Table 3.

Association between road traffic mortality and exposures, by model

		Model 1	Model 2	Model 3
Population measures
	Population density in built-up area, 2010	0·90 (0·86–0·94)	0·93 (0·89–0·98)	0·94 (0·90–0·98)
	Population growth 2010–16*	1·03 (0·99–1·07)	1·02 (1·00–1·06)	1·03 (1·00–1·06)
Social environment index		0·90 (0·85–0·94)	0·96 (0·91–1·02)	0·96 (0·91–1·02)
Gross domestic product per capita in 2010		0·96 (0·93–0·99)	0·96 (0·94–0·98)	0·96 (0·94–0·98)
Urban landscape
	Patch density (fragmentation)	0·91 (0·88–0·94)	0·99 (0·95–1·03)	..
	Area-weighted mean nearest neighbour (isolation)	1·08 (1·05–1·12)	1·05 (1·01–1·08)	1·05 (1·02–1·09)
	Proportion of city built up	0·89 (0·86–0·91)	0·99 (0·91–1·07)	..
Road network measures
	Intersection density	0·88 (0·86–0·91)	0·94 (0·87–1·02)	0·92 (0·89–0·95)
	Average street length (metres)	1·03 (1·00–1·07)	0·96 (0·92–1·00)	0·96 (0·92–1·00)
	Average streets per node	1·06 (1·02–1·09)	1·02 (0·99–1·05)	1·02 (0·99–1·05)
Transportation measures
	Urban travel delay index	0·89 (0·86–0·92)	0·98 (0·94–1·02)	0·98 (0·94–1·02)
	Presence of bus rapid transit system or subway (vs none)	0·77 (0·71–0·82)	0·93 (0·86–0·99)	0·92 (0·86–0·99)

Data are rate ratio (95% CI). All models are adjusted for a fixed effect for country, sex, and 5-year age group. Model 1 is a single exposure model of each exposure and covariate. Model 2 includes all exposures and covariates in a multivariable model. Model 3 includes all exposures and covariates except patch density and percent built-up area. Rate ratios were estimated using mixed-effects generalised linear model regression with negative binomial distribution and robust standard errors. All rate ratios and 95% CIs reflect a difference of 1 SD except for bus rapid transit system or subway.

^*Measured for 2010–14 in El Salvador and 2012–16 in Panama to match the years of outcome data available.