Retrospective Dataset Research Articles

Machine learning approach for predicting inhalation injury in patients with burns

BackgroundThe coronavirus disease pandemic has had a tangible impact on bronchoscopy for burn inpatients due to isolation and triage measures. We utilised the machine-learning approach to identify risk factors for predicting mild and severe inhalation injury and whether patients with burns experienced inhalation injury. We also examined the ability of two dichotomous models to predict clinical outcomes including mortality, pneumonia, and duration of hospitalisation. MethodsA retrospective 14-year single-centre dataset of 341 intubated patients with burns with suspected inhalation injury was established. The medical data on day one of admission and bronchoscopy-diagnosed inhalation injury grade were compiled using a gradient boosting-based machine-learning algorithm to create two prediction models: model 1, mild vs. severe inhalation injury; and model 2, no inhalation injury vs. inhalation injury. ResultsThe area under the curve (AUC) for model 1 was 0·883, indicating excellent discrimination. The AUC for model 2 was 0·862, indicating acceptable discrimination. In model 1, the incidence of pneumonia (P < 0·001) and mortality rate (P < 0·001), but not duration of hospitalisation (P = 0·1052), were significantly higher in patients with severe inhalation injury. In model 2, the incidence of pneumonia (P < 0·001), mortality (P < 0·001), and duration of hospitalisation (P = 0·021) were significantly higher in patients with inhalation injury. ConclusionsWe developed the first machine-learning tool for differentiating between mild and severe inhalation injury, and the absence/presence of inhalation injury in patients with burns, which is helpful when bronchoscopy is not available immediately. The dichotomous classification predicted by both models was associated with the clinical outcomes.

Generalizable and robust deep learning algorithm for atrial fibrillation diagnosis across geography, ages and sexes

To drive health innovation that meets the needs of all and democratize healthcare, there is a need to assess the generalization performance of deep learning (DL) algorithms across various distribution shifts to ensure that these algorithms are robust. This retrospective study is, to the best of our knowledge, an original attempt to develop and assess the generalization performance of a DL model for AF events detection from long term beat-to-beat intervals across geography, ages and sexes. The new recurrent DL model, denoted ArNet2, is developed on a large retrospective dataset of 2,147 patients totaling 51,386 h obtained from continuous electrocardiogram (ECG). The model’s generalization is evaluated on manually annotated test sets from four centers (USA, Israel, Japan and China) totaling 402 patients. The model is further validated on a retrospective dataset of 1,825 consecutives Holter recordings from Israel. The model outperforms benchmark state-of-the-art models and generalized well across geography, ages and sexes. For the task of event detection ArNet2 performance was higher for female than male, higher for young adults (less than 61 years old) than other age groups and across geography. Finally, ArNet2 shows better performance for the test sets from the USA and China. The main finding explaining these variations is an impairment in performance in groups with a higher prevalence of atrial flutter (AFL). Our findings on the relative performance of ArNet2 across groups may have clinical implications on the choice of the preferred AF examination method to use relative to the group of interest.

Deep Learning Body Region Classification of MRI and CT Examinations.

This study demonstrates the high performance of deep learning in identification of body regions covering the entire human body from magnetic resonance (MR) and computed tomography (CT) axial images across diverse acquisition protocols and modality manufacturers. Pixel-based analysis of anatomy contained in image sets can provide accurate anatomic labeling. For this purpose, a convolutional neural network (CNN)-based classifier was developed to identify body regions in CT and MRI studies. Seventeen CT (18 MRI) body regions covering the entire human body were defined for the classification task. Three retrospective datasets were built for the AI model training, validation, and testing, with a balanced distribution of studies per body region. The test datasets originated from a different healthcare network than the train and validation datasets. Sensitivity and specificity of the classifier was evaluated for patient age, patient sex, institution, scanner manufacturer, contrast, slice thickness, MRI sequence, and CT kernel. The data included a retrospective cohort of 2891 anonymized CT cases (training, 1804 studies; validation, 602 studies; test, 485 studies) and 3339 anonymized MRI cases (training, 1911 studies; validation, 636 studies; test, 792 studies). Twenty-seven institutions from primary care hospitals, community hospitals, and imaging centers contributed to the test datasets. The data included cases of all sexes in equal proportions and subjects aged from 18years old to + 90years old. Image-level weighted sensitivity of 92.5% (92.1-92.8) for CT and 92.3% (92.0-92.5) for MRI and weighted specificity of 99.4% (99.4-99.5) for CT and 99.2% (99.1-99.2) for MRI were achieved. Deep learning models can classify CT and MR images by body region including lower and upper extremities with high accuracy.

Dynamical Predictability of Leading Interannual Variability Modes of the Asian-Australian Monsoon in Climate Models

The dynamical prediction of the Asian-Australian monsoon (AAM) has been an important and long-standing issue in climate science. In this study, the predictability of the first two leading modes of the AAM is studied using retrospective prediction datasets from the seasonal forecasting models in four operational centers worldwide. Results show that the model predictability of the leading AAM modes is sensitive to how they are defined in different seasonal sequences, especially for the second mode. The first AAM mode, from various seasonal sequences, coincides with the El Niño phase transition in the eastern-central Pacific. The second mode, initialized from boreal summer and autumn, leads El Niño by about one year but can exist during the decay phase of El Niño when initialized from boreal winter and spring. Our findings hint that ENSO, as an early signal, is conducive to better performance of model predictions in capturing the spatiotemporal variations of the leading AAM modes. Still, the persistence barrier of ENSO in spring leads to poor forecasting skills of spatial features. The multimodel ensemble (MME) mean shows some advantage in capturing the spatiotemporal variations of the AAM modes but does not provide a significant improvement in predicting its temporal features compared to the best individual models in predicting its temporal features. The BCC_CSM1.1M shows promising skill in predicting the two AAM indices associated with two leading AAM modes. The predictability demonstrated in this study is potentially useful for AAM prediction in operational and climate services.

External Validation of an Artificial Intelligence Device for Intracranial Hemorrhage Detection

Artificial intelligence applications have gained traction in the field of cerebrovascular disease by assisting in the triage, classification, and prognostication of both ischemic and hemorrhagic stroke. The Caire ICH system aims to be the first device to move into the realm of assisted diagnosis for intracranial hemorrhage (ICH) and its subtypes. A single-center retrospective dataset of 402 head noncontrast CT scans (NCCT) with an intracranial hemorrhage were retrospectively collected from January 2012 to July 2020; an additional 108 NCCT scans with no intracranial hemorrhage findings were also included. The presence of an ICH and its subtype were determined from the International Classification of Diseases-10 code associated with the scan and validated by an expert panel. We used the Caire ICH vR1 to analyze these scans, and we evaluated its performance in terms of accuracy, sensitivity, and specificity. We found the Caire ICH system to have an accuracy of 98.05% (95% confidence interval [CI]: 96.44%-99.06%), a sensitivity of 97.52% (95% CI: 95.50%-98.81%), and a specificity of 100% (95% CI: 96.67%-100.00%) in the detection of ICH. Experts reviewed the 10 incorrectly classified scans. The Caire ICH vR1 algorithm was highly accurate, sensitive, and specific in detecting the presence or absence of an ICH and its subtypes in NCCTs. This work suggests that the Caire ICH device has potential to minimize clinical errors in ICH diagnosis that could improve patient outcomes and current workflows as both a point-of-care tool for diagnostics and as a safety net for radiologists.

Trends in HIV incidence following scale-up of harm reduction interventions among people who inject drugs in Kachin, Myanmar, 2008–2020: analysis of a retrospective cohort dataset

People who inject drugs (PWID) in Kachin, Myanmar, have a high HIV prevalence (>40%), but there is no data on incidence. We used HIV testing data from three harm reduction drop-in centres (DIC) in Kachin (2008-2020) to determine HIV incidence trends among PWID and associations with intervention uptake. Individuals were HIV-tested at first DIC visit and periodically thereafter, during which demographic and risk behaviour data were collected. Two DIC provided opioid agonist therapy (OAT) from 2008. Monthly DIC-level needle/syringe provision (NSP) data was available from 2012. Site-level 6-monthly NSP coverage was denoted low, high, or medium if it was below the lower quartile, above upper quartile, between these quartiles of provision levels over 2012-2020, respectively. HIV incidence was estimated by linking subsequent test records for those initially testing HIV-negative. Associations with HIV incidence were examined using Cox regression. Follow-up HIV testing data was available for 31.4% (2227) of PWID initially testing HIV-negative, with 444 incident HIV infections during 6266.5 person years (py) of follow-up. Overall HIV incidence was 7.1 per 100 py (95% confidence interval 6.5-7.8), which decreased from 19.3 (13.3-28.2) in 2008-11 to 5.2 per 100 py (4.6-5.9) in 2017-20. In the full PWID incidence dataset after adjustment for various factors, recent (≤6 weeks) injecting (aHR 1.74, 1.35-2.25) and needle sharing (aHR 2.00, 1.48-2.70) were associated with higher incidence, while longer injection careers were associated with reduced incidence (aHR 0.54, 0.34-0.86, for 2-5 yrs vs<2 yrs). In a reduced dataset including data on OAT access and NSP coverage (2012-2020 for two DIC providing OAT), being on OAT during follow-up was associated with reduced HIV incidence (aHR 0.36, 0.27-0.48, compared to never being on OAT) as was high NSP coverage (aHR 0.64, 0.48-0.84, compared to medium syringe coverage). Although HIV incidence is high among PWID in Kachin, data suggests it has decreased since the scale-up in harm reduction interventions. US NIH, Médecins du Monde.

Management disparities in metastatic castrate-sensitive prostate cancer (mCSPC).

79 Background: Recent approvals in management of high-volume mCSPC offer major improvements in patient outcomes. Triplet therapy combining androgen deprivation therapy (ADT) with docetaxel (D) and novel hormonal therapy (NHT) is the new standard of care in this patient population. Patients ineligible for triplet therapy should receive doublet therapy with ADT plus D or NHT based on category 1 evidence. Multiple retrospective datasets examining the “real-world” mCSPC treatment landscape note significant underutilization of therapies beyond ADT and discordance with treatment guidelines. Methods: A 15-question survey to assess management patterns was electronically distributed to various mCSPC providers in California and Washington. Results: A total of 40 responses were received, with 62.5% identifying as medical oncologists (MO), 20.0% as urologists, 5.0% as radiation oncologists, and 12.5% as advanced practitioners/other. Out of physician responders, 48.6% were academic and 51.4% were community-based. When treating newly diagnosed fit mCSPC patients, 20.0% would prescribe triplet therapy, 42.5% ADT/NHT, 5.0% ADT/D, 5.0% ADT/bicalutamide, 0% ADT alone, 10.0% other therapies, and 17.5% would refer out for therapies beyond ADT. When comparing academic versus community oncologists, 66.7% versus 0% would prescribe triplet therapy, while 22.2% versus 68.8% would prescribe ADT/NHT. Some perceived barriers to following mCSPC treatment guidelines included difficulty keeping up with rapidly evolving data/too many treatment options in 27.5%, insurance coverage concerns in 22.5%, difficulty obtaining early referrals for these therapies in 12.5%, patient factors in 10.0%, and toxicity concerns in 2.5%. When making referrals to MO, 37.5% of non-MO responders would do so at development of mCSPC and 6.3% at mCRPC. Conversely, 24.2% of MO perceived receiving referrals at development of mCSPC and 33.3% at mCRPC. Patterns of ordering genomic testing are listed, with biggest perceived barriers being cost, inadequate access to sequencing platforms and genetic counselors, lack of confidence in counseling about results, and patient refusal. Upon progression from mCSPC to mCRPC, ordering of repeat tumor biopsies would occur “sometimes” by 40.0% of providers, “rarely” by 20.0%, and “never” by 20.0%. Conclusions: Multiple provider disparities when managing mCSPC patients still exist. Our survey demonstrated higher compliance with treatment guidelines than in previous retrospective datasets, possibly due to higher proportion of academic providers. Increased provider awareness of genomic testing recommendations is also needed. [Table: see text]

Ventilation and Complications after Lung Resection: Reply.

We thank Dr. Hoftman and Dr. Umar for their comments1 and their interest in our focused review.2 In their letter to the editor, they refer to their interesting work on the utility of vital capacity to estimate tidal volume in contrast to the widely used predicted body weight in patients undergoing thoracic surgery. Their work did not report on clinical outcomes.3 As these authors acknowledge, their data were derived from a small retrospective dataset, and their results are exploratory. Large clinical trials should be conducted before revising clinical care pathways.1,3 In the continuously enhanced world of personalized medicine, dogma most certainly should be challenged. Our review presented clinical outcome data that showed conflicting results on the association of not only tidal volume but also other parameters of protective lung ventilation bundles and morbidity in patients undergoing lung resection. We stressed the importance of proven risk factors for postoperative pulmonary complications after lung resection, such as decrements in forced expiratory volume in 1 s or diffusion capacity of the lung for carbon monoxide and greater fluid administration during surgery, to be considered when interpreting observational data or in the design of future studies.The expert in mechanical ventilation will stylize many factors including the inspiratory pressures and time, positive end-expiratory pressure, lung compliance, and respiratory rate to optimize breath delivery, which can change hour by hour during surgery. Promoting novices to experts certainly depends on education and experience. Algorithms and consultations with experts can improve critical care delivery. Designing ventilators with machine learning computers could potentially further personalize care.The authors declare no competing interests.

Comparison of Chest Radiograph Captions Based on Natural Language Processing vs Completed by Radiologists

Artificial intelligence (AI) can interpret abnormal signs in chest radiography (CXR) and generate captions, but a prospective study is needed to examine its practical value. To prospectively compare natural language processing (NLP)-generated CXR captions and the diagnostic findings of radiologists. A multicenter diagnostic study was conducted. The training data set included CXR images and reports retrospectively collected from February 1, 2014, to February 28, 2018. The retrospective test data set included consecutive images and reports from April 1 to July 31, 2019. The prospective test data set included consecutive images and reports from May 1 to September 30, 2021. A bidirectional encoder representation from a transformers model was used to extract language entities and relationships from unstructured CXR reports to establish 23 labels of abnormal signs to train convolutional neural networks. The participants in the prospective test group were randomly assigned to 1 of 3 different caption generation models: a normal template, NLP-generated captions, and rule-based captions based on convolutional neural networks. For each case, a resident drafted the report based on the randomly assigned captions and an experienced radiologist finalized the report blinded to the original captions. A total of 21 residents and 19 radiologists were involved. Time to write reports based on different caption generation models. The training data set consisted of 74 082 cases (39 254 [53.0%] women; mean [SD] age, 50.0 [17.1] years). In the retrospective (n = 8126; 4345 [53.5%] women; mean [SD] age, 47.9 [15.9] years) and prospective (n = 5091; 2416 [47.5%] women; mean [SD] age, 45.1 [15.6] years) test data sets, the mean (SD) area under the curve of abnormal signs was 0.87 (0.11) in the retrospective data set and 0.84 (0.09) in the prospective data set. The residents' mean (SD) reporting time using the NLP-generated model was 283 (37) seconds-significantly shorter than the normal template (347 [58] seconds; P < .001) and the rule-based model (296 [46] seconds; P < .001). The NLP-generated captions showed the highest similarity to the final reports with a mean (SD) bilingual evaluation understudy score of 0.69 (0.24)-significantly higher than the normal template (0.37 [0.09]; P < .001) and the rule-based model (0.57 [0.19]; P < .001). In this diagnostic study of NLP-generated CXR captions, prior information provided by NLP was associated with greater efficiency in the reporting process, while maintaining good consistency with the findings of radiologists.

A DSC Test for the Early Detection of Neoplastic Gastric Lesions in a Medium-Risk Gastric Cancer Area.

In this study, we aimed to assess the accuracy of the proposed novel, noninvasive serum DSC test in predicting the risk of gastric cancer before the use of upper endoscopy. To validate the DSC test, we enrolled two series of individuals living in Veneto and Friuli-Venezia Giulia, Italy (n = 53 and n = 113, respectively), who were referred for an endoscopy. The classification used for the DSC test to predict gastric cancer risk combines the coefficient of the patient's age and sex and serum pepsinogen I and II, gastrin 17, and anti-Helicobacter pylori immunoglobulin G concentrations in two equations: Y1 and Y2. The coefficient of variables and the Y1 and Y2 cutoff points (>0.385 and >0.294, respectively) were extrapolated using regression analysis and an ROC curve analysis of two retrospective datasets (300 cases for the Y1 equation and 200 cases for the Y2 equation). The first dataset included individuals with autoimmune atrophic gastritis and first-degree relatives with gastric cancer; the second dataset included blood donors. Demographic data were collected; serum pepsinogen, gastrin G17, and anti-Helicobacter pylori IgG concentrations were assayed using an automatic Maglumi system. Gastroscopies were performed by gastroenterologists using an Olympus video endoscope with detailed photographic documentation during examinations. Biopsies were taken at five standardized mucosa sites and were assessed by a pathologist for diagnosis. The accuracy of the DSC test in predicting neoplastic gastric lesions was estimated to be 74.657% (65%CI; 67.333% to 81.079%). The DSC test was found to be a useful, noninvasive, and simple approach to predicting gastric cancer risk in a population with a medium risk of developing gastric cancer.

Magnetic resonance imaging organ at risk delineation for nasopharyngeal radiotherapy: Measuring the effectiveness of an educational intervention.

Magnetic resonance imaging (MRI) demonstrates superior soft tissue contrast and is increasingly being used in radiotherapy planning. This study evaluated the impact of an education workshop in minimising inter-observer variation (IOV) for nasopharyngeal organs at risk (OAR) delineation on MRI. Ten observers delineated 14 OARs on 4 retrospective nasopharyngeal MRI data sets. Standard contouring guidelines were provided pre-workshop. Following an education workshop on MRI OAR delineation, observers blinded to their original contours repeated the 14 OAR delineations. For comparison, reference volumes were delineated by two head and neck radiation oncologists. IOV was evaluated using dice similarity coefficient (DSC), Hausdorff distance (HD) and relative volume. Location of largest deviations was evaluated with centroid values. Observer confidence pre- and post-workshop was also recorded using a 6-point Likert scale. The workshop was deemed beneficial for an OAR if ≥50% of observers mean scores improved in any metric and ≥50% of observers' confidence improved. All OARs had ≥50% of observers improve in at least one metric. Base of tongue, larynx, spinal cord and right temporal lobe were the only OARs achieving a mean DSC score of ≥0.7. Base of tongue, left and right lacrimal glands, larynx, left optic nerve and right parotid gland all exhibited statistically significant HD improvements post-workshop (P < 0.05). Brainstem and left and right temporal lobes all had statistically significant relative volume improvements post-workshop (P < 0.05). Post-workshop observer confidence improvement was observed for all OARs (P < 0.001). The educational workshop reduced IOV and improved observers' confidence when delineating nasopharyngeal OARs on MRI.

Abstract TP88: Deep Learning-based Automatic Detection And Segmentation Of Hyperdense Artery Sign On Non-contrast Computed Tomography In Acute Ischemic Stroke

Introduction: CT- or MR-angiography(CTA/MRA) is the golden-standard to evaluate the site of large vessel occlusion (LVO). However, there are cases in which occluded vessels can be recognized as a hyperdense artery sign (HAS) on non-contrast CT (NCCT). If the HAS is detectable from NCCT, it may lead to a significant reduction in the time to start treatment. In this study, we developed and validated the deep learning model to detect the HAS on NCCT. Methods: This study used three multicenter retrospective datasets; ischemic stroke patients with LVO (occlusions of the intracranial internal carotid artery (ICA) and middle cerebral artery M1/M2 segments) who underwent mechanical thrombectomy, ischemic stroke patients without LVO, non-stroke patients. For developing the deep neural network model, positive samples (182 ICA/M1 and 57 M2 occlusion patients) were selected from stroke patients with LVO wherein the HAS was visible on NCCT. The decision for HAS visibility was made by two Neuro-interventionalists who referred to paired MRA/CTA of the same patient. While doing so, they also manually labeled the HAS regions on NCCT. Negative samples (592 patients) were collected from stroke patients without LVO and non-stroke patients. We split the data into training, validation and test sets. After we trained and validated the model, its performance was evaluated on the test set. The performance was measured using the metrics of sensitivity and specificity. Results: The sensitivity of the model was 0.78 (49/63) for ICA/M1 and 0.77 (10/13) for M2 occlusion patients. The specificity of the model was 0.61 (33/54) for stroke patients without LVO and non-stroke patients. Conclusions: The performance of our deep learning-based method to automatically detect the HAS on NCCT is close to that of expert physicians reported in earlier studies, and our method may lead to further improvement of patient prognosis by shortening the time to start treatment.

Breast cancer survival analysis agents for clinical decision support

Personalized support and assistance are essential for cancer survivors, given the physical and psychological consequences they have to suffer after all the treatments and conditions associated with this illness. Digital assistive technologies have proved to be effective in enhancing the quality of life of cancer survivors, for instance, through physical exercise monitoring and recommendation or emotional support and prediction. To maximize the efficacy of these techniques, it is challenging to develop accurate models of patient trajectories, which are typically fed with information acquired from retrospective datasets. This paper presents a Machine Learning-based survival model embedded in a clinical decision system architecture for predicting cancer survivors’ trajectories. The proposed architecture of the system, named PERSIST, integrates the enrichment and pre-processing of clinical datasets coming from different sources and the development of clinical decision support modules. Moreover, the model includes detecting high-risk markers, which have been evaluated in terms of performance using both a third-party dataset of breast cancer patients and a retrospective dataset collected in the context of the PERSIST clinical study.

Validation of a deep neural network-based algorithm supporting clinical management of adnexal mass.

Conservative management of adnexal mass is warranted when there is imaging-based and clinical evidence of benign characteristics. Malignancy risk is, however, a concern due to the mortality rate of ovarian cancer. Malignancy occurs in 10-15% of adnexal masses that go to surgery, whereas the rate of malignancy is much lower in masses clinically characterized as benign or indeterminate. Additional diagnostic tests could assist conservative management of these patients. Here we report the clinical validation of OvaWatch, a multivariate index assay, with real-world evidence of performance that supports conservative management of adnexal masses. OvaWatch utilizes a previously characterized neural network-based algorithm combining serum biomarkers and clinical covariates and was used to examine malignancy risk in prospective and retrospective samples of patients with an adnexal mass. Retrospective data sets were assembled from previous studies using patients who had adnexal mass and were scheduled for surgery. The prospective study was a multi-center trial of women with adnexal mass as identified on clinical examination and indeterminate or asymptomatic by imaging. The performance to detect ovarian malignancy was evaluated at a previously validated score threshold. In retrospective, low prevalence (N = 1,453, 1.5% malignancy rate) data from patients that received an independent physician assessment of benign, OvaWatch has a sensitivity of 81.8% [95% confidence interval (CI) 65.1-92.7] for identifying a histologically confirmed malignancy, and a negative predictive value (NPV) of 99.7%. OvaWatch identified 18/22 malignancies missed by physician assessment. A prospective data set had 501 patients where 106 patients with adnexal mass went for surgery. The prevalence was 2% (10 malignancies). The sensitivity of OvaWatch for malignancy was 40% (95% CI: 16.8-68.7%), and the specificity was 87% (95% CI: 83.7-89.7) when patients were included in the analysis who did not go to surgery and were evaluated as benign. The NPV remained 98.6% (95% CI: 97.0-99.4%). An independent analysis set with a high prevalence (45.8%) the NPV value was 87.8% (95% CI: 95% CI: 75.8-94.3%). OvaWatch demonstrated high NPV across diverse data sets and promises utility as an effective diagnostic test supporting management of suspected benign or indeterminate mass to safely decrease or delay unnecessary surgeries.

Standardising Breast Radiotherapy Structure Naming Conventions: A Machine Learning Approach.

In progressing the use of big data in health systems, standardised nomenclature is required to enable data pooling and analyses. In many radiotherapy planning systems and their data archives, target volumes (TV) and organ-at-risk (OAR) structure nomenclature has not been standardised. Machine learning (ML) has been utilised to standardise volumes nomenclature in retrospective datasets. However, only subsets of the structures have been targeted. Within this paper, we proposed a new approach for standardising all the structures nomenclature by using multi-modal artificial neural networks. A cohort consisting of 1613 breast cancer patients treated with radiotherapy was identified from Liverpool & Macarthur Cancer Therapy Centres, NSW, Australia. Four types of volume characteristics were generated to represent each target and OAR volume: textual features, geometric features, dosimetry features, and imaging data. Five datasets were created from the original cohort, the first four represented different subsets of volumes and the last one represented the whole list of volumes. For each dataset, 15 sets of combinations of features were generated to investigate the effect of using different characteristics on the standardisation performance. The best model reported 99.416% classification accuracy over the hold-out sample when used to standardise all the nomenclatures in a breast cancer radiotherapy plan into 21 classes. Our results showed that ML based automation methods can be used for standardising naming conventions in a radiotherapy plan taking into consideration the inclusion of multiple modalities to better represent each volume.

Retrospective dataset and survey analyses identify gaps in data collection for craniopharyngioma and priorities of patients and families affected by the disease

IntroductionCraniopharyngioma is a rare, low-grade tumor located in the suprasellar region of the brain, near critical structures like the pituitary gland. Here, we concurrently investigate the status of clinical and genomic data in a retrospective craniopharyngioma cohort and survey-based data to better understand patient-relevant outcomes associated with existing therapies and provide a foundation to inform new treatment strategies. MethodsClinical, genomic, and outcome data for a retrospective cohort of patients with craniopharyngioma were collected and reviewed through the Children's Brain Tumor Network (CBTN) database. An anonymous survey was distributed to patients and families with a diagnosis of craniopharyngioma to understand their experiences throughout diagnosis and treatment. ResultsThe CBTN repository revealed a large proportion of patients (40 – 70%) with specimens that are available for sequencing but lacked relevant quality of life (QoL) and functional outcomes. Frequencies of reported patient comorbidities ranged from 20–35%, which is significantly lower than historically reported. Survey results from 159 patients/families identified differences in treatment considerations at time of diagnosis versus time of recurrence. In retrospective review, patients and families identified preference for therapy that would improve QoL, rather than decrease risk of recurrence (mean 3.9 vs. 4.4 of 5) and identified endocrine issues as having the greatest impact on patients’ lives. ConclusionsThis work highlights the importance of prospective collection of QoL and functional metrics alongside robust clinical and molecular correlates in individuals with craniopharyngioma. Such comprehensive measures will facilitate biologically relevant therapeutic strategies that also prioritize patient needs.

Stability of dosiomic features against variations in dose calculation: An analysis based on a cohort of prostate external beam radiotherapy patients.

Interest in using higher order features of the planned 3D dose distributions (i.e., dosiomics) to predict radiotherapy outcomes is growing. This is driving many retrospective studies where historical data are mined to train machine learning models; however, recent decades have seen considerable advances in dose calculation that could have a direct impact on the dosiomic features such studies seek to extract. Is it necessary to recalculate planned dose distributions using a common algorithm if retrospective datasets from different institutions are included? Does a change in dose calculation grid size part way through a retrospective cohort, introduce bias in the extracted dosiomic features? The purpose of this study is to assess the stability of dosiomic features against variations in three factors: the dose calculation algorithm type, version, and dose grid size. Dose distributions for 27 prostate patients who received EBRT were recalculated in the Eclipse Treatment Planning System (Varian Medical Systems, Palo Alto, California, USA) using two algorithms (AAA and Acuros XB), two versions (version 13.6 and 15.6), and three dose grids (2, 2.5s, and 3mm) - 12 dose distributions for each patient. Ninety-three dosiomic features were extracted from each dose distribution and each of the following regions-of-interest: high dose PTV (PTV_High), 1cm rind around PTV_High (PTV_Ring), low dose PTV (PTV_Low), rectum, and bladder using PyRadiomics. The coefficient of variation (CV) was calculated for each dosiomic feature. Hierarchical clustering was used to group features with high and low variability. Three-way repeated measures ANOVA was performed to investigate the effect of the three different factors on dosiomic features that were classified with high variation. Additionally, CVs were calculated for cumulative dose volume histograms (DVHs) to test their ability to detect the variations in dose distributions. For PTV_Ring, PTV_Low, and rectum, all the dosiomic features had low CV (average CV ≤ 0.26) across the varying dose calculation conditions. For PTV_High, six dosiomic features showed CV>0.26, and dose calculation algorithm type and grid size were the major sources of within-patient variation. For bladder, one dosiomic feature had average CV>0.26, but none of the three dose calculation-related factors led to a statistically significant variation. The CVs for all the DVHs were very small (CV<0.05). For all the regions-of-interest examined in this study, the majority of the dosiomic features were stable against variations in dose calculation; however, some of the dosiomic features for PTV_High and bladder had significant variations due to differences in dose calculation details. DVHs were detecting less variation than dosiomic features.

Use of claims data to predict the inpatient length of stay among U.S. stroke patients

BackgroundThe prediction of stroke inpatient length of stay (LOS) attracts much attention from researchers worldwide as it facilitates resource management and improves care. However, studies largely underutilized the claims data for predicting stroke LOS. The predictive models for the U.S. stroke population are understudied. PurposeTo evaluate the feasibility of using claims data for stroke LOS prediction and to identify new important predictors of LOS in U.S. stroke patients. MethodsData preparation, analyses, and predictive modeling processes were conducted on a retrospective dataset, including claims and EHR data about acute care stroke admissions during 2010–2018. Two tree-based models (i.e., the eXtreme Gradient Boosting (XGBoost) model and Categorical Boosting (CatBoost) model) were trained through 10-fold cross-validation and compared with a baseline model that did not include any predictors. The predictive performance was evaluated on the holdout set using mean absolute error (MAE) and root mean squared error (RMSE). Importance plots and SHAP (SHapley Additive exPlanations) plots were used to identify the important predictors. ResultsA total of 6102 stroke patients were included, with an average LOS of 6.4 days. The predictive models built using claims data (RMSE: 1.627; MAE: 0.530) performed similarly well as those built on the entire dataset, including additional variables from EHR (RMSE: 1.622; MAE: 0.533). Important predictors were admission channel and type, comorbidities (e.g., acute respiratory failure), medical services used (e.g., critical care, ambulance), facility characteristics (e.g., type, size), patient demographics, and patient socioeconomic status. Among these important predictors, admission channel and type, medical services including critical care and ambulance, facility type, and patient socioeconomic status were newly identified predictors not studied before. ConclusionClaims data are suitable for stroke LOS prediction. The newly identified important predictors from this study could be integrated with other existing key predictors identified in previous research to improve the prediction, thereby aiding in better stroke care management.

Bodyweight Changes During COVID-19 for Patients Diagnosed with Depression: A Retrospective Cohort Study.

Context: The COVID-19 pandemic led to an unprecedented lockdown of millions of Americans from the spring of 2020 to the fall of 2020. Studies done on the impact of COVID-19 on mental health and body weight have been important to our understanding of the effects of the pandemic. However, these studies on depression and BMI change have not identified a possible direction of the causality of the relationship between depression and body weight as affected by lockdown measures during a pandemic. Objective: The purpose of this study was to examine whether a diagnosis of depression is associated with changes in BMI during the COVID-19 pandemic for adults (aged ≥18 years). Study Design and Analysis: Retrospective cohort study Setting or Dataset: EHR data from a family medicine university clinic. Population Studied: Adults ≥18 years who visited the clinic within a 6-month period prior to lockdown (October 2019-March 2020) and at least once in the 6-month post-lockdown period (September 2020-March 2021). The lockdown period started in March 2020. 1,211 patients were included. Outcome Measures: Dependent variable: change in BMI; Primary independent variable: diagnosis of depression; Confounding variables: age, race/ethnicity, sex, medications, and chronic conditions Results: Mean age was 59.9 (sd=16.5). Patients were mostly female (n=770, 63.6%), white (n=678, 56.0%), and non-Hispanic (n=622, 51.4%). 18.7% (n=227) had a diagnosis of depression. There was a significant difference in BMI change (p<0.001) between the group diagnosed with depression (mean change=2.11, sd=1.9) and the group with no depression diagnosis (mean change=1.67, sd=1.9). Similarly, a diagnosis of depression significantly predicted BMI changes (p >0.001]). This association remained while including confounding variables in the model (p=0.009). Further statistical analysis showed that age between 31 and 50 significantly predicted BMI changes in those patients with no depression diagnosis while controlling for confounding variables (p=0.027). Conclusion: Individuals with depression had significant changes in BMI during the COVID-19 pandemic, and age predicted these changes in middle-aged adults (30-50 years old). These findings highlight the importance of identifying and following up with individuals with a diagnosis of depression to alleviate effects on their BMI during extended isolation. Identifying patients who might be susceptible to these changes could lead to patient health outcomes.

Why Did Home Care Personal Support Service Volumes Drop During the COVID-19 Pandemic? The Contributions of Client Choice and Personal Support Worker Availability.

Home care personal support service delivery decreased during the COVID-19 pandemic, and qualitative studies have suggested many potential contributors to these reductions. This paper provides insight into the source (client or provider) of reductions in home care service volumes early in the pandemic through analysis of a retrospective administrative dataset from a large provider organization. The percentage of authorized services not delivered was 17.2% in Wave 1, 12.6% in Wave 2 and 10.5% in Wave 3, nearing the pre-pandemic baseline of 8.9%. The dominant contribution to reduced home care service volumes was client-initiated holds and cancellations, collectively accounting for 99.3% of the service volume; missed care visits by the provider accounted for 0.7%. Worker availability also declined due to long-term absences (which increased 5-fold early in Wave 1 and remained 4× above baseline in Waves 2 and 3); short-term absences rose sharply for 6 early-pandemic weeks, then dropped below the pre-pandemic baseline. These data reveal that service volume reductions were primarily driven by client-initiated holds and cancellations; despite unprecedented decreases in Personal Support Worker availability, missed care did not increase, indicating that the decrease in demand was more substantial and occurred earlier than the decrease in worker availability.