A cohort study was performed from January 2014 to December 2016 in a Brazilian neonatal intensive care unit, including neonates with high risk for infection and death. We estimated bloodstream infection (BSI) incidence and conducted a survival analysis, considering the time to death and to the first episode of BSI as outcomes, comparing very low birth weight (VLBW) neonates with the remaining neonates. An extended Cox model was performed and the hazard ratio (HR) was calculated for different time periods. The study had 1560 neonates included, the incidence and the incidence density of BSI was 22% and 18.6 per 1000 central venous catheter-days, respectively. Considering VLBW neonates as the reference group, the HR for time to death was 4.06 (95% CI 2.75–6.00, P < 0.01) from day 0 to 60 and for time to the first episode of BSI was 1.76 (95% CI 1.31–2.36, P < 0.01) from day 0 to 36. Having the heavier neonates group as reference, the HR for time to the first episode of BSI was 2.94 (95% CI 1.92–4.34, P < 0.01) from day 37 to 90. Late-onset neonatal sepsis prevention measures should consider the differences in risk during time, according to neonates' birth weight.