Assumptions Patient population and setting This analysis is conducted among ambulatory HIV-infected patients in South Africa. Test qualities and diagnostic costs Estimates of sensitivity and specificity of diagnostic procedures used in the model are shown in Table 1 and are based on reports from the literature (chest x-ray alone, oral wash with DQ, PCR, nested PCR, and rtPCR; expectorated sputum with GMS, TBO, CW; induced sputum with DQ, GMS, TBO, IS, IFA, PCR, nPCR; BAL with DQ, GMS, TBO, CW, IFA, PCR, nPCR, rtPCR) [29], [30], [31], [36], [38], [39], [40], [44], [47], [48], [49], [50], [52], [54], [55], [57], [59], [60], [61], [62], [63], [64], [65], [66], [67], [68], [69], [70], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80], [81], [82], [83], [84], or estimated by the authors (oral wash with IFA, GMS, CW, TBO; expectorated sputum with DQ, IFA, PCR, nPCR, or rtPCR; induced sputum with rtPCR). Although all of the above-referenced literature was consulted, for diagnostic procedures involving any form of PCR, only studies which targeted the mitochondrial large subunit ribosomal RNA were used for estimations of sensitivity and specificity [28], [29], [36], [48], [51], [54], [55], [57], [63], [64], [67], [68], [69], [71], [72], [75], [76], [78], [85], [86]. For tests for which data did not exist in the literature, estimations of test characteristics were based on interpolation and pre-existing knowledge of the sensitivity and specificity of other tests in the same diagnostic category (e.g., we assumed that expectorated sputum with DQ, for which we did not find published reports, would be intermediate in sensitivity between oral wash with DQ and induced sputum with DQ, for which we were able to reference published reports.) Table 1 Model inputs and costs: Sensitivity and specificity of diagnostic procedures, based on estimates derived from existing studies (see text) or, when reference studies not available, from author estimation. Diagnostic Specimen collection Sensitivity Specificity Cost (USD) CXR None 0.86 0.40 $40.00 DQ Oral wash 0.30 1.00 $2.32 Expectorated sputum 0.60 1.00 $2.22 Induced sputum 0.75 1.00 $8.72 Bronchoalveolar lavage 0.75 1.00 $77.12 GMS Oral wash 0.30 1.00 $4.21 Expectorated sputum 0.52 0.95 $4.11 Induced sputum 0.70 0.96 $10.61 Bronchoalveolar lavage 0.82 0.98 $79.01 TBO Oral wash 0.30 1.00 $0.93 Expectorated sputum 0.71 1.00 $0.83 Induced sputum 0.75 1.00 $7.33 Bronchoalveolar lavage 0.80 1.00 $75.73 CW Oral wash 0.30 1.00 $2.94 Expectorated sputum 0.33 1.00 $2.84 Induced sputum 0.57 1.00 $9.34 Bronchoalveolar lavage 0.78 1.00 $77.74 IFA Oral wash 0.30 1.00 $20.79 Expectorated sputum 0.50 1.00 $20.69 Induced sputum 0.81 1.00 $27.19 Bronchoalveolar lavage 1.00 1.00 $95.59 PCR Oral wash 0.71 0.99 $8.78 Expectorated sputum 0.85 0.99 $8.68 Induced sputum 0.94 0.99 $15.18 Bronchoalveolar lavage 1.00 0.94 $83.58 nPCR Oral wash 0.83 1.00 $10.32 Expectorated sputum 0.91 1.00 $10.22 Induced sputum 1.00 1.00 $16.72 Bronchoalveolar lavage 1.00 0.89 $85.12 rtPCR Oral wash 0.89 0.94 $13.84 Expectorated sputum 0.92 0.94 $13.74 Induced sputum 0.95 0.90 $20.24 Bronchoalveolar lavage 0.99 0.80 $88.64 CXR: Chest x-ray; DQ: Diff-Quick; GMS: Grocott's Methenamine Silver Stain; TBO: Toluidine Blue O; CW: Calcofluor white stain; IFA: Immunofluorescence; PCR: Polymerase chain reaction; nPCR: nested PCR; rtPCR: real-time (quantitative) PCR. Estimates of costs include required materials and personnel time (Tables S1 and S2). Total costs for diagnostic procedures are included in Table 1. Estimated salaries for laboratory and health care workers are available in Appendix S1. Except where stated, all cost and time estimates were provided by the National Institute of Communicable Diseases in South Africa. Others The value referred to as ‘prevalence’ refers specifically to the prevalence of disease among patients with signs and symptoms of PCP who would normally warrant testing at a given hospital or clinic. It does not refer to the population prevalence of disease. This value will differ regionally; some hospitals or clinics might test all patients with respiratory disease and negative AFB smears, while others will test only patients who have a chest x-ray typical for PCP. Three models are also presented, with prevalences set at 5%, 20%, and 50%. Treatment failure, whether related to insufficient adherence to treatment or breakthrough infections during treatment to which the patient is adherent, is assumed to occur among 10% of patients (Table S3). Treatment costs are based on a single, 21-day regimen with oral CTX (Table S3). Patients are assumed to not be taking CTX at the time of diagnosis. Life-years gained In studies carried out before the year 2000, median survival time after AIDS diagnosis among patients in developing countries not on antiretroviral therapy was calculated to be approximately one year [87]. In the absence of treatment, PCP is generally accepted to lead to rapid death. Therefore, we assumed that diagnosis and appropriate treatment led to a single life-year gained among patients with PCP, compared with patients who were not diagnosed correctly. Model flow An example of model flow with sample values is depicted in Figure 1. ‘Ill patients’ refers to patients with PCP; ‘well persons’ refers to persons without PCP (although persons in this group likely have another illness, since they are undergoing testing). At a given PCP prevalence among persons tested, the number of ill patients correctly diagnosed is calculated as the sensitivity of the diagnostic procedure (Table 1) multiplied by the total number of ill patients. The number of well persons incorrectly classified as ill is equal to the total number of well persons, minus the procedural specificity (Table 1) multiplied by the total number of well persons. The total number of persons classified as ill is the sum of these values. Total diagnostic procedural costs are calculated as a sum of the health care worker and laboratory staff costs and material costs for the specimen collection and the diagnostic test procedures (Table 1 and Tables S1 and S2). Figure 1 Model flow. ‘Ill patients’ refers to patients with PCP. ‘Well persons’ refers to persons without PCP, regardless of their health status otherwise. Patients successfully treated are assumed to gain one life-year. All persons diagnosed as PCP-positive (correctly or incorrectly) are assumed to receive a full course of treatment. Treatment failure rates are considered as a combination of failure-to-adhere and breakthrough infection rates (Table S3). The number of patients who fail treatment is equal to the number of ill patients correctly classified as ill who undergo treatment, multiplied by the treatment failure rate. Because each patient is assumed to gain a single year of life from correct treatment, total life-years gained is equal to the number of ill patients correctly diagnosed minus those for whom treatment did not successfully treat infection (Figure 1). The proportion of ill patients successfully treated is represented by the number of patients successfully treated divided by the number ill, while the proportion unnecessarily treated is equal to the number of well persons treated divided by the total number of well persons. Total treatment costs are equal to the total number of well persons and ill patients who receive treatment, multiplied by the estimated treatment cost. Finally, the total diagnostic and treatment cost per life-year gained (the cost-effectiveness ratio) is equal to the sum of the total diagnostic costs and the total treatment costs, divided by the number of ill patients successfully treated. The incremental cost-effectiveness ratios of the most effective options were then calculated. Relapse rates are not considered. Start-up and indirect costs (building costs, laboratory equipment purchase, electricity, training) are also not considered, as they will differ greatly by region and available pre-existing infrastructure. Sensitivity analyses Sensitivity analyses were performed by varying specific parameters, including treatment costs, treatment failure rates, and costs of diagnostic procedures, over a range of plausible values to determine the impact of uncertainty in the data, and the robustness of results.