Choosing algorithms for TB screening: a modelling study to compare yield, predictive value and diagnostic burden

BMC Infectious Diseases

Table 2 Model Parameters: sensitivity and specificity of screening methods and confirmatory tests

Screen	Population (No. of studies)*	Sensitivity [95% CI]^†	Specificity [95% CI]^†	Reference
Symptom screening
Prolonged Cough (2-3 weeks or longer)	Community TB prevalence surveys (8)	0.351 [0.244; 0.457]	0.947 [0.925; 0.968]	[5]
	SSA-high HIV prevalence^§ (4)	0.492 [0.389; 0.597]	0.923 [0.891; 0.956]	[5]
	Asia-low HIV prevalence^§ (4)	0.247 [0.176; 0.317]	0.963 [0.947; 0.979]	[5]
Any TB Symptom (out of 4-7 symptoms)	Combined (8)	0.770 [0.680; 0.860]	0.677 [0.502; 0.851]	[5]
	SSA-high HIV prevalence^§ (4)	0.842 [0.756; 0.927]	0.740 [0.531; 0.949]	[5]
	Asia-low HIV prevalence^§ (4)	0.698 [0.579; 0.818]	0.606 [0.347; 0.866]	[5]
Chest X-ray screening
Any CXR abnormality	(3)	0.978 [0.951; 1.00]	0.754 [0.720; 0.788]	[5]
CXR abnormality suggestive of TB	(4)	0.868 [0.792; 0.945]	0.894 [0.867; 0.920]	[5]
Chest X-ray screening as a 2nd screen
Any CXR abnormality	(1)	0.90 [0.81; 0.96]	0.56 [0.54; 0.58]	[5, 13]
Confirmatory test
Sputum Smear microscopy	(30)	0.61 [0.31; 0.89]	0.98 [0.93; 1.0]	[14]
Xpert MTB/RIF	Multi-sites (1)	0.89 [0.63; 0.97]	0.99 [0.90; 1.00]	[15]
Clinical Diagnosis (PE), algorithm including trial of broad spectrum antibiotics and/or CXR and/or clinical judgment	Smear-negative presumptive TB patients from India, Uganda, South Africa, average of 3 sites, Lima	0.24 [0.10; 0.51]^‡	0.94 [0.79; 0.97]^‡	[16, 17]
Clinical Diagnosis (alternative) based on CXR highly consistent for TB	(1)	0.49 [0.45; 0.53]^‡	0.90 [0.88; 0.92]^‡	[18]

PE = point estimate; NPV = negative predictive value; SSA = Sub-Saharan Africa; TB = tuberculosis; CXR = chest X-ray.
*Number of studies included in the estimate.
^†The values in between brackets reflect the 95% confidence interval, except for Xpert MTB/RIF the 95% prediction interval was used, and for SSM the range across studies (see Methods section).
^§the 4 SSA-high HIV prevalence studies are from Zimbabwe, Zambia, South Africa and Kenya. The Asia-low HIV studies are from Vietnam, Myanmar, India and Cambodia.
^‡An assumption is made that in an active screening program only a proportion of patients with a negative confirmatory SSM or Xpert MTB/RIF result receive clinical diagnosis, and this proportion depends on the NPV (rounded to 2 decimals as follows: (1-NPV)*10 If NPV ≥ 99.5% then the proportion is 5%. This is equivalent to multiplying the sensitivity parameter by (1-NPV)*10. The number of false-positive diagnoses is adjusted as follows: if S is the specificity parameter, the proportion of false-positives is [(1-S)*((1-NPV)*10)]. In algorithms 3 and 4 all persons with prolonged cough and a CXR abnormality and negative confirmatory tests are assumed to be further evaluated clinically.

ISSN: 1471-2334