PREreview of The long-term risk of tuberculosis among individuals with Xpert Ultra “trace” screening results: a longitudinal follow-up study

The long-term risk of tuberculosis among individuals with Xpert Ultra “trace” screening results: a longitudinal follow-up study 

1. Summary and Overall Impression

Recommendation: Minor revisions are required. 

Tuberculosis (TB) remains the leading single-agent infectious cause of death globally, and early detection could reduce transmission and improve outcomes, especially in high-burden settings. The longitudinal cohort study by Sung et al. investigates the long-term risk of TB among individuals identified with Xpert MTB/RIF Ultra “trace” sputum results through community-wide screening in Kampala, Uganda. “Trace” results indicate extremely low levels of Mtb DNA and often go microbiologically unconfirmed, but can be biologically relevant and indicative of a need to consider empiric treatment. Participants with trace-positive results underwent extensive baseline evaluation, including molecular testing, culture, HIV testing, chest X-rays (CXR), and chest CTs. Individuals without a microbiological confirmation or clinical diagnosis of TB at baseline were prospectively followed for up to two years with serial microbiological and radiographic assessments. Among 129 individuals with trace-positive screening results, 35% were diagnosed with TB at baseline. Of the 76 individuals not referred to treatment at baseline, 26% subsequently developed TB over two years, yielding a cumulative hazard of 35% (95% CI: 19-52%). This was markedly higher than the 2% observed in Ultra-negative controls. Incident TB risk was strongly associated with abnormal CXR at baseline (HR 15.0, 95% CI: 3.4-65.1) but showed no correlation with symptom status, HIV infection, sex, or prior TB history. Computer-aided detection (CAD) of baseline CXR showed predictive performance (AUC 0.78-0.85) for identifying individuals who later progressed to TB, particularly in those without prior history of TB. The findings challenge the perception that Ultra trace results represent false positives in active case finding or community screening contexts, but methodological weaknesses must be addressed prior to publication. Sung et al. present compelling evidence of an increased risk of progression to active TB in trace-positive people, particularly when combining trace results with CXR. The authors advocate for consideration of empiric TB treatment initiation or preventative therapy for individuals identified as “trace”-positive during screening, but the recommendation to treat most trace-positive individuals requires further validation through future studies.

1. Section-by-Section Review

Methods

Subheading: Symptom-Neutral Tuberculosis Screening and Participant Recruitment

Major issues: 

1. The manuscript mentions that negative controls were age- and sex-matched, but it lacks crucial details on the matching procedure. We recommend that the authors provide complete details of the matching methodology, including whether individual or frequency matching was used, the matching ratio, and the algorithm for selecting matches. Describe any post-matching statistical adjustments performed to account for residual imbalances in known TB risk factors. Address how matching on only age and sex might impact the validity of between-group comparisons. Consider performing sensitivity analyses using propensity score methods to account for potential selection bias. 

Subheading: Evaluation and Follow-Up of Study Participants

Major issues:

1. The manuscript should clarify blinding procedures for the physician panel making treatment recommendations, which constitute the primary outcome measure. While radiologists were explicitly described as "blinded to clinical information," there is no mention of whether the physician panel members were blinded to the initial Ultra trace result when determining if a participant should receive TB treatment. If consultants were not blinded, acknowledge this as a limitation and discuss how it might have influenced treatment decisions.

2. The implementation of substantially different follow-up protocols between trace-positive participants and controls warrants careful consideration, as it introduces a risk of surveillance bias. Trace-positive participants underwent comprehensive evaluations at 1, 3, 6, 12, and 24 months, while negative controls received only "symptom assessment at 6 months and repeat sputum testing at 12 and 24 months." This differential intensity of surveillance might artificially inflate the relative incidence in the trace-positive group by detecting cases that would have remained undiagnosed with less intensive monitoring. The ascertainment of the outcome (tuberculosis diagnosis) depends partly on the intensity of follow-up, which differed systematically between exposure groups. This methodological issue is particularly concerning for the primary analysis comparing the cumulative incidence between groups. Please acknowledge the differential follow-up as a study limitation and discuss its potential impact on the hazard ratio estimates. Consider performing a sensitivity analysis restricting outcome ascertainment in trace-positive participants to the same timepoints used for controls (6, 12, and 24 months). Discuss how the effect size might be adjusted to account for this surveillance bias.

Minor issues:

1. The study would benefit from enhanced clarity regarding the standardized criteria used by the physician panel for making treatment recommendations. Explicitly describe the criteria or decision algorithm that guided treatment recommendations, including how consultants weighed different types of evidence (clinical, radiological, microbiological). 

Subheading: Statistical analysis

Major issues:

1. The statistical approach appears to lack multivariable adjustment to control for the confounding factors that were measured. While the authors report hazard ratios for individual risk factors (e.g., abnormal CXR, HR 15.0), this univariate approach fails to isolate the independent effect of trace-positive status from other measured risk factors. Conduct and report multivariable Cox regression analyses, adjusting simultaneously for all measured potential confounders. Perform formal interaction testing for effect modification by baseline CXR status and other key variables. Consider presenting stratified hazard ratios for trace-positive status across different subgroups (e.g., normal vs. abnormal CXR). 

Results

Major issues:

1. Line 194 and Figure 2 report that n=129 participants with trace-positive sputum (PWTS), which is misaligned with n=121 in Table 1, as well as the header of Table 2, which indicates "Trace Ultra (n=123)". The subgroups of the “Trace Ultra (n=123)” sum to 117, not 123. If these discrepancies are due to attrition, please indicate this (e.g., why n=123 in Table 2 differs from n=129 elsewhere). 

2. The study identifies a strong effect modification by baseline CXR status (HR 15.0, 95% CI 3.4-65.1) but would benefit from incorporating this finding into a more comprehensive analytic framework. This substantial interaction suggests that the value of trace-positive results varies dramatically depending on CXR findings, which has critical implications for clinical decision-making. Without any formal interaction testing, it remains unclear whether the broad recommendation to treat most individuals with trace-positive results is appropriate for all subgroups, particularly those with normal chest imaging. Please conduct formal statistical testing for interaction between trace-positive status and CXR findings. It may be helpful to present stratified hazard ratios and cumulative incidence curves for trace-positive participants with normal versus abnormal CXR. 

Minor issues:

1. Consider adding statistical significance indicators to Figure 4. It is difficult for the reader to interpret which differences between groups are statistically meaningful after Bonferroni correction.

2. The manuscript emphasizes that “a CAD score threshold of 0.5 resulted in a sensitivity of 75% and a specificity of 64%, and a threshold of 0.2 had a sensitivity of 86% and specificity of 45%.” Figure 5 could be strengthened by visually denoting the specific threshold points (0.5 and 0.2) that are discussed in the text. 

Discussion

Minor issues:

1. The study's interpretation of "incident tuberculosis" would benefit from a more nuanced discussion of whether these cases represent true progression of early disease or delayed detection of prevalent tuberculosis that was below detection thresholds at baseline. While the authors frame trace results as potentially representing early tuberculosis that could progress over time, the study design cannot definitively distinguish between early TB that progresses, missed prevalent TB that was below the detection threshold, and reinfection during follow-up. Without molecular methods to confirm whether detected TB progression represents the same infection as detected by the initial trace result, alternative explanations for the observed "incident" cases remain plausible. Acknowledge the limitations in distinguishing true progression from delayed detection of prevalent disease.

Competing interests

The authors declare that they have no competing interests.

Use of Artificial Intelligence (AI)

The authors declare that they used generative AI to come up with new ideas for their review.