Evaluation of a Remote Patient Monitoring Program During the COVID-19 Pandemic: Retrospective Case Study With a Mixed Methods Explanatory Sequential Design.

BACKGROUND: Community health center (CHC) patients experience a disproportionately high prevalence of chronic conditions and barriers to accessing technologies that might support the management of these conditions. One such technology includes tools used for remote patient monitoring (RPM), the use of which surged during the COVID-19 pandemic. OBJECTIVE: The aim of this study was to assess how a CHC implemented an RPM program during the COVID-19 pandemic. METHODS: This retrospective case study used a mixed methods explanatory sequential design to evaluate a CHC's implementation of a suite of RPM tools during the COVID-19 pandemic. Analyses used electronic health record-extracted health outcomes data and semistructured interviews with the CHC's staff and patients participating in the RPM program. RESULTS: The CHC enrolled 147 patients in a hypertension RPM program. After 6 months of RPM use, mean systolic blood pressure (BP) was 13.4 mm Hg lower and mean diastolic BP 6.4 mm Hg lower, corresponding with an increase in hypertension control (BP<140/90 mm Hg) from 33.3% of patients to 81.5%. Considerable effort was dedicated to standing up the program, reinforced by organizational prioritization of chronic disease management, and by a clinician who championed program implementation. Noted barriers to implementation of the RPM program were limited initial training, lack of sustained support, and complexities related to the RPM device technology. CONCLUSIONS: While RPM technology holds promise for addressing chronic disease management, successful RPM program requires substantial investment in implementation support and technical assistance.

Patient-Portal Compared with Supplemental In-Office Tablet Screening for Health-Related Social Needs in Primary Care.

BACKGROUND: Screening for health-related social needs (HRSN) has become more widespread but the best method of delivering the screening tool is not yet known. OBJECTIVE: Describe HRSN screening completion rate, specifically portal-based and in-person tablet-based screening. DESIGN: Cross-sectional retrospective observational study. PARTICIPANTS: Adults age 18 or older who had a non-acute primary care visit at one of three internal medicine primary care clinics at a large, urban, academic medical center between July 2022 and July 2023. MAIN MEASURES: We identified the proportion of individuals who were screened using the HRSN questionnaire, whether screening was completed by patient-portal or tablet, as well as the degree of burden of HRSN. Using the electronic health record, we explored associations between sociodemographic characteristics and HRSN attributes. KEY RESULTS: Our study included 24,597 patients, of whom 37% completed the HRSN questionnaire. A smaller proportion of Black/African American patients and those with Medicaid insurance completed the questionnaire, yet they comprised a greater percentage of those who screened positive for unmet HRSN (p ≤ 0.001). Most patients completed the questionnaire by patient-portal (86.1%) compared with in-office tablets (14.0%). A larger proportion of those who completed screening by tablet screened positive for HRSN. Of all patients screened, 21.8% were positive for an unmet HRSN and 11.5% had more than one unmet HRSN. CONCLUSIONS: A majority of patients are not being screened for HRSN and results illustrate disparities when screening patients for HRSN through portal-based compared with supplemental in-office tablet-based screening. Prevalence of unmet HRSN varied by demographics such as race and insurance status.

Equivalence of Alcohol Use Disorder Symptom Assessments in Routine Clinical Care When Completed Remotely via Online Patient Portals Versus In Clinic via Paper Questionnaires: Psychometric Evaluation.

BACKGROUND: The National Institute on Alcohol Abuse and Alcoholism (NIAAA) recommends the paper-based or computerized Alcohol Symptom Checklist to assess alcohol use disorder (AUD) symptoms in routine care when patients report high-risk drinking. However, it is unknown whether Alcohol Symptom Checklist response characteristics differ when it is administered online (eg, remotely via an online electronic health record [EHR] patient portal before an appointment) versus in clinic (eg, on paper after appointment check-in). OBJECTIVE: This study evaluated the psychometric performance of the Alcohol Symptom Checklist when completed online versus in clinic during routine clinical care. METHODS: This cross-sectional, psychometric study obtained EHR data from the Alcohol Symptom Checklist completed by adult patients from an integrated health system in Washington state. The sample included patients who had a primary care visit in 2021 at 1 of 32 primary care practices, were due for annual behavioral health screening, and reported high-risk drinking on the behavioral health screen (Alcohol Use Disorder Identification Test-Consumption score ≥7). After screening, patients with high-risk drinking were typically asked to complete the Alcohol Symptom Checklist-an 11-item questionnaire on which patients self-report whether they had experienced each of the 11 AUD criteria listed in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) over a past-year timeframe. Patients could complete the Alcohol Symptom Checklist online (eg, on a computer, smartphone, or tablet from any location) or in clinic (eg, on paper as part of the rooming process at clinical appointments). We examined sample and measurement characteristics and conducted differential item functioning analyses using item response theory to examine measurement consistency across these 2 assessment modalities. RESULTS: Among 3243 patients meeting eligibility criteria for this secondary analysis (2313/3243, 71% male; 2271/3243, 70% White; and 2014/3243, 62% non-Hispanic), 1640 (51%) completed the Alcohol Symptom Checklist online while 1603 (49%) completed it in clinic. Approximately 46% (752/1640) and 48% (764/1603) reported ≥2 AUD criteria (the threshold for AUD diagnosis) online and in clinic (P=.37), respectively. A small degree of differential item functioning was observed for 4 of 11 items. This differential item functioning produced only minimal impact on total scores used clinically to assess AUD severity, affecting total criteria count by a maximum of 0.13 criteria (on a scale ranging from 0 to 11). CONCLUSIONS: Completing the Alcohol Symptom Checklist online, typically prior to patient check-in, performed similarly to an in-clinic modality typically administered on paper by a medical assistant at the time of the appointment. Findings have implications for using online AUD symptom assessments to streamline workflows, reduce staff burden, reduce stigma, and potentially assess patients who do not receive in-person care. Whether modality of DSM-5 assessment of AUD differentially impacts treatment is unknown.

Optimizing word embeddings for small dataset: a case study on patient portal messages from breast cancer patients.

Patient portal messages often relate to specific clinical phenomena (e.g., patients undergoing treatment for breast cancer) and, as a result, have received increasing attention in biomedical research. These messages require natural language processing and, while word embedding models, such as word2vec, have the potential to extract meaningful signals from text, they are not readily applicable to patient portal messages. This is because embedding models typically require millions of training samples to sufficiently represent semantics, while the volume of patient portal messages associated with a particular clinical phenomenon is often relatively small. We introduce a novel adaptation of the word2vec model, PK-word2vec (where PK stands for prior knowledge), for small-scale messages. PK-word2vec incorporates the most similar terms for medical words (including problems, treatments, and tests) and non-medical words from two pre-trained embedding models as prior knowledge to improve the training process. We applied PK-word2vec in a case study of patient portal messages in the Vanderbilt University Medical Center electric health record system sent by patients diagnosed with breast cancer from December 2004 to November 2017. We evaluated the model through a set of 1000 tasks, each of which compared the relevance of a given word to a group of the five most similar words generated by PK-word2vec and a group of the five most similar words generated by the standard word2vec model. We recruited 200 Amazon Mechanical Turk (AMT) workers and 7 medical students to perform the tasks. The dataset was composed of 1389 patient records and included 137,554 messages with 10,683 unique words. Prior knowledge was available for 7981 non-medical and 1116 medical words. In over 90% of the tasks, both reviewers indicated PK-word2vec generated more similar words than standard word2vec (p = 0.01).The difference in the evaluation by AMT workers versus medical students was negligible for all comparisons of tasks' choices between the two groups of reviewers ( p = 0.774 under a paired t-test). PK-word2vec can effectively learn word representations from a small message corpus, marking a significant advancement in processing patient portal messages.

Roger's diffusion of innovations theory and the adoption of a patient portal's digital anamnesis collection tool: study protocol for the MAiBest project.

BACKGROUND: German hospitals are legally obliged to implement digital patient portals within the next years. Systematic reviews show that the use of patient portals may be associated with improved patient-centeredness and workflows. However, mandatory digital healthcare innovations are sometimes not used by the target group as planned or even completely rejected. Based on Roger's theory of innovation diffusion, it can be assumed that the time factor is of particular importance for the adoption of the patient portal. The aim of the project is to assess determinants of patient portal adoption and to examine whether Roger's theory can be confirmed. METHODS: The project investigates the use of the patient portal in three different clinics of a large academic teaching hospital in Germany using a longitudinal study design with three cross-sectional time points (pre, post, post). Doctors and patients are surveyed about factors that predict the use of the patient portal and whether the strength of these factors changes over time. They are also interviewed about possible barriers they experience when using the patient portal or about the reasons why the patient portal is not used. Regression models and content analyses are used to answer the research questions. DISCUSSION: Determinants of patient portal use will be discussed under the light of the temporal component of Roger's theory. At the same time, it is expected that some determinants will remain unchanged over time. Identifying determinants independent of time allows targeting the groups, enabling specific communication strategies to empower these groups to use the patient portal, contributing to an equal health care system. TRIAL REGISTRATION: The study was prospectively registered in the German register of clinical trials (DRKS00033125) in May 2024.

A novel approach to patient portal activation data to power equity improvements.

BACKGROUND: There are significant disparities in access and utilization of patient portals by age, language, race, and ethnicity. MATERIALS AND METHODS: We developed ambulatory and inpatient portal activation equity dashboards to understand disparities in initial portal activation, identify targets for improvement, and enable monitoring of interventions over time. We selected key metrics focused on episodes of care and filters to enable high-level overviews and granular data selection to meet the needs of health system leaders and individual clinical units. RESULTS: In addition to highlighting disparities by age, preferred language, race and ethnicity, and insurance payor, the dashboards enabled development and monitoring of interventions to improve portal activation and equity. DISCUSSION AND CONCLUSIONS: Data visualization tools that provide easily accessible, timely, and customizable data can enable a variety of stakeholders to understand and address healthcare disparities, such as patient portal activation. Further institutional efforts are needed to address the persistent inequities highlighted by these dashboards.

Patient- and Provider-Level Factors Associated with Patient Portal Usage Among Medicaid Recipients.

Background: Patient portals can improve access to electronic health information and enhance patient engagement. However, disparities in patient portal utilization remain, affecting disadvantaged communities disproportionately. This study examined patient- and provider-level factors associated with portal usage among Medicaid recipients in a large federally qualified health center (FQHC) network in Texas. Methods: Deidentified electronic medical records of patients 18 years or older from a large Texas FQHC network were analyzed. The dependent variable was a binary flag indicating portal usage during the study period. Independent variables included patient- and provider-level factors. Patient-level factors included sociodemographic, geographic, and clinical characteristics. Provider characteristics included primary service line, provider type, provider language, and years in practice. Because the analysis was at the individual level, a multivariable logistic regression model focused on adjusted associations between independent variables and portal usage. Results: The analytic sample consisted of 9,271 individuals. Compared with individuals 18-39 years, patients 50 years and older had lower odds (50-64 OR: 0.60, p < 0.001; 65+ OR: 0.51, p < 0.001) of portal usage. Males were less likely to use portals (OR: 0.44, p = 0.03), and compared to Non-Hispanic Whites, Non-Hispanic Black (OR: 0.86, p = 0.02) and Hispanics (OR: 0.83, p < 0.001) were significantly less likely to use portals. Individuals with 1 or more telemedicine consults had a two-times greater odds of portal usage (OR: 1.97, p < 0.001). Compared to individuals who had clinic visits in December 2018, portal usage was significantly higher in the pandemic months (March 2020-November 2020, all p's < 0.01). Importantly, the behavioral health service line had the greatest odds (OR: 1.52, p < 0.001), whereas the dental service line had the lowest odds (OR: 0.69, p = 0.01) compared to family practice. No other provider characteristics were significant. Conclusion: Our finding of significant patient-level factors is important and can contribute to developing appropriate patient-focused health information technology approaches to ensure equitable access and maximize the potential benefits of patient portals in health care delivery.

Enhancing the Management of Benign Prostatic Hyperplasia: The Role of Electronic Health Record Patient Portal Distribution of the International Prostate Symptom Score.

INTRODUCTION: Recent AUA guidelines for the management of benign prostatic hyperplasia (BPH) recommend routine collection of the International Prostate Symptom Score (IPSS) data, but routine collection can be challenging to fully implement. We investigated the impact of distributing the IPSS by electronic patient portal (EPP) on IPSS completion and its impact on BPH management. METHODS: We performed a retrospective, longitudinal study of men undergoing a new patient visit (NPV) for BPH at our academic medical center. From September 2019 to November 2022, we identified patients undergoing an NPV for BPH. Prior to January 2021, the IPSS was collected in person at NPVs via paper forms; afterwards, the IPSS was distributed before the NPV using the EPP. Our primary outcome was IPSS completion; secondary outcomes were new BPH medications and BPH surgery ordered within 6 months. RESULTS: We identified 485 patients who underwent an NPV for BPH. EPP implementation significantly increased IPSS questionnaire completion (36.5% vs 56.9%, P < .0001). Following EPP implementation, we found that new BPH medications ordered at time of NPV decreased (10.4% vs 4.7%, P = .02). Although BPH surgery ordered within 6 months was similar, patients following EPP implementation had shorter time to BPH surgery compared to prior. CONCLUSIONS: Our study revealed that EPP distribution of the IPSS improves IPSS collection compliance, aligning our practice closer with AUA guidelines. Routine collection of the IPSS may impact clinical practice through the detection of more severe BPH, which reduces medical BPH management and time to definitive BPH therapy. Further work is needed to confirm findings.

Association Between Patient Portal Activities and End-of-Life Outcomes Among Deceased Patients in the Last 12 Months of Life.

Objective: The objective of this study was to examine the association between portal use and end-of-life (EOL) outcomes in the last year of life. Methods: A retrospective cohort (n = 6,517) study at Kaiser Permanente Colorado among adults with serious illness deceased between January 1, 2016, and June 30, 2019. Portal use was categorized into engagement types: no use, nonactive, active without a provider, and active with a provider. EOL outcomes were hospitalizations in the month before death, last-year advance directive completion, and hospice use. Association between EOL outcomes and levels of portal use was assessed using χ2 statistics and generalized linear models. Results: Higher portal engagement types were associated with higher rates of hospitalizations (p = 0.0492), advance directive completion (p = 0.0226), and hospice use (p = 0.0070). Conclusion: Portal use in the last year of life was associated with increases in a poor EOL outcome, hospitalizations, and beneficial EOL outcomes, advance directives, and hospice care.

Using large language model to guide patients to create efficient and comprehensive clinical care message.

OBJECTIVE: This study aims to investigate the feasibility of using Large Language Models (LLMs) to engage with patients at the time they are drafting a question to their healthcare providers, and generate pertinent follow-up questions that the patient can answer before sending their message, with the goal of ensuring that their healthcare provider receives all the information they need to safely and accurately answer the patient's question, eliminating back-and-forth messaging, and the associated delays and frustrations. METHODS: We collected a dataset of patient messages sent between January 1, 2022 to March 7, 2023 at Vanderbilt University Medical Center. Two internal medicine physicians identified 7 common scenarios. We used 3 LLMs to generate follow-up questions: (1) Comprehensive LLM Artificial Intelligence Responder (CLAIR): a locally fine-tuned LLM, (2) GPT4 with a simple prompt, and (3) GPT4 with a complex prompt. Five physicians rated them with the actual follow-ups written by healthcare providers on clarity, completeness, conciseness, and utility. RESULTS: For five scenarios, our CLAIR model had the best performance. The GPT4 model received higher scores for utility and completeness but lower scores for clarity and conciseness. CLAIR generated follow-up questions with similar clarity and conciseness as the actual follow-ups written by healthcare providers, with higher utility than healthcare providers and GPT4, and lower completeness than GPT4, but better than healthcare providers. CONCLUSION: LLMs can generate follow-up patient messages designed to clarify a medical question that compares favorably to those generated by healthcare providers.

Automatic uncovering of patient primary concerns in portal messages using a fusion framework of pretrained language models.

OBJECTIVES: The surge in patient portal messages (PPMs) with increasing needs and workloads for efficient PPM triage in healthcare settings has spurred the exploration of AI-driven solutions to streamline the healthcare workflow processes, ensuring timely responses to patients to satisfy their healthcare needs. However, there has been less focus on isolating and understanding patient primary concerns in PPMs-a practice which holds the potential to yield more nuanced insights and enhances the quality of healthcare delivery and patient-centered care. MATERIALS AND METHODS: We propose a fusion framework to leverage pretrained language models (LMs) with different language advantages via a Convolution Neural Network for precise identification of patient primary concerns via multi-class classification. We examined 3 traditional machine learning models, 9 BERT-based language models, 6 fusion models, and 2 ensemble models. RESULTS: The outcomes of our experimentation underscore the superior performance achieved by BERT-based models in comparison to traditional machine learning models. Remarkably, our fusion model emerges as the top-performing solution, delivering a notably improved accuracy score of 77.67 ± 2.74% and an F1 score of 74.37 ± 3.70% in macro-average. DISCUSSION: This study highlights the feasibility and effectiveness of multi-class classification for patient primary concern detection and the proposed fusion framework for enhancing primary concern detection. CONCLUSIONS: The use of multi-class classification enhanced by a fusion of multiple pretrained LMs not only improves the accuracy and efficiency of patient primary concern identification in PPMs but also aids in managing the rising volume of PPMs in healthcare, ensuring critical patient communications are addressed promptly and accurately.

Improving Asthma Control Test completion rates across a multi-site pediatric pulmonary clinic network: a quality improvement initiative.

OBJECTIVE: Assessing asthma control is an essential part of the outpatient management of children with asthma and can be performed through validated questionnaires such as the Asthma Control Test (ACT). Systematic approaches to incorporating the ACT in outpatient visits are often lacking, contributing to inconsistent completion rates. We conducted a quality improvement initiative to increase the proportion of visits where the ACT is completed for children with asthma in our multi-site pediatric pulmonary clinic network. METHODS: We developed an intervention of sending the ACT questionnaire to patients and caregivers through the electronic patient portal to complete prior to their visits. This strategy was first piloted at one clinic beginning in July 2020 and then expanded to 5 other clinics in the network in October 2020. Our outcome measure was average monthly proportion of visits with a completed ACT, tracked using statistical process control charts. The process measure was method of ACT completion tracked using run charts. RESULTS: At the pilot clinic, average monthly completion rate rose within 3 months of the intervention from 27% to 72% and was sustained more than 22 months. Completion across all clinics increased from 57% pre-intervention to 76% post-intervention. Importantly, the intervention did not rely on clinic staff to administer the questionnaire and did not interfere with existing clinic flow. CONCLUSION: An intervention of delivering the ACT electronically to patients and caregivers for completion prior to visits led to a rapid and sustained improvement in ACT completion rates across a large, pediatric pulmonary clinic network.

Care partners and consumer health information technology: A framework to guide systems-level initiatives in support of digital health equity.

Introduction: Consumer-oriented health information technologies (CHIT) such as the patient portal have a growing role in care delivery redesign initiatives such as the Learning Health System. Care partners commonly navigate CHIT demands alongside persons with complex health and social needs, but their role is not well specified. Methods: We assemble evidence and concepts from the literature describing interpersonal communication, relational coordination theory, and systems-thinking to develop an integrative framework describing the care partner's role in applied CHIT innovations. Our framework describes pathways through which systematic engagement of the care partner affects longitudinal work processes and multi-level outcomes relevant to Learning Health Systems. Results: Our framework is grounded in relational coordination, an emerging theory for understanding the dynamics of coordinating work that emphasizes role-based relationships and communication, and the Systems Engineering Initiative for Patient Safety (SEIPS) model. Cross-cutting work systems geared toward explicit and purposeful support of the care partner role through CHIT may advance work processes by promoting frequent, timely, accurate, problem-solving communication, reinforced by shared goals, shared knowledge, and mutual respect between patients, care partners, and care team. We further contend that systematic engagement of the care partner in longitudinal work processes exerts beneficial effects on care delivery experiences and efficiencies at both individual and organizational levels. We discuss the utility of our framework through the lens of an illustrative case study involving patient portal-mediated pre-visit agenda setting. Conclusions: Our framework can be used to guide applied embedded CHIT interventions that support the care partner role and bring value to Learning Health Systems through advancing digital health equity, improving user experiences, and driving efficiencies through improved coordination within complex work systems.

Differential utilization of the online patient portal for completion of health-related social needs screening during routine gynecologic cancer care.

BACKGROUND: Telehealth technologies offer efficient ways to deliver health-related social needs (HRSN) screening in cancer care, but these methods may not reach all populations. The authors examined patient characteristics associated with using an online patient portal (OPP) to complete HRSN screening as part of gynecologic cancer care. METHODS: From June 2021 to June 2023, patients in a gynecologic oncology clinic completed validated HRSN screening questions either (1) using the OPP (independently before the visit) or (2) in person (verbally administered by clinic staff). The authors examined the prevalence of HRSN according to activated OPP status and, in a restricted subgroup, used stepwise multivariate Poisson regression to identify associations between patient and visit characteristics and using the OPP. RESULTS: Of 1616 patients, 87.4% (n = 1413) had an activated OPP. Patients with inactive OPPs (vs. activated OPPs) more frequently reported two or more needs (10% vs 5%; p < .01). Of 986 patients in the restricted cohort, 52% used the OPP to complete screening. The final multivariable model indicated that patients were less likely to use the OPP if they were Black (vs. White; adjusted relative risk [aRR], 0.70; 95% confidence interval [CI], 0.59-0.83); not employed (vs. employed; aRR, 0.81; 95% CI, 0.68-0.97), or had low measures of OPP engagement (aRR, 0.80; 95% CI, 0.68-0.92). New versus established patients were 21% more likely to use the OPP (aRR, 1.21; 95% CI, 1.06-1.38). CONCLUSIONS: Differential use of the OPP suggested that over-reliance on digital technologies could limit the ability to reach those populations that have social factors already associated with cancer outcome disparities. Cancer centers should consider using multiple delivery methods for HRSN screening to maximize reach to all populations.

Using patient portals for large-scale recruitment of individuals underrepresented in biomedical research: an evaluation of engagement patterns throughout the patient portal recruitment process at a single site within the All of Us Research Program.

OBJECTIVE: To evaluate the use of patient portal messaging to recruit individuals historically underrepresented in biomedical research (UBR) to the All of Us Research Program (AoURP) at a single recruitment site. MATERIALS AND METHODS: Patient portal-based recruitment was implemented at Columbia University Irving Medical Center. Patient engagement was assessed using patient's electronic health record (EHR) at four recruitment stages: Consenting to be contacted, opening messages, responding to messages, and showing interest in participating. Demographic and socioeconomic data were also collected from patient's EHR and univariate logistic regression analyses were conducted to assess patient engagement. RESULTS: Between October 2022 and November 2023, a total of 59 592 patients received patient portal messages inviting them to join the AoURP. Among them, 24 445 (41.0%) opened the message, 8983 (15.1%) responded, and 3765 (6.3%) showed interest in joining the program. Though we were unable to link enrollment data with EHR data, we estimate about 2% of patients contacted ultimately enrolled in the AoURP. Patients from underrepresented race and ethnicity communities had lower odds of consenting to be contacted and opening messages, but higher odds of showing interest after responding. DISCUSSION: Patient portal messaging provided both patients and recruitment staff with a more efficient approach to outreach, but patterns of engagement varied across UBR groups. CONCLUSION: Patient portal-based recruitment enables researchers to contact a substantial number of participants from diverse communities. However, more effort is needed to improve engagement from underrepresented racial and ethnic groups at the early stages of the recruitment process.

Minors' and guardian access to and use of a national patient portal: A retrospective comparative case study of Sweden and Finland.

BACKGROUND: Approaches to implementing online record access (ORA) via patient portals for minors and guardians vary internationally, as more countries continue to develop patient-accessible electronic health records (PAEHR) systems. Evidence of ORA usage and country-specific practices to allow or block minors' and guardians' access to minors' records during adolescence (i.e. access control practices) may provide a broader understanding of possible approaches and their implications for minors' confidentiality and guardian support. AIM: To describe and compare minors' and guardian proxy users' PAEHR usage in Sweden and Finland. Furthermore, to investigate the use of country-specific access control practices. METHODS: A retrospective, observational case study was conducted. Data were collected from PAEHR administration services in Sweden and Finland and proportional use was calculated based on population statistics. Descriptive statistics were used to analyze the results. RESULTS: In both Sweden and Finland, the proportion of adolescents accessing their PAEHR increased from younger to older age-groups reaching the proportion of 59.9 % in Sweden and 84.8 % in Finland in the age-group of 17-year-olds. The PAEHR access gap during early adolescence in Sweden may explain the lower proportion of users among those who enter adulthood. Around half of guardians in Finland accessed their minor children's records in 2022 (46.1 %), while Swedish guardian use was the highest in 2022 for newborn children (41.8 %), and decreased thereafter. Few, mainly guardians, applied for extended access in Sweden. In Finland, where a case-by-case approach to access control relies on healthcare professionals' (HCPs) consideration of a minor's maturity, 95.8 % of minors chose to disclose prescription information to their guardians. CONCLUSION: While age-based access control practices can hamper ORA for minors and guardians, case-by-case approach requires HCP resources and careful guidance to ensure equality between patients. Guardians primarily access minors' records during early childhood and adolescents show willingness to share their PAEHR with parents.

Optimizing Word Embeddings for Patient Portal Message Datasets with a Small Number of Samples.

Background: Patient portal messages often relate to specific clinical phenomena (e.g., patients undergoing treatment for breast cancer) and, as a result, have received increasing attention in biomedical research. These messages require natural language processing and, while word embedding models, such as word2vec, have the potential to extract meaningful signals from text, they are not readily applicable to patient portal messages. This is because embedding models typically require millions of training samples to sufficiently represent semantics, while the volume of patient portal messages associated with a particular clinical phenomenon is often relatively small. Objective: We introduce a novel adaptation of the word2vec model, PK-word2vec, for small-scale messages. Methods: PK-word2vec incorporates the most similar terms for medical words (including problems, treatments, and tests) and non-medical words from two pre-trained embedding models as prior knowledge to improve the training process. We applied PK-word2vec on patient portal messages in the Vanderbilt University Medical Center electric health record system sent by patients diagnosed with breast cancer from December 2004 to November 2017. We evaluated the model through a set of 1000 tasks, each of which compared the relevance of a given word to a group of the five most similar words generated by PK-word2vec and a group of the five most similar words generated by the standard word2vec model. We recruited 200 Amazon Mechanical Turk (AMT) workers and 7 medical students to perform the tasks. Results: The dataset was composed of 1,389 patient records and included 137,554 messages with 10,683 unique words. Prior knowledge was available for 7,981 non-medical and 1,116 medical words. In over 90% of the tasks, both reviewers indicated PK-word2vec generated more similar words than standard word2vec (p=0.01).The difference in the evaluation by AMT workers versus medical students was negligible for all comparisons of tasks' choices between the two groups of reviewers (p = 0.774 under a paired t-test). Conclusions: PK-word2vec can effectively learn word representations from a small message corpus, marking a significant advancement in processing patient portal messages.