Since their inception, HIV treatment programmes in Sub-Saharan Africa (SSA) have implemented a number of changes in the initiation, mode of delivery, and monitoring of antiretroviral therapy (ART). The initial scale-up was facilitated by the availability of inexpensive, effective fixed-dose combination drug regimens. Stavudine (d4T), in combination with lamivudine and efavirenz or nevirapine, was an important component of the initial regimens (Wester et al. 2005; Stringer et al. 2006; Assefa & Kloos 2008). However, the short- and long-term toxicities of d4T constrained its continued use. In 2006, the World Health Organization (WHO) recommended the use of newer generation, less toxic, but more expensive antiretroviral drugs (WHO 2006). Following this recommendation, tenofovir disoproxil fumarate (TDF), which has a favourable toxicity profile and activity against hepatitis B virus, an important co-infection in SSA, has gradually replaced d4T. While randomised trials comparing patient outcomes with d4T vs. TDF have been undertaken (Gallant et al. 2004), there have been limited observational cohort data on this comparison in SSA. Using data from a single large HIV clinic in Johannesburg, South Africa, Brennan and colleagues compared 24-month outcomes between the two drugs (Brennan et al. 2014). Comparing ART outcomes between two or more drug regimens using observational data is susceptible to various forms of confounding, including that of secular trends. Secular trends confound observational analysis when changes in patient and programme characteristics are paralleled by changes in the probability that a patient receives one treatment vs. another (i.e. as a new policy is implemented). In their paper, Brennan and colleagues leveraged the rapid implementation of South Africa's 2010 switch from d4T to TDF to reduce the impact of secular trends. In Themba Lethu Clinic, implementation of this policy change was very rapid with the proportion of patients initiating d4T-containing regimens decreasing from approximately 90% to 20% within 3 months. This allowed Brennan and colleagues to utilise a before and after study design, making the assumption that patient characteristics before and after implementation were similar (i.e. that secular trends were minimal). In the primary analysis, the authors compared those starting d4T in the 12 months before with TDF in the 12 months after implementation. The authors excluded from analysis patients who received TDF before the guideline change, an action that could have biased their results, as that group of patients may have been sicker and their exclusion could have altered average outcomes in the TDF group. Patients in the TDF group had higher CD4+ T-cell counts and lower WHO disease stages than their d4T counterparts, reflecting the substantial secular changes in patient and HIV programme characteristics that have been described in SSA (Geng et al. 2011). To account for these secular trends, Brennan and colleagues adjusted multivariable models for available patient characteristics. Another approach could have been to adjust models for calendar month. Importantly, the authors conducted a sensitivity analysis on the cohort that initiated ART within 3 months of implementation and reported similar outcomes. This lent evidence that secular trends were not influencing their results. So, what were their results? The two regimens were similar; however, there was a trend towards lower mortality rates, fewer losses to follow-up and better viral suppression in the TDF group. Whether the magnitude of the difference was clinically significant and/or generalisable is a subject for further studies in different settings. These results provide important evidence that the policy change did not adversely affect patient outcomes. Further analysis of cost-effectiveness and quality-adjusted life years would be required to fully understand the benefits of policy implementation to the South African HIV programme. In conclusion, Brennan and colleagues' paper demonstrates how the rapid implementation of a treatment policy can be leveraged to reduce bias in observational data. By selecting a short interval around the guideline change, the impact of secular trends was reduced. Similar approaches could be employed to answer other important clinical and implementation questions.