Toronto restricted access at one site and the effect held for seventeen years

In Toronto, access at one site changed on a fixed date. Construction of a barrier on the Bloor Viaduct ran from April 2002 to June 2003, and from July 2003 the structure stood. Before it, the viaduct accounted for about nine deaths a year – half of all bridge-related suicides in the city, and 4% of every suicide recorded in Toronto. Nobody was screened, scored or placed on a list for this to happen. Access closed on a calendar date, and the question Mark Sinyor and colleagues put seventeen years later was whether the city's count stayed down.

The coroner series, 1998 to 2020

Coroner records for the City of Toronto covering April 1998 to December 2020 hold 5,219 suicide deaths, 303 of them bridge-related. At the viaduct itself the split is 48 deaths in the five years before completion against 2 in the seventeen and a half years after.

The city-wide figure carries more weight, because a barrier that merely moved deaths to the next bridge would leave the city total untouched. Interrupted time-series Poisson regression, adjusted for season, Ontario unemployment, the Toronto consumer price index and the city's population, puts the step change at 49% in the quarter after completion (IRR 0.51, 95% CI 0.30 to 0.86). The crude model gives IRR 0.47 (0.30 to 0.75). Counts had been stable in the years immediately before, so there was no pre-existing decline for the barrier to ride.

Then it held. The post-intervention trend term is IRR 0.99 (95% CI 0.96 to 1.03) – a flat line across seventeen years.

Other bridges, other methods

The standard reason to set means restriction aside is that people move to another site or another method. This group had published that very finding themselves: their first analysis, covering four years after the barrier, showed a rise at other bridges. Their ten-year follow-up did not. The present analysis, at nearly two decades, measures no method substitution immediately (IRR 1.04, 95% CI 0.90 to 1.20) and none over the long run (IRR 1.00, 95% CI 0.99 to 1.01). The authors read their own early result as short-term fluctuation in small counts, plausibly driven by press coverage of the construction.

One number cuts against a clean reading. Ottawa and Hamilton were analysed as control cities, where the barrier could not have acted. Hamilton returned IRR 1.17 (95% CI 0.44 to 3.43). Ottawa returned IRR 0.50 (95% CI 0.26 to 1.01) – a trend toward the same reduction, in a city the structure never touched. It misses the significance threshold, and the authors list it in their own limitations as a sign that other confounders may be at work.

Prediction and place

Set this beside the arithmetic of risk prediction. To act on a person, a service must first identify that person, and at a base rate in the hundredths of a percent that identification is mostly wrong in both directions. The barrier required none of it. It generates no false positives because it makes no classification, and it ran for seventeen years without a clinician's hour, a follow-up call or a register.

The authors estimate more than 150 lives. That figure is an extrapolation rather than a count: roughly nine deaths a year multiplied by 17.5 years, resting on the assumption that a death prevented at the site was a death prevented altogether. An ecological natural experiment cannot establish with full confidence that the structure prevented suicides in Toronto, though the authors argue the pattern fits the Bradford Hill criteria.

For a service, the practical shape is unusual. This is capital spending with a completion date, not a clinical programme that consumes staff time and decays when attention moves elsewhere. Its evidence sits in coroner records rather than on a scale.