Neighbourhood prevalence-to-notification ratios for adult bacteriologically-confirmed tuberculosis reveals hotspots of underdiagnosis in Blantyre, Malawi

Abstract

Local information is needed to guide targeted interventions for respiratory infections such as tuberculosis (TB). Case notification rates (CNRs) are readily available, but systematically underestimate true disease burden in neighbourhoods with high diagnostic access barriers. We explored a novel approach, adjusting CNRs for under-notification (P:N ratio) using neighbourhood-level predictors of TB prevalence-to-notification ratios. We analysed data from 1) a citywide routine TB surveillance system including geolocation, confirmatory mycobacteriology, and clinical and demographic characteristics of all registering TB patients in Blantyre, Malawi during 2015–19, and 2) an adult TB prevalence survey done in 2019. In the prevalence survey, consenting adults from randomly selected households in 72 neighbourhoods had symptom-plus-chest X-ray screening, confirmed with sputum smear microscopy, Xpert MTB/Rif and culture. Bayesian multilevel models were used to estimate adjusted neighbourhood prevalence-to-notification ratios, based on summarised posterior draws from fitted adult bacteriologically-confirmed TB CNRs and prevalence. From 2015–19, adult bacteriologically-confirmed CNRs were 131 (479/371,834), 134 (539/415,226), 114 (519/463,707), 56 (283/517,860) and 46 (258/578,377) per 100,000 adults per annum, and 2019 bacteriologically-confirmed prevalence was 215 (29/13,490) per 100,000 adults. Lower educational achievement by household head and neighbourhood distance to TB clinic was negatively associated with CNRs. The mean neighbourhood P:N ratio was 4.49 (95% credible interval [CrI]: 0.98–11.91), consistent with underdiagnosis of TB, and was most pronounced in informal peri-urban neighbourhoods. Here we have demonstrated a method for the identification of neighbourhoods with high levels of under-diagnosis of TB without the requirement for a prevalence survey; this is important since prevalence surveys are expensive and logistically challenging. If confirmed, this approach may support more efficient and effective targeting of intensified TB and HIV case-finding interventions aiming to accelerate elimination of urban TB.

Prevalence of Bacteriologically-Confirmed Tuberculosis in Urban Blantyre, Malawi 2019-20: Substantial Decline Compared to 2013-14 National Survey

“…[26,27]. As such, the pronounced decline in the prevalence of undiagnosed TB that we infer from comparing our results to the 2013-14 national survey is almost certainly correct, although our survey still demonstrates under-diagnosis of TB in Blantyre with a prevalence to case-notification ratio of 4.49 (95% CI: 0.98-11.91) as reported elsewhere [10]. Undiagnosed infectious TB remains well above TB elimination targets, underscoring the need to continue appropriately targeted case-finding activities in Blantyre.…”

“…Between the 2013-2014 NTP prevalence survey and this 2019-2020 survey, annual TB case notification rates, including bacteriologically confirmed TB, had been declining in Blantyre [10] with concurrent steep reductions in the percentage of primary care clinic attendees with bacteriologically-confirmed TB following self-presentation for investigation of TB symptoms. [26,27].…”

“…TB case-notification rates (CNRs) have been declining in Blantyre [10], likely reflecting the combined effects of increasing coverage with antiretroviral therapy (ART) [11,12] and use of TB preventive therapy for people living with HIV (PLHIV), as well as TB case-finding and prevention activities. HIV testing and treatment services have been successfully scaled-up to reach UNAIDS 90-90-90 targets for 2020 in Malawi [12,13].…”

Treatment preferences among people at risk of developing tuberculosis: a discrete choice experiment

“…In constructing the choice sets for the DCE, using a full factorial design would be the most balanced and orthogonal as it contains all possible combinations of attributes and levels (23). However, with a total of 7 attributes and 25 levels this would result in 1024 (4 3 X 2 4 X 1 5 ) choice sets, which was impractically large and would result in respondent fatigue (23). Therefore, to make the task manageable, we used a fractional factorial design (23,24), whereby participants were exposed to a random selection of 10 choice sets from the full factorial design.…”

“…Although some progress has been made in reducing TB incidence and death, particularly in countries that have experienced HIV-driven epidemics in Africa, this has fallen well short of global targets, and has further been worsened by the disruptions to diagnosis and treatment caused by the COVID-19 pandemic (2,3). Additionally, the TB epidemic has become increasingly concentrated in marginalised groups as the overall TB prevalence has reduced (4,5). Therefore, new, person-centred, and effective approaches must be considered to diagnose and treat individuals early.…”