Exploring Prediabetes Pathways Using Explainable AI on Data from Electronic Medical Records.

This study leverages data from a Canadian database of primary care Electronic Medical Records to develop machine learning models predicting type 2 diabetes mellitus (T2D), prediabetes, or normoglycemia. These models are used as a basis for extracting counterfactual explanations and derive personalized changes in biomarkers to prevent T2D onset, particularly in the still reversible prediabetic state. The models achieve satisfactory performance. Furthermore, feature importance analysis underscores the significance of fasting blood sugar and glycated hemoglobin, while counterfactuals explanations emphasize the centrality of keeping body mass index and cholesterol indicators within or close to the clinically desirable ranges. This research highlights the potential of machine learning and counterfactual explanations in guiding preventive interventions that may help slow down the progression from prediabetes to T2D on an individual basis, eventually fostering a recovery from prediabetes to a normoglycemic state.

Measuring and Managing Healthcare Supply and Demand in Real-Time.

Measuring the supply and demand for access to and wait-times for healthcare is key to managing healthcare services and allocating resources appropriately. Yet, few jurisdictions in distributed, socialized medicine settings have any way to do so. In this paper, we propose the requirements for a jurisdictional patient scheduling system that can measure key metrics, such as supply of and demand for regulated health care professional care, access to and wait times for care, real-time health system utilization and provide the data to compute patient journeys. The system is also capable of tracking new supply of providers and who does not have access to a primary care provider. Benefits, limitations and risks of the model are discussed.

A Framework for Implementing Disease Prevention and Behavior Change Evidence at Scale.

The current corpus of evidence-based information for chronic disease prevention and treatment is vast and growing rapidly. Behavior change theories are increasingly more powerful but difficult to operationalize in the current healthcare system. Millions of Canadians are unable to access personalized preventive and behavior change care because our in-person model of care is running at full capacity and is not set up for mass education and behavior change programs. We propose a framework to utilize data from electronic medical records to identify patients at risk of developing chronic disease and reach out to them using digital health tools that are overseen by the primary care team. The framework leverages emerging technologies such as artificial intelligence, digital health tools, and patient-generated data to deliver evidence-based knowledge and behavior change to patients across Canada at scale. The framework is flexible to enable new technologies to be added without overwhelming providers, patients or implementers.

Towards a Regulatory Framework for Workflow Improvement in Electronic Medical Records.

Physicians have to complete several time-consuming and burnout-inducing tasks in their EMRs for everyday care of patients. Poor workflow design generates increased effort for physicians. In this study, we measure time doctors take to retrieve and review information in the patient chart at the beginning of a visit; one of approximately 12 tasks a doctor must do in the EMR during the visit. Information retrieval takes approximately 40 minutes per day. Automation could save 75% of that time. We estimate that if every family doctor in Canada could save 30 minutes through automation of just this one process, we could free up time equivalent to >3000 physicians and >5 million patients; enough to absorb the vast majority of patients who currently do not have a doctor. We know of no more powerful intervention than workflow automation in Canadian EMRs to increase the supply of doctors while simultaneously reducing a major cause of burnout. We recommend an accelerated research program to identify additional opportunities for workflow automation and a regulatory program to ensure that every physician has access to workflow automation in their EMR.

Lack of Data Access, but Not Availability, Hinders AI Training for High-Risk Conditions in Ontario.

Advanced disease prediction is an important step toward achieving a proactive healthcare system. New technologies such as artificial intelligence are very promising in their ability to predict the onset of future disease much earlier than has been possible in the past. However, artificial intelligence requires training and training requires data. In this study, we report on the ready availability, but lack of accessibility and real-time access to healthcare data required to treat five high-cost diseases that are predictable using AI and preventable using well-established evidence-based therapies. There is urgent need for action on the part of governments and other interest holders to define and invest in the infrastructure required to make data for training and deploying AI at scale more accessible.

Towards a Regulatory Framework for Electronic Medical Record Interoperability in Canada.

All complex systems are potentially predisposed to failure. Healthcare systems are complex systems that are prone to many errors that can result in dire consequences for patients and healthcare providers. The healthcare system in Canada is under unprecedented strain due to shortages of healthcare providers, provider burnout, inefficient workflows, and a lack of appropriate digital infrastructure. We used failure mode and effects analysis (FMEA) to identify the failure modes for care provided in primary care settings. We identified failure modes in appointment scheduling, patient-provider communications, referrals, laboratory and diagnostic procedures, and medication prescriptions as the main failure modes. To mitigate the detected risks, we recommend solutions to 'close the loop' on failure modes to prevent patients from falling through the cracks, as vulnerable patients who cannot advocate for themselves are most likely to do so. We provide preliminary requirements for a regulatory regime for electronic health records that can reduce provider burnout, improve regulatory compliance, and improve system efficiency, all while improving patient safety, experience, and outcomes.

Towards a Regulatory Framework for Electronic Medical Record Data Visualization.

Physicians struggle to retrieve data from electronic medical records. We evaluated a digital tool that enhances physician efficiency in retrieving and analyzing patient information for treatment decision-making. Our use case is the care of diabetic patients. Evaluation results showed that healthcare providers who used the i4C (Insights for Care) dashboard experienced greater time efficiency than those who used traditional EMR information retrieval methods. A comprehensive evaluation of the i4C Dashboard confirms its effectiveness in facilitating diabetic care data management, as well as its potential application to a wide range of healthcare scenarios. In order to further maximize its effectiveness on clinical efficiency and patient care, future research should focus on improving its usability and scalability.

Towards a Unified Framework for Information and Interoperability Governance.

Challenges in health data interoperability have highlighted overall health care system inefficiencies. Many organizations struggle to establish a robust data governance infrastructure to meet the increasing demands of advanced data uses, let alone sharing it with a large number of other organizations. There is a need for health care organizations to adopt information governance frameworks that encapsulates interoperability as a core attribute as this can improve data processing, knowledge translation and participation in the larger health data ecosystem. To establish interoperability between healthcare organizations, standards must exist in relation to how information is governed and circulates in the healthcare system, not just on how it is structured, stored and used within an organization. In this paper we demonstrate that interoperability between organizations cannot coherently exist without consideration of information governance within organizations. Lack of coherence can lead to lack of data accessibility, decreased organizational efficiencies, and poor data quality. With this in mind, we propose a unified framework that integrates the principles of both information and interoperability governance to increase the adaptability, flexibility, and efficiency of health information usage across the entire healthcare system.

AI Can Improve the Economics of Blindness Prevention in Canada.

Diabetic retinopathy is a leading cause of vision loss in Canada and creates significant economic and social burden on patients. Diabetic retinopathy is largely a preventable complication of diabetes mellitus. Yet, hundreds of thousands of Canadians continue to be at risk and thousands go on to develop vision loss and disability. Blindness has a significant impact on the Canadian economy, on families and the quality of life of affected individuals. This paper provides an economic analysis on two potential interventions for preventing blindness and concludes that use of AI to identify high-risk individuals could significantly decrease the costs of identifying, recalling, and screening patients at risk of vision loss, while achieving similar results as a full-fledged screening and recall program. We propose that minimal data interoperability between optometrists and family physicians combined with artificial intelligence to identify and screen those at highest risk of vision loss can lower the costs and increase the feasibility of screening and treating large numbers of patients at risk of going blind in Canada.

Validation of a Design Architecture to Deliver Health Management and Behavior Change Evidence at Scale.

Forty-four percent of Canadians over the age of 20 have a non-communicable disease (NCD). Millions of Canadians are at risk of developing the complications of NCDs; millions have already experienced those complications. Fortunately, the evidence base for NCD prevention and behavior change is large and growing and digital technologies can deliver them at scale and with high fidelity. However, the current model of in-person primary care is not designed nor capable of operationalizing that evidence. New developments in artificial intelligence that can predict who will develop NCD or the complications of NCD are increasingly available, making the challenge of delivering disease prevention even more urgent. This paper presents findings from stakeholder engagement on a design architecture to address three initial barriers to large-scale deployment of health management and behavior change evidence: 1) the challenges of regulating mobile health apps, 2) the challenge of creating a value-based rationale for payers to invest in deploying mobile health apps at scale, and 3) the high cost of customer acquisition for delivering mobile health apps to those at risk.

Association of a Heart Failure Management Incentive in Primary Care With Clinical Outcomes: A Retrospective Cohort Study.

BACKGROUND: We aim to examine the association between primary care physicians' billing of Q050A, a pay-for-performance heart failure (HF) management incentive fee code, and the composite outcome of mortality, hospitalization, and emergency department visits. METHODS AND RESULTS: This population-based cohort study linked administrative health databases in Ontario, Canada, for patients with HF aged >66 years between January 1, 2008, and March 31, 2020. Cases were patients with HF who had a Q050A fee code billed. Cases and controls were matched 1:1 on age, sex, patient status on being rostered to a primary care physician, cardiologist, or internist visit in the 6 months before study enrollment, Johns Hopkins Adjusted Clinical Group resource use bands, days between HF diagnosis and study enrollment (±2 years), and the logit of the propensity score. A Cox proportional hazards model assessed the association of Q050A with the outcome. A total of 59 664 cases had a Q050A billed, whereas 244 883 patients did not. Before matching, patients who had a Q050A billed were more likely to be men (52% versus 49%), were rostered to a primary care physician (100% versus 96%), had a higher Charlson Comorbidity Index, and had higher health care costs. The mean follow-up was 481 days for cases and 530 days for controls. The composite outcome (hazard ratio, 1.11 [95% CI, 1.09-1.12]) was significantly higher for cases than controls. CONCLUSIONS: The Q050A incentive improved financial compensation for primary care physicians managing patients with HF but was not associated with improvements in the outcome. Research on promoting evidence-based HF management is warranted.

Multi-Input Multi-Output Dynamic Modelling of Type 2 Diabetes Progression.

Type 2 Diabetes Mellitus (T2D) is a chronic health condition that affects millions of people globally. Early identification of risk can support preventive intervention and therefore slow down disease progression. Risk characterization is also necessary to monitor the mechanisms behind the pathology through the analysis of the interrelationships between the predictors and their time course. In this work, a multi-input multi-output Gaussian Process model is proposed to describe the evolution of different biomarkers in patients who will/will not develop T2D considering the interdependencies between outputs. The preliminary results obtained suggest that the trends in biomarkers captured by the model are coherent with the literature and with real-world data, demonstrating the value of multi-input multi-output approaches. In future developments, the proposed method could be applied to assess how the biomarkers evolve and interact with each other in groups of patients having in common one or more risk factors.

Epidemiology and treatment of heart failure with chronic obstructive pulmonary disease in Canadian primary care.

AIMS: Heart failure (HF) and chronic obstructive pulmonary disease (COPD) are largely managed in primary care, but their intersection in terms of disease burden, healthcare utilization, and treatment is ill-defined. METHODS AND RESULTS: We examined a retrospective cohort including all patients with HF or COPD in the Canadian Primary Care Sentinel Surveillance Network from 2010 to 2018. The population size in 2018 with HF, COPD, and HF with COPD was 15 778, 27 927, and 4768 patients, respectively. While disease incidence declined, age-sex-standardized prevalence per 100 population increased for HF alone from 2.33 to 3.63, COPD alone from 3.44 to 5.96, and COPD with HF from 12.70 to 15.67. Annual visit rates were high and stable around 8 for COPD alone but declined significantly over time for HF alone (9.3-8.1, P = 0.04) or for patients with both conditions (14.3-11.9, P = 0.006). For HF alone, cardiovascular visits were common (29.4%), while respiratory visits were infrequent (3.5%), with the majority of visits being non-cardiorespiratory. For COPD alone, respiratory and cardiovascular visits were common (16.4% and 11.3%) and the majority were again non-cardiorespiratory. For concurrent disease, 39.0% of visits were cardiorespiratory. The commonest non-cardiorespiratory visit reasons were non-specific symptoms or signs, endocrine, musculoskeletal, and mental health. In patients with HF with and without COPD, angiotensin-converting enzyme inhibitor/angiotensin receptor blocker/angiotensin receptor-neprilysin inhibitor use was similar, while mineralocorticoid receptor antagonist use was marginally higher with concurrent COPD. Beta-blocker use was initially lower with concurrent COPD compared with HF alone (69.3% vs. 74.0%), but this progressively declined by 2018 (74.5% vs. 73.5%). CONCLUSIONS: The prevalence of HF and COPD continues to rise. Although patients with either or both conditions are high utilizers of primary care, the majority of visits relate to non-cardiorespiratory comorbidities. Medical therapy for HF was similar and the initially lower beta-blocker utilization disappeared over time.

Artificial intelligence with temporal features outperforms machine learning in predicting diabetes.

Diabetes mellitus type 2 is increasingly being called a modern preventable pandemic, as even with excellent available treatments, the rate of complications of diabetes is rapidly increasing. Predicting diabetes and identifying it in its early stages could make it easier to prevent, allowing enough time to implement therapies before it gets out of control. Leveraging longitudinal electronic medical record (EMR) data with deep learning has great potential for diabetes prediction. This paper examines the predictive competency of deep learning models in contrast to state-of-the-art machine learning models to incorporate the time dimension of risk. The proposed research investigates a variety of deep learning models and features for predicting diabetes. Model performance was appraised and compared in relation to predominant features, risk factors, training data density and visit history. The framework was implemented on the longitudinal EMR records of over 19K patients extracted from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN). Empirical findings demonstrate that deep learning models consistently outperform other state-of-the-art competitors with prediction accuracy of above 91%, without overfitting. Fasting blood sugar, hemoglobin A1c and body mass index are the key predictors of future onset of diabetes. Overweight, middle aged patients and patients with hypertension are more vulnerable to developing diabetes, consistent with what is already known. Model performance improves as training data density or the visit history of a patient increases. This study confirms the ability of the LSTM deep learning model to incorporate the time dimension of risk in its predictive capabilities.

Characterization of Inclination Analysis for Predicting Onset of Heart Failure from Primary Care Electronic Medical Records.

The aim of this study is to characterize the performance of an inclination analysis for predicting the onset of heart failure (HF) from routinely collected clinical biomarkers extracted from primary care electronic medical records. A balanced dataset of 698 patients (with/without HF), including a minimum of five longitudinal measures of nine biomarkers (body mass index, diastolic and systolic blood pressure, fasting glucose, glycated hemoglobin, low-density and high-density lipoproteins, total cholesterol, and triglycerides) is used. The proposed algorithm achieves an accuracy of 0.89 (sensitivity of 0.89, specificity of 0.90) to predict the inclination of biomarkers (i.e., their trend towards a 'survival' or 'collapse' as defined by an inclination analysis) on a labeled, balanced dataset of 40 patients. Decision trees trained on the predicted inclination of biomarkers have significantly higher recall (0.69 vs. 0.53) and significantly higher negative predictive value (0.60 vs. 0.55) than those trained on the average values computed from the measures of biomarkers available before the onset of the disease, suggesting that an inclination analysis can help identify the onset of HF in the primary care patient population from routinely available clinical data. This exploratory study provides the basis for further investigations of inclination analyses to identify at-risk patients and generate preventive measures (i.e., personalized recommendations to reverse the trend of biomarkers towards collapse).

A novel method to derive personalized minimum viable recommendations for type 2 diabetes prevention based on counterfactual explanations.

Despite the growing availability of artificial intelligence models for predicting type 2 diabetes, there is still a lack of personalized approaches to quantify minimum viable changes in biomarkers that may help reduce the individual risk of developing disease. The aim of this article is to develop a new method, based on counterfactual explanations, to generate personalized recommendations to reduce the one-year risk of type 2 diabetes. Ten routinely collected biomarkers extracted from Electronic Medical Records of 2791 patients at low risk and 2791 patients at high risk of type 2 diabetes were analyzed. Two regions characterizing the two classes of patients were estimated using a Support Vector Data Description classifier. Counterfactual explanations (i.e., minimal changes in input features able to change the risk class) were generated for patients at high risk and evaluated using performance metrics (availability, validity, actionability, similarity, and discriminative power) and a qualitative survey administered to seven expert clinicians. Results showed that, on average, the requested minimum viable changes implied a significant reduction of fasting blood sugar, systolic blood pressure, and triglycerides and a significant increase of high-density lipoprotein in patients at risk of diabetes. A significant reduction in body mass index was also recommended in most of the patients at risk, except in females without hypertension. In general, greater changes were recommended in hypertensive patients compared to non-hypertensive ones. The experts were overall satisfied with the proposed approach although in some cases the proposed recommendations were deemed insufficient to reduce the risk in a clinically meaningful way. Future research will focus on a larger set of biomarkers and different comorbidities, also incorporating clinical guidelines whenever possible. Development of additional mathematical and clinical validation approaches will also be of paramount importance.

High cardiovascular disease risk-associated with the incidence of Type 2 diabetes among prediabetics.

BACKGROUND: Prediabetes is a risk factor for developing Type 2 diabetes mellitus (T2D). We report on the first cohort study of the association between high cardiovascular diseases (CVD) risk with the incidence of T2D in prediabetics. First, estimate the direct effect of developing T2D on patients with prediabetes who have high CVDs risk; and 2) assess the potential increased risk of developing T2D mediated by statins. METHODS: We conducted a population-based cohort study using a subset of data from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) from 2000 to 2015. Cox proportional hazards (PH) regressions were conducted to estimate our primary outcome, which is the time to T2D among patients with prediabetes. RESULTS: From the 4995 filtered prediabetic participants identified between 2000 and 2015, 2800 participants were diagnosed with high CVDs risk scores as measured by the Framingham risk score. 2195 participants were non-high CVDs risk controls. The covariate-adjusted hazard ratio (HR) of 1.24 [95% confidence interval (CI), 1.10-1.31] for T2D by CVDs risk among prediabetics was observed. The total effect of CVDs risk on developing T2D was decomposed to a natural direct effect of high CVDs risk HR= 1.18 [95% CI, 1.01-1.48] and an indirect effect through statin therapy of HR= 1.06 [95% CI, 0.97-1.30]. CONCLUSION: Patients with prediabetes and high CVDs risk had a 24% higher chance of developing T2D. The high CVDs risk effect was mediated by statin therapy. Regular monitoring and counselling of prediabetics using statins is likely warranted to prevent the incidence of T2D.