Investigating the Effect of Return-to-Play Timing After Injury on Performance: Does the Analysis Answer the Research Objective?

Return-to-play decision making should be based on all the advantages and disadvantages of return to play for athletes, not just the risk of injury. For competitive athletes, this includes the effect of early versus delayed return to sport on performance. In this paper, we address the questions "How can I estimate the effect of injury on the individual's performance at return to play?" and "What is the effect of delaying return to sport on the individual's performance?". To address these questions, we describe (1) some foundational concepts, design and analytical challenges related to estimating the causal effect of return to play timing on performance in the athlete, (2) additional challenges if one is interested in the effects of delaying return to play and (3) differences when the questions relate to the team's performance. Although the analytical strategies described appear complicated, coaches and athletes make these judgements informally every day without explicitly stating their assumptions. Using a formal approach should help analysts provide the most valid answers to the questions asked by athletes and coaches. In brief, the choice of a comparison group depends on the research question and requires that one consider the hypothetical performance trajectory of the athlete had they never been injured. Thus, the optimal comparison group depends on the shape of the expected trajectory and the specific research question being asked.

Implementing Multiple Imputation for Missing Data in Longitudinal Studies When Models are Not Feasible: An Example Using the Random Hot Deck Approach.

Purpose: Researchers often use model-based multiple imputation to handle missing at random data to minimize bias. However, constraints within the data may sometimes result in implausible values, making model-based imputation infeasible. In these contexts, we illustrate how random hot deck imputation can allow for plausible multiple imputation in longitudinal studies. Patients and Methods: Our motivating example is the Childhood Health, Activity, and Motor Performance School Study Denmark (CHAMPS-DK), a prospective cohort study that measured weekly sports participation for 1700 Danish schoolchildren. Using observed data on 4 variables (pain, activity frequency, sport, sport counts), we created a gold-standard data set without missing data. We then created a synthetic data set by setting some variable values to missing based on a prediction model that mimicked real-data missingness patterns. To create 5 imputed data sets, we matched each record with missing data to several fully observed records, generated probabilities from matched records, and sampled from these records based on the probability of each occurring. We assessed variability and agreement (kappa) between the imputed data sets and the gold-standard data set. We compare results to common model-based imputation methods. Results: Variability across data sets appeared reasonable. The range of kappa for the random hot deck approach was moderate for activity frequency (0.65 to 0.71) and sport (0.59 to 0.85), and poor for common model-based approaches (range 0.00 to 0.11). The range of kappas for sport count was strong (0.87 to 0.97) for random hot deck imputation and weak to moderate (0.55 to 0.71) for common model-based imputation. Agreement was higher when more information was present, and when prevalence was higher for our binary variable sport. Conclusion: Random hot deck imputation should be considered as an alternative method when model-based approaches are infeasible, specifically where there are constraints within and between covariates.

Causal effects, workload and injury risk: The importance of specifying the research question.

OBJECTIVES: To illustrate why the research question determines whether and how sport medicine investigators should adjust for workload when interested in interventions or causal risk factors for injury. DESIGN: Theoretical conceptualization. METHODS: We use current concepts of causal inference to demonstrate the advantages and disadvantages of adjusting for workload through different analytic approaches when evaluating causal effects on injury risk. RESULTS: When a risk factor of interest changes workload, including workload in the regression will cause bias. When workload represents time-at-risk (e.g. games played, minutes run), including workload as an offset in Poisson regression provides a comparison of injury rates (injuries per unit time). This is equivalent to including log(workload) as an independent variable with the coefficient fixed to 1. If workload is included as an independent variable instead of an offset, using log(workload) rather than workload is more consistent with theory. This practice is similar to the principles of allometric scaling. When workload represents a combination of both time-at-risk and intensity, such as with session ratings of perceived exertion, the optimal analytical strategy may require modeling time-at-risk and intensity separately rather than as one factor. CONCLUSIONS: Whether to account for recent workload or not, and how to account for recent workload, depends on the research question and the causal assumptions, both of which should be explicitly stated.

Predicting Injury Risk Over Changes in Physical Activity in Children Using the Acute:Chronic Workload Ratio.

Limited research exists on the relationship between changes in physical activity levels and injury in children. In this study, we investigated the prognostic relationship between changes in activity, measured by the acute:chronic workload ratio (ACWR), and injury in children. We used data from the Childhood Health, Activity, and Motor Performance School Study Denmark (2008-2014), a prospective cohort study of 1,660 children aged 6-17 years. We modeled the relationship between the uncoupled 5-week ACWR and injury, defined as patient-reported musculoskeletal pain, using generalized additive mixed models. These methods accounted for repeated measures, and they improved model fit and precision compared with previous studies that used logistic models. The prognostic model predicted an injury risk of approximately 3% between decreases in activity level of up to 60% and increases of up to 30%. Predicted risk was lower when activity decreased by more than 60% (minimum of 0.5% with no recreational activity). Predicted risk was higher when activity increased by more than 30% (4.5% with a 3-fold increase in activity). Girls were at significantly higher risk of injury than boys. We observed similar patterns but lower absolute risks when we restricted the outcome to clinician-diagnosed injury. Predicted increases in injury risk with increasing activity were much lower than those of previous studies carried out in adults.

Causal simulation experiments: Lessons from bias amplification.

Recent theoretical work in causal inference has explored an important class of variables which, when conditioned on, may further amplify existing unmeasured confounding bias (bias amplification). Despite this theoretical work, existing simulations of bias amplification in clinical settings have suggested bias amplification may not be as important in many practical cases as suggested in the theoretical literature. We resolve this tension by using tools from the semi-parametric regression literature leading to a general characterization in terms of the geometry of OLS estimators which allows us to extend current results to a larger class of DAGs, functional forms, and distributional assumptions. We further use these results to understand the limitations of current simulation approaches and to propose a new framework for performing causal simulation experiments to compare estimators. We then evaluate the challenges and benefits of extending this simulation approach to the context of a real clinical data set with a binary treatment, laying the groundwork for a principled approach to sensitivity analysis for bias amplification in the presence of unmeasured confounding.

Analyzing Activity and Injury: Lessons Learned from the Acute:Chronic Workload Ratio.

Injuries occur when an athlete performs a greater amount of activity than what their body can withstand. To maximize the positive effects of training while avoiding injuries, athletes and coaches need to determine safe activity levels. The International Olympic Committee has recommended using the acute:chronic workload ratio (ACWR) to monitor injury risk and has provided thresholds to minimize risk when designing training programs. However, there are several limitations to the ACWR and how it has been analyzed which impact the validity of current recommendations and should discourage its use. This review aims to discuss previously published and novel challenges with the ACWR, and strategies to improve current analytical methods. In the first part of this review, we discuss challenges inherent to the ACWR. We explain why using a ratio to represent changes in activity may not always be appropriate. We also show that using exponentially weighted moving averages to calculate the ACWR results in an initial load problem, and discuss their inapplicability to sports where athletes taper their activity. In the second part, we discuss challenges with how the ACWR has been implemented. We cover problems with discretization, sparse data, bias in injured athletes, unmeasured and time-varying confounding, and application to subsequent injuries. In the third part, conditional on well-conceived study design, we discuss alternative causal-inference based analytical strategies that may avoid major flaws in studies on changes in activity and injury occurrence.