Table 3

Estimated average impact of ice cleat distribution programmes in 73 Swedish municipalities on snow and ice-related fall injuries among older adults (primary analysis) and fall injuries unrelated to snow and ice (negative control analysis)

Model and estimateEstimated impact per 1000 person-winters (95% CI)Impact in relative termsP value for impactP value from parallel pre-trends test*
Primary analysis0.143
 Intention to treat†−0.24 (−0.49, 0.02)−8.2%0.066
 Efficacy‡−0.38 (−0.76, −0.09)−12.5%<0.001
Negative control analysis§0.096
 Intention to treat†0.06 (−0.76, 0.79)0.3%0.881
 Efficacy‡0.02 (−0.62, 0.65)0.1%0.995
Synthetic control analysis¶N/A
 Intention to treat†−0.22 (−0.44, 0.00)−7.6%0.055
Survey respondents only**0.169
 Intention to treat†−0.27 (−0.53, −0.01)−9.2%0.043
 Efficacy‡−0.42 (−0.72, −0.11)−13.6%0.006
  • *A joint significance test (F-test) on time-specific preintervention impact estimates up to 10 winters before the intervention takes place. A significant test suggests a risk of bias. Not estimable in the synthetic control analysis.

  • †Average effectiveness estimate using a binary intervention variable. The estimate reflects the average expected impact under typical conditions. Based on data from all 73 programme municipalities.

  • ‡Average efficacy estimate estimated by scaling municipality-specific impact estimates by municipality-specific reach (ie, the number of ice cleat pairs distributed per eligible citizen). The estimate reflects the expected impact when one ice cleat pair distributed per citizen. For reference, the average programme municipality distributed 0.4 ice cleat pairs per eligible citizen (table 1). This estimate is based on data from the 66 programme municipalities within non-missing distribution data.

  • §The negative control outcome is the incidence of fall injuries unrelated to snow and ice per 1000 population (ICD-10 codes W01-W18), which should not be affected by ice cleat distribution. A significant impact suggests a risk of bias.

  • ¶Results from an analysis that is more robust potential violations of the parallel trend assumption (see online supplemental file 1 for details). Efficacy could not be estimated in this analysis.

  • **Results from an analysis that restricts the control group to non-programme municipalities that responded to our survey (n controls=148).

  • ICD, International Classification of Diseases, 10th revision; N/A, not available.