Healthcare Needs AI, AI Needs Causality

There's much to be excited about with artificial intelligence (AI) in healthcare: Google AI is improving the workflow of clinicians with predictive models for diabetic retinopathy, many new approaches are achieving expert-level performance in tasks such as classification of skin cancer, and others surpassing the capabilities of doctors -- notably the recent report of DeepMind's AI for predicting acute kidney disease, capable of detecting potentially fatal kidney injuries 48 hours before symptoms are recognized by doctors.

Yet medical practitioners and researchers at the intersection of machine learning (ML) and medicine are quick to point out these successes are not representative of the more nuanced, non-trivial challenges presented by medical research and clinical applications. These ML success stories (notably all deep learning) are disease prediction problems, learning patterns that map well-defined inputs to well-labeled outputs.

The full Forbes article can be viewed at this link.  

Vital Signs: Pharmacy-Based Naloxone Dispensing — United States, 2012–2018

The CDC Guideline for Prescribing Opioids for Chronic Pain recommends considering prescribing naloxone when factors that increase risk for overdose are present (e.g., history of overdose or substance use disorder, opioid dosages ≥50 morphine milligram equivalents per day [high-dose], and concurrent use of benzodiazepines). In light of the high numbers of drug overdose deaths involving opioids, 36% of which in 2017 involved prescription opioids, improving access to naloxone is a public health priority. CDC examined trends and characteristics of naloxone dispensing from retail pharmacies at the national and county levels in the United States.

CDC analyzed 2012–2018 retail pharmacy data from IQVIA, a health care, data science, and technology company, to assess U.S. naloxone dispensing by U.S. Census region, urban/rural status, prescriber specialty, and recipient characteristics, including age group, sex, out-of-pocket costs, and method of payment. Factors associated with naloxone dispensing at the county level also were examined.

The number of naloxone prescriptions dispensed from retail pharmacies increased substantially from 2012 to 2018, including a 106% increase from 2017 to 2018 alone. Nationally, in 2018, one naloxone prescription was dispensed for every 69 high-dose opioid prescriptions. Substantial regional variation in naloxone dispensing was found, including a twenty-fivefold variation across counties, with lowest rates in the most rural counties. A wide variation was also noted by prescriber specialty. Compared with naloxone prescriptions paid for with Medicaid and commercial insurance, a larger percentage of prescriptions paid for with Medicare required out-of-pocket costs.

Despite substantial increases in naloxone dispensing, the rate of naloxone prescriptions dispensed per highdose opioid prescription remains low, and overall naloxone dispensing varies substantially across the country. Naloxone distribution is an important component of the public health response to the opioid overdose epidemic. Health care providers can prescribe or dispense naloxone when overdose risk factors are present and counsel patients on how to use it. Efforts to improve naloxone access and distribution work most effectively with efforts to improve opioid prescribing, implement other harm-reduction strategies, promote linkage to medications for opioid use disorder treatment, and enhance public health and public safety partnerships.

This report can be downloaded below.  

Patients’ willingness to share digital health and non-health data for research: a cross-sectional study

Patients generate large amounts of digital data through devices, social media applications, and other online activities. Little is known about patients’ perception of the data they generate online and its relatedness to health, their willingness to share data for research, and their preferences regarding data use.

Patients at an academic urban emergency department were asked if they would donate any of 19 different types of data to health researchers and were asked about their views on data types’ health relatedness. Factor analysis was used to identify the structure in patients’ perceptions of willingness to share different digital data, and their health relatedness.

Of 595 patients approached 206 agreed to participate, of whom 104 agreed to share at least one types of digital data immediately, and 78% agreed to donate at least one data type after death. EMR, wearable, and Google search histories (80%) had the highest percentage of reported health relatedness. 72% participants wanted to know the results of any analysis of their shared data, and half wanted their healthcare provider to know.

Patients in this study were willing to share a considerable amount of personal digital data with health researchers. They also recognize that digital data from many sources reveal information about their health. This study opens up a discussion around reconsidering US privacy protections for health information to reflect current opinions and to include their relatedness to health.

The full article can be downloaded below.  

Practical guidance on artificial intelligence for health-care data

Advances in machine learning and artificial intelligence (AI) offer the potential to provide personalised care that is equal to or better than the performance of humans for several health-care tasks. AI models are often powered by clinical data that are generated and managed via the medical system, for which the primary purpose of data collection is to support care, rather than facilitate subsequent analysis. Thus, the direct application of AI approaches to health care is associated with both challenges and opportunities.

The full article can be downloaded below.  

Mixed methods evaluation of implementation and outcomes in a community-based cancer prevention intervention

Community-based educational programs can complement clinical strategies to increase cancer screenings and encourage healthier lifestyles to reduce cancer burden. However, implementation quality can influence program outcomes and is rarely formally evaluated in community settings. This mixed-methods study aimed to characterize implementation of a community-based cancer prevention program using the Consolidated Framework for Implementation Research (CFIR), determine if implementation was related to participant outcomes, and identify barriers and facilitators to implementation that could be addressed.

This study utilized quantitative participant evaluation data (n = 115) and quantitative and qualitative data from semi-structured interviews with program instructors (N = 13). At the participant level, demographic data (age, sex, insurance status) and behavior change intention were captured. Instructor data included implementation of program components and program attendance to create a 7-point implementation score of fidelity and reach variables. Degree of program implementation (high and low) was operationalized based on these variables (low: 0–4, high: 5–7). Relationships among degree of implementation, participant demographics, and participant outcomes (e.g., intent to be physically active or limit alcohol) were assessed using linear or ordinal logistic mixed effects models as appropriate. Interview data were transcribed and coded deductively for CFIR constructs, and constructs were then rated for magnitude and valence. Patterns between ratings of high and low implementation programs were used to determine constructs that manifested as barriers or facilitators.

Program implementation varied with scores ranging from 4 to 7. High implementation was related to greater improvements in intention to be physically active (p < 0.05), achieve a healthy weight (p < 0.05), and limit alcohol (p < 0.01). Eight constructs distinguished between high and low implementation programs. Design quality and packaging, compatibility, external change agents, access to knowledge and information, and experience were facilitators of implementation and formally appointed internal implementation leaders was a barrier to implementation.

As higher implementation was related to improved participant outcomes, program administrators should emphasize the importance of fidelity in training for program instructors. The CFIR can be used to identify barriers and/or facilitators to implementation in community interventions, but results may be unique from clinical contexts.

The full article can be downloaded below.  

The Evolution of Elderly Telehealth and Health Informatics

Many elderly individuals experience memory loss and often dementia as they age. This causes problems for the elderly due to diminished skills and increase in medical problems and natural decline. The Veterans Health Administration (VHA) introduced a national home telehealth program, Care Coordination/Home Telehealth (CCHT). Its purpose was to coordinate the care of veteran patients with chronic conditions and avoid their unnecessary admission to long-term institutional care. Such programs are cost-effective. Long-term care insurance companies are likely to cover these services. Home care and nursing home corporations are following the VHA’s lead. We have recently witnessed significant advances in technology. Internet and mobile applications have opened a new world, providing information and opportunities for individuals to learn more information about illness and at a much faster rate. Smart home technology has evolved. Elderly patients often encounter difficulties using these technologies. Despite the advances in telehealth and telemedicine and the evolution of the technology, many individuals cannot afford the treatment or the technology. These same individuals and families are part of the digital divide, and they have not embraced the new technology. Federal programs have been developed and implemented to help this portion of the population.

The full chapter can be downloaded below.  

Characterising complex healthcare systems using network science: The small world of emergency surgery

Hospitals are complex systems and optimising their function is critical to the provision of high quality, cost effective healthcare. Nevertheless, metrics of performance have to date focused on the performance of individual elements rather than the system as a whole. Manipulation of individual elements of a complex system without an integrative understanding of its function is undesirable and may lead to counter-intuitive outcomes and a holistic metric of hospital function might help design more efficient services. We aimed to characterise the system of peri-operative care for emergency surgical admissions in our tertiary care hospital using network analysis. We used retrospective electronic health record data to construct a weighted directional network of the system. For this we selected all unplanned admissions during a 3.5 year period involving a surgical intervention during the inpatient stay and obtained a set of 16,500 individual inpatient episodes. We then constructed and analysed the structure of this network using established methods from network science such as degree distribution, betweenness centrality and small-world characteristics. The analysis showed the service to be a complex system with scale-free, small-world network properties. This finding has implications for the structure and resilience of the service as such networks, whilst being robust in general, may be vulnerable to outages at specific key nodes. We also identified such potential hubs and bottlenecks in the system based on a variety of network measures. It is hoped that such a holistic, system-wide description of a hospital service may provide better metrics for hospital strain and serve to help planners engineer systems that are as robust as possible to external shocks.

The full article can be downloaded below.

Developing Open‑Source Models for the US Health System: Practical Experiences and Challenges to Date with the Open‑Source Value Project

The Innovation and Value Initiative started the Open-Source Value Project with the aim to improve the credibility and relevance of model-based value assessment in the context of the US healthcare environment. As a core activity of the OpenSource Value Project, the Innovation and Value Initiative develops and provides access to flexible open-source economic models that are developed iteratively based on public feedback and input. In this article, we describe our experience to date with the development of two currently released, Open-Source Value Project models, one in rheumatoid arthritis and one in epidermal growth factor receptor-positive non-small-cell lung cancer. We developed both Open-Source Value Project models using the statistical programming language R instead of spreadsheet software (i.e., Excel), which allows the models to capture multiple model structures, model sequential treatment with individual patient simulations, and improve integration with formal evidence synthesis. By developing the models in R, we were also able to use version control systems to manage changes to the source code, which is needed for iterative and collaborative model development. Similarly, OpenSource Value Project models are freely available to the public to provide maximum transparency and facilitate collaboration. Development of the rheumatoid arthritis and non-small-cell lung cancer model platforms has presented multiple challenges. The development of multiple components of the model platform tailored to different audiences, including web interfaces, required more resources than a cost-effectiveness analysis for a publication would. Furthermore, we faced methodological hurdles, in particular related to the incorporation of multiple competing model structures and novel elements of value. The iterative development based on public feedback also posed some challenges during the review phase, where methodological experts did not always understand feedback from clinicians and vice versa. Response to the Open-Source Value Project by the modeling community and patient organizations has been positive, but feedback from US decision makers has been limited to date. As we progress with this project, we hope to learn more about the feasibility, benefits, and challenges of an open-source and collaborative approach to model development for value assessment.

The full article can be downloaded below.  

Willingness to Participate in Health Information Networks with Diverse Data Use: Evaluating Public Perspectives

Health information generated by health care encounters, research enterprises, and public health is increasingly interoperable and shareable across uses and users. This paper examines the US public’s willingness to be a part of multi-user health information networks and identifies factors associated with that willingness.

Using a probability-based sample (n = 890), we examined the univariable and multivariable relationships between willingness to participate in health information networks and demographic factors, trust, altruism, beliefs about the public’s ethical obligation to participate in research, privacy, medical deception, and policy and governance using linear regression modeling.

Willingness to be a part of a multi-user network that includes health care providers, mental health, social services, research, or quality improvement is low (26 percent–7.4 percent, depending on the user). Using stepwise regression, we identified a model that explained 42.6 percent of the variability in willingness to participate and included nine statistically significant factors associated with the outcome: Trust in the health system, confidence in policy, the belief that people have an obligation to participate in research, the belief that health researchers are accountable for conducting ethical research, the desire to give permission, education, concerns about insurance, privacy, and preference for notification.

Our results suggest willingness to be a part of multi-user data networks is low, but that attention to governance may increase willingness. Building trust to enable acceptance of multi-use data networks will require a commitment to aligning data access practices with the expectations of the people whose data is being used.

The full article can be downloaded below.

Helping patients help themselves: A systematic review of self-management support strategies in primary health care practice

This review highlights core components of successful interventions showing positive clinical and/or humanistic outcomes. Whilst it was difficult to directly correlate individual strategies to outcomes and effectiveness, there was a clear distinction of strategies across the conditions studied. This review provides encouraging groundwork for the design and evaluation of practical strategies for evidence-based practice and the construction of self-management support processes in primary healthcare practice. This review may assist in determining the breadth and focus of the support primary care professionals provide. Application of a theoretical perspective provides a strong base for the development of SMS interventions. The developed model sets the foundation for the design and evaluation of practical strategies for the construct of self-management support in primary healthcare practice. These results may be used to justify additional research investigating self-management interventions delivered in the primary care setting. In response, primary care providers can begin to deeply reflect on current practice and become involved in a dialogue to improve self-management support. Critically, these results should stimulate informed discussion for the future delivery of self-management support in primary care and the requirements for upskilling healthcare providers to effectively support patients in this collaborative process.

The full article can be downloaded below.