Sampling-Based Approaches to Improve Estimation of Mortality among Patient Dropouts: Experience from a Large PEPFAR-Funded Program in Western Kenya

Abstract

BackgroundMonitoring and evaluation (M&E) of HIV care and treatment programs is impacted by losses to follow-up (LTFU) in the patient population. The severity of this effect is undeniable but its extent unknown. Tracing all lost patients addresses this but census methods are not feasible in programs involving rapid scale-up of HIV treatment in the developing world. Sampling-based approaches and statistical adjustment are the only scaleable methods permitting accurate estimation of M&E indices.Methodology/Principal FindingsIn a large antiretroviral therapy (ART) program in western Kenya, we assessed the impact of LTFU on estimating patient mortality among 8,977 adult clients of whom, 3,624 were LTFU. Overall, dropouts were more likely male (36.8% versus 33.7%; p = 0.003), and younger than non-dropouts (35.3 versus 35.7 years old; p = 0.020), with lower median CD4 count at enrollment (160 versus 189 cells/ml; p<0.001) and WHO stage 3–4 disease (47.5% versus 41.1%; p<0.001). Urban clinic clients were 75.0% of non-dropouts but 70.3% of dropouts (p<0.001). Of the 3,624 dropouts, 1,143 were sought and 621 had their vital status ascertained. Statistical techniques were used to adjust mortality estimates based on information obtained from located LTFU patients. Observed mortality estimates one year after enrollment were 1.7% (95% CI 1.3%–2.0%), revised to 2.8% (2.3%–3.1%) when deaths discovered through outreach were added and adjusted to 9.2% (7.8%–10.6%) and 9.9% (8.4%–11.5%) through statistical modeling depending on the method used. The estimates 12 months after ART initiation were 1.7% (1.3%–2.2%), 3.4% (2.9%–4.0%), 10.5% (8.7%–12.3%) and 10.7% (8.9%–12.6%) respectively.Conclusions/Significance AbstractAssessment of the impact of LTFU is critical in program M&E as estimated mortality based on passive monitoring may underestimate true mortality by up to 80%. This bias can be ameliorated by tracing a sample of dropouts and statistically adjust the mortality estimates to properly evaluate and guide large HIV care and treatment programs.

Estimated mortality of adult HIV-infected patients starting treatment with combination antiretroviral therapy

“…Gaps in care, where a patient discontinued attending the clinic for some time but subsequently returned to care, were not considered as losses to follow-up. On the other hand, patients who were outreached were essentially treated as having been lost to follow-up regardless of the length of time between their last visit and the outreach encounter date as, once traced successfully, their loss to follow-up status would never be determined with certainty 17. This only applies to cohorts with documented outreach programmes (Kenya) or established linkages to vital registries (South Africa).…”

“…17 For South African data, we used similar methods, but lost patients were instead ‘searched’ in the death registries of the country 28. Only passively determined death information was available in all other cohorts.…”

“…Two reports from patient cohorts in the US and the UK also cited high levels of patient loss14 15 suggesting that this phenomenon is not limited to resource-limited settings. Moreover, it appears that, at least in sub-Saharan Africa, the risks for mortality overlap to a large extent with the risks of loss from follow-up16 17 and that the hardest to reach patients may also be those at the highest risk of mortality 18. Thus, it comes as no surprise that programmes which undertake more intensive efforts to locate patients who have been lost have higher levels of mortality 2.…”

“…17 19 These efforts are emanating from the work on double sampling 20 by Frangakis and Rubin21 and inverse probability of treatment weighting schemes introduced by Robins and others 2223 They are based on the availability of vital status information on a random sample of patients among those who have been lost to follow-up (the second or ‘double’ sample).…”

Implementation and outcomes of an active defaulter tracing system for HIV, prevention of mother to child transmission of HIV (PMTCT), and TB patients in Kibera, Nairobi, Kenya

“…death, transfer out) is unknown as LTFU contributes to inaccurate conclusions about the effectiveness of a given programme. 23,24 A 2009 systematic review and metaanalysis of 17 point-in-time ART LTFU studies concluded that only 63% of patients could be found to ascertain their vital status, and 46% of patients in African studies had died by the time they were traced. 19 Existing LTFU studies have been limited to retrospectively examining the true outcomes of an entire ART cohort at one point-in-time months or years after patients actually defaulted.…”