A systematic review of prediction models used in tuberculosis contact tracing. (13th November 2019)

• Record Type: Journal Article
• Authors: Kidy, F; Bruno-McClung, E; Shantikumar, S; Proto, W; Oyebode, O
• Is Part Of: European journal of public health. Volume 29(2019)Supplement 4
• Publication Date: 2019-11-13
• Subjects: Epidemiology -- Europe -- Periodicals; Public health -- Europe -- Periodicals; 362.109405
• DOI: 10.1093/eurpub/ckz186.495
• Languages: English
• ISSNs: 1101-1262

Abstract

Abstract: Background: Contact tracing forms a key part of tuberculosis (TB) control in high-income, low-incidence settings. It aims to reduce morbidity, mortality and onward transmission of TB. Contact tracing is a complex and resource intensive intervention. Risk assessment of contacts is needed to ensure appropriate allocation of resources and greatest possible impact. Current prioritisation procedures are based on expert opinion and consensus. Prognostic prediction models offer a way to synthesise evidence about this decision. Methods: We searched Medline, Embase, BNI, CINAHL, HMIC, and the Cochrane Library for peer reviewed publications in English about TB contact tracing prediction models. Studies were included if there was statistical combination of predictors. No date, age or other restrictions were applied. Study selection was carried out by two independent reviewers. Data were extracted using the CHARMS checklist and studies evaluated for risk of bias using PROBAST. Results: Five reports were selected from a total of 16, 585 non-identical returns. Each study was carried out in demographically distinct settings (Peru, USA, France, Taiwan). The choice and definition of outcomes and predictors varied. All the models included external validation and some included internal validation. Calibration and discrimination measures were variably reported. The models were at high risk of bias due to challenges in defining TB disease and due to statistical approaches taken: there was poor reporting of sample size considerations, universal use of univariable analysis to select predictors, and dichotomisation of data. There were some concerns about applicability due to differing populations and diagnostic approaches. None of the models included social risk factors. Conclusions: The use of existing models is problematic. There are constraints upon resources which means that contact tracing needs to be carried out efficiently. A robust prediction model is urgently needed to achieve this. Key messages: Contact tracing for tuberculosis would benefit from more robust prioritisation tools to save resources and increase impact. Existing prognostic prediction models are at high risk of bias and there are concerns about applicability in high-income, low-incidence settings.