Bayesian inference for a principal stratum estimand on recurrent events truncated by death

Recurrent events are often important endpoints in randomized clinical trials. For example, the number of recurrent disease-related hospitalizations may be considered as a clinically meaningful endpoint in cardiovascular studies. In some settings, the recurrent event process may be terminated by an event such as death, which makes it more challenging to define and estimate a causal treatment effect on recurrent event endpoints. In this paper, we focus on the principal stratum estimand, where the treatment effect of interest on recurrent events is defined among subjects who would be alive regardless of the assigned treatment. For the estimation of the principal stratum effect in randomized clinical trials, we propose a Bayesian approach based on a joint model of the recurrent event and death processes with a frailty term accounting for within-subject correlation. We also present Bayesian posterior predictive check procedures for assessing the model fit. The proposed approaches are demonstrated in the randomized Phase III chronic heart failure trial PARAGON-HF (NCT01920711).

An Efficient Transformer Model for Synthesizing Dual Energy CT from Single Energy Scanner

PURPOSE/OBJECTIVE(S)

Dual-energy CT can be used to optimize radiation treatment. Recently, deep learning has been demonstrated to synthesize high-energy CT images from low-energy ones for dose reduction and lower CT system burden. As the state-of-the-art deep learning architecture, the computation burden of Transformer increases quadratically with the feature size, making the model training resource-demanding or even infeasible. Here, we introduce an efficient transformer for the balance between CT image synthesis quality and computational burden.

MATERIALS/METHODS

The model is a U-shape deep neural network with encoders and decoders built by Transformer blocks. The model input is low-energy 100kVp CT image and the output is high-energy 140kVp one. Each block has a Self Channel Correlation Unit (SCCU) and a Self Spatial Attention Unit (SSAU). Local shortcuts are applied for both units. Under-sampling operation achieved by pixel shuffling is used to obtain multi-scale feature maps, and the transformer block is applied on each feature scale. Symmetric skip connection sending features from shallow layers to deep layers, thus an additional 1 × 1 convolution is used for feature fusion in each decoder. In a SCCU, the feature is first mapped to one Query, one Key, and one Value. Then the Query and the Key tensors perform matrix multiplication to compute cross covariance of feature channels. The channel correlation score can be obtained by normalization of the covariance, and it is used to weight the Value tensor. As a result, the model complexity only increases linearly with the feature size. Besides the channel weighting, we enhance spatial information using SSAU, where the feature is mapped to two tensors. One tensor after activation is used to point-wisely calibrate another tensor. Additional Transformer blocks are cascaded to the last decoder for feature refinement. Because of the structure similarity of low- and high-energy CT images, a global shortcut is used to ease model training. Clinical iodine contrast-enhanced dual energy CT image datasets of 19 patients are used in this study. The dual-energy scanning is performed by a SOMATOM Definition Flash DECT scanner. We split the datasets into training dataset of 15 patients, validation dataset of 1 patient, and testing dataset of 3 patients. The image size is 512 × 512 with pixel size 0.5 × 0.5 mm2.

RESULTS

The U-Net model with 1.95M parameters and 44.87G FLOPS achieved the averaged PSNR value of 44.55 dB (s.t.d. 1.34) and averaged RMSE value of 0.0060 (s.t.d. 0.001). In comparison, our efficient Transformer with 1.408M parameters and 31.375G FLOPS achieved the averaged PSNR value of 44.78 dB (s.t.d. 1.37) and RMSE value of 0.0059 (s.t.d. 0.001), demonstrating our model has better performance with small model size and less computation.

CONCLUSION

The efficient Transformer model allows high-resolution CT image synthesis with small model scale and computation burden from low-energy CT image.

Efficacy and safety of omalizumab in Chinese patients with anti-histamine refractory chronic spontaneous urticaria

Chronic spontaneous urticaria (CSU) is characterized by the spontaneous development of wheals, itching, and/or angioedema, for ≥6 weeks. In China, non-sedating H1-antihistamines (H1AH) are the recommended first-line treatment, with escalation up to 4× the standard dose in symptomatic patients to achieve control. Treatment options for Chinese patients who remain symptomatic on H1AH treatment are limited. This 20-week randomized, double-blind, placebo-controlled, parallel-group study investigated the efficacy and safety of omalizumab as an add-on therapy for the treatment of patients with CSU who remained symptomatic despite H1AH treatment in China. Adult patients (N = 418) diagnosed with refractory CSU for ≥6 months were randomized (2:2:1) to receive omalizumab 300 mg (OMA300), omalizumab 150 mg (OMA150) or placebo, subcutaneously, every 4 weeks. Primary outcome was change from baseline to Week 12 in weekly itch severity score (ISS7). Safety was assessed by rates of adverse events (AEs). Demographic and disease characteristics at baseline were comparable across treatment groups. At week 12, statistically significant greater decreases from baseline were observed in ISS7 with OMA300 (least square mean difference [LSM]: -4.23; 95% confidence interval [CI]: -5.70, -2.77; P < 0.001) and OMA150 (LSM: -3.79; 95% CI: -5.24, -2.33; P < 0.001) vs. placebo. Incidence of treatment-emergent AEs over 20 weeks was slightly higher with OMA300 (71.3%) compared to OMA150 and placebo groups (64.7% and 63.9%, respectively). The incidences of serious AEs were balanced between groups. This study demonstrated the efficacy and safety of omalizumab in Chinese adult patients with CSU who remained symptomatic despite H1AH therapy. Trials registered at clinicaltrials.gov NCT03328897 Date of registration: 1 November 2017.

Additive-Multiplicative Rates Model for Recurrent Event Data with Intermittently Observed Time-Dependent Covariates

Regression methods, including the proportional rates model and additive rates model, have been proposed to evaluate the effect of covariates on the risk of recurrent events. These two models have different assumptions on the form of the covariate effects. A more flexible model, the additive-multiplicative rates model, is considered to allow the covariates to have both additive and multiplicative effects on the marginal rate of recurrent event process. However, its use is limited to the cases where the time-dependent covariates are monitored continuously throughout the follow-up time. In practice, time-dependent covariates are often only measured intermittently, which renders the current estimation method for the additive-multiplicative rates model inapplicable. In this paper, we propose a semiparametric estimator for the regression coefficients of the additive-multiplicative rates model to allow intermittently observed time-dependent covariates. We present the simulation results for the comparison between the proposed method and the simple methods, including last covariate carried forward and linear interpolation, and apply the proposed method to an epidemiologic study aiming to evaluate the effect of time-varying streptococcal infections on the risk of pharyngitis among school children. The R package implementing the proposed method is available at www.github.com/TianmengL/rectime.

Additive rates model for recurrent event data with intermittently observed time-dependent covariates

Various regression methods have been proposed for analyzing recurrent event data. Among them, the semiparametric additive rates model is particularly appealing because the regression coefficients quantify the absolute difference in the occurrence rate of the recurrent events between different groups. Estimation of the additive rates model requires the values of time-dependent covariates being observed throughout the entire follow-up period. In practice, however, the time-dependent covariates are usually only measured at intermittent follow-up visits. In this paper, we propose to kernel smooth functions involving time-dependent covariates across subjects in the estimating function, as opposed to imputing individual covariate trajectories. Simulation studies show that the proposed method outperforms simple imputation methods. The proposed method is illustrated with data from an epidemiologic study of the effect of streptococcal infections on recurrent pharyngitis episodes.

Predicting drug approvals: The Novartis data science and artificial intelligence challenge

We describe a novel collaboration between academia and industry, an in-house data science and artificial intelligence challenge held by Novartis to develop machine-learning models for predicting drug-development outcomes, building upon research at MIT using data from Informa as the starting point. With over 50 cross-functional teams from 25 Novartis offices around the world participating in the challenge, the domain expertise of these Novartis researchers was leveraged to create predictive models with greater sophistication. Ultimately, two winning teams developed models that outperformed the baseline MIT model-areas under the curve of 0.88 and 0.84 versus 0.78, respectively-through state-of-the-art machine-learning algorithms and the use of newly incorporated features and data. In addition to validating the variables shown to be associated with drug approval in the earlier MIT study, the challenge also provided new insights into the drivers of drug-development success and failure.

[Research of artificial intelligence-based clinical decision support system for primary hepatocellular carcinoma]

Objective: To apply artificial intelligence technology in clinical real-world data of patients with primary hepatocellular carcinoma, explore the precise treatment of disease and build up artificial intelligence-based clinical decision support system. Methods: A total of 5 642 patients with primary hepatocellular carcinoma admitted to West China Hospital from July 2004 to June 2016 with complete follow-up records were included in the study. A merged model composed of multiple sub-classifiers was adopted to calculate therapy recommendation coefficient, and receiver operator characteristic curve was analyzed. Survival risk and recurrence risk were predicted by DeepSurv algorithm, and Kaplan-Meier survival curves were further compared among low, middle and high risk groups. Siamese-Net was applied to find similar patients. Results: The Top-1 and Top-2 accuracy of therapy recommendation coefficient reached 82.36% and 94.13% respectively. In internal verification of West China Hospital, the above-mentioned value reached 95.10% in accordance with multi-disciplinary team results. The C-index derived from survival risk model was 0.735 (95%CI:0.70-0.77), and the difference of Kaplan-Meier in pairwise comparison was of statistical significance under log-rank test (P<0.001). Meanwhile, the C-index derived from recurrence risk model was 0.705 (95%CI:0.68-0.73), and the difference of Kaplan-Meier in pairwise comparison was of statistical significance under log-rank test (P<0.001). Conclusions: The artificial intelligence-based clinical decision support system for primary hepatocellular carcinoma has can accurately make therapy recommendation and prognosis prediction for primary hepatocellular carcinoma.

[Early diagnosis and treatment for burn complicated with severe paroxysmal sympathetic hyper-activity]

Objective: To investigate the early diagnosis and treatment for burn complicated with severe paroxysmal sympathetic hyperactivity (PSH). Methods: Medical records of patients with burn complicated with severe PSH, admitted to our department from April 2016 to March 2019 and meeting the inclusion criteria were analyzed retrospectively. There were 4 males and 1 female, aged 17 months to 39 years, with an average of (21±16) years. During occurrence of PSH, the vital signs of patients were routinely monitored and oxygen were given. Other treatment included central venous catheterization and infusion of electrolyte solution, infusion of plasma according to patients' condition, use of opioid analgesics and benzodiazepine sedatives, physical cooling and drug cooling, and establishment or maintenance of artificial airway and use of ventilator. Heart rate was controlled below 120 beats per minute in adults and 140 beats per minute in children with comprehensive treatment dominated by analgesia and sedation. Besides, single or multiple vasoactive agents, even in large doses were used to maintain normal blood pressure of patients. The occurrence characteristics, time, and treatment outcome of PSH were analyzed. Results: PSH happened rapidly, with a sharp increase in several minutes to dozens of minutes. Five patients were with symptoms such as high body temperature, shortness of breath, very fast heart rate, normal or elevated systolic blood pressure, hyperhidrosis, and dystonia at the onset. The symptoms occurred simultaneously or successively. According to the Clinical Feature Scale, the above-mentioned 6 indexes achieved the highest score of 3 points except of systolic blood pressure. Four patients showed dilated pupils and impaired consciousness. Among the patients, PSH occurred in the acute exudation stage in 3 patients, in the fluid reabsorption stage in 1 patient, and in the late repair stage in 1 patient. PSH of patients lasted for 3 hours to 12 days. The symptoms of 4 patients were effectively controlled, and 1 patient died of deterioration. No PSH occurred in the cured patients during follow-up of 3 to 14 months. Conclusions: Burn complicated with PSH can occur at any time before wound repair and in patients with different injury conditions. The causes of PSH include sudden burn, persistent pain, fright and fear, strange environment, low blood volume, and other adverse stimuli, and PSH is more likely to occur in children with underdeveloped brain function. Intravenous infusion of analgesics sedatives, physical therapy and medication to lower body temperature, stabilizing blood pressure and respiration are effective measures to treat PSH. PSH should be distinguished from the common complications of burns, such as sepsis, cerebral edema, hyperpyretic convulsion, transfusion response, stress disorder, etc.

Sensitivity of Volumetric Magnetic Resonance Imaging and Magnetic Resonance Spectroscopy to Progression of Spinocerebellar Ataxia Type 1

Background

Spinocerebellar ataxia type 1 (SCA1) causes progressive degeneration of the cerebellum and brainstem. Volumetric magnetic resonance imaging (MRI) was shown to be more sensitive to disease progression than the most sensitive clinical measure, the Scale for the Assessment and Rating of Ataxia (SARA), in longitudinal studies, and magnetic resonance spectroscopy (MRS) was shown to detect neurochemical abnormalities with high sensitivity cross-sectionally in SCA1.

Objectives

The objectives of this study were to compare the sensitivities to change of volumetric MRI, MRS, and SARA in a 3-year longitudinal study in SCA1.

Methods

A total of 16 early-to-moderate stage patients with SCA1 (SARA 0-14) and 21 matched healthy participants were scanned up to 3 times with 1.5-year intervals. Ataxia severity was assessed with SARA. T₁-weighted images and magnetic resonance spectra from the cerebellar vermis, cerebellar white matter, and pons were acquired at 3T.

Results

The pontine total N-acetylaspartate-to-myo-inositol ratio was the most sensitive MRS measure to change (-3.9 ± 4.6%/yr in SCA1 vs. -0.3 ± 3.5%/yr in controls; P < 0.02), and the pontine volume was the most sensitive MRI measure to change (-2.6 ± 1.2%/yr in SCA1 vs. -0.1 ± 1.2 in controls; P < 0.02). Effect size (mean percent change/standard deviation of percent change) of pontine volume was highest (-2.13) followed by pontine N-acetylaspartate-to-myo-inositol ratio (-0.84) and SARA (+0.60). The pontine N-acetylaspartate-to-myo-inositol ratio was abnormal for 1 premanifest patient at all visits and predicted study withdrawal as a result of disease progression in 3 patients.

Conclusion

Both MRI and MRS were more sensitive to disease progression than SARA in SCA1. Pontine volume was most sensitive to change, whereas MRS may have more sensitivity at the premanifest stage and predictive value for disease progression.

[The downbeat nystagmus in 2 cases with unilateral Meniere's disease]

We retrospectively analyzed 2 patients of unilateral Meniere's disease with downbeat nystagmus. Audiometric test, the Dix-Hallpike maneuver and the supine roll test were performed. Both patients presented with spontaneous nystagmus with downbeat components, and downbeat components were exaggerated by Dix-Hallpike maneuver. Spontaneous and positional nystagmus with downbeat component can occur in Meniere's disease, which may be resulted from excitation of three semicircular canals.

[One case of severely burned patient complicated by acute hemorrhagic necrotizing enteritis and fungal infection]

One severely burned patient, caused by heat lead slag and combined with shock, was hospitalized in our burn unit on 2nd June, 2016. The patient received treatments including anti-shock, intensive care, anti-infection, and organ protection. On post injury day 16, the patient suffered outbreak of acute hemorrhagic necrotizing enteritis after eating dumplings. Plasma and albumin were given, octreotide was intravenously infused to inhibit the secretion of intestinal fluid, the broad-spectrum antibiotics were used for anti-infection, abdominal puncture and drainage were performed, sodium tanshinone ⅡA sulfonate was applied to improve the intestinal microcirculation, ulinastatin was applied to alleviate inflammatory reaction, somatostatin was given to reduce intestinal bleeding, and voriconazole was given for antifungal treatment. The patient gradually recovered and was finally cured and discharged. Among critically ill patients, gastrointestinal tract is not only the initiating organ of sepsis, but also one of the target organs which can be easily damaged during sepsis. This case reminds us the importance of gastrointestinal management in severely burned patients.

Neurochemical abnormalities in premanifest and early spinocerebellar ataxias

OBJECTIVE

To investigate whether early neurochemical abnormalities are detectable by high-field magnetic resonance spectroscopy (MRS) in individuals with spinocerebellar ataxias (SCAs) 1, 2, 3, and 6, including patients without manifestation of ataxia.

METHODS

A cohort of 100 subjects (N = 18-21 in each SCA group, including premanifest mutation carriers; mean score on the Scale for the Assessment and Rating of Ataxia [SARA] <10 for all genotypes, and 22 matched controls) was scanned at 7 Tesla to obtain neurochemical profiles of the cerebellum and brainstem. A novel multivariate approach (distance-weighted discrimination) was used to combine regional profiles into an "MRS score."

RESULTS

MRS scores robustly distinguished individuals with SCA from controls, with misclassification rates of 0% (SCA2), 2% (SCA3), 5% (SCA1), and 17% (SCA6). Premanifest mutation carriers with estimated disease onset within 10 years had MRS scores in the range of early-manifest SCA subjects. Levels of neuronal and glial markers significantly correlated with SARA and an Activities of Daily Living score in subjects with SCA. Regional neurochemical alterations were different between SCAs at comparable disease severity, with SCA2 displaying the most extensive neurochemical abnormalities, followed by SCA1, SCA3, and SCA6.

INTERPRETATION

Neurochemical abnormalities are detectable in individuals before manifest disease, which may allow premanifest enrollment in future SCA trials. Correlations with ataxia and quality-of-life scores show that neurochemical levels can serve as clinically meaningful endpoints in trials. Ranking of SCA types by degree of neurochemical abnormalities indicates that the neurochemistry may reflect synaptic function or density. Ann Neurol 2018;83:816-829.

Induced smoothing for rank-based regression with recurrent gap time data

Various semiparametric regression models have recently been proposed for the analysis of gap times between consecutive recurrent events. Among them, the semiparametric accelerated failure time (AFT) model is especially appealing owing to its direct interpretation of covariate effects on the gap times. In general, estimation of the semiparametric AFT model is challenging because the rank-based estimating function is a nonsmooth step function. As a result, solutions to the estimating equations do not necessarily exist. Moreover, the popular resampling-based variance estimation for the AFT model requires solving rank-based estimating equations repeatedly and hence can be computationally cumbersome and unstable. In this paper, we extend the induced smoothing approach to the AFT model for recurrent gap time data. Our proposed smooth estimating function permits the application of standard numerical methods for both the regression coefficients estimation and the standard error estimation. Large-sample properties and an asymptotic variance estimator are provided for the proposed method. Simulation studies show that the proposed method outperforms the existing nonsmooth rank-based estimating function methods in both point estimation and variance estimation. The proposed method is applied to the data analysis of repeated hospitalizations for patients in the Danish Psychiatric Center Register.

The lung tissue microbiota of mild and moderate chronic obstructive pulmonary disease

BACKGROUND

Oral taxa are often found in the chronic obstructive pulmonary disease (COPD) lung microbiota, but it is not clear if this is due to a physiologic process such as aspiration or experimental contamination at the time of specimen collection.

METHODS

Microbiota samples were obtained from nine subjects with mild or moderate COPD by swabbing lung tissue and upper airway sites during lung lobectomy. Lung specimens were not contaminated with upper airway taxa since they were obtained surgically. The microbiota were analyzed with 16S rRNA gene qPCR and 16S rRNA gene hypervariable region 3 (V3) sequencing. Data analyses were performed using QIIME, SourceTracker, and R.

RESULTS

Streptococcus was the most common genus in the oral, bronchial, and lung tissue samples, and multiple other taxa were present in both the upper and lower airways. Each subject's own bronchial and lung tissue microbiota were more similar to each other than were the bronchial and lung tissue microbiota of two different subjects (permutation test, p = 0.0139), indicating more within-subject similarity than between-subject similarity at these two lung sites. Principal coordinate analysis of all subject samples revealed clustering by anatomic sampling site (PERMANOVA, p = 0.001), but not by subject. SourceTracker analysis found that the sources of the lung tissue microbiota were 21.1% (mean) oral microbiota, 8.7% nasal microbiota, and 70.1% unknown. An analysis using the neutral theory of community ecology revealed that the lung tissue microbiota closely reflects the bronchial, oral, and nasal microbiota (immigration parameter estimates 0.69, 0.62, and 0.74, respectively), with some evidence of ecologic drift occurring in the lung tissue.

CONCLUSION

This is the first study to evaluate the mild-moderate COPD lung tissue microbiota without potential for upper airway contamination of the lung samples. In our small study of subjects with COPD, we found oral and nasal bacteria in the lung tissue microbiota, confirming that aspiration is a source of the COPD lung microbiota.

[Validation of the Chinese version of the uterine fibroid symptom and health-related quality of life]

Objective: To evaluate the reliability and validity of the Chinese version of the uterine fibroid symptom and health-related quality of life (UFS-QOL) in patients with uterine fibroid. Methods: The original English UFS-QOL was translated into Chinese and linguistically validated following the cross-cultural adaptation of health-related quality of life measures. Patients recruited randomly from Peking Union Medical College Hospital from June 2013 to December 2016 were scheduled for two visits with 2 weeks apart, and they were surveyed through the Chinese version of UFS-QOL and short-form 12-item health survey (SF-12). The reliability and validity of the Chinese version of UFS-QOL were evaluated. Results: A total of 190 uterine fibroid patients who met the criteria participated this study. The Chinese version of UFS-QOL had a high internal consistency (Cronbach's α, uterine fibroid symptom severity scale:0.912, health-related quality of life scale: 0.976) and high test-retest reliability (intraclass correlation coefficient, 0.572-0.951, P<0.01). The symptom severity scores and SF-12 scores were negatively correlated (r=-0.813, P<0.01); the health-related quality of life scores and SF-12 scores were positively correlated (r=0.620, P<0.01). The factor analysis showed good construct validity. Conclusion: Psychometric testing supports the reliability and validity of the Chinese version of UFS-QOL as an disease-specific measure of health-related quality of life.

Discriminating sample groups with multi-way data

High-dimensional linear classifiers, such as distance weighted discrimination (DWD) and versions of the support vector machine (SVM), are commonly used in biomedical research to distinguish groups of subjects based on a large number of features. However, their use is limited to applications where a single vector of features is measured for each subject. In practice, data are often multi-way, or measured over multiple dimensions. For example, metabolite abundance may be measured over multiple regions or tissues, or gene expression may be measured over multiple time points, for the same subjects. We propose a framework for linear classification of high-dimensional multi-way data, in which coefficients can be factorized into weights that are specific to each dimension. More generally, the coefficients for each measurement in a multi-way dataset are assumed to have low-rank structure. This framework extends existing classification techniques from single vector to multi-way features, and we have implemented multi-way versions of SVM and DWD. We describe informative simulation results, and apply multi-way DWD to data for two very different clinical research studies. The first study uses magnetic resonance spectroscopy metabolite data over multiple brain regions to compare participants with and without spinocerebellar ataxia; the second uses publicly available gene expression time-course data to compare degrees of treatment response among patients with multiple sclerosis. Our multi-way method can improve performance and simplify interpretation over naive applications of full rank linear and non-linear classification to multi-way data. The R package is available at https://github.com/lockEF/MultiwayClassification.

Test-retest reproducibility of neurochemical profiles with short-echo, single-voxel MR spectroscopy at 3T and 7T

PURPOSE

To determine the test-retest reproducibility of neurochemical concentrations obtained with a highly optimized, short-echo, single-voxel proton MR spectroscopy (MRS) pulse sequence at 3T and 7T using state-of-the-art hardware.

METHODS

A semi-LASER sequence (echo time = 26-28 ms) was used to acquire spectra from the posterior cingulate and cerebellum at 3T and 7T from six healthy volunteers who were scanned four times weekly on both scanners. Spectra were quantified with LCModel.

RESULTS

More neurochemicals were quantified with mean Cramér-Rao lower bounds (CRLBs) ≤20% at 7T than at 3T despite comparable frequency-domain signal-to-noise ratio. Whereas CRLBs were lower at 7T (P < 0.05), between-session coefficients of variance (CVs) were comparable at the two fields with 64 transients. Five metabolites were quantified with between-session CVs ≤5% at both fields. Analysis of subspectra showed that a minimum achievable CV was reached with a lower number of transients at 7T for multiple metabolites and that between-session CVs were lower at 7T than at 3T with fewer than 64 transients.

CONCLUSION

State-of-the-art MRS methodology allows excellent reproducibility for many metabolites with 5-min data averaging on clinical 3T hardware. Sensitivity and resolution advantages at 7T are important for weakly represented metabolites, short acquisitions, and small volumes of interest. Magn Reson Med 76:1083-1091, 2016. © 2015 Wiley Periodicals, Inc.