Devising a Pace-Based Definition for “The Wall”: An Observational Analysis of Marathoners' Subjective Experiences of Fatigue

Participants were recruited at convenience during the Dublin Marathon Expo 2018 held October 26–27, 2018. Prospective participants in the KBC Dublin Marathon 2018, which was held on October 28, 2018, were required to attend this event to pick up their race numbers and race chips, which were needed to obtain an official time in the race. Floor space was rented in the preceding months, and members of the research team were stationed at a stand during the event. Attendees who approached the stand were informed of the researchers' intent to conduct a study of marathon running. Volunteers were not eligible for the study if they were <18 years of age. All participants provided written informed consent, and the study was approved by the Human Research Ethics Committee-Sciences of our university (reference: LS-17-77).

Participants were instructed to complete a short on-site questionnaire, documenting their race number, age, basic anthropometric measures (height and mass), training behavior, most recent race times (10 km, 10 mi, or half marathon), marathon experiences, and target finish times and provide their e-mail address. They were informed that their e-mail addresses would be used to send a postrace questionnaire related to their marathon experience. Runners with target times >4 hours and 30 minutes were deemed ineligible to exclude individuals with a walk-run marathon-pacing strategy.

Racers were sent the postrace questionnaire 48 hours after the marathon. The postrace questionnaire was used to ascertain symptoms of HTW, as identified by Buman et al,¹²  and delivered via a Google Form (Google LLC, Mountain View, CA) link (Table 1).

Participants were also instructed to provide a means to access, where available, any global positioning system (GPS) tracking data acquired via a smartphone or wearable device and subsequently logged in a self-monitoring platform, such as Strava (San Francisco, CA). Participants' race numbers were required to facilitate cross-checking of their in-race split times via the 2018 Dublin Marathon's Web site.¹⁴ 

The Wall phenomenon was not discussed with prospective participants. Only postrace responses recorded within 1 week of the race were eligible for analysis to minimize the potential effect of recall bias. Recruits were not compensated for taking part in the study.

The marathon started at 9:00 am on October 28, 2018. The temperature during the race was between 3°C (9:00 am) and 10°C (1:30 pm). Wind conditions were northerly at 11 to 17 km/h, and the relative humidity was 76% to 87% with scattered clouds.

We input data gathered from the prerace questionnaire into a spreadsheet to document each participant's actual finish time based on the race records; target time; and a predicted time, derived from race and training data, using the equations described by Vickers and Vertosick.¹⁵  The postrace questionnaire asked whether participants experienced HTW during the marathon. They were instructed to report whether they experienced any of the 4 symptoms: generalized fatigue, unintentionally slowing of pace, desire to walk, or shifting focus to survival.¹²  They were deemed to have experienced HTW if they reported any 3 of the 4 symptoms described by Buman et al.¹² 

We retrieved the times for all marathon finishers (10 672 men, 5537 women). Using their race numbers, we identified the subset of respondents to the postrace questionnaire and extracted their in-race split and overall finish times. The time in each split (10 km, 21.1 km, 30 km, and 42.2 km) was normalized to the distance covered so that each split could be compared over a distance of 10 km. Specifically, the time spent in segment 2 was divided by 1.10975 (21.0975 km – 10 km), in segment 3 was divided by 0.89025 (30 km − 21.0975 km), and in segment 4 was divided by 1.2195 (42.195 km − 30 km). Each split time was then divided by the associated finish time to facilitate pooling of participants with different finish times.

When self-monitored GPS tracking data were also available for each 1-km segment of the race, they were cross-referenced with the official in-race split times to ensure accuracy, and, if they matched, the 2 datasets were combined. The 1-km split times were adjusted for course gradient using the method described by Minetti et al.¹⁶ 

To maximize the utility of a pace-based threshold for HTW, we devised thresholds for split resolutions of 1-km, 5-km, 10-km, or first- versus second-half race times where available. To achieve this, in-race split data were discretized in the following manner:

• (1)

  by calculating the standard deviation (SD) in the split time (1-km, 5-km, 10-km, and first- versus second-half split resolutions separately);

• (2)

  by dividing the time to complete the slowest split by the average of the remaining splits (1-km, 5-km, and 10-km split resolutions separately);

• (3)

  by dividing the second-half race time by the first-half race time (first- versus second-half split resolution only).

The 10-km and 1-km split-time data before discretization are presented in Figure 1.

Respondents and nonrespondents to the questionnaire were compared using independent-samples t tests for each variable to assess representativeness. The α level for this analysis was set at .05.

The predictive capacity of discretized data in classifying HTW instances was quantified via logistic regression and a receiver operating characteristic (ROC) curve analysis. Accuracy was quantified via the C statistic, which measures the area under the ROC curve.¹⁷  The α level for this analysis was set a priori using a Bonferroni-adjusted correction for multiple comparisons at .0167 (0.05/3 [discretization methods]). All data were analyzed using SPSS (version 18; SPSS Inc, Chicago, IL).

A total of 237 marathoners (83 [35%] women, 154 [65%] men) who were registered for and intended to take part in the marathon were recruited at the expo event. Of these, 83 (35%) marathoners (28 women, 55 men) completed the postrace questionnaire within 1 week of administration. The characteristics of the respondents and nonrespondents, in addition to the results of the statistical comparison for each characteristic, are presented in Table 2. In summary, we observed no differences between respondents and nonrespondents to the questionnaire.

Excluding the 83 respondents to our postrace questionnaire, 16 126 runners took part in the marathon (5509 [34%] women, 10 617 [66%] men), the average finish time was 256 minutes (4 hours, 16 minutes) ± 50 minutes, and the mode finish time was 238 minutes (3 hours, 58 minutes). Formal statistical comparison between respondents and nonparticipants was not deemed appropriate, as we sought to exclude individuals with a walk-run pacing strategy for the purpose of defining HTW.

Fifteen (18%) of the 83 runners were deemed to have experienced HTW. The GPS data, including 1-km split times, were available for 56 (67%) of the 83 respondents (14 respondents who experienced HTW and 42 respondents who did not experience HTW; Table 3).

Results of the direct logistic regression and ROC curve analyses identified predictive values for the following discretized variables:

• (1)

  the time to complete the slowest split relative to the remaining splits (1-km [accuracy = 84.6%, sensitivity = 1, specificity = 0.6, P < .001] and 5-km [accuracy = 84.6%, sensitivity = 1, specificity = 0.644, P < .001] splits);

• (2)

  the SD in the split time (1-km [accuracy = 83%, sensitivity = 0.818, specificity = 0.8, P = .001] and 5-k [accuracy = 82%, sensitivity = 0.909, specificity = 0.622, P = .001] split resolutions).

The results of the ROC curve analysis for all discretized variables are presented in Table 4. The cutoff for HTW at the 1-km split resolution for SD was 0.0532 and at the 5-km resolution was 0.0346. The cutoff for HTW at the 1-km split resolution for the slowest relative segment was 1.1107 and at the 5-km resolution was 1.0731. The ROC curves are plotted for the combined set of discretized variables in Figure 2.

The purpose of our study was to devise a pace-based criterion for HTW. Instances of HTW were identified via a postrace questionnaire, based on the findings of Buman et al.¹²  Using their results and specifically any 3 of the 4 symptoms of generalized fatigue, unintentionally slowing of pace, desire to walk, or shifting focus to survival, we observed a 16% prevalence of HTW in our group of postrace questionnaire respondents. Using logistic regression and ROC curve analyses, we identified the pacing profiles associated with HTW for the 1-km and 5-km split-time resolutions. Specifically, by comparing any single 1-km segment with the remaining 41 1-km segments of the marathon (1-km split-time resolution) or any single 5-km segment with the remaining 7 5-km segments of the marathon (5-km split-time resolution), we classified instances of HTW by cross-referencing questionnaire responses with these split resolutions for each participant.

Based on a 1-km split-time resolution, runners were classified as having experienced HTW if they ran any single 1-km segment 11.07% slower than the average of the remaining 41 1-km segments of the race (accuracy = 84.6%, sensitivity = 1, specificity = 0.6) or if the SD of the normalized 1-km split times exceeded 0.0532 (accuracy = 83%, sensitivity = 0.818, specificity = 0.8). Based on a 5-km split-time resolution, runners were classified as having experienced HTW if they ran any single 5-km segment 7.3% slower than the average of the remaining seven 5-km segments of the race (accuracy = 84.6%, sensitivity = 1, specificity = 0.644) or if the SD of the normalized 5-km split times exceeded 0.0346 (accuracy = 82%, sensitivity = 0.909, specificity = 0.622). Although none of the 21.1-km or 10-km split-resolution thresholds were different based on our Bonferroni-adjusted a priori α (P = .0167), completing the second half of the race 4.3% slower than the first could be used to correctly classify 70.7% of HTW cases (sensitivity = 0.818, specificity = 0.533, P = .04) at the level of P < .05. These pace-based thresholds could be valuable to researchers evaluating the prevalence of HTW in cohorts for whom they lack subjective questionnaire data.

Subjective reports of perceived fatigue are an important foundation for defining HTW because runners experience fatigue differently. Researchers have presented HTW as a phenomenon that is only experienced by recreational or subelite runners,¹⁸  and the runners' physiology and background in endurance events are both likely to influence this experience.¹⁹  Although this might imply that HTW is to some extent unmeasurable, the degree of accuracy with which discretized measures of pacing could be used to classify HTW instances in our study, defined by high ratings of perceived exertion, suggests that the phenomenon is to some extent quantifiable, both subjectively and by observable performance, despite the underlying subjectivity of the experience. This is further evidenced by our analysis of each group's target, predicted, and actual finish times. Although we found no between-groups difference between target and actual finish times (ie, both groups reached their target finish times), participants who experienced HTW finished slower. From this information, we could deduce that participants who experienced HTW finished slower because they were simply less capable or trained less well for the race. However, the participants who experienced HTW were also less likely to achieve their predicted times, which were derived based on their self-reported training histories (Table 3). These predicted times can be considered to represent the extent to which participants fulfilled their potential, whereby faster predicted times were derived by having higher average weekly training distances and faster historical race times. Combining these findings, we posit that the greater difference between the predicted and actual finish times of participants who experienced HTW represents a negative effect of HTW, whereby, despite training adequately to achieve their predicted times, an HTW event disrupted this prediction and was associated with a slower finish. Although the causality of the relationship between pacing and HTW cannot be inferred using our results, previous researchers²⁰  demonstrated that aggressive early pacers are more likely to experience large slowdowns later in the race. Therefore, pacing and the occurrence of HTW are likely interdependent, whereby incorrect pacing in the early stages of a marathon might increase one's risk for HTW later, which itself manifests in an undesirable slowdown in pace. Our findings provide a means of quantifying how much of a pace slowdown is associated with HTW and how this is linked with slower finish times.

Despite the insights provided by the pace-based thresholds for HTW, the results of our study are not definitive and cannot be used to prevent instances of HTW or ascertain at what stage of the race people experience HTW or be leveraged to draw any causal inferences as to why it occurs. Given the personal and contextual factors associated with each marathon, a comprehensive understanding of The Wall will vary with each subjective experience. Variables not measured as part of our analysis, such as a marathoner's prerace²¹  and in-race nutrition strategies,²²  are probably important contributors to overall performance on race day. We illustrated that a definitive pace-based classification of The Wall (with accuracy levels approaching 100%) is probably an unrealistic pursuit due to the variability of its underlying subjective perception but that this subjectivity is nonetheless important when seeking to devise any such classification. In our study, by classifying HTW using subjective symptoms, we captured the heterogeneity of its physical manifestation, improving the external validity of these results.

In a practical sense, runners and coaches should acknowledge that although HTW cannot be fully understood, they should not be dissuaded from attempting to mitigate its symptoms during their preparation for a marathon event. Specifically, using cognitive strategies, including if-and-then implementation strategies, has been advocated by researchers.²³  Elite runners who used associative cognitive strategies coped better with pain during a marathon compared with nonelite runners who tended to rely on dissociative strategies.¹⁹  Furthermore, a few investigators²²  have shown that runners who consumed carbohydrate fuel during the race were less likely to experience HTW. Although we could not monitor our participants' adherence or nonadherence to such strategies, our data showed that it may be possible to prevent HTW and, by extension, decreased performance. Extrapolating these findings to our results, marathoners would be more likely to reach their predicted finish times by adopting an appropriate in-race nutrition strategy because they would be less likely to experience HTW.

We sought to maximize the external validity of our results through a variety of means, but some potential limitations should be acknowledged. First, the accuracy of the estimates of training behavior provided by our participants was an important consideration, as self-report sampling is potentially associated with more measurement error.²⁴  However, self-report training logs and diaries have consistently been used to determine the frequency, duration, intensity, and type of exercise exposures for a variety of outcomes²⁵  and have been accurate when describing exercise.²⁶  Attempts to reconcile data garnered via self-reporting with direct measures, such as GPS tracking devices, have indicated that the error associated with subjective determination is random.²⁷  Therefore, we are confident that any potential “noise” associated with self-report measures is unlikely to have made a meaningful difference in the accuracy or representativeness of the dataset.

Second, our pace-based classification of The Wall was derived from the grade-adjusted pacing profiles¹⁶  of the respondents to our questionnaire. The advantage of grade-adjusted pace is that it includes the elevation profile of the marathon course. Despite our efforts to limit the effect of this confounder, it is possible that our participants' paces, and, therefore, the pace-derived classification of The Wall, depended in part on the elevation profile of the course, thereby limiting its applicability to other marathons. It is also possible that the definition of HTW, whether determined by pace or the administration of a subjective questionnaire, will always be partially dependent on the marathon itself. Defining The Wall on a per-marathon basis may not be feasible; however, it would be prudent to investigate the potential variability of what constitutes The Wall for a group of marathons with various elevation profiles and environmental conditions in future studies. Researchers have elucidated a relative performance difference for marathons in different locations under different environmental conditions,²⁸  substantiating this presumption.

Third, although we sought to recruit a large representative sample covering the full spectrum of running ability (the fastest runner in our cohort finished in 2 hours and 30 minutes, and the slowest runner finished in 5 hours and 58 minutes), the external validity of our pace-based threshold of The Wall was limited by the relatively small sample of participants from only 1 marathon and the resulting constraints. For instance, the environmental conditions may have substantially affected the pace-based threshold of The Wall, but this cannot be elucidated using our dataset. We also sought to limit our cohort to participants who intended to run the entire marathon, hence our decision to exclude runners aiming for a finish time of >4 hours and 30 minutes. Future researchers building on our findings should replicate this study in larger samples of marathoners at a more diverse group of marathons.

We presented a pace-based threshold for HTW that can be applied when the subjectively reported data required to definitively determine the presence of this phenomenon on an individual basis are not available. Specifically, our findings suggested that, based on a 1-km split-time resolution, runners can be classified as having experienced HTW if they ran any 1-km segment 11.07% slower than the average of the remaining segments of the race or if the SD of the normalized 1-km split times exceeded 0.0532. For a 5-km split-time resolution, runners can be classified as having experienced HTW if they ran any 5-km segment 7.3% slower than the average of the remaining 5-km segments of the race or if the SD of the normalized 5-km split times exceeded 0.0346. These thresholds can be adopted by athletes, coaches, and researchers in any evaluation of HTW prevalence among groups of athletes for whom subjective questionnaire data are not available; however, future researchers should expand the methods presented here to include a greater variety of marathon courses, environmental conditions, and participant characteristics if a more definitive pace-based definition of The Wall is to be devised.

This research was cofunded by the European Regional Development Fund (ERDF) under Ireland's European Structural and Investment Funds Programmes 2014–2020, under the “Pace-Man” project title (B.C.).