written by Jess G. Fiedorowicz, Editor-in-Chief, Journal of Psychosomatic Research, November 2019
The Editor’s Choice – Smoking into Depression
This quarter’s Editor’s Choice from the Journal of Psychosomatic Research continues a recent tradition of selecting longitudinal studies. Current smokers with type 2 diabetes mellitus are about twice as likely to be depressed . While this association is well-established, there is considerable debate as to whether smoking leads to depression, depression leads to smoking, or both. The selected paper by Rachel McGihon and colleagues discerns the temporal relationships between smoking, measured as cigarettes per day, and depression, measured by the 9-item Patient Health Questionnaire (PHQ-9), in a community sample who smoke cigarettes and have type 2 diabetes mellitus . Senior and corresponding author Norbert Schmitz provided the following concise summary for the team of Canadian investigators from Toronto, Ottawa, and Montréal:
It is unclear how the number of cigarettes smoked per day (CPD) and depressive symptoms are temporally related in persons with type 2 diabetes who smoke. Three theoretical models describe the direction of this association: 1) CPD predicts depressive symptoms, 2) depressive symptoms predict CPD, and 3) CPD and depressive symptoms have a bidirectional relation. In a longitudinal study, we used model fit statistics to compare the three models in 296 adult smokers with type 2 diabetes. We found that a primary smoking model, in which CPD precedes depressive symptoms, best explains the association between CPD and depressive symptoms over time.
In their introduction, the authors note that a direct comparison of the three theoretical models had been done only once before and in a general population sample. This analysis also supported what the authors refer to as a “primary smoking model” wherein nicotine dependence preceded depressive symptoms . The analysis of McGihon and colleagues adds another innovation by looking at the quantity of cigarettes in those who smoke. The authors collected information on cigarettes smoked per day and depressive symptoms by telephone at five time points (baseline and four annual follow-ups). From this, they created four path models, corresponding to the three theoretical models and a “stability model” that did not include lagged associations between smoking and depression. Each of these models were adjusted for several potential clinical and sociodemographic confounding variables. The best fitting model, based on Akaike’s Information Criterion, was the primary smoking model where cigarettes smoked per day predicted subsequent depressive symptoms. This model appeared strongly superior to both the “primary depression model” and the “stability model” and moderately superior to the bidirectional model.
Longitudinal analyses are critically needed to advance the field of psychosomatic medicine. We are ultimately interested in causal relations between variables, in this case, depression and smoking. Often, the clinically relevant questions that we seek answers to are not amenable to experimental designs. In the case of the authors study question, it would be unethical to randomize participants to a year of smoking cigarettes. Randomization to depression is similarly dubious and implausible. Thus, studies designed to allow us to discern temporality become essential. One important challenge to such study involves identifying an optimal intensity of follow-up. How frequently much each variable be assessed to discern temporal relationships? Ultimately, this depends on how long it takes a given exposure to influence outcome. The time window selected should reflect the timeline and mechanisms by which exposure plausibly leads to the outcome. If too much time passes between assessments, the variables may appear to co-occur and a distant time point may no longer be or less strongly be associated. This may be especially problematic for dynamic variables that are prone to change between sampled timepoints. If too little time passes between assessments, we may not observe an extant association. Further, in the case of bidirectional relationships, this exposure time window may not be overlapping. How well the frequency of follow-up assessments corresponds to the exposure time window, which unfortunately is often uncertain, is prudent to consider in interpreting a given study and a given body of literature.
Returning to the study of McGihon and colleagues, the duration of one year between assessments was not likely preselected based on hypothesized timelines for this research question. Rather, the frequency of assessment was limited to what was available in the Evaluation of Diabetes Treatment (EDIT) study. In this study, the number of cigarettes smoked per day was a stronger predictor of itself at each subsequent time point than depressive symptoms were. This suggests that depressive symptoms were more dynamic than daily cigarettes smoked. If the timeline by which depressive symptoms influence smoking was strongest on a scale of less than a year and depressive symptoms fluctuated, this could have biased the results away from a primary depression model. Thus, the primary depression model shouldn’t be discarded based on this study alone, particularly given the inconsistent results from previous longitudinal studies of smoking and depression or anxiety . Regardless, McGihon and colleagues provide compelling evidence for a dose-dependent relationship between the quantity of cigarettes smoked and subsequent depressive symptoms in type 2 diabetes mellitus. These findings are useful to inform future research and direct public health efforts toward smoking cessation.
References: L.E. Egede, D. Zheng, Independent factors associated with major depressive disorder in a national sample of individuals with diabetes, Diabetes Care 26 (1) (2003) 104–111.  R.E. McGihon, R.J. Burns, S.S. Deschênes, N. Schmitz, Longitudinal associations between number of cigarettes per day and depressive symptoms in adult smokers with type 2 diabetes: A path analysis approach, Journal of Psychosomatic Research 125 (2019) 109737.  J.M. Boden, D.M. Fergusson, L.J. Horwood, Cigarette smoking and depression: tests of causal linkages using a longitudinal birth cohort, British Journal of Psychiatry196 (6) (2010) 440–446.  M. Fluharty, A.E. Taylor, M. Grabski, M.R. Munafò, The association of cigarette smoking with depression and anxiety: a systematic review, Nicotine & Tobacco.Research. 19 (1) (2016) 3–13.
Rachel E. McGihon, Rachel J. Burns, Sonya S. Deschênes, Norbert Schmitz
Vol. 125, October 2019
Corresponding Author’s Commentary
It is unclear how the number of cigarettes smoked per day (CPD) and depressive symptoms are temporally related in persons with type 2 diabetes (T2D) who smoke. Three theoretical models describe the direction of this association: 1) CPD predicts depressive symptoms, 2) depressive symptoms predict CPD, and 3) CPD and depressive symptoms have a bidirectional relation. In a longitudinal study, we used model fit statistics to compare the three models in 296 adult smokers with T2D. We found that a primary smoking model, in which CPD precedes depressive symptoms, best explains the association between CPD and depressive symptoms over time.