Negation detection in Dutch clinical texts: an evaluation of rule-based and machine learning methods.

When developing models for clinical information retrieval and decision support systems, the discrete outcomes required for training are often missing. These labels need to be extracted from free text in electronic health records. For this extraction process one of the most important contextual properties in clinical text is negation, which indicates the absence of findings. We aimed to improve large scale extraction of labels by comparing three methods for negation detection in Dutch clinical notes. We used the Erasmus Medical Center Dutch Clinical Corpus to compare a rule-based method based on ContextD, a biLSTM model using MedCAT and (finetuned) RoBERTa-based models. We found that both the biLSTM and RoBERTa models consistently outperform the rule-based model in terms of F1 score, precision and recall. In addition, we systematically categorized the classification errors for each model, which can be used to further improve model performance in particular applications. Combining the three models naively was not beneficial in terms of performance. We conclude that the biLSTM and RoBERTa-based models in particular are highly accurate accurate in detecting clinical negations, but that ultimately all three approaches can be viable depending on the use case at hand.

Microarray analysis of autoantibodies can identify future Systemic Lupus Erythematosus patients.

OBJECTIVE: Reliable early ascertainment in patients with SLE is important to prevent the accumulation of irreversible organ damage. Autoantibodies are often present in the serum of patients before the first symptoms arise, therefore they are of potential use as early diagnostic tools. METHODS: We used a custom-made antibody microarray containing 57 autoantigens to analyze serum samples of 1519 patients previously tested for anti-dsDNA and 361 samples of self-reported healthy blood bank donors (BBD). The 1519 patients included 483 patients with SLE, 346 patients with other immune mediated inflammatory diseases (IMID), 218 patient controls without relevant clinical symptoms (Non-IMID), and 472 patients that did not fit in any of the previous groups (Rest). The Non-IMID and BBD groups were used individually to create multivariable prediction models to distinguish samples of patients with SLE from these control groups. We subsequently used these models to predict the outcome for samples of patients who developed SLE while in follow-up (pre-SLE). RESULTS: Out of 1036 patients with no diagnosis of SLE at the moment of sample collection, 17 patients developed SLE while in follow-up (mean time to diagnosis 7.2 months). The best performing model (AUC 0.83) identified 9 out of 17 (53%) pre-SLE samples as SLE, with a specificity of 94%. CONCLUSION: Serum samples of patients who will develop SLE in the future already show a shift of the autoantibody profile prior to diagnosis. In this study, we show that these autoantibody profiles can be used to identify these future SLE patients.

Effects of tDCS on the attentional blink revisited: A statistical evaluation of a replication attempt.

The attentional blink (AB) phenomenon reveals a bottleneck of human information processing: the second of two targets is often missed when they are presented in rapid succession among distractors. In our previous work, we showed that the size of the AB can be changed by applying transcranial direct current stimulation (tDCS) over the left dorsolateral prefrontal cortex (lDLPFC) (London & Slagter, Journal of Cognitive Neuroscience, 33, 756-68, 2021). Although AB size at the group level remained unchanged, the effects of anodal and cathodal tDCS were negatively correlated: if a given individual's AB size decreased from baseline during anodal tDCS, their AB size would increase during cathodal tDCS, and vice versa. Here, we attempted to replicate this finding. We found no group effects of tDCS, as in the original study, but we no longer found a significant negative correlation. We present a series of statistical measures of replication success, all of which confirm that both studies are not in agreement. First, the correlation here is significantly smaller than a conservative estimate of the original correlation. Second, the difference between the correlations is greater than expected due to sampling error, and our data are more consistent with a zero-effect than with the original estimate. Finally, the overall effect when combining both studies is small and not significant. Our findings thus indicate that the effects of lDPLFC-tDCS on the AB are less substantial than observed in our initial study. Although this should be quite a common scenario, null findings can be difficult to interpret and are still under-represented in the brain stimulation and cognitive neuroscience literatures. An important auxiliary goal of this paper is therefore to provide a tutorial for other researchers, to maximize the evidential value from null findings.

Identification and prediction of difficult-to-treat rheumatoid arthritis patients in structured and unstructured routine care data: results from a hackathon.

BACKGROUND: The new concept of difficult-to-treat rheumatoid arthritis (D2T RA) refers to RA patients who remain symptomatic after several lines of treatment, resulting in a high patient and economic burden. During a hackathon, we aimed to identify and predict D2T RA patients in structured and unstructured routine care data. METHODS: Routine care data of 1873 RA patients were extracted from the Utrecht Patient Oriented Database. Data from a previous cross-sectional study, in which 152 RA patients were clinically classified as either D2T or non-D2T, served as a validation set. Machine learning techniques, text mining, and feature importance analyses were performed to identify and predict D2T RA patients based on structured and unstructured routine care data. RESULTS: We identified 123 potentially new D2T RA patients by applying the D2T RA definition in structured and unstructured routine care data. Additionally, we developed a D2T RA identification model derived from a feature importance analysis of all available structured data (AUC-ROC 0.88 (95% CI 0.82-0.94)), and we demonstrated the potential of longitudinal hematological data to differentiate D2T from non-D2T RA patients using supervised dimension reduction. Lastly, using data up to the time of starting the first biological treatment, we predicted future development of D2TRA (AUC-ROC 0.73 (95% CI 0.71-0.75)). CONCLUSIONS: During this hackathon, we have demonstrated the potential of different techniques for the identification and prediction of D2T RA patients in structured as well as unstructured routine care data. The results are promising and should be optimized and validated in future research.

Sustaining attention for a prolonged period of time increases temporal variability in cortical responses.

Our ability to stay focused is limited: prolonged performance of a task typically results in mental fatigue and decrements in performance over time. This so-called vigilance decrement has been attributed to depletion of attentional resources, though other factors such as reductions in motivation likely also play a role. In this study, we examined three electroencephalography (EEG) markers of attentional control, to elucidate which stage of attentional processing is most affected by time-on-task and motivation. To elicit the vigilance decrement, participants performed a sustained attention task for 80 min without breaks. After 60 min, participants were motivated by an unexpected monetary incentive to increase performance in the final 20 min. We found that task performance and self-reported motivation declined rapidly, reaching stable levels well before the motivation manipulation was introduced. Thereafter, motivation increased back up to the initial level, and remained there for the final 20 min. While task performance also increased, it did not return to the initial level, and fell to the lowest level overall during the final 10 min. This pattern of performance changes was mirrored by the trial-to-trial consistency of the phase of theta (3-7 Hz) oscillations, an index of the variability in timing of the neural response to the stimulus. As task performance decreased, temporal variability increased, suggesting that attentional stability is crucial for sustained attention performance. The effects of attention on our two other EEG measures-early P1/N1 event-related potentials (ERPs) and pre-stimulus alpha (9-14 Hz) power-did not change with time-on-task or motivation. In sum, these findings show that the vigilance decrement is accompanied by a decline in only some facets of attentional control, which cannot be fully brought back online by increases in motivation. The vigilance decrement might thus not occur due to a single cause, but is likely multifactorial in origin.

No Evidence that Predictions and Attention Modulate the First Feedforward Sweep of Cortical Information Processing.

Predictive coding models propose that predictions (stimulus likelihood) reduce sensory signals as early as primary visual cortex (V1), and that attention (stimulus relevance) can modulate these effects. Indeed, both prediction and attention have been shown to modulate V1 activity, albeit with fMRI, which has low temporal resolution. This leaves it unclear whether these effects reflect a modulation of the first feedforward sweep of visual information processing and/or later, feedback-related activity. In two experiments, we used electroencephalography and orthogonally manipulated spatial predictions and attention to address this issue. Although clear top-down biases were found, as reflected in pre-stimulus alpha-band activity, we found no evidence for top-down effects on the earliest visual cortical processing stage (<80 ms post-stimulus), as indexed by the amplitude of the C1 event-related potential component and multivariate pattern analyses. These findings indicate that initial visual afferent activity may be impenetrable to top-down influences by spatial prediction and attention.

No Evidence That Frontal Eye Field tDCS Affects Latency or Accuracy of Prosaccades.

Transcranial direct current stimulation (tDCS) may be used to directly affect neural activity from outside of the skull. However, its exact physiological mechanisms remain elusive, particularly when applied to new brain areas. The frontal eye field (FEF) has rarely been targeted with tDCS, even though it plays a crucial role in control of overt and covert spatial attention. Here, we investigate whether tDCS over the FEF can affect the latency and accuracy of saccadic eye movements. Twenty-six participants performed a prosaccade task in which they made eye movements to a sudden-onset eccentric visual target (lateral saccades). After each lateral saccade, they made an eye movement back to the center (center saccades). The task was administered before, during, and after anodal or cathodal tDCS over the FEF, in a randomized, double-blind, within-subject design. One previous study (Kanai et al., 2012) found that anodal tDCS over the FEF decreased the latency of saccades contralateral to the stimulated hemisphere. We did not find the same effect: neither anodal nor cathodal tDCS influenced the latency of lateral saccades. tDCS also did not affect accuracy of lateral saccades (saccade endpoint deviation and saccade endpoint variability). For center saccades, we found some differences between the anodal and cathodal sessions, but these were not consistent across analyses (latency, endpoint variability), or were already present before tDCS onset (endpoint deviation). We tried to improve on the design of Kanai et al. (2012) in several ways, including the tDCS duration and electrode montage, which could explain the discrepant results. Our findings add to a growing number of null results, which have sparked concerns that tDCS outcomes are highly variable. Future studies should aim to establish the boundary conditions for FEF-tDCS to be effective, in addition to increasing sample size and adding additional controls such as a sham condition. At present, we conclude that it is unclear whether eye movements or other aspects of spatial attention can be affected through tDCS of the frontal eye fields.

No Differential Effects of Two Different Alpha-Band Electrical Stimulation Protocols Over Fronto-Parietal Regions on Spatial Attention.

In a previous study using transcranial alternating current stimulation (tACS), we found preliminary evidence that phase coherence in the alpha band (8-12 Hz) within the fronto-parietal network may critically support top-down control of spatial attention (van Schouwenburg et al., 2017). Specifically, synchronous alpha-band stimulation over the right frontal and parietal cortex (0° relative phase) was associated with changes in performance and fronto-parietal coherence during a spatial attention task as compared to sham stimulation. In the current study, we firstly aimed to replicate these findings with synchronous tACS. Second, we extended our previous protocol by adding a second tACS condition in which the right frontal and parietal cortex were stimulated in a desynchronous fashion (180° relative phase), to test the specificity of the changes observed in our previous study. Participants (n = 23) were tested in three different sessions in which they received either synchronous, desynchronous, or sham stimulation over the right frontal and parietal cortex. In contrast to our previous study, we found no spatially selective effects of stimulation on behavior or coherence in either stimulation protocol compared to sham. We highlight some of the differences in study design that may have contributed to this discrepancy in findings and more generally may determine the effectiveness of tACS.

Contributions of the Ventral Striatum to Conscious Perception: An Intracranial EEG Study of the Attentional Blink.

The brain is limited in its capacity to consciously process information, necessitating gating of information. While conscious perception is robustly associated with sustained, recurrent interactions between widespread cortical regions, subcortical regions, including the striatum, influence cortical activity. Here, we examined whether the ventral striatum, given its ability to modulate cortical information flow, contributes to conscious perception. Using intracranial EEG, we recorded ventral striatum activity while 7 patients performed an attentional blink task in which they had to detect two targets (T1 and T2) in a stream of distractors. Typically, when T2 follows T1 within 100-500 ms, it is often not perceived (i.e., the attentional blink). We found that conscious T2 perception was influenced and signaled by ventral striatal activity. Specifically, the failure to perceive T2 was foreshadowed by a T1-induced increase in α and low ß oscillatory activity as early as 80 ms after T1, indicating that the attentional blink to T2 may be due to very early T1-driven attentional capture. Moreover, only consciously perceived targets were associated with an increase in θ activity between 200 and 400 ms. These unique findings shed new light on the mechanisms that give rise to the attentional blink by revealing that conscious target perception may be determined by T1 processing at a much earlier processing stage than traditionally believed. More generally, they indicate that ventral striatum activity may contribute to conscious perception, presumably by gating cortical information flow. SIGNIFICANCE STATEMENT: What determines whether we become aware of a piece of information or not? Conscious access has been robustly associated with activity within a distributed network of cortical regions. Using intracranial electrophysiological recordings during an attentional blink task, we tested the idea that the ventral striatum, because of its ability to modulate cortical information flow, may contribute to conscious perception. We find that conscious perception is influenced and signaled by ventral striatal activity. Short-latency (80-140 ms) striatal responses to a first target determined conscious perception of a second target. Moreover, conscious perception of the second target was signaled by longer-latency (200-400 ms) striatal activity. These results suggest that the ventral striatum may be part of a subcortical network that influences conscious experience.