As time goes by

In clinical science, we take snapshots and then try to understand mechanisms or predict clinical outcomes from these single moments in time. It is not surprising that prediction models, for whatever disease, all share a far from perfect performance. It is as though we are trying to predict the adult face from one photograph taken in childhood by the school photographer. The same holds true for etiologic research. A putative cause of disease is measured at a certain time-point and associated with the subsequent occurrence of disease, but much may have happened in between. When the causative factor remains fixed (e.g. a genetic variant) time will have no direct effect, but such an intransient factor is invariably mediated through many processes that are influenced by other factors that do vary over time (i.e. environmental effects on protein expression). Most putative causes are not intransient and do change over time. Statistically, the effect of imprecision on the independent variables is called regression dilution and will lead to an underestimation of the true relationship of a putative cause with a certain outcome. One may argue that this imprecision is not just measurement error, but will always be present, because of changes that occurred during the time that elapsed between the measurement and the event, and the consequent difference between the measured factor and the true factor (e.g. the level of the protein and all its associated factors at the time of disease onset). When the variability of a certain factor (e.g. a clotting protein concentration) is known, the magnitude of the error due to this variability can be estimated. However, when we assume that virtually all diseases are the result of a complex interplay of genetic and environmental causes, many of which are still unknown, we are clueless about how wrong we may be. Because it will remain impossible to predict if and when someone will encounter a virus or consume a vegetable that affects clotting protein concentration, there is a natural limit to the associative capability. This is now all captured in the ‘random variation’, which is usually quantified in a standard error or confidence interval, beyond which there also must be true randomness. Molecules drift through the plasma, concentrations will vary in small ways between one place and another, and this may be decisive in whether a clotting reaction takes off or not. So, to become a bit more practical: why do individuals with a certain type of thrombophilia have their first thrombosis at a certain age, and some earlier, some later and some never? Why do patients with a clotting deficiency bleed on a specific day and not on another day, and why do they differ in the frequency of bleeding? Every prediction we make based on genetic defects and protein levels will be limited by events happening in between that may alter those levels or those of other proteins that play a role, or anatomical structures, such as the vessel wall or venous valves. To measure more is to know more. Ideally, we would measure all variables in all individuals continuously. This is unlikely to happen and therefore our predictions and associations will remain imperfect. Two articles in this issue of the Journal of Thrombosis and Haemostasis deal with time. In the first article, by Brunner-Ziegler and colleagues, volunteers took rivaroxaban, a direct oral anticoagulant with a rather short half-life, either in the evening or in the morning. Twelve hours after the evening intake, concentrations of rivaroxaban and the anticoagulant effect were more pronounced than 12 hours after the morning intake. Interestingly, a similar finding has been reported previously for aspirin intake. Time matters. In the second article, patients with cancer were followed during anti-neoplastic therapy with monthly blood draws as reported by Reitter and colleagues. Not only were the time trends in analytes such as factor VIII (FVIII), P-selectin and endogenous thrombin generation different between patients with various types of cancer and those with and without progression of the malignancy, but also in those who developed venous thrombosis: in particular, FVIII concentrations went up more over time than in those who did not develop thrombosis. Therefore this time trend will predict thrombosis better than a single baseline measurement. Although the theoretical ideal of measuring everything continuously will not be reached in the foreseeable future, these two studies present examples of integrating time and time trends in studies that may well be implemented in clinical care.