Introduction

Safeguarding the health and well-being of athletes is essential, not only to maintain their physical health but also to allow them to pursue their personal aspirations and athletic ambitions without disruption. Therefore, prevention of injury and illness is a foundational principle in modern sports medicine and the core objective of many research centers, including the International Olympic Committee’s (IOC) Health, Medicine, and Science Department1,2,3. In elite sports, the margins of victory are often very thin, and therefore, the ability to maintain optimal health has been proposed to be a key determinant of success. Compromised athlete health, the argument suggests, can reduce an athlete or team’s competitive potential and the broader performance of the sporting programs4,5,6,7. These effects can extend beyond sport itself, as athlete success is often tied to financial incentives, including medal bonuses, endorsement opportunities, and future career prospects. Therefore, even small changes in place finishing have the potential to impact long-term earning potential and post-competition opportunities.

The impact of injury and illness on performance is partly captured by the concept of injury burden, which has been proposed to convey the average amount of time lost per incident of injury or illness8. Time-loss injuries are typically more visible and straightforward in their impact on competition, as an athlete who is unable to compete is entirely removed from the event, and thus the probability of success is reduced to zero. However, in elite sport, even subclinical or non-time-loss conditions such as mild respiratory infections or muscle/tendon injuries may result in changes in endurance, biomechanical, or functional capacity that may ultimately have profound effects on performance outcomes. Despite these factors, athletes competing at pivotal events like the Olympic Games often choose to compete despite sustaining injuries and illnesses, driven by the reality that, for most athletes, such opportunities may not come again. Indeed, at the 2024 Paris Summer Olympic Games, only 13.2% of injuries and illnesses sustained by Team USA athletes were considered time-loss injuries9. This suggests that athletes at the highest level of sport frequently compete while physically compromised, posing the question of how these suboptimal conditions affect performance outcomes.

To date, no comprehensive study has quantified the impact of injuries or illnesses sustained during the Summer Olympic Games on performance outcomes. While robust surveillance systems now provide detailed data on incidence and burden of injury and illness during the Summer Olympic Games10,11,12,13, little is known about how these conditions influence actual performance beyond time loss in real-world, high-stakes settings. Addressing this gap is essential for translating epidemiological data into targeted prevention efforts that ensure athletes remain healthy and capable of performing at their best. The present study seeks to address this gap by examining the impact of injuries and illnesses sustained by Team USA athletes during the 2024 Paris Summer Olympic Games on sports performance. Specifically, we endeavored to answer two research questions: RQ1) What is the impact of sustaining an injury and/or illness on the probability of winning a medal at the 2024 Paris Summer Olympic Games?; and RQ2) What is the impact of sustaining an injury and/or illness on an athlete’s finishing place at the 2024 Paris Summer Olympic Games? For both research questions, we theorized that athletes who experienced injuries or illnesses during the competition would perform below their expected performance. We specifically hypothesized that injuries and illnesses would significantly decrease an athlete’s probability of winning a medal and would negatively affect an athlete’s percentile ranking.

Methods

Participants

All Team USA athletes competing at the Paris 2024 Summer Olympic Games were included in the present analysis (N = 635). In order to establish the requisite predicted performance (i.e., ex ante performance), especially for establishing percentile rank finish, performance data for all Paris 2024 Summer Olympic athletes were included in the analysis. However, because injury and illness data were available only for Team USA athletes, the final analysis was completed only for Team USA athletes. All Team USA athletes or legal guardians provided informed consent for their data to be used in retrospective analysis. This study was reviewed and approved by the Institutional Review Board at the University of North Carolina at Greensboro (IRB-FY22-218) and the University of Utah (IRB_00187889), and all procedures and methods were carried out in accordance with relevant guidelines. The datasets generated and analyzed during the current study are not publicly available due to concerns regarding athlete privacy, but are available from the corresponding author upon reasonable request.

Ex ante performance

The present study investigates the impact of injury and/or illness (independent variable) on athlete performance (dependent variable). Previous work has analyzed the unadjusted relationship between injuries and finishing place or points total4,5, but these analyses are limited by a number of confounding factors. For example, athlete or team resources can differ among competitors; fewer financial (or other) resources may be associated with both injury risk and health status and performance outcomes. Therefore, accurately estimating the true effect of injury and/or illness on performance requires addressing the fact that athletes differ in their ex ante probability of success. Stated another way, some athletes are inherently more likely to win than others, and this expected performance may, in some cases, also be associated with their risk of sustaining injury or illness. As such, expected performance functions as a potential confounding variable in this analysis and needs to be accounted for. To robustly quantify the impact of injury and/or illness, therefore, one must first seek to understand each athlete’s expected performance. To understand why controlling for expected performance is important, consider a model that examines only the relationship between injury or illness and final performance. Such a model estimates an association between health and performance but does not account for the possibility that athletes who perform better may also have a higher risk of injury, because they play more minutes, adopt more aggressive styles, or are more frequently targeted with tackles, for example. Conversely, athletes with lower expected performance might face greater injury or illness risk due to limited access to medical resources, recovery modalities, or other factors. In either case, the athlete’s expected performance level could theoretically influence both the likelihood of injury and illness as well as the eventual observed outcome, and thus confounds the effect of interest. Therefore, to more appropriately interpret the relationship between injury and illness and performance, it is important to consider the counterfactual scenario (i.e., what would the athlete’s performance have been had they not been injured or ill?).

To address this consideration, from July 16, 2024, through August 11, 2024, all available Paris 2024 Olympic Summer Games sports betting odds were recorded from a publicly facing website of a commercial betting company. These betting odds were used only for this strict academic and educational purpose, and were manually recorded into a locally stored spreadsheet. We chose to use the betting odds posted on July 25, 2024 (the day preceding the Opening Ceremony). For events that commenced prior to the Opening Ceremony (rugby sevens, football [soccer], and handball), betting odds were extracted from the day preceding the start of that competition. In cases where betting odds were not available on July 25, 2024, the earliest available date on which betting odds could be extracted for the respective event was used.

The rationale for using sports gambling odds to represent ex ante performance is intrinsic to the incentive structures of commercial for-profit sports betting companies (or “bookmakers”). Namely, there is a vast financial incentive for the sports bookmaker to provide the most accurate possible betting odds to customers. As such, sophisticated forecasting models are deployed to provide the most accurate projection of future athlete performance as possible. Further, intelligent sports gamblers can capitalize upon inefficiencies in these betting odds markets, and thus, the betting lines can be adjusted to reflect the wisdom of the sports betting public. As a result, betting markets synthesize expert modeling with real-time adjustments based on collective market behavior, and therefore, we believe that these betting odds represent the most clear and unbiased forecasts of expected athlete performance during the Paris 2024 Summer Olympic Games. Of note, the odds extracted from these websites contain the vigorish (or “vig”), or the commission charged by the bookmaker. Therefore, the sum of the probabilities across the same event does not equal 1. Without being privy to the degree of proportionality of the vig across the individual athlete odds, we did not attempt to correct the odds to be “fair” odds.

The extracted odds consisted of odds that an athlete would win a gold medal, and when available, the odds that an athlete would win a medal (i.e., finish first, second, or third). The American odds, commonly referred to as “betting lines” or “money lines”, were extracted and converted to the probability of the outcome (Eq. 1) and, for the purposes of this manuscript, will be referred to as market-derived probabilities (MDP).

$$\:Market\:Derived\:Probability=\:\left\{\begin{array}{c}\frac{\left|Odds\right|}{\left|Odds\right|+100},\:if\:Odds<0\\\:\frac{100}{\left|Odds\right|+100},\:if\:Odds>0\end{array}\right.$$

Equation 1. Formula for converting American betting odds to implied market-derived probabilities.

Injury and illness surveillance

Team USA medical providers at the Paris 2024 Summer Olympic Games utilized the United States Olympic & Paralympic Committee’s (USOPC) Injury and Illness Surveillance system14 to record all injuries and illnesses at the Games, including an 11 day pre-competition period, the competition period, and post-Games period (July 15, 2024 – July 26, 2024, July 27 to August 11, 2024, and August 12 to August 15, 2024, respectively). This system, described in detail elsewhere14, required Team USA medical providers to record all injuries and illnesses (both time-loss and non-time-loss) that result in a clinical diagnosis, including the date of onset, diagnostic code, and other relevant details specific to the injury or illness. To ensure completeness, the USOPC Sports Medicine research team conducted daily reviews of surveillance entries and cross-referenced them against the electronic medical record. Following the Games, all injuries and illnesses were linked with athlete-event-level ex ante MDP data.

Observed results

The actual outcomes from the Paris 2024 Summer Olympic Games were obtained for all athletes across all sports and events from the Olympic World Library Results Books15. Each result (medal outcome for RQ1 for Team USA athletes, and rank finish for RQ2 for all athletes at the 2024 Paris Games) was manually extracted from the available sport-specific files, linked by athlete name and event with the MDP and injury and illness data to construct the final analytic dataset.

Analytical approach

Research Question 1: What is the impact of sustaining an injury and/or illness on the probability of winning a medal at the Paris 2024 Summer Olympic Games?

This analysis sought to quantify the extent to which sustaining an injury and/or illness during the Paris 2024 Summer Olympic Games influenced an athlete’s probability of winning a medal. To do so, we leveraged market-derived probabilities (MDP) as an ex ante estimate of medal probability. While MDP values were consistently available for the probability of winning a gold medal (MDP(Gold)), missing data were observed for MDP estimates concerning the larger probability of medaling (i.e., placing first, second, or third; MDP(Medal)). Owing to the absence of a fixed or proportional relationship between MDP(Gold) and MDP(Medal), a random forest regression model (500 decision trees) was employed to predict MDP(Medal) based on observed MDP(Gold) values, conditioned on sport. The model was trained on complete cases and applied to impute missing MDP(Medal) values for subsequent analyses (Supplemental Fig. 1). Our analytic approach for this research question then proceeded in two stages: (1) validating the accuracy of MDP(Medal) values, and (2) estimating the effect of injury and/or illness on medal outcomes, adjusting for each athlete’s expected probability of success.

To evaluate the predictive validity of the MDP(Medal), the dataset was randomly partitioned into a 70% training set and a 30% test set. The 70%/30% split was chosen over other alternatives (e.g., 80%/20%) to permit a sufficient number of observed medals in the test set. A binomial generalized linear model (GLM) was fit to the training data, regressing observed medal outcomes (binary: medal vs. no medal) on the imputed or observed MDP(Medal). In addition, to account for potential non-linear associations between MDP(Medal) and observed medal outcomes, a random forest (RF) classification model was fit using the same 70%/30% train-test split. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve metric was used to evaluate the predictive accuracy of the RF and GLM models in predicting medal attainment, computed on the hold-out test data. As a final validation step, the distribution of the percentage of actual medals won was visually compared against the distribution of MDP. Under the assumption that the sum of the MDP(Medal) probabilities across all Team USA athletes should approximate the total number of medals won, we aggregated the MDPs into quartiles. The proportion of athletes in each quartile who actually won medals was compared to the expected value implied by their MDP(Medal) quartile. For instance, if 20 athletes each had an MDP(Medal) of 50%, the expected number of medals won would be approximately 10. Similarly, for athletes with MDP(Medal) between 50% and 75%, the expected number of medals would range from 50% to 75% of the athletes in that quartile, or equivalently, between [N ⋅ 0.5] and [N ⋅ 0.75] medals, where N represents the number of athletes within the quartile. This quartile-based comparison allowed visual assessment of the alignment between predicted and observed outcomes, providing another evaluation of the accuracy of the medal prediction model.

Upon establishing the validity of the MDP(Medal) estimates, a series of logistic regression models were fit to test the effect of injury and/or illness on medal probability, while adjusting for each athlete’s MDP(Medal). The primary model (Model 1.1) included an interaction term between MDP(Medal) and a binary variable indicating the presence or absence of an injury or illness. Two additional models were fit to examine the independent effects of injuries (Model 1.2) and illnesses (Model 1.3), respectively. All models had an applied Firth correction to account for the small number of cases (injuries and illnesses) relative to the whole Team USA delegation.

All statistical analyses were conducted using R Statistical Software16. An alpha level of p < 0.05 was set for all hypothesis tests, with Bonferroni correction applied to control the experiment-wise error rate across the three models under consideration.

Exploratory sub-analysis

Recognizing that the physiological and performance consequences of injury and illness are unlikely to be uniform across all medical conditions, this analysis was designed to systematically evaluate the extent to which distinct combinations of injury and illness types influence medaling probability at the Paris 2024 Summer Olympic Games. Specifically, our working assumption was that some health conditions, or combinations thereof, exert a more profound negative effect on an athlete’s probability of medaling than others. Although a pre-planned analysis, this analysis constituted a purely exploratory approach, and thus, for this reason, while we describe our methods here, the full results are included only in an Online Supplemental File.

The analysis employed the same athlete-level dataset used in RQ1, but was adjusted to focus on all observed unique combinations of injury and illness types. Since individual injury and illness types may result in additive effects on performance, we modeled the impact of all possible combinations to identify those with disproportionately negative consequences on performance. Injuries were classified by etiology, and illnesses by affected medical system, in accordance with the Orchard Sports Injury and Illness Classification System (OSIICS, version 1517). A total of 26 distinct injury etiologies and illness system classifications were identified within the dataset (Supplemental Table 1).

The total number of possible non-empty combinations was calculated by 2k − 1, where k represents the number of unique injury and illness categories. For k = 26, this yielded 67,108,863 distinct combinations. For each combination, a logistic regression model with Firth correction was fit to evaluate the interaction between the presence of the specified combination and MDP(Medal) in predicting medal outcome. From each model, the following parameters were extracted: the combination tested, the number of athletes who sustained an injury or illness within the relevant combination, and the model coefficient for the interaction term and its associated p-value. Given the computational burden of fitting over 67 million models, batch processing was employed to optimize efficiency, with each batch containing a maximum of 4 million combinations.

Upon completion of model fitting and summary extraction, the results were collated and further analyzed to minimize spurious inferences arising from multiple statistical inference testing. To be considered further, combinations were required to (1) demonstrate a negative interaction coefficient (β3 < 0), meaning the direction of effect was consistent with theoretical expectations of injury or illness reducing the probability of medaling; and (2) include at least 20 injury and/or illness observations. We then applied a Benjamin and Hochberg p-value adjustment (padj) to control the false discovery rate (FDR)18. We acknowledge that applying this correction to only the subset of models that met theoretically plausible constraints (i.e., direction of effect and number of injuries and illnesses included) is not a convential application of the FDR procedure, and it would be more statistically sound to have applied the FDR across all ~ 67 million models. However, by doing so, we only reduced the number of models to approximately 37.8 million, which, while potentially leading to the possibility of a false discovery, allowed us to retain statistical power for detecting large effects while ensuring computational feasibility. Thus, caution is still warranted when interpreting any padj < 0.05. All analyses and data processing were conducted using Python and R Statistical Software16.

Research Question 2: What is the impact of sustaining an injury and/or illness on an athlete’s finishing place at the Paris 2024 Summer Olympic Games?

A key limitation of the analyses undertaken in RQ1 and the exploratory sub-analysis lies in the binary nature of the outcome variable, namely, whether or not an athlete won a medal. Given that only a small proportion of athletes win medals, and even fewer of those experience an injury and/or illness, the binary nature of the RQ1 and sub-analysis outcomes reduces statistical power and may obscure more subtle relationships between injury, illness, and competitive outcomes. To address this, RQ2 explored the relationship between injury and illness status and an athlete’s final event ranking (converted to a percentile finish), rather than a binary medal/no medal outcome. The primary predictor of expected performance remained the MDP values described above. However, rank-based analysis necessitates knowledge of not only the MDP values for Team USA athletes but also for all competitors at the Paris 2024 Summer Olympic Games. Therefore, MDP data were systematically extracted and transcribed for all available sports, events, and athletes. For athletes without an available MDP, we imputed expected performance by assigning the mean MDP of all athletes who finished in the same rank position, thereby preserving the structure of rank-based performance expectations.

To normalize across events with different numbers of competitors (e.g., a rank place finish of 12 is meaningfully different in a field of 12 athletes as opposed to a field of 80 competitors), all rankings were converted to a percentile finish, where finishing at the 100th percentile was first place (i.e. gold medal), and the 0th percentile was last place, to create a quasi-continuous metric. After the MDP-informed predicted percentile finish was established, all observed final rankings were extracted from the same Olympic World Library database and aligned with each event and athlete, and converted to percentile finish. Because we have access only to the USOPC Injury and Illness Surveillance system, we then merged injury and illness data and limited the dataset to only Team USA athletes for analysis.

A general linear model was fit to the data, with observed percentile as the dependent variable. The independent variables included the MDP-derived ex ante percentile, a binary indicator for injury or illness status, and the interaction between these two variables. Model diagnostics suggested the presence of heteroscedasticity, particularly at the lower performance range, where model predictions appeared less precise. To address this issue, robust standard errors were estimated using the heteroscedasticity-consistent covariance matrix, calculated via the vcovHC function from the sandwich package19. Model diagnostics also suggested small deviations from normality, and thus the rlm() function from the MASS package20 was used to refit the model using Huber’s M estimation to be more resistant to non-normality of residuals.

Additionally, because the same athlete could compete in multiple events, the assumption of residual independence was likely violated. Although the primary model allowed only a single observation per athlete per event (thereby limiting the effect of multiple injuries within the same event), athletes with multiple event entries appeared more than once in the dataset. To account for the non-independence of observations within athletes and thus robustly interrogate these results, hierarchical linear models were fit using random intercepts for each athlete, allowing for intra-individual correlation of residuals using the lme4 package21, to confirm findings. All statistical analyses were completed in R Statistical Software, and the alpha level was set at p < 0.05 for inferential tests.

Results

Team USA athletes (N = 635, 53.2% female) sustained 174 injuries and 99 illnesses during the Paris 2024 Summer Olympic Games. The team earned 40 gold, 44 silver, and 42 bronze medals, with a total of 257 individual medalists.

Research question 1

What is the impact of sustaining an injury and/or illness on the probability of winning a medal at the Paris 2024 Summer Olympic Games?

Ex ante MDP was a strong predictor of an athlete’s likelihood of medaling. The AUC for the ROC curve was 0.801 for the GLM and 0.881 for the RF model, indicating strong model performance, with MDP correctly predicting medal outcomes in 80.1% (GLM) to 88.1% (RF) of cases (Fig. 1).

Fig. 1
figure 1

Receiver operating characteristic (ROC) curves illustrating the predictive accuracy of the generalized linear model (A) and random forest (B) estimating the likelihood of winning a medal based on market-derived medal probabilities.

Full size image

An analysis of medal distribution across quartiles of ex ante MDP further supported the predictive validity of the model. Medal counts generally aligned with model expectations across MDP quartiles, with a slight overperformance observed in the second quartile (25th to 50th percentile), where Team USA athletes won 51% of medals (Fig. 2).

Fig. 2
figure 2

Observed medal-winning percentage across market-derived probability quartiles.

Full size image

The generalized linear models showed no statistically significant interaction between injury or illness status and MDP(Medal) (interaction term β3 = 1.00, 95% CI: 0.98 to 1.10, p = 0.945; Fig. 3). When modeled independently, neither injuries (β3 = 0.99, 95% CI: 0.97 to 1.01, p = 0.253) nor illnesses (β3 = 1.01, 95% CI: 0.99 to 1.03, p = 0.337) were significantly associated with medal outcomes.

Fig. 3
figure 3

Interaction between market-derived medal probability and injury or illness on medal outcomes. Shaded regions represent 95% confidence intervals.

Full size image

Exploratory sub-analyses

A total of 67,108,863 logistic regression models were fit to the data, encompassing all non-zero combinations of injury and illness types. The results for this sub-analysis can be found in Supplemental Material 1.

Research question 2

What is the impact of sustaining an injury and/or illness on an athlete’s finishing place at the Paris 2024 Summer Olympic Games?

To accurately generate ex ante rank percentiles for all events, interpolation was applied to 14.3% of MDP estimates. This approach enabled the inclusion of 29 additional Team USA athletes, including 6 who had sustained injuries or illnesses. After carefully deliberating this trade-off, all subsequent analyses were conducted using the interpolated dataset. Observed rank percentiles were significantly associated with predicted rank percentiles (β = 0.59, p < 0.001, R2 = 0.350; Fig. 4). The inclusion of the interpolated values reduced the explained variance between ex ante and observed percentiles from 39.5% to 35.0%, suggesting the interpolated values slightly reduced the accuracy of predicted percentile finish.

Fig. 4
figure 4

Comparison of observed and predicted rank percentiles after interpolation of missing data. The dashed line indicates perfect agreement between observed and predicted percentiles (line of identity) and the black solid line is the fitted general linear regression line.

Full size image

The general linear model results (Table 1) indicated that sustaining an injury and/or illness was associated with a significant decline in rank percentile, and this effect varied by predicted performance level (Fig. 5).

Table 1 General linear model results testing the impact of injury and illness on percentile rank finish.
Full size table
Fig. 5
figure 5

The effect of sustaining an injury or illness on the relationship between predicted and observed percentiles. The dashed line represents perfect prediction (line of identity). Figure annotations denote what would be considered under- and overperforming the ex ante prediction.

Full size image

Given evidence of heteroscedasticity in the residuals (see Supplemental Fig. 2 for fitted vs. residual plots), model coefficients were re-estimated using a heteroscedasticity-consistent covariance matrix. The adjusted estimates confirmed the significance of both the main effect (β2 = −18.02, 95%CI = −32.1 to −3.94, p = 0.012) and the effect for the interaction between expected percentile rank and presence of injury or illness (β3 = 0.20, 95%CI = 0.03 to 0.37, p = 0.022). Likewise, after diagnostic plot suggested non-normaily of residuals (see Supplemental Fig. 3 for histogram and QQ-plots), when using robust linear regression, the interpretation of the model remains consistent for both the main (β2 = −21.61, 95%CI = −31.88 to −11.33, p < 0.001) and interaction (β3 = 0.24, 95%CI = 0.111 to 0.38, p < 0.001) effects.

Discussion

The present study introduces a novel approach to linking injury and illness epidemiology with sports performance. To the authors’ knowledge, this approach to associating injury and illness with performance has not previously been attempted in Olympic-level athletes, despite its critical importance for several reasons. First, it provides empirical evidence supporting the value of injury and illness prevention strategies to athletes, National Olympic Committees, National and International Federations, and the broader athlete-support community. Injury and illness burden can now be communicated to these stakeholders in terms beyond time-loss metrics, allowing for a more accurate articulation of the financial and performance-related incentives for decision-makers. Second, although the current analyses are descriptive in nature and do not permit causal inferences regarding the impact of health status on performance, they nonetheless demonstrate an association between maintaining health and achieving better outcomes. Highlighting this relationship may serve as a powerful behavior-change mechanism for medical personnel aiming to implement injury and illness prevention programs. Finally, the methods presented here (namely the use of MDP to adjust for the confounds of expected performance in elite sport) offer a foundation for more ambitious future projects and provide a roadmap for identifying meaningful outcomes for injury and illness prevention initiatives beyond simple incidence and prevalence metrics.

In answering our research questions, we failed to find evidentiary support for our first hypothesis or exploratory sub-analyses (that injury and/or illness would impact medal outcomes and that specific injuries and illnesses would significantly impact medal outcomes). However, we did observe that injuries and illnesses were associated with an athlete’s percentile finish. When interpreting the significant interaction effects from the model in RQ2, one can observe that those athletes expected to finish first (100th percentile), their actual performance was on average at the 87th percentile, increasing to the 89th percentile if they were injured or ill (albeit this is basically no effect). However, injury and illness drops performance from 73rd to 70th percentile for those expected to finish at the 75th percentile, from 58th to 50th percentile for those expected to finish at the 50th percentile (median Team USA athlete), and from the 44th to the 31 st percentile for athletes expected to finish at the 25th percentile. In short, injury and illness resulted in worse performance for Team USA athletes, and the drop in performance was smallest among the most elite athletes and largest among those with worse expected finishes, perhaps reflecting the narrower margins for error in this latter group. Stated in another way, for the athlete expected to finish in the median position (e.g. 40th out of 80 athletes), by not sustaining an injury, their actual performance is expected to be, on average, at the 58th percentile, or approximately 33rd – 34th, a gain of 6–7 places.

Previous work has suggested that injury and illness have performance consequences, but primarily through the function of the athlete being unavailable for training or competition4,5,6,7,22,23,24. It is likely from this notion of injuries resulting in time-loss that the modern concept of injury burden metric was introduced, suggesting that the burden of an injury can be captured by the product of how often the injury occurs and the resulting time loss8. We have previously explored alternative models of injury burden that attempt to account for the holistic impact that sports injuries can have on an athlete25,26. In support of these alternative conceptualizations of injury burden, the present work supports the need for the burden of non-time-loss injuries to be considered due simply to the performance implications. Despite the exploratory sub-analyses not identifying statistically significant clusters of injuries and illnesses that have a disproportionate association with performance outcomes, these methods could be replicated on larger sample sizes and thus identify injury and illness types that have the greatest performance burden.

Limitations

From a theoretical perspective, the ex ante MDP used here assumes that information regarding an athlete’s health status is not included in the forecast. This is achieved to some degree by using odds that were set prior to the study period, and recorded injuries. However, it does not preclude these MDPs from embedding some degree of injury or illness prediction, even if on a population level (e.g., some percentage of athletes stratified across expected outcomes are expected to sustain a performance-impacting injury or illness). The present analyses assume that these MDPs are health-status naive, but since we are using MDPs set by an external party, we are unable to confirm this.

A major limitation of this analysis is that we did not put time restrictions on when the injuries and illnesses occurred during the study period. While we strongly considered doing so, we ultimately chose to time-bound the epidemiological data to the observed Games period for a number of reasons. Firstly, we conceptualized it is as unlikely that athletes would sustain new sport-related injuries after completing their event competitions, although post-competition acquisition of infectious illnesses is plausible, particularly given relaxed prevention behaviors once competition concludes. More consequentially, however, we chose not to apply timing restrictions primarily due to the uncertainty around the temporal precision of the surveillance system, which depends on accurate clinician documentation and timely athlete reporting. While we are confident that the system mostly captures what occurred during the Games, it is less clear that it is accurate in capturing when the injury or illness occurred. This is partly because it relies on accurate data entry by clinicians, but more so relies on accurate observation or reporting of the injury or illness by athletes. In high-stakes environments like the Olympic Games, athletes may delay reporting symptoms to avoid jeopardizing their eligibility to compete. Thus, we can not reliably say whether the injury or illness truly occurred at the reported time. Moreover, illnesses may be contracted and impact performance even before conscious acknowledgement of symptoms. In looking forward to future analyses, it is important to consider that the time from which to anchor the injury and illness observation is highly variable. For example, if two tennis players contract an illness on the same day, but one athlete has already been eliminated and the other is still competing, the impact of that illness on performance differs considerably. One is affected after competition, while the other may experience performance impairments before or during their next match. Additionally, if an athlete recovers from an illness but later sustains an injury during a critical round, the sequence and timing of health events become even more complex. These scenarios highlight the importance of considering each athlete’s schedule, sport, and event structure when linking health conditions to performance outcomes. Future work should strive to construct a more temporally accurate replication of these analyses, restricting to only injuries and illnesses known to have occurred prior to the rank-affecting competition for a given athlete within a specific event. Further, the observation period for this study did not include the full preparatory period prior to the Games. This is clearly an important window of time for injuries and illness that may affect subsequent performance. Unfortunately, the surveillance system deployed for Team USA is not able, at this time, to be broadly extended to individual athletes or National Governing Bodies outside of the Games period. As a final consideration, the present analyses considered only physical injuries and illnesses; the epidemiology of athlete mental health is a critical and necessary addition to these and other sports medicine efforts27. Mental health surveillance systems should continue to be developed and deployed to permit the analysis of the impact of psychological factors on performance outcomes.

In addition, we imputed MDP for athletes for whom no MDP was available. This unavailability may have been due to the lack of a market for that bet or lack of available data for the bookmaker to set an accurate betting line. When considering imputation, we considered a uniform value (e.g. 1% probability) or using the lowest set probability for that event for all remaining missing values for that event. However, since these options do not consider the possibility that if data were available and if a sufficient market was present, the MDP may be substantially greater, we instead chose to impute by using the mean MDP stratified by rank finish of other athletes in the population. By doing so, we are in some sense reinforcing the accuracy of the MDP – following imputation, the explained variance dropped by ~ 4.5%, suggesting that the input MDP was less accurate than the observed MDP alone. Nonetheless, this method permitted the inclusion of 29 Team USA athletes in the analysis that would otherwise be dropped, and permitted a more reasonable calculation of the percentile finish. Future work should consider how missing MDP could be better handled (including bespoke solutions for deriving the initial MDP to avoid missingness). Additionally, robust MDPs were not available for Paralympic athletes, and thus could not be included in the present analysis. We suggest that these analyses be replicated in a Paralympic cohort should the MDP become more readily available.

An additional limitation relates to how team-based and individual sports were handled in this analysis. In team settings (e.g., football or field hockey) injury or illness to one athlete may not have the same performance consequences as in an individual sport, due to player substitutions, replacements, or redistribution of effort among teammates. However, replacing athletes is rarely performance-neutral, and the overall win probability of a team may still be affected by the health status of a key player. In some Olympic events (e.g., synchronized diving, mixed archery, or relays), athletes compete in both individual and team or partner formats, further blurring this distinction. Ideally, understanding the individual-level effects of injury/illness on performance in team contexts would involve quantifying the counterfactual contribution of a healthy athlete to the team’s result; unfortunately, no validated methods currently exist for this across all sports. For these reasons, we included all athletes regardless of sport or event type and assigned performance outcomes at the event level. Future work with larger and more detailed datasets should investigate how the effect of injury and illness varies across sport formats and event structures, including the role of team size and athlete-specific importance to outcomes in team sports.

Finally, our analyses implicitly assume that percentile shifts among Team USA athletes are attributable solely to their own injury or illness status. In reality, competitors’ performances may also be influenced by similar health events. Thus, some observed changes in rank percentile may reflect improvements due to competitor setbacks rather than true overperformance within Team USA. We recognize this limitation and submit that our work should be treated as a proof-of-concept and an outline of a methodology that could be used examine these same effects across the entire Olympic athlete community at upcoming Games. To our understanding, the IOC injury and illness surveillance system currently only collects deidentified data, thus removing the possibility of aligning with ex ante expected performance data. However, the process outlined in this manuscript could be completed by each National Olympic Committee (of some minimal size) independently and have model summaries (with or without random noise-injected model coefficients) submitted to the IOC for collation using meta-analytical techniques. This would retain privacy for athletes whilst permitting a much larger number of injuries and illnesses to be included in the analyses and thus detect even marginal effects on performance.

It is worthwhile noting that while the observed association between injury and illness and performance was statistically significant, we acknowledge that the limitations and confounds listed here may bias the estimated effect. Given this, the model coefficients and effect sizes should be interpreted with caution, and emphasis should be placed not only on statistical significance but also on the plausibility and theoretical consistency of the findings. The persistence of the observed effect despite these limitations suggests a potentially meaningful signal. Future research with more robust covariate adjustment is needed to better isolate the causal effect. Further, future work may also consider investigating the impact of the severity of an injury in terms of time lost from sport, even if availability or team selection is not impacted.

Conclusion

Our analysis demonstrated a measurable and statistically significant association between injury and illness and deviations from expected performance outcomes among Team USA athletes at the Paris 2024 Summer Olympic Games. Notably, this effect was most pronounced among athletes whose pre-Games performance projections placed them in the lower percentiles of expected rankings, suggesting that health-related disruptions may be particularly detrimental for athletes who are, for example, competing for qualification to quarter-final rounds or similar. This insight presents a critical strategic opportunity for performance optimization: targeted injury and illness prevention initiatives may not only preserve baseline performance but may also improve the likelihood of converting marginal medal contenders into podium finishers by helping athletes maintain their predicted competitive capacities.

While Team USA will foreseeably continue fielding athletes with strong medal prospects regardless of health status, our data suggest that the slight gains achievable through preventive strategies are disproportionately concentrated among athletes whose projected performances cluster around or below the median. This finding carries particular relevance in anticipation of the Los Angeles 2028 Summer Olympic Games, where, as the host nation, Team USA will receive automatic qualification slots. In this setting, even modest improvements in health, or prevention of health detriments, among mid-tier qualifiers could yield meaningful gains in medal count, potentially influencing the nation’s overall standing on the Olympic medal table.