Extracting social determinants of health from electronic health records using natural language processing: a systematic review

ObjectiveSocial determinants of health (SDoH) are nonclinical dispositions that impact patient health risks and clinical outcomes. Leveraging SDoH in clinical decision-making can potentially improve diagnosis, treatment planning, and patient outcomes. Despite increased interest in capturing SDoH in electronic health records (EHRs), such information is typically locked in unstructured clinical notes. Natural language processing (NLP) is the key technology to extract SDoH information from clinical text and expand its utility in patient care and research. This article presents a systematic review of the state-of-the-art NLP approaches and tools that focus on identifying and extracting SDoH data from unstructured clinical text in EHRs.Materials and MethodsA broad literature search was conducted in February 2021 using 3 scholarly databases (ACL Anthology, PubMed, and Scopus) following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A total of 6402 publications were initially identified, and after applying the study inclusion criteria, 82 publications were selected for the final review.ResultsSmoking status (n = 27), substance use (n = 21), homelessness (n = 20), and alcohol use (n = 15) are the most frequently studied SDoH categories. Homelessness (n = 7) and other less-studied SDoH (eg, education, financial problems, social isolation and support, family problems) are mostly identified using rule-based approaches. In contrast, machine learning approaches are popular for identifying smoking status (n = 13), substance use (n = 9), and alcohol use (n = 9).ConclusionNLP offers significant potential to extract SDoH data from narrative clinical notes, which in turn can aid in the development of screening tools, risk prediction models, and clinical decision support systems.     

Impact of event notification services on timely follow-up and rehospitalization among primary care patients at two Veterans Affairs Medical Centers

ObjectiveTo examine the effectiveness of event notification service (ENS) alerts on health care delivery processes and outcomes for older adults.Materials and methodsWe deployed ENS alerts in 2 Veterans Affairs (VA) medical centers using regional health information exchange (HIE) networks from March 2016 to December 2019. Alerts targeted VA-based primary care teams when older patients (aged 65+ years) were hospitalized or attended emergency departments (ED) outside the VA system. We employed a concurrent cohort study to compare postdischarge outcomes between patients whose providers received ENS alerts and those that did not (usual care). Outcome measures included: timely follow-up postdischarge (actual phone call within 7 days or an in-person primary care visit within 30 days) and all-cause inpatient or ED readmission within 30 days. Generalized linear mixed models, accounting for clustering by primary care team, were used to compare outcomes between groups.ResultsCompared to usual care, veterans whose primary care team received notification of non-VA acute care encounters were 4 times more likely to have phone contact within 7 days (AOR = 4.10, P < .001) and 2 times more likely to have an in-person visit within 30 days (AOR = 1.98, P = .007). There were no significant differences between groups in hospital or ED utilization within 30 days of index discharge (P = .057).DiscussionENS was associated with increased timely follow-up following non-VA acute care events, but there was no associated change in 30-day readmission rates. Optimization of ENS processes may be required to scale use and impact across health systems.ConclusionGiven the importance of ENS to the VA and other health systems, this study provides guidance for future research on ENS for improving care coordination and population outcomes.Trial RegistrationClinicalTrials.gov {"type":"clinical-trial","attrs":{"text":"NCT02689076","term_id":"NCT02689076"}}NCT02689076. “Regional Data Exchange to Improve Care for Veterans After Non-VA Hospitalization.” Registered February 23, 2016.     

Towards clinical data-driven eligibility criteria optimization for interventional COVID-19 clinical trials

ObjectiveThis research aims to evaluate the impact of eligibility criteria on recruitment and observable clinical outcomes of COVID-19 clinical trials using electronic health record (EHR) data.Materials and MethodsOn June 18, 2020, we identified frequently used eligibility criteria from all the interventional COVID-19 trials in ClinicalTrials.gov (n = 288), including age, pregnancy, oxygen saturation, alanine/aspartate aminotransferase, platelets, and estimated glomerular filtration rate. We applied the frequently used criteria to the EHR data of COVID-19 patients in Columbia University Irving Medical Center (CUIMC) (March 2020–June 2020) and evaluated their impact on patient accrual and the occurrence of a composite endpoint of mechanical ventilation, tracheostomy, and in-hospital death.ResultsThere were 3251 patients diagnosed with COVID-19 from the CUIMC EHR included in the analysis. The median follow-up period was 10 days (interquartile range 4–28 days). The composite events occurred in 18.1% (n = 587) of the COVID-19 cohort during the follow-up. In a hypothetical trial with common eligibility criteria, 33.6% (690/2051) were eligible among patients with evaluable data and 22.2% (153/690) had the composite event.DiscussionBy adjusting the thresholds of common eligibility criteria based on the characteristics of COVID-19 patients, we could observe more composite events from fewer patients.ConclusionsThis research demonstrated the potential of using the EHR data of COVID-19 patients to inform the selection of eligibility criteria and their thresholds, supporting data-driven optimization of participant selection towards improved statistical power of COVID-19 trials.     

Seriously ill pediatric patient,parent, and clinician perspectives on visualizing symptom data

ObjectiveThis study examined the perspectives on the use of data visualizations and identified key features seriously ill children, their parents, and clinicians prefer to see when visualizing symptom data obtained from mobile health technologies (an Apple Watch and smartphone symptom app).Materials and MethodsChildren with serious illness and their parents were enrolled into a symptom monitoring study then a subset was interviewed for this study. A study team member created symptom data visualizations using the pediatric participant’s mobile technology data. Semi-structured interviews were conducted with a convenience sample of participants (n = 14 children; n = 14 parents). In addition, a convenience sample of clinicians (n = 30) completed surveys. Pediatric and parent participants shared their preferences and perspectives on the symptom visualizations.ResultsWe identified 3 themes from the pediatric and parent participant interviews: increased symptom awareness, communication, and interpretability of the symptom visualizations. Clinicians preferred pie charts and simple bar charts for their ease of interpretation and ability to be used as communication tools. Most clinicians would prefer to see symptom visualizations in the electronic health record.DiscussionMobile health tools offer a unique opportunity to obtain patient-generated health data. Effective, concise symptom visualizations can be used to synthesize key clinical information to inform clinical decisions and promote patient-clinician communication to enhance symptom management.ConclusionsEffectively visualizing complex mobile health data can enhance understanding of symptom dynamics and promote patient-clinician communication, leading to tailored personalized symptom management strategies.     

Feeling better on hemodialysis: user-centered design requirements for promoting patient involvement in the prevention of treatment complications

ObjectiveHemodialysis patients frequently experience dialysis therapy sessions complicated by intradialytic hypotension (IDH), a major patient safety concern. We investigate user-centered design requirements for a theory-informed, peer mentoring-based, informatics intervention to activate patients toward IDH prevention.MethodsWe conducted observations (156 hours) and interviews (n = 28) with patients in 3 hemodialysis clinics, followed by 9 focus groups (including participatory design activities) with patients (n = 17). Inductive and deductive analyses resulted in themes and design principles linked to constructs from social, cognitive, and self-determination theories.ResultsHemodialysis patients want an informatics intervention for IDH prevention that collapses distance between patients, peers, and family; harnesses patients’ strength of character and resolve in all parts of their life; respects and supports patients’ individual needs, preferences, and choices; and links “feeling better on dialysis” to becoming more involved in IDH prevention. Related design principles included designing for: depth of interpersonal connections; positivity; individual choice and initiative; and comprehension of connections and possible actions.DiscussionFindings advance the design of informatics interventions by presenting design requirements for outpatient safety and addressing key design opportunities for informatics to support patient involvement; these include incorporation of behavior change theories. Results also demonstrate the meaning of design choices for hemodialysis patients in the context of their experiences; this may have applicability to other populations with serious illnesses.ConclusionThe resulting patient-facing informatics intervention will be evaluated in a pragmatic cluster-randomized controlled trial in 28 hemodialysis facilities in 4 US regions.     

The quality of social determinants data in the electronic health record: a systematic review

ObjectiveThe aim of this study was to collect and synthesize evidence regarding data quality problems encountered when working with variables related to social determinants of health (SDoH).Materials and MethodsWe conducted a systematic review of the literature on social determinants research and data quality and then iteratively identified themes in the literature using a content analysis process.ResultsThe most commonly represented quality issue associated with SDoH data is plausibility (n = 31, 41%). Factors related to race and ethnicity have the largest body of literature (n = 40, 53%). The first theme, noted in 62% (n = 47) of articles, is that bias or validity issues often result from data quality problems. The most frequently identified validity issue is misclassification bias (n = 23, 30%). The second theme is that many of the articles suggest methods for mitigating the issues resulting from poor social determinants data quality. We grouped these into 5 suggestions: avoid complete case analysis, impute data, rely on multiple sources, use validated software tools, and select addresses thoughtfully.DiscussionThe type of data quality problem varies depending on the variable, and each problem is associated with particular forms of analytical error. Problems encountered with the quality of SDoH data are rarely distributed randomly. Data from Hispanic patients are more prone to issues with plausibility and misclassification than data from other racial/ethnic groups.ConclusionConsideration of data quality and evidence-based quality improvement methods may help prevent bias and improve the validity of research conducted with SDoH data.     

Persistent inequitable design and implementation of patient portals for users at the margins

ObjectiveDiane Forsythe and other feminist scholars have long shown how system builders’ tacit assumptions lead to the systematic erasure of certain users from the design process. In spite of this phenomena being known in the health informatics literature for decades, recent research shows how patient portals and electronic patients health records continue to reproduce health inequalities in Western societies. To better understand this discrepancy between scholarly awareness of such inequities and mainstream design, this study unravels the (conceptual) assumptions and practices of designers and others responsible for portal implementation in the Netherlands and how citizens living in vulnerable circumstances are included in this process.Materials and methodsWe conducted semistructured interviews (n = 24) and questionnaires (n = 14) with portal designers, health professionals, and policy advisors.ResultsIn daily design practices, equity is seen as an “end-of-the-pipeline” concern. Respondents identify health care professionals rather than patients as their main users. If patients are included in the design, this generally entails patients in privileged positions. The needs of citizens living in vulnerable circumstances are not prioritized in design processes. Developers legitimize their focus with reference to the innovation-theoretical approach of the Diffusion of Innovations.Discussion and conclusionAlthough feminist scholars have developed important understandings of the exclusion of citizens living in vulnerable circumstances from portal design, other academic efforts have profoundly shaped daily practices of portal development. Diane Forsythe would likely have taken up this discrepancy as a challenge by finding ways to translate these insights into mainstream systems design.     

A survey of extant organizational and computational setups for deploying predictive models in health systems

ObjectiveArtificial intelligence (AI) and machine learning (ML) enabled healthcare is now feasible for many health systems, yet little is known about effective strategies of system architecture and governance mechanisms for implementation. Our objective was to identify the different computational and organizational setups that early-adopter health systems have utilized to integrate AI/ML clinical decision support (AI-CDS) and scrutinize their trade-offs.Materials and MethodsWe conducted structured interviews with health systems with AI deployment experience about their organizational and computational setups for deploying AI-CDS at point of care.ResultsWe contacted 34 health systems and interviewed 20 healthcare sites (58% response rate). Twelve (60%) sites used the native electronic health record vendor configuration for model development and deployment, making it the most common shared infrastructure. Nine (45%) sites used alternative computational configurations which varied significantly. Organizational configurations for managing AI-CDS were distinguished by how they identified model needs, built and implemented models, and were separable into 3 major types: Decentralized translation (n = 10, 50%), IT Department led (n = 2, 10%), and AI in Healthcare (AIHC) Team (n = 8, 40%).DiscussionNo singular computational configuration enables all current use cases for AI-CDS. Health systems need to consider their desired applications for AI-CDS and whether investment in extending the off-the-shelf infrastructure is needed. Each organizational setup confers trade-offs for health systems planning strategies to implement AI-CDS.ConclusionHealth systems will be able to use this framework to understand strengths and weaknesses of alternative organizational and computational setups when designing their strategy for artificial intelligence.     

Machine-learning model to predict the cause of death using a stacking ensemble method for observational data

ObjectiveCause of death is used as an important outcome of clinical research; however, access to cause-of-death data is limited. This study aimed to develop and validate a machine-learning model that predicts the cause of death from the patient’s last medical checkup.Materials and MethodsTo classify the mortality status and each individual cause of death, we used a stacking ensemble method. The prediction outcomes were all-cause mortality, 8 leading causes of death in South Korea, and other causes. The clinical data of study populations were extracted from the national claims (n = 174 747) and electronic health records (n = 729 065) and were used for model development and external validation. Moreover, we imputed the cause of death from the data of 3 US claims databases (n = 994 518, 995 372, and 407 604, respectively). All databases were formatted to the Observational Medical Outcomes Partnership Common Data Model.ResultsThe generalized area under the receiver operating characteristic curve (AUROC) of the model predicting the cause of death within 60 days was 0.9511. Moreover, the AUROC of the external validation was 0.8887. Among the causes of death imputed in the Medicare Supplemental database, 11.32% of deaths were due to malignant neoplastic disease.DiscussionThis study showed the potential of machine-learning models as a new alternative to address the lack of access to cause-of-death data. All processes were disclosed to maintain transparency, and the model was easily applicable to other institutions.ConclusionA machine-learning model with competent performance was developed to predict cause of death.     

Optimizing the electronic health record: An inpatient sprint addresses provider burnout and improves electronic health record satisfaction

ObjectiveWe sought reduce electronic health record (EHR) burden on inpatient clinicians with a 2-week EHR optimization sprint.Materials and MethodsA team led by physician informaticists worked with 19 advanced practice providers (APPs) in 1 specialty unit. Over 2 weeks, the team delivered 21 EHR changes, and provided 39 one-on-one training sessions to APPs, with an average of 2.8 hours per provider. We measured Net Promoter Score, thriving metrics, and time spent in the EHR based on user log data.ResultsOf the 19 APPs, 18 completed 2 or more sessions. The EHR Net Promoter Score increased from 6 to 60 postsprint (1.0; 95% confidence interval, 0.3-1.8; P = .01). The NPS for the Sprint itself was 93, a very high rating. The 3-axis emotional thriving, emotional recovery, and emotional exhaustion metrics did not show a significant change. By user log data, time spent in the EHR did not show a significant decrease; however, 40% of the APPs responded that they spent less time in the EHR.ConclusionsThis inpatient sprint improved satisfaction with the EHR.     

Influence of an acetate- and a lactate-based balanced infusion solution on acid base physiology and hemodynamics: an observational pilot study

Background

The current pilot study compares the impact of an intravenous infusion of Ringer’s lactate to an acetate-based solution with regard to acid–base balance. The study design included the variables of the Stewart approach and focused on the effective strong ion difference. Because adverse hemodynamic effects have been reported when using acetate buffered solutions in hemodialysis, hemodynamics were also evaluated.

Methods

Twenty-four women who had undergone abdominal gynecologic surgery and who had received either Ringer’s lactate (Strong Ion Difference 28 mmol/L; n = 12) or an acetate-based solution (Strong Ion Difference 36.8 mmol/L; n = 12) according to an established clinical protocol and its precursor were included in the investigation. After induction of general anesthesia, a set of acid–base variables, hemodynamic values and serum electrolytes was measured three times during the next 120 minutes.

Results

Patients received a mean dose of 4,054 ± 450 ml of either one or the other of the solutions. In terms of mean arterial blood pressure and norepinephrine requirements there were no differences to observe between the study groups. pH and serum HCO₃^- concentration decreased slightly but significantly only with Ringer’s lactate. In addition, the acetate-based solution kept the plasma effective strong ion difference more stable than Ringer’s lactate.

Conclusions

Both of the solutions provided hemodynamic stability. Concerning consistency of acid base parameters none of the solutions seemed to be inferior, either. Whether the slight advantages observed for the acetate-buffered solution in terms of stability of pH and plasma HCO₃^- are clinically relevant, needs to be investigated in a larger randomized controlled trial.     

Retrospective analysis of the frequency of centrofacial telangiectasia in systemic sclerosis patients treated with bosentan or ilomedin

Background

Bosentan is a dual endothelin receptor antagonist initially introduced for the treatment of pulmonary arterial hypertension and recently approved for the treatment of digital ulcers in patients with systemic sclerosis (SSc). Our clinical observations indicate that bosentan therapy may be associated with an increased frequency of centrofacial telangiectasia (TAE). Here, we sought to analyze the frequency of TAE in patients with SSc who were treated with either bosentan or the prostacyclin analog iloprost.

Methods

We conducted a retrospective analysis in 27 patients with SSc undergoing therapy with either bosentan (n = 11) or iloprost (n = 16). Standardized photodocumentations of all patients (n = 27) were obtained at a time point ten months after therapy initiation and analyzed. A subgroup of patients (bosentan: n = 6; iloprost: n = 6) was additionally photodocumented prior to therapy initiation, enabling an intraindividual analysis over the course of therapy.

Results

After ten months of therapy patients with SSc receiving bosentan showed a significantly (P = 0.0028) higher frequency of centrofacial TAE (41.6 ± 27.8) as compared to patients with SSc receiving iloprost (14.3 ± 13.1). Detailed subgroup analysis revealed that the frequency of TAE in the bosentan group (n = 6 patients) increased markedly and significantly (P = 0.027) by 44.4 after ten months of therapy (TAE at therapy initiation: 10.8 ± 5.1; TAE after ten months of therapy: 55.2 ± 29.8), whereas an only minor increase of 1.9 was observed in the iloprost group (n = 6 patients; TAE at therapy initiation: 18.3 ± 14.5; TAE after ten months of therapy: 20.2 ± 15.5), yet without reaching statistical significance (P = 0.420).

Conclusions

The use of bosentan may be associated with an increased frequency of TAE in patients with SSc. Patients should be informed about this potential adverse effect prior to therapy. Treatment options may include camouflage or laser therapy.     

Patient interest in and barriers to telemedicine video visits in a multilingual urban safety-net system

ObjectiveTo determine interest in and barriers to video visits in safety-net patients with diverse age, racial/ethnic, or linguistic background.Materials and MethodsWe surveyed patients in an urban safety-net system to assess: interest in video visits; ability to successfully complete test video visits; and barriers to successful completion of test video visits.ResultsAmong 202 participants, of which 177 (87.6%) were persons of color and 113 (55.9%) preferred non-English languages, 132 (65.3%) were interested in and 109 (54.0%) successfully completed a test video visit. Younger age, non-English preference, and prior smartphone application use were associated with interest. Over half (n = 112) reported barriers to video visits; Internet/data access was the most common barrier (n = 50, 24.8%).ConclusionSafety-net patients are interested in video visits and able to successfully complete test visits. Internet or mobile data access is a common barrier in even urban safety-net settings and may impact equitable telemedicine access.     

Bias and fairness assessment of a natural language processing opioid misuse classifier: detection and mitigation of electronic health record data disadvantages across racial subgroups

ObjectivesTo assess fairness and bias of a previously validated machine learning opioid misuse classifier.Materials & MethodsTwo experiments were conducted with the classifier’s original (n = 1000) and external validation (n = 53 974) datasets from 2 health systems. Bias was assessed via testing for differences in type II error rates across racial/ethnic subgroups (Black, Hispanic/Latinx, White, Other) using bootstrapped 95% confidence intervals. A local surrogate model was estimated to interpret the classifier’s predictions by race and averaged globally from the datasets. Subgroup analyses and post-hoc recalibrations were conducted to attempt to mitigate biased metrics.ResultsWe identified bias in the false negative rate (FNR = 0.32) of the Black subgroup compared to the FNR (0.17) of the White subgroup. Top features included “heroin” and “substance abuse” across subgroups. Post-hoc recalibrations eliminated bias in FNR with minimal changes in other subgroup error metrics. The Black FNR subgroup had higher risk scores for readmission and mortality than the White FNR subgroup, and a higher mortality risk score than the Black true positive subgroup (P < .05).DiscussionThe Black FNR subgroup had the greatest severity of disease and risk for poor outcomes. Similar features were present between subgroups for predicting opioid misuse, but inequities were present. Post-hoc mitigation techniques mitigated bias in type II error rate without creating substantial type I error rates. From model design through deployment, bias and data disadvantages should be systematically addressed.ConclusionStandardized, transparent bias assessments are needed to improve trustworthiness in clinical machine learning models.     

Finding commonalities in rare diseases through the undiagnosed diseases network

ObjectiveWhen studying any specific rare disease, heterogeneity and scarcity of affected individuals has historically hindered investigators from discerning on what to focus to understand and diagnose a disease. New nongenomic methodologies must be developed that identify similarities in seemingly dissimilar conditions.Materials and MethodsThis observational study analyzes 1042 patients from the Undiagnosed Diseases Network (2015-2019), a multicenter, nationwide research study using phenotypic data annotated by specialized staff using Human Phenotype Ontology terms. We used Louvain community detection to cluster patients linked by Jaccard pairwise similarity and 2 support vector classifier to assign new cases. We further validated the clusters’ most representative comorbidities using a national claims database (67 million patients).ResultsPatients were divided into 2 groups: those with symptom onset before 18 years of age (n = 810) and at 18 years of age or older (n = 232) (average symptom onset age: 10 [interquartile range, 0-14] years). For 810 pediatric patients, we identified 4 statistically significant clusters. Two clusters were characterized by growth disorders, and developmental delay enriched for hypotonia presented a higher likelihood of diagnosis. Support vector classifier showed 0.89 balanced accuracy (0.83 for Human Phenotype Ontology terms only) on test data.DiscussionsTo set the framework for future discovery, we chose as our endpoint the successful grouping of patients by phenotypic similarity and provide a classification tool to assign new patients to those clusters.ConclusionThis study shows that despite the scarcity and heterogeneity of patients, we can still find commonalities that can potentially be harnessed to uncover new insights and targets for therapy.     

A qualitative analysis of communication workflows between adult day service centers and primary care providers

ObjectivesOur study documented communication workflows across adult day care centers (ADCs) and primary care providers (PCPs) around complex needs of persons living with dementia (PLWD). We also identified barriers and facilitators to productive communication in clinical decision support and clinical information systems.Materials and MethodsWe conducted 6 focus groups with ADC staff (N = 33) and individual semistructured interviews with PCPs (N = 22) in California. The eHealth Enhanced Chronic Care Model was used to frame the directed qualitative content analysis.ResultsOur results captured cumbersome and ineffective workflows currently used to exchange information across PCPs and ADCs. Stakeholders characterized current communication as (1) infrequent, (2) delayed, (3) incomplete, (4) unreliable, (5) irrelevant, and (6) generic. Conversely, communication that was bidirectional, relevant, succinct, and interdisciplinary was needed to elevate the standard of care for PLWD.Discussion and ConclusionADCs possess a wealth of information that can support clinical decision-making across community-based providers involved in the care of PLWD, especially PCPs. However, effective information exchange is mired by complicated workflows that rely on antiquated technologies (eg, facsimile) and standard templates. Current information exchange largely focuses on satisfying regulatory guidelines rather than supporting clinical decision-making. Integrating community-based services into the health care continuum is a necessary step in elevating the standard of care for PLWD. In the absence of interoperable electronic health records, which may not be financially viable for ADCs, other options, such as mobile health, should be explored to facilitate productive information exchange of personalized relevant information.     

A theory-based meta-regression of factors influencing clinical decision support adoption and implementation

ObjectiveThe purpose of the study was to explore the theoretical underpinnings of effective clinical decision support (CDS) factors using the comparative effectiveness results.Materials and MethodsWe leveraged search results from a previous systematic literature review and updated the search to screen articles published from January 2017 to January 2020. We included randomized controlled trials and cluster randomized controlled trials that compared a CDS intervention with and without specific factors. We used random effects meta-regression procedures to analyze clinician behavior for the aggregate effects. The theoretical model was the Unified Theory of Acceptance and Use of Technology (UTAUT) model with motivational control.ResultsThirty-four studies were included. The meta-regression models identified the importance of effort expectancy (estimated coefficient = −0.162; P = .0003); facilitating conditions (estimated coefficient = 0.094; P = .013); and performance expectancy with motivational control (estimated coefficient = 1.029; P = .022). Each of these factors created a significant impact on clinician behavior. The meta-regression model with the multivariate analysis explained a large amount of the heterogeneity across studies (R2 =88.32%).DiscussionThree positive factors were identified: low effort to use, low controllability, and providing more infrastructure and implementation strategies to support the CDS. The multivariate analysis suggests that passive CDS could be effective if users believe the CDS is useful and/or social expectations to use the CDS intervention exist.ConclusionsOverall, a modified UTAUT model that includes motivational control is an appropriate model to understand psychological factors associated with CDS effectiveness and to guide CDS design, implementation, and optimization.