Table 4.

Approaches to impact evaluation for TB screening interventions (adapted from¹⁰²)

Impact being evaluated	Study designs	Expected outputs and examples
Any high-risk group
Comparison of performance characteristics between different tests and algorithms	Cross-sectional studies evaluating new TB diagnostics	Estimates sensitivity and specificity; can inform on robustness of new diagnostic systems in resource-poor settings30,31,78 Provides number-needed-to-screen per TB patient diagnosed in different populations
HIV care clinics and household contacts
Direct cohort follow-up to evaluate patient-important outcomes following screening	Cohort studies comparing outcomes according to whether or not/how screened Appropriate comparator populations provided by randomisation (e.g., step-wedge), or historical or non-randomly selected comparison cohorts	Outcomes post-screening can include numbers diagnosed with TB at and after screen, vital status, retention in care, time to TB treatment58 Cost effectiveness and consequences of false-negative and false-positive screening results can also be assessed
Well-defined high TB incidence populations
Time trends in TB case notification rates compared to non-intervention comparator populations	Time trend analysis for: initial peak in case notifications attributable to screening, and ‘additionality’ compared to comparator and historic trends post-peak accelerated rate of decline in new cases	Need accurate routine case notification system; can be confounded by other changes in routine TB diagnosis/reporting Requires disaggregation of routine case notification data to subdistrict level (unless intervention is district-wide)
Time trends in deaths from diagnosed/undiagnosed TB	Aiming for reduced diagnosed +/− undiagnosed TB deaths ideally routine as well as intervention participants	Need complete TB registration and outcome data for diagnosed TB deaths Accurate capture of undiagnosed TB deaths requires autopsy
Prevalence surveys for undiagnosed TB in the general population	Repeated before-after, or cluster-randomised cross-sectional outcomes Requires very large sample sizes and consistent survey methods	Change/difference in undiagnosed TB provides outcome (less is always better). Several recent examples7,22,39,43,107 Before:after change does not prove causality; aim for major change in short time period to provide strongest evidence104
TB transmission rates	Before-after estimates, or cluster randomised cohort or cross-sectional outcomes; requires large sample sizes and consistent methods	Evidence of impact on TB transmission is an ideal outcome, but difficult to measure and so not often evaluated in high TB incidence settings; results affected by HIV status One recent example used an incidence cohort design in schoolchildren22,43
TB prevalence in HIV-infected patients	Assessed through repeated before-after design, regular surveillance with time trends, or cluster-randomised cross-sectional outcomes New clinic attendees, or routine post-mortem	A key indicator of population-level TB control, and also the main target of prevention for TB-HIV interventions Occurs at high prevalence and clearly linked to TB transmission

TB = tuberculosis; HIV = human immunodeficiency virus