Introduction Active-case finding (ACF) programs have an important role in addressing case detection gaps and halting tuberculosis (TB) transmission. Evidence is limited on the cost-effectiveness of ACF interventions, particularly on how their value is impacted by different operational, epidemiological and patient care-seeking patterns. Methods We evaluated the costs and cost-effectiveness of a combined facility and community-based ACF intervention in Zambia that utilized mobile chest X-ray with computer-aided reading/interpretation software and laboratory-based Xpert MTB/RIF testing. Programmatic costs (in 2018 US dollars) were assessed from the health system perspective using prospectively collected cost and operational data. Cost-effectiveness of the ACF intervention was assessed as the incremental cost per TB death averted over a five-year time horizon using a multi-stage Markov state-transition model reflecting patient symptom-associated care-seeking and TB care under ACF compared to passive care. Results Over 18 months of field operations, the ACF intervention costed $435 to diagnose and initiate treatment for one person with TB. After accounting for patient symptom-associated care-seeking patterns in Zambia, we estimate that this one-time ACF intervention would incrementally diagnose 407 (7,207 versus 6,800) TB patients and avert 502 (611 versus 1,113) TB-associated deaths compared to the status quo (passive case finding), at an incremental cost of $2,284 per death averted over the next five-year period. HIV/TB mortality rate, patient symptom-associated care-seeking probabilities in the absence of ACF, and the costs of ACF patient screening were key drivers of cost-effectiveness. Conclusions A one-time comprehensive ACF intervention simultaneously operating in public health clinics and corresponding catchment communities can have important medium-term impact on case-finding and be cost-effective in Zambia. The value of such interventions increases if targeted to populations with high HIV/TB mortality, substantial barriers (both behavioral and physical) to care-seeking exist, and when ACF interventions can optimize screening by achieving operational efficiency.
The CIDRZ ACF project operated in the catchment area of the George primary health care centre (GPHC, covering a peri-urban settlement of 172,550 people), a Lusaka province public-sector facility offering TB diagnostic and treatment services. The intervention was operationalized by two distinct teams—community-based outreach and facility-based—which screened individuals for presumptive TB and facilitated linkage to care for diagnosis and treatment of TB. Diagnostic procedures included mobile X-ray followed by laboratory-based Xpert MTB/RIF (Xpert) testing. The community-based outreach team (which rotated through different areas of the GPHC catchment region, 3–4 days per area) employed a group of trained community health workers (CHWs) and drama teams who conducted community TB awareness events, community and door-to-door symptom screening, and chest X-rays taken in a mobile X-ray (mCXR) unit installed in a truck and interpreted by a computer-aided reading/interpretation software (CAD4TB Version 1.5, Delft Images, hereafter CAD4TB). The facility-based team operated out of an open-access tent at GPHC where four trained staff members conducted TB symptom screening and patient referral to TB services (including treatment initiation). In addition to screening people presenting (or referred) to the tent, the facility-based team also made regular visits to the antiretroviral therapy (ART) clinic and other departments (e.g., maternal health or general outpatient clinics) at GPHC to identify patients with TB symptoms and/or those patients otherwise indicated for TB screening (e.g., people living with HIV). Both community-based outreach and facility-based team collected sputum samples on-spot from the patients showing abnormal symptoms or Chest Xray results for Xpert testing in the GPHC laboratory once a patient was identified as a presumptive TB patient after initial symptom and mCXR screening. When on-site mCXR was not available (either in-use by one of the teams or inoperable), all symptomatic clients received Xpert without CXR screening. Patients with a positive Xpert test result were immediately followed-up by the ACF team and referred to the TB clinic for treatment initiation. Cost and resource-use data were collected using a standardized cost data collection and analysis tool developed by our team. This tool allows for cost and resource-use data collection by key activity component (i.e. training, screening, diagnosis, and treatment) for analysis using a top-down method [19]. Human resource costs for each activity component were estimated based on their estimated level of effort (LOE), approximated as proportional time allocation (%) of their full-time work spent on each activity during the program operations, assessed periodically (each quarter) using a workload survey. For the costs of goods, equipment, and services, we divided direct costs into capital and recurrent costs. Common programmatic costs (indirect and overhead costs that were shared across various ACF activities) were first calculated as total cost and were apportioned into each major ACF activity category based on direct human resource contribution (ratios of LOE across all ACF activities, weighted based the total person-time contribution assessed for each activity category). Cost data and program operation statistics were collected on a quarterly basis from July 2017 to December 2018 based on program financial documents and interviews with program managers. We evaluated these data against the respective service utilization and program statistics (e.g. number of patients) both quarterly and over the entire program period (total of 18 operational months). The main outcomes of the analysis were 1) unit costs of key service/activity components of the ACF intervention (calculated based on direct total costs of each discrete service/activity divided by the total number of patients who received the service); and 2) cost per direct program yield (cost per presumptive TB patient identified and per confirmed TB diagnosis). Capital assets were annualized based on the relevant expected life years and discounted at a 3% annual rate. All costs were assessed as economic costs from the health system perspective and reported in 2018 United States Dollars (USD) with cost data in local currencies converted based on the average United Nations operational exchange rates for 2018 [20]. Costs and effect estimates were estimated using a multi-stage Markov state-transition model (Fig 1) using monthly time steps. Two stages included 1) a symptom transition model to calculate symptom-based care-seeking probabilities without ACF and 2) a decision-analytic model representing ACF and Passive Case Finding (PCF) as a status quo diagnostic and treatment algorithm. In the symptom transition model (developed using Microsoft Excel), we calibrated the underlying monthly care-seeking probabilities and symptom transition rates to reproduce a population consistent with current TB epidemiology in Zambia for each of three symptom levels: 1) TB-asymptomatic (patients unaware of or without any TB-specific symptoms); 2) non-TB-specific symptoms (symptoms for which an Xpert test would be ordered, if patients were to present passively to the clinic) and 3) classical TB symptoms (symptoms for which TB treatment would be started empirically, even if Xpert testing were negative). Additional details of this symptom transition model are described in the Supporting Information (S2 and S3 Tables in S1 File and S2 and S3 Figs in S1 File) and elsewhere [21]. Symptom-specific care-seeking probabilities from the symptom transition model were incorporated into the decision-analytic model to estimate the incremental effect (diagnosis and averted TB mortality) of the ACF intervention over the five-year time horizon compared to the status quo. The decision-analytic model, built-in TreeAge Pro 2018 (Williamstown, MA, USA), represented the overall TB care cascade for both passive and active case finding for a simulated population of 100,000 individuals. Epidemiologic parameters for our symptom-transition model were obtained from national estimates and published literature [22, 23–27]. Cost-effectiveness of the CIDRZ ACF operation in a setting with TB epidemiology and economic conditions consistent with national averages was evaluated based on the incremental cost-effectiveness ratio (ICER), calculated as incremental cost per TB death averted over a five-year time horizon (chosen as the minimum interval that might occur between serial intensive ACF campaigns in practice), relative to the status quo (PCF). Complete list of parameters used in the model can be found in Table 1 and S3 Table in S1 File. In our symptom-based care seeking model, we defined three TB-symptom levels—asymptomatic, nonspecific, classic—based on the corresponding probability of diagnostic evaluation for TB. We calculated monthly transition rates between these symptom levels based on three constraints: 1. Probability of progression is 2 times that of regression; 2. Lifetime probability of TB self-cure equals that of death in the absence of treatment (untreated case fatality ratio of 0.5); and 3. Mean duration of asymptomatic period is 9 months. These values (monthly transition rates between symptom levels and monthly probabilities of seeking care) were inputted into a decision tree Markov model which was constructed to reflect the diagnostic algorithm (CXR and Xpert) used for Active Case Finding (ACF) in the Zambia TB REACH program. 100,000 individuals defined by TB/HIV status and symptom level and modeled as having a one-time chance to attend ACF (86% for nonspecific and classic symptom). Those who did not access ACF were modeled as seeking routine care with a monthly probability based on symptom development (20% for nonspecific and 40% for classic symptom) throughout the duration of the analysis. Individuals with untreated TB at the end of each monthly cycle experienced a monthly probability of symptom level transition (progression or regression). More detailed model structure and clinical diagnostic algorithms are described in the supporting information S2 and S3 Tables in S1 File; S2 and S3 Figs in S1 File. a. While multi-drug resistant (MDR) TB is not a part of standard monitoring indicators of the TB REACH program in Zambia, we incorporated a probability of MDR TB for the Markov state-transition model based on the country estimates. b. “High bacterial load” is defined as TB that, if tested with a single sputum smear under programmatic conditions, would test positive; “Low bacterial load” is defined as TB that, if tested with a single sputum smear under programmatic conditions, would test negative. c. Monthly mortality rate was estimated based on 1-EXP(-annual rate/12 months) from Vassall et al.[25] and WHO Zambia TB country profile (TB case fatality ratio as 31% in 2018)20. To test the robustness of our cost-effectiveness estimates, we performed a suite of sensitivity analyses (one-way, three-way, and probabilistic sensitivity analyses) based on the uncertainty estimates of each parameter. Parameters for the three-way sensitivity analysis were selected based on the ranking of the top three most influential parameters identified from one-way sensitivity analysis. For the Probabilistic Sensitivity Analysis (PSA), all model parameter values were randomly sampled over 10,000 Monte Carlo simulations based on pre-specified distributions of each data parameter to generate 95% uncertainty ranges. ICER estimates were evaluated against different willingness-to-pay (WTP) thresholds representing a range of financial and budgetary constraints in Zambia for public health interventions [28]. Neither ethical approval nor informed consent was required for this analysis which did not involve human subjects research. Neither patients nor the public were involved in the design, conduct, reporting, or dissemination plans of our research.
DIMA AI Care
