAg screening due to laboratory error or assay stockout and that 10% of patients who screen CrAg-positive do not return to clinic to begin pre-emptive fluconazole, putting them at higher risk for progression to CM.
Base model: CrAg screening at CD4 101-200 cells/µL, treatment for both ART-naïve and ART-experienced The base model treatment module (see Figure S2 in extended data 13 and key parameter assumptions in Table 1) includes outcomes for patients (1) with a CD4 count of 101-200 cells/µL who do not receive CrAg screening, (2) who are screened and CrAg-positive but do not receive follow-up to initiate pre-emptive therapy, and (3) who are screened and started on pre-emptive fluconazole therapy.
Full modeling assumptions are detailed in a Microsoft Excel file accessible online 13 . Risk of progression to CM is dictated by whether a patient has a high-(>1:160) or low (≤1:160) CrAg titre 14 . For patients who either don't receive CrAg screening or receive screening but do not subsequently initiate fluconazole therapy, given the comparatively lower CrAg titre distribution in patients with higher CD4 counts of 101-200 cells/µL, we expect a longer delay until progression to CM in the absence of pre-emptive fluconazole compared to CrAg-positive patients with CD4 cell counts ≤100 cells/µL (Figure S1 in extended data 13 ). However, initiation of fluconazole therapy in CrAg-positive patients further reduces the risk of progression to CM in the population with CD4 101-200 cells/µL.
Outcomes of patients who develop incident CM are informed by local mortality data from Botswana, with approximately 50% of patients dying within 10 weeks of CM diagnosis under routine care conditions 9,17 . Patients who are recognized as CrAg-positive and started on pre-emptive fluconazole but subsequent fail therapy and are admitted to the hospital for the management of CM are assumed to have better clinical outcomes (25% versus 50% 10-week mortality) based on limited data from South Africa 14 . Some patients who develop CM and survive hospitalization may develop relapsed CM. Given the small proportion of these patients and small clinical and public health impact, we do not consider them further in our models.
With reflex CrAg screening, patients receive CrAg screening based on CD4 count regardless of prior ART status. However, most (85%) patients with a CD4 count of 101-200 cells/µL are now ART-experienced according to recent cohort data from Botswana 2018-2019 12 . Very little outcome data exist in this disparate sub-population, which consists of patients: (1) recently started on ART; (2) ART-experienced who defaulted and are now re-establishing care; and (3) ARTexperienced but with treatment failure. From local 2018-2019 cohort data in Botswana, approximately 75% of these ARTexperienced patients are considered to have recently started on ART (with an undetectable HIV viral load in the previous three months), 20% are on ART but with a recent unsuppressed HIV viral load signifying treatment failure, and 5% have a history of recent ART use without a recent HIV viral load signifying likely ART default 18 . For those recently started on ART, we assumed a 33% reduction in risk of CM for those with CD4 101-200 cells/µL compared to our previous estimates for those with CD4 ≤100 cells/µL. In our base model, based on prospective cohort data 15 , those recently started on ART with a suppressed HIV viral load have a low risk of progression to CM without pre-emptive fluconazole therapy (7%), with a greater risk in those with ART treatment failure (60%) and ART defaulters (33%). The combined risk of progression to CM for all ART-experienced patients in the CD4 101-200 cells/µL group is assumed to be 19% without pre-emptive treatment.
We estimate an 87.5% reduction in risk of incident CM with pre-emptive fluconazole (factoring in a relatively low baseline CrAg titre distribution in this group) 19 .

CrAg screening and treatment unit costs
Costing data for CrAg screening, pre-emptive therapy, and CM treatment costs are derived using local costing data when available (

Sensitivity analyses
Three main sensitivity analyses are reported to account for key areas of parameter uncertainty. The complete Excel-based model is provided as underlying data 13 so that alternative sensitivity analyses can be completed by interested readers.
Sensitivity analysis 1 (SA1): In this analysis, we assume that in some real world settings a lower proportion of CrAg-positive patients are started on pre-emptive fluconazole after laboratory testing (50% versus 90% in the base model) because of programmatic barriers such as inadequate communication of test results to clinics, a lack of fluconazole availability in clinics, lack of provider awareness of treatment guidelines, or for other reasons. This analysis still assumes that 90% of patients attended in outpatient clinics and receiving CD4 testing will stay engaged in health care. Other parameters remain the same as the base model.
Sensitivity analysis 2 (SA2): In this model, we assume less benefit of pre-emptive fluconazole in CrAg-positive patients, with a 75% rather than 87.5% reduction in incident CM. This is to account for significant uncertainty in the benefits of pre-emptive fluconazole in this population with a higher CD4 count, and for possible sub-optimal adherence to therapy.
Other parameters remain the same as the base model.
Sensitivity analysis 3 (SA3): In this model, we test our parameters with a higher proportion of ART-naïve patients receiving CD4 testing and CrAg screening (75% versus 15% in the base model). Other parameters remain the same as the base model. This is to provide estimates applicable to settings with less mature ART programmes where a higher proportion of individuals with CD4 counts of 101-200 cells/µL are likely to be ART-naïve.
Sensitivity analysis 4 (SA4): In this model, we expand model 3 which assumes a greater ART-naïve population than observed in Botswana. We also consider lower costs of CM care in other settings. Based on a costing analysis from a cryptococcal meningitis clinical trial that enrolled patients from four countries in sub-Saharan Africa 24 , we use a reduced cost of $2125.00 for two weeks of hospitalization with amphotericin B and fluconazole therapy for incident CM. This model also includes a lower cost of fluconazole therapy used for either CM treatment or targeted preventive treatment for CrAg-positive patients. As in the base model, here we assume a lower cost of care in patients who die during hospitalization (75%). Other costs remain unchanged compared to the base model.

Results
Cryptococcal meningitis cases and costs without screening Without CrAg screening (   (Table 5). Compared to no screening, high coverage of CrAg screening and pre-emptive treatment for CrAg-positive individuals in this population is associated with a cost of $2440 per one death averted or $114 per DALY saved (Table 5).
Sensitivity analyses SA1 and SA2 assume a lower proportion of CrAg positive are started on pre-emptive fluconazole and a reduced benefit of pre-emptive fluconazole therapy for CrAg-positive patients with a CD4 101-200 cells/µL, respectively, which may be more realistic under many routine care conditions. Both models will therefore result in a smaller public health benefit to CrAg screening and a higher incremental cost per death or DALY saved. For SA1, an estimated 25% (21/85) of deaths are averted with treatment of both ART-naïve and ART-experienced with a cost per death averted of $7476 or $349 per DALY saved ( Figure 2). For SA2, 52% (44/85) of deaths are averted with treatment of ART-naïve and experienced at a cost per death No screening 85 368,982 n/a n/a n/a n/a n/a  Table 4. Outcomes with CrAg screening and pre-emptive fluconazole for ART-naïve and ART-experienced (base model).

Cost for patients (USD)
Identified for preemptive treatment (but did not receive), but did not develop CM - survives 6 Overall estimated costs, number of CM cases, number of deaths averted, and DALYs saved for the base model and sensitivity analyses are summarized in Figure 3.

Discussion
We  12 . We included a sensitivity analysis assuming that a majority (75%) of patients who received CD4 testing and CrAg screening were  ART-naïve, which may inform other health systems with a higher proportion of ART-naïve patients receiving CrAg screening with ART initiation. This sensitivity analysis showed a slightly better impact and cost-effectiveness compared to the base model assuming most patients were ART-experienced although screening was still not cost-neutral or cost-saving. This project contains the following extended data: -Figures S1 (flowchart of screening module) and S2 (flowchart of treatment module) Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).

Beyene T, Zewde AG, Balcha A, et al.: High Dose Fluconazole Monotherapy is
The authors use a previously published adapted model of which all the data is online available. The performed sensitivity analyses for several uncertainties in their data. It is an interesting topic as the WHO changed the CD4 threshold for cryptococcal screening.
I have some minor comments/considerations: It is unclear from the methods how much simulations are run in the model ○ Table 2 says "hotel costs" I think this should be hospital ○ What is the willingness to pay threshold in Botswana. It is hard to make the conclusion whether screening people with a CD4 cellcount between 101-200 is cost-effective or even cost-saving if someone is not familiar with the numbers. Maybe also add the total HIV/AIDS care and treatment program budget just to place the amount in perspective ○ Limitation 1 of the study can be addressed by changing this in a sensitivity analysis. That would make the work more useful for other countries too ○

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Partly © 2021 Nalintya E. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Elizabeth Nalintya
Infectious Diseases Institute, College of Health Sciences, Makerere University, Kampala, Uganda The study title is short and smart and gives a glimpse into what the paper is about. The study is well introduced however I would reorganize the paragraphs to get a better flow, that is: The first sentence of paragraph two can be moved to the beginning of the introduction to draw the picture on how big the HIV problem is in Botswana upfront and then talk about the rationale after. I would merge and reorganize the first two paragraphs.
In the methods section, a great deal of work has gone into explaining how the different estimates were arrived at and what type of data was used to arrive at the estimates. This however blurs the description of the actual modeling. It would be beneficial to the reader to get a quick snap short of the final model (could be in a summarized figure placed within the text explaining the methods) This is especially because the figure S1 is very busy and can get confusing, the reader needs to understand what the final model is before trying to understand the smaller details of how the model was arrived at.
The screening model clearly talks about the estimates included and how they are arrived at. The viral load(VL) test is used as the proxy for ART experience, it would be good to know if all these VL tests are done after 6 months of ART or at what time point this is done. Pragmatically with all logistical challenges in resource limited settings, a recommendation to do a six month viral load will mean the viral load was done at about month eight or nine after ART start. Wondering if this choice of defining ART experience could have lumped many ART experience persons as non experienced.
Page 3, the last paragraph talks about the assumptions for those who did not receive CrAG screening, however its not clear where these estimates are derived from. Is this lab data or data from the prospective cohort. Also it seems to belong under the next subheading and yet has been placed under the screening model.
The base model, the cost analysis and the sensitivity analysis are presented well and in detail and are supplemented by the tables giving a clear picture of what was being done. The discussion of study findings is comprehensive and puts them in context and the conclusions have been derived systematically from the data presented.

Is the study design appropriate and is the work technically sound? Yes
Are sufficient details of methods and analysis provided to allow replication by others?

Yes
If applicable, is the statistical analysis and its interpretation appropriate? I cannot comment. A qualified statistician is required.
description of the actual modeling. It would be beneficial to the reader to get a quick snap short of the final model (could be in a summarized figure placed within the text explaining the methods) This is especially because the figure S1 is very busy and can get confusing, the reader needs to understand what the final model is before trying to understand the smaller details of how the model was arrived at.
Author response: We thank the reviewer for this suggestion. We have added a summary Figure 1 to broadly describe the model as an orientation and added some overview discussion at the beginning of methods. We agree that Figure S1 is very busy, and thus was moved to the supplementary materials. The base model, the cost analysis and the sensitivity analysis are presented well and in detail and are supplemented by the tables giving a clear picture of what was being done. The discussion of study findings is comprehensive and puts them in context and the conclusions have been derived systematically from the data presented.
Author response: Thank you again for this positive and constructive feedback which we have incorporated to improve the manuscript.

Reviewer #2
This research describes a decision based model to analyse the cost-effectiveness of cryptococcal antigen screening among people living with HIV with an CD4 cell-count between 101-200 cells.
The authors use a previously published adapted model of which all the data is online available. The performed sensitivity analyses for several uncertainties in their data. It is an interesting topic as the WHO changed the CD4 threshold for cryptococcal screening.
I have some minor comments/considerations: It is unclear from the methods how much simulations are run in the model  What is the willingness to pay threshold in Botswana. It is hard to make the conclusion whether screening people with a CD4 cellcount between 101-200 is cost-effective or even cost-saving if someone is not familiar with the numbers. Maybe also add the total HIV/AIDS care and treatment program budget just to place the amount in perspective In this analysis, the lower cost of fluconazole treatment for CrAg-positive patients is offset by the lower cost of hospitalization in patients hospitalized with incident cryptococcal meningitis. The impact of screening and costs per death and DALY averted therefore remain similar to sensitivity analysis 3. We included further mention of these findings in the results and discussion.

Competing Interests:
The authors have no competing interests to disclose.