The healthcare domain is witnessing a remarkable shift with the introduction of “TBorNotTB,” a state-of-the-art clinical decision support system (CDSS) specifically designed for the evaluation of suspected tuberculosis (TB) cases in hospitals. The primary function of TBorNotTB is to aid healthcare professionals in making informed decisions about discontinuing airborne infection isolation (AII) without compromising the detection accuracy or overall care of TB patients. With hospitals grappling with immense pressures on their capacities and staffing, implementing such a sophisticated tool comes at an incredibly opportune moment. By revolutionizing the way TB evaluations are conducted, TBorNotTB promises to alleviate some of the burdens on infection prevention and control (IPC) personnel, making it a vital advancement in modern healthcare settings.
Background and Need for TBorNotTB
Tuberculosis incidence in the United States has seen a dramatic decrease over the past few decades, plummeting from 10.4 cases per 100,000 people in 1992 to 2.2 cases per 100,000 in 2020. Nevertheless, this progress met an unexpected challenge with a rise to 2.9 cases per 100,000 people in 2023. This resurgence has exacerbated the already strained hospital environments that are dealing with capacity and staffing challenges, underscoring the urgent need for rapid and accurate TB case evaluations. Traditional TB evaluation guidelines include testing sputum samples for mycobacterial culture and acid-fast bacilli (AFB) smear while placing suspected cases under AII. However, these methods present significant drawbacks. Cultures can take a minimum of two weeks to generate results, with nearly half of pulmonary TB cases displaying negative smears during this period. Although nucleic acid amplification tests (NAATs) offer better sensitivity compared to AFB smears, they still do not match the accuracy of cultures.
The extended time frame required for obtaining culture results typically necessitates that IPC personnel undergo an additional review of patients’ records before discontinuing AII. This process involves consultations with medical specialists and the primary care team to evaluate the likelihood of TB, which is both labor-intensive and time-consuming. The TBorNotTB system addresses these challenges head-on by expediting the evaluation and decision-making processes to facilitate more efficient and accurate determinations on discontinuing AII.
Development and Methodology
The development of the TBorNotTB system was spearheaded by a panel of experts who employed the Delphi consensus method, which relies on clinical guidelines and epidemiological data to form and refine a set of specific questions. The initial testing phase for these questions was conducted on hospitalized patients at Massachusetts General Hospital (MGH). The CDSS operates by assigning scores based on a myriad of factors such as risk elements, TB symptoms, patient medical history, and results from bronchoscopy or sputum tests, along with chest imaging reports. If a patient’s score surpasses a predefined threshold, the tool automatically recommends the discontinuation of AII; if not, it suggests further evaluation.
The CDSS’s validation process included a case-control analysis using data from the Mass General Brigham (MGB) system. Researchers applied the system retrospectively to patients confirmed with culture-positive TB and compared it against matched culture-negative controls. During this phase, variables that did not significantly enhance predictive accuracy were either removed or adjusted. The final model’s performance underwent fine-tuning by recalibrating the weights assigned to key TB predictors and assessing its sensitivity, specificity, and the area under the curve (AUC).
To ensure the robustness of the tool, the study also included a sub-group analysis to assess the impact of the COVID-19 pandemic on model performance. Besides enhancing predictive accuracy, the study estimated the total IPC person-hours saved annually at MGH due to the CDSS.
Key Findings and Implications
The extensive study behind the TBorNotTB system unveiled several critical predictors of TB, greatly enhancing the efficiency and accuracy of TB evaluations. Chest radiology reports and epidemiologic risk factors were among the leading indicators of active TB. Findings such as cavitary lesions or other suspicious indicators on chest radiology reports were strongly correlated with active TB cases. Additionally, a history of residence in a TB-endemic country emerged as a significant epidemiologic risk factor. The presence of a positive interferon-γ release assay (IGRA) was another substantial predictor of active TB, while a negative IGRA or recent tuberculin skin test exhibited a negative association with TB.
Among traditional TB symptoms, the study found that only a history of weight loss was a significant predictor of active TB. An interesting discovery was that a noticeable improvement or resolution of symptoms following treatment for an alternative diagnosis showed a strong negative association with TB. Initially, the model displayed a high sensitivity of 100% but had a low specificity of 16%. However, incremental revisions over time saw the final model’s specificity improve to 27%, while maintaining its 100% sensitivity and an AUC score of 0.87, denoting solid predictive performance.
Crucially, the model demonstrated consistent performance both before and after the COVID-19 pandemic, reinforcing its reliability in different healthcare contexts. Researchers estimated that TBorNotTB could save over 40 IPC person-hours annually at MGH, highlighting its potential efficacy and efficiency in hospital settings.
Potential Impact on Clinical Practice
The TBorNotTB system was developed by a group of experts using the Delphi consensus method, which uses clinical guidelines and epidemiologic data to develop and refine specific questions. The initial questions were tested on patients at Massachusetts General Hospital (MGH). The clinical decision support system (CDSS) assigns scores based on risk factors, TB symptoms, patient history, bronchoscopy or sputum test results, and chest imaging reports. If a patient’s score surpasses a certain threshold, the tool recommends ending airborne infection isolation (AII); otherwise, it suggests further evaluation.
To validate the CDSS, a case-control analysis was performed using data from the Mass General Brigham (MGB) system. Researchers applied the tool retrospectively to patients with culture-positive TB and compared it to matched culture-negative controls. Variables that didn’t significantly enhance predictive accuracy were removed or adjusted. The final model’s performance was fine-tuned by recalibrating the weights of key TB predictors and assessing sensitivity, specificity, and the area under the curve (AUC).
A sub-group analysis was included to evaluate the COVID-19 pandemic’s impact on model performance. The study also estimated the total IPC person-hours saved annually at MGH due to the CDSS, which bolstered predictive accuracy.
