Selection criteria and generalizability within the counterfactual framework: explaining the paradox of antidepressant-induced suicidality?