Predicting the presence of depressive symptoms in the HIV-HCV co-infected population in Canada using supervised machine learning.

BACKGROUND: Depression is common in the human immunodeficiency virus (HIV)-hepatitis C virus (HCV) co-infected population. Demographic, behavioural, and clinical data collected in research settings may be of help in identifying those at risk for clinical depression. We aimed to predict the presence of depressive symptoms indicative of a risk of depression and identify important classification predictors using supervised machine learning. METHODS: We used data from the Canadian Co-infection Cohort, a multicentre prospective cohort, and its associated sub-study on Food Security (FS). The Center for Epidemiologic Studies Depression Scale-10 (CES-D-10) was administered in the FS sub-study; participants were classified as being at risk for clinical depression if scores ≥ 10. We developed two random forest algorithms using the training data (80%) and tenfold cross validation to predict the CES-D-10 classes-1. Full algorithm with all candidate predictors (137 predictors) and 2. Reduced algorithm using a subset of predictors based on expert opinion (46 predictors). We evaluated the algorithm performances in the testing data using area under the receiver operating characteristic curves (AUC) and generated predictor importance plots. RESULTS: We included 1,934 FS sub-study visits from 717 participants who were predominantly male (73%), white (76%), unemployed (73%), and high school educated (52%). At the first visit, median age was 49 years (IQR:43-54) and 53% reported presence of depressive symptoms with CES-D-10 scores ≥ 10. The full algorithm had an AUC of 0.82 (95% CI:0.78-0.86) and the reduced algorithm of 0.76 (95% CI:0.71-0.81). Employment, HIV clinical stage, revenue source, body mass index, and education were the five most important predictors. CONCLUSION: We developed a prediction algorithm that could be instrumental in identifying individuals at risk for depression in the HIV-HCV co-infected population in research settings. Development of such machine learning algorithms using research data with rich predictor information can be useful for retrospective analyses of unanswered questions regarding impact of depressive symptoms on clinical and patient-centred outcomes among vulnerable populations.

Measuring progress towards reaching zero new HIV acquisitions among key populations in Québec (Canada) using routine surveillance data: a mathematical modelling study.

INTRODUCTION: Men who have sex with men (MSM) and people who inject drugs (PWID) are disproportionately impacted by the HIV epidemic in Canada. Having the second-highest provincial diagnosis rate, an improved understanding of the epidemic among these populations in Québec could aid ongoing elimination efforts. We estimated HIV incidence and other epidemic indicators among MSM and PWID in Montréal and across Québec using a back-calculation model synthesizing surveillance data. METHODS: We developed a deterministic, compartmental mathematical model stratified by age, HIV status and disease progression, and clinical care stages. Using AIDS and HIV diagnoses data, including self-reported time since the last negative test and laboratory results of CD4 cell count at diagnosis, we estimated HIV incidence in each population over 1975-2020 by modelling a cubic M-spline. The prevalence, undiagnosed fraction, fraction diagnosed that started antiretroviral treatment (ART) and median time to diagnosis were also estimated. Since the COVID-19 pandemic disrupted testing, we excluded 2020 data and explored this in sensitivity analyses. RESULTS: HIV incidence in all populations peaked early in the epidemic. In 2020, an estimated 97 (95% CrI: 33-227) and 266 (95% CrI: 103-508) HIV acquisitions occurred among MSM in Montréal and Québec, respectively. Among PWID, we estimated 2 (95% CrI: 0-14) and 6 (95% CrI: 1-26) HIV acquisitions in those same regions. With 2020 data, unless testing rates were reduced by 50%, these estimates decreased, except among Québec PWID, whose increased. Among all, the median time to diagnosis shortened to <2 years before 2020 and the undiagnosed fraction decreased to <10%. This fraction was higher in younger MSM, with 22% of 15-24 year-olds living with HIV in Montréal (95% CrI: 9-39%) and 31% in Québec (95% CrI: 17-48%) undiagnosed by 2020 year-end. Finally, ART access neared 100% in all diagnosed populations. CONCLUSIONS: HIV incidence has drastically decreased in MSM and PWID across Québec, alongside significant improvements in diagnosis and treatment coverage-and the 2013 introduction of pre-exposure prophylaxis. Despite this, HIV transmission continued. Effective efforts to halt this transmission and rapidly diagnose people who acquired HIV, especially among younger MSM, are needed to achieve elimination. Further, as the impacts of the COVID-19 pandemic on HIV transmission are understood, increased efforts may be needed to overcome these.

National HIV testing and diagnosis coverage in sub-Saharan Africa: a new modeling tool for estimating the 'first 90' from program and survey data

HIV testing services (HTS) are a crucial component of national HIV responses. Learning one's HIV diagnosis is the entry point to accessing life-saving antiretroviral treatment and care. Recognizing the critical role of HTS, the Joint United Nations Programme on HIV/AIDS (UNAIDS) launched the 90-90-90 targets stipulating that by 2020, 90% of people living with HIV know their status, 90% of those who know their status receive antiretroviral therapy, and 90% of those on treatment have a suppressed viral load. Countries will need to regularly monitor progress on these three indicators. Estimating the proportion of people living with HIV who know their status (i.e. the 'first 90'), however, is difficult. Methods: We developed a mathematical model (henceforth referred to as 'Shiny90') that formally synthesizes population-based survey and HTS program data to estimate HIV status awareness over time. The proposed model uses country-specific HIV epidemic parameters from the standard UNAIDS Spectrum model to produce outputs that are consistent with other national HIV estimates. Shiny90 provides estimates of HIV testing history, diagnosis rates, and knowledge of HIV status by age and sex. We validate Shiny90 using both in-sample comparisons and out-of-sample predictions using data from three countries: Côte d'Ivoire, Malawi, and Mozambique. Results: In-sample comparisons suggest that Shiny90 can accurately reproduce longitudinal sex-specific trends in HIV testing. Out-of-sample predictions of the fraction of people living with HIV ever tested over a 4-to-6-year time horizon are also in good agreement with empirical survey estimates. Importantly, out-of-sample predictions of HIV knowledge of status are consistent (i.e. within 4% points) with those of the fully calibrated model in the three countries when HTS program data are included. The model's predictions of knowledge of status are higher than available self-reported HIV awareness estimates, however, suggesting - in line with previous studies - that these self-reports could be affected by nondisclosure of HIV status awareness. Conclusion: Knowledge of HIV status is a key indicator to monitor progress, identify bottlenecks, and target HIV responses. Shiny90 can help countries track progress towards their 'first 90' by leveraging surveys of HIV testing behaviors and annual HTS program data.