Skip to main content

Table 2 Sample size required to detect significant difference in costs between standard and GX arms—in each country

From: GeneXpert or chest-X-ray or tuberculin skin testing for household contact assessment (GXT): protocol for a cluster-randomized trial

Estimated costs associated with standarda CAD$ 2017

Estimated costs associated with GXb CAD$ 2017

Power to detect effect sizesd (effect size = the detectable difference/SD)

Patient’s perspectivec

Health system perspective

Total

Patient perspectivec

Health system perspective

Total

0.3

0.33

0.5

Benin

 20

121

141

16

111

127

0.72

0.8

0.99

Brazil

 36

319

355

28

278

306

0.72

0.8

0.99

  1. aFor standard scenario: We assumed that HHC has two visits for TST (administration and reading). Half have three more visits for medical evaluation and CXR and 20% of these have an added two more visits to collect sputum samples; 25% have all of the above, plus 1 visit for LTBI treatment initiation and 3 more visits for LTBI treatment follow-up
  2. bFor GX scenario: We assume that HHC has two visits for TST (administration and reading). Half have one more visit for medical evaluation and GX. One quarter also have one visit for LTBI treat initiation and 3 more visits for LTBI treatment follow-up
  3. cCosts from the patient perspective: Expenses associated with medical visits assumed to be $4.00 per visit in Benin and $7.50 per visit in Brazil. This accounts for travel costs and additional expenses during travel or at medical visit[4]
  4. dTo estimate power, we assume alpha = 0.05, and 455/3 = 152 analyzable subjects per group in each country, and we considered the effect size (detectable difference/SD). We do not know the standard deviation but can estimate approximate costs, based on prior work in each country. As an example, based on costs collected previously in Ghana (neighboring country to BENIN, that also participated in our prior RCT of 4RIF vs 9INH) [1], we expect a difference in total costs of $28 between standard and GX arms. If the standard deviation is $84, then the effect size will be ($28/$84) = 0.33. After accounting for clustering by household, assuming an ICC of 0.33 and 4 subjects per household this effect size will result in estimated power of 80%. If the SD is actually smaller (SD = $56), then for the same expected difference in costs, we will have an effect size = 0.5, providing 99% power to detect a significant difference