No abstract

Background Randomized trials show that medical male circumcision (MC) reduces high-risk human papillomavirus (HR-HPV) infection in men. We assessed the efficacy of MC to reduce HR-HPV in female partners. Methods HIV-negative men were randomized to immediate MC (intervention) or MC delayed for 24 months (control). HIV-uninfected female partners of married men (648 intervention and 597 control arm) were simultaneously enrolled and provided interview information and self-collected vaginal swabs at baseline, 12 and 24 months. Female HPV infection was a secondary trial end point. Vaginal swabs were evaluated for HR-HPV by Roche HPV Linear Array. An intention-to-treat analysis estimated prevalence risk and incident rate ratios (PRR and IRR) and 95% confidence intervals (95%CI) of HR-HPV by Poisson multiple regression. In women with pre-existing HR-HPV, we estimated the risk ratio (RR) of cleared infection (i.e., loss of detection). The trials were registered with ClinicalTrials.gov, NCT00425984 and NCT00124878.) Findings Female characteristics and HPV prevalence were similar between arms at enrollment. Two year retention rates were 84.7% (549/648) in intervention arm and 84.1% (502/597) in control arm spouses. Year 2 female HR-HPV prevalence was 27.8% (151/544) in the intervention and 38.7% (189/488) in the control arm (PRR=0.72, 95%CI 0.60–0.85, p=0.001). HR-HPV incidence was 20.7/100py in the intervention arm and 26.9/100py in the control arm wives (IRR=0.77, 95%CI 0.63-0.93, p=0.008). HR-HPV incidence was lower in intervention than control arm wives for 13 of 14 (92.9%) HR-HPV genotypes and in most demographic/behavioral subgroups. Genotype specific HR-HPV clearance was higher in the wives of men in the intervention arm (66.2%, 376/568) than the control arm (59.2%, 339/573, RR=1.12, 95%CI 1.02-1.22). Interpretation MC reduces the prevalence and incidence and increases clearance of HR-HPV infections in female partners. Funding Bill & Melinda Gates Foundation with additional laboratory and training support from National Institutes of Health and Fogarty International Center.

The BED capture enzyme immunoassay (BED-CEIA) was developed for estimating HIV incidence from cross-sectional data. This assay misclassifies some individuals with nonrecent HIV infection as recently infected, leading to overestimation of HIV incidence. We analyzed factors associated with misclassification by the BED-CEIA. We analyzed samples from 383 men who were diagnosed with HIV infection less than 1 year after a negative HIV test (Multicenter AIDS Cohort Study). Samples were collected 2-8 years after HIV seroconversion, which was defined as the midpoint between the last negative and first positive HIV test. Samples were analyzed using the BED-CEIA with a cutoff of OD-n ≤ 0.8 for recent infection. Logistic regression was used to identify factors associated with misclassification. Ninety-one (15.1%) of 603 samples were misclassified. In multivariate models, misclassification was independently associated with highly active antiretroviral treatment (HAART) for >2 years, HIV RNA <400 copies/ml, and CD4 cell count <50 or <200 cells/mm(3); adjusted odds ratios (OR) and 95% confidence intervals (CI) were 4.72 (1.35-16.5), 3.96 (1.53-10.3), 6.85 (2.71-17.4), and 11.5 (3.64-36.0), respectively. Among 220 men with paired samples, misclassification 2-4 years after seroconversion was significantly associated with misclassification 6-8 years after seroconversion [adjusted OR: 25.8 (95% CI: 8.17-81.5), p<0.001] after adjusting for race, CD4 cell count, HIV viral load, and HAART use. Low HIV viral load, low CD4 cell count, and >2 years of HAART were significantly associated with misclassification using the BED-CEIA. Some men were persistently misclassified as recently infected up to 8 years after HIV seroconversion.

Physicians make critical time-constrained decisions every day. Clinical predictive models can help physicians and administrators make decisions by forecasting clinical and operational events. Existing structured data-based clinical predictive models have limited use in everyday practice owing to complexity in data processing, as well as model development and deployment1–3. Here we show that unstructured clinical notes from the electronic health record can enable the training of clinical language models, which can be used as all-purpose clinical predictive engines with low-resistance development and deployment. Our approach leverages recent advances in natural language processing4,5 to train a large language model for medical language (NYUTron) and subsequently fine-tune it across a wide range of clinical and operational predictive tasks. We evaluated our approach within our health system for five such tasks: 30-day all-cause readmission prediction, in-hospital mortality prediction, comorbidity index prediction, length of stay prediction, and insurance denial prediction. We show that NYUTron has an area under the curve (AUC) of 78.7–94.9%, with an improvement of 5.36–14.7% in the AUC compared with traditional models. We additionally demonstrate the benefits of pretraining with clinical text, the potential for increasing generalizability to different sites through fine-tuning and the full deployment of our system in a prospective, single-arm trial. These results show the potential for using clinical language models in medicine to read alongside physicians and provide guidance at the point of care.