Baker 2016.

Methods	Design: Controlled ITS Timing: Retrospective Allocation to group: Quote: "A natural experiment study design was adopted since the intervention (ratification of a US‐FTA) is not amenable to experimental manipulation, and it was possible to compare an intervention country (one that had ratified a US‐FTA) and a control country (one that had not). The control country was matched to the intervention country against economic, demographic and trade indicators reported in the literature to effect soft drink consumption, including population and income growth (…). We also wanted to include a TPP country in the analysis to demonstrate what has previously happened in the country as a consequence of trade liberalization, thereby providing a richer understanding of the context in which the TPP will operate. Against these criteria Peru was selected as the intervention country and Bolivia as the control country" Number of clusters or sites: 2 countries Number of individuals: Peru (the intervention country) had, in 2013, a population of 30 million. Bolivia, the control country, had a population of 10 million Length of intervention: 4 ‐ 6 years (from enforcement and ratification onwards, respectively)
Participants	General description of participants: Inhabitants of Peru Age: All ages Inclusion criteria: N/A Exclusion criteria: N/A Recruitment: N/A Weight status at baseline: Quote: "Obesity is a pressing public health concern in Peru and Bolivia where rates of adult obesity are high, 15.7 and 17.9 % respectively" SSB consumption at baseline: Baseline (2006) SSB consumption was approx. 50 l/capita/year in Peru (the IG) and 45 l/capita/year in Boliva (the CG) Equity considerations: N/R
Interventions	Intervention: A bilateral trade and investment agreement. (Quote: "The US‐Peru FTA [free trade agreement], which is similar to the preceding NAFTA and CAFTA agreements, resulted in the preferential elimination of tariffs on soft‐drink imports (previously 25 %), as well as stronger protections for US investors, and strengthened intellectual property rights") Behavioural co‐intervention: None reported Control: No intervention. (Quote: "[W]e adopted a natural experiment design to quantify the effects of the US‐Peru FTA, ratified in 2006 and enforced in 2009, in contrast to Bolivia, a suitably matched country having no such agreement, on Peru’s soft drink market")
Outcomes	Measures of SSB intake: Rate of change in SSB, and sports and energy drinks sales in litre per capita, and in sugar from soft drinks in kg per capita, assessed continuously throughout the study period with data from the commercial Euromonitor International Passport Global Market Information database Measures of intake of alternatives to SSB: Rate of change in bottled water and juice sales in litre per capita, assessed continuously throughout the study period with data from the commercial Euromonitor International Passport Global Market Information database Anthropometric measures: N/R Adverse outcomes: The study does not report how data on adverse outcomes were collected, and if adverse outcomes were observed or not, but notes per‐capita sales of sports and energy drinks increased (see above) Other outcomes: None included in this review
Context and implementation	Setting: Low‐ and middle‐income countries Sector: Trade and investment policy Country: Peru Year(s) when implemented: 2006 ‐ 2009 Mode of implementation: Mandatory government regulation Level of implementation: Policy‐level intervention
Declarations	COI: "The authors declare that they have no competing interests" Funding: "SF and PB are supported by funding from an Australian Research Council Discovery Project, “Trade policy: Maximising benefits for nutrition, food security, human health and the economy” (DP130101478). RL is an investigator on the same Project but does not receive funding from it, and is supported through the Canada Research Chairs program" Trial registration: N/R Protocol availability: Protocol mentioned in the report but not publicly available
Notes	None
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	This was a controlled ITS, and allocation was not random and not concealed
Allocation concealment (selection bias)	High risk	This was a controlled ITS, and allocation was not random and not concealed
Similarity of baseline outcome measurements (selection bias)	Low risk	For the outcomes of interest to us, baseline trends were similar for Bolivia and Peru (see table 4 of the study's primary report)
Similarity of other baseline characteristics (selection and performance bias)	High risk	As discussed by the study authors, Peru and Bolivia differed in a number of relevant ways, which may have influenced investment, production, trade, marketing and consumption patterns, including population size, GDP per capita, general trade and FDI intensity, and other demographic, cultural, political and economic factors. The analysis did adjust for GDP and population growth, but not for other factors
Incomplete outcome data (attrition bias) All outcomes	Low risk	While the commercially available Euromonitor sales figures are not perfect, they are generally considered reliable, and there is no evidence that problems regarding incomplete outcome data affected the 2 countries differently
Blinding (performance and detection bias) Objective outcomes	Low risk	The outcome data (Euromonitor sales data) can be considered objective
Contamination (performance bias)	Unclear risk	While trade and investment rules did evolve in Bolivia, too, no major free trade agreement with the US was concluded at the time of the study. The study does, however, note that the US‐Peru FTA may also have affected Bolivia Quote: "Interestingly we also observed an increase, albeit small, in soft drink imports into Bolivia from Peru following enforcement, suggesting the US‐Peru FTA may have had regional effects on soft drink sales"
Selective reporting (reporting bias)	Low risk	The study has not been registered and there is no protocol available; however, given that some of the authors have taken in past publications a critical stance towards trade liberalisation, and that they found in this study positive effects strengthens confidence that outcomes were not reported selectively
Independence of the intervention from other changes (performance bias)	High risk	Study authors note that a number of factors not attributable to the intervention may have affected outcomes. Quote: "[I]t is important to note that the study design makes it difficult to draw inferences and the results may simply reflect changes in the strategic decisions made by soft‐drink manufacturers in response to consumer demand for different products rather than changes in FDI in‐flows and production resulting from the FTA. Others have also acknowledged the difficulty of separating the effects of trade liberalization from other social and economic influences on food markets and consumer behaviour"
Pre‐specification of the intervention effect (detection bias)	Low risk	Study authors discuss the difficulties of establishing a clear time point when the intervention occurred, and report results both for the time point when the free trade agreement was ratified, and for the time point when it was enforced. In both cases, the point of analysis is the point of intervention. Quotes: "Another challenge in using the natural experiment design in this analysis was to determine where to introduce the intervention time‐point. For example, although the US‐Peru FTA was ratified in 2006 and enforced in 2009, changes in soft‐drink markets may have occurred earlier when, for example, the Office of the US Trade Representative notified US Congress of the Government’s intention to initiate negotiations with Peru in November 2003, with negotiations beginning in May 2004. Because this may have enhanced US investor confidence in Peru (a so‐called ‘market signal’) trade and investment flows may have begun as early as 2004. The staggered timing of the agreement and associated processes (e.g. notification of intent to congress → negotiation → ratification → enforcement) made it difficult to define a single intervention time‐point. For these reasons we included two time‐points in our models: 2006 as the ratification year and 2009 as the enforcement year"
Intervention effects on data collection (detection bias)	Low risk	Data were collected routinely and independently of the intervention and the study. It is unlikely that the intervention affected data collection
Other bias	Low risk	No other concerns