Is more research always needed? Estimating optimal sample sizes for trials of retention in care interventions for HIV-positive East Africans. Issue 2 (20th July 2017)

• Record Type: Journal Article
• Authors: Uyei, Jennifer; Li, Lingfeng; Braithwaite, R Scott
• Is Part Of: BMJ global health. Volume 2:Issue 2(2017)
• Publication Date: 2017-07-20
• Subjects: value of information -- sample size -- expected value of sample information EVSI -- HIV -- antiretroviral therapy -- lost to follow up -- Kenya -- East Africa; World health -- Periodicals; 362.105
• DOI: 10.1136/bmjgh-2016-000195
• Languages: English
• ISSNs: 2059-7908

Abstract

Abstract : Introduction: Given the serious health consequences of discontinuing antiretroviral therapy, randomised control trials of interventions to improve retention in care may be warranted. As funding for global HIV research is finite, it may be argued that choices about sample size should be tied to maximising health. Methods: For an East African setting, we calculated expected value of sample information and expected net benefit of sampling to identify the optimal sample size (greatest return on investment) and to quantify net health gains associated with research. Two hypothetical interventions were analysed: (1) one aimed at reducing disengagement from HIV care and (2) another aimed at finding/relinking disengaged patients. Results: When the willingness to pay (WTP) threshold was within a plausible range (1–3 × GDP; US$1377–4130/QALY), the optimal sample size was zero for both interventions, meaning that no further research was recommended because the pre-research probability of an intervention's effectiveness and value was sufficient to support a decision on whether to adopt the intervention and any new information gained from additional research would likely not change that decision. In threshold analyses, at a higher WTP of $5200 the optimal sample size for testing a risk reduction intervention was 2750 per arm. For the outreach intervention, the optimal sample size remained zero across a wide range of WTP thresholds and was insensitive to variation. Limitations, including not varying all inputs in the model, may have led to an underestimation of the value of investing in new research. Conclusion: In summary, more research is not always needed, particularly when there is moderately robust prestudy belief about intervention effectiveness and little uncertainty about the value (cost-effectiveness) of the intervention. Users can test their own assumptions at http://torchresearch.org.