A process mining- deep learning approach to predict survival in a cohort of hospitalized COVID-19 patients

Background

Various machine learning and artificial intelligence methods have been used to predict outcomes of hospitalized COVID-19 patients. However, process mining has not yet been used for COVID-19 prediction. We developed a process mining/deep learning approach to predict mortality among COVID-19 patients and updated the prediction in 6-h intervals during the first 72 h after hospital admission.

Methods

The process mining/deep learning model produced temporal information related to the variables and incorporated demographic and clinical data to predict mortality. The mortality prediction was updated in 6-h intervals during the first 72 h after hospital admission. Moreover, the performance of the model was compared with published and self-developed traditional machine learning models that did not use time as a variable. The performance was compared using the Area Under the Receiver Operator Curve (AUROC), accuracy, sensitivity, and specificity.

Results

The proposed process mining/deep learning model outperformed the comparison models in almost all time intervals with a robust AUROC above 80% on a dataset that was imbalanced.

Conclusions

Our proposed process mining/deep learning model performed significantly better than commonly used machine learning approaches that ignore time information. Thus, time information should be incorporated in models to predict outcomes more accurately.

Number needed to screen for TB in clinical, structural or occupational risk groups

BACKGROUND: Screening for active TB using active case-finding (ACF) may reduce TB incidence, prevalence, and mortality; however, yield of ACF interventions varies substantially across populations. We systematically reviewed studies reporting on ACF to calculate the number needed to screen (NNS) for groups at high risk for TB.METHODS: We conducted a literature search for studies reporting ACF for adults published between November 2010 and February 2020. We determined active TB prevalence detected through various screening strategies and calculated crude NNS for - TB confirmed using culture or Xpert^® MTB/RIF, and weighted mean NNS stratified by screening strategy, risk group, and country-level TB incidence.RESULTS: We screened 27,223 abstracts; 90 studies were included (41 in low/moderate and 49 in medium/high TB incidence settings). High-risk groups included inpatients, outpatients, people living with diabetes (PLWD), migrants, prison inmates, persons experiencing homelessness (PEH), healthcare workers, and miners. Screening strategies included symptom-based screening, chest X-ray and Xpert testing. NNS varied widely across and within incidence settings based on risk groups and screening methods. Screening tools with higher sensitivity (e.g., Xpert, CXR) were associated with lower NNS estimates.CONCLUSIONS: NNS for ACF strategies varies substantially between adult risk groups. Specific interventions should be tailored based on local epidemiology and costs.

A systematic review of the number needed to screen for active TB among people living with HIV

BACKGROUND: Systematic screening for active TB is recommended for all people living with HIV (PLWH); however, case detection remains poor globally. We investigated the yield of active case finding (ACF) by calculating the number needed to screen (NNS) to detect a case of active TB among PLWH.METHODS: We identified studies reporting ACF for TB among PLWH published from November 2010 to February 2020. We calculated crude NNS for Xpert- or culture-confirmed TB and weighted mean NNS stratified by screening approach, population/risk group, and country TB burden.RESULTS: Of the 27,221 abstracts screened, we identified 58 studies eligible for inclusion, including 5 in low/moderate TB incidence settings and 53 in medium/high incidence settings. Populations screened for TB included inpatients, outpatients not receiving antiretroviral therapy (ART), outpatients receiving ART, those with CD4 < 200 cells/µL, children aged ≤15 years, pregnant PLWH, and PLWH in prisons. Screening tools included symptom-based screening, chest X-ray, C-reactive protein levels, and Xpert. The weighted mean NNS varied across groups but was consistently low, ranging from 4 among inpatients in moderate/high TB burden settings to 137 among pregnant PLWH in moderate/high TB burden settings.CONCLUSIONS: ACF is a high yield intervention among PLWH. Approaches to screening should be tailored to local epidemiological and health-system contexts, and sensitive screening tools such as Xpert should be implemented where feasible.

Feasibility of a streamlined tuberculosis diagnosis and treatment initiation strategy

OBJECTIVE:

To assess the feasibility of a streamlined strategy for improving tuberculosis (TB) diagnostic evaluation and treatment initiation among patients with presumed TB.

DESIGN:

Single-arm interventional pilot study at five primary care health centers of a streamlined, SIngle-saMPLE (SIMPLE) TB diagnostic evaluation strategy: 1) examination of two smear results from a single spot sputum specimen using light-emitting diode fluorescence microscopy, and 2) daily transportation of smear-negative sputum samples to Xpert® MTB/RIF testing sites.

RESULTS:

Of 1212 adults who underwent sputum testing for TB, 99.6% had two smears examined from the spot sputum specimen. Sputum was transported for Xpert testing within 1 clinic day for 83% (907/1091) of the smear-negative patients. Of 157 (13%) patients with bacteriologically positive TB, 116 (74%) were identified using sputum smear microscopy and 41 (26%) using Xpert testing of smear-negative samples. Anti-tuberculosis treatment was initiated in 142 (90%) patients with bacteriologically positive TB, with a median time to treatment of 1 day for smear-positive patients and 6 days for smear-negative, Xpert-positive patients.

CONCLUSION:

The SIMPLE TB strategy led to successful incorporation of Xpert testing and rapid treatment initiation in the majority of patients with bacteriologically confirmed TB in a resource-limited setting.

Evaluation of mobile digital light-emitting diode fluorescence microscopy in Hanoi, Viet Nam

SETTING

Hanoi Lung Hospital, Hanoi, Viet Nam.

OBJECTIVE

To compare the accuracy of CellScopeTB, a manually operated mobile digital fluorescence microscope, with conventional microscopy techniques.

DESIGN

Patients referred for sputum smear microscopy to the Hanoi Lung Hospital from May to September 2013 were included. Ziehl-Neelsen (ZN) smear microscopy, conventional light-emitting diode (LED) fluorescence microscopy (FM), CellScopeTB-based LED FM and Xpert(®) MTB/RIF were performed on sputum samples. The sensitivity and specificity of microscopy techniques were determined in reference to Xpert results, and differences were compared using McNemar's paired test of proportions.

RESULTS

Of 326 patients enrolled, 93 (28.5%) were Xpert-positive for TB. The sensitivity of ZN microscopy, conventional LED FM, and CellScopeTB-based LED FM was respectively 37.6% (95%CI 27.8-48.3), 41.9% (95%CI 31.8-52.6), and 35.5% (95%CI 25.8-46.1). The sensitivity of CellScopeTB was similar to that of conventional LED FM (difference -6.5%, 95%CI -18.2 to 5.3, P = 0.33) and ZN microscopy (difference -2.2%, 95%CI -9.2 to 4.9, P = 0.73). The specificity was >99% for all three techniques.

DISCUSSION

CellScopeTB performed similarly to conventional microscopy techniques in the hands of experienced TB microscopists. However, the sensitivity of all sputum microscopy techniques was low. Options enabled by digital microscopy, such as automated imaging with real-time computerized analysis, should be explored to increase sensitivity.

Impact of mycobacterial culture among HIV-infected adults with presumed TB in Uganda: a prospective cohort study

BACKGROUND

Implementation of new tuberculosis (TB) diagnostic strategies in resource-constrained settings is challenging. We measured the impact of solid and liquid mycobacterial cultures on treatment practices for patients undergoing TB evaluation in Kampala, Uganda.

METHODS

We enrolled consecutive smear-negative, human immunodeficiency virus positive adults with cough of ⩾2 weeks from September 2009 to April 2010. Laboratory technicians performed mycobacterial cultures on solid and liquid media. We compared empiric treatment decisions with solid and liquid culture in terms of diagnostic yield and time to results, and assessed impact on patient management.

RESULTS

Of 200 patients enrolled, 26 (13%) had culture-confirmed TB: 22 (85%) on solid culture alone, 2 (8%) on liquid culture alone, and 2 (8%) on both solid and liquid culture. Thirty-four patients received empiric anti-tuberculosis treatment, but only 10 (29%) were culture-positive. Median time to a positive result on solid culture was 92 days (interquartile range [IQR] 69-148) compared to 106 days (IQR 66-157) for liquid culture. No patients initiated treatment following a positive result on liquid culture.

CONCLUSION

The introduction of mycobacterial culture did not influence care for patients undergoing evaluation for TB in Kampala, Uganda. Attention to contextual factors surrounding implementation is needed to ensure the effective introduction of new testing strategies in low-income countries.