ACM BCB 2023: Predicting Postoperative Pain and Opioid Use with Machine Learning Applied to Longitudinal Electronic Health Record and Wearable Data.

BACKGROUND: Managing acute postoperative pain and minimizing chronic opioid use is crucial for patient recovery and long-term well-being. OBJECTIVE: This study explored using preoperative electronic health records (EHR) and wearable device data for machine-learning models that predict postoperative acute pain and chronic opioid use. METHODS: The study cohort consisted of ~347 All of Us Research Program participants who underwent one of eight surgical procedures and shared EHR and wearable device data. We developed four machine learning models and used the Shapley additive explanations (SHAP) technique to identify the most relevant predictors of acute pain and chronic opioid use. RESULTS: The stacking ensemble model achieved the highest accuracy in predicting acute pain (0.68) and chronic opioid use (0.89). The area under the curve (AUC) score for severe pain vs. other pain was highest (0.88) when predicting acute post-operative pain. Values of logistic regression, random forest, extreme gradient boosting, and stacking ensemble ranged from 0.74 to 0.90 when predicting postoperative chronic opioid use. Variables from wearable devices played a prominent role in predicting both outcomes. CONCLUSIONS: SHAP detection of individual risk factors for severe pain can help healthcare providers tailor pain management plans. Accurate prediction of postoperative chronic opioid use before surgery can help mitigate the risk for the outcomes we studied. Prediction can also reduce the chances of opioid overuse and dependence. Such mitigation can promote safer and more effective pain control for patients during their recovery.

Incorporation of emergent symptoms and genetic covariates improves prediction of aromatase inhibitor therapy discontinuation.

Objectives: Early discontinuation is common among breast cancer patients taking aromatase inhibitors (AIs). Although several predictors have been identified, it is unclear how to simultaneously consider multiple risk factors for an individual. We sought to develop a tool for prediction of AI discontinuation and to explore how predictive value of risk factors changes with time. Materials and Methods: Survival machine learning was used to predict time-to-discontinuation of AIs in 181 women who enrolled in a prospective cohort. Models were evaluated via time-dependent area under the curve (AUC), c-index, and integrated Brier score. Feature importance was analysis was conducted via Shapley Additive Explanations (SHAP) and time-dependence of their predictive value was analyzed by time-dependent AUC. Personalized survival curves were constructed for risk communication. Results: The best-performing model incorporated genetic risk factors and changes in patient-reported outcomes, achieving mean time-dependent AUC of 0.66, and AUC of 0.72 and 0.67 at 6- and 12-month cutoffs, respectively. The most significant features included variants in ESR1 and emergent symptoms. Predictive value of genetic risk factors was highest in the first year of treatment. Decrease in physical function was the strongest independent predictor at follow-up. Discussion and Conclusion: Incorporation of genomic and 3-month follow-up data improved the ability of the models to identify the individuals at risk of AI discontinuation. Genetic risk factors were particularly important for predicting early discontinuers. This study provides insight into the complex nature of AI discontinuation and highlights the importance of incorporating genetic risk factors and emergent symptoms into prediction models.

Individual- and Neighborhood-Level Disparities in Audio-Only Telemedicine Utilization Across a Large Academic Health System.

Introduction: The objective of this study was to understand whether use of audio-only telemedicine visits differed by individual- and neighborhood-level patient characteristics during the COVID-19 pandemic. Methods: We conducted a retrospective cross-sectional study of telemedicine encounter data from a large academic health system. The primary outcome was rate of audio-only versus video visits. The exposures of interest were individual- (age, race, insurance, preferred language) and neighborhood-level (Social Deprivation Index [SDI]) patient characteristics. Results: Our study included 1,054,465 patient encounters from January 1, 2020 to December 31, 2021, of which 18.33% were completed via audio-only. Encounters among adults 75 years or older, Black patients, Spanish-speakers, and those with public insurance were more frequently conducted by audio-only (p < 0.001). Overall, populations showed decreasing rates of audio-only visits over time. We also observed an increase in the rate of audio-only encounters as SDI scores increased. Discussion: We found that audio-only disparities exist in telemedicine utilization by individual and zip code level characteristics. Though these disparities have improved over time as seen by our temporal analysis, marginalized and minority groups still showed the lowest rates of video utilization. In conclusion, access to audio-only care is a critical component to ensure that telemedicine is accessible to all populations. State and federal policy should support continued reimbursement of audio-only care to ensure equitable access to care while the implications of different care modalities are further studied.

Enabling the clinical application of artificial intelligence in genomics: a perspective of the AMIA Genomics and Translational Bioinformatics Workgroup.

OBJECTIVE: Given the importance AI in genomics and its potential impact on human health, the American Medical Informatics Association-Genomics and Translational Biomedical Informatics (GenTBI) Workgroup developed this assessment of factors that can further enable the clinical application of AI in this space. PROCESS: A list of relevant factors was developed through GenTBI workgroup discussions in multiple in-person and online meetings, along with review of pertinent publications. This list was then summarized and reviewed to achieve consensus among the group members. CONCLUSIONS: Substantial informatics research and development are needed to fully realize the clinical potential of such technologies. The development of larger datasets is crucial to emulating the success AI is achieving in other domains. It is important that AI methods do not exacerbate existing socio-economic, racial, and ethnic disparities. Genomic data standards are critical to effectively scale such technologies across institutions. With so much uncertainty, complexity and novelty in genomics and medicine, and with an evolving regulatory environment, the current focus should be on using these technologies in an interface with clinicians that emphasizes the value each brings to clinical decision-making.

Will the Doctor "See" You Now? The Development and Implementation of a Targeted Telemedicine System for Primary Care.

Objectives: The coronavirus disease 2019 (COVID-19) pandemic led to a rapid adoption of telehealth. For underserved populations lacking internet access, telemedicine was accomplished by phone rather than an audio-video connection. The latter is presumed a more effective form and better approximation of an in-person visit. We sought to provide a telehealth platform to overcome barriers for underserved groups to hold video visits with their health care providers and evaluate differences between the two telehealth modalities as assessed by physicians and patients. Methods: We designed a simplified tablet solution for video visits and piloted its use among patients who otherwise would have been completing audio-only visits. Patients consented to participation and were randomized in a 1:1 fashion to continue with their scheduled phone visit (control) versus being shipped a tablet to facilitate a video visit (intervention). Participants and providers completed communication and satisfaction surveys. Results: Tablet and connectivity design features included removal of all functions but for the telemedicine program, LTE always-on wireless internet connectivity, absence of external equipment (cords chargers and keyboard), and no registration with a digital portal. In total, 18 patients were enrolled. Intervention patients with video-enabled devices compared to control patients agreed more strongly that they were satisfied with their visits (4.75/5 vs. 3.75/5, p = 0.02). Conclusion: The delivered simplified tablet solution for video visits holds promise to improve access to video visits for underserved groups. Strategies to facilitate patient acceptance of devices are needed to expand the scope and potential impact of this effort.

Feasibility of the Genetic Information Assistant Chatbot to Provide Genetic Education and Study Genetic Test Adoption Among Pancreatic Cancer Patients at Johns Hopkins Hospital.

Genetic testing is a valuable tool to guide care of pancreatic cancer patients, yet personal and family uncertainty about the benefits of genetic testing (i.e., decisional conflict) may lead to low adoption. Enabling patients to learn more about genetic testing before their scheduled appointments may help to address this decisional conflict problem. We completed a feasibility assessment of a chatbot to provide genetic education (GEd) with 60 pancreatic cancer patients and using the chatbot to deliver surveys to assess: (a) opinions about the GEd, and (b) decisional conflict about genetic testing. Findings demonstrate intervention and study feasibility with about 80% of participants engaging with the GEd chatbot, 71% of which completed at least one survey. Overall, participants appear to have favorable opinions of the chatbot-delivered education and thought it was helpful to decide about genetic testing. Furthermore, patients who chose to get genetic testing spent more time interacting with the chatbot. Findings will be used to improve chatbot design and to facilitate a well-powered future trial.

Factors associated with resistance to SARS-CoV-2 infection discovered using large-scale medical record data and machine learning.

There have been over 621 million cases of COVID-19 worldwide with over 6.5 million deaths. Despite the high secondary attack rate of COVID-19 in shared households, some exposed individuals do not contract the virus. In addition, little is known about whether the occurrence of COVID-19 resistance differs among people by health characteristics as stored in the electronic health records (EHR). In this retrospective analysis, we develop a statistical model to predict COVID-19 resistance in 8,536 individuals with prior COVID-19 exposure using demographics, diagnostic codes, outpatient medication orders, and count of Elixhauser comorbidities in EHR data from the COVID-19 Precision Medicine Platform Registry. Cluster analyses identified 5 patterns of diagnostic codes that distinguished resistant from non-resistant patients in our study population. In addition, our models showed modest performance in predicting COVID-19 resistance (best performing model AUROC = 0.61). Monte Carlo simulations conducted indicated that the AUROC results are statistically significant (p < 0.001) for the testing set. We hope to validate the features found to be associated with resistance/non-resistance through more advanced association studies.

Deep Learning Model for Static Ocular Torsion Detection Using Synthetically Generated Fundus Images.

Purpose: The objective of the study is to develop deep learning models using synthetic fundus images to assess the direction (intorsion versus extorsion) and amount (physiologic versus pathologic) of static ocular torsion. Static ocular torsion assessment is an important clinical tool for classifying vertical ocular misalignment; however, current methods are time-intensive with steep learning curves for frontline providers. Methods: We used a dataset (n = 276) of right eye fundus images. The disc-foveal angle was calculated using ImageJ to generate synthetic images via image rotation. Using synthetic datasets (n = 12,740 images per model) and transfer learning (the reuse of a pretrained deep learning model on a new task), we developed a binary classifier (intorsion versus extorsion) and a multiclass classifier (physiologic versus pathologic intorsion and extorsion). Model performance was evaluated on unseen synthetic and nonsynthetic data. Results: On the synthetic dataset, the binary classifier had an accuracy and area under the receiver operating characteristic curve (AUROC) of 0.92 and 0.98, respectively, whereas the multiclass classifier had an accuracy and AUROC of 0.77 and 0.94, respectively. The binary classifier generalized well on the nonsynthetic data (accuracy = 0.94; AUROC = 1.00). Conclusions: The direction of static ocular torsion can be detected from synthetic fundus images using deep learning methods, which is key to differentiate between vestibular misalignment (skew deviation) and ocular muscle misalignment (superior oblique palsies). Translational Relevance: Given the robust performance of our models on real fundus images, similar strategies can be adopted for deep learning research in rare neuro-ophthalmologic diseases with limited datasets.

Risk for Poor Post-Operative Quality of Life Among Wearable Use Subgroups in an All of Us Research Cohort.

The objective of this research was to build and assess the performance of a prediction model for post-operative recovery status measured by quality of life among individuals experiencing a variety of surgery types. In addition, we assessed the performance of the model for two subgroups (high and moderately consistent wearable device users). Study variables were derived from the electronic health records, questionnaires, and wearable devices of a cohort of individuals with one of 8 surgery types and that were part of the NIH All of Us research program. Through multivariable analysis, high frailty index (OR 1.69, 95% 1.05-7.22, p<0.006), and older age (OR 1.76, 95% 1.55-4.08, p<0.024) were found to be the driving risk factors of poor recovery post-surgery. Our logistic regression model included 15 variables, 5 of which included wearable device data. In wearable use subgroups, the model had better accuracy for high wearable users (81%). Findings demonstrate the potential for models that use wearable measures to assess frailty to inform clinicians of patients at risk for poor surgical outcomes. Our model performed with high accuracy across multiple surgery types and were robust to variable consistency in wearable use.

An Automated Strategy to Calculate Medication Regimen Complexity.

Understanding medication regimen complexity is important to understand what patients may benefit from pharmacist interventions. Medication Regimen Complexity Index (MRCI), a 65-item tool to quantify the complexity by incorporating the count, dosage form, frequency, and additional administration instructions of prescription medicines, provides a more nuanced way of assessing complexity. The goal of this study was to construct and validate a computational strategy to automate the calculation of MRCI. The performance of our strategy was evaluated by comparing our calculated MRCI values with gold-standard values, using correlation coefficients and population distributions. The results revealed satisfactory performance to calculate the sub-score of MRCI that includes dosage form and frequency (76 to 80% match with gold standard), and fair performance for sub-score related to additional direction (52% match with gold standard). Our automated strategy shows potential to help reduce the effort for manually calculating MRCI and highlights areas for future development efforts.

Assessing Associations Between COVID-19 Symptomology and Adverse Outcomes After Piloting Crowdsourced Data Collection: Cross-sectional Survey Study.

BACKGROUND: Crowdsourcing is a useful way to rapidly collect information on COVID-19 symptoms. However, there are potential biases and data quality issues given the population that chooses to participate in crowdsourcing activities and the common strategies used to screen participants based on their previous experience. OBJECTIVE: The study aimed to (1) build a pipeline to enable data quality and population representation checks in a pilot setting prior to deploying a final survey to a crowdsourcing platform, (2) assess COVID-19 symptomology among survey respondents who report a previous positive COVID-19 result, and (3) assess associations of symptomology groups and underlying chronic conditions with adverse outcomes due to COVID-19. METHODS: We developed a web-based survey and hosted it on the Amazon Mechanical Turk (MTurk) crowdsourcing platform. We conducted a pilot study from August 5, 2020, to August 14, 2020, to refine the filtering criteria according to our needs before finalizing the pipeline. The final survey was posted from late August to December 31, 2020. Hierarchical cluster analyses were performed to identify COVID-19 symptomology groups, and logistic regression analyses were performed for hospitalization and mechanical ventilation outcomes. Finally, we performed a validation of study outcomes by comparing our findings to those reported in previous systematic reviews. RESULTS: The crowdsourcing pipeline facilitated piloting our survey study and revising the filtering criteria to target specific MTurk experience levels and to include a second attention check. We collected data from 1254 COVID-19-positive survey participants and identified the following 6 symptomology groups: abdominal and bladder pain (Group 1); flu-like symptoms (loss of smell/taste/appetite; Group 2); hoarseness and sputum production (Group 3); joint aches and stomach cramps (Group 4); eye or skin dryness and vomiting (Group 5); and no symptoms (Group 6). The risk factors for adverse COVID-19 outcomes differed for different symptomology groups. The only risk factor that remained significant across 4 symptomology groups was influenza vaccine in the previous year (Group 1: odds ratio [OR] 6.22, 95% CI 2.32-17.92; Group 2: OR 2.35, 95% CI 1.74-3.18; Group 3: OR 3.7, 95% CI 1.32-10.98; Group 4: OR 4.44, 95% CI 1.53-14.49). Our findings regarding the symptoms of abdominal pain, cough, fever, fatigue, shortness of breath, and vomiting as risk factors for COVID-19 adverse outcomes were concordant with the findings of other researchers. Some high-risk symptoms found in our study, including bladder pain, dry eyes or skin, and loss of appetite, were reported less frequently by other researchers and were not considered previously in relation to COVID-19 adverse outcomes. CONCLUSIONS: We demonstrated that a crowdsourced approach was effective for collecting data to assess symptomology associated with COVID-19. Such a strategy may facilitate efficient assessments in a dynamic intersection between emerging infectious diseases, and societal and environmental changes.

A research agenda to support the development and implementation of genomics-based clinical informatics tools and resources.

OBJECTIVE: The Genomic Medicine Working Group of the National Advisory Council for Human Genome Research virtually hosted its 13th genomic medicine meeting titled "Developing a Clinical Genomic Informatics Research Agenda". The meeting's goal was to articulate a research strategy to develop Genomics-based Clinical Informatics Tools and Resources (GCIT) to improve the detection, treatment, and reporting of genetic disorders in clinical settings. MATERIALS AND METHODS: Experts from government agencies, the private sector, and academia in genomic medicine and clinical informatics were invited to address the meeting's goals. Invitees were also asked to complete a survey to assess important considerations needed to develop a genomic-based clinical informatics research strategy. RESULTS: Outcomes from the meeting included identifying short-term research needs, such as designing and implementing standards-based interfaces between laboratory information systems and electronic health records, as well as long-term projects, such as identifying and addressing barriers related to the establishment and implementation of genomic data exchange systems that, in turn, the research community could help address. DISCUSSION: Discussions centered on identifying gaps and barriers that impede the use of GCIT in genomic medicine. Emergent themes from the meeting included developing an implementation science framework, defining a value proposition for all stakeholders, fostering engagement with patients and partners to develop applications under patient control, promoting the use of relevant clinical workflows in research, and lowering related barriers to regulatory processes. Another key theme was recognizing pervasive biases in data and information systems, algorithms, access, value, and knowledge repositories and identifying ways to resolve them.

Inverting the model of genomics data sharing with the NHGRI Genomic Data Science Analysis, Visualization, and Informatics Lab-space.

The NHGRI Genomic Data Science Analysis, Visualization, and Informatics Lab-space (AnVIL; https://anvilproject.org) was developed to address a widespread community need for a unified computing environment for genomics data storage, management, and analysis. In this perspective, we present AnVIL, describe its ecosystem and interoperability with other platforms, and highlight how this platform and associated initiatives contribute to improved genomic data sharing efforts. The AnVIL is a federated cloud platform designed to manage and store genomics and related data, enable population-scale analysis, and facilitate collaboration through the sharing of data, code, and analysis results. By inverting the traditional model of data sharing, the AnVIL eliminates the need for data movement while also adding security measures for active threat detection and monitoring and provides scalable, shared computing resources for any researcher. We describe the core data management and analysis components of the AnVIL, which currently consists of Terra, Gen3, Galaxy, RStudio/Bioconductor, Dockstore, and Jupyter, and describe several flagship genomics datasets available within the AnVIL. We continue to extend and innovate the AnVIL ecosystem by implementing new capabilities, including mechanisms for interoperability and responsible data sharing, while streamlining access management. The AnVIL opens many new opportunities for analysis, collaboration, and data sharing that are needed to drive research and to make discoveries through the joint analysis of hundreds of thousands to millions of genomes along with associated clinical and molecular data types.

Recurrent preterm birth risk assessment for two delivery subtypes: A multivariable analysis.

OBJECTIVE: The study sought to develop and apply a framework that uses a clinical phenotyping tool to assess risk for recurrent preterm birth. MATERIALS AND METHODS: We extended an existing clinical phenotyping tool and applied a 4-step framework for our retrospective cohort study. The study was based on data collected in the Genomic and Proteomic Network for Preterm Birth Research Longitudinal Cohort Study (GPN-PBR LS). A total of 52 sociodemographic, clinical and obstetric history-related risk factors were selected for the analysis. Spontaneous and indicated delivery subtypes were analyzed both individually and in combination. Chi-square analysis and Kaplan-Meier estimate were used for univariate analysis. A Cox proportional hazards model was used for multivariable analysis. RESULTS: : A total of 428 women with a history of spontaneous preterm birth qualified for our analysis. The predictors of preterm delivery used in multivariable model were maternal age, maternal race, household income, marital status, previous caesarean section, number of previous deliveries, number of previous abortions, previous birth weight, cervical insufficiency, decidual hemorrhage, and placental dysfunction. The models stratified by delivery subtype performed better than the naïve model (concordance 0.76 for the spontaneous model, 0.87 for the indicated model, and 0.72 for the naïve model). DISCUSSION: The proposed 4-step framework is effective to analyze risk factors for recurrent preterm birth in a retrospective cohort and possesses practical features for future analyses with other data sources (eg, electronic health record data). CONCLUSIONS: We developed an analytical framework that utilizes a clinical phenotyping tool and performed a survival analysis to analyze risk for recurrent preterm birth.

Willingness to Share Wearable Device Data for Research Among Mechanical Turk Workers: Web-Based Survey Study.

BACKGROUND: Wearable devices that are used for observational research and clinical trials hold promise for collecting data from study participants in a convenient, scalable way that is more likely to reach a broad and diverse population than traditional research approaches. Amazon Mechanical Turk (MTurk) is a potential resource that researchers can use to recruit individuals into studies that use data from wearable devices. OBJECTIVE: This study aimed to explore the characteristics of wearable device users on MTurk that are associated with a willingness to share wearable device data for research. We also aimed to determine whether compensation was a factor that influenced the willingness to share such data. METHODS: This was a secondary analysis of a cross-sectional survey study of MTurk workers who use wearable devices for health monitoring. A 19-question web-based survey was administered from March 1 to April 5, 2018, to participants aged ≥18 years by using the MTurk platform. In order to identify characteristics that were associated with a willingness to share wearable device data, we performed logistic regression and decision tree analyses. RESULTS: A total of 935 MTurk workers who use wearable devices completed the survey. The majority of respondents indicated a willingness to share their wearable device data (615/935, 65.8%), and the majority of these respondents were willing to share their data if they received compensation (518/615, 84.2%). The findings from our logistic regression analyses indicated that Indian nationality (odds ratio [OR] 2.74, 95% CI 1.48-4.01, P=.007), higher annual income (OR 2.46, 95% CI 1.26-3.67, P=.02), over 6 months of using a wearable device (OR 1.75, 95% CI 1.21-2.29, P=.006), and the use of heartbeat and pulse tracking monitoring devices (OR 1.60, 95% CI 0.14-2.07, P=.01) are significant parameters that influence the willingness to share data. The only factor associated with a willingness to share data if compensation is provided was Indian nationality (OR 0.47, 95% CI 0.24-0.9, P=.02). The findings from our decision tree analyses indicated that the three leading parameters associated with a willingness to share data were the duration of wearable device use, nationality, and income. CONCLUSIONS: Most wearable device users indicated a willingness to share their data for research use (with or without compensation; 615/935, 65.8%). The probability of having a willingness to share these data was higher among individuals who had used a wearable for more than 6 months, were of Indian nationality, or were of American (United States of America) nationality and had an annual income of more than US $20,000. Individuals of Indian nationality who were willing to share their data expected compensation significantly less often than individuals of American nationality (P=.02).

Preferences for Updates on General Research Results: A Survey of Participants in Genomic Research from Two Institutions.

There is a need for multimodal strategies to keep research participants informed about study results. Our aim was to characterize preferences of genomic research participants from two institutions along four dimensions of general research result updates: content, timing, mechanism, and frequency. METHODS: We conducted a web-based cross-sectional survey that was administered from 25 June 2018 to 5 December 2018. RESULTS: 397 participants completed the survey, most of whom (96%) expressed a desire to receive research updates. Preferences with high endorsement included: update content (brief descriptions of major findings, descriptions of purpose and goals, and educational material); update timing (when the research is completed, when findings are reviewed, when findings are published, and when the study status changes); update mechanism (email with updates, and email newsletter); and update frequency (every three months). Hierarchical cluster analyses based on the four update preferences identified four profiles of participants with similar preference patterns. Very few participants in the largest profile were comfortable with budgeting less money for research activities so that researchers have money to set up services to send research result updates to study participants. CONCLUSION: Future studies may benefit from exploring preferences for research result updates, as we have in our study. In addition, this work provides evidence of a need for funders to incentivize researchers to communicate results to participants.

Infobuttons for Genomic Medicine: Requirements and Barriers.

OBJECTIVES: The study aimed to understand potential barriers to the adoption of health information technology projects that are released as free and open source software (FOSS). METHODS: We conducted a survey of research consortia participants engaged in genomic medicine implementation to assess perceived institutional barriers to the adoption of three systems: ClinGen electronic health record (EHR) Toolkit, DocUBuild, and MyResults.org. The survey included eight barriers from the Consolidated Framework for Implementation Research (CFIR), with additional barriers identified from a qualitative analysis of open-ended responses. RESULTS: We analyzed responses from 24 research consortia participants from 18 institutions. In total, 14 categories of perceived barriers were evaluated, which were consistent with other observed barriers to FOSS adoption. The most frequent perceived barriers included lack of adaptability of the system, lack of institutional priority to implement, lack of trialability, lack of advantage of alternative systems, and complexity. CONCLUSION: In addition to understanding potential barriers, we recommend some strategies to address them (where possible), including considerations for genomic medicine. Overall, FOSS developers need to ensure systems are easy to trial and implement and need to clearly articulate benefits of their systems, especially when alternatives exist. Institutional champions will remain a critical component to prioritizing genomic medicine projects.

Facilitating Genetics Aware Clinical Decision Support: Putting the eMERGE Infrastructure into Practice.

This editorial provides context for a series of published case reports in ACI Open by summarizing activities and outputs of joint electronic health record integration and pharmacogenomics workgroups in the NIH-funded electronic Medical Records and Genomics (eMERGE) Network. A case report is a useful tool to describe the range of capabilities that an IT infrastructure or a particular technology must support. The activities we describe have informed infrastructure requirements used during eMERGE phase III, provided a venue to share experiences and ask questions among other eMERGE sites, summarized potential hazards that might be encountered for specific clinical decision support (CDS) implementation scenarios, and provided a simple framework that captured progress toward implementing CDS at eMERGE sites in a consistent format.

Feature engineering with clinical expert knowledge: A case study assessment of machine learning model complexity and performance.

Incorporating expert knowledge at the time machine learning models are trained holds promise for producing models that are easier to interpret. The main objectives of this study were to use a feature engineering approach to incorporate clinical expert knowledge prior to applying machine learning techniques, and to assess the impact of the approach on model complexity and performance. Four machine learning models were trained to predict mortality with a severe asthma case study. Experiments to select fewer input features based on a discriminative score showed low to moderate precision for discovering clinically meaningful triplets, indicating that discriminative score alone cannot replace clinical input. When compared to baseline machine learning models, we found a decrease in model complexity with use of fewer features informed by discriminative score and filtering of laboratory features with clinical input. We also found a small difference in performance for the mortality prediction task when comparing baseline ML models to models that used filtered features. Encoding demographic and triplet information in ML models with filtered features appeared to show performance improvements from the baseline. These findings indicated that the use of filtered features may reduce model complexity, and with little impact on performance.

Genomic Information for Clinicians in the Electronic Health Record: Lessons Learned From the Clinical Genome Resource Project and the Electronic Medical Records and Genomics Network.

Genomic knowledge is being translated into clinical care. To fully realize the value, it is critical to place credible information in the hands of clinicians in time to support clinical decision making. The electronic health record is an essential component of clinician workflow. Utilizing the electronic health record to present information to support the use of genomic medicine in clinical care to improve outcomes represents a tremendous opportunity. However, there are numerous barriers that prevent the effective use of the electronic health record for this purpose. The electronic health record working groups of the Electronic Medical Records and Genomics (eMERGE) Network and the Clinical Genome Resource (ClinGen) project, along with other groups, have been defining these barriers, to allow the development of solutions that can be tested using implementation pilots. In this paper, we present "lessons learned" from these efforts to inform future efforts leading to the development of effective and sustainable solutions that will support the realization of genomic medicine.