Improving estimators of HIV mortality and retention in care by linking clinical cohort data with national registries in Mexico.

Abstract

Introduction: Loss to follow-up (LTFU) bias mortality estimates in cohort studies. This study aimed to estimate the corrected incidence of LTFU, retention in care, and mortality rates in people with HIV (PWH) in an HIV/AIDS Clinic in Mexico City, by linking clinic and national registry data, and to identify characteristics associated with LTFU.

Methods: We linked adult PWH who met lost to follow-up criteria while enrolled at a Mexico City clinic with the national death registry, a national administrative HIV database and a clinical national registry in May 2019 and January 2021. We defined lost to follow-up as absence of a documented clinical contact in the previous 180 days. We estimated corrected LTFU, retention in care, mortality rates, and survival, and identified characteristics associated to LTFU.

Results: Among 2,826 PWH followed by a median of 5.9 years (IQR: 1.9-7.1), 804 (28.4%) were lost and 149 (5.3%) died. After the second linkage, of the 804 lost, 257 (32%) died, 175 (21.8%) transferred, 40 (4.9%) were retained in care, and 332 (41.3%) remained LTFU. Post second linkage, the corrected cumulative proportion of LTFU slightly decreased (n = 764, 27%), but deaths substantially increased (n = 459, 16.2%). A subset of PWH transferred to other centres (n = 184, 6.5%). The unadjusted LTFU rate decreased from 4.0 to 2.9 per 100 person-years (PY), while mortality increased from 0.7 to 1.7 per 100 PY. Younger age (aHR 1.5, 95%CI 1.38-1.63 per 10-year change), and lower education (aHR 1.05, 95%CI 1.01-1.08 per year) were associated to a higher risk of LTFU.

Conclusion: Datasets linkages revealed mortality rates three times higher than the original estimates, highlighting the limitations of relying in single-source data. At the individual level, identifying predictors of LTFU can help target interventions to improve retention and may reduce mortality. However, at the system-level, our study demonstrates the feasibility and value of cross-institutional data linkage in fragmented health systems. Key lessons include the importance of long-term, privacy-preserving data-sharing collaboration between institutions to improve patient tracking across institutions, mortality surveillance and care continuity.