Association between Sports Participation, Factor VIII Levels and Bleeding in Hemophilia A

Background  Little is known on how sports participation affects bleeding risk in hemophilia. This study aimed to examine associations between sports participation, factor VIII (FVIII) levels and bleeding in persons with hemophilia A. Methods  In this observational, prospective, single-center study, persons with hemophilia A who regularly participated in sports were followed for 12 months. The associations of patient characteristics, FVIII levels, and type/frequency of sports participation with bleeding were analyzed by repeated time-to-event modelling. Results  One hundred and twelve persons (median age: 24 years [interquartile range:16–34], 49% severe, 49% on prophylaxis) were included. During follow-up, 70 bleeds of which 20 sports-induced were observed. FVIII levels were inversely correlated with the bleeding hazard; a 50% reduction of the baseline bleeding hazard was observed at FVIII levels of 3.1 and a 90% reduction at 28.0 IU/dL. The bleeding hazard did not correlate with sports participation. In addition, severe hemophilia, prestudy annual bleeding rate, and presence of arthropathy showed a positive association with the bleeding hazard. Conclusion  This analysis showed that FVIII levels were an important determinant of the bleeding hazard, but sports participation was not. This observation most likely reflects the presence of adequate FVIII levels during sports participation in our study. Persons with severe hemophilia A exhibited a higher bleeding hazard at a similar FVIII levels than nonsevere, suggesting that the time spent at lower FVIII levels impacts overall bleeding hazard. These data may be used to counsel persons with hemophilia regarding sports participation and the necessity of adequate prophylaxis.

In this observational, prospective, single-center Sports Participation and Injuries in people with hemophilia (SPRAIN) study from University Medical Center Utrecht (The Netherlands), persons with hemophilia who regularly participated in sports were followed for 1 year. For the present analysis, data from 13 people with hemophilia B were excluded, as the exposure-effect relation of factor VIII (FVIII) and factor IX concentrates could be different. Injuries and bleeds were assessed proactively, i.e., participants were contacted bi-weekly, and information about nature, involvement of sports participation, mechanism of injury leading to a bleed, and, in case of prophylactic treatment, last factor concentrate dose and timing of dosing and event were recorded. Bleeding was defined according to the International Society on Thrombosis and Haemostasis definitions. 1 A bleed was classified as occurring during sports when the bleed developed following a sports injury and required treatment with factor concentrates with or without a consultation with the hemophilia treatment center. Participation in high-risk sports was based on a National Hemophilia Foundation score >2. 2 The hemophilia joint health score was assessed at study initiation. 3,4 Repeated Time-to-Event Model A repeated time-to-event (RTTE) model is able to characterize the occurrence of time-varying events (bleeding over time) together with event predictors (e.g., factor activity levels and sports activities). In an RTTE model, the bleeding probability over time is described by the hazard function. First, a median bleeding hazard was estimated which describes the bleeding hazard for a median person using data from the whole population simultaneously. Differences in bleeding hazard between persons were evaluated by inclusion of the inter-individual variability on the hazard.
Exponential (Equation 1), Gompertz (Equation 2), and Weibull (Equation 3) hazard functions were tested to describe the distribution of time to bleeding. An exponential hazard function describes a constant hazard over time, while Gompertz and Weibull hazard functions can describe increasing or decreasing bleeding hazards over time. 5 The final individual hazard function was described by Equation 4: in which the bleeding hazard of the ith patient at time t is described by h i (t). λ describes the scale, γ the shape, FVIII the FVIII activity level at time t, IC 50 the FVIII activity level at which 50% of the maximal inhibition on the bleeding hazard occurs and ηi the inter-individual variability in bleeding hazard with mean 0 and variance ω 2 .
For persons not receiving a FVIII dose on the day of study initiation, FVIII levels at start of study were calculated based on the previous dose administered prior to study inclusion.
The λ and IC50 values were parameterized to describe the bleeding hazard for a FVIII level of 0.5 and 20 IU/dL following Equations 5 and 6. 6 The survival function describes the probability of not having a bleed within a specific time interval. By taking the integral of the hazard, the cumulative hazard can be calculated, which is used to calculate the survival function (Equation 7): in which the survival function of the ith patient within the time interval 0 to t j is described by S i (t). In this example, 0 is taken as start of the time interval and t j as the end of the time interval, and h i (t) is the individual bleeding hazard.
For some bleeds only the day of the bleeding was known, but not the exact time of the bleeding event. Interval censoring was applied for these bleeds. The probability that these bleeds occurred can be described by the probability that the event occurred between t j and t j þ24 hours, following equation 8:

Covariate Analysis
A full random-effects model (FREM) was used to identify covariates with an effect on individual bleeding hazard. 7 This covariate analysis method can characterize the correlation between model parameters-such as the bleeding hazardand all patient characteristics of interest simultaneously. Herewith, the correlation between the bleeding hazard (including the effect of FVIII levels) and sports activities can be evaluated independently of other patient factors.
Furthermore, problems with correlations between covariates and multiplicity are avoided with this method. Covariates are described by the mean and variance, handled as observations into the dataset. The mean is included as fixed effect and the variance as a random effect. The FREM model estimates the random effects of the parameters and covariates and the covariance between those two in a full covariance matrix. The covariance between the parameter and covariates describes the covariate effect. An exponential covariate parameter relationship was used.
During model development, we did not evaluate injuries as a covariate, as all bleeds except for four spontaneous bleeds were related to an injury. Furthermore, when an Supplementary Fig. S1 Illustration of the relationship between factor VIII (FVIII) levels and bleeding hazard for two individuals from the dataset. In the top panels, individual FVIII level over time is plotted, while the bottom panels show the corresponding model-predicted individual bleeding hazard. Patient A (10 years, 33 kg, treated with 3Â per week 750 IU Elocta) did not experience any bleeds, while patient B (34 years, 73 kg, treated with 3Â per week 1,000 IU Novoeight) experienced two bleeds (dots). The bleeding hazard is inversely related to the FVIII levels and is in general higher for patients who experience more bleeds. injury occurred without a bleed, the timing of the last concentrate dose was not explicitly recorded.

Model Development and Assessment
The RTTE model was developed in NONMEM (v7.4.1, Icon Development Solutions, Gaithersburg, Maryland, United States). The model was estimated with the Monte Carlo importance sampling assisted by mode a posteriori (IMPMAP) method. R v4.1.1, Piranha v2.9.9 and PsN v5.2.6 were used for data handling, visualization, model management, and evaluation.