A clinical score for identifying active tuberculosis while awaiting microbiological results: Development and validation of a multivariable prediction model in sub- Saharan Africa
Loading...
Date
2020
Journal Title
Journal ISSN
Volume Title
Publisher
PLoS medicine
Abstract
In highly resource-limited settings, many clinics lack same-day microbiological testing for
active tuberculosis (TB). In these contexts, risk of pretreatment loss to follow-up is high, and
a simple, easy-to-use clinical risk score could be useful.
Methods and findings
We analyzed data from adults tested for TB with Xpert MTB/RIF across 28 primary health
clinics in rural South Africa (between July 2016 and January 2018). We used least absolute
shrinkage and selection operator regression to identify characteristics associated with
Xpert-confirmed TB and converted coefficients into a simple score. We assessed discrimination
using receiver operating characteristic (ROC) curves, calibration using Cox linear
logistic regression, and clinical utility using decision curves. We validated the score externally
in a population of adults tested for TB across 4 primary health clinics in urban Uganda
(between May 2018 and December 2019). Model development was repeated de novo with
the Ugandan population to compare clinical scores. The South African and Ugandan cohorts
included 701 and 106 individuals who tested positive for TB, respectively, and 686 and 281
randomly selected individuals who tested negative. Compared to the Ugandan cohort, the
South African cohort was older (41% versus 19% aged 45 years or older), had similar breakdown
of biological sex (48% versus 50% female), and had higher HIV prevalence (45% versus
34%). The final prediction model, scored from 0 to 10, included 6 characteristics: age,
sex, HIV (2 points), diabetes, number of classical TB symptoms (cough, fever, weight loss,and night sweats; 1 point each), and >14-day symptom duration. Discrimination was moderate
in the derivation (c-statistic = 0.82, 95% CI = 0.81 to 0.82) and validation (c-statistic =
0.75, 95% CI = 0.69 to 0.80) populations. A patient with 10% pretest probability of TB would
have a posttest probability of 4% with a score of 3/10 versus 43% with a score of 7/10. The
de novo Ugandan model contained similar characteristics and performed equally well. Our
study may be subject to spectrum bias as we only included a random sample of people without
TB from each cohort. This score is only meant to guide management while awaiting
microbiological results, not intended as a community-based triage test (i.e., to identify individuals
who should receive further testing).
Conclusions
In this study, we observed that a simple clinical risk score reasonably distinguished individuals
with and without TB among those submitting sputum for diagnosis. Subject to prospective
validation, this score might be useful in settings with constrained diagnostic resources
where concern for pretreatment loss to follow-up is high.
Description
Keywords
Clinical score, Tuberculosis, microbiological results
Citation
Baik Y, Rickman HM, Hanrahan CF, Mmolawa L, Kitonsa PJ, Sewelana T, et al. (2020) A clinical score for identifying active tuberculosis while awaiting microbiological results: Development and validation of a multivariable prediction model in sub-Saharan Africa. PLoS Med 17(11): e1003420. https://doi.org/10.1371/journal. pmed.1003420