Clinical risk prediction models to pre-screen populations at risk for sexually transmitted and blood-borne infections

Background: Globally, 1 million people are infected with a sexually transmitted or blood-borne infection (STBBI) each day.In Canada, 63,000 and 246,000 people are currently living with human immunodeficiency (HIV) and Hepatitis C (HCV) viruses, respectively.Syphilis rates have also increased by 178% in the past 10 years.People who use injection drugs (PWID) and someLGBTQ2+ individuals are at increased risk for these three STBBIs, including co-infections.Despite advances in screening technologies, many remain unaware of their status.Thus, current screening programs would benefit from deployment of clinical risk prediction models (CRPM) that estimate STBBI risk based on individual characteristics to support clinicians in screening those most at risk.With increasing development of data-driven methods (i.e. machine learning [ML] and Bayesian statistics), existing evidence should be supplemented with models developed using these methods.Objective(s): The objective of this manuscript-based thesis was to assess STBBI CRPMs as prescreening tools by 1) conducting a review of existing CRPMs, 2) developing and validating a CRPM for HIV/syphilis/HCV in Canadian key populations using Bayesian methods.Methods: 1) PubMed was searched for HCV and/or chlamydia and gonorrhea (CT/NG) CRPMs published since January 2021 and since 2016 for syphilis to extract information on data sources, STBBIs predicted, population targeted, statistical methods, and performance.2) A CRPM was developed using cross-sectional data of a study evaluating multiplex immunological tests for HIV, HCV, and syphilis.20 clinical, socio-demographic, behavioural, and individual categorical variables were extracted from participants (𝑛 = 400) recruited in two clinics serving PWIDs and LGBTQ2+ people in two Canadian provinces.Missing data for 77 individuals was imputed with multiple imputation.Pooled imputed datasets were split into development (𝑛 = iv 300) and validation (𝑛 = 100) data using clinic-stratified sampling.Predictors of STBBI (i.e.HIV, HCV, and/or syphilis) were selected using Bayesian predictive projection with regularized horseshoe priors (𝜏 = 0.02) and leave-one out cross-validation.The performance of the nine smallest sub-models were evaluated using area under the receiver operating curve (AUC), sensitivity, and specificity along with 89% Credible Intervals (89%CrI).All analyses were done in R (≥v4.2.3).Results: 1) Data on 27 CRPMs from 19 countries was abstracted.Out of which, 10 predicted HCV, 6 CT/NG, 3 syphilis, and 8 multiple STBBIs.Most CRPMs targeted specific populations (𝑛 = 18), and 9 were not validated.Methods for development included frequentist statistics (𝑛 = 17) and ML (𝑛 = 9).Only 4 CRPMs made use of Bayesian statistics.Performance (i.e.AUC) and quality varied.2) Out of 400, 73 participants were infected with HIV (𝑛 = 16), HCV (𝑛 = 60) and/or syphilis (𝑛= 5).A sub-model with two predictors (i.e.Drug injection history, past STI testing type) displayed the highest AUC (0.79; 89%CrI: 0.66-0.79)with validation data.Its sensitivity (0.85; 89%CrI: 0.79-0.91)was higher than its specificity (0.3; 89%CrI: 0.15-0.5).Discussion: Overall, the review highlighted the need to develop and validate targeted CRPMs for key populations by harnessing data-driven methods in a rigorous methodological way.The internally validated CRPM had an acceptable AUC and high sensitivity.Bayesian methods enabled integration of prior information into the analysis and quantification of parameter uncertainty.Conclusion: This is the first CRPM developed with Bayesian methods for HIV, HCV, and syphilis screening.Combining it with novel screening methods could increase access to care for key populations and help decrease.vRésumé Contexte : Chaque jour, un million de personnes sont infectées mondialement par une infection transmise sexuellement ou par le sang (ITSS).Au Canada, 63 000 et 246 000 personnes vivent actuellement avec le virus de l'immunodéficience humaine (VIH) et le virus de l'hépatite C (VHC), respectivement.Les taux de syphilis ont également augmenté de 178 % au cours des dix dernières années.Les personnes qui s'injectent des drogues (PID) et certaines personnes LGBTQ2+ courent un risque plus élevé de contracter ces trois ITSSs, y compris des co-infections.Malgré les progrès en matière de technologies de dépistage, nombreuses sont les personnes infectées qui ignorent leur statut.Les programmes de dépistage actuels bénéficieraient donc du déploiement de modèles de prédiction du risque clinique (MPRC) qui estiment le risque d'ITSS en fonction de caractéristiques individuelles, afin d'aider les cliniciens à dépister les personnes les plus vulnérables.Avec le développement croissant des méthodes basées sur les données (comme l'apprentissage automatique [AA] et les statistiques Bayésiennes), les évidences existantes devraient être renforcées par des modèles développés à l'aide de ces méthodes.Objectif(s) : L'objectif de cette thèse basée sur un manuscrit était d'évaluer les MPRC des ITSS en tant qu'outils de présélection pour le dépistage en 1) effectuant un bilan des MPRC existants, 2) développant et validant un MPRC pour le VIH/syphilis/VHC dans les populations clés canadiennes à l'aide de méthodes Bayésiennes.Méthodes : 1) Une recherche a été effectuée sur PubMed pour trouver les études sur les MPRCs pour le VHC et/ou la chlamydiose et la gonorrhée (CT/NG) publiées depuis janvier 2021 et depuis 2016 pour la syphilis afin d'extraire de l'information sur les sources de données, les ITSSs prédites, la population ciblée, les méthodes statistiques et la performance.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.