Predicting decompression surgery by applying multimodal deep learning to patients' structured and unstructured health data.

BACKGROUND: Low back pain (LBP) is a common condition made up of a variety of anatomic and clinical subtypes. Lumbar disc herniation (LDH) and lumbar spinal stenosis (LSS) are two subtypes highly associated with LBP. Patients with LDH/LSS are often started with non-surgical treatments and if those are not effective then go on to have decompression surgery. However, recommendation of surgery is complicated as the outcome may depend on the patient's health characteristics. We developed a deep learning (DL) model to predict decompression surgery for patients with LDH/LSS. MATERIALS AND METHOD: We used datasets of 8387 and 8620 patients from a prospective study that collected data from four healthcare systems to predict early (within 2 months) and late surgery (within 12 months after a 2 month gap), respectively. We developed a DL model to use patients' demographics, diagnosis and procedure codes, drug names, and diagnostic imaging reports to predict surgery. For each prediction task, we evaluated the model's performance using classical and generalizability evaluation. For classical evaluation, we split the data into training (80%) and testing (20%). For generalizability evaluation, we split the data based on the healthcare system. We used the area under the curve (AUC) to assess performance for each evaluation. We compared results to a benchmark model (i.e. LASSO logistic regression). RESULTS: For classical performance, the DL model outperformed the benchmark model for early surgery with an AUC of 0.725 compared to 0.597. For late surgery, the DL model outperformed the benchmark model with an AUC of 0.655 compared to 0.635. For generalizability performance, the DL model outperformed the benchmark model for early surgery. For late surgery, the benchmark model outperformed the DL model. CONCLUSIONS: For early surgery, the DL model was preferred for classical and generalizability evaluation. However, for late surgery, the benchmark and DL model had comparable performance. Depending on the prediction task, the balance of performance may shift between DL and a conventional ML method. As a result, thorough assessment is needed to quantify the value of DL, a relatively computationally expensive, time-consuming and less interpretable method.

RImmPort: an R/Bioconductor package that enables ready-for-analysis immunology research data.

Summary: : Open access to raw clinical and molecular data related to immunological studies has created a tremendous opportunity for data-driven science. We have developed RImmPort that prepares NIAID-funded research study datasets in ImmPort (immport.org) for analysis in R. RImmPort comprises of three main components: (i) a specification of R classes that encapsulate study data, (ii) foundational methods to load data of a specific study and (iii) generic methods to slice and dice data across different dimensions in one or more studies. Furthermore, RImmPort supports open formalisms, such as CDISC standards on the open source bioinformatics platform Bioconductor, to ensure that ImmPort curated study datasets are seamlessly accessible and ready for analysis, thus enabling innovative bioinformatics research in immunology. Availability and Implementation: RImmPort is available as part of Bioconductor (bioconductor.org/packages/RImmPort). Contact: rshankar@stanford.edu. Supplementary information: Supplementary data are available at Bioinformatics online.

A Comparison of Natural Language Processing Methods for the Classification of Lumbar Spine Imaging Findings Related to Lower Back Pain.

RATIONALE AND OBJECTIVES: The use of natural language processing (NLP) in radiology provides an opportunity to assist clinicians with phenotyping patients. However, the performance and generalizability of NLP across healthcare systems is uncertain. We assessed the performance within and generalizability across four healthcare systems of different NLP representational methods, coupled with elastic-net logistic regression to classify lower back pain-related findings from lumbar spine imaging reports. MATERIALS AND METHODS: We used a dataset of 871 X-ray and magnetic resonance imaging reports sampled from a prospective study across four healthcare systems between October 2013 and September 2016. We annotated each report for 26 findings potentially related to lower back pain. Our framework applied four different NLP methods to convert text into feature sets (representations). For each representation, our framework used an elastic-net logistic regression model for each finding (i.e., 26 binary or "one-vs.-rest" classification models). For performance evaluation, we split data into training (80%, 697/871) and testing (20%, 174/871). In the training set, we used cross validation to identify the optimal hyperparameter value and then retrained on the full training set. We then assessed performance based on area under the curve (AUC) for the test set. We repeated this process 25 times with each repeat using a different random train/test split of the data, so that we could estimate 95% confidence intervals, and assess significant difference in performance between representations. For generalizability evaluation, we trained models on data from three healthcare systems with cross validation and then tested on the fourth. We repeated this process for each system, then calculated mean and standard deviation (SD) of AUC across the systems. RESULTS: For individual representations, n-grams had the best average performance across all 26 findings (AUC: 0.960). For generalizability, document embeddings had the most consistent average performance across systems (SD: 0.010). Out of these 26 findings, we considered eight as potentially clinically important (any stenosis, central stenosis, lateral stenosis, foraminal stenosis, disc extrusion, nerve root displacement compression, endplate edema, and listhesis grade 2) since they have a relatively greater association with a history of lower back pain compared to the remaining 18 classes. We found a similar pattern for these eight in which n-grams and document embeddings had the best average performance (AUC: 0.954) and generalizability (SD: 0.007), respectively. CONCLUSION: Based on performance assessment, we found that n-grams is the preferred method if classifier development and deployment occur at the same system. However, for deployment at multiple systems outside of the development system, or potentially if physician behavior changes within a system, one should consider document embeddings since embeddings appear to have the most consistent performance across systems.

Usefulness of mobile apps for communication of genetic test results to at-risk family members in a U.S. integrated health system: a qualitative approach from user-testing.

OBJECTIVE: To assess the usefulness a mobile based application to send genetic test results to at-risk family members in a U.S. integrated health system. METHODS: We conducted semi-structured in-person interviews with members of Kaiser Permanente Washington who had enrolled in a prospective study and received genetic test results. Participants were given the task to use the app and comment on the experience. The moderator asked participants to share perspectives on the usefulness of a mobile based app and their lived experiences of sharing their test results with family members. RESULTS: Fourteen study participants who had undergone genetic testing were interviewed. Four primary themes emerged as relevant to the use of mobile-based apps as a tool for communicating genetic test results to at-risk family members: (i) Participants felt a sense of obligation to share positive test results with relatives; (ii) Participants felt that the advantages of using email were similar to those of the app; (iii) Participants felt that younger individuals would be more comfortable with an app; and, (iv) Participants felt they could use the app independently and in their own time. CONCLUSION: A mobile based app could be used as a tool to improve cascade screening for pathogenic/likely pathogenic test results. The benefits of such a tool are likely greatest among relatives still at the stage of family planning, as well as among family members with strained relationships. There would be minimal burden on the system to offer a mobile based app as a tool.

ShareDNA: a smartphone app to facilitate family communication of genetic results.

BACKGROUND: Genetic testing allows patients and clinicians to understand the risk of hereditary diseases. By testing early, individuals can make informed medical decisions about management which may minimize the risk of developing certain diseases. Importantly, genetic test results may also be applicable to patients' biological relatives; thus, these results could also lead to minimizing their risk of disease. However, sharing genetic test results between patients and their relatives is scarce. The most frequently reported problems are that patients cannot clearly explain this information and relatives misinterpret the results. Smartphone apps in the healthcare field are a possible solution as they allow patients to accurately share sensitive information to others, while providing educational material to support understanding the information. However, these apps may not provide security to protect patients' identifiable information. We developed ShareDNA, a smartphone app that (1) allows patients to securely share their genetic test results with others, (2) provides information on how to interpret these results, and (3) minimizes the amount of patient information needed to use the service. RESULTS: We recruited thirteen participants to test the usability of our app and provide feedback. We found overall that participants were comfortable with using this app and could easily learn each app function when filling out our questionnaire. Additionally, based on vocalized impressions of the usefulness of the app, participants indicated that the user-interface could be more intuitive and that we needed to add more text within the app to explain why ShareDNA is a secure service. CONCLUSIONS: ShareDNA is a free smartphone app that allows patients to share their genetic test results with others, including their biological relatives. Sharing these results along with educational material will enable relatives to share accurate information and discuss their possible risk for disease with their clinical providers. As a result, appropriate testing in relatives could be improved.

MetaCyto: A Tool for Automated Meta-analysis of Mass and Flow Cytometry Data.

While meta-analysis has demonstrated increased statistical power and more robust estimations in studies, the application of this commonly accepted methodology to cytometry data has been challenging. Different cytometry studies often involve diverse sets of markers. Moreover, the detected values of the same marker are inconsistent between studies due to different experimental designs and cytometer configurations. As a result, the cell subsets identified by existing auto-gating methods cannot be directly compared across studies. We developed MetaCyto for automated meta-analysis of both flow and mass cytometry (CyTOF) data. By combining clustering methods with a silhouette scanning method, MetaCyto is able to identify commonly labeled cell subsets across studies, thus enabling meta-analysis. Applying MetaCyto across a set of ten heterogeneous cytometry studies totaling 2,926 samples enabled us to identify multiple cell populations exhibiting differences in abundance between demographic groups. Software is released to the public through Bioconductor (http://bioconductor.org/packages/release/bioc/html/MetaCyto.html).

A Transient Developmental Hematopoietic Stem Cell Gives Rise to Innate-like B and T Cells.

The generation of distinct hematopoietic cell types, including tissue-resident immune cells, distinguishes fetal from adult hematopoiesis. However, the mechanisms underlying differential cell production to generate a layered immune system during hematopoietic development are unclear. Using an irreversible lineage-tracing model, we identify a definitive hematopoietic stem cell (HSC) that supports long-term multilineage reconstitution upon transplantation into adult recipients but does not persist into adulthood in situ. These HSCs are fully multipotent, yet they display both higher lymphoid cell production and greater capacity to generate innate-like B and T lymphocytes as compared to coexisting fetal HSCs and adult HSCs. Thus, these developmentally restricted HSCs (drHSCs) define the origin and generation of early lymphoid cells that play essential roles in establishing self-recognition and tolerance, with important implications for understanding autoimmune disease, allergy, and rejection of transplanted organs.