Modeling the patient and health system impacts of alternative xpert® MTB/RIF algorithms for the diagnosis of pulmonary tuberculosis in Addis Ababa, Ethiopia

The study was conducted for public health facilities which provide the Directly Observed Treatment (DOT) services to TB patients in Addis Ababa , Ethiopia. We compared the impact of eight diagnostic algorithms. These were chosen based on WHO [15] and Ethiopian ministry of Health [6] recommendations. Routine diagnostic algorithms, that are common in the country, were also considered. Under the ZN-Spot-Morning-Spot scenario, Ziehl-Neelsen (ZN) microscopy using three sputum samples provided within two days was modeled. This scenario represents the common diagnostic algorithms for presumptive TB patients when there is no access to Xpert MTB/RIF (expert opinion). In this model, it was considered as the base case for comparison. FN-Spot-Morning-Spot scenario replaces ZN microscopy with light-emitting diode (LED) fluorescence microscopy. FN-Spot-Morning and FN-Spot-Spot scenarios use LED Fluorescence microscopy using two sputum samples provided within two days (on the SPOT-early MORNING) and on the same day (on the SPOT-SPOT), respectively. Full-Xpert scenario tests all patients with presumptive tuberculosis with Xpert. Targeted-Xpert- MDR-HIV-ZN-Spot-Morning-Spot scenario uses ZN microscopy (SPOT-MORNING-SPOT) as the primary test for tuberculosis diagnosis and target Xpert for presumptive TB cases that are HIV positive, known contact of MDR-TB patients, and retreatment cases. But Targeted-Xpert-ZN-Negative-Spot-Morning-Spot scenario targets the use of Xpert to a patient with presumptive TB who are smear negative in addition to a patient who is HIV positive, known contact of MDR-TB patients, and retreatment case. Targeted-Xpert-MDR-HIV-FN-Spot-Spot scenario uses LED fluorescence microscopy on the same day (SPOT-SPOT) as the primary test for tuberculosis diagnosis and target Xpert for presumptive TB cases who are HIV positive, known contact of MDR-TB patients, and retreatment cases. Figure 1 Illustrate the diagnostic algorithms used in the analysis.

We adopted a modeling framework pioneered by Langley et al. in Tanzania [5] to represent the operational and epidemiological context of Addis Ababa, Ethiopia. The model used the discrete-event simulation (DES) approach and Witness software to represent the patient pathways that individual patients follow and the sputum sample pathways in the laboratory.

The patient pathways, the patient was categorized as either new cases or retreatment cases, and HIV+ or HIV- with some knowing their status and some not. Multiple visits to the diagnostic center are required by each new patient. The number of visits for each patient is dependent on a number of factors which were modeled. These factors included the diagnostic test; whether the patient became diagnostic lost to follow-up (LTFU); the backlog in the laboratory that could lead to diagnosis not being available; the time the patient was willing to wait in the diagnostic centre for results; whether the initial test result was positive; whether X-ray and a short course of antibiotics was required following a negative sputum test; and whether TB treatment would be required. Once TB has been diagnosed a further visit to the diagnostic center is required to receive the first batch of medication and for treatment education. Patients return to the diagnostic center during treatment for sputum smear microscopy testing at the end of each phase of treatment. Patients are finally categorized as cured, completed, failed treatment, or treatment lost to follow-up.

In sputum sample pathways, sputum samples are collected in the diagnostic center from presumptive TB cases and individuals undergoing treatment for TB. Samples follow a pathway through the laboratory depending on the patient characteristics, diagnostic technique, and the diagnostic algorithm being used. There were two different types of diagnostic test that were modeled – microscopy and Xpert. These were modeled separately with control logic defining which test process would be used for which patient sample based on the ‘attributes’ of the patient. This logic enabled a number of different diagnostic algorithms to be evaluated. Samples are prepared for the appropriate test, the test is then conducted and a test result assigned.

For simplicity, we assumed a constant population with no immigration or emigration and constant transmission dynamics. Further details on the model and parameter descriptions are given in the Additional file 1 and other references [5, 11]. The model was calibrated by using real data from diagnostic centers in Addis Ababa. The outputs from the model were validated against the observed results from diagnostic centers.

Considering the natural history and pathogenesis of tuberculosis that take generations for the effects of intervention to be seen and situational changes due to new diagnostics technologist in pipeline, we choose a timeframe of 10 years for prediction.

A survey of one central referral laboratory and 20 public health facilities that provide TB services provided (randomly selected) estimates for the diagnostic and cost parameters to be used in the model (Additional file 1: Table B). Sensitivity and specificity of diagnostic tools were collated from the literature [16‐18] (Additional file 1: Table A). Data were collected from August to December 2014. Additional file 1: part A and B lists the model variables and their sources.

Cost variables and source of the data are presented in Additional file 1: part C. Unit costs were input in US dollars and are based on published data and estimates from Ethiopian National Tuberculosis program, Addis Ababa Health Bureau, and Health facilities. We used public sector salary scale of 2014 to estimate annual employment cost using National Bank of Ethiopia exchange rate (19.7 Birr= 1 USD). Consolidated Health Economic Evaluation Reporting Standards (CHEERS) [19] was used for reporting of economic evaluation results.

We conducted uncertainty analysis to estimate the 95% credible interval of projected outcome, accounting for the uncertainty of input parameter values. In the operational component, each simulated year involved running the model for over 2,500 simulated days. Outcomes were recorded for each simulated 90 day period and the mean and standard deviation calculated. From these, 95% confidence limits for key outcome measures were calculated.

Cost-effectiveness measures were calculated to compare different diagnostic options. The incremental cost of implementing each alternative diagnostic option was derived from the Addis Ababa health system perspective, and included the additional annual running costs (eg, consumables, susceptible and drug resistant -tuberculosis drugs, radiographs, equipment maintenance, and laboratory personnel) and the investment costs (eg, microscopes and the equipment related to Xpert implementation) (Additional file 1). Other overhead costs were assumed to be unaffected by a change in the diagnostic algorithm. The introduction of new tuberculosis diagnostics is expected to improve the survival of patients co-infected with tuberculosis and HIV, therefore we estimated the incremental costs from additional antiretroviral therapy (ART) on the basis of the projected increase in HIV prevalence resulting from reductions in the number of deaths from tuberculosis with HIV co-infection. The population effect on tuberculosis epidemiology was summarized using disability- adjusted life years (DALYs). Costs and DALYs were calculated over 10 years with an annual discount rate of 3%. We calculated the average cost-effectiveness ratio (ACER) by comparing each alternative diagnostic option to the base case scenario. Because the alternative diagnostic options are mutually exclusive interventions that compete for the same resources, we also calculated the incremental cost- effectiveness ratios (ICER) to compare one option with the next less-effective option. The estimated ICER was compared with the willingness-to-pay threshold for Ethiopia (US$690 based on the gross domestic product [GDP] per capita in 2015) [20]. Interventions with an ICER of less than the Gross Domestic Product (GDP) per capita considered as cost-effective interventions [21] but need to be considered alongside affordability, budget impact, fairness, feasibility and any other criteria considred important in the local.

The full roll-out of Xpert MTB/RIF is projected to produce the greatest patient-level benefits among the eight alternative diagnostic options (Table 1). The modeling result showed that mean patient visits for diagnosis reduces by an average of 1·4 visits, time to start treatment reduces by an average of 7.8 days, and the diagnostic lost to follow-up rate reduces by 9% compared to the base case (ZN-Spot-Morning-Spot). The likelihood that a patient with tuberculosis will successfully complete diagnosis and treatment is increased by 8.3%. The next best algorithms from the patient perspective involve targeting of Xpert MTB/RIF to HIV-positive cases and high MDR risk groups alongside same day LED for the other TB presumptive cases (Targeted-Xpert-MDR-HIV-FN-Spot-Spot) and Xpert as a secondary test for smear negative new TB suspects (Targeted-Xpert-ZN-Negative-Spot-Morning-Spot).

Under full scale-up of Xpert, the required number of sputum samples per year is 54,000 compared to 113,000 under the base case. The full roll-out of Xpert requires only 33.3% of the laboratory staff time that is needed for the base case (Table 2). There is no significant change in drug sensitive tuberculosis notification among the alternative diagnosis algorithms. After false-positive cases are excluded, full Xpert scale-up is projected to increase the number of true tuberculosis cures by 346 (Table 2). The use of Xpert would also identify 207 rifampicin-resistant cases during the diagnostic process compared with only 59 MDR tuberculosis cases under the base case. Xpert reduces missed TB cases (patients with TB but not diagnosed or lost to follow-up) by 341 patients compared with the base case.

The modeling analysis indicated that full roll-out of Xpert is expected to avert approximately 9117 DALYs per years at average cost-effectiveness ratio (ACER) of $127 per disabilityadjusted life year (DALY) averted compared to the base case (Table 3). The average cost-effectiveness ratio (ACER) of other alternative diagnostic options ranged from $2 to $203 per DALY averted compared with the base case. Xpert as the primary test for HIV-positive cases and high MDR risk group alongside LED same day for all others (Targeted-Xpert-MDR-HIV-FN-Spot-Spot) is expected to avert 7360 DALYs per year and Xpert as secondary tests for smear negative new presumptive TB cases (Targeted-Xpert-ZN-Negative-Spot-Morning-Spot) is expected to avert 4724 DALYs. Other diagnostic algorithms would have less effect on DALYs, with the benefits ranging from 1747 (FN-Spot-Morning) to 4358 (FN-Spot-Spot) DALYs averted per year. The full rollout of Xpert requires an additional budget of an average US$ 1,161,133 budget per year over 10 years. Incremental cost for the other alternative algorithms varies between –US$ 52,727 and US$950,669 over the next 10 years (Table 3, Fig. 2)

.

In the analysis of the incremental cost-effectiveness ratio (ICER) options FN-Spot-Morning-Spot, Targeted-Xpert-ZN-Negative-Spot-Morning-Spot, and Targeted-Xpert-MDR-HIV-ZN-Spot-Morning-Spot are strongly dominated by same-day LED fluorescence microscopy (FN-Spot-Spot), LED same day with Xpert (Targeted-Xpert-MDR-HIV-FN-Spot-Spot) , and Xpert full rollout; because they produce DALY gains at a higher incremental cost . The ICER of Xpert full rollout is $370.1 per DALY averted , followed by LED Same Day + Xpert for HIV positive and high MDR risk group (Targeted-Xpert-MDR-HIV-FN-Spot-Spot) ($176.7), same-day LED fluorescence microscopy ($12.6) and LED fluorescence microscopy with two sample (-$30.2) (Table 3). The ICERs for the above alternative algorithms are lower than the annual per-capita GDP of Ethiopia (US$690) [20] and are deemed cost-effective.