White et al ^{Reference White, Zaninotto, Walters, Kivimäki, Demakakos and Biddulph1} examined the relationship between the duration of depressive symptoms and mortality in adults aged 50 or older in a follow-up study. The authors assessed depressive symptom duration as the sum of screen-positive number by an eight-item Center for Epidemiologic Studies Depression Scale (CES-D) score of ⩾3. Adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) of screen-positive depressive symptoms with numbers from 1 to 4 against never-reported symptoms for mortality were 1.41 (1.15–1.74), 1.80 (1.44–2.26), 1.97 (1.57–2.47) and 2.48 (1.90–3.23), respectively. The authors concluded that the duration of depressive symptoms was significantly associated with mortality in a dose–response manner. I have some concerns regarding their study.

First, the same authors recently investigated the association between depressive symptom severity and mortality. ^{Reference White, Zaninotto, Walters, Kivimäki, Demakakos and Shankar2} Depressive symptom severity was assessed by the positive number of an eight-item CES-D score. Adjusted HRs (95% CIs) of 2, 4 and 8 positive scores on CES-D for mortality were 1.59 (1.40–1.82), 1.80 (1.52–2.13) and 2.27 (1.69–3.04), respectively. I understand that mortality risk existed at low levels of depressive symptoms, but the authors concluded that and the risk became a plateau thereafter. The authors conducted the risk assessment of duration and severity of depressive symptoms for subsequent mortality in these two papers, and there is a need of additional explanation for the relationship with and without a dose–response manner.

Second, there are reports that the time (duration) of depression was associated with subsequent mortality in patients with acute coronary syndrome. ^{Reference Grace, Abbey, Kapral, Fang, Nolan and Stewart3,Reference Parker, Hyett, Walsh, Owen, Brotchie and Hadzi-Pavlovic4} These references handled different events, such as all-cause mortality and cardiac events. I recommend that White et al conduct a sensitivity analysis by dividing the cause of death such as neoplasms, cardiovascular disease and cerebrovascular disease (stroke).

Finally, the authors adopted Cox regression analysis, and evaluation of depressive symptoms by CES-D was repeatedly made. In this situation, I strongly recommend the authors conduct a time-dependent Cox model for their analysis, which had been reported. ^{Reference Fan, Sarnak, Tighiouart, Drew, Kantor and Lou5–Reference Wassertheil-Smoller, Applegate, Berge, Chang, Davis and Grimm7} Among them, Wassertheil-Smoller et al ^{Reference Wassertheil-Smoller, Applegate, Berge, Chang, Davis and Grimm7} reported that baseline depressive symptoms were not related to subsequent mortality, but Cox proportional hazards regression analyses with the CES-D scale as a time-dependent variable indicated a significant increase of death with increase in the CES-D score thereafter. As the significance of HR disappeared when all covariates were adjusted for the association between the number of screen-positive depressive symptoms and mortality, further study is needed to confirm the association.