Many sensitive domains — such as the clinical domain — lack widely available datasets due to privacy risks. The increasing generative capabilities of large language models (LLMs) have made synthetic datasets a viable path forward. In this study, we domain-adapt LLMs to the clinical domain and generate synthetic clinical texts that are machine-annotated with tags for personally identifiable information using capable encoder-based NER models. The synthetic corpora are then used to train synthetic NER models. The results show that training NER models using synthetic corpora incurs only a small drop in predictive performance. The limits of this process are investigated in a systematic ablation study — using both Swedish and Spanish data. Our analysis shows that smaller datasets can be sufficient for domain-adapting LLMs for data synthesis. Instead, the effectiveness of this process is almost entirely contingent on the performance of the machine-annotating NER models trained using the original data.

Research suggests that using generic language models in specialized domains may be sub-optimal due to significant domain differences. As a result, various strategies for developing domain-specific language models have been proposed, including techniques for adapting an existing generic language model to the target domain, e.g. through various forms of vocabulary modifications and continued domain-adaptive pretraining with in-domain data. Here, an empirical investigation is carried out in which various strategies for adapting a generic language model to the clinical domain are compared to pretraining a pure clinical language model. Three clinical language models for Swedish, pretrained for up to ten epochs, are fine-tuned and evaluated on several downstream tasks in the clinical domain. A comparison of the language models’ downstream performance over the training epochs is conducted. The results show that the domain-specific language models outperform a general-domain language model; however, there is little difference in performance of the various clinical language models. However, compared to pretraining a pure clinical language model with only in-domain data, leveraging and adapting an existing general-domain language model requires fewer epochs of pretraining with in-domain data.

Automatic de-identification is a cost-effective and straightforward way of removing large amounts of personally identifiable information from large and sensitive corpora. However, these systems also introduce errors into datasets due to their imperfect precision. These corruptions of the data may negatively impact the utility of the de-identified dataset. This paper de-identifies a very large clinical corpus in Swedish either by removing entire sentences containing sensitive data or by replacing sensitive words with realistic surrogates. These two datasets are used to perform domain adaptation of a general Swedish BERT model. The impact of the de-identification techniques is assessed by training and evaluating the models using six clinical downstream tasks. The results are then compared to a similar BERT model domain-adapted using an unaltered version of the clinical corpus. The results show that using an automatically de-identified corpus for domain adaptation does not negatively impact downstream performance. We argue that automatic de-identification is an efficient way of reducing the privacy risks of domain-adapted models and that the models created in this paper should be safe to distribute to other academic researchers.

An abundance of electronic health records (EHR) is produced every day within healthcare. The records possess valuable information for research and future improvement of healthcare. Multiple efforts have been done to protect the integrity of patients while making electronic health records usable for research by removing personally identifiable information in patient records. Supervised machine learning approaches for de-identification of EHRs need annotated data for training, annotations that are costly in time and human resources. The annotation costs for clinical text is even more costly as the process must be carried out in a protected environment with a limited number of annotators who must have signed confidentiality agreements. In this paper is therefore, a semi-supervised method proposed, for automatically creating high-quality training data. The study shows that the method can be used to improve recall from 84.75% to 89.20% without sacrificing precision to the same extent, dropping from 95.73% to 94.20%. The model’s recall is arguably more important for de-identification than precision.

Adverse drug events (ADEs) are heavily under-reported in electronic health records (EHRs). Alerting systems that are able to detect potential ADEs on the basis of patient-specific EHR data would help to mitigate this problem. To that end, the use of machine learning has proven to be both efficient and effective; however, challenges remain in representing the heterogeneous EHR data, which moreover tends to be high-dimensional and exceedingly sparse, in a manner conducive to learning high-performing predictive models. Prior work has shown that distributional semantics - that is, natural language processing methods that, traditionally, model the meaning of words in semantic (vector) space on the basis of co-occurrence information - can be exploited to create effective representations of sequential EHR data of various kinds. When modeling data in semantic space, an important design decision concerns the size of the context window around an object of interest, which governs the scope of co-occurrence information that is taken into account and affects the composition of the resulting semantic space. Here, we report on experiments conducted on 27 clinical datasets, demonstrating that performance can be significantly improved by modeling EHR data in ensembles of semantic spaces, consisting of multiple semantic spaces built with different context window sizes. A follow-up investigation is conducted to study the impact on predictive performance as increasingly more semantic spaces are included in the ensemble, demonstrating that accuracy tends to improve with the number of semantic spaces, albeit not monotonically so. Finally, a number of different strategies for combining the semantic spaces are explored, demonstrating the advantage of early (feature) fusion over late (classifier) fusion. Semantic space ensembles allow multiple views of (sparse) data to be captured (densely) and thereby enable improved performance to be obtained on the task of detecting ADEs in EHRs.

The enormous amounts of data that are continuously recorded in electronic health record systems offer ample opportunities for data science applications to improve healthcare. There are, however, challenges involved in using such data for machine learning, such as high dimensionality and sparsity, as well as an inherent heterogeneity that does not allow the distinct types of clinical data to be treated in an identical manner. On the other hand, there are also similarities across data types that may be exploited, e.g., the possibility of representing some of them as sequences. Here, we apply the notions underlying distributional semantics, i.e., methods that model the meaning of words in semantic (vector) space on the basis of co-occurrence information, to four distinct types of clinical data: free-text notes, on the one hand, and clinical events, in the form of diagnosis codes, drug codes and measurements, on the other hand. Each semantic space contains continuous vector representations for every unique word and event, which can then be used to create representations of, e.g., care episodes that, in turn, can be exploited by the learning algorithm. This approach does not only reduce sparsity, but also takes into account, and explicitly models, similarities between various items, and it does so in an entirely data-driven fashion. Here, we report on a series of experiments using the random forest learning algorithm that demonstrate the effectiveness, in terms of accuracy and area under ROC curve, of the proposed representation form over the commonly used bag-of-items counterpart. The experiments are conducted on 27 real datasets that each involves the (binary) classification task of detecting a particular adverse drug event. It is also shown that combining structured and unstructured data leads to significant improvements over using only one of them.

• A corpus of Swedish clinical notes was annotated for adverse drug event information. • Detecting adverse drug events in clinical notes can support pharmacovigilance. • Modeling context with distributional semantics yielded better predictive models. • Distributed word representations allowed more context information to be incorporated. • Inter-sentential relations between drugs and disorders/findings are hard to detect. For the purpose of post-marketing drug safety surveillance, which has traditionally relied on the voluntary reporting of individual cases of adverse drug events (ADEs), other sources of information are now being explored, including electronic health records (EHRs), which give us access to enormous amounts of longitudinal observations of the treatment of patients and their drug use. Adverse drug events, which can be encoded in EHRs with certain diagnosis codes, are, however, heavily underreported. It is therefore important to develop capabilities to process, by means of computational methods, the more unstructured EHR data in the form of clinical notes, where clinicians may describe and reason around suspected ADEs. In this study, we report on the creation of an annotated corpus of Swedish health records for the purpose of learning to identify information pertaining to ADEs present in clinical notes. To this end, three key tasks are tackled: recognizing relevant named entities (disorders, symptoms, drugs), labeling attributes of the recognized entities (negation, speculation, temporality), and relationships between them (indication, adverse drug event). For each of the three tasks, leveraging models of distributional semantics – i.e., unsupervised methods that exploit co-occurrence information to model, typically in vector space, the meaning of words – and, in particular, combinations of such models, is shown to improve the predictive performance. The ability to make use of such unsupervised methods is critical when faced with large amounts of sparse and high-dimensional data, especially in domains where annotated resources are scarce.

Creating sufficiently large annotated resources for supervised machine learning, and doing so for every problem and every domain, is prohibitively expensive. Techniques that leverage large amounts of unlabeled data, which are often readily available, may decrease the amount of data that needs to be annotated to obtain a certain level of performance, as well as improve performance when large annotated resources are indeed available. Here, the development of one such method is presented, where semantic features are generated by exploiting the available annotations to learn prototypical (vector) representations of each named entity class in semantic space, constructed by employing a model of distributional semantics (random indexing) over a large, unannotated, in-domain corpus. Binary features that describe whether a given word belongs to a specific named entity class are provided to the learning algorithm; the feature values are determined by calculating the (cosine) distance in semantic space to each of the learned prototype vectors and ascertaining whether they are below or above a given threshold, set to optimize Fβ-score. The proposed method is evaluated empirically in a series of experiments, where the case is health-record deidentification, a task that involves identifying protected health information (PHI) in text. It is shown that a conditional random fields model with access to the generated semantic features, in addition to a set of orthographic and syntactic features, significantly outperforms, in terms of F1-score, a baseline model without access to the semantic features. Moreover, the quality of the features is further improved by employing a number of slightly different models of distributional semantics in an ensemble. Finally, the way in which the features are generated allows one to optimize them for various Fβ-scores, giving some degree of control to trade off precision and recall. Methods that are able to improve performance on named entity recognition tasks by exploiting large amounts of unlabeled data may substantially reduce costs involved in creating annotated resources for every domain and every problem.

OBJECTIVE: The ability of a cue-based system to accurately assert whether a disorder is affirmed, negated, or uncertain is dependent, in part, on its cue lexicon. In this paper, we continue our study of porting an assertion system (pyConTextNLP) from English to Swedish (pyConTextSwe) by creating an optimized assertion lexicon for clinical Swedish. METHODS AND MATERIAL: We integrated cues from four external lexicons, along with generated inflections and combinations. We used subsets of a clinical corpus in Swedish. We applied four assertion classes (definite existence, probable existence, probable negated existence and definite negated existence) and two binary classes (existence yes/no and uncertainty yes/no) to pyConTextSwe. We compared pyConTextSwe's performance with and without the added cues on a development set, and improved the lexicon further after an error analysis. On a separate evaluation set, we calculated the system's final performance. RESULTS: Following integration steps, we added 454 cues to pyConTextSwe. The optimized lexicon developed after an error analysis resulted in statistically significant improvements on the development set (83% F-score, overall). The system's final F-scores on an evaluation set were 81% (overall). For the individual assertion classes, F-score results were 88% (definite existence), 81% (probable existence), 55% (probable negated existence), and 63% (definite negated existence). For the binary classifications existence yes/no and uncertainty yes/no, final system performance was 97%/87% and 78%/86% F-score, respectively. CONCLUSIONS: We have successfully ported pyConTextNLP to Swedish (pyConTextSwe). We have created an extensive and useful assertion lexicon for Swedish clinical text, which could form a valuable resource for similar studies, and which is publicly available.

Automatic recognition of clinical entities in the narrative text of health records is useful for constructing applications for documentation of patient care, as well as for secondary usage in the form of medical knowledge extraction. There are a number of named entity recognition studies on English clinical text, but less work has been carried out on clinical text in other languages. This study was performed on Swedish health records, and focused on four entities that are highly relevant for constructing a patient overview and for medical hypothesis generation, namely the entities: Disorder, Finding, Pharmaceutical Drug and Body Structure. The study had two aims: to explore how well named entity recognition methods previously applied to English clinical text perform on similar texts written in Swedish; and to evaluate whether it is meaningful to divide the more general category Medical Problem, which has been used in a number of previous studies, into the two more granular entities, Disorder and Finding. Clinical notes from a Swedish internal medicine emergency unit were annotated for the four selected entity categories, and the inter-annotator agreement between two pairs of annotators was measured, resulting in an average F-score of 0.79 for Disorder, 0.66 for Finding, 0.90 for Pharmaceutical Drug and 0.80 for Body Structure. A subset of the developed corpus was thereafter used for finding suitable features for training a conditional random fields model. Finally, a new model was trained on this subset, using the best features and settings, and its ability to generalise to held-out data was evaluated. This final model obtained an F-score of 0.81 for Disorder, 0.69 for Finding, 0.88 for Pharmaceutical Drug, 0.85 for Body Structure and 0.78 for the combined category Disorder+Finding. The obtained results, which are in line with or slightly lower than those for similar studies on English clinical text, many of them conducted using a larger training data set, show that the approaches used for English are also suitable for Swedish clinical text. However, a small proportion of the errors made by the model are less likely to occur in English text, showing that results might be improved by further tailoring the system to clinical Swedish. The entity recognition results for the individual entities Disorder and Finding show that it is meaningful to separate the general category Medical Problem into these two more granular entity types, e.g. for knowledge mining of co-morbidity relations and disorder-finding relations.