Dementia risk prediction using decision-focused content selection from medical notes.

Several general-purpose language model (LM) architectures have been proposed with demonstrated improvement in text summarization and classification. Adapting these architectures to the medical domain requires additional considerations. For instance, the medical history of the patient is documented in the Electronic Health Record (EHR) which includes many medical notes drafted by healthcare providers. Direct processing of these notes may not be possible because the computational complexity of LMs imposes a limit on the length of input text. Therefore, previous applications resorted to content selection using truncation or summarization of the text. Unfortunately, these text processing techniques may lead to information loss, redundancy or irrelevance. In the present paper, a decision-focused content selection technique is proposed. The objective of this technique is to select a subset of sentences from the medical notes of a patient that are relevant to the target outcome over a predefined observation period. This decision-focused content selection methodology is then used to develop a dementia risk prediction model based on the Longformer LM architecture. The results show that the proposed framework delivers an AUC of 78.43 when the summary is restricted to 1024 tokens, outperforming previously proposed content selection techniques. This performance is notable given that the model estimates dementia risk with a one year prediction horizon, relies on an observation period of only one year and solely uses medical notes without other EHR data modalities. Moreover, the proposed techniques overcome the limitation of machine learning models that use a tabular representation of the text by preserving contextual content, enable feature engineering from raw text and circumvent the computational complexity of language models.

Implementing a pragmatic clinical trial to tailor opioids for chronic pain on behalf of the IGNITE ADOPT PGx investigators.

Chronic pain is a prevalent condition with enormous economic burden. Opioids such as tramadol, codeine, and hydrocodone are commonly used to treat chronic pain; these drugs are activated to more potent opioid receptor agonists by the hepatic CYP2D6 enzyme. Results from clinical studies and mechanistic understandings suggest that CYP2D6-guided therapy will improve pain control and reduce adverse drug events. However, CYP2D6 is rarely used in clinical practice due in part to the demand for additional clinical trial evidence. Thus, we designed the ADOPT-PGx (A Depression and Opioid Pragmatic Trial in Pharmacogenetics) chronic pain study, a multicenter, pragmatic, randomized controlled clinical trial, to assess the effect of CYP2D6 testing on pain management. The study enrolled 1048 participants who are taking or being considered for treatment with CYP2D6-impacted opioids for their chronic pain. Participants were randomized to receive immediate or delayed (by 6 months) genotyping of CYP2D6 with clinical decision support (CDS). CDS encouraged the providers to follow the CYP2D6-guided trial recommendations. The primary study outcome is the 3-month absolute change in the composite pain intensity score assessed using Patient-Reported Outcomes Measurement Information System (PROMIS) measures. Follow-up will be completed in July 2024. Herein, we describe the design of this trial along with challenges encountered during enrollment.

Prescriptions of opioid-containing drugs in patients with chronic cough.

BACKGROUND: Chronic cough (CC) affects about 10% of adults, but opioid use in CC is not well understood. OBJECTIVES: To determine the use of opioid-containing cough suppressant (OCCS) prescriptions in patients with CC using electronic health records. DESIGN: Retrospective cohort study. METHODS: Through retrospective analysis of Midwestern U.S. electronic health records, diagnoses, prescriptions, and natural language processing identified CC - at least three medical encounters with cough, with 56-120 days between first and last encounter - and a 'non-chronic cohort'. Student's t-test, Pearson's chi-square, and zero-inflated Poisson models were used. RESULTS: About 20% of 23,210 patients with CC were prescribed OCCS; odds of an OCCS prescription were twice as great in CC. In CC, OCCS drugs were ordered in 38% with Medicaid insurance and 15% with commercial insurance. CONCLUSION: Findings identify an important role for opioids in CC, and opportunity to learn more about the drugs' effectiveness.

Rationale and design for a pragmatic randomized trial to assess gene-based prescribing for SSRIs in the treatment of depression.

Specific selective serotonin reuptake inhibitors (SSRIs) metabolism is strongly influenced by two pharmacogenes, CYP2D6 and CYP2C19. However, the effectiveness of prospectively using pharmacogenetic variants to select or dose SSRIs for depression is uncertain in routine clinical practice. The objective of this prospective, multicenter, pragmatic randomized controlled trial is to determine the effectiveness of genotype-guided selection and dosing of antidepressants on control of depression in participants who are 8 years or older with ≥3 months of depressive symptoms who require new or revised therapy. Those randomized to the intervention arm undergo pharmacogenetic testing at baseline and receive a pharmacy consult and/or automated clinical decision support intervention based on an actionable phenotype, while those randomized to the control arm have pharmacogenetic testing at the end of 6-months. In both groups, depression and drug tolerability outcomes are assessed at baseline, 1 month, 3 months (primary), and 6 months. The primary end point is defined by change in Patient-Reported Outcomes Measurement Information System (PROMIS) Depression score assessed at 3 months versus baseline. Secondary end points include change inpatient health questionnaire (PHQ-8) measure of depression severity, remission rates defined by PROMIS score < 16, medication adherence, and medication side effects. The primary analysis will compare the PROMIS score difference between trial arms among those with an actionable CYP2D6 or CYP2C19 genetic result or a CYP2D6 drug-drug interaction. The trial has completed accrual of 1461 participants, of which 562 were found to have an actionable phenotype to date, and follow-up will be complete in April of 2024.

The INGENIOUS trial: Impact of pharmacogenetic testing on adverse events in a pragmatic clinical trial.

Adverse drug events (ADEs) account for a significant mortality, morbidity, and cost burden. Pharmacogenetic testing has the potential to reduce ADEs and inefficacy. The objective of this INGENIOUS trial (NCT02297126) analysis was to determine whether conducting and reporting pharmacogenetic panel testing impacts ADE frequency. The trial was a pragmatic, randomized controlled clinical trial, adapted as a propensity matched analysis in individuals (N = 2612) receiving a new prescription for one or more of 26 pharmacogenetic-actionable drugs across a community safety-net and academic health system. The intervention was a pharmacogenetic testing panel for 26 drugs with dosage and selection recommendations returned to the health record. The primary outcome was occurrence of ADEs within 1 year, according to modified Common Terminology Criteria for Adverse Events (CTCAE). In the propensity-matched analysis, 16.1% of individuals experienced any ADE within 1-year. Serious ADEs (CTCAE level ≥ 3) occurred in 3.2% of individuals. When combining all 26 drugs, no significant difference was observed between the pharmacogenetic testing and control arms for any ADE (Odds ratio 0.96, 95% CI: 0.78-1.18), serious ADEs (OR: 0.91, 95% CI: 0.58-1.40), or mortality (OR: 0.60, 95% CI: 0.28-1.21). However, sub-group analyses revealed a reduction in serious ADEs and death in individuals who underwent pharmacogenotyping for aripiprazole and serotonin or serotonin-norepinephrine reuptake inhibitors (OR 0.34, 95% CI: 0.12-0.85). In conclusion, no change in overall ADEs was observed after pharmacogenetic testing. However, limitations incurred during INGENIOUS likely affected the results. Future studies may consider preemptive, rather than reactive, pharmacogenetic panel testing.

Development and Temporal Validation of an Electronic Medical Record-Based Insomnia Prediction Model Using Data from a Statewide Health Information Exchange.

This study aimed to develop and temporally validate an electronic medical record (EMR)-based insomnia prediction model. In this nested case-control study, we analyzed EMR data from 2011-2018 obtained from a statewide health information exchange. The study sample included 19,843 insomnia cases and 19,843 controls matched by age, sex, and race. Models using different ML techniques were trained to predict insomnia using demographics, diagnosis, and medication order data from two surveillance periods: -1 to -365 days and -180 to -365 days before the first documentation of insomnia. Separate models were also trained with patient data from three time periods (2011-2013, 2011-2015, and 2011-2017). After selecting the best model, predictive performance was evaluated on holdout patients as well as patients from subsequent years to assess the temporal validity of the models. An extreme gradient boosting (XGBoost) model outperformed all other classifiers. XGboost models trained on 2011-2017 data from -1 to -365 and -180 to -365 days before index had AUCs of 0.80 (SD 0.005) and 0.70 (SD 0.006), respectively, on the holdout set. On patients with data from subsequent years, a drop of at most 4% in AUC is observed for all models, even when there is a five-year difference between the collection period of the training and the temporal validation data. The proposed EMR-based prediction models can be used to identify insomnia up to six months before clinical detection. These models may provide an inexpensive, scalable, and longitudinally viable method to screen for individuals at high risk of insomnia.

Feature engineering from medical notes: A case study of dementia detection.

Background and objectives: Medical notes are narratives that describe the health of the patient in free text format. These notes can be more informative than structured data such as the history of medications or disease conditions. They are routinely collected and can be used to evaluate the patient's risk for developing chronic diseases such as dementia. This study investigates different methodologies for transforming routine care notes into dementia risk classifiers and evaluates the generalizability of these classifiers to new patients and new health care institutions. Methods: The notes collected over the relevant history of the patient are lengthy. In this study, TF-ICF is used to select keywords with the highest discriminative ability between at risk dementia patients and healthy controls. The medical notes are then summarized in the form of occurrences of the selected keywords. Two different encodings of the summary are compared. The first encoding consists of the average of the vector embedding of each keyword occurrence as produced by the BERT or Clinical BERT pre-trained language models. The second encoding aggregates the keywords according to UMLS concepts and uses each concept as an exposure variable. For both encodings, misspellings of the selected keywords are also considered in an effort to improve the predictive performance of the classifiers. A neural network is developed over the first encoding and a gradient boosted trees model is applied to the second encoding. Patients from a single health care institution are used to develop all the classifiers which are then evaluated on held-out patients from the same health care institution as well as test patients from two other health care institutions. Results: The results indicate that it is possible to identify patients at risk for dementia one year ahead of the onset of the disease using medical notes with an AUC of 75% when a gradient boosted trees model is used in conjunction with exposure variables derived from UMLS concepts. However, this performance is not maintained with an embedded feature space and when the classifier is applied to patients from other health care institutions. Moreover, an analysis of the top predictors of the gradient boosted trees model indicates that different features inform the classification depending on whether or not spelling variants of the keywords are included. Conclusion: The present study demonstrates that medical notes can enable risk prediction models for complex chronic diseases such as dementia. However, additional research efforts are needed to improve the generalizability of these models. These efforts should take into consideration the length and localization of the medical notes; the availability of sufficient training data for each disease condition; and the variabilities resulting from different feature engineering techniques.

Racial disparities in the pharmacological treatment of insomnia: A time-to-event analysis using real-world data.

OBJECTIVE: Examine the association between race and time to pharmacologic treatment of insomnia in a large multi-institutional cohort. METHODS: Retrospective analysis of electronic medical records from a regional health information exchange. Eligible patients included adults with at least one healthcare visit per year from 2010 to 2019, a new insomnia diagnosis code during the study period, and no prior insomnia diagnosis codes or medications. A Cox frailty model was used to examine the association between race and time to an insomnia medication after diagnosis. RESULTS: In total, 9557 patients were analyzed, 7773 (81.3%) of whom where White, 1294 (13.5%) Black, 238 (2.5%) Other, and 252 (2.6%) unknown race. About 6.2% of Black and 8% of Other race patients received an order for a Food and Drug Administration-approved insomnia medication after diagnosis compared with 13.5% of White patients. Black patients were significantly less likely to have an order for a Food and Drug Administration-approved insomnia medication at all time points (adjusted hazard ratio [aHR] range: 0.37-0.73), and patients reporting Other race were less likely to have received an order at 2 (aHR 0.51, 95% confidence interval [CI] 0.28-0.94), 3 (aHR 0.33, 95% CI 0.13-0.79), and 4 years (aHR 0.21, 95% CI 0.06-0.71) of follow-up. Similar results were observed in a sensitivity analysis including off-label medications. CONCLUSIONS: Patients belonging to racial minority groups are less likely to be prescribed an insomnia medication than White patients after accounting for sociodemographic and clinical factors. Further research is needed to determine the extent to which patient preferences and physician perceptions affect these prescribing patterns and investigate potential disparities in nonpharmacologic treatment.

Association of Antipsychotic-Related Weight Gain With Treatment Adherence and Switching Using Electronic Medical Records Data.

Objective: To leverage electronic health record (EHR) data to explore the relationship between weight gain and antipsychotic adherence among patients with schizophrenia and bipolar disorder (BD).Methods: EHR data were used to identify individuals with at least 60 days of continuous antipsychotic use between 2005 and 2019. Patients were diagnosed with schizophrenia, schizoaffective disorder, BD, or neither diagnosis (psychiatric controls). We examined the association of weight gain in the first 90 days with the proportion of days covered (PDC) with an antipsychotic and with the frequency of medication switching or stopping.Results: We identified 590 adults with schizophrenia or schizoaffective disorder, 819 adults with BD, and 642 psychiatric controls. In the first 90 days, the percentages of patients with a PDC ≥ 0.80 were 76.8% (schizophrenia), 77.1% (BD), and 70.7% (controls). Logistic regression models revealed that weight gain of ≥ 7% trended toward being significantly associated with greater adherence in the first 90 days (odds ratio = 1.29, P = .077) and was significantly associated with an increased likelihood of a medication switch in the first 180 days (odds ratio = 1.60, P = .003).Discussion: Patients whose weight increased by 7% or more in the first 90 days were more adherent but were also more likely to switch medications during the first 180 days.

Modeling acute care utilization: practical implications for insomnia patients.

Machine learning models can help improve health care services. However, they need to be practical to gain wide-adoption. In this study, we investigate the practical utility of different data modalities and cohort segmentation strategies when designing models for emergency department (ED) and inpatient hospital (IH) visits. The data modalities include socio-demographics, diagnosis and medications. Segmentation compares a cohort of insomnia patients to a cohort of general non-insomnia patients under varying age and disease severity criteria. Transfer testing between the two cohorts is introduced to demonstrate that an insomnia-specific model is not necessary when predicting future ED visits, but may have merit when predicting IH visits especially for patients with an insomnia diagnosis. The results also indicate that using both diagnosis and medications as a source of data does not generally improve model performance and may increase its overhead. Based on these findings, the proposed evaluation methodologies are recommended to ascertain the utility of disease-specific models in addition to the traditional intra-cohort testing.

Health-related experiences of adults with chronic cough: Empirical research mixed methods.

AIM: To describe adults' health-related experiences with chronic cough. DESIGN: Survey and interviews. METHODS: Participants completed questionnaires and interviews, to explore chronic cough's impact and management. DATA SOURCES: Patients aged 18-85 years with at least three cough-related encounters within 56-120 days. RESULTS: Forty-one patients were surveyed. Mean cough severity was 4.5 (scale 0-9). Chronic cough-related problems included embarrassment (66%), fatigue (56%), and anxiety or depression (49%). Testing was judged insufficient by 44%. Only 28% were satisfied with treatment; 20% reported abandoning treatment due to ineffectiveness. Interview themes (N = 30) included frustration with diagnostic uncertainty, and feelings of therapeutic futility. Some reported psychological distress. Work and socializing were commonly disrupted. CONCLUSION: Diagnostic uncertainty, perceived limitations of testing, and treatment failures suggest needs for better approaches to evaluating and treating chronic cough. Special attention to identifying and addressing mental health issues appears warranted.

The impact of antipsychotic adherence on acute care utilization.

BACKGROUND: Non-adherence to psychotropic medications is common in schizophrenia and bipolar disorders (BDs) leading to adverse outcomes. We examined patterns of antipsychotic use in schizophrenia and BD and their impact on subsequent acute care utilization. METHODS: We used electronic health record (EHR) data of 577 individuals with schizophrenia, 795 with BD, and 618 using antipsychotics without a diagnosis of either illness at two large health systems. We structured three antipsychotics exposure variables: the proportion of days covered (PDC) to measure adherence; medication switch as a new antipsychotic prescription that was different than the initial antipsychotic; and medication stoppage as the lack of an antipsychotic order or fill data in the EHR after the date when the previous supply would have been depleted. Outcome measures included the frequency of inpatient and emergency department (ED) visits up to 12 months after treatment initiation. RESULTS: Approximately half of the study population were adherent to their antipsychotic medication (a PDC ≥ 0.80): 53.6% of those with schizophrenia, 52.4% of those with BD, and 50.3% of those without either diagnosis. Among schizophrenia patients, 22.5% switched medications and 15.1% stopped therapy. Switching and stopping occurred in 15.8% and 15.1% of BD patients and 7.4% and 20.1% of those without either diagnosis, respectively. Across the three cohorts, non-adherence, switching, and stopping therapy were all associated with increased acute care utilization, even after adjusting for baseline demographics, health insurance, past acute care utilization, and comorbidity. CONCLUSION: Non-continuous antipsychotic use is common and associated with high acute care utilization.

Digital detection of dementia (D3): a study protocol for a pragmatic cluster-randomized trial examining the application of patient-reported outcomes and passive clinical decision support systems.

BACKGROUND: Early detection of Alzheimer's disease and related dementias (ADRD) in a primary care setting is challenging due to time constraints and stigma. The implementation of scalable, sustainable, and patient-driven processes may improve early detection of ADRD; however, there are competing approaches; information may be obtained either directly from a patient (e.g., through a questionnaire) or passively using electronic health record (EHR) data. In this study, we aim to identify the benefit of a combined approach using a pragmatic cluster-randomized clinical trial. METHODS: We have developed a Passive Digital Marker (PDM), based on machine learning algorithms applied to EHR data, and paired it with a patient-reported outcome (the Quick Dementia Rating Scale or QDRS) to rapidly share an identified risk of impairment to a patient's physician. Clinics in both south Florida and Indiana will be randomly assigned to one of three study arms: 1200 patients in each of the two populations will be administered either the PDM, the PDM with the QDRS, or neither, for a total of 7200 patients across all clinics and populations. Both incidence of ADRD diagnosis and acceptance into ADRD diagnostic work-up regimens is hypothesized to increase when patients are administered both the PDM and QDRS. Physicians performing the work-up regimens will be blind to the study arm of the patient. DISCUSSION: This study aims to test the accuracy and effectiveness of the two scalable approaches (PDM and QDRS) for the early detection of ADRD among older adults attending primary care practices. The data obtained in this study may lead to national early detection and management program for ADRD as an efficient and beneficial method of reducing the current and future burden of ADRD, as well as improving the annual rate of newly documented ADRD in primary care practices. TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT05231954 . Registered February 9, 2022.

Implementing a pragmatic clinical trial to tailor opioids for acute pain on behalf of the IGNITE ADOPT PGx investigators.

Opioid prescribing for postoperative pain management is challenging because of inter-patient variability in opioid response and concern about opioid addiction. Tramadol, hydrocodone, and codeine depend on the cytochrome P450 2D6 (CYP2D6) enzyme for formation of highly potent metabolites. Individuals with reduced or absent CYP2D6 activity (i.e., intermediate metabolizers [IMs] or poor metabolizers [PMs], respectively) have lower concentrations of potent opioid metabolites and potentially inadequate pain control. The primary objective of this prospective, multicenter, randomized pragmatic trial is to determine the effect of postoperative CYP2D6-guided opioid prescribing on pain control and opioid usage. Up to 2020 participants, age ≥8 years, scheduled to undergo a surgical procedure will be enrolled and randomized to immediate pharmacogenetic testing with clinical decision support (CDS) for CYP2D6 phenotype-guided postoperative pain management (intervention arm) or delayed testing without CDS (control arm). CDS is provided through medical record alerts and/or a pharmacist consult note. For IMs and PM in the intervention arm, CDS includes recommendations to avoid hydrocodone, tramadol, and codeine. Patient-reported pain-related outcomes are collected 10 days and 1, 3, and 6 months after surgery. The primary outcome, a composite of pain intensity and opioid usage at 10 days postsurgery, will be compared in the subgroup of IMs and PMs in the intervention (n = 152) versus the control (n = 152) arm. Secondary end points include prescription pain medication misuse scores and opioid persistence at 6 months. This trial will provide data on the clinical utility of CYP2D6 phenotype-guided opioid selection for improving postoperative pain control and reducing opioid-related risks.

Design and rationale of GUARDD-US: A pragmatic, randomized trial of genetic testing for APOL1 and pharmacogenomic predictors of antihypertensive efficacy in patients with hypertension.

RATIONALE AND OBJECTIVE: APOL1 risk alleles are associated with increased cardiovascular and chronic kidney disease (CKD) risk. It is unknown whether knowledge of APOL1 risk status motivates patients and providers to attain recommended blood pressure (BP) targets to reduce cardiovascular disease. STUDY DESIGN: Multicenter, pragmatic, randomized controlled clinical trial. SETTING AND PARTICIPANTS: 6650 individuals with African ancestry and hypertension from 13 health systems. INTERVENTION: APOL1 genotyping with clinical decision support (CDS) results are returned to participants and providers immediately (intervention) or at 6 months (control). A subset of participants are re-randomized to pharmacogenomic testing for relevant antihypertensive medications (pharmacogenomic sub-study). CDS alerts encourage appropriate CKD screening and antihypertensive agent use. OUTCOMES: Blood pressure and surveys are assessed at baseline, 3 and 6 months. The primary outcome is change in systolic BP from enrollment to 3 months in individuals with two APOL1 risk alleles. Secondary outcomes include new diagnoses of CKD, systolic blood pressure at 6 months, diastolic BP, and survey results. The pharmacogenomic sub-study will evaluate the relationship of pharmacogenomic genotype and change in systolic BP between baseline and 3 months. RESULTS: To date, the trial has enrolled 3423 participants. CONCLUSIONS: The effect of patient and provider knowledge of APOL1 genotype on systolic blood pressure has not been well-studied. GUARDD-US addresses whether blood pressure improves when patients and providers have this information. GUARDD-US provides a CDS framework for primary care and specialty clinics to incorporate APOL1 genetic risk and pharmacogenomic prescribing in the electronic health record. TRIAL REGISTRATION: ClinicalTrials.govNCT04191824.

Application of unsupervised deep learning algorithms for identification of specific clusters of chronic cough patients from EMR data.

BACKGROUND: Chronic cough affects approximately 10% of adults. The lack of ICD codes for chronic cough makes it challenging to apply supervised learning methods to predict the characteristics of chronic cough patients, thereby requiring the identification of chronic cough patients by other mechanisms. We developed a deep clustering algorithm with auto-encoder embedding (DCAE) to identify clusters of chronic cough patients based on data from a large cohort of 264,146 patients from the Electronic Medical Records (EMR) system. We constructed features using the diagnosis within the EMR, then built a clustering-oriented loss function directly on embedded features of the deep autoencoder to jointly perform feature refinement and cluster assignment. Lastly, we performed statistical analysis on the identified clusters to characterize the chronic cough patients compared to the non-chronic cough patients. RESULTS: The experimental results show that the DCAE model generated three chronic cough clusters and one non-chronic cough patient cluster. We found various diagnoses, medications, and lab tests highly associated with chronic cough patients by comparing the chronic cough cluster with the non-chronic cough cluster. Comparison of chronic cough clusters demonstrated that certain combinations of medications and diagnoses characterize some chronic cough clusters. CONCLUSIONS: To the best of our knowledge, this study is the first to test the potential of unsupervised deep learning methods for chronic cough investigation, which also shows a great advantage over existing algorithms for patient data clustering.

An Automated Line-of-Therapy Algorithm for Adults With Metastatic Non-Small Cell Lung Cancer: Validation Study Using Blinded Manual Chart Review.

BACKGROUND: Extraction of line-of-therapy (LOT) information from electronic health record and claims data is essential for determining longitudinal changes in systemic anticancer therapy in real-world clinical settings. OBJECTIVE: The aim of this retrospective cohort analysis is to validate and refine our previously described open-source LOT algorithm by comparing the output of the algorithm with results obtained through blinded manual chart review. METHODS: We used structured electronic health record data and clinical documents to identify 500 adult patients treated for metastatic non-small cell lung cancer with systemic anticancer therapy from 2011 to mid-2018; we assigned patients to training (n=350) and test (n=150) cohorts, randomly divided proportional to the overall ratio of simple:complex cases (n=254:246). Simple cases were patients who received one LOT and no maintenance therapy; complex cases were patients who received more than one LOT and/or maintenance therapy. Algorithmic changes were performed using the training cohort data, after which the refined algorithm was evaluated against the test cohort. RESULTS: For simple cases, 16 instances of discordance between the LOT algorithm and chart review prerefinement were reduced to 8 instances postrefinement; in the test cohort, there was no discordance between algorithm and chart review. For complex cases, algorithm refinement reduced the discordance from 68 to 62 instances, with 37 instances in the test cohort. The percentage agreement between LOT algorithm output and chart review for patients who received one LOT was 89% prerefinement, 93% postrefinement, and 93% for the test cohort, whereas the likelihood of precise matching between algorithm output and chart review decreased with an increasing number of unique regimens. Several areas of discordance that arose from differing definitions of LOTs and maintenance therapy could not be objectively resolved because of a lack of precise definitions in the medical literature. CONCLUSIONS: Our findings identify common sources of discordance between the LOT algorithm and clinician documentation, providing the possibility of targeted algorithm refinement.

Applying interpretable deep learning models to identify chronic cough patients using EHR data.

BACKGROUND AND OBJECTIVE: Chronic cough (CC) affects approximately 10% of adults. Many disease states are associated with chronic cough, such as asthma, upper airway cough syndrome, bronchitis, and gastroesophageal reflux disease. The lack of an ICD code specific for chronic cough makes it challenging to identify such patients from electronic health records (EHRs). For clinical and research purposes, computational methods using EHR data are urgently needed to identify chronic cough cases. This research aims to investigate the data representations and deep learning algorithms for chronic cough prediction. METHODS: Utilizing real-world EHR data from a large academic healthcare system from October 2005 to September 2015, we investigated Natural Language Representation of the EHR data and systematically evaluated deep learning and traditional machine learning models to predict chronic cough patients. We built these machine learning models using structured data (medication and diagnosis) and unstructured data (clinical notes). RESULTS: The sensitivity and specificity of a transformer-based deep learning algorithm, specifically BERT with attention model, was 0.856 and 0.866, respectively, using structured data (medication and diagnosis). Sensitivity and specificity improved to 0.952 and 0.930 when we combined structured data with symptoms extracted from clinical notes. We further found that the attention mechanism of deep learning models can be used to extract important features that drive the prediction decisions. Compared with our previously published rule-based algorithm, the deep learning algorithm can identify more chronic cough patients with structured data. CONCLUSIONS: By applying deep learning models, chronic cough patients can be reliably identified for prospective or retrospective research through medication and diagnosis data, widely available in EHR and electronic claims data, thus improving the generalizability of the patient identification algorithm. Deep learning models can identify chronic cough patients with even higher sensitivity and specificity when structured and unstructured EHR data are utilized. We anticipate language-based data representation and deep learning models developed in this research could also be productively used for other disease prediction and case identification.

Strategies to Integrate Genomic Medicine into Clinical Care: Evidence from the IGNITE Network.

The complexity of genomic medicine can be streamlined by implementing some form of clinical decision support (CDS) to guide clinicians in how to use and interpret personalized data; however, it is not yet clear which strategies are best suited for this purpose. In this study, we used implementation science to identify common strategies for applying provider-based CDS interventions across six genomic medicine clinical research projects funded by an NIH consortium. Each project's strategies were elicited via a structured survey derived from a typology of implementation strategies, the Expert Recommendations for Implementing Change (ERIC), and follow-up interviews guided by both implementation strategy reporting criteria and a planning framework, RE-AIM, to obtain more detail about implementation strategies and desired outcomes. We found that, on average, the three pharmacogenomics implementation projects used more strategies than the disease-focused projects. Overall, projects had four implementation strategies in common; however, operationalization of each differed in accordance with each study's implementation outcomes. These four common strategies may be important for precision medicine program implementation, and pharmacogenomics may require more integration into clinical care. Understanding how and why these strategies were successfully employed could be useful for others implementing genomic or precision medicine programs in different contexts.