Tutorial in Biostatistics: The use of generalized additive models to evaluate alcohol consumption as an exposure variable

Clinical Utility of Machine Learning Methods Using Regression Models for Diagnosing Eosinophilic Chronic Rhinosinusitis

Objective

Machine learning methods using regression models can predict actual values of histological eosinophil count from blood eosinophil levels. Therefore, these methods might be useful for diagnosing eosinophilic chronic rhinosinusitis, but their utility still remains unclear. We compared 2 statistical approaches, and investigated the utility of machine learning methods for diagnosing eosinophilic chronic rhinosinusitis.

Study Design

Retrospective study.

Setting

Medical center.

Methods

Data, including eosinophilic levels, obtained from blood and sinonasal samples of 264 patients with chronic rhinosinusitis (257 with and 57 without nasal polyps) were analyzed. We determined factors affecting histopathological eosinophil count in regression models. We also investigated optimal cutoff values for blood eosinophil percentages/absolute eosinophil counts (AECs) through receiver operating characteristic curves and machine-learning methods based on regression models. A histopathological eosinophil count ≥10/high-power field was defined as eosinophilic chronic rhinosinusitis.

Results

Blood eosinophil levels, nasal polyp presence, and comorbid asthma were factors affecting histopathological eosinophil count. Cutoffs between the 2 statistical approaches differed in the group with nasal polyps, but not in one without nasal polyps. Machine-learning methods identified blood eosinophil percentages ≥1% or AEC ≥100/μL as cut-offs for eosinophilic chronic rhinosinusitis with nasal polyps, while ≥6% or ≥400/μL for one without nasal polyps.

Conclusion

Cut-offs of blood eosinophil levels obtained by machine-learning methods might be useful when suspecting eosinophilic chronic rhinosinusitis prior to biopsy because of their ability to adjust covariates, dealing with overfitting, and predicting actual values of histological eosinophil count.

Association between remnant cholesterol and progression of bioprosthetic valve degeneration

AIMS

Remnant cholesterol (RC) seems associated with native aortic stenosis. Bioprosthetic valve degeneration may share similar lipid-mediated pathways with aortic stenosis. We aimed to investigate the association of RC with the progression of bioprosthetic aortic valve degeneration and ensuing clinical outcomes.

METHODS AND RESULTS

We enrolled 203 patients with a median of 7.0 years (interquartile range: 5.1-9.2) after surgical aortic valve replacement. RC concentration was dichotomized by the top RC tertile (23.7 mg/dL). At 3-year follow-up, 121 patients underwent follow-up visit for the assessment of annualized change in aortic valve calcium density (AVCd). RC levels showed a curvilinear relationship with an annualized progression rate of AVCd, with increased progression rates when RC >23.7 mg/dL (P = 0.008). There were 99 deaths and 46 aortic valve re-interventions in 133 patients during a median clinical follow-up of 8.8 (8.7-9.6) years. RC >23.7 mg/dL was independently associated with mortality or re-intervention (hazard ratio: 1.98; 95% confidence interval: 1.31-2.99; P = 0.001).

CONCLUSION

Elevated RC is independently associated with faster progression of bioprosthetic valve degeneration and increased risk of all-cause mortality or aortic valve re-intervention.

Assessing the feasibility of SUVindex (a metric derived from FDG PET/CT) for the diagnosis of polymyalgia rheumatica

AIM

To evaluate the feasibility of standard uptake value (SUV) index (ratio lesional maximum SUV [SUVmax] to liver mean SUV [SUVmean]) as a metabolic parameter for diagnosing polymyalgia rheumatica (PMR).

MATERIALS AND METHODS

A retrospective group of patients with PMR and controls with symptoms similar to PMR but diagnosed with other diseases. Semiquantitative and qualitative analysis of 2-[18F]-fluoro-2-deoxy-d-glucose (18F-FDG) uptake at 18 sites was undertaken for all patients. The diagnostic value of positron-emission tomography/computed tomography (PET/CT) for PMR was assessed by R software using logistic regression and a generalised additive model (GAM). All images were examined independently by two nuclear medicine physicians with extensive work experience.

RESULTS

The characteristic sites of PMR were the ischial tuberosity, interspinous bursa, periarticular hip, and symphysis pubis enthesis. The area under the curve (AUC) of the characteristic site SUV index was 0.930, and the best cut-off value was 1.685 with a sensitivity of 84.6% and a specificity of 92.6%. After adjusting for potential confounders, the probability of PMR diagnosis increased as the characteristic site SUV index increased and there was a nonlinear correlation between the two. When the characteristic site SUV index was ≥2.56, the probability of PMR gradually reached the threshold effect, which was as high as 90% or more.

CONCLUSION

The characteristic site SUV index is an independent factor for diagnosing PMR, and PMR should be highly suspected when it is ≥ 1.685. Nonetheless, it is important to note that these findings are based on an initial retrospective single-centre study and require external validation and further prospective evaluation before being translated into clinical practice.

Multicultural emergency medicine epidemiology: A health economic analysis of patient visits

OBJECTIVE

There is growing evidence to suggest that culturally and linguistically diverse (CALD) patients cost the health system more than non-CALD patients because of a higher burden of disease and increased resource consumption. The present study aimed to compare the ED resource utilisation of CALD and non-CALD patients at a tertiary hospital in Sydney, Australia.

METHODS

The total ED resource utilisation was calculated by separating each visit into diagnostic test cost and time spent in ED components. The time component was calculated using the product of the total length of stay and a resource cost per unit time measure. Diagnostic tests were costed using the Australian Medicare Benefit Schedule. A generalised additive model was developed to estimate the isolated effect of CALD status on the resource utilisation during an ED visit.

RESULTS

CALD patients had a higher median resource utilisation than non-CALD patients ($736.93 vs $701.36, P < 0.0001); however, the generalised additive model demonstrated that CALD status was not independently associated with increased resource utilisation.

CONCLUSION

CALD status is not an independent influence on ED resource utilisation but other explanatory variables such as increased age and altered case-mix appear to have a much greater influence. There may, however, be other reasons to consider CALD loading such as equity in healthcare and to address poorer overall health outcomes for CALD patients.