Leung and colleagues (1) appreciate that “chronotype, that is, an individual's biological preference as a “morning,” “intermediate,” or “evening” person … is potentially an important effect modifier of the relationship between shift work and cancer.” Their investigation avoids the misconception to include chronotype as a potential confounder (2, 3). Yet, we wish to comment on the use of stratification and interaction terms to measure effects of shift work (5 operationalizations) modified by chronotype (3 preferences):
Splitting information yields many coefficients which jointly describe effects of shift work with low power to discriminate between hypotheses.
Multiplicative interactions are difficult to interpret because causal (biological) interactions are expressed on an additive scale.
We recommended defining one single time-dependent variable (i.e., chronodisruption CDi) that integrates both chronotype and working history of an individual i (3, 4). We define on the time scale t (e.g., age or calendar time) time points t0, t1 with birth of i ≤ t0 < t1 ≤ death of i:
where bi(t) is an indicator of the biological night of i at t [i.e., bi(t) = 1 if t is within a biological night of i and 0 otherwise] and wi(t) is an indicator of work of i at t [i.e., wi(t) = 1 if t is within a working period of i and 0 otherwise].
CDi(t0,t1) simplifies evaluation (one coefficient, increased power).
CDi(t0,t1) entails the “interaction” of bi(t) and wi(t).
CDi(t0,t1) is the sum of the lengths of pairwise disjoint time intervals. The union of these time intervals is the intersection of the union of all (disjoint) biological nights and the union of all (disjoint) working periods of individual i.
Sophistications include the following: (i) bi(t) can reflect a change in chronotype with time (e.g., changing from an intermediate to an earlier midpoint of sleep) and varying length of sleep periods. (ii) wi(t) avoids the narrow, reductionist, focus on night shifts, or shift work only [indeed, working at civil day time should be considered when assessing chronodisruption (5)] and wi(t) may consider further work-related times such as time spent travelling to and from work (3). (iii) the integral can be expanded to ∫ [t0,t1] fi(t)bi(t)wi(t) dt with fi(t) as an a-priori defined factor weighting bi(t)wi(t) appropriately; the factor fi(t) may be used to reflect, for example, the direction of shift work (backward-rotating vs. forward-rotating).
Such chronobiology-targeted analysis strategy may also guide how we collect chronotype and working time information, preferably in a time-dependent manner.
See the Response, p. 1405
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.