Tag: statistics

World Neurosurg. 2023 Aug 11:S1878-8750(23)01126-9. doi: 10.1016/j.wneu.2023.08.023. Online ahead of print.

ABSTRACT

BACKGROUND: Meningiomas are common intracranial tumors. Machine learning (ML) algorithms are emerging to improve accuracy in four primary domains: classification, grading, outcome prediction, and segmentation. Such algorithms include both, traditional approaches that rely on hand-crafted features and deep learning techniques that utilize automatic feature extraction.

OBJECTIVE: To evaluate the performance of published traditional ML vs. deep learning algorithms in classification, grading, outcome prediction, and segmentation of meningiomas.

METHODS: A systematic review and meta-analysis were conducted. Major databases were searched through September 2021 for publications evaluating traditional ML vs. deep learning models on meningioma management. Performance measures including pooled sensitivity, specificity, F1-score, area under the receiver-operating characteristic curve (AUC), positive and negative likelihood ratios (LR+, LR-) along with their respective 95% confidence intervals (95%CIs) were derived using random-effects models.

RESULTS: 534 records were screened, and 43 articles were included, regarding classification (3 articles), grading (29), outcome prediction (7), and segmentation (6) of meningiomas. Of the 29 studies that reported on grading, 10 could be meta-analyzed with two deep learning models (sensitivity 0.89, 95%CI 0.74-0.96; specificity 0.91, 95%CI 0.45-0.99; LR+ 10.1, 95%CI 1.33-137; LR- 0.12, 95%CI 0.04-0.59) and eight traditional ML (sensitivity 0.74, 95%CI 0.62-0.83; specificity 0.93, 95%CI 0.79-0.98; LR+ 10.5, 95%CI 2.91-39.5; and LR- 0.28, 95%CI 0.17-0.49). The insufficient performance metrics reported precluded further statistical analysis of other performance metrics.

CONCLUSION: Machine learning on meningiomas is mostly carried out with traditional methods. For meningioma grading, traditional machine learning methods generally had a higher LR+, while deep learning models a lower LR-.

PMID:37574189 | DOI:10.1016/j.wneu.2023.08.023

J Pain. 2023 Aug 11:S1526-5900(23)00496-0. doi: 10.1016/j.jpain.2023.08.001. Online ahead of print.

ABSTRACT

Elevated levels of anxiety in relation to chronic pain have been consistently associated with greater distress and disability. Thus, accurate measurement of pain-related anxiety is an important requirement in modern pain services. The Pain Anxiety Symptom Scale (PASS) was introduced over 30 years ago, with a shortened 20 item version introduced 10 years later. Both versions of the PASS were derived using principal components analysis, an established method of measure development with roots in classical test theory. Item Response Theory is a complementary approach to measure development that can reduce the number of items needed and maximize item utility with minimal loss of statistical and clinical information. The present study used IRT to shorten the 20 item PASS (PASS-20) in a large sample of people with chronic pain (N = 2669). Two shortened versions were evaluated, one composed of the single best performing item from each of its four subscales (PASS-4) and the other with the two best performing items from each subscale (PASS-8). Several supplementary analyses were performed, including comparative item convergence evaluations based on sample characteristics (i.e., female/male sex; clinical/online sample), factor invariance testing, and criterion validity evaluation of the 4, 8, and 20 item version of the PASS in hierarchical regression models predicting pain-related distress and interference. Overall, both shortened PASS versions performed adequately across these supplemental tests, although the PASS-4 had more consistent item convergence between samples, stronger evidence for factor invariance, and accounted for 83% of the variance accounted for by the PASS-20 and 92% of the variance accounted for by the PASS-8 in criterion variables. Consequently, the PASS-4 is recommended for use in situations where a briefer evaluation of pain-related anxiety is appropriate. PERSPECTIVE: The Pain Anxiety Symptom Scale (PASS) is an established measure of pain-related fear. This study derived four and eight item versions of the PASS using Item Response Theory. Both versions showed strong psychometric properties, stability of factor structure, and relation to important aspects of pain-related functioning.

PMID:37574179 | DOI:10.1016/j.jpain.2023.08.001

J Thorac Oncol. 2023 Aug 11:S1556-0864(23)00730-X. doi: 10.1016/j.jtho.2023.08.010. Online ahead of print.

ABSTRACT

INTRODUCTION: This open-label, phase 3 trial (ALTA-3; NCT03596866) compared efficacy and safety of brigatinib versus alectinib for ALK+ NSCLC after disease progression on crizotinib.

METHODS: Patients with advanced ALK+ NSCLC that progressed on crizotinib were randomized 1:1 to brigatinib 180 mg once daily (7-day lead-in, 90 mg) or alectinib 600 mg twice daily, aiming to test superiority. The primary endpoint was blinded independent review committee (BIRC)-assessed progression-free survival (PFS). Interim analysis for efficacy and futility was planned at approximately 70% of 164 expected PFS events.

RESULTS: The population (N=248; brigatinib, n=125; alectinib, n=123) was notable for long median duration of prior crizotinib (16.0-16.8 months) and low rate of ALK fusion in baseline circulating tumor DNA (ctDNA; 78/232 [34%]). Median BIRC-assessed PFS was 19.3 months with brigatinib and 19.2 months with alectinib (hazard ratio: 0.97 [95% confidence interval [CI]: 0.66-1.42]; p=0.8672]). The study met futility criterion. Overall survival was immature (41 events [17%]). Exploratory analyses pooled across treatment groups demonstrated median PFS of 11.1 versus 22.5 months in patients with versus without ctDNA-detectable ALK fusion at baseline (hazard ratio: 0.48 [95% CI: 0.32-0.71]). Treatment-related adverse events in >30% of patients (brigatinib/alectinib) were elevated blood creatine phosphokinase (70%/29%), aspartate aminotransferase (53%/38%), and alanine aminotransferase (40%/36%).

CONCLUSION: Brigatinib was not superior to alectinib for PFS in crizotinib-pretreated ALK+ NSCLC. Safety was consistent with the well-established and unique profiles of each drug. The low proportion of patients with ctDNA-detectable ALK fusion may account for prolonged PFS with both drugs in ALTA-3.

PMID:37574132 | DOI:10.1016/j.jtho.2023.08.010

Can J Cardiol. 2023 Aug 11:S0828-282X(23)01582-9. doi: 10.1016/j.cjca.2023.08.008. Online ahead of print.

NO ABSTRACT

PMID:37574131 | DOI:10.1016/j.cjca.2023.08.008

Neuroimage. 2023 Aug 11:120319. doi: 10.1016/j.neuroimage.2023.120319. Online ahead of print.

ABSTRACT

Human cognitive performance is a key function whose biological foundations have been partially revealed by genetic and brain imaging studies. The sleep electroencephalogram (EEG) is tightly linked to structural and functional features of the central nervous system and serves as another promising biomarker. We used data from MrOS, a large cohort of older men and cross-validated regularized regression to link sleep EEG features to cognitive performance in cross-sectional analyses. In independent validation samples 2.5-10% of variance in cognitive performance can be accounted for by sleep EEG features, depending on the covariates used. Demographic characteristics accounted for more covariance between sleep EEG and cognition than health variables, and consequently reduced this association by a greater degree, but even with the strictest covariate sets a statistically significant association was present. Sigma power in NREM and beta power in REM sleep were associated with better cognitive performance, while theta power in REM sleep was associated with worse performance, with no substantial effect of coherence and other sleep EEG metrics. Our findings show that cognitive performance is associated with the sleep EEG (r=0.283), with the strongest effect ascribed to spindle-frequency activity. This association becomes weaker after adjusting for demographic (r=0.186) and health variables (r=0.155), but its resilience to covariate inclusion suggest that it also partially reflects trait-like differences in cognitive ability.

PMID:37574121 | DOI:10.1016/j.neuroimage.2023.120319

Neuroimage. 2023 Aug 11:120312. doi: 10.1016/j.neuroimage.2023.120312. Online ahead of print.

ABSTRACT

Activity-dependent myelination is a fundamentally important mode of brain plasticity which significantly influences function. We recently discovered that absence seizures, which occur in multiple forms of generalized epilepsy, can induce activity-dependent myelination, which in turn promotes further progression of epilepsy. Structural alterations of myelin are likely to be widespread, given that absence seizures arise from an extensive thalamocortical network involving frontoparietal regions of the bilateral hemispheres. However, the temporal course and spatial extent of myelin plasticity is unknown, due to limitations of gold-standard histological methods such as electron microscopy (EM). In this study, we leveraged magnetization transfer and diffusion MRI for estimation of g-ratios across major white matter tracts in a mouse model of generalized epilepsy with progressive absence seizures. Electron microscopy was performed on the same brains after MRI. After seizure progression, we found increased myelination (decreased g-ratios) throughout the anterior portion (genu-to-body) of the corpus callosum but not in the posterior portion (body-splenium) nor in the fornix or the internal capsule. Curves obtained from averaging g-ratio values at every longitudinal point of the corpus callosum were statistically different with p<0.0001. Seizure-associated myelin differences found in the corpus callosum body with MRI were statistically significant (p = 0.0027) and were concordant with EM in the same region (p = 0.01). Notably, these differences were not detected by diffusion tensor imaging. This study reveals widespread myelin structural change that is specific to the absence seizure network.Furthermore, our findings demonstrate the potential utility and importance of MRI-based g-ratio estimation to non-invasively detect myelin plasticity.

PMID:37574120 | DOI:10.1016/j.neuroimage.2023.120312

Osteoarthritis Cartilage. 2023 Aug 11:S1063-4584(23)00880-4. doi: 10.1016/j.joca.2023.07.011. Online ahead of print.

ABSTRACT

OBJECTIVE: To assess (i) the impact of changes in body weight on changes in joint-adjacent subcutaneous fat (SCF) and cartilage thickness over 4-years and (ii) the relation between changes in joint-adjacent SCF and knee cartilage thickness.

DESIGN: Individuals from the Osteoarthritis Initiative (total=399) with >10% weight gain (n=100) and >10% weight loss (n=100) over 4 years were compared to a matched control cohort with less than 3% change in weight (n=199). 3.0T MRI of the right knee was performed at baseline and after 4 years to quantify joint-adjacent SCF and cartilage thickness. Linear regression models were used to evaluate the associations between the (i) weight change group and 4-year changes in both knee SCF and cartilage thickness, and (ii) 4-year changes in knee SCF and in cartilage thickness. Analyses were adjusted for age, sex, baseline BMI, tibial diameter (and weight change group in analysis (ii)).

RESULTS: Individuals who lost weight over 4-years had significantly less joint-adjacent SCF (beta range, medial/lateral joint sides: 2.2mm to 4.2mm, p<0.001) than controls; individuals who gained weight had significantly greater joint-adjacent SCF than controls (beta range: -1.4mm- -3.9mm, p<0.001). No statistically significant associations were found between weight change and cartilage thickness change. However, increases in joint-adjacent SCF over 4-years were significantly associated with decreases in cartilage thickness (p=0.04).

CONCLUSIONS: Weight change was associated with joint-adjacent SCF, but not with change in cartilage thickness. However, 4-year increases in joint-adjacent SCF were associated with decreases in cartilage thickness independent of baseline BMI and weight change group.

PMID:37574110 | DOI:10.1016/j.joca.2023.07.011

Ann Endocrinol (Paris). 2023 Aug 11:S0003-4266(23)00685-6. doi: 10.1016/j.ando.2023.08.002. Online ahead of print.

ABSTRACT

BACKGROUND AND AIMS: We describe mortality-related risk factors of inpatients with diabetes and coronavirus disease 2019 (COVID-19) in Belgium.

METHODS: We conducted a multicenter retrospective study from March to May, 2020, in 8 Belgian centers. Data on admission of patients with diabetes and COVID-19 were collected. Survivors were compared to non-survivors to identify prognostic risk factors for in-hospital death, using multivariate analysis in both the total population and in the subgroup of patients admitted to the intensive care unit (ICU).

RESULTS: The study included 375 patients. The mortality rate was 26.4% (99/375) in the total population and 40% (27/67) in the ICU. Multivariate analysis identified older age (HR 1.05 [CI 1.03-1.07], p <0.0001) and male gender (HR 2.01 [1.31-3.07], p 0.0013) as the main independent risk factors for in-hospital death in the total population. Use of metformin (HR 0.51 [0.34-0.78], p 0.0018) and renin-angiotensin-aldosterone system blockers (HR 0.56 [0.36-0.86], p 0.0088) before admission were independent protective factors. In the ICU, chronic kidney disease (CKD) was identified as an independent risk factor for death (HR 4.96 [2.14-11.5], p <0.001).

CONCLUSION: In-hospital mortality due to the first wave of the COVID-19 pandemic in Belgium was high in patients with diabetes. We found that advanced age and male gender were independent risk factors for in-hospital death. We also showed that metformin use before admission was associated with significantly lower COVID-19-related in-hospital mortality. Finally, we showed that CKD is a COVID-19-related mortality risk factor in patients with diabetes admitted to the ICU.

PMID:37574109 | DOI:10.1016/j.ando.2023.08.002

Sci Total Environ. 2023 Aug 11:166148. doi: 10.1016/j.scitotenv.2023.166148. Online ahead of print.

ABSTRACT

Common beech (Fagus sylvatica) is one of the most important deciduous tree species in European forests. However, climate-change-induced drought may threaten its dominant position. The Sonian Forest close to Brussels (Belgium) is home to some of the largest beech trees in the world. This UNESCO world heritage site is famous for its high density of very large beech trees as a result of its climatic suitability, fertile soil conditions, and past management. Here we utilized tree-ring data from increment cores to investigate the growth of these old and monumental beech trees, evaluating their growth trends, response to past climate, and the effect of mast years on 39 living and 16 recently wind-thrown trees. Our analysis reveals that the sampled trees were generally sensitive to spring and summer droughts but recovered quickly after such an extreme climatic event. The growth trend of living trees has remained high and only shows a slight, statistically insignificant, decline over the past 50 years. Although the overall growth rate remains strong (BAI 50 cm²/year), the past five decades have shown strong inter-annual growth variations due to frequent and more intense droughts combined with an increased frequency of mast years. We also found notable differences in growth patterns between the living trees and those that had recently been wind-thrown. While there were no significant differences between living and wind-thrown trees in response to droughts, heatwaves, or mast years when examining year-to-year growth changes, the wind-thrown trees did exhibit considerably lower overall growth rates and a significant downward trend in growth (BAI -0.57 cm²/year). This difference in growth trends has been apparent since at least the 1980s. Overall, the findings of this study can provide valuable insights for understanding the long-term dynamics of lowland beech forests and their responses to climate change.

PMID:37574075 | DOI:10.1016/j.scitotenv.2023.166148

Sci Total Environ. 2023 Aug 11:166163. doi: 10.1016/j.scitotenv.2023.166163. Online ahead of print.

ABSTRACT

Wastewater-based epidemiology (WBE) relies on representative sampling that is typically achieved with autosamplers that collect time, flow, or volume proportional samples. The expense, resources and operational know-how associated with autosampler operation means they are only typically available at major wastewater treatment plants (WWTPs). This results in a lack of data on consumption levels in regional and remote areas, or in countries that lack the financial means. The aim of this study was to estimate and investigate trends in drug consumption across varying levels of remoteness in Australia. Field-calibrated, microporous polyethylene passive samplers were deployed over 2 periods (Aug/Sept 2019 and 2020) at 43 treatment plants covering all five categories of remoteness, as per Australian Bureau of Statistics definitions (Major cities, Inner regional, Outer regional, Remote, and Very remote). The per capita consumption of cocaine, methylamphetamine, nicotine, oxycodone and MDMA were estimated. No spatial trends between remoteness and drug consumption were observed, except for cocaine, where Major cities had a 5-to-10-fold higher consumption compared to the other levels of remoteness in 2019 and 2020, respectively. Outer regional sites had the highest and lowest methylamphetamine consumption. The variance in drug use among sites was much higher in Remote (and Inner/Outer regional) sites when compared with Major cities. A significant and consistent decrease in oxycodone consumption was observed at all sites between 2019 and 2020, possibly related to regulatory changes and the COVID-19 pandemic where elective surgeries were suspended. The majority of sites experienced a decrease in cocaine and methylamphetamine consumption, possibly due to border restrictions or changes in supply and demand dynamics. This was the first extensive passive sampling study to assess drug consumption in urban, regional, and remote locations, demonstrating that passive samplers can facilitate extension of wastewater-based drug monitoring programs to sites where other representative sampling options are very difficult to implement.

PMID:37574069 | DOI:10.1016/j.scitotenv.2023.166163