Retention on antiretroviral therapy remained a major challenge in this Northern Ghana cohort. The investigators followed 329 adults who started ART at Nalerigu Baptist Medical Centre between January 2013 and December 2019. They calculated incidence rates and cumulative incidence while treating death and loss to follow‑up as competing events. The observed incidence for loss to follow‑up was 2.2 per 100 person‑months (95% CI: 1.9–2.6). The incidence for death was lower: 0.74 per 100 person‑months (95% CI: 0.57–0.98). At 36 months the cohort showed a cumulative incidence of 50.2% for LTFU and 16.0% for death. These figures indicate that, in this population and setting, LTFU was substantially more frequent than mortality over the three‑year follow‑up window.
The authors emphasized that conventional survival methods can misestimate the probability of either outcome when events compete, which motivated use of a subdistribution hazard approach. The data source was routine clinical records from the ART unit; public release of the full patient‑level data is restricted by ethics approval, though a minimal de‑identified dataset and code are available under controlled access.
The study team used the Fine‑Gray subdistribution hazard model to estimate associations between baseline characteristics and the cumulative incidence of each competing event. For LTFU, patients residing closer to the clinic had a lower subdistribution hazard compared with those living farther away (SHR = 0.68, 95% CI: 0.49–0.93). Baseline weight had a small positive association with LTFU (SHR = 1.02 per kg, 95% CI: 1.00–1.04), according to the authors. Those two variables emerged as meaningful correlates in the competing‑risks framework.
Mortality predictors differed. The investigators reported that receiving an efavirenz‑based regimen was associated with a lower subdistribution hazard of death (SHR = 0.27, 95% CI: 0.15–0.50). Female sex was associated with reduced mortality risk (SHR = 0.39, 95% CI: 0.18–0.85). Higher baseline weight was linked with a lower mortality hazard (SHR = 0.93 per kg, 95% CI: 0.90–0.96). The authors presented these as independent associations after accounting for the competing event.
These results suggest that drivers of LTFU and death are not identical in this programme. Geographic access — measured as distance to the treatment facility — appeared especially important for retention, whereas regimen choice, sex, and nutritional status (proxied by weight) were associated with survival.
To assess robustness, the investigators fitted cause‑specific Cox proportional hazards models as a sensitivity analysis. The manuscript states that the findings were broadly consistent between the Fine‑Gray subdistribution models and the cause‑specific Cox analyses. The authors did not report discrepant effect directions between the two approaches; they used the dual modelling strategy to address the distinct inferential aims of cumulative incidence versus instantaneous risk.
The preprint includes code and simulation material in public repositories to permit reproducibility of the modelling approach, while the underlying patient‑level data remain available only under controlled access in line with ethical restrictions.
The cohort results point to retention as the dominant programmatic failure relative to mortality over three years. Investigators recommended expanding access through decentralized and community‑based ART service delivery to reduce travel burden and improve retention. The study team framed travel distance to the clinic as a modifiable barrier; they also suggested that regimen selection and attention to baseline nutritional status could influence survival outcomes. Those practical recommendations derive directly from the associations the authors report in their competing‑risks models.
Patient‑level clinical records used for the analysis are governed by ethical approval from the Committee on Human Research, Publication and Ethics (CHRPE), School of Medicine and Dentistry, Kwame Nkrumah University of Science and Technology (Approval No. CHRPE/AP/1181/26). The authors deposited a minimal de‑identified dataset, data dictionary, and R scripts in Zenodo and shared analytical code on GitHub under controlled access; contact details for the ART unit at Baptist Medical Centre, Nalerigu, are provided in the submission for data requests.
The authors declared no competing interests and confirmed that participant consent and institutional oversight were in place. This article is a medRxiv preprint; the authors and the posting note that the work has not undergone peer review and therefore should not be used to guide clinical practice. No regulatory milestones or peer‑review outcomes are reported in the preprint.