Syphilis is an infectious disease caused by Treponema pallidum subsp. pallidum (T. Pallidum) infection that can be sexually transmitted (Acquired Syphilis—AS) or through vertical transmission during pregnancy (Congenital Syphilis—CS) (1–3). Although curable and preventable through barrier methods (such as condoms), syphilis has been neglected and still represents a global public health concern due to inadequate diagnosis and treatment, resulting in morbidity and mortality in newborns and untreated infected people (4, 5).

According to the World Health Organization (WHO), over 357 million new cases of curable sexually transmitted infections (STIs) were diagnosed among young adults (15–49 years) in 2016 alone, of which 6 million were associated with syphilis (6). Currently, Brazil, Europe Union, and the USA are facing a silent syphilis epidemic that affects millions of patients annually (7–10). The 2022 Epidemiological Bulletin of Syphilis reported that, between January 1 and June 30, 2021, 167,523 new cases of AS were identified, followed by 74,095 cases of syphilis in pregnancy (SIP) and 27,019 cases of CS (11). In Brazil, there was on average one new case of AS every 1 min and 40 s, 1 new case of SIP every 4 min and 15 s, and 1 new case of CS every 11 min (12, 13).

Syphilis is diagnosed through serological tests, such as the Venereal Disease Research Laboratory (VDRL), a non-treponemal test. If a non-treponemal test is reactive, a treponemal test, e.g., T. Pallidum hemagglutination assay (TPHA), is performed to confirm the diagnosis. However, serological tests have limitations, such as the time of infection, which may present false-negative results in cases of early or late infections (14).

Adequately treated patients are expected to show significantly reducing non-treponemal antibody titers, but there are cases where titers persist for months to years and may represent a false-positive result when retested (15, 16). Furthermore, it is also possible to observe false-positive results in patients with an autoimmune condition, such as systemic lupus erythematosus, with other infectious diseases, such as brucellosis, or even in pregnancy (15). As syphilis shares several clinical manifestations and clinical characteristics with other treponemal and non-treponemal diseases, a safe clinical diagnosis is necessary, always performed by well-prepared and highly accurate laboratory tests (17).

In parallel, computational methods have been applied in health to aid diagnosis and treatment decisions, including in the diagnosis of STIs, recommendation of adequate treatment, and predictions on the probability of infection (18–21). Predictive analytics is a method for predicting future risks based on current and prior data, assisted often by data mining, machine learning, and novel statistical techniques (22). These techniques are used to develop an inference mechanism, a set of rules that can be applied to a dataset to render a mathematical function that can predict or infer knowledge about that data (19).

Artificial intelligence (AI) has been used to determine characteristics of individuals who are more prone to STIs, such as men who have sex with men (MSM), transgender people, sex workers, those who use stimulants to enhance and prolong sexual experiences (known as chemsex practitioners), and pre-exposure prophylaxis users (PrEP) who do not use condoms (23). For AI systems to be deployed, they need to be trained using data generated from clinical interactions. These data can be collected during clinical activities such as screening, diagnosis, and treatment of patients so that the AI systems can learn the similarities between groups and associations between the characteristics of subjects. This data can also include demographic data, clinical notes of health professionals, electronic records from medical devices, data from physical exams, and laboratory and imaging results. AI includes, among others, machine learning (ML) techniques that analyze structured data, such as images and genetic data, and natural language processing (NLP) that can use and integrate data in various forms, such as text, waveform, and images (24).

Basic ML algorithms can be categorized as supervised and unsupervised. Supervised ML methods work by gathering many training cases, which contain labeled inputs and the desired outputs (25). By analyzing the patterns in all the labeled input-output pairs for new cases, the algorithm learns how to produce the correct output for a given input (26, 27). Unsupervised learning infers the underlying patterns by applying similarity measures to unlabeled data to find subclusters of the original data, identify outliers, or produce low-dimensional representations of the data (24).

Against this background, this systematic literature review (SLR) aims to analyze published studies that use computational methods with the application of AI, ML, or other statistical methods to predict the occurrence of syphilis in critical populations and also identify potential gaps and opportunities for future research on different areas for programmatic response to syphilis, such as management of surveillance and comprehensive care.

This research was developed based on the systematic review guidelines proposed by Kitchenham (28) and the PRISMA checklist (29). Initially, as a fundamental part of the protocol, 3 Research Questions (RQ) were formulated (Table 1).

The process of identifying primary studies related to the research object of this SLR consisted of searches in six repositories: Science Direct, Web of Science, Springer, Scopus, ACM Digital Library, and PubMed. Searches in all databases were performed on August 9, 2022. The following search string (SS01) was used in searches:

After identifying and defining the initial set of records, screening was performed to select a subset of eligible primary studies. This process was organized and executed based on the application of three basic procedures: (i) Inclusion Criteria—IC; (ii) Exclusion Criteria—EC; and (iii) Quality Assessment Criteria—QA.

In the first procedure (i), a subset of primary studies was defined from the IC and applied through the filters available in the repositories. In the subsequent step (ii), a screening guided by the EC based on reading the title, abstract, and keywords was performed on the subset of primary studies. Rayyan (30), a web application for systematic reviews, helped carry out step (ii). The search used two inclusion and three exclusion criteria, as shown in Table 2.

To determine the final set of eligible studies to seek answers to the RQ (Table 1), a screening guided by the QA criteria was carried out from the complete reading of the primary articles (Table 3). An evaluation metric called score was used to qualify and classify the studies (as presented in Equation 1). The score is the arithmetic mean of the weights (w) assigned to each QA criterion. The weight (w), which can vary between 0, 0.5, and 1.0, measures how satisfactory the response of that article is to a specific QA criterion, as shown in Equation (2). The preliminary reports that obtained a score ≥ 0.5 (i.e., 0.5 ≤ score ≤ 1.0) were considered eligible for this SLR.

where:

The scores were assigned by two independent reviewers and elementary data of the final set of eligible studies, extracted based on the RQ, were summarized in Table 4. Studies were included via another method, based on a simple and active search in Science Direct, Springer, and PubMed (Figure 1). This search used the following descriptors: syphilis AND model AND diagnosis.

The quantitative results of the execution of the SLR protocol are presented in Figure 1. After identification and screening, 26 primary studies were selected as eligible and included in this SLR to respond to the RQ (Table 1). Relevant data were extracted from the eligible studies and described in Table 4.

Although completely curable, syphilis is a sexually transmitted infection caused by T. Pallidum, which is responsible for a silent epidemic wave worldwide (3, 7–9, 58, 59). Even though it is relatively easy to diagnose syphilis through routine laboratory methods, the tests available around the world still present problems, mainly because the most qualified tests are difficult to access, especially in poorer countries. Therefore, the application of computational methods can contribute to the development of new, more accessible (point of care), cheaper, and more accurate tests for the diagnosis of syphilis (60).

Previous studies analyzed the sensitivity and specificity of Syphilis Health Check, a rapid qualitative test to detect human antibodies to T. Pallidum (61), or explored the prevalence of syphilis in men who have sex with men (MSM), identifying critical geographic mapping, trends, and data gaps in Latin America and the Caribbean (62). However, in the current paper, we present a systematic review that investigates the application of computational methods as technological tools to support and induce strategies in the context of syphilis. The analysis revealed a diverse set of studies that used computational methods for epidemiological surveillance of syphilis, diagnosis of syphilis, and assessing the impact of public policies.

In this sense, our review shows the utility of computational tools in furthering the general understanding of syphilis which is worsening a global problem, to guide policy and practice to target better public health interventions such as surveillance and prevention, health care service delivery, and the optimal use of diagnostic tools. For instance, Joshi et al. (41) utilized an ARIMA model to investigate the impact of the COVID-19 pandemic on the diagnosis and reporting of STIs, aiming to inform sexual health program planning. The study analyzed New York State STI surveillance data from January 2015 to December 2019 and found that stay-at-home orders contributed to a decline in sexual activity with casual partners, and adversely affected sexual health services, including a reduction in access to diagnostic testing for STIs.

Zhang et al. (38) showed that disease surveillance data could be used to understand syphilis behavior over time using a time-series models. The study revealed a long-term seasonal and increasing trend for the infection, with secondary syphilis showing more significant seasonal fluctuation than other types of the disease. They concluded that patient's likelihood of seeking treatment for secondary syphilis, which is more severe than the other types, was one reason to explain the observed seasonality. Using logistic regression models, Cuffe et al. (34) revealed that several risk factors were associated with a CS case. This finding may potentially support epidemiological surveillance and healthcare services in directing prevention efforts for CS.

Bao et al. (49) demonstrated that it is possible to use ML techniques to predict syphilis infection using datasets that should be available in most settings, such as STIs symptoms, previous syphilis infection, length of residence in the current place, frequency of condom use with casual male sex partners during receptive anal sex, and the number of sex partners.

Dexter et al. (50) alerted to the limitations of predictive models, especially regarding the low generalization power using health data. They cautioned on generalizing the model's performance in the test and validation dataset to general population use. Understanding the descriptors and how to render the model with high generalizability in the test and validation datasets allows the development of reliable models that reach a favorable result within the scope for which it was intended. Algorithmic bias is an important consideration when applying algorithms generated using learning sets and restricted data, as they can further reinforce and augment prevailing inequalities in health systems (63).

There is a need for establishing population-level integrated data sets that are representative, inclusive, and incorporate public health and surveillance data with health service delivery and socio-economic data to improve the utility of AI and ML techniques to strengthen health systems in general and to improve control of syphilis (64, 65). For this disease, there is encouraging development of technological platforms aimed to minimize errors generated by the fragmentation of data used to survey, diagnose and treat syphilis. For example, in Brazil, the Salus Platform Integrates surveillance data with primary health care data and applies ML to improve work processes and response in health crisis scenarios (66, 67). This Platform has also integrated a model of Research on Knowledge, Attitudes, and Practices in the Population into its technological architecture, adapted from the national survey carried out by the Ministry of Health, the Search of Knowledge, Attitudes, and Practices in the Brazilian population (PCAP) (68). With this, it is possible to investigate patient's knowledge, attitudes, and practices related to syphilis, HIV, and other STIs infection.

There are great possibilities with ML to improve and better target surveillance and testing for syphilis and to help inform the development of more efficient and timely diagnostic processes for syphilis and in health surveillance. These developments can help benefit the fight against syphilis, but also other infectious diseases by paving the way for the development of rapid incidence assays to characterize emerging and worsening epidemics (69).

In the context of Brazil, which has the Brazilian National Health System (SUS), with a tripartite governance framework underpinned by a regionalized and hierarchical network of healthcare providers organized according to the complexity of care, behavioral surveys can be carried out when patients seek health care (70, 71). However, for this to happen in SUS, health policies, public health, surveillance, and healthcare service delivery activities need to operate more effectively in an integrated manner (72). Brazil's suboptimal response to COVID-19 has shown the need for better coordination of health policies, public and healthcare delivery, and integrated datasets that can be harnessed for the application of ML methods (73–75).

Results of this study show that analysis using computational techniques could help inform public health and healthcare delivery responses to the worsening syphilis epidemic around the world. But for this to happen, surveillance and policies developed to inform public health and healthcare delivery interventions must be better coordinated. The fact is that health sciences have advanced a lot, particularly with digital health, surpassing the analog world. Therefore, we come from a place where health was more restricted in terms of access to care, as diagnosis methods were only carried out using expensive and difficult-to-access equipment that required super specialists to operate and issue medical reports.

Surveillance actions for STIs such as syphilis, coupled with novel AI-based technologies and tools, contribute toward overcoming the delays in reports drawn on case notification and the shortcomings in current STI data collection. Optimal STI surveillance is contingent on timely and accurate data, yet surveillance data are generally delayed or unavailable (76). In Brazil, which has experienced a syphilis epidemic since 2016 (77), epidemiological reports on syphilis have usually been released belatedly, usually by more than a year. Thus, in this case, how to make decisions that rely only on delayed data that reflects previous scenarios? (78).

Against this background, IA may enhance surveillance, serving as a tool to support decisions about public health interventions in the context of STIs. Therefore, this could provide part of the answer to this public health problem. According to Young et al. (76), available research on STIs has shown that AI can predict syphilis rates at the small-town level by parsing publicly available social media data regarding people's sexual attitudes and behaviors associated with syphilis. This method, known as Rumor analysis, is highly cumbersome through traditional surveillance methods. However, this is not the case when AI-based tools are used because they allow the same analyses to be performed within seconds (79).

Today we are living the transition from this analogical world of health to a fully digital world; the world is experiencing an important process of digital transformation in health. However, for this movement in digital health to be successful and achieve better social results, science must also look at neglected diseases such as syphilis. Advances in health with AI cannot only be used to increase the profits of the health industry; they must also target social inequities and injustices and develop new diagnostic methods to increase access to health for all who need it. This is the way of the future. Using digital health, based on computational methods such as AI, and all its potential to create new diagnostic methods, new tests, and new forms of prevention against STIs, for example, would be a great advance.

Cheaper technologies at the point-of-care that can be operated at distances—telemedicine and telediagnosis—will certainly contribute to reducing inequalities health access, an important contribution to global health (60). Syphilis is a secular disease, but there are indications it is an ancient ailment, rendering senseless the fact it is still a neglected disease by global science. It is necessary to move forward in the present—right now—so that in the future there will be no more children dying from congenital syphilis. This is a very noble goal for science, for health, for digital health and for those who study the application of AI in health.

This article investigates the literature, based on a systematic review protocol, to identify and highlight studies exploring applications based on computational methods or approaches in the context of syphilis. The execution of the SLR protocol yielded 26 primary studies, which were considered eligible. Our findings reveal a substantial diversification of algorithmic models, regardless of the purpose of application, and three subareas of concentration in the field of health sciences: (1) surveillance (16 studies—61.54%) (31–46); (2) diagnosis (nine studies—34.62%) (47–55); and (3) evaluation of health policies (one study—3.85%) (56).

By showing computational models capable of being tools to support STI control and surveillance actions, the studies show promising outcomes. The use of several ML models in the context of syphilis, for example, exhibit a tendency toward consolidation of algorithms for classification and regression tasks. However, there are still ambitious challenges to be explored, such as evaluating the generalization capacity of models considering different global populations, identifying biases in data, and investigating universal access to applications.

A limitation of this review was the impossibility of defining the best predictors for the analysis of syphilis due to the diversity of methods, datasets, and variables used. In addition, the review findings could not establish a technique with good generalizability for the implemented models.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

GA, FF, and RV contributed to conception and design of the study. GA and FF did collection, organizing, and review of the literature. GA, DB, and FF wrote the first draft of the manuscript. GA, FF, DB, IB, LG-L, AS-M, TP, and RV wrote sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.