Physiological testing is widely used in sports to simulate and estimate athletic performance in controlled conditions (1). These tests provide valuable insights into an athlete's physiological status, enabling the assessment of training interventions over time (1, 2) and helping optimize performance in competitive settings (3). For these tests to be practical and meaningful, their protocols must demonstrate both reliability and validity (4, 5). Reliability refers to the consistency or stability of the test outcomes, while validity indicates how accurately a test measures or predicts what it is designed to assess (1). In this context, competitive physical performance serves as the dependent (predicted) variable, while physiological test results function as the independent (predicting) variables.

Predicting alpine skiing performance using common physiological tests has proven challenging (6, 7). The complexity of the sport itself (8–10), coupled with difficulties in establishing reliable response variables (e.g., competitive performance), complicates the identification of valid and reliable tests for predicting success in competition. While several studies (11–14) have identified relationships between anthropometric and physiological test variables and ranked skiing performance, such as Fédération Internationale de Ski (FIS) rankings, many of these studies employed questionable statistical methods (e.g., using parametric instead of non-parametric statistics), undermining the validity of their findings (7).

Additionally, the FIS ranking system itself can be problematic as a performance indicator due to the algorithm used for calculating ranks, which may encourage tactical and opportunistic behavior (15). Other studies have employed alternative performance outcomes, such as time trials (which lack competitive elements) or the number of top-three finishes (dependents on participant strength in each race) (16–18). Each of these response variables has its own limitations and may not accurately reflect long-term competitive performance.

Without a reliable response variable, the selection of physiological tests, while valid in general, may not accurately predict competitive performance (1). This uncertainty can lead to misguided evaluations of training interventions and flawed decisions when selecting athletes for teams or competitions.

Reliable and valid physical tests are essential for optimizing athletic performance.

In various sports, efforts to quantify performance profiles reveal that elite athletes exhibit distinct key performance variables compared to their less skilled counterparts (19, 20). However, significant differences in performance variables are also observed among elite athletes, even those with similar rankings (21). These individual differences among elite athletes significantly affect their physiological and psychological capabilities, training requirements, and performance outcomes (22). Factors such as strength, endurance, mobility, coordination, confidence, and mental stamina can all influence an athlete's overall competitive success. Therefore, training plans and test protocols should be tailored to each athlete's unique needs and goals, depending on the sport and personal ambitions (22, 23).

Individualized training is generally considered essential for maximizing performance, preventing injuries, and maintaining long-term motivation and commitment (24).

However, when implementing physiological tests, clubs, sports organizations, and federations often use generalized protocols to assess athletes’ physiological status. For elite athletes, this approach is likely suboptimal and rarely provides a clear indication of which physiological qualities are critical for sport-specific performance at an individual level.

In the worst-case scenario, these tests could recommend a training regime that is counterproductive to the athlete's competitive abilities.

A significant knowledge gap exists in the identification of valid physical tests that can accurately predict alpine skiing performance and assess the efficacy of training interventions.

The necessity for reliable and valid test batteries in elite sports, along with the development of individualized prediction profiles based on generated results, is crucial.

Therefore, the primary objective of this study is to investigate the predictive power of anthropometric and physiological tests for sport-specific performance in world-class female alpine skiers over time.

A secondary objective is to address the challenges of predicting alpine skiing performance at a group level and to explore the potential for individualized profiling models to enhance the accuracy of performance predictions.

We hypothesis that multivariable statistical method will generate statistically significant, and valid models for prediction of competitive alpine skiing performance.

No multivariable model (OPLS) could accurately predict (i.e., all have low Q²) competitive performance over time in SL and GS among elite female alpine skiers, based on included (Table 1) physiological testing results (Figures 1A–D, 2A–D).

Our stated research hypothesis is thus rejected.

Generated OPLS loading plots (Figures 1B, 2B) indicate separated clustering of FIS points (ranking) and physiological testing results (i.e., X and Y are not located in the same area of the plot). Goodness of regression (R²) and goodness of prediction (Q²) for FIS points (ranking) displayed in the X/Y overview plots (Figures 1C, 2C) indicate a low predictive power (R² = 0.39–0.40, Q² = 0.21–0.15) for SL and GS, respectively. Cross-validations by permutation (Figures 1D, 2D) confirm the validity of generated models. In this context, a non-significant model can be valid as being non-significant.

Data from competitive performance and physiological testing results recorded over time can be used to generate valid models for individual skiers (Figure 3) with higher predictive power (R² = 0.88–0.99 and Q² = 0.64–0.96) compared to when using the same data on a group level. Individual variation in physiological testing results, with similar/unchanged FIS points (ranking) (partially indicated by Figures 1A, 2A), is the main reason for the lack of high predictive models (R²/Q² >0.6) on group level. To exemplify the unique individual profiles of equally top-ranked skiers, individual OPLS models for Slalom and Giant Slalom were generated and subsequently compared with the Shared and Unique Structures (SUS)-plot analysis (Figures 4A,B). Both SUS-plots indicate that each skier's competitive performance [FIS points (ranking)] for Slalom and Giant Slalom depends on individual, partially different physiological qualities.

The outcomes of this investigation reveal that the incorporated tests yielded valid OPLS models for both slalom (SL) and giant slalom (GS) (Figures 1C,D, 2C,D), but models failed to withstand validation. Specifically, the models exhibited low predictive power for competitive performance at a group level (Figures 1C, 2C). In other words, models are true only for the dataset in which they are trained, but not applicable to a novel dataset. These findings are partially discordant with our previous research, in which no valid models for the investigated disciplines could be generated (6, 7), yet they converge in that all models demonstrated low predictive power (i.e., low Q²).

Consequently, these results imply that the competitive performance of elite alpine skiers cannot be predicted at a group level, regardless of the physiological performance tests employed.

In contrast, when physiological testing results and competitive performance data are analyzed at an individual level, robust models with high predictive power for competitive performance can be generated (Figures 3A–D). These findings corroborate the notion that multiple physiological qualities (e.g., muscle strength, balance, anaerobic and aerobic capacity) contribute to the competitive performance of alpine skiers (11), but also highlight that the relative importance of these qualities varies significantly between individuals. In the present study, FVBT PP (W) and IAT were found to be crucial for skiers A and K, whereas Body Stature was only significant for Skier A. Hand Grip Strength was important for skiers F and H, while Body Stature was only relevant for Skier H and A (Figure 4A). Historically, physiological performance research in alpine skiing has primarily focused on distinguishing between athletes at different performance levels (an indirect measure of performance differences) (16 33–35) or attempted to correlate physiological variables with ranking data (18, 30) and time trials (16, 17). Some studies have also used dubious statistical methods without validation of results (16, 30).

Our results indicate that normative data and non-validated, generalized research results are of limited use for the absolute top athletes.

Overall, our present results may provide important implications when implementing physiological performance tests for alpine skiers. While our previous studies showed that the included tests were unable to generate valid predictive models, our new results suggest that the problem may lie in how data are analyzed. Specifically, the new tests implemented in this study generated valid OPLS models for both SL and GS but also had low predictive power for competitive performance on a group level. Therefore, rather than implementing new tests to generate models with higher predictive power, it may be necessary to re-evaluate the analytical techniques used in this context.

Undoubtedly, elite performance in alpine skiing is influenced by multiple interdependent physiological qualities and other factors, including skiing technique, equipment, and mental dimensions (8, 10, 36).

A more extensive test battery, encompassing all these factors, may improve the test battery's predictive power regarding individual performance, even when applied at a group level.

Accordingly, future studies should adopt a more comprehensive and holistic approach to testing, with the aim of creating reliable and valid tests for the prediction and evaluation of competitive performance in alpine skiing.

Of outermost importance is to validate models with independent data, not used in the original model.

Also, the same test battery should be administered repeatedly over time rather than conducting new tests each year, as this approach may facilitate individual profiling and enable the identification of discriminatory variables for each athlete.

Moreover, other analytical methods, such as machine learning (ML) and artificial intelligence (AI), may mitigate the challenges associated with predicting group-level alpine skiing performance by analyzing large amounts of data from various sources, including different physiological and performance metrics, as well as environmental and race-specific factors (37). As a result, these methods can identify patterns and relationships difficult even for human experts to detect, possibly providing a more accurate and comprehensive assessment of an athlete's performance potential. In addition, ML and AI can continuously learn and improve from new data, leading to more accurate predictions and personalized training recommendations for individual athletes as they progress (37). As such, these analytical methods have the potential to enhance the accuracy and validity of the predictive models used in alpine skiing, enabling coaches and athletes to make more informed decisions regarding training and competition strategies over time.

To our knowledge, this is the first study using multivariable statistical methods to create individual profiles of world-class alpine skiers. However, while the presented results may provide valuable insights regarding its potential, further studies are needed to validate the approach using data sets from other sports. As such, the generalizability of this method can hopefully be determined and, therefore, also help to gain a more comprehensive understanding of its ability to predict performance in competitive sports.

In future studies, a more holistic approach for performance in alpine skiing should be considered, possibly including other variables than physical tests. Also, a defined sport performance parameter should always be included.

The practical implications of these findings offer several important considerations for the development of performance testing and predictive modeling in competitive alpine skiing: