Preface
Sexually transmitted infections (STIs), including syphilis, gonorrhoea and chlamydia, exact a heavy toll on women in the developing world. Syphilis is a significant cause of adverse pregnancy outcomes, including stillbirth. Gonorrhoea and chlamydia are important causes of pelvic inflammatory disease and other debilitating conditions, especially among high-risk populations such as commercial sex workers (CSWs). The use of existing diagnostics for STIs is limited in the developing world due to cost, technical requirements and performance problems. This paper examines the potential impact of the introduction of better-performing and more-accessible diagnostics for STIs among two targeted populations of women in the developing world.
Introduction
Curable bacterial STIs such as gonorrhoea, chlamydia and syphilis, exact a heavy toll in terms of morbidity and mortality in the developing world. The long-term complications of STIs are disproportionately borne by women and their infants. In 1999, there were an estimated 3.5 million new (incident) syphilis cases, caused by the spirochete Treponema pallidum, involving adult women in sub-Saharan Africa, China and southeast Asia, representing 
70% of the global incidence1, 2. There were also 34 million and 50 million incident cases of gonorrhoea and chlamydia, respectively, in women, >75% of which occurred in these three regions1, 2, 3. Syphilis is a significant cause of adverse pregnancy outcomes (including spontaneous abortion, stillbirth, intrauterine growth retardation, premature delivery and perinatal deaths) as well as congenital disease in infected infants and children, but it rarely causes any symptoms in women4. However, syphilis and its sequelae can be successfully managed, and the latter prevented, if infected mothers are diagnosed and treated appropriately by the middle of the second trimester4, 5.
A significant proportion of the gonorrhoea and chlamydia disease burden in low-resource countries is also borne by women. Those who have multiple sexual contacts, such as CSWs, and those whose partners have visited CSWs are at substantial risk of acquiring gonorrhoea and chlamydia. Untreated gonorrhoea and chlamydia infections are important causes of pelvic inflammatory disease, tubal infertility, ectopic pregnancy and chronic pelvic pain among women; they can also cause complications through neonatal transmission and might facilitate human immunodeficiency virus (HIV) transmission6, 7, 8, 9, 10. As >50% of gonorrhoea and chlamydia infections are asymptomatic, most infected women fail to seek care. CSWs are at a particularly high risk of acquiring STIs, and of transmitting gonorrhoea and chlamydia to others in their communities11, 12; therefore, the control of these diseases among CSWs and their partners is recognized as an important strategy in preventing incident cases of gonorrhoea, chlamydia and HIV infection6, 7, 8, 9, 10, 13, 14, 15, 16, 17.
Photo by Liz Gilbert courtesy of the Bill & Melinda Gates Foundation
Despite the recognized need for early diagnosis and treatment of these curable STIs, detection strategies for syphilis, gonorrhoea and chlamydia are currently limited in low-resource countries. Options for the effective screening and serodiagnosis of syphilis infections in pregnant women involve two types of test: first, a nontreponemal test that detects cardiolipin antibodies, such as the rapid plasma reagin (RPR) test; and second, a test that confirms positive RPR results using T. pallidum-specific antigens, such as the T. pallidum haemagglutination test or the T. pallidum particle-agglutination test5. However, these tests require laboratory equipment, refrigeration, and trained personnel to perform and interpret the results, all of which are often unavailable in low-resource areas. Therefore, these tests rarely guide patient therapy in the developing world5, 18, 19. Newer rapid syphilis tests (RSTs) using finger-stick or venous blood take only 15–20 min, require minimal training and no equipment to use, and are easily interpretable by health-care staff20. A disadvantage of current RSTs, however, is that they use treponemal biomarkers, and antibodies to these biomarkers are retained in the body for years. Hence, all women who have ever had syphilis, as well as those with active untreated infections that might be passed on to their infants, can test positive by RSTs. These tests cannot therefore be used to distinguish between those women who have been previously screened and treated, and those who are at risk of adverse pregnancy outcomes and require treatment. Moreover, the use of RSTs might result in overtreatment in areas where women have been previously screened and treated.
The detection of Neisseria gonorrhoeae and Chlamydia trachomatis (the causative bacterial agents of gonorrhoea and chlamydia, respectively) among high-risk women in resource-limited settings also faces serious limitations21, 22. Microscopic examination of Gram staining is the only currently available laboratory method offering a relatively simple and inexpensive diagnosis of gonorrhoea. However, Gram staining is not widely used in resource-limited countries to detect gonorrhoea in women, because it requires the collection of cervical specimens during an invasive pelvic examination carried out by trained health-care workers. Moreover, Gram staining is relatively insensitive in women23. In practice, although highly sensitive and specific laboratory assays, such as nucleic-acid amplification tests (NAATs), are available for gonorrhoea and chlamydia23, 24, 25, 26, their high costs and technical requirements make routine use impossible in resource-limited settings22. Finally, although currently available rapid point-of-care (POC) tests for gonorrhoea and chlamydia have good reported specificity (98%) in field trials, their sensitivity in symptomatic women (50–70%) is considered to be too low to merit widespread use27, and their utility in asymptomatic women has not been fully examined28. Because of the lack of laboratory services for the diagnosis of gonorrhoea and chlamydia, providers in many resource-limited settings rely on syndromic algorithms and presumptively treat for all major causes, resulting in substantial overtreatment25, 28. To improve the specificity of syndromic management, a process of risk scoring has been introduced in an attempt to identify those individuals who are at highest risk of gonorrhoea and chlamydia infection. Unfortunately, for women with vaginal discharge and lower abdominal pain, such syndromic algorithms have been shown to have low sensitivity, fail to diagnose infection in many cases29, and have poor specificity resulting in continued overdiagnosis and unnecessary treatment in individuals without either or both diseases. Moreover, this approach fails to identify the majority of women with asymptomatic gonorrhoea and chlamydia.
Photo by Prashant Panjiar courtesy of the Bill & Melinda Gates Foundation
Recognizing the significant burden of STIs in developing countries, and the limitations of existing diagnostic strategies, the Global Health Diagnostics Forum of the Bill & Melinda Gates Foundation convened a working group of international experts to model and evaluate the potential health benefits of hypothetical new diagnostics for STIs. As a result of a series of discussions surrounding the clinical diagnosis, care and management of STIs among women, we determined that the most critical interventions in guiding clinical care would come from the following: first, a new syphilis screening diagnostic for antenatal women to reduce syphilis-attributable mortality and morbidity in infants in sub-Saharan Africa, where the disease burden is particularly high; and second, new gonorrhoea and chlamydia diagnostic tests for symptomatic and asymptomatic CSWs to reduce morbidity and incident cases of gonorrhoea, chlamydia and HIV infection in sub-Saharan Africa, China and southeast Asia, which are the three regions that account for the majority of incident infections. We considered a number of other possible scenarios and sub-populations for analysis, including gonorrhoea and chlamydia screening in antenatal and other lower-risk women, and syphilis screening in asymptomatic CSWs who attend STI clinics. However, we determined the two scenarios described above to be the intervention points of highest priority for reducing the disability associated with these diseases.
Methods
Analytic overview
Separate decision-tree models were developed to depict current and potential new diagnostic and treatment strategies for syphilis, and for gonorrhoea and chlamydia infection. The models calculate the benefits of improved tests relative to the current diagnostic standard of care (referred to as the status quo) in terms of specific health outcomes. The antenatal syphilis screening model reports outcomes in terms of annual congenital syphilis cases and stillbirths averted in sub-Saharan Africa, as well as adjusted lives saved, which accounts for potential harm associated with treatment.
The gonorrhoea and chlamydia model, which focuses on CSWs in sub-Saharan Africa, southeast Asia and China, presents primary outcomes in terms of disability-adjusted life years (DALYs) saved by treating the index case (the initial patient in an epidemiological investigation) and by preventing downstream cases due to transmission. DALYs saved are also further adjusted to account for the harm of treatment.
For both models, we varied diagnostic screening scenarios by the diagnostic technology requirements (advanced/moderate, minimal or no infrastructure) and the corresponding level of health-care access, and by test performance characteristics (sensitivity and specificity). Access to different levels of health care for both models was estimated from a multinomial logistic-regression model informed by data obtained from the Demographic and Health Surveys (DHS) for 17 African and six Asian countries between 2000 and 2005 (ref. 30). In our models, advanced infrastructure implies consistent access to running water and electricity, well-trained staff and good laboratory facilities. Moderate infrastructure implies consistent access to running water and electricity, some equipment (such as a microscope) and one or two trained technicians. Minimal infrastructure refers to settings without reliably available water and electricity, a physical location with no laboratory equipment and a minimally trained health provider (such as a traditional birth attendant). No infrastructure refers to settings without water or electricity, and makes no assumptions about the training or literacy of the caregiver30.
We measured the comparative health benefits of various new diagnostic scenarios in each model by comparing the difference in health outcomes between the new diagnostic test and the status quo. For antenatal syphilis, we calculated the incremental number of identified and appropriately treated cases of syphilis relative to the status quo, and translated them into health outcomes using published data on syphilis-attributable adverse events. For gonorrhoea and chlamydia, we calculated the incremental number of cases averted relative to the status quo, and used published estimates of total gonorrhoea and chlamydia DALYs lost in each region to derive the DALYs saved by identifying and appropriately treating infections. In our model, the DALYs saved measure consists of two separate components: those saved by treating the immediate gonorrhoea and/or chlamydia index case, and those saved by preventing subsequent downstream cases. Therefore, to capture the population effect of gonorrhoea and chlamydia diagnosis and treatment in CSWs, we account for the fact that treating individual infections in a highly sexually active population interrupts transmission and prevents future infections in the community. As there is evidence that gonorrhoea and chlamydia facilitate the transmission of HIV, we also report the number of HIV infections that could be averted with appropriate gonorrhoea and chlamydia diagnosis and treatment as a secondary outcome (we did not consider HIV transmission in the antenatal syphilis model). The calculation of outcomes for the gonorrhoea and chlamydia model is described in detail elsewhere31.
Finally, we varied the assumptions about each of the model input parameters extensively in sensitivity analyses to test the robustness of our results and conducted separate model calibration analyses to evaluate the ability of our status quo models to predict reasonable estimates of disease-specific outcomes. Details of the calibration are presented elsewhere31, 32.
Modelling antenatal syphilis diagnosis and management
The decision tree shown in Fig. 1a is a simplified characterization of the current status quo antenatal syphilis diagnostic and treatment algorithm in sub-Saharan Africa. Syphilis screening for pregnant women in sub-Saharan Africa is mainly provided in the context of antenatal care. Consequently, the decision tree first divides the population into pregnant women with and without antenatal care. Women with access to ante-natal care are then either screened with an RPR test or not screened, as many antenatal care sites in sub-Saharan Africa do not currently offer syphilis screening. We allow for a range of RPR test performance characteristics in the model, depending on whether the diagnostic is performed in a laboratory, clinic or field-based setting. The RPR test is not often used as a POC test; hence, there is typically a delay between the test administration and the receipt of results, and significant numbers of patients might be lost to follow-up. Therefore, we assume that only women who return for their results are eligible for treatment. Women who test positive and receive treatment are assumed to face a reduced or negligible probability of an adverse outcome, such as stillbirth or congenital syphilis. Women who test positive and do not receive treatment are assumed to experience the same frequency of adverse outcomes as women whose infection is undetected. Finally, women without access to antenatal care are assumed not to be eligible for screening or treatment.
Figure 1: Antenatal syphilis decision trees.
(a) Antenatal syphilis screening status quo decision tree. (b) Antenatal syphilis screening new diagnostic decision tree. FN, false negative; FP, false positive; RPR, rapid plasma reagin; TN, true negative; TP, true positive.
High resolution image and legend (104K)We have made some additional simplifying assumptions about the course of infection in pregnant women. We assume that syphilis is present at the beginning of the pregnancy, and that early testing will detect all active cases; we therefore discount the relatively small proportion of women who might acquire syphilis during pregnancy. We also assume that women with access to antenatal care who are screened for active syphilis and who receive treatment do so at the first antenatal visit, which is optimally at 2 months, and that treatment is 100% effective.
We have simulated the introduction of a hypothetical new antenatal syphilis test, as depicted in the decision tree in Fig. 1b. We assume that some proportion of pregnant women will have access to this new test, and that as the technological requirements decrease, the diagnostic becomes more accessible to our target population of pregnant women. Conditional on access to the new diagnostic, women are again divided into those with and without access to antenatal care, as in the status quo. Women with access to antenatal care, however, might now be tested either with an RPR test or with the new diagnostic. Women receive the new test only if it is associated with better outcomes relative to the status quo.
Syphilis model parameters
Table 1 summarizes the syphilis model parameter values and plausible ranges used in sensitivity analyses. To estimate the prevalence of active syphilis (that is, the cases that require treatment during pregnancy), we relied on published studies of infection rates among antenatal care attendees and assumed that the same rates apply to women who do not have access to antenatal care. Specifically, following Gloyd and colleagues, we assumed a population-weighted mean prevalence of antenatal active syphilis of 8% across 14 sub-Saharan African countries18.
To capture mortality and morbidity associated with maternal syphilis, we calculated the numbers of stillbirths and congenital syphilis cases due to active syphilis. Based on estimates from the published literature, a stillbirth risk of 22% was assigned to women with active syphilis who are either undetected or detected but fail to receive treatment32, 33, 34. The risk of stillbirth and congenital syphilis for treated and untreated active syphilis, and the probability of receiving effective treatment, are reported in Table 1.
In addition to the congenital syphilis and stillbirth cases averted, we also report health outcomes in terms of adjusted lives saved with a new diagnostic. Specifically, when a pregnant woman is treated with antibiotics for active syphilis, the possibility of some harm to society (for example, adverse events associated with treatment or the development of antibiotic resistance) is considered. Inappropriate treatment results in a similar penalty in the model. Conceptually, if a test for active syphilis is widely accepted and adopted by the medical community, then using it must be better than the alternative of treating every woman or treating no one. Consistent with the neoclassical economics principle that the preferences of rational people are revealed by the choices they make, we assume that these preferences provide adequate information to estimate how much weight the medical community assigns to specificity relative to sensitivity. This revealed preference is then translated into an estimate for the harm of treatment, which is referred to as parameter C and represents the proportion of lives lost as a result of treating one woman for active syphilis. Details of this approach and calculation are presented elsewhere30.
Modelling gonorrhoea and chlamydia diagnosis and management
Figure 2a represents a status quo decision tree based upon gonorrhoea and chlamydia tests that are theoretically available to CSWs in sub-Saharan Afr ica, China and southeast Asia. A small proportion of these women presenting to an STI or other health-care clinic might be screened for gonorrhoea and chlamydia with a laboratory test accompanied by a clinical evaluation. Alternatively, in the absence of a laboratory-based evaluation, a woman presenting with signs or symptoms of gonorrhoea and chlamydia might receive a clinical evaluation, following the syndromic management approach. Finally, women who present to clinics that do not offer gonorrhoea and chlamydia testing or clinical assessment of any kind will not be evaluated, and some might not have access to a clinic.
Figure 2: Decision trees for gonorrhoea and chlamydia.
(a) Gonorrhoea and chlamydia status quo decision tree. (b) Gonorrhoea and chlamydia new diagnostic decision tree.
High resolution image and legend (120K)For simplicity, the sensitivity and specificity of the laboratory test modelled here are a composite of culture, microscopy and NAATs, which are used sporadically in isolated clinics or health-care settings, as well as in research and field studies. As microscopy accounts for most of the small number of laboratory-based gonorrhoea tests used in resource-limited settings (despite its low sensitivity in women), the sensitivity and specificity of the composite laboratory test are weighted to most closely resemble microscopy. For the minority who receive both laboratory and clinical evaluations, we assume that a woman has gonorrhoea and chlamydia if she is positive by either test, and that she must test negative by both tests to be considered infection-free (that is, the laboratory and clinical evaluations are mathematically independent).
We assume that women who test positive for gonorrhoea and chlamydia by laboratory diagnosis and/or clinical evaluation and receive therapy are treated according to the World Health Organization (WHO) guidelines25. However, due to potential antibiotic supply stock-outs, or failure to use effective antimicrobials when available, the possibility that some proportion of women who test positive will not receive treatment is allowed15. We also assume that some proportion of women without access to a clinic will treat themselves, and that only a fraction of these individuals will select the appropriate regimen with regard to antibiotic, dose and duration of therapy. Because genitourinary symptoms are a primary driver of health-seeking behaviour for sexually active high-risk women, we assume that most CSWs will present to a clinic with symptoms (such as vaginal discharge, lower abdominal pain or ulcers). A small minority of women without symptoms might receive a clinical work-up (for example, if they were referred due to partner contact or were offered routine monthly examinations). However, because many gonorrhoea and chlamydia cases in women are asymptomatic, and some ignore or self-treat their symptoms, most women in our hypothetical model population do not seek clinic-based care.
Figure 2b illustrates the introduction of a hypothetical new diagnostic for gonorrhoea and chlamydia infection to which some proportion of CSWs will have access. The determination of access to the new diagnostic is similar to that in the syphilis model described above. Women are divided into those with and without access to a clinic, and then into the three further groups that mimic the status quo decision tree in Fig. 2a. Women who present to a clinic, however, now have the option of being tested with the new diagnostic. Similar to the syphilis model, women receive the new test only if it is associated with better health outcomes (in terms of DALYs) than the status quo test it replaces30.
Gonorrhoea and chlamydia model parameters
Table 2 summarizes the values and plausible ranges used in the gonorrhoea and chlamydia model base case and sensitivity analyses for sub-Saharan Africa, China and southeast Asia.
Until recently, the number of CSWs in a given region has been difficult to quantify. Although data are available on the prevalence of STIs among CSWs in many countries, and numerous studies in such populations provide empirical evidence about screening, diagnosis and treatment effectiveness16, 35, 36, 37, 38, 39, 40, 41, 42, they generally rely upon small samples of self-identified CSWs, and therefore do not permit inferences about the overall proportion of sexually active women who might engage in commercial sex work. However, a recent study by Vandepitte and colleagues estim-ated the prevalence of CSWs in female pop-ulations in different regions of the world based on survey data and the published literature, and provided reasonable baseline values and ranges for our three regions of interest43.
Similar to the antenatal syphilis screening model, health outcomes were adjusted to account for harm that might result from gonorrhoea and chlamydia treatment (for example, antibiotic drug resistance or social stigma). Unlike syphilis, however, the estimate of harm of treatment for gonorrhoea and chlamydia represents the proportion of DALYs lost (instead of the proportion of lives lost) as a result of treating one woman.
Estimates of DALYs lost due to gonorrhoea and chlamydia were derived from Murray and Lopez44, and the number of downstream gonorrhoea, chlamydia and HIV infections generated by an index case was based on a dynamic deterministic mathematical model previously developed by Vickerman and colleagues45, 46. Their model was fitted to available gonorrhoea, chlamydia, HIV and genital ulcer disease (GUD) prevalence data to estimate the impact of STI treatment interventions in two populations of female CSWs in South Africa and Benin. We used that model to estimate the impact of hypothetical gonorrhoea and chlamydia diagnostic tests with different performance characteristics on the number of downstream cases of gonorrhoea, chlamydia and HIV averted over 4 years. For each hypothetical test, the different model fits produced a range of estimates for the number of subsequent infections that would be averted among CSWs, their clients and the general population for every gonorrhoea and chlamydia case treated. Details of the approach are provided elsewhere31, 45, 46.
Sensitivity analyses
To account for a high degree of uncertainty in the models, the effects of parameter estimates on outcomes for specific syphilis, gonorrhoea and chlamydia tests were evaluated in a series of one-way sensitivity analyses, in which each parameter was varied independently. Elasticity — an economic concept that measures the responsiveness of one variable to another in terms of percentage change — was used to quantify the impact of uncertainty in the model parameters on health outcomes.
A series of two-way sensitivity analyses was also performed to explore continuous trade-offs between key variables, such as access to care, and the sensitivity and specificity of the new diagnostics. Finally, Monte-Carlo simulation was used to estimate the probable range of expected outcomes. All input parameters were allowed to vary independently over a defined range unless they were explicitly linked in the model47. Ranges were derived from the published literature and expert opinion. Input variables were randomly drawn 1,000 times from defined probability distributions, and a complete run of the model was executed for each draw. Outcomes were recorded and collated as distributions of health outcomes for each scenario.
Results
Tables 3 and 4 present the incremental health benefits of hypothetical new diagnostics, relative to the status quo strategies, for antenatal syphilis screening in pregnant women in sub-Saharan Africa, and for gonorrhoea and chlamydia screening in CSWs, respectively. Aggregate results for sub-Saharan Africa, China and south east Asia are presented for the gonorrhoea and chlamydia model. Each numbered row in Tables 3 and 4 corresponds to a particular hypothetical new diagnostic. For syphilis, new tests are defined by sensitivity, specificity, infrastructure requirements (advanced/moderate, minimal or no health-care infrastructure), patient return rate (60 or 100%) and RPR-positive treatment rate (70 or 100%). For example, faster tests generate results during the same diagnostic visit, thereby resulting in no patient loss to follow-up and an effective patient return rate of 100%. For gonorrhoea and chlamydia, hypothetical new tests are defined by sensitivity, specificity and infrastructure requirements. In each model, the level of infrastructure determines the probability that an individual woman has access to the new test30. For instance, a new test requiring no health-care infrastructure is assumed to be available to all women, whereas a new test requiring advanced infrastructure is assumed to be available only to those with access to a hospital or clinic with laboratory services. The numbers in parentheses in Tables 3 and 4 are standard deviations generated from a Monte-Carlo analysis that quantifies the uncertainty around each estimate.
High patient return and treatment rates improve syphilis outcomes
The results indicate that improvements in test sensitivity and specificity for syphilis, without a corresponding increase in access to treatment or patient return rate, do not yield substantial gains in health outcomes compared with the status quo laboratory-based RPR test. As a measure of the relative effectiveness of each hypothetical new diagnostic, Table 3 reports adjusted lives saved as a proportion of the maximum adjusted lives saved associated with an optimal new syphilis screening diagnostic (Table 3, test 25). This test saves 461,500 adjusted lives compared with the status quo, which is the maximum preventable disease burden accounted for by the model. In comparison, tests that are 100% sensitive and 100% specific, but do not improve upon the status quo patient return or treatment rate, alleviate only 2% (Table 3, test 3), 24% (Table 3, test 12) and 36% (Table 3, test 21) of the disease burden, depending on the level of required infrastructure. Notably, the incremental health benefits are substantial even with tests that are less sensitive and less specific than the status quo laboratory-based RPR test, as long as the test is rapid and the treatment rate is high. A minimal infrastructure test that is only 75% sensitive and 65% specific (Table 3, test 11) will still alleviate 47% of the disease burden. Maximal gains are therefore realized by faster tests that result in no patient loss to follow-up and are accompanied by high treatment rates for women who test positive.
Infrastructure requirement of syphilis RPR test limits health benefits
The results of the analysis also indicate that as long as a new diagnostic requires advanced-to-moderate levels of infrastructure, its health benefit will be limited. Even in the case of a perfectly sensitive and specific test, which can be performed during the patient visit (a rapid test) and is accompanied by 100% treatment for women who test positive (Table 3, test 7), only 20% of the total possible adjusted lives will be saved if the diagnostic requires advanced-to-moderate infrastructure. Although this is not an insignificant benefit, improved results can be achieved when new diagnostics require only minimal infrastructure and can therefore be accessed by women who do not currently have access to the status quo laboratory-based RPR test. For example, minimal infrastructure tests that are 86% sensitive and 72% specific, which have either 100% treatment or 100% return rates (Table 3, tests 13 and 14), prevent up to 30% and 37% of disease burden, respectively. Figure 3 shows that significant gains in syphilis-related pregnancy health outcomes over the status quo are realized with minimal infrastructure tests that have comparable sensitivity and specificity to the laboratory-based RPR (that is, 86% sensitivity and 72% specificity), and that even greater gains can be achieved by tests that require no infrastructure.
Figure 3: Incremental congenital syphilis cases averted, stillbirths averted and adjusted lives saved for a 1-year period (compared with the status quo) across different infrastructure requirements. Results are shown for a test with sensitivity and specificity comparable to the laboratory-based rapid plasma reagin test (86% sensitive and 72% specific), but which also generates results in the same diagnostic visit.
![Figure 3 : Incremental congenital syphilis cases averted, stillbirths averted and adjusted lives saved for a 1-year period (compared with the status quo) across different infrastructure requirements. Results are shown for a test with sensitivity and specificity comparable to the laboratory-based rapid plasma reagin test (86|[percnt]| sensitive and 72|[percnt]| specific), but which also generates results in the same diagnostic visit. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nature/journal/v444/n1s/images/nature05447-f3.0.jpg)
High resolution image and legend (30K)
Increased access to better gonorrhoea and chlamydia tests improves outcomes
The results in Table 4 indicate that there are limited gains from expanding access to the current gonorrhoea and chlamydia diagnostic strategies. For example, even if the current strategy of clinical evaluation were available to all CSWs, regardless of gonorrhoea and chlamydia signs or symptoms, only 22% of the disease burden accounted for in the model would be alleviated (Table 4, test 15). The results also show that diagnostic tests that improve upon the status quo strategies of clinical evaluation alone or in conjunction with a laboratory test can alleviate a significant proportion of the gonorrhoea and chlamydia disease burden when more women have access to them. Specifically, new tests that offer significant improvements in sensitivity and specificity compared with the status quo strategies (for example, tests that are 85% sensitive and 90% specific), and that require minimal or no infrastructure, reduce the burden of disease by 58% and 77% (Table 4, tests 12 and 19), respectively, saving 3–4 million adjusted DALYs. These gains are substantial compared with the 1.5 million adjusted DALYs saved, or 30% of the total disease burden, associated with a test with 100% sensitivity and 100% specificity requiring advanced or moderate infrastructure (Table 4, test 7). Moreover, significant increases in access and test sensitivity, both of which increase the detection and treatment rate, can substantially outweigh the loss of DALYs that arises from the harm associated with treatment, even when there is no corresponding improvement in test specificity that would reduce the number of women who are treated inappropriately. Test 18 in Table 4 is a case in point: when sensitivity is increased to 85% relative to the status quo strategy of clinical evaluation, the test specificity remains 70% and the test is universally accessible, 65% of the maximum adjusted DALYs are saved.
When we consider the effect of improved access to better performance gonorrhoea and chlamydia diagnostics on disease transmission, analogous benefits in terms of incident gonorrhoea, chlamydia and HIV cases averted are realized. For example, a new diagnostic for gonorrhoea and chlamydia that increases sensitivity to 85%, specificity to 90% and requires minimal infrastructure (Table 4, test 12), would avert >12 million incident gonorrhoea and chlamydia infections, and prevent >161,000 HIV infections among female CSWs in sub-Saharan Africa, China and southeast Asia over a 4-year period. A similar test requiring no infrastructure (Table 4, test 19) would avert >16.6 million incident gonorrhoea and chlamydia infections, and prevent >212,000 new HIV infections.
Results are sensitive to new diagnostic assumptions
Figure 4 presents results of one-way sensitivity analyses for two hypothetical antenatal syphilis screening tests in sub-Saharan Africa, both of which are faster than the laboratory-based RPR but have the same sensitivity and specificity. The analyses show that, in terms of elasticity, assumptions about access to the new diagnostic as well as its sensitivity, and associated treatment and patient return rates, have the greatest impact on adjusted lives saved. For gonorrhoea and chlamydia, health outcomes associated with new diagnostics in sub-Saharan Africa that are 85% sensitive and 90% specific are most influenced by assumptions about sensitivity, the calculation of average DALYs per gonorrhoea and chlamydia infection, access and the size of the CSW population (Fig. 5). Disaggregated one-way sensitivity analyses for China and southeast Asia can be found elsewhere31. Note that the outcomes in both models are less sensitive to assumptions about the harm of treatment, which is one of the most uncertain estimates in both sets of analyses.
Figure 4: Impact of a 10% change in a given input parameter on adjusted lives saved for two different hypothetical antenatal syphilis screening tests compared with the status quo laboratory-based rapid plasma reagin (RPR) test.
![Figure 4 : Impact of a 10|[percnt]| change in a given input parameter on adjusted lives saved for two different hypothetical antenatal syphilis screening tests compared with the status quo laboratory-based rapid plasma reagin (RPR) test. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nature/journal/v444/n1s/images/nature05447-f4.0.jpg)
Both new diagnostics have the same sensitivity and specificity as the laboratory-based RPR, but are faster and result in no patient loss to follow-up. (a) Results for a test that requires minimal infrastructure. (b) Results for a test that requires no infrastructure. The treatment rate for both diagnostics is the same as in the status quo (70%). On the vertical axes, individual parameters are ranked from top to bottom in order of increasing impact on outcomes. The horizontal axes present elasticity in the model outcome of adjusted lives saved. For example, the elasticity of the stillbirth rate for false negatives (that is, for untreated syphilis) is 1.2 in (a). By implication, if we underestimate this rate by 10%, we will underestimate the net health benefit by 12%. ANC, antenatal care; RPR, rapid plasma reagin test.
Figure 5: Impact of a 10% change in a given input parameter on adjusted disability-adjusted life years (DALYs) saved for two different hypothetical gonorrhoea and chlamydia diagnostic tests compared with the status quo diagnostic strategies.
![Figure 5 : Impact of a 10|[percnt]| change in a given input parameter on adjusted disability-adjusted life years (DALYs) saved for two different hypothetical gonorrhoea and chlamydia diagnostic tests compared with the status quo diagnostic strategies. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nature/journal/v444/n1s/images/nature05447-f5.0.jpg)
Both new tests are 85% sensitive and 90% specific. Results are shown for sub-Saharan Africa only. (a) Results for a test that requires minimal infrastructure. (b) Results for a test that requires no infrastructure. On the vertical axes, individual parameters are ranked from top to bottom in order of increasing impact on outcomes. The horizontal axes present elasticity in the model outcome of adjusted DALYs saved. For example, the elasticity of the specificity of the new diagnostic is 0.7. CSWs, commercial sex workers; TP, true positives.
Two-way sensitivity analyses that explore continuous trade-offs among access to the new diagnostic (determined by the different levels of infrastructure), its sensitivity and specificity, and treatment rates generally support the finding that as access increases to a test that is faster than the current RPR status quo (which generates results during the same patient visit), the syphilis-related pregnancy health outcomes are significantly improved. In addition, the sensitivity of the test appears to have a greater influence than specificity on outcomes (Fig. 6). Due primarily to the low harm associated with treatment, reducing inappropriate use (that is, improving specificity) does not appear to affect outcomes as much as effectively diagnosing and treating more people with disease (that is, improving sensitivity). Two-way sensitivity analyses for gonorrhoea and chlamydia (data not shown) indicate that diagnostic tests that improve upon the status quo can alleviate a significant proportion of the disease burden if they are widely accessible31.
Figure 6: Adjusted lives saved (in thousands) in sub-Saharan Africa for antenatal syphilis screening tests that are comparable in performance characteristics to the laboratory-based rapid plasma reagin (RPR) test, but are faster. The y-axes varies access to the new diagnostic (a function of the level of infrastructure requirements) from 0 to 100% (that is, no universal access). The x-axes in (a) varies the sensitivity of the new diagnostic, holding the specificity constant at 72%. The x-axes in (b) varies the specificity of the new diagnostic, holding the sensitivity constant at 86%. Isocurves are presented as a function of access to the test and either sensitivity or specificity on a colour gradient from white to green, where white is the maximum number of adjusted lives saved and green is the minimum.
![Figure 6 : Adjusted lives saved (in thousands) in sub-Saharan Africa for antenatal syphilis screening tests that are comparable in performance characteristics to the laboratory-based rapid plasma reagin (RPR) test, but are faster. The y-axes varies access to the new diagnostic (a function of the level of infrastructure requirements) from 0 to 100|[percnt]| (that is, no universal access). The x-axes in (a) varies the sensitivity of the new diagnostic, holding the specificity constant at 72|[percnt]|. The x-axes in (b) varies the specificity of the new diagnostic, holding the sensitivity constant at 86|[percnt]|. Isocurves are presented as a function of access to the test and either sensitivity or specificity on a colour gradient from white to green, where white is the maximum number of adjusted lives saved and green is the minimum. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nature/journal/v444/n1s/images/nature05447-f6.0.jpg)
High resolution image and legend (63K)
Discussion
The potential effects of hypothetical new diagnostics for syphilis, and for gonorrhoea and chlamydia, were quantified for two respective high-priority populations in the developing world: an antenatal population generally representative of sub-Saharan Africa; and a population of CSWs generally representative of sub-Saharan Africa, China and southeast Asia. By evaluating multiple scenarios involving different infrastructure requirements, we explored the benefits of new diagnostics under the assumption that women with geographic access to the new tests would receive them. Current socioeconomic, cultural and behavioural barriers to care were qualitatively controlled for in the assessment.
The analyses of different diagnostic intervention scenarios highlight the impact of a variety of factors on the reduction of syphilis, gonorrhoea, chlamydia and HIV cases, including access to a hypothetical new diagnostic (as indicated by infrastructure requirements), patient return rates associated with a new test, the availability of effective treatment, and the sensitivity and specificity of the diagnostic. The interaction of these factors indicates that imperfect syphilis diagnostic tests with sensitivity and specificity well below 100% can have substantial health benefits for pregnant women in sub-Saharan Africa. If a rapid test with sensitivity and specificity characteristics similar to the current laboratory-based RPR test could be readily used in settings with minimal infrastructure requirements, or as a POC test in settings with no infrastructure, >50% of the maximum disease burden could be averted.
Our results for gonorrhoea and chlamydia point to different conclusions for high-risk women, indicating that a significant proportion of the disease burden is attributable to the poor sensitivity and specificity of current laboratory and clinical evaluation strategies, as well as the lack of access. Consequently, >75% of the disease burden accounted for could be averted with a test that outperforms the current strategies and is suitable for minimal-to-no infrastructure settings. If the impact of preventing gonorrhoea and chlamydia transmission is con-sidered, as well as the possible influence on HIV incidence, analogous reductions in new HIV cases from improved gonorrhoea and chlamydia diagnostics over a 4-year time period are seen.
These analyses have several important limitations. First, our choice of analytic framework necessarily informs the kinds of inferences that can be made about the benefit of a new diagnostic. A static decision-tree model was employed rather than a dynamic transmission model, in order to focus on the individual trade-offs and health consequences associated with different test outcomes. As such, the analysis of syphilis probably underestimates the true benefits of a new test, because it does not consider the population benefits from averting syphilis transmission to sexual partners of pregnant women and beyond to wider sexual networks, or of reducing the impact (positive and negative) of partner notification in averting male sequelae and further decreasing secondary transmission. Although transmission effects were incorporated into the gonorrhoea and chlamydia analysis, our approach only estimated the downstream cases of gonorrhoea/chlamydia and HIV averted over a 4-year period. The multipliers used in the model might therefore underestimate the potential effect on gonorrhoea, chlamydia and HIV incidence over a longer time period. Moreover, as empirical estimates of the number of gonorrhoea, chlamydia and HIV infections resulting from index cases of gonorrhoea and chlamydia in communities in China and southeast Asia were not available, the estimates reported by Vickerman and colleagues (derived from CSWs in Benin and South Africa) were applied uniformly to the population of CSWs in sub-Saharan Africa, China and southeast Asia. The model should be updated as these data become available.
Second, although our analyses focused on syphilis, and gonorrhoea and chlamydia, we recognize that GUDs (including genital herpes) and vaginal infections cause substantial morbidity, and have also been implicated in increasing the risk of HIV acquisition or transmission48. Randomized controlled trials are currently examining the role of herpes simplex virus type-2 suppression in preventing the acquisition or transmission of HIV infection. The potential role of improved diagnostic tests in averting the morbidity attributable to GUDs also warrants further analyses.
Third, because the gonorrhoea and chlamydia analysis focused on a narrow segment of the female population, and maintained the assumption of imperfect treatment rates, it should be recognized that any new diagnostic used in this context will prevent only a fraction of the true disease burden.
Fourth and finally, given the high degree of uncertainty in all of the model parameters, the standard deviation for each estimate of potential benefit is relatively large, and the coefficients of variation (that is, the ratio between the standard deviation and the mean) range from 30 to 50% among most diagnostic testing scenarios evaluated. Although the conclusions do not change as a result of this uncertainty, it is reasonable to expect that they might be affected by differences in country-specific characteristics, such as disease prevalence, the syphilis stillbirth rate, access to antenatal care and treatment rates.
Significant health benefits in resource-limited settings could be realized by new diagnostics for syphilis, gonorrhoea and chlamydia, conditional on treatment availability. For syphilis, there is a specific need for diagnostics that can definitively identify the presence of T. pallidum in antenatal women regardless of outward signs or symptoms, so that treatment can be more effectively targeted to those who can pass the infection to their infants in utero and during delivery. Furthermore, significant gains in syphilis-related pregnancy health outcomes can be realized with a diagnostic that is comparable to the current RPR tests in terms of performance characteristics (86% sensitive and 72% specific), returns results within 2 h, and is linked to widely available and effective syphilis treatment that can be delivered at the time of testing. In order to ensure broad and routine access in settings with minimal health-care infrastructure, the test should be simple enough to be performed by a health-care worker with minimal training, should require no clean water, electricity or refrigeration, and should utilize easily collected clinical samples (such as finger-stick blood, urine or saliva). Although tests of T. pallidum biomarkers have not been identified in urine or saliva specimens, their identification could expedite the development of new diagnostics that might avert hundreds of thousands of stillbirths and congenital syphilis cases in sub-Saharan Africa.
Similarly, there is an urgent need for the development of new tests for gonorrhoea and chlamydia diagnosis in highly sexually active women that are 
85% sensitive and 90% specific, and that can produce results in <2 h by caregivers with minimal training, without electricity or refrigeration, using patient-collected urine or vaginal swabs in low-resource STI clinics and other settings. These targets are feasible using current and emerging technologies, and, if realized and adopted into clinical practice in the developing world, will alleviate significant STI morbidity and HIV-related cases (see Box 1 for key messages).