Table 2 Fallacy in Descriptive Epidemiology: Bringing Machine Learning to the Table DOI Open Access
Christoffer Dharma, Rui Fu, Michael Chaiton

et al.

International Journal of Environmental Research and Public Health, Journal Year: 2023, Volume and Issue: 20(13), P. 6194 - 6194

Published: June 21, 2023

There is a lack of rigorous methodological development for descriptive epidemiology, where the goal to describe and identify most important associations with an outcome given large set potential predictors. This has often led Table 2 fallacy, one presents coefficient estimates all covariates from single multivariable regression model, which are uninterpretable in analysis. We argue that machine learning (ML) solution this problem. illustrate power ML example analysis identifying predictors alcohol abuse among sexual minority youth. The framework we propose as follows: (1) Identify few methods analysis, (2) optimize parameters using whole data nested cross-validation approach, (3) rank variables variable importance scores, (4) present partial dependence plots (PDP) association between outcome, (5) strength interaction terms PDPs. discuss strengths weaknesses future directions research. R codes reproduce these analyses provided, invite other researchers use.

Language: Английский

Prevalence of probable substance use disorders among children in Ugandan health facilities DOI Creative Commons
Harriet Aber,

Juliet N. Babirye,

Ingunn Marie Stadskleiv Engebretsen

et al.

BMC Public Health, Journal Year: 2024, Volume and Issue: 24(1)

Published: Jan. 29, 2024

Abstract Background Globally, there is a concerning surge in the prevalence of substance use among adolescents and children, creating substantial public health problem. Despite magnitude this issue, accessing healthcare explicitly for remains challenging, even though many users frequently visit institutions other health-related issues. To address gap, proactive screening disorders has emerged as critical strategy identifying engaging patients at risk use. The purpose study was to investigate probable alcohol disorders, associated factors, children aged 6 17 years old attending facilities Mbale, Uganda. Methods We conducted facility cross-sectional study, involving 854 6–17 years. assessed using validated Car, Relax, Alone, Forget, Friends, Trouble (CRAFFT) tool. Univariable multivariable modified Poisson regression analyses were performed STATA 15 software. Results overall (AUD) (SUD) 27.8% (95% CI 1.24–1.31) while that AUD alone 25.3% 1.22–1.28). Peer (APR = 1.24, 95% 1.10–1.32), sibling 1.14, 1.06–1.23), catholic caregiver religion 1.07 1.01–1.13), income more than $128 0.90, 0.82–0.98), having no parental reprimand 1.05, 1.01–1.10) knowledge how decline an offer substances 1.06, 1.01–1.12) found be significantly with AUD/SUD. Conclusions Our findings suggest high SUD visiting conditions, along strong link between social factors. implication our system actively screen treat these conditions primary facilities.

Language: Английский

Citations

4
Community-engaged artificial intelligence research: A scoping review DOI Creative Commons
Tyler J. Loftus, Jeremy A. Balch, Kenneth L. Abbott

et al.

PLOS Digital Health, Journal Year: 2024, Volume and Issue: 3(8), P. e0000561 - e0000561

Published: Aug. 23, 2024

The degree to which artificial intelligence healthcare research is informed by data and stakeholders from community settings has not been previously described. As communities are the principal location of delivery, engaging them could represent an important opportunity improve scientific quality. This scoping review systematically maps what known unknown about community-engaged identifies opportunities optimize generalizability these applications through involvement throughout model development, validation, implementation. Embase, PubMed, MEDLINE databases were searched for articles describing or machine learning with in Model architecture performance, nature engagement, barriers facilitators engagement reported according PRISMA extension Scoping Reviews guidelines. Of approximately 10,880 applications, 21 (0.2%) described involvement. All derived settings, most commonly leveraging existing datasets sources that included subjects, often bolstered internet-based acquisition subject recruitment. Only one article inclusion designing application–a natural language processing detected cases likely child abuse 90% accuracy using harmonized electronic health record notes both hospital practice settings. primary barrier including community-derived was small sample sizes, may have affected 11 studies (53%), introducing substantial risk overfitting threatens generalizability. Community application implementation rare. delivery occurs primarily investigators should consider user-centered design, usability, clinical

Language: Английский

Citations

2
Medical Metaverse, Part 2: Artificial Intelligence Algorithms and Large Language Models in Psychiatry and Clinical Neurosciences DOI

Wilfredo López-Ojeda,

Robin A. Hurley

Journal of Neuropsychiatry, Journal Year: 2023, Volume and Issue: 35(4), P. 316 - 320

Published: Oct. 1, 2023

Language: Английский

Citations

4
Table 2 Fallacy in Descriptive Epidemiology: Bringing Machine Learning to the Table DOI Open Access
Christoffer Dharma, Rui Fu, Michael Chaiton

et al.

International Journal of Environmental Research and Public Health, Journal Year: 2023, Volume and Issue: 20(13), P. 6194 - 6194

Published: June 21, 2023

There is a lack of rigorous methodological development for descriptive epidemiology, where the goal to describe and identify most important associations with an outcome given large set potential predictors. This has often led Table 2 fallacy, one presents coefficient estimates all covariates from single multivariable regression model, which are uninterpretable in analysis. We argue that machine learning (ML) solution this problem. illustrate power ML example analysis identifying predictors alcohol abuse among sexual minority youth. The framework we propose as follows: (1) Identify few methods analysis, (2) optimize parameters using whole data nested cross-validation approach, (3) rank variables variable importance scores, (4) present partial dependence plots (PDP) association between outcome, (5) strength interaction terms PDPs. discuss strengths weaknesses future directions research. R codes reproduce these analyses provided, invite other researchers use.

Language: Английский

Citations

3