Burkitt’s lymphoma (BL) is common in sub-Saharan Africa, yet diagnosis is often delayed due to limited pathology capacity. Here we evaluated blood-based liquid biopsies from 377 children and young adults with clinically suspected lymphoma at four hospitals in Tanzania and Uganda, assessing diagnostic accuracy and turnaround time (TAT).
Burkitt’s lymphoma is common in sub-Saharan Africa, yet diagnosis is often delayed due to limited pathology capacity. Here we evaluated blood-based liquid biopsies from 377 children and young adults with clinically suspected lymphoma at four hospitals in Tanzania and Uganda, assessing diagnostic accuracy and turnaround time .
After extensive pathology capacity building, a gold-standard diagnosis was established using tissue morphology, a limited validated immunohistochemistry panel and independent dual histopathologist review. Using clinical features and circulating tumor DNA markers . Diagnostic accuracy, yield and TAT were compared head to head between liquid biopsy and the gold standard in 58 participants. The comprehensive model achieved the highest performance 0.95, 95% confidence interval 0.901–0.981, sensitivity 0.86, specificity 0.95), confirmed by external validation . Liquid biopsy was the only diagnostic result available at the multidisciplinary review in 42% of participants and reduced median diagnostic TAT from 46.8 d to 6.5 d is a common cancer in children, with an aggressive course, but a high cure rate when treatment is initiated early and correctly. It is associated with Epstein–Barr virus infection in 95% of cases in sub-Saharan Africa and shows a strong correlation with malaria endemic areas. To differentiate BL from other non-Hodgkin’s lymphomas, standard-of-care diagnosis requires the integration of morphology, immunophenotype and demonstration of a. However, these methods are not readily available in many low-income countries, where the burden of BL is highest and the availability of timely, reliable pathology services remains an important challenge. In addition to a shortage of histopathologists, access to immunohistochemistry is limited and, even when present, often inconsistently performed due to difficulties maintaining a stable supply of reagents and antibodies. As a result, diagnosis typically relies solely on hematoxylin and eosin morphology, making precise differential diagnosis impossible. Moreover, delays in diagnosis and treatment initiation remain a critical barrier to care, with a median delay of 91 d for BL from first onset of symptoms and a 51% probability of treatment not starting until 90 d after presentation A three-phase diagnostic scoring system, based on a limited IHC panel, has been proposed to support diagnosis in environments where IHC is either unavailable or inconsistently implemented. This scoring system relies on a minimal set of markers that are more accessible in low-resource laboratories. This scoring system, designed for settings with limited pathology infrastructure, demonstrated 81% concordance with SoC BL pathology diagnosis, supporting its use as a practical and reliable reference standard in such contexts.. In principle, targeted sequencing of circulating cell-free DNA extracted from plasma enables identification of genetic aberrations associated with different tumor types. However, for SSA, the application of cfDNA analysis in cancer care is an under-researched topic. Despite a plethora of literature from high-income countries on the utility of this approach for the detection of minimal residual disease and accelerated approvals by the US Food and Drug Administration of some early cancer detection testsWe previously proposed a liquid biopsy diagnostic model as a minimally invasive yet accurate modality for the detection of BL in children and young adults with suspected lymphomaintron 1 single nucleotide variants , tumor fraction of circulating cfDNA and autosomal DNA entropy, were significantly associated with BL diagnosis. When integrated into a penalized logistic regression model, these variables demonstrated strong discriminative performance, achieving an area under the curve of 0.95 for the detection of BL–Ig translocation in only 50% of cases of BL, highlighting the need for additional molecular markers to further improve diagnostic accuracy., it is important to consider whether EBV DNA characteristics, such as fragment size and relative abundance, could meaningfully enhance the performance of diagnostic models. For example, in nasopharyngeal carcinoma, a blood-based screening test combining EBV DNA quantity expressed as the EBV proportion with EBV fragment size ratios, achieved superior diagnostic performance compared to using EBV quantity aloneHere we present the diagnostic accuracy study of this liquid biopsy approach in a larger cohort of children and young adults with suspected lymphoma incorporating multiple molecular attributes, followed by an independent, prospective, real-world evaluation, to provide further evidence of its clinical utility. Overall diagnostic performance was assessed comprehensively based on accuracy , diagnostic yield and TAT, compared to the best local gold-standard pathology as previously describedIn phase I of the study, we enrolled a total of 313 children and young adults clinically suspected of having a diagnosis of lymphoma . A tissue biopsy was collected and processed for a histopathology diagnosis with H&E staining in 89.5% of these patients. For the remaining 33 participants, 8 died before undergoing a tissue biopsy, 5 declined the procedure or were lost to follow-up and 20 had inadequate biopsy samples for histopathological assessment. Among the 280 samples, only 212 achieved a GSP tissue biopsy diagnosis consisting of IHC with a limited paneland review by at least 2 pathologists. The remaining 68 samples were not of adequate quality or sufficient quantity to allow IHC. The failure rate of H&E diagnosis was 10.5% , whereas the failure rate of GSP tissue biopsy was 24.3% . Liquid biopsy was collected from all the 313 participants. However, sequencing was prioritized for samples with a GSP tissue biopsy because this was the gold-standard comparator in the clinical validation. In total, samples from 247 participants were sequenced. Although a total of 4 runs initially failed during the first attempt, all were successfully repeated and sequencing results were subsequently generated for all samples. For clinical validation of the liquid biopsy test, we analyzed data from the 212 samples in this cohort with a GSP tissue biopsy diagnosis 9–17 years), with a male-to-female ratio of 2:1. Eleven participants had a serologically confirmed HIV diagnosis and were on antiretroviral treatment, whereas nine had a history of having been diagnosed and treated for tuberculosis. Only 15 participants reported a positive family history of cancer in their family. A jaw mass was the presenting feature in 25% of participants, peripheral lymph node enlargement was present in 68% and at least half the participants had an abdominal organomegaly. Participants had a median hemoglobin of 9.81 g dl). The most common GSP tissue biopsy diagnosis was classic Hodgkin’s lymphoma , followed by BL and diffuse large B cell lymphoma . Other childhood cancers were present in 9.6% , whereas 9.3% of participants had benign conditions. The demographic, clinical and laboratory characteristics of participants for both phases of the study are summarized in Extended Data TablesAmong the 212 participants, where GSP tissue biopsy diagnosis was achieved, 38.2% had a diagnosis of BL and 61.8% had a non-BL diagnosis. For the non-BL participants, 36.7% had HL, 16.8% had DLBCL, 18.3% had a benign diagnosis and the remaining 28.2% had a mixture of other types of cancer. A younger age, presence of an abdominal or jaw mass and elevated LDH were significantly associated with a diagnosis of BL . For the liquid biopsy variables, diagnosis of BL was associated with higher ctDNA levels, greater median number of mutations incopies per cell, higher EBV DNA fragment size ratio and EBVP, increased EBV DNA fragment size entropy and autosomal fragment size entropy and the presence of alocus occurred within the class II and class III regions. On themutations were present in both BL and non-BL samples, the distribution differed markedly between the two groups . A fourfold difference was also observed in the median VAF between the BL and non-BL samples. The VAF distribution by diagnosis for each sample is shown in Extended Data Fig.): a clinical model comprising clinical parameters only; an EBV quantitative model based on quantitative PCR-derived metrics; an EBV quantitative plus clinical model; an EBV model incorporating additional qualitative EBV features ; a liquid biopsy model including cfDNA-derived variables only; and a comprehensive model combining liquid biopsy and clinical variables. Tenfold crossvalidation showed that all models demonstrated good discriminative ability for BL , with the liquid biopsy model and the comprehensive model showing excellent discriminative abilitywith AUC values of 0.92 and 0.94, respectively. The sensitivity of the models ranged from 0.57 to 0.86 , whereas specificity ranged from 0.78 to 0.95 . Quantification of EBV in copies per cell, which was previously shown to correlate strongly with EBV DNA qPCR, was performed inferiorly to the liquid biopsy model both when clinical parameters were included and when they were not . Overall, the comprehensive model demonstrated the best performance, with an AUC of 0.95, sensitivity of 0.86 and specificity of 0.95 . Among 81 patients with a confirmed local gold-standard diagnosis of BL, 11 were not detected by the liquid biopsy model , whereas 70 were correctly identified . False-negative cases were comparable to true positives with respect to age, sex distribution and LDH levels. However, they were significantly less likely to present with jaw or abdominal tumors regression analysis. Most important predictors for the other models are included in Extended Data Fig.In the phase II arm, we enrolled 64 participants, of whom 1 patient died and another was lost to follow-up before the collection of a tissue or liquid biopsy, leaving 62 patients eligible for analysis. With respect to tissue biopsy, GSP tissue biopsy diagnosis was not obtained for six patients: one patient sample was inadequate for processing and five patient samples were insufficient for IHC. In addition, the liquid biopsy sample for one patient was not sequenced because it hemolyzed, disqualifying it for downstream analysis. Consequently, a GSP tissue biopsy diagnosis was reached for 56 patient tissue biopsy samples, whereas a diagnostic report was generated for 61 patient liquid biopsy samples . We conducted an external validation of our best-performing model on these 56 patients from the phase II cohort after addressing missing values by appropriate imputation methods. The comprehensive model demonstrated excellent discriminative ability, with an AUC of 0.97 , a sensitivity of 0.94 and a specificity of 0.85 . When the analysis was focused on the 44 patients with a complete dataset, the AUC was 0.97 with a sensitivity of 0.93 and a specificity of 0.90 . The heatmap shows agreement between predicted and actual diagnoses and the ROC curve summarizes overall discriminative performance .. Sensitivity and specificity are derived from the confusion matrices and the diagonal line in ROC plots indicates no discrimination. Freq., frequency.). Liquid biopsy was the only test available for making a diagnosis in 42.6% of cases , in the first MDT meeting, 1 was diagnosed using tissue biopsy alone, 6 were diagnosed with both liquid and tissue biopsy available, whereas 8 were diagnosed using liquid biopsy alone. One child with BL was not diagnosed in the first MDT meeting, but was confirmed in subsequent MDT meetings . Six samples were diagnosed with tissue biopsy alone at the first MDT meeting, out of which five were non-BL. This demonstrates that liquid biopsy enabled diagnosis in an additional 53.3% of cases of BL at the first MDT meeting, increasing the overall diagnostic yield of BL to 93.3% , underscoring its potential as a complementary and timely diagnostic tool, especially where tissue biopsy access is limited or delayed. The percentage of occasional decisions taken based on each pathway are denoted on Fig.. In 16.4% of cases, neither the liquid biopsy nor the tissue biopsy was available at the first MDT meeting and diagnosis was therefore made in the subsequent MDT meeting or liquid biopsy at the first MDT meeting. In 16% of cases, diagnosis was established at a subsequent MDT meeting, not shown., Sankey diagram illustrating the availability of liquid biopsy and tissue biopsy results at the first MDT meeting and the corresponding diagnostic outcomes . The median time from sample collection to its receipt at the histopathology lab was 1 d , whereas sample processing to the generation of an H&E slide ready for interpretation took 3.5 d . The median time for H&E slide reporting was 3.7 d , whereas tissue processing for GSP tissue biopsy diagnosis had a median TAT of 43 d . Sequencing had a median duration of 1 d , whereas the time from sequencing completion to bioinformatics analysis and diagnostic report generation had a median of 2 d was 40.3 d earlier than the TAT for tissue biopsy and specificity , likely reflecting the inclusion of other EBV-associated malignancies such as HL and DLBCL rather than healthy controls. Similarly, a recent, large, multi-country study reported high diagnostic accuracy for EBV DNA quantification using digital droplet PCR, but again relied on comparisons with healthy population controls, limiting applicability to real-world clinical settings where multiple EBV-associated malignancies coexist. Together, these findings suggest that, although EBV quantification is a useful screening tool, it lacks sufficient specificity for diagnostic confirmation. Incorporating additional EBV attributes alongside mutation-based and translocation-based markers therefore enhances diagnostic specificity and biological interpretability consistent with observations in EBV-associated nasopharyngeal carcinoma Although acute EBV infection can be associated with elevated EBV copy numbers, integration of both quantitative and qualitative EBV parameters enabled reliable distinction between BL and reactive cases . Notably, model performance was primarily driven by molecular features, rather than clinical parameters , automation and health system integration remain uncertain and may affect turnaround times at scale. Second, study conditions likely benefitted from enhanced oversight and quality control. The 10% tissue biopsy failure rate observed in phase II may underestimate rates in routine care, particularly in resource-limited settings, potentially limiting generalizability. The shorter TAT of liquid biopsy compared with IHC reflects differences in workflow, staffing and infrastructure. In Tanzania, existing molecular skills among laboratory scientists enabled rapid training for cfDNA workflows, supported by automated bioinformatics and clinician-led interpretation using a predefined diagnostic algorithm. In contrast, the longer TAT observed for tissue-based diagnosis reflects constrained histopathology capacity, with few pathologists and prolonged training requirements. Diagnostic delays are largely procedure related, driven by surgical biopsy processing time and delayed IHC interpretation due to heavy workload and occasional repeat biopsies Implementing cfDNA diagnostics in SSA will require alignment with existing clinical pathways. CfDNA testing could complement limited histopathology by providing rapid, minimally invasive diagnosis deployable from peripheral hospitals with centralized sequencing. Leveraging established molecular networks for HIV or tuberculosis testing may facilitate integration. Implementation research, including pragmatic pilots and health-economic analyses, will be essential to address scalability, cost-effectiveness and sustainability adoption within national cancer control programs. Ongoing studies are extending this cfDNA approach to the quantitative detection of minimal residual disease in children receiving therapy for EBVFinally, there are considerable costs associated with next-generation sequencing technology. In a previously published health-economic microcosting analysis conducted by our group, the average per-patient cost of histopathology was estimated at US$185.01, driven primarily by staining and the biopsy procedure . In the same analysis, liquid biopsy cost $710.15 per patient at current throughput, with sequencing reagents representing the largest component. . Although liquid biopsy is more expensive, improved clinical outcomes could justify the higher upfront cost. Ongoing modeling work by our group is evaluating the cost-effectiveness of earlier diagnosis using liquid biopsy, with preliminary findings supporting further evaluation in implementation studies. Adoption of additional sequencing tests for other indications across infectious and noncommunicable diseases using high-throughput sequencing platforms is likely to lead to substantial cost reduction, similar to that observed with HIV viral load or human papillomavirus testing. Ultimately this might even allow the expansion of liquid biopsy testing for BL into primary care settings in endemic regions.This study was approved by the Oxford Tropical Research Ethics Committee , the National Institute of Medical Research in Tanzania and the Uganda National Council of Science and Technology , and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from parents or legal guardians of all participating children, with age-appropriate assent obtained where applicable. Participants did not receive financial compensation for study participation., which was conducted in two phases. In phase I, we trained different liquid biopsy models and assessed their performance for the diagnosis of EBVBL using histopathology with a limited IHC panel and review by a minimum of 2 histopathologists as the GSP in 212 participants with clinically suspected lymphoma. In phase II, we performed a head-to-head comparison between the liquid biopsy and gold-standard tissue biopsy to assess the TAT for the two methods was installed and maintained at all participating laboratories for the duration of the study and 3-monthly internal quality control procedures were implemented, including the use of known positive controls. Tissue samples were fixed in 10% neutral buffered formalin and embedded in paraffin according to standard protocols. H&E staining was performed on all samples to evaluate morphology and IHC was done using the previously described limited IHC antibody panel for BL. This diagnostic approach utilizes a three-phase scoring system. The first phase combines typical BL morphology with immunostains BCL2, CD10 and CD20 to establish a diagnosis. Cases that remain inconclusive proceed to a second phase with CD38, CD44 and Ki67, whereas unresolved cases undergo a third phase with FISH analysis for–Ig translocation. In our study, we used the first-phase immunostains to support the diagnosis of BL, because this phase has previously been shown to achieve 81% concordance with the World Health Organization SoC pathology and is well suited for the limited-resource settings. During the study, we purchased and procured all relevant antibodies and ensured that these were available. Finally, we introduced digital whole-slide imaging to enable slide review by a minimum of two local, study histopathologists and to gauge external third opinions as requiredCfDNA extraction, library preparation and targeted sequencing were performed at two dedicated study laboratories; the Haematology clinical Research Laboratory at the Muhimbili University of Health and Allied Sciences in Tanzania and the Central Public Health Laboratory.Approximately 8 ml of whole blood was collected in Roche circulating cfDNA tubes or PAXgene blood ccfDNA tubes. Plasma was separated by centrifugation at 1,600CfDNA was extracted from plasma using the QIAamp Circulating Nucleic Acid Kit according to the manufacturer’s instructions. Briefly, it involved four main steps; first, samples are lyzed to inactivate DNases and RNases and allow complete release of nucleic acids from bound proteins, lipids and vesicles. Second, the lysates are transferred onto a QIAamp Mini column and circulating nucleic acids are adsorbed from a large volume on to the small silica membrane as the lysate is drawn through by vacuum pressure. Third, while the nucleic acids remain bound to the membrane, contaminants are washed away in a three-step wash process. The final step involves elution of highly purified nucleic acid. The quantification of cfDNA was done with a Qubit fluorimeter 3.0 using the qubit high-sensitivity assay .Libraries were constructed using 50 ng of extracted cfDNA with the ThruPLEX Tag-Seq HV kit according to the manufacturer’s protocol. The process involved addition of unique ThruPLEX HV Unique Dual indexed PCR primers to aid with sample tracking. Seven cycles were performed to ensure a yield of >500 ng depending on the concentration of the input DNA. This was followed by a purification step through magnetic separation using AMPure XP beads . The final library was quantified and validated using the Qubit HS kit and the Bio-analyser High Sensitivity DNA kit, according to the manufacturer’s instructions. Library hybridization and capture were done using the xGen hybridization capture kit according to the manufacturer’s instructions.. The final panel manufactured by Integrated DNA Technologies consisted of 731 probes at a size of 148 kb to permit compatibility with either the iSeq100 or the MiSeq sequencing platforms. The final panel design with the genomic coordinates of our genes of interest is listed in Supplementary Table. Target enrichment was performed using a hybrid capture approach and sequencing was conducted twice weekly on the MiSeq platform using the MiSeq reagent kit v2 at a loading concentration of 10 pM, with 6 samples per run.Panel of normal plasmas: we excluded variants that occurred in ≥2 samples in a separate cohort of 12 healthy controls.Population databases: all variants were annotated against gnomAD and those with a population allele frequency ) >1% were excluded.VAF: to minimize the possibility of including germline variants, we excluded all variants with VAF ≥ 40%, in addition to the initial exclusion of low-level artefacts with VAF< 1%.Sequencing depth and read support: only variants with a minimum sequencing depth of 500× and at least 5 mutant reads were retained for downstream analysis.Copy number alterations : we excluded variants that showed evidence of being affected by CNAs. This was done through an internal normalization approach in which the expected relationship between VAF and log₁₀ was modeled usingSamples that had GSP tissue biopsy results were prioritized for sequencing to allow for clinical validation. For phase II samples, batching was not applied; instead, samples were transported to the sequencing laboratory immediately after collection and sequenced in a head-to-head comparison against GSP tissue biopsy, allowing us to measure TAT prospectively. The same twice-weekly sequencing schedule was maintained, with each MiSeq run limited to a maximum of six samples. Details on the target coordinates in a Health Insurance Portability and Accountability Act of 1996-compliant, Amazon Web Services cloud. Paired reads were combined and statistics generated on the total number of reads and invalid reads, using a customized tool . The raw sequence data in FASTQ format were aligned to GRCh37 with BWA-MEM2. Sorting was done using samtools. De-duplication of reads and generation of error rates and family sizes were done by a custom-made tool . The de-duplicated FASQ was re-aligned to GRCh37 with BWA-MEM2 and fragment sizes calculated using read pairs via a customized script . Another customized script was used to map the reads on to the genomic regions of interest and to calculate copies per cell for the EBV genes . The last base at the end of each read was trimmed using a customized tool and the trimmed SAM file converted into a BAM file and indexed. A customized script was used to generate summary statistics for coverage of all targets at different coverage depths and for different genes. Customized scripts were developed for variant calling using VarScan and annotation of the variants using Ensembl Variant Effect Predictor. IgCaller was used to comprehensively analyze the rearrangements of the Ig genes and identify oncogenic translocations and the Genomic Rearrangement Identification Software Suite, with additional Picards options and Samtools as aligner , was also included as a structural variation caller. To visualize and distinguish true variants from false variants, Integrated Genome Viewer was used. EBV DNA size ratio was calculated as the proportion of EBV DNA fragments with size between 180 bp and 200 bp /). Calculation of the size ratio and selection of the fragment size window were determined based on the methodology of a previous study looking at EBV DNA in nasopharyngeal carcinoma. The lower the EBV size ratio, the lower the proportion of EBV DNA molecules of size 180–200 bp. The distribution of the fragment sizes for reads that map to regions of EBV and for reads that map to the autosomes was calculated and recorded as the EBV entropy and autosome entropy, respectively. This gives a measure of how wide or clustered the distribution of fragment sizes is. The EBV DNA size ratio, EBV entropy and autosome entropy were calculated using customized scripts quantity : calculated as the mean depth of the respective EBV gene divided by the mean depth for all other targeted genes and expressed as copies per cell.EBV fragment size ratio : EBV DNA size ratio was calculated as the proportion of EBV DNA fragments with size between 180 bp and 200 bp /). Calculation of EBVSR and selection of the fragment size window was determined based on the methodology of a previous study looking at EBV DNA in nasopharyngeal carcinoma EBVP: represents the number of EBV DNA reads divided by the total number of sequenced reads , expressed as a proportion EBV DNA and autosomal entropy: calculated as a measure of the diversity of fragment length distributions, using Shannon entropy computed across the full distribution of paired-end sequencing fragment sizes mapping to the EBV genome and the autosomal regions , respectively. Fragment sizes were binned into 20-bp intervals and the proportional frequency of fragments in each bin was used to derive entropy, reflecting the overall variability and dispersion of the fragment size distribution.To further evaluate the potential clinical utility of the liquid biopsy diagnostic test, we established a weekly virtual MDT meeting consisting of pediatric hematologists and oncologists, hematologists, pathologists and laboratory scientists from all study sites, as well as a senior study bioinformatician. When tissue biopsy was reported based on H&E stain only and the liquid biopsy result indicated BL, treatment was initiated as BL . When liquid biopsy indicated non-BL, treatment was deferred until GSP tissue biopsy results were available. If the liquid biopsy report was unavailable at the first MDT meeting, treatment decisions were deferred until GSP tissue biopsy results were available.). When liquid biopsy indicated non-BL, treatment was deferred until the GSP tissue biopsy result was available. If neither tissue biopsy nor liquid biopsy results were available at the first MDT meeting, treatment decisions were deferred until the second MDT meeting in the following week. Our analysis focused on the initial MDT review to capture the immediate outcome of the two tests in the early decision-making process for cases of BL, although enhanced follow-up protocols in phase II ensured that results from both liquid biopsy and tissue biopsy were obtained for all cases and integrated into patient management plans.For the tissue biopsy TAT analysis, we assessed the time from patient presentation at the hospital to the issuance of the GSP diagnostic report, divided into key intervals: time from presentation to sample collection; collection to laboratory receipt; receipt to H&E slide processing; time to issuing the first diagnostic report ; and time to completing the final GSP diagnostic report. For the liquid biopsy TAT, we measured the time from sample receipt in the laboratory to issuance of the diagnostic report, broken down into the intervals of cfDNA extraction and library preparation, targeted sequencing, bioinformatic analysis and final report generation. For the direct TAT comparison between GSP tissue biopsy and liquid biopsy, we specifically analyzed the time from laboratory sample receipt at the respective laboratories to the issuance of the diagnostic report for each method.We assessed the diagnostic yield for BL at the first MDT meeting, expressed as a percentage and calculated from the number of BL cases diagnosed at the first MDT meeting by either the GSP tissue biopsy or the liquid biopsy method divided by the total number of confirmed cases of BL.All statistical analyses were conducted in R v4.2.3. Normality of continuous variables was assessed using the Shapiro–Wilk test and visual inspection of histograms and Q–Q plots. As the data were not normally distributed, nonparametric summaries and tests were used where appropriate. Descriptive statistics were summarized as proportions for categorical variables or medians with IQRs for continuous variables. The Pearson’stest and Wilcoxon’s rank-sum test were used to assess associations between variables. We applied the Benjamini–Hochberg procedure to adjust for multiple comparisons across 19 variables. False recovery rate-adjustedvalues are presented and values<0.05 were considered statistically significant. All variables showing a significant association with the diagnosis of BL were considered for inclusion in a penalized logistic regression model that we previously described and validated using a smaller cohort. Before model fitting, candidate predictors were examined for multicollinearity using pairwise correlation and variance inflation factor analysis and highly collinear variables were excluded to improve model stability and interpretability. Priority was given for those that were most biologically or clinically relevant. Six different models were created based on a combination of different variables, as shown in Table) was selected via tenfold crossvalidation to minimize classification error and ROC curves were generated. The best-performing model was identified based on the highest AUC and pairwise comparisons of AUCs were conducted using DeLong’s test to assess the statistical significance of performance differences between models. The relative importance of variables contributing to the performance of the comprehensive model was derived from the absolute values of the coefficients obtained through LASSO regression. In this penalized logistic regression framework, variables with larger absolute coefficients exert a greater influence on the model’s predictions, reflecting their relative contribution to distinguishing cases of BL from non-BL cases. External validation was done by predicting the diagnosis on a different cohort based on the best-performing model. In this analysis, missing data for LDH in 12 samples was imputed using multiple imputation by chained equations, implemented in the mice R package . Predictive mean matching was applied. Comparison of the median TAT between the liquid biopsy and the tissue biopsy was done using Wilcoxon’s signed-rank test .A formal sample size calculation was conducted to ensure sufficient power for assessing diagnostic accuracy. Using a binomial proportion approach , a minimum of 62 cases with BL was estimated, corresponding to approximately 124 participants, assuming a 1:1 case–control ratio. For multivariable model development, an events-per-variable threshold of 20, with up to 18 candidate predictors, indicated a target sample size of 720 participants. Owing to the prospective design and disruptions related to the COVID-19 pandemic, this target was not reached. The final cohort comprised 212 participants, including 81 cases of BL. To mitigate overfitting in the context of a constrained sample size and multiple candidate predictors, LASSO regression was used for variable selection and regularization. Model performance was internally validated using tenfold crossvalidation to assess robustness and generalizability, in accordance with recommended best practices for predictive model development. Participants were enrolled consecutively as part of routine clinical care. No randomization was performed and investigators were not blinded to allocation during experiments or outcome assessment. Participants with missing outcome data were excluded from the analysis; this exclusion was predefined to preserve the integrity of outcome-based comparisons and was limited to individuals lacking key clinical endpoints or follow-up information required for the primary analyses. For other missing data, including covariates, appropriate imputation methods were applied to minimize bias and retain statistical power. Reproducibility was supported through the use of standardized laboratory protocols, predefined diagnostic algorithms established a priori and automated bioinformatics pipelines applied consistently across all samples. Source data and analysis code are made available as described in ‘Data availability’.This research was conducted through equitable collaboration between institutions in Tanzania, Uganda and the UK, with a deliberate focus on capacity strengthening, technology transfer and shared scientific leadership. NGS platforms were transferred and implemented locally, enabling all cfDNA sequencing to be performed in-country. Laboratory scientists, clinicians and bioinformaticians participated in structured, crosscountry training and reciprocal site visits, supporting hands-on skills development in library preparation, sequencing, bioinformatics analysis and clinical interpretation. Bioinformatics pipelines and reporting frameworks were deployed with local oversight, ensuring full access to data and analytical workflows for all collaborating investigators. The program supported formal academic training, including PhD, DPhil and masters training for local scientists, contributing to sustainable research capacity beyond the duration of the study. Study design, implementation, data interpretation and authorship reflected shared leadership and local ownership.The raw sequencing data and individual-level clinical data generated in this study cannot be made publicly available owing to ethical and data protection restrictions, because they include information from human participants collected under institutional approvals that do not permit open public deposition. Data access is governed by the study consortium and collaborating centers in Tanzania and Uganda. Requests to access the underlying anonymized data will be reviewed by the consortium’s data access committee in consultation with all participating institutions. If a request is deemed scientifically sound and compliant with applicable institutional, national and international data protection regulations, de-identified and anonymized data will be shared after the execution of a data transfer agreement. Processed and anonymized data supporting the findings of this study are available in the GitHub repository: https://github.com/ClaraClaudius/CLINICAL-VALIDATION-OF-LIQUID-BIOPSY-FOR-FASTER-DIAGNOSIS-OF-EBV-POSITIVE-BURKITT-LYMPHOMA.git . This repository is provided to ensure transparency and reproducibility. All data-sharing requests should be addressed to the corresponding author . Data access requests will be reviewed within 8 weeks subject to institutional, national and crossborder regulatory requirements. https://github.com/ClaraClaudius/CLINICAL-VALIDATION-OF-LIQUID-BIOPSY-FOR-FASTER-DIAGNOSIS-OF-EBV-POSITIVE-BURKITT-LYMPHOMA.git . This repository contains fully annotated R scripts used to reproduce all main analyses, figures and tables reported in the paper.Grande, B. M. et al. Genome-wide discovery of somatic coding and noncoding mutations in pediatric endemic and sporadic Burkitt lymphoma.Geser, A. et al. Epstein-Barr virus markers in a series of Burkitt’s lymphomas from the West Nile District, Uganda.Hesseling, P. B. et al. Burkitt lymphoma: the effect of age, sex and delay to diagnosis on treatment completion and outcome of treatment in 934 patients in Cameroon.Mawalla, W. F. et al. Treatment delays in children and young adults with lymphoma: report from an East Africa lymphoma cohort study.Buckle, G. C. et al. Factors influencing time to diagnosis and initiation of treatment of endemic Burkitt lymphoma among children in Uganda and western Kenya: a cross-sectional survey.Antel, K. et al. The determinants and impact of diagnostic delay in lymphoma in a TB and HIV endemic setting.Naresh, K. N. et al. Lymphomas in sub-Saharan Africa—what can we learn and how can we help in improving diagnosis, managing patients and fostering translational research?.Heitzer, E., Haque, I. S., Roberts, C. E. S. & Speicher, M. R. Current and future perspectives of liquid biopsies in genomics-driven oncology.Kurtz, D. M. et al. Enhanced detection of minimal residual disease by targeted sequencing of phased variants in circulating tumor DNA.Si, H. Q. et al. Cancer liquid biopsies by Oxford Nanopore Technologies sequencing of cell-free DNA: from basic research to clinical applications.Chamba, C., Mbulaiteye, S. M., Balandya, E. & Schuh, A. Clinical application of circulating cell-free lymphoma DNA for fast and precise diagnosis of Burkitt lymphoma: precision medicine for sub-Saharan Africa.Cescon, D. W., Bratman, S. V., Chan, S. M. & Siu, L. L. Circulating tumor DNA and liquid biopsy in oncology.Lam, W. K. J. et al. Sequencing-based counting and size profiling of plasma Epstein–Barr virus DNA enhance population screening of nasopharyngeal carcinoma.Carter, J. V., Pan, J., Rai, S. N. & Galandiuk, S. ROC-ing along: evaluation and interpretation of receiver operating characteristic curves.El-Mallawany, N. K. et al. Beyond endemic Burkitt lymphoma: navigating challenges of differentiating childhood lymphoma diagnoses amid limitations in pathology resources in Lilongwe, Malawi.Volesky-Avellaneda, K. et al. Circulating plasma EBV DNA: a potential tool to facilitate diagnosis of pediatric Burkitt lymphoma in sub-Saharan Africa.Raez, L. E. et al. Liquid biopsy versus tissue biopsy to determine front line therapy in metastatic non-small cell lung cancer .Morrell, L. et al. Diagnosing Burkitt lymphoma in sub-Saharan Africa by sequencing of circulating tumor DNA: a comparative microcosting study.Ruhago, G. M., Tungu, M., Morrell, L. & The AL-REAL Consortium. Understanding micro costing a genomics-based diagnostic in a resource-limited setting: experiences and challenges from a Burkitt lymphoma study in sub-Saharan Africa.Stover, J. et al. What is required to end the AIDS epidemic as a public health threat by 2030? The cost and impact of the fast-track approach.Legason, I. D. et al. A protocol to clinically evaluate liquid biopsies as a tool to speed up diagnosis of children and young adults with aggressive infection-related lymphoma in East Africa ‘’.Naresh, K. N. et al. Diagnosis of Burkitt lymphoma using an algorithmic approach—applicable in both resource-poor and resource-rich countries.Mremi, A. et al. Diagnostic validation of a portable whole slide imaging scanner for lymphoma diagnosis in resource-constrained setting: a cross-sectional study.We thank the members of the AI-REAL Scientific Advisory Board—S.M.M., M. Du, R. Siebert, D. Kurtz, C. Oakes, D. Bentley, S. Gopal, D. Lo, K. Naresh and L. Leoncini—for providing their expert guidance throughout the implementation of the study and for reviewing the final draft of the manuscript. We also thank T. Scanlan, O. Henke and K. Schroeder for their mentorship role and guidance throughout the course of this study. This work was supported by the National Institute for Health and Care Research using UK International Development funding from the UK Government to support global health research and support from the Intramural Research Program , National Cancer Institute, National Institutes of Health, US Department of Health and Human Services. The funder had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.Clara Chamba, Heavenlight Christopher, Emmanuel Josephat, William Mawalla, Rehema Shungu, Lulu Chirande, Malale Tungu, Faraja Chiwanga, Emmanuel Balandya & Anna SchuhKieran Howard, Helene Dreau, Adam Burns, Sıla Gerlevik, Claire El Mouden, Anthony Cutts, Liz Morrell, Dimitrios Vavoulis & Anna SchuhThe study was led by A.S., F.C., C.C., M.O. and C.E.M. S.M., G.S., H.M., L.C., P.N., E.M. and I.O. contributed to the collection of clinical data at the different sites. Liquid biopsy wet lab analysis was conducted by P.M., H.C., E.J., J.S., A.A., I.L., I.O., A.B., H.D. and A.C. D.M. and E.M. processed the tissue biopsy samples and histopathology evaluation of samples was done by L.M., A.N., E.E., A.M. and C.A. W.M. managed the data for this study on the redcap database. K.H. performed the bioinformatics analysis and R.S., C.C., S.G., K.H. and D.V. conducted the statistical analysis. M.T. and L.M. contributed to the cost-effectiveness aspect of the study. C.C., A.S., S.M.M., D.V. and E.B. wrote the manuscript with contributions from all authors.thanks Daniel Hodson, Vincent Ribrag and the other, anonymous, reviewer for their contribution to the peer review of this work.Flowchart showing enrolment, sample collection, processing and diagnostic outcomes for 377 participants. In Phase I , tissue biopsy and liquid biopsy samples were collected, processed and sequenced, with generation of gold standard pathology tissue biopsy reports and liquid biopsy diagnostic reports, including losses due to death, loss to follow-up and inadequate sample quality. In Phase II , corresponding tissue and liquid biopsy workflows are shown, including exclusions due to inadequate or hemolyzed samples. Numbers indicate samples at each step. GSP denotes gold standard pathology.Box-and-whisker plots show EBV markers in BL and non-BL cases, including EBER1, EBER2, EBNA2, EBVmax , EBVSR , EBVP , EBV entropy , and autosomal fragment entropy . Centre lines indicate medians; boxes show the interquartile range ; whiskers extend to the minimum and maximum values within 1.5×IQR; points represent individual samples. Group comparisons were performed using two-sided Wilcoxon rank-sum tests, with P values shown for each marker .gene and immunoglobulin loci in Burkitt lymphoma . Percentages indicate the proportion of cases with breakpoints in each annotated region.breakpoints are distributed across exon 1 , intron 1, exon 2 , exon 3 and regions distal to the 3′ end, corresponding to class I–III translocations.variants; the shaded band indicates the expected range. Variants outlined in black fall outside this range and are flagged as copy number alteration –affected .variants per sample, stratified by diagnosis. Each dot represents an individual variant; blue dots indicate sample medians and solid lines denote group medians. Differences in median between groups was assessed with values calculated using a two-sided Wilcoxon rank-sum test. Hodges–Lehmann estimator is shown with 95% confidence intervals. Only samples with detectable variants are shown .intron 1 mutations per sample in BL and non-BL groups. Bars represent individual samples; dashed lines indicate group medians. Only samples withintron 1 mutation counts in non-BL samples stratified by diagnosis, with the accompanying table showing proportions of mutated and non-mutated samples. Source Data are provided as source data 4A, 4B, 4C, 4D and 4E files.Extended Data Fig. 6 Importance of key predictors of Burkitt lymphoma identified by LASSO models, n = 212 samples. Bar plots show predictors ranked by absolute coefficient magnitude from LASSO-regularised logistic regression models. Models include: comprehensive , liquid biopsy only, EBV markers only, EBV quantitative markers only, EBV quantitative plus clinical variables, and clinical variables only. Larger absolute coefficients indicate stronger contributions to classification. EBV, Epstein–Barr virus; EBVP, EBV proportion; EBVmax, maximum EBV copy number; EBER1/2, EBV-encoded RNA 1/2; EBNA2, EBV nuclear antigen 2; EBVSR, EBV size ratio; EBVent, EBV fragment entropy; autoent, autosomal fragment entropy; Tumorsite, anatomical tumour site; SympMon, symptom duration in months; LDH, lactate dehydrogenase.Schematic overview of parallel diagnostic pathways. The liquid biopsy pathway comprises venous blood collection, plasma separation and cfDNA extraction, library preparation, hybridization capture, targeted sequencing, bioinformatic analysis and generation of a diagnostic report. The tissue biopsy pathway comprises excisional biopsy, formalin fixation and paraffin embedding, haematoxylin and eosin staining, immunohistochemistry, histopathology review and diagnostic reporting. Extended Data Table 1 Baseline demographic and clinical presentation of children and young adults with clinically suspected lymphoma ,Extended Data Table 2 Laboratory parameters and final diagnoses of children and young adults with clinically suspected lymphoma ,This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
New Balance’s Futuristic Lifestyle Sneaker Is Getting New Multicolor Styles for SpringNew Balance is giving its futuristic Abzorb 2000 sneaker two new 'Liquid Metal' colorways with gradient fade multicolor looks. Find the release info.
Read more »
Mars Was Once a World of RainMars today is a frozen, barren world where liquid water can briefly appear on its surface but evaporates almost instantly in the thin atmosphere, unable to persist in any meaningful quantity.
Read more »
CyberOne humanoid robot’s hand mimics human sweat glands for thermal managementFull-palm tactile sensing, liquid cooling channels, and high dexterity aims to improve humanoid robot's bionic hands for long operations.
Read more »
Scientists Discover Hidden Behavior in Water That Helps Explain Why Ice FloatsWater exists in two different liquid forms, which might be why it behaves differently than other substances.
Read more »