Artificial intelligence (AI) may offer insights into women’s health by uncovering patterns in data that have historically eluded clinical observation and pathophysiologic understanding, according to an editorial published in JAMA Internal Medicine by Eve Rittenberg, MD, of Brigham and Women’s Hospital and Harvard Medical School, Boston, and colleagues. Women spend 25% more of their lives in poor health than men, the researchers noted, and AI could help address this disparity.
“The reasons for the women’s health gap are myriad,” including biologic sex differences, gender effects and bias, and inadequate research, noted Dr. Rittenberg and colleagues. “AI has the potential to both bridge these gaps and surface new knowledge through an understanding of intergroup differences (sex-specific disease presentations and outcomes) and intragroup differences (heterogeneous phenotypes of gynecologic diseases among women), ultimately enabling personalized care.”
Intergroup Differences
According to the authors, AI may be used to identify sex-specific disease presentations and outcomes that contribute to disparities in care. Cardiovascular disease was highlighted as an example, as women have been found to be less likely than men to receive effective diagnosis and treatment.
Analyses using AI could help identify sex-specific biomarkers, refine risk stratification and prognostic schema, and inform therapeutic targets. As an example, the editorial authors referenced an AI-enabled analysis of electrocardiograms that was able to identify women at elevated risk of subsequently developing cardiovascular disease. AI-generated evidence-based and validated treatment algorithms, if trained on inclusive and appropriate data sets, could reduce gender bias by adding objectivity to clinical decision-making.
Intragroup Differences
Dr. Rittenberg and colleagues also described the potential of AI to improve understanding of intragroup differences in conditions that primarily affect women and are often dismissed or misdiagnosed, including uterine fibroids, polycystic ovary syndrome, and menopausal symptoms. Endometriosis was cited to illustrate the diagnostic challenges common in women’s health, with definitive diagnosis estimated to take an average of 7 to 10 years.
AI-enabled analysis of multisource data—including natural language processing of patient reports and clinical notes, and computer vision–enabled imaging analysis, coupled with novel developments in biomarkers and -omics data sets—could identify patterns clinicians may miss and shorten the diagnostic process, the researchers wrote.
Future Directions
The potential of AI applications in women’s health depends on prioritizing fairness, accountability, transparency, and equity throughout their development and deployment, noted the authors. Algorithms trained on nonrepresentative data may reinforce rather than resolve disparities, and privacy concerns exist, particularly as wearable devices or menstrual trackers could be used to monitor reproductive health decisions.
“Newer machine learning approaches use bias mitigation techniques to undo the historical biases of the past, and novel differential privacy frameworks can ensure an individual’s data are at minimal risk. In fact, the historic underrepresentation of women in health data presents a unique opportunity to make women’s health the demonstrative case for building, testing, and validating AI systems that center fairness and validate effective bias mitigation and privacy strategies," wrote the editorial authors.
“Researchers must include sex- and gender-specific variables in AI model development,” they wrote, adding that clinicians should engage patients and the public to support ethical implementation of AI tools that meet the needs of all parties. “In this way, AI can contribute to a better understanding of women’s health, more personalized and effective health care, and better health outcomes for women to the benefit of all.”
Disclosure: For full disclosures of the editorial authors visit jamanetwork.com.
Source: JAMA Internal Medicine
