Use of electronic health record data to examine administrations of pro re nata analgesics during hip fracture post-acute care.

PURPOSE: Medications prescribed to older adults in US skilled nursing facilities (SNF) and administrations of pro re nata (PRN) "as needed" medications are unobservable in Medicare insurance claims. There is an ongoing deficit in our understanding of medication use during post-acute care. Using SNF electronic health record (EHR) datasets, including medication orders and barcode medication administration records, we described patterns of PRN analgesic prescribing and administrations among SNF residents with hip fracture. METHODS: Eligible participants resided in SNFs owned by 11 chains, had a diagnosis of hip fracture between January 1, 2018 to August 2, 2021, and received at least one administration of an analgesic medication in the 100 days after the hip fracture. We described the scheduling of analgesics, the proportion of available PRN doses administered, and the proportion of days with at least one PRN analgesic administration. RESULTS: Among 24 038 residents, 57.3% had orders for PRN acetaminophen, 67.4% PRN opioids, 4.2% PRN non-steroidal anti-inflammatory drugs, and 18.6% PRN combination products. The median proportion of available PRN doses administered per drug was 3%-50% and the median proportion of days where one or more doses of an ordered PRN analgesic was administered was 25%-75%. Results differed by analgesic class and the number of administrations ordered per day. CONCLUSIONS: EHRs can be leveraged to ascertain precise analgesic exposures during SNF stays. Future pharmacoepidemiology studies should consider linking SNF EHRs to insurance claims to construct a longitudinal history of medication use and healthcare utilization prior to and during episodes of SNF care.

Relationship between youth cardiometabolic health and physical activity in medical records.

BACKGROUND: Thers is limited research examining modifiable cardiometabolic risk factors with a single-item health behavior question obtained during a clinic visit. Such information could support clinicians in identifying patients at risk for adverse cardiometabolic health. We investigated if children meeting physical activity or screen time recommendations, collected during clinic visits, have better cardiometabolic health than children not meeting recommendations. We hypothesized that children meeting either recommendation would have fewer cardiometabolic risk factors. METHODS AND FINDINGS: This cross-sectional study used data from electronic medical records (EMRs) between January 1, 2013 through December 30, 2017 from children (2-18 years) with a well child visits and data for ≥1 cardiometabolic risk factor (i.e., systolic and diastolic blood pressure, glycated hemoglobin, alanine transaminase, high-density and low-density lipoprotein, total cholesterol, and/or triglycerides). Physical activity and screen time were patient/caregiver-reported. Analyses included EMRs from 63,676 well child visits by 30,698 unique patients (49.3% female; 41.7% Black, 31.5% Hispanic). Models that included data from all visits indicated children meeting physical activity recommendations had reduced risk for abnormal blood pressure (odds ratio [OR] = 0.91, 95%CI 0.86, 0.97; p = 0.002), glycated hemoglobin (OR = 0.83, 95%CI 0.75, 0.91; p = 0.00006), alanine transaminase (OR = 0.85, 95%CI 0.79, 0.92; p = 0.00001), high-density lipoprotein (OR = 0.88, 95%CI 0.82, 0.95; p = 0.0009), and triglyceride values (OR = 0.89, 95%CI 0.83, 0.96; p = 0.002). Meeting screen time recommendations was not associated with abnormal cardiometabolic risk factors. CONCLUSION: Collecting information on reported adherence to meeting physical activity recommendations can provide clinicians with additional information to identify patients with a higher risk of adverse cardiometabolic health.

Identifying subtypes of type 2 diabetes mellitus with machine learning: development, internal validation, prognostic validation and medication burden in linked electronic health records in 420 448 individuals.

INTRODUCTION: None of the studies of type 2 diabetes (T2D) subtyping to date have used linked population-level data for incident and prevalent T2D, incorporating a diverse set of variables, explainable methods for cluster characterization, or adhered to an established framework. We aimed to develop and validate machine learning (ML)-informed subtypes for type 2 diabetes mellitus (T2D) using nationally representative data. RESEARCH DESIGN AND METHODS: In population-based electronic health records (2006-2020; Clinical Practice Research Datalink) in individuals ≥18 years with incident T2D (n=420 448), we included factors (n=3787), including demography, history, examination, biomarkers and medications. Using a published framework, we identified subtypes through nine unsupervised ML methods (K-means, K-means++, K-mode, K-prototype, mini-batch, agglomerative hierarchical clustering, Birch, Gaussian mixture models, and consensus clustering). We characterized clusters using intracluster distributions and explainable artificial intelligence (AI) techniques. We evaluated subtypes for (1) internal validity (within dataset; across methods); (2) prognostic validity (prediction for 5-year all-cause mortality, hospitalization and new chronic diseases); and (3) medication burden. RESULTS: Development: We identified four T2D subtypes: metabolic, early onset, late onset and cardiometabolic. Internal validity: Subtypes were predicted with high accuracy (F1 score >0.98). Prognostic validity: 5-year all-cause mortality, hospitalization, new chronic disease incidence and medication burden differed across T2D subtypes. Compared with the metabolic subtype, 5-year risks of mortality and hospitalization in incident T2D were highest in late-onset subtype (HR 1.95, 1.85-2.05 and 1.66, 1.58-1.75) and lowest in early-onset subtype (1.18, 1.11-1.27 and 0.85, 0.80-0.90). Incidence of chronic diseases was highest in late-onset subtype and lowest in early-onset subtype. Medications: Compared with the metabolic subtype, after adjusting for age, sex, and pre-T2D medications, late-onset subtype (1.31, 1.28-1.35) and early-onset subtype (0.83, 0.81-0.85) were most and least likely, respectively, to be prescribed medications within 5 years following T2D onset. CONCLUSIONS: In the largest study using ML to date in incident T2D, we identified four distinct subtypes, with potential future implications for etiology, therapeutics, and risk prediction.

Variation in clinical presentation of pediatric-onset and adult-onset atopic dermatitis: a retrospective, single-center, chart review of adults with atopic dermatitis from the United States.

Atopic dermatitis (AD) is a chronic skin condition that can manifest in childhood and persist into adulthood or can present de novo in adults. The clinical presentation of adults with AD may differ among those with pediatric-onset versus adult-onset disease and potential differences between both groups remain to be better characterized. These atypical features might not be encompassed as part of current diagnostic criteria for AD, such as the Hanifin-Rajka (H-R) and the U.K. Working Party (UKWP) criteria. We conducted a retrospective chart review of the electronic medical records of a large, single, academic center to compare the clinical characteristics between adult-onset and pediatric onset AD and examine the proportion of patients who meet the H-R and/or UKWP criteria. Our single-center retrospective chart review included adults (≥ 18 years of age) with any AD-related ICD-10 codes, ≥ 2 AD-related visits, and a recorded physician-confirmed AD diagnosis. Descriptive statistics were used to compare adults with pediatric-onset (< 18 years of age) and adult-onset (≥ 18 years of age) AD. Logistic regression and x2 test were used to compare groups. We found that, compared to pediatric-onset AD, adults with adult-onset AD had less flexural involvement, flexural lichenification and a personal and family history of other atopic diseases. Compared to adults with pediatric-onset AD, adults with adult-onset AD had greater involvement of the extensor surfaces and more nummular eczema compared to pediatric-onset AD. In our cohort, adults with adult-onset AD were less likely to meet H-R and UKWP criteria compared to pediatric-onset AD. Adults with adult-onset AD may present with a clinical presentation that is different from those with pediatric-onset AD, which may not be completely captured by current AD criteria such as the H-R and UWKP criteria. This can lead to possibly mis- or underdiagnosing AD in adults. Thus, understanding the differences and working towards modifying criteria for adult-onset AD has the potential to improve accurate diagnosis of adults with AD.

Practice Facilitation and Clinical Performance Feedback Using the Electronic Health Record Improved Blood Pressure and Cholesterol Management in Small Primary Care Practices in New York City.

Cardiovascular disease (CVD) disproportionately affects people of color and those with lower household income. Improving blood pressure (BP) and cholesterol management for those with or at risk for CVD can improve health outcomes. The New York City Department of Health implemented clinical performance feedback with practice facilitation (PF) in 134 small primary care practices serving on average over 84% persons of color. Facilitators reviewed BP and cholesterol management data on performance dashboards and guided practices to identify and outreach to patients with suboptimal BP and cholesterol management. Despite disruptions from the COVID-19 pandemic, practices demonstrated significant improvements in BP (68%-75%, P < .001) and cholesterol management (72%-78%, P = .01). Prioritizing high-need neighborhoods for impactful resource investment, such as PF and data sharing, may be a promising approach to reducing CVD and hypertension inequities in areas heavily impacted by structural racism.

Preparing to Address Social Determinants of Health (SDOH): Approaches to Clinic Transformation.

CONTEXT: Pennsylvanians' health is influenced by numerous social determinants of health (SDOH). Integrating SDOH data into electronic health records (EHRs) is critical to identifying health disparities, informing public health policies, and devising interventions. Nevertheless, challenges remain in its implementation within clinical settings. In 2018, the Pennsylvania Department of Health (PADOH) received the Centers for Disease Control and Prevention's DP18-1815 "Improving the Health of Americans Through Prevention and Management of Diabetes and Heart Disease and Stroke" grant to strengthen SDOH data integration in Pennsylvania practices. IMPLEMENTATION: Quality Insights was contracted by PADOH to provide training tailored to each practice's readiness, an International Classification of Diseases, Tenth Revision (ICD-10) guide for SDOH, Continuing Medical Education on SDOH topics, and introduced the PRAPARE toolkit to streamline SDOH data integration and address disparities. Dissemination efforts included a podcast highlighting success stories and lessons learned from practices. From 2019 to 2022, Quality Insights and the University of Pittsburgh Evaluation Institute for Public Health (Pitt evaluation team) executed a mixed-methods evaluation. FINDINGS: During 2019-2022, Quality Insights supported 100 Pennsylvania practices in integrating SDOH data into EHR systems. Before COVID-19, 82.8% actively collected SDOH data, predominantly using PRAPARE tool (62.7%) and SDOH ICD-10 codes (80.4%). Amidst COVID-19, these statistics shifted to 65.1%, 45.2%, and 42.7%, respectively. Notably, the pandemic highlighted the importance of SDOH assessment and catalyzed some practices' utilization of SDOH data. Progress was evident among practices, with additional contribution to other DP18-1815 objectives. The main challenge was the variable understanding, utilization, and capability of handling SDOH data across practices. Effective strategies involved adaptable EHR systems, persistent efforts by Quality Insights, and the presence of change champions within practices. DISCUSSION: The COVID-19 pandemic strained staffing in many practices, impeding SDOH data integration into EHRs. Addressing the diverse understanding and use of SDOH data requires standardized training and procedures. Customized support and sustained engagement by facilitating organizations are paramount in ensuring practices' efficient SDOH data collection and integration.

Effect of integration of prescription drug monitoring program data in the electronic health record on queries by primary care providers.

Objective: To evaluate the impact of PDMP integration in the EHR on provider query rates within twelve primary care clinics in one academic medical center. Methods: Using linked data from the EHR and state PDMP program, we evaluated changes in PDMP query rates using a stepped-wedge observational design where integration was implemented in three waves (four clinics per wave) over a five-month period (May, July, September 2019). Multivariable negative binomial general estimating equations (GEE) models assessed changes in PDMP query rates, overall and across several provider and clinic-level subgroups. Results: Among 206 providers in PDMP integrated clinics, the average number of queries per provider per month increased significantly from 1.43 (95% CI 1.07 - 1.91) pre-integration to 3.94 (95% CI 2.96 - 5.24) post-integration, a 2.74-fold increase (95% CI 2.11 to 3.59; p < .0001). Those in the lowest quartile of PDMP use pre-integration increased 36.8-fold (95% CI 16.91 - 79.95) after integration, significantly more than other pre-integration PDMP use quartiles. Conclusions: Integration of the PDMP in the EHR significantly increased the use of the PDMP overall and across all studied subgroups. PDMP use increased to a greater degree among providers with lower PDMP use pre-integration.

Prognostic model of in-hospital ischemic stroke mortality based on an electronic health record cohort in Indonesia.

Stroke patients rarely have satisfactory survival, which worsens further if comorbidities develop in such patients. Limited data availability from Southeast Asian countries, especially Indonesia, has impeded the disentanglement of post-stroke mortality determinants. This study aimed to investigate predictors of in-hospital mortality in patients with ischemic stroke (IS). This retrospective observational study used IS medical records from the National Brain Centre Hospital, Jakarta, Indonesia. A theoretically driven Cox's regression and Fine-Gray models were established by controlling for age and sex to calculate the hazard ratio of each plausible risk factor for predicting in-hospital stroke mortality and addressing competing risks if they existed. This study finally included 3,278 patients with IS, 917 (28%) of whom had cardiovascular disease and 376 (11.5%) suffered renal disease. Bivariate exploratory analysis revealed lower blood levels of triglycerides, low density lipoprotein, and total cholesterol associated with in-hospital-stroke mortality. The average age of patients with post-stroke mortality was 64.06 ± 11.32 years, with a mean body mass index (BMI) of 23.77 kg/m2 and a median Glasgow Coma Scale (GCS) score of 12 and an IQR of 5. Cardiovascular disease was significantly associated with IS mortality risk. NIHSS score at admission (hazard ratio [HR] = 1.04; 95% confidence interval [CI]: 1.00-1.07), male sex (HR = 1.51[1.01-2.26] and uric acid level (HR = 1.02 [1.00-1.03]) predicted survivability. Comorbidities, such as cardiovascular disease (HR = 2.16 [1.37-3.40], pneumonia (HR = 2.43 [1.42-4.15] and sepsis (HR = 2.07 [1.09-3.94, had higher hazards for post-stroke mortality. Contrarily, the factors contributing to a lower hazard of mortality were BMI (HR = 0.94 [0.89-0.99]) and GCS (HReye = 0.66 [0.48-0.89]. In summary, our study reported that male sex, NIHSS, uric acid level, cardiovascular diseases, pneumonia, sepsis. BMI, and GCS on admission were strong determinants of in-hospital mortality in patients with IS.

Inclusion of Drinking Water Source and Testing Questions into Electronic Medical Records: Advancing Environmental Medicine and Public Health Outreach in New Jersey.

Chronic arsenic exposure is associated with adverse health outcomes, and early life exposure is particularly damaging. Households with pregnant people and young children drinking from unregulated wells in arsenic-prevalent regions are therefore a public health priority for outreach and intervention. A partnership between Columbia University, New Jersey government partners, and Hunterdon Healthcare has informed Hunterdon County residents of the risks faced from drinking arsenic-contaminated water and offered free well testing through a practice-based water test kit distribution and an online patient portal outreach. Encouraged by those successes, Hunterdon Healthcare incorporated questions about drinking water source and arsenic testing history into the electronic medical record (EMR) template used by most primary care practices in Hunterdon County. The new EMR fields allow for additional targeting of risk-based outreach and water test kit distribution, offering promising new opportunities for public health and environmental medicine outreach, surveillance, and research.

Leveraging Electronic Health Records to Assess Residential Mobility Among Veterans in the Veterans Health Administration.

BACKGROUND: Residential mobility, or a change in residence, can influence health care utilization and outcomes. Health systems can leverage their patients' residential addresses stored in their electronic health records (EHRs) to better understand the relationships among patients' residences, mobility, and health. The Veteran Health Administration (VHA), with a unique nationwide network of health care systems and integrated EHR, holds greater potential for examining these relationships. METHODS: We conducted a cross-sectional analysis to examine the association of sociodemographics, clinical conditions, and residential mobility. We defined residential mobility by the number of VHA EHR residential addresses identified for each patient in a 1-year period (1/1-12/31/2018), with 2 different addresses indicating one move. We used generalized logistic regression to model the relationship between a priori selected correlates and residential mobility as a multinomial outcome (0, 1, ≥2 moves). RESULTS: In our sample, 84.4% (n=3,803,475) veterans had no move, 13.0% (n=587,765) had 1 move, and 2.6% (n=117,680) had ≥2 moves. In the multivariable analyses, women had greater odds of moving [aOR=1.11 (95% CI: 1.10,1.12) 1 move; 1.27 (1.25,1.30) ≥2 moves] than men. Veterans with substance use disorders also had greater odds of moving [aOR=1.26 (1.24,1.28) 1 move; 1.77 (1.72,1.81) ≥2 moves]. DISCUSSION: Our study suggests about 16% of veterans seen at VHA had at least 1 residential move in 2018. VHA data can be a resource to examine relationships between place, residential mobility, and health.

MSGene: a multistate model using genetic risk and the electronic health record applied to lifetime risk of coronary artery disease.

Coronary artery disease (CAD) is the leading cause of death among adults worldwide. Accurate risk stratification can support optimal lifetime prevention. Current methods lack the ability to incorporate new information throughout the life course or to combine innate genetic risk factors with acquired lifetime risk. We designed a general multistate model (MSGene) to estimate age-specific transitions across 10 cardiometabolic states, dependent on clinical covariates and a CAD polygenic risk score. This model is designed to handle longitudinal data over the lifetime to address this unmet need and support clinical decision-making. We analyze longitudinal data from 480,638 UK Biobank participants and compared predicted lifetime risk with the 30-year Framingham risk score. MSGene improves discrimination (C-index 0.71 vs 0.66), age of high-risk detection (C-index 0.73 vs 0.52), and overall prediction (RMSE 1.1% vs 10.9%), in held-out data. We also use MSGene to refine estimates of lifetime absolute risk reduction from statin initiation. Our findings underscore our multistate model's potential public health value for accurate lifetime CAD risk estimation using clinical factors and increasingly available genetics toward earlier more effective prevention.

Development of a Longitudinal Prostate Cancer Transcriptomic and Clinical Data Linkage.

Importance: Although tissue-based gene expression testing has become widely used for prostate cancer risk stratification, its prognostic performance in the setting of clinical care is not well understood. Objective: To develop a linkage between a prostate genomic classifier (GC) and clinical data across payers and sites of care in the US. Design, Setting, and Participants: In this cohort study, clinical and transcriptomic data from clinical use of a prostate GC between 2016 and 2022 were linked with data aggregated from insurance claims, pharmacy records, and electronic health record (EHR) data. Participants were anonymously linked between datasets by deterministic methods through a deidentification engine using encrypted tokens. Algorithms were developed and refined for identifying prostate cancer diagnoses, treatment timing, and clinical outcomes using diagnosis codes, Common Procedural Terminology codes, pharmacy codes, Systematized Medical Nomenclature for Medicine clinical terms, and unstructured text in the EHR. Data analysis was performed from January 2023 to January 2024. Exposure: Diagnosis of prostate cancer. Main Outcomes and Measures: The primary outcomes were biochemical recurrence and development of prostate cancer metastases after diagnosis or radical prostatectomy (RP). The sensitivity of the linkage and identification algorithms for clinical and administrative data were calculated relative to clinical and pathological information obtained during the GC testing process as the reference standard. Results: A total of 92 976 of 95 578 (97.2%) participants who underwent prostate GC testing were successfully linked to administrative and clinical data, including 53 871 who underwent biopsy testing and 39 105 who underwent RP testing. The median (IQR) age at GC testing was 66.4 (61.0-71.0) years. The sensitivity of the EHR linkage data for prostate cancer diagnoses was 85.0% (95% CI, 84.7%-85.2%), including 80.8% (95% CI, 80.4%-81.1%) for biopsy-tested participants and 90.8% (95% CI, 90.5%-91.0%) for RP-tested participants. Year of treatment was concordant in 97.9% (95% CI, 97.7%-98.1%) of those undergoing GC testing at RP, and 86.0% (95% CI, 85.6%-86.4%) among participants undergoing biopsy testing. The sensitivity of the linkage was 48.6% (95% CI, 48.1%-49.1%) for identifying RP and 50.1% (95% CI, 49.7%-50.5%) for identifying prostate biopsy. Conclusions and Relevance: This study established a national-scale linkage of transcriptomic and longitudinal clinical data yielding high accuracy for identifying key clinical junctures, including diagnosis, treatment, and early cancer outcome. This resource can be leveraged to enhance understandings of disease biology, patterns of care, and treatment effectiveness.

Characteristics of Korean medical care utilization in patients with traffic injury: Analysis of 3 hospital electronic health record databases.

This retrospective study aimed to identify the characteristics of Korean medical care utilization in patients with traffic injury (TI) and to explore the clinical effectiveness of Korean medical interventions for TI through a multicenter chart review. This multicenter, retrospective registry study gathered electronic health records from 3 hospitals between January 1, 2018 and December 31, 2021. Data included treatment dates, demographic information, the Korean Standard Classification of Diseases codes, collision data, Korean medicine treatment modalities, and treatment outcomes. In total, 384 patients (182 inpatients and 202 outpatients) were included in the analysis. Patients were categorized into acute (207 patients, 53.9%), subacute (77 patients, 20.1%), and chronic (100 patients, 26.0%) phases based on the period until the visit. The most frequent Korean Standard Classification of Diseases code was "sprain and strain of cervical spine (S13.4)." All patients, except one, received Korean physiotherapy, followed by acupuncture and cupping. Comparative intragroup analysis revealed significant pain reduction in patients treated with the combination of Chuna manual therapy, herbal medicine, and pharmacopuncture and those treated with pharmacopuncture and herbal medicine only. This study highlights the characteristics of patients with TI visiting medical institutions providing Korean medicine and describes the effectiveness of Korean medicine interventions. Further comprehensive analysis with more data is necessary for future research.

Doubly robust estimation and sensitivity analysis for marginal structural quantile models.

The marginal structure quantile model (MSQM) provides a unique lens to understand the causal effect of a time-varying treatment on the full distribution of potential outcomes. Under the semiparametric framework, we derive the efficiency influence function for the MSQM, from which a new doubly robust estimator is proposed for point estimation and inference. We show that the doubly robust estimator is consistent if either of the models associated with treatment assignment or the potential outcome distributions is correctly specified, and is semiparametric efficient if both models are correct. To implement the doubly robust MSQM estimator, we propose to solve a smoothed estimating equation to facilitate efficient computation of the point and variance estimates. In addition, we develop a confounding function approach to investigate the sensitivity of several MSQM estimators when the sequential ignorability assumption is violated. Extensive simulations are conducted to examine the finite-sample performance characteristics of the proposed methods. We apply the proposed methods to the Yale New Haven Health System Electronic Health Record data to study the effect of antihypertensive medications to patients with severe hypertension and assess the robustness of the findings to unmeasured baseline and time-varying confounding.

Characteristics of a transgender and gender-diverse patient population in Utah: Use of electronic health records to advance clinical and health equity research.

Transgender and gender-diverse (TGD) people, individuals whose gender identity differs from their sex assigned at birth, face unique challenges in accessing gender-affirming care and often experience disparities in a variety of health outcomes. Clinical research on TGD health is limited by a lack of standardization on how to best identify these individuals. The objective of this retrospective cohort analysis was to accurately identify and describe TGD adults and their use of gender-affirming care from 2003-2023 in a healthcare system in Utah, United States. International Classification of Disease (ICD)-9 and 10 codes and surgical procedure codes, along with sexual orientation and gender identity data were used to develop a dataset of 4,587 TGD adults. During this time frame, 2,985 adults received gender-affirming hormone therapy (GAHT) and/or gender-affirming surgery (GAS) within one healthcare system. There was no significant difference in race or ethnicity between TGD adults who received GAHT and/or GAS compared to TGD adults who did not receive such care. TGD adults who received GAHT and/or GAS were more likely to have commercial insurance coverage, and adults from rural communities were underrepresented. Patients seeking estradiol-based GAHT tended to be older than those seeking testosterone-based GAHT. The first GAS occurred in 2013, and uptake of GAS have doubled since 2018. This study provides a methodology to identify and examine TGD patients in other health systems and offers insights into emerging trends and access to gender-affirming care.

Data-Driven Exploration of National Health Service Talking Therapies Care Pathways Using Process Mining: Retrospective Cohort Study.

BACKGROUND: The National Health Service (NHS) Talking Therapies program treats people with common mental health problems in England according to "stepped care," in which lower-intensity interventions are offered in the first instance, where clinically appropriate. Limited resources and pressure to achieve service standards mean that program providers are exploring all opportunities to evaluate and improve the flow of patients through their service. Existing research has found variation in clinical performance and stepped care implementation across sites and has identified associations between service delivery and patient outcomes. Process mining offers a data-driven approach to analyzing and evaluating health care processes and systems, enabling comparison of presumed models of service delivery and their actual implementation in practice. The value and utility of applying process mining to NHS Talking Therapies data for the analysis of care pathways have not been studied. OBJECTIVE: A better understanding of systems of service delivery will support improvements and planned program expansion. Therefore, this study aims to demonstrate the value and utility of applying process mining to NHS Talking Therapies care pathways using electronic health records. METHODS: Routine collection of a wide variety of data regarding activity and patient outcomes underpins the Talking Therapies program. In our study, anonymized individual patient referral records from two sites over a 2-year period were analyzed using process mining to visualize the care pathway process by mapping the care pathway and identifying common pathway routes. RESULTS: Process mining enabled the identification and visualization of patient flows directly from routinely collected data. These visualizations illustrated waiting periods and identified potential bottlenecks, such as the wait for higher-intensity cognitive behavioral therapy (CBT) at site 1. Furthermore, we observed that patients discharged from treatment waiting lists appeared to experience longer wait durations than those who started treatment. Process mining allowed analysis of treatment pathways, showing that patients commonly experienced treatment routes that involved either low- or high-intensity interventions alone. Of the most common routes, >5 times as many patients experienced direct access to high-intensity treatment rather than stepped care. Overall, 3.32% (site 1: 1507/45,401) and 4.19% (site 2: 527/12,590) of all patients experienced stepped care. CONCLUSIONS: Our findings demonstrate how process mining can be applied to Talking Therapies care pathways to evaluate pathway performance, explore relationships among performance issues, and highlight systemic issues, such as stepped care being relatively uncommon within a stepped care system. Integration of process mining capability into routine monitoring will enable NHS Talking Therapies service stakeholders to explore such issues from a process perspective. These insights will provide value to services by identifying areas for service improvement, providing evidence for capacity planning decisions, and facilitating better quality analysis into how health systems can affect patient outcomes.

Healthcare utilization patterns of individuals with depression after national policy to increase the mental health workforce in primary care: a data linkage study.

BACKGROUND: The deployment of the mental health nurse, an additional healthcare provider for individuals in need of mental healthcare in Dutch general practices, was expected to substitute treatments from general practitioners and providers in basic and specialized mental healthcare (psychologists, psychotherapists, psychiatrists, etc.). The goal of this study was to investigate the extent to which the degree of mental health nurse deployment in general practices is associated with healthcare utilization patterns of individuals with depression. METHODS: We combined national health insurers' claims data with electronic health records from general practices. Healthcare utilization patterns of individuals with depression between 2014 and 2019 (N = 31,873) were analysed. The changes in the proportion of individuals treated after depression onset were assessed in association with the degree of mental health nurse deployment in general practices. RESULTS: The proportion of individuals with depression treated by the GP, in basic and specialized mental healthcare was lower in individuals in practices with high mental health nurse deployment. While the association between mental health nurse deployment and consultation in basic mental healthcare was smaller for individuals who depleted their deductibles, the association was still significant. Treatment volume of general practitioners was also lower in practices with higher levels of mental health nurse deployment. CONCLUSION: Individuals receiving care at a general practice with a higher degree of mental health nurse deployment have lower odds of being treated by mental healthcare providers in other healthcare settings. More research is needed to evaluate to what extent substitution of care from specialized mental healthcare towards general practices might be associated with waiting times for specialized mental healthcare.

Predicting patients with dementia most at risk of needing psychiatric in-patient or enhanced community care using routinely collected clinical data: retrospective multi-site cohort study.

BACKGROUND: Dementia is a common and progressive condition whose prevalence is growing worldwide. It is challenging for healthcare systems to provide continuity in clinical services for all patients from diagnosis to death. AIMS: To test whether individuals who are most likely to need enhanced care later in the disease course can be identified at the point of diagnosis, thus allowing the targeted intervention. METHOD: We used clinical information collected routinely in de-identified electronic patient records from two UK National Health Service (NHS) trusts to identify at diagnosis which individuals were at increased risk of needing enhanced care (psychiatric in-patient or intensive (crisis) community care). RESULTS: We examined the records of a total of 25 326 patients with dementia. A minority (16% in the Cambridgeshire trust and 2.4% in the London trust) needed enhanced care. Patients who needed enhanced care differed from those who did not in age, cognitive test scores and Health of the Nation Outcome Scale scores. Logistic regression discriminated risk, with an area under the receiver operating characteristic curve (AUROC) of up to 0.78 after 1 year and 0.74 after 4 years. We were able to confirm the validity of the approach in two trusts that differed widely in the populations they serve. CONCLUSIONS: It is possible to identify, at the time of diagnosis of dementia, individuals most likely to need enhanced care later in the disease course. This permits the development of targeted clinical interventions for this high-risk group.

Development, evaluation and validation of machine learning models to predict hospitalizations of patients with coronary artery disease within the next 12 months.

BACKGROUND: Improved survival of patients after acute coronary syndromes, population growth, and overall life expectancy rise have led to a significant increase in the proportion of patients with stable coronary artery disease (CAD), creating a significant load on the entire healthcare system. The disease often progresses with the development of many complications while significantly increasing the likelihood of hospitalization. Developing and applying a machine learning model for predicting hospitalizations of patients with CAD to an inpatient medical facility will allow for close monitoring of high-risk patients, early preventive interventions, and optimized medical care. AIMS: Development and external validation of personalized models for predicting the preventable hospitalizations of patients with stable CAD and its complications using ML algorithms and data of real-world clinical practice. METHODS: 135,873 depersonalized electronic health records of 49,103 patients with stable CAD were included in the study. Anthropometric measurements, physical examination results, laboratory, instrumental, anamnestic, and socio-demographic data, widely used in routine medical practice, were considered as potential predictors, a total of 73 features. Logistic regression, decision tree-based methods including gradient boosting (AdaBoost, LightGBM, XGBoost, CatBoost) and bagging (RandomForest and ExtraTrees), discriminant analysis (LinearDiscriminant, QuadraticDiscriminant), and naive Bayes classifier were compared. External validation was performed on the data of a separate region. RESULTS: The best results and stability to external validation data were shown by the CatBoost model with an AUC of 0.875 (95% CI 0.865-0.885) for the internal testing and 0.872 (95% CI 0.856-0.886) for the external validation. The best model showed good performance evaluated through AUROC, Brier score and standardized net benefit (for the target NPV threshold) for the validation dataset that was only slightly similar to the train data. CONCLUSION: The metrics of the best model were superior to previously published studies. The results of external validation demonstrated the relative stability of the model to new data from another region that confirms the possibility of the model's application in real clinical practice.

Comparative performance of intensive care mortality prediction models based on manually curated versus automatically extracted electronic health record data.

INTRODUCTION: Benchmarking intensive care units for audit and feedback is frequently based on comparing actual mortality versus predicted mortality. Traditionally, mortality prediction models rely on a limited number of input variables and significant manual data entry and curation. Using automatically extracted electronic health record data may be a promising alternative. However, adequate data on comparative performance between these approaches is currently lacking. METHODS: The AmsterdamUMCdb intensive care database was used to construct a baseline APACHE IV in-hospital mortality model based on data typically available through manual data curation. Subsequently, new in-hospital mortality models were systematically developed and evaluated. New models differed with respect to the extent of automatic variable extraction, classification method, recalibration usage and the size of collection window. RESULTS: A total of 13 models were developed based on data from 5,077 admissions divided into a train (80%) and test (20%) cohort. Adding variables or extending collection windows only marginally improved discrimination and calibration. An XGBoost model using only automatically extracted variables, and therefore no acute or chronic diagnoses, was the best performing automated model with an AUC of 0.89 and a Brier score of 0.10. DISCUSSION: Performance of intensive care mortality prediction models based on manually curated versus automatically extracted electronic health record data is similar. Importantly, our results suggest that variables typically requiring manual curation, such as diagnosis at admission and comorbidities, may not be necessary for accurate mortality prediction. These proof-of-concept results require replication using multi-centre data.