In elite football, injury risk is considerable. Consequently, optimal rehabilitation and an adequate time point of return-to-play (RTP) are crucial. The elevated injury risk in the weeks after RTP (e.g. Orchard & Best, 2002; Wangensteen et al., 2016) and the time course of biological healing processes (Orchard & Best, 2002) nourished the speculation that players generally return to early. While the previously reported negative association between rehab time and injury risk after RTP (e.g. Beischer et al., 2020; Della Villa et al., 2021; Orchard & Best, 2002) points in the same direction, the truism “Correlation does not imply causation.” must be considered. To the best of our knowledge, the causal effect of rehab time on injury risk has not been investigated so far. This work aims to contribute towards closing this gap in knowledge. Specifically, to scrutinize if a reduction in injury risk may be expected when RTP is delayed. Previous time-loss injuries and match exposures of players currently playing in the European top leagues (Germany, Austria, France, Spain, Italy, Netherlands, United Kingdom) were collected from public sources and episodes (index injury-RTP-subsequent injury) constructed. Data analysis was conducted in R. Considering the time dependence of the excess injury risk after RTP, we analyzed injury occurrence after RTP within a fixed time frame (28 days). A robust scaling approach was used for making rehab time comparable between index injury diagnoses. Exposure within the 28-day period was adjusted for eventual time loss due to subsequent injury. The association between rehab time and injury was analyzed using logistic regression. Odds ratios (OR) were calculated from regression coefficients and adjusted for baseline probability to avoid overestimation of relative risk (RR). Two approaches were used to delineate a potential causal effect: (a) sensitivity analysis for the putative confounder (using Cornfield’s inequality) and (b) path analysis which due to the simple causal diagram could be implemented by including exposure in the logistic regression model. Eight thousand six hundred and forty-four episodes could be included among which 1,245 (14 %) had a subsequent injury within 28 days. There was a highly significant negative association between scaled rehab time and injury risk (RR 0,896), which was more pronounced for recurrences (RR 0,815). Cornfield’s inequality indicates that exposure would have to be lower by at least ~10% for every interquartile range of rehab time of the respective index injury to account for the observed association. While this is substantial, early return is in fact closely associated with higher exposure and this association accounts entirely for the observed association between rehab time and injury risk (Figure 1). Note that Figure 1 also shows that high exposures are observed exclusively after an early return. We conclude that the negative association between rehab time and injury risk is largely accounted for by higher exposures after early return. However, following from the limitations associated with predicting outside the range of previously observed values, the virtual absence of high exposures after long rehab (for the respective diagnosis) impedes on predicting the effect of prolonging rehab time.
Read full abstract