Best Paper Selection

Appendix A: Search queries constructed for PubMed to identify candidate papers for review.

((((“precision prevention”) OR (“precision public health”)) OR (“precision epidemiology”)) OR (“precision nutrition”)) AND ((informatics) OR (“information science”) OR (implementation) OR (predictive) OR (artificial intelligence))

Limits: Published between 1/1/2023 and 12/31/2023, English

Search performed on January 16, 2024

Appendix B: Content Summaries of Selected Best Papers for the 2024 IMIA Yearbook, Special Section on Digital Health for Precision in Prevention

Andrade AQ, Calabretto J-P, Pratt NL, Kalisch-Ellett LM, Le Blanc VT, Roughead ER

Precision public-health intervention for care coordination: a real-world study

British Journal of General Practice, 2023, 23;73(728):e220-e230

This paper from Australia sought to ascertain if digital interventions delivered to the clinical software used by general practitioners (GPs) were more effective at promoting primary care appointments during the COVID-19 pandemic, compared to interventions that were delivered by standard mail. The rationale for the study was rooted in concern that patients were hesitant or even avoided leaving their homes, especially when they were vulnerable, such as being elderly, having a chronic disease, or physical disability. The study was developed and implemented with cooperation from the Veteran's Medicines Advice and Therapeutics Education Services (MATES) program, taking advantage of the program's strong emphasis on theory-driven health behavior research. The intervention was developed after conducting an epidemiologic survey, followed by the creation and evaluation of educational material with user-stakeholders, and ultimately the delivery of the intervention by one of the two venues. The theoretical framework was based on a diffusion of innovation model. The intervention was delivered in April 2020, and tested using a non-randomized experimental study. Eligibility was determined from a health claims database, based on candidate subjects' age, conditions that could lead to complications from COVID-19. GPs were identified through a proprietary algorithm. A total of 77,911 veterans and 18,577 GPs were enrolled in the trial, with 51,052 and 26,859 in the digital and postal groups, respectively. Those in the digital group had significantly more GP visits over the observation period. In addition, a time-to-event analysis showed that they had significantly more such visits in the first three months following the intervention. This study showed that it was possible to mount a clinical informatics-based, theory-driven intervention to activate patient health behavior early in a pandemic or other global health crisis.

Hwang U, Kim SW, Jung D, Kim S, Lee H, Seo SW, Seong JK, Yoon S. Alzheimer's Disease Neuroimaging Initiative

Real-world prediction of preclinical Alzheimer's disease with a deep generative model

Artif Intell Med. 2023 Oct;144:102654

In this study, the authors developed a modification of an existing deep generative model (HexaGAN) to create a deep neural network capable of predicting if cognitively normal individuals are at risk of preclinical Alzheimer's disease through the detection of amyloid deposits. Data included a set of 538 tabular records drawn from the Alzheimer's Disease Neuroimaging Initiative (ADNI) that included demographic, genetic, and clinical variables. Also drawn from the ADNI database were 53 MRI scans; amyloid positivity was obtained from PET scans. The data were artificially corrupted to include missing values and labels and to introduce class imbalance. Using image and tabular data, the deep learners performed classification modestly well, with the author's modification of the HexaGAN model performing best when using imaging and tabular data (AUROC=0.86). As with the comparison methods, the authors' model did not perform as well when using only image or only tabular data. The results indicate that using these types of data in conjunction with a deep GAN can successfully identify cognitively normal people with amyloid dysfunction. As such, the methods described in this paper could be a key component to prevention of the advancement of Alzheimer's disease in those who appear to be cognitively intact. The methods could be further applied to other progressive diseases to identify at-risk individuals for screening and/or early intervention to slow disease progression.

Lee S, Kim S, Yoon DS, Park JS, Woo H, Lee D, Cho S-Y, Park C, Yoo YK, Lee K-B, Lee JH

Sample-to-answer platform for the clinical evaluation of COVID-19 using a deep learning-assisted smartphone-based assay

Nat Commun. 2023;14(1):2361

This paper from Korea describes the development and validation of deep learning trained models and a smartphone application of the models to evaluate images of lateral flow assays for the SARS-CoV-2 virus. Lateral flow assays (LFAs) are rapid, inexpensive, and convenient tests for viruses, including influenza and SARS-CoV-2. During the COVID-19 pandemic, LFAs enabled point-of-care testing as well as at home testing. Although LFAs proved to have reliable performance, many clinicians and patients struggled to interpret results. The paper details the development and optimization of a two-step convolutional neural network (CNN) model that accurately classifies LFA results for SARS-CoV-2 (e.g., positive, negative, invalid) based on images captured on a smartphone. The model was validated using a broad range of samples analyzed with myriad LFAs from different manufacturers, and testing included images of variable quality. The robust optimization and validation process resulted in extremely accurate interpretation of results by the model such that secondary interpretation by a human was not necessary. The article illustrates how artificial intelligence and informatics approaches could be used to create precision applications that could rapidly diagnose viral infection, increasing accuracy and preventing spread of disease.

Wang B, Liu F, Deveaux L, Ash A, Gerber B, Allison J, Herbert C, Poitier M, MacDonell K, L X, Stanton B

Predicting Adolescent Intervention Non-responsiveness for Precision HIV Prevention Using Machine Learning

AIDS and behavior. 2023;27(5):1392-402

This paper from Unites States and Bahamian authors describes the development of machine learning models to identify adolescents who are unlikely to respond to HIV prevention interventions. Despite multiple intervention programs in many nations designed to prevent HIV transmission among adolescents, incidence of HIV has remained stable over the past two decades. In this study, the authors used longitudinal data from a randomized controlled trial in the Bahamas that tested various prevention interventions. The goal was to predict non-response to the myriad interventions. The researchers tested different machine learning approaches, including support vector machines and decision trees, and used a typical 80/20 split for training vs. validation data. Random forest approaches performed better, and self-efficacy proved to be the most influential feature to predict non-response. The paper demonstrates how informatics techniques can be used to examine different prevention approaches to increase precision in recommending at-risk patients to an intervention but also how they can be used to find behavioral constructs associated with achievement of success from a given prevention intervention. This approach could be used across multiple conditions and populations to better optimize referral and construction of prevention programs in clinical and public health settings.

No conflict of interest has been declared by the author(s).

Publication History

Article published online:
08 April 2025