Table 2: GRADE evidence profile for the effect of community-based suicide prevention programs, including screening for depression, on the incidence of suicide
Quality assessmentNo. of patients/person- yearsEffectQualityImportance
No. of studiesDesignRisk of biasInconsistencyIndirectnessImprecisionOther considerationsScreening
(older persons)
ControlRelative
(95% CI)
Absolute
Overall (follow-up 4–20 yr; assessed with community-based depression screening)
51721ObservationalNo serious risk of bias*No serious inconsistency†Very serious‡No serious imprecision§None¶65/70 053
(0.09%)
145/113324
(0.13%)
RRR 0.50 (0.32–0.78)1 fewer per 1000 (range 0 fewer to 1 fewer)Very lowCritical

Note: CI = confidence interval, RRR = ratio of rate ratios. 
*The quality assessment tools identified a few concerns (e.g., selection of non-exposed cohort, blinding and reporting of withdrawals and drop-outs); however, the evidence was not downgraded for these reasons.
†Heterogeneity statistics were not significant: τ2 = 0.05, χ2 = 5.04, df = 4 (p = 0.28); I2 = 2%.
‡Directness was downgraded because of concerns about population characteristics. The included papers all involved older populations in rural areas of Japan, which are unlikely to be representative of Canadians at average or high risk for depression. Directness was downgraded further because of concerns regarding community-based screening for depression: The studies included in the analysis evaluated the effectiveness of community-based programs to screen for depression, which incorporated screening for depression, follow-up with mental health care or psychiatric treatment, and health education in the community setting. As such, any observed reduction in suicide rates could not be attributed solely to the screening component of these programs. 
§The number of events was small (< 300, which is the threshold rule-of-thumb value for dichotomous outcomes); however, with regard to the specific outcome, the evidence was not downgraded.
¶Funnel plot of the comparison indicated potential asymmetry and thus potential publication bias. However, the number of papers (n = 5) was too small to assess publication bias with confidence (≥ 10 papers being the threshold rule-of-thumb value).