Pooling biological specimens in primary studies compromises evidence generation in test accuracy reviews: a case study
Yemisi Takwoingi, University of Birmingham
2Christian Medical College, Vellore, India
3Institute of Applied Health Research, University of Birmingham, UK
Background:
Pooling multiple biological specimens collected at different time points to increase the quantity and quality of the sample from an individual for a single test may improve the diagnostic yield of an index test (test being evaluated). However, this may overestimate test accuracy in primary studies, as well as in meta-analysis, and lead to a high concern about applicability if only a single sample will be tested in clinical practice.
Objectives:
To explore the frequency of pooling biological specimens in primary studies and the effect on estimates of test accuracy using two Cochrane diagnostic test accuracy reviews assessing rapid diagnostic tests for tuberculosis.
Methods:
We contacted the authors of included studies for additional data when the papers reported results were available for pooled and single (unpooled) samples from individuals. We performed meta-analysis separately for the two groups in each review when possible. We also examined individual studies and compared estimates of sensitivity and specificity if a study included results from both pooled and unpooled samples.
Results:
Contrary to the diagnostic algorithm of tuberculosis programs or routine clinical practice, four of the eight included studies used pooled sputum samples from one or more patient cohorts. Larger studies tended to pool sputum (Pillay 2022; Inbaraj 2023). One of the four studies assessed the Truenat MTB assay for the detection of pulmonary tuberculosis in two cohorts using pooled sputum in one and unpooled sputum in the other cohort (Figure 1) (Gomathi 2020). From this within-study comparison of unpooled versus pooled, the differences in sensitivity (-11 percentage points, 95% CI -15 to -6.2) and specificity (15 percentage points, 95% CI 10 to 20) were large. Analyses are ongoing and further results will be available at the Colloquium.
Conclusions:
Test accuracy studies should ensure that samples used for evaluating tests reflect how the tests will be performed in clinical practice to ensure robust estimates of real-world performance and applicability. This will also reduce research waste.