Picture: Bryan Goodchild
Synthetic intelligence (AI) is already altering the way in which health care is offered, with enhanced diagnostic accuracy of pictures, predictive analyses of affected person outcomes from massive knowledge units that may direct therapy plans and analyzing particular person affected person knowledge to tailor interventions to private wants.
UMass Chan Medical College researcher Feifan Liu, PhD, affiliate professor of inhabitants & quantitative health sciences, is a part of a nationwide effort to spearhead one other essential utility of AI: to advance health equity.
In 2022, Dr. Liu was among the many first cohort of management fellows within the National Institutes of Health’s Artificial Intelligence/Machine Learning Consortium to Advance Health Equity and Research Diversity (AIM-AHEAD) program, a partnership to improve participation and illustration of researchers and communities underrepresented within the growth of AI and machine studying fashions, and to enhance the capabilities of this rising know-how to handle health disparities and inequities.
“Feifan’s work with AIM-AHEAD and being a part of this AIM-AHEAD construction is leading the way in which by way of considering, not simply what the worth is of AI, however how does this work within the health care system and society at massive?” mentioned Ben S. Gerber, MD, MPH, professor of inhabitants & quantitative health sciences, who works with Liu on a number of tasks. “What are the dangers? How will we deal with equity, bias and belief, and different moral problems with synthetic intelligence?”
Liu is principal investigator on two main analysis initiatives that grew out of the AIM-AHEAD fellowship. The primary, DETERMINE (Diabetes Prediction and Equity by way of Accountable Machine Studying), is a $1.4 million, two-year NIH AIM-AHEAD consortium growth grant in partnership with College of Illinois Chicago and Temple College to develop an AI-powered multivariable threat prediction mannequin to combine social, demographic and medical elements for correct, honest, generalizable and interpretable sort 2 diabetes prediction. Dr. Gerber is co-principal investigator of the examine, now in its second 12 months.
“The principle purpose is to construct a responsive AI mannequin predicting the danger of creating sort 2 diabetes and consider how nicely the mannequin generalizes throughout completely different establishments in addition to how equitably the mannequin performs throughout completely different demographic subgroups,” mentioned Liu. “We can even conduct simulation analyses and illustrate the potential impression on real-world medical apply, and bettering entry to preventive drugs or prevention applications, particularly for minority teams disproportionately affected by sort 2 diabetes.”
The guts of AI purposes is the algorithms on which machine studying fashions are primarily based.
Present medical pointers for sort 2 diabetes preventive measures depend on a simplified, imprecise prediabetes definition, which depends on restricted measures corresponding to glycemia and physique mass index, Liu defined. The researchers are integrating nonmedical socioeconomic knowledge involving neighborhood, surroundings and financial traits into the DETERMINE algorithm that they hope will extra precisely establish folks in danger and lead to extra equitable distribution of prevention and therapy sources.
The second examine, AI2Equity, is a $3 million, four-year grant funded by the Nationwide Coronary heart, Lung and Blood Institute in 2024. In partnership with OCHIN, a nationwide group health community, and Temple College, the multidisciplinary group of researchers goals to construct a deep studying mannequin incorporating social determinants of health, structured digital health information and medical notes to enhance prediction of heart problems. The challenge gives a strong basis for advancing equitable heart problems prevention, in accordance to Liu.
The mannequin might be in contrast with at the moment used cardiovascular threat prediction instruments.
“For each tasks, we’ll assess and enhance the generalizability and mannequin equity throughout completely different establishments and completely different settings,” Liu mentioned. “To mitigate bias, we’ll develop coaching algorithms that guarantee mannequin coaching excludes info carefully linked to delicate attributes corresponding to race or ethnicity. Research present that AI might unintentionally amply indicators from bias knowledge, exacerbating disparities for marginalized teams. Lastly, we would like to present that the mannequin has higher interpretability to higher assist medical resolution making.”
Liu and Gerber mentioned that the earlier and extra precisely the danger of creating diabetes or heart problems might be recognized, the higher the health consequence—stopping or delaying illness with way of life and medicine remedy.