TY - JOUR
T1 - Using unsupervised machine learning to classify behavioral risk markers of bacterial vaginosis
AU - Rodriguez, Violeta J
AU - Pan, Yue
AU - Salazar, Ana S
AU - Nogueira, Nicholas Fonseca
AU - Raccamarich, Patricia
AU - Klatt, Nichole R
AU - Jones, Deborah L
AU - Alcaide, Maria L
N1 - Publisher Copyright:
© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.
PY - 2024/3
Y1 - 2024/3
N2 - Introduction: This study used an unsupervised machine learning algorithm, sidClustering and random forests, to identify clusters of risk behaviors of Bacterial Vaginosis (BV), the most common cause of abnormal vaginal discharge linked to STI and HIV acquisition. Methods: Participants were 391 cisgender women in Miami, Florida, with a mean of 30.8 (SD = 7.81) years of age; 41.7% identified as Hispanic; 41.7% as Black and 44.8% as White. Participants completed measures of demographics, risk behaviors [sexual, medical, and reproductive history, substance use, and intravaginal practices (IVP)], and underwent collection of vaginal samples; 135 behavioral variables were analyzed. BV was diagnosed using Nugent criteria. Results: We identified four clusters, and variables were ranked by importance in distinguishing clusters: Cluster 1: nulliparous women who engaged in IVPs to clean themselves and please sexual partners, and used substances frequently [n = 118 (30.2%)]; Cluster 2: primiparous women who engaged in IVPs using vaginal douches to clean themselves (n = 112 (28.6%)]; Cluster 3: primiparous women who did not use IVPs or substances [n = 87 (22.3%)]; and Cluster 4: nulliparous women who did not use IVPs but used substances [n = 74 (18.9%)]. Clusters were related to BV (p < 0.001). Cluster 2, the cluster of women who used vaginal douches as IVPs, had the highest prevalence of BV (52.7%). Conclusions: Machine learning methods may be particularly useful in identifying specific clusters of high-risk behaviors, in developing interventions intended to reduce BV and IVP, and ultimately in reducing the risk of HIV infection among women.
AB - Introduction: This study used an unsupervised machine learning algorithm, sidClustering and random forests, to identify clusters of risk behaviors of Bacterial Vaginosis (BV), the most common cause of abnormal vaginal discharge linked to STI and HIV acquisition. Methods: Participants were 391 cisgender women in Miami, Florida, with a mean of 30.8 (SD = 7.81) years of age; 41.7% identified as Hispanic; 41.7% as Black and 44.8% as White. Participants completed measures of demographics, risk behaviors [sexual, medical, and reproductive history, substance use, and intravaginal practices (IVP)], and underwent collection of vaginal samples; 135 behavioral variables were analyzed. BV was diagnosed using Nugent criteria. Results: We identified four clusters, and variables were ranked by importance in distinguishing clusters: Cluster 1: nulliparous women who engaged in IVPs to clean themselves and please sexual partners, and used substances frequently [n = 118 (30.2%)]; Cluster 2: primiparous women who engaged in IVPs using vaginal douches to clean themselves (n = 112 (28.6%)]; Cluster 3: primiparous women who did not use IVPs or substances [n = 87 (22.3%)]; and Cluster 4: nulliparous women who did not use IVPs but used substances [n = 74 (18.9%)]. Clusters were related to BV (p < 0.001). Cluster 2, the cluster of women who used vaginal douches as IVPs, had the highest prevalence of BV (52.7%). Conclusions: Machine learning methods may be particularly useful in identifying specific clusters of high-risk behaviors, in developing interventions intended to reduce BV and IVP, and ultimately in reducing the risk of HIV infection among women.
KW - Female
KW - Humans
KW - Vaginosis, Bacterial/diagnosis
KW - HIV Infections/diagnosis
KW - Unsupervised Machine Learning
KW - Vagina/microbiology
KW - Sexual Behavior
KW - Women
KW - Bacterial vaginosis
KW - Unsupervised machine learning
KW - Sexual behavior
UR - http://www.scopus.com/inward/record.url?scp=85184197100&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85184197100&partnerID=8YFLogxK
U2 - 10.1007/s00404-023-07360-7
DO - 10.1007/s00404-023-07360-7
M3 - Article
C2 - 38310145
SN - 0932-0067
VL - 309
SP - 1053
EP - 1063
JO - Archives of Gynecology and Obstetrics
JF - Archives of Gynecology and Obstetrics
IS - 3
ER -