Search

Abstract Background Efforts toward tuberculosis management and control are challenged by the emergence of Mycobacterium tuberculosis (MTB) resistance to existing anti-TB drugs. This study aimed to explore the potential of machine learning algorithms in predicting drug resistance of four anti-TB drugs (rifampicin, isoniazid, streptomycin, and ethambutol) in MTB using whole-genome sequence and clinical data from Uganda. We also assessed the model’s generalizability on another dataset from South Africa. Results We trained ten machine learning algorithms on a dataset comprising of 182 MTB isolates with clinical data variables (age, sex, HIV status) and SNP mutations across the entire genome as predictor variables and phenotypic drug-susceptibility data for the four drugs as the outcome variable. Model performance varied across the four anti-TB drugs after a five-fold cross validation. The best model was selected considering the highest Mathews Correlation Coefficient (MCC) and Area Under the Receiver Operating Characteristic Curve (AUC) score as key metrics. The Logistic regression excelled in predicting rifampicin resistance (MCC: 0.83 (95% confidence intervals (CI) 0.73–0.86) and AUC: 0.96 (95% CI 0.95–0.98) and streptomycin (MCC: 0.44 (95% CI 0.27–0.58) and AUC: 0.80 (95% CI 0.74–0.82), Extreme Gradient Boosting (XGBoost) for ethambutol (MCC: 0.65 (95% CI 0.54–0.74) and AUC: 0.90 (95% CI 0.83–0.96) and Gradient Boosting (GBC) for isoniazid (MCC: 0.69 (95% CI 0.61–0.78) and AUC: 0.91 (95% CI 0.88–0.96). The best performing model per drug was only trained on the SNP dataset after excluding the clinical data variables because intergrating them with SNP mutations showed a marginal improvement in the model’s performance. Despite the high MCC (0.18 to 0.72) and AUC (0.66 to 0.95) scores for all the best models with the Uganda test dataset, LR model for rifampicin and streptomycin didn’t generalize with the South Africa dataset compared to the GBC and XGBoost models. Compared to TB profiler, LR for RIF was very sensitive and the GBC for INH and XGBoost for EMB were very specific on the Uganda dataset. TB profiler outperformed all the best models on the South Africa dataset. We identified key mutations associated with drug resistance for these antibiotics. HIV status was also identified among the top significant features in predicting drug resistance. Conclusion Leveraging machine learning applications in predicting antimicrobial resistance represents a promising avenue in addressing the global health challenge posed by antimicrobial resistance. This work demonstrates that integration of diverse data types such as genomic and clinical data could improve resistance predictions while using machine learning algorithms, support robust surveillance systems and also inform targeted interventions to curb the rising threat of antimicrobial resistance.

BackgroundIt is projected that drug-resistant infections will lead to 10 million deaths annually by 2050 if left unabated. Despite this threat, surveillance data from resource-limited settings are scarce and often lack antimicrobial resistance (AMR)–related clinical outcomes and economic burden. We aim to build an AMR and antimicrobial use (AMU) data warehouse, describe the trends of resistance and antibiotic use, determine the economic burden of AMR in Uganda, and develop a machine learning algorithm to predict AMR-related clinical outcomes. ObjectiveThe overall objective of the study is to use data-driven approaches to optimize antibiotic use and combat antimicrobial-resistant infections in Uganda. We aim to (1) build a dynamic AMR and antimicrobial use and consumption (AMUC) data warehouse to support research in AMR and AMUC to inform AMR-related interventions and public health policy, (2) evaluate the trends in AMR and antibiotic use based on annual antibiotic and point prevalence survey data collected at 9 regional referral hospitals over a 5-year period, (3) develop a machine learning model to predict the clinical outcomes of patients with bacterial infectious syndromes due to drug-resistant pathogens, and (4) estimate the annual economic burden of AMR in Uganda using the cost-of-illness approach. MethodsWe will conduct a study involving data curation, machine learning–based modeling, and cost-of-illness analysis using AMR and AMU data abstracted from procurement, human resources, and clinical records of patients with bacterial infectious syndromes at 9 regional referral hospitals in Uganda collected between 2018 and 2026. We will use data curation procedures, FLAIR (Findable, Linkable, Accessible, Interactable and Repeatable) principles, and role-based access control to build a robust and dynamic AMR and AMU data warehouse. We will also apply machine learning algorithms to model AMR-related clinical outcomes, advanced statistical analysis to study AMR and AMU trends, and cost-of-illness analysis to determine the AMR-related economic burden. ResultsThe study received funding from the Wellcome Trust through the Centers for Antimicrobial Optimisation Network (CAMO-Net) in April 2023. As of October 28, 2024, we completed data warehouse development, which is now under testing; completed data curation of the historical Fleming Fund surveillance data (2020-2023); and collected retrospective AMR records for 599 patients that contained clinical outcomes and cost-of-illness economic burden data across 9 surveillance sites for objectives 3 and 4, respectively. ConclusionsThe data warehouse will promote access to rich and interlinked AMR and AMU data sets to answer AMR program and research questions using a wide evidence base. The AMR-related clinical outcomes model and cost data will facilitate improvement in the clinical management of AMR patients and guide resource allocation to support AMR surveillance and interventions. International Registered Report Identifier (IRRID)PRR1-10.2196/58116

Abstract Background Antimicrobial resistance (AMR) remains a significant global health threat particularly impacting low- and middle-income countries (LMICs). These regions often grapple with limited healthcare resources and access to advanced diagnostic tools. Consequently, there is a pressing need for innovative approaches that can enhance AMR surveillance and management. Machine learning (ML) though underutilized in these settings, presents a promising avenue. This study leverages ML models trained on whole-genome sequencing data from England, where such data is more readily available, to predict AMR in E. coli, targeting key antibiotics such as ciprofloxacin, ampicillin, and cefotaxime. A crucial part of our work involved the validation of these models using an independent dataset from Africa, specifically from Uganda, Nigeria, and Tanzania, to ascertain their applicability and effectiveness in LMICs. Results Model performance varied across antibiotics. The Support Vector Machine excelled in predicting ciprofloxacin resistance (87% accuracy, F1 Score: 0.57), Light Gradient Boosting Machine for cefotaxime (92% accuracy, F1 Score: 0.42), and Gradient Boosting for ampicillin (58% accuracy, F1 Score: 0.66). In validation with data from Africa, Logistic Regression showed high accuracy for ampicillin (94%, F1 Score: 0.97), while Random Forest and Light Gradient Boosting Machine were effective for ciprofloxacin (50% accuracy, F1 Score: 0.56) and cefotaxime (45% accuracy, F1 Score:0.54), respectively. Key mutations associated with AMR were identified for these antibiotics. Conclusion As the threat of AMR continues to rise, the successful application of these models, particularly on genomic datasets from LMICs, signals a promising avenue for improving AMR prediction to support large AMR surveillance programs. This work thus not only expands our current understanding of the genetic underpinnings of AMR but also provides a robust methodological framework that can guide future research and applications in the fight against AMR.

Abstract Genetic characterisation of circulating influenza viruses directs annual vaccine strain selection and mitigation of infection spread. We used next-generation sequencing to locally generate whole genomes from 116 A(H1N1)pdm09 and 118 A(H3N2) positive patient swabs collected across Uganda between 2010 and 2018. We recovered sequences from 92% (215/234) of the swabs, 90% (193/215) of which were whole genomes. The newly-generated sequences were genetically and phylogenetically compared to the WHO-recommended vaccines and other Africa strains sampled since 1994. Uganda strain hemagglutinin (n = 206), neuraminidase (n = 207), and matrix protein (MP, n = 213) sequences had 95.23–99.65%, 95.31–99.79%, and 95.46–100% amino acid similarity to the 2010–2020 season vaccines, respectively, with several mutated hemagglutinin antigenic, receptor binding, and N-linked glycosylation sites. Uganda influenza type-A virus strains sequenced before 2016 clustered uniquely while later strains mixed with other Africa and global strains. We are the first to report novel A(H1N1)pdm09 subclades 6B.1A.3, 6B.1A.5(a,b), and 6B.1A.6 (± T120A) that circulated in Eastern, Western, and Southern Africa in 2017–2019. Africa forms part of the global influenza ecology with high viral genetic diversity, progressive antigenic drift, and local transmissions. For a continent with inadequate health resources and where social distancing is unsustainable, vaccination is the best option. Hence, African stakeholders should prioritise routine genome sequencing and analysis to direct vaccine selection and virus control.

Self-Medication (SM) involves the utilization of medicines to treat self-recognized symptoms or diseases without consultation and the irrational use of over-the-counter drugs. During the COVID-19 pandemic, the lack of definitive treatment led to increased SM. We aimed to estimate the extent of SM for drugs used to treat COVID-19 symptoms by collecting data from pharmacy sale records. The study was conducted in Kampala, Uganda, where we extracted data from community pharmacies with functional Electronic Health Records between January 2018 and October 2021 to enable a comparison of pre-and post-COVID-19. The data included the number of clients purchasing the following drugs used to treat COVID-19 and its symptoms: Antibiotics included Azithromycin, Erythromycin, and Ciprofloxacin; Supplements included Zinc and vitamin C, while Corticosteroids included dexamethasone. A negative binomial model was used to estimate the incident rate ratios for each drug to compare the effect of COVID-19 on SM. In the pre- COVID-19 period (1st January 2018 to 11th March 2020), 19,285 customers purchased antibiotics which included; Azithromycin (n = 6077), Ciprofloxacin (n = 6066) and Erythromycin (n = 997); health supplements including Vitamin C (430) and Zinc (n = 138); and Corticosteroid including Dexamethasone (n = 5577). During the COVID-19 pandemic (from 15th March 2020 to the data extraction date in October 2021), we observed a 99% increase in clients purchasing the same drugs. The number of clients purchasing Azithromycin increased by 19.7% to 279, Ciprofloxacin reduced by 58.8% to 96 clients, and those buying Erythromycin similarly reduced by 35.8% to 492 clients. In comparison, there were increases of 170%, 181%, and 377% for Vitamin C, Zinc, and Dexamethasone, respectively. The COVID-19 pandemic underscored the extent of SM in Uganda. We recommend future studies with a representation of data from pharmacies located in rural and urban areas to further study pandemics' effect on antimicrobials prescriptions, including obtaining pharmacists' perspectives using mixed methods approaches.

Understanding treatment-seeking behavior is critical to the treatment and control of sexually transmitted infections (STIs), yet current data on STI treatment seeking in low-resource settings is rare. This population-based study aimed to describe STI treatment-seeking behavior and identify factors associated with seeking treatment at a clinic among adults with STI-related symptoms in rural Uganda. The STI prevalence study (STIPS) conducted a survey and STI testing among all consenting adults aged 18-49 in two communities in rural south-central Uganda. Of 1,825 participants, 962 individuals self-reported STI symptoms in the past six months; we present descriptive data on treatment seeking and STI prevalence among these individuals. We used multivariable Poisson regressions with robust variance to determine the sociodemographic and symptom-related factors independently associated with seeking STI treatment at a clinic and assessed the association with previous clinic treatment seeking and current STI diagnosis. Forty-three percent of adults who reported STI-related symptoms in the past six months said they did not seek any treatment. Among those who did, 58% sought treatment at a private clinic, 28% at a government clinic, 9% at a pharmacy/drug store, 3% at a traditional healer, 2% at a market/shop, and 5% at another location. Among both males and females, having multiple STI related symptoms was positively associated with clinic treatment seeking (males = PRR: 1.73, 95%CI: 1.36-2.21; females = PR: 1.41, 95%CI: 1.12-1.78). Approximately one-third of males and females who reported previously seeking clinic treatment for their symptoms were diagnosed with a curable STI at the time of the survey. In this setting, nearly half of adults with STI-related symptoms are not seeking clinical care and many who report having sought treatment for recent STI symptoms have curable STIs. Future studies should explore barriers to care-seeking and strategies to improve STI services.

Abstract Background Host population structure is a key determinant of pathogen and infectious disease transmission patterns. Pathogen phylogenetic trees are useful tools to reveal the population structure underlying an epidemic. Determining whether a population is structured or not is useful in informing the type of phylogenetic methods to be used in a given study. We employ tree statistics derived from phylogenetic trees and machine learning classification techniques to reveal an underlying population structure. Results In this paper, we simulate phylogenetic trees from both structured and non-structured host populations. We compute eight statistics for the simulated trees, which are: the number of cherries; Sackin, Colless and total cophenetic indices; ladder length; maximum depth; maximum width, and width-to-depth ratio. Based on the estimated tree statistics, we classify the simulated trees as from either a non-structured or a structured population using the decision tree (DT), K-nearest neighbor (KNN) and support vector machine (SVM). We incorporate the basic reproductive number ( $$R_0$$ R 0 ) in our tree simulation procedure. Sensitivity analysis is done to investigate whether the classifiers are robust to different choice of model parameters and to size of trees. Cross-validated results for area under the curve (AUC) for receiver operating characteristic (ROC) curves yield mean values of over 0.9 for most of the classification models. Conclusions Our classification procedure distinguishes well between trees from structured and non-structured populations using the classifiers, the two-sample Kolmogorov-Smirnov, Cucconi and Podgor-Gastwirth tests and the box plots. SVM models were more robust to changes in model parameters and tree size compared to KNN and DT classifiers. Our classification procedure was applied to real -world data and the structured population was revealed with high accuracy of $$92.3\%$$ 92.3 % using SVM-polynomial classifier.

Abstract Background Globally, tuberculosis disease (TB) is more common among males than females. Recent research proposes that differences in social mixing by sex could alter infection patterns in TB. We examine evidence for two mechanisms by which social-mixing could increase men’s contact rates with TB cases. First, men could be positioned in social networks such that they contact more people or social groups. Second, preferential mixing by sex could prime men to have more exposure to TB cases. Methods We compared the networks of male and female TB cases and healthy matched controls living in Kampala, Uganda. Specifically, we estimated their positions in social networks (network distance to TB cases, degree, betweenness, and closeness) and assortativity patterns (mixing with adult men, women, and children inside and outside the household). Results The observed network consisted of 11,840 individuals. There were few differences in estimates of node position by sex. We found distinct mixing patterns by sex and TB disease status including that TB cases have proportionally more adult male contacts and fewer contacts with children. Conclusions This analysis used a network approach to study how social mixing patterns are associated with TB disease. Understanding these mechanisms may have implications for designing targeted intervention strategies in high-burden populations.

Abstract Tuberculosis (TB) is the leading cause of death in humans by a single infectious agent worldwide with approximately two billion humans latently infected with the bacterium Mycobacterium tuberculosis. Currently, the accepted method for controlling the disease is Tuberculosis Directly Observed Treatment Shortcourse (TB‐DOTS). This program is not preventative and individuals may transmit disease before diagnosis, thus better understanding of disease transmission is essential. Using whole‐genome sequencing and single nucleotide polymorphism analysis, we analyzed genomes of 145 M. tuberculosis clinical isolates from active TB cases from the Rubaga Division of Kampala, Uganda. We established that these isolates grouped into M. tuberculosis complex (MTBC) lineages 1, 2, 3, and 4, with the most isolates grouping into lineage 4. Possible transmission pairs containing ≤12 SNPs were identified in lineages 1, 3, and 4 with the prevailing transmission in lineages 3 and 4. Furthermore, investigating DNA codon changes as a result of specific SNPs in prominent virulence genes including plcA and plcB could indicate potentially important modifications in protein function. Incorporating this analysis with corresponding epidemiological data may provide a blueprint for the integration of public health interventions to decrease TB transmission in a region.

Genomic characterization of circulating influenza type-A viruses (IAVs) directs the selection of appropriate vaccine formulations and early detection of potentially pandemic virus strains. However, longitudinal data on the genomic evolution and transmission of IAVs in Africa are scarce, limiting Africa’s benefits from potential influenza control strategies. We searched seven databases: African Journals Online, Embase, Global Health, Google Scholar, PubMed, Scopus, and Web of Science according to the PRISMA guidelines for studies that sequenced and/or genomically characterized Africa IAVs. Our review highlights the emergence and diversification of IAVs in Africa since 1993. Circulating strains continuously acquired new amino acid substitutions at the major antigenic and potential N-linked glycosylation sites in their hemagglutinin proteins, which dramatically affected vaccine protectiveness. Africa IAVs phylogenetically mixed with global strains forming strong temporal and geographical evolution structures. Phylogeographic analyses confirmed that viral migration into Africa from abroad, especially South Asia, Europe, and North America, and extensive local viral mixing sustained the genomic diversity, antigenic drift, and persistence of IAVs in Africa. However, the role of reassortment and zoonosis remains unknown. Interestingly, we observed substitutions and clades and persistent viral lineages unique to Africa. Therefore, Africa’s contribution to the global influenza ecology may be understated. Our results were geographically biased, with data from 63% (34/54) of African countries. Thus, there is a need to expand influenza surveillance across Africa and prioritize routine whole-genome sequencing and genomic analysis to detect new strains early for effective viral control.

Helminth infection may affect vaccine immunogenicity and efficacy. Adolescents, a target population for tuberculosis booster vaccines, often have a high helminth burden. We investigated effects of Schistosoma mansoni (Sm) on the immunogenicity and safety of MVA85A, a model candidate tuberculosis vaccine, in BCG-vaccinated Ugandan adolescents.In this phase II open label trial we enrolled 36 healthy, previously BCG-vaccinated adolescents, 18 with no helminth infection detected, 18 with Sm only. The primary outcome was immunogenicity measured by Ag85A-specific interferon gamma ELISpot assay. Tuberculosis and schistosome-specific responses were also assessed by whole-blood stimulation and multiplex cytokine assay, and by antibody ELISAs.Ag85A-specific cellular responses increased significantly following immunisation but with no differences between the two groups. Sm infection was associated with higher pre-immunisation Ag85A-specific IgG4 but with no change in antibody levels following immunisation. There were no serious adverse events. Most reactogenicity events were of mild or moderate severity and resolved quickly.The significant Ag85A-specific T cell responses and lack of difference between Sm-infected and uninfected participants is encouraging for tuberculosis vaccine development. The implications of pre-existing Ag85A-specific IgG4 antibodies for protective immunity against tuberculosis among those infected with Sm are not known. MVA85A was safe in this population.ClinicalTrials.gov NCT02178748.

HIV viral load is recommended for monitoring antiretroviral treatment and identifying treatment failure. We assessed the durability of viral suppression after viral load-triggered adherence counseling among patients with HIV viremia 6 months after ART initiation.Observational cohort enrolled in an antiretroviral treatment program in rural Uganda.Participants who underwent routine viral load determination every 24 weeks and had at least 48 weeks of follow-up were included in this analysis. Patients with viral loads >400 copies/ml at 24 weeks of treatment were given additional adherence counseling, and all patients were followed to assess the duration of viral suppression and development of virologic failure.1,841 participants initiating antiretroviral therapy were enrolled in the Rakai Health Sciences Program between June 2005 and June 2011 and were followed with viral load monitoring every 24 weeks. 148 (8%) of patients did not achieve viral suppression at 24 weeks and were given additional adherence counseling. 85 (60%) of these patients had undetectable viral loads at 48 weeks, with a median duration of viral suppression of 240 weeks (IQR 193-288 weeks). Failure to achieve an undetectable viral load at 48 weeks was associated with age 2,000 copies/ml in multivariate logistic regression analysis.The majority of patients with persistent viremia who were provided adherence counseling achieved robust viral suppression for a median 4.6 years. Access to virologic monitoring and adherence counseling is a priority in resource-limited settings.