Hemoglobin Mass for Detecting Autologous Blood Doping
Hemoglobin Mass for Detecting Autologous Blood Doping
During the observation period of 42 wk, 206 tHb measurements were obtained in 21 subjects. The maximal number of 10 samples per subject was ordered by the DCI for all but two test subjects for whom only eight tests were requested. Applying strict quality measures, seven measurements had to be excluded because of leakage or other measurement errors, so that a total of 199 (96.6%) valid tHb measurements (93 in control, 106 in doped) were included in the final analysis. Using an individual CO dose of 1.0 mL·kg body weight, [COHb] increased from 1.2% ± 0.5% before CO administration to 6.3% ± 0.8% after CO administration.
The WS and BS variances for tHb and the OFFmass are presented in Table 1. The use of the adaptive model for tHb and OFFmass allowed identification of doped subjects as displayed in Table 1 (B) and Table 2 (A, B).
The individual strikes of all individuals are presented in Table 2 (A), whereas the sensitivities, specificities, and positive and negative predictive values of each of the markers are displayed in Table 2 (B). The weighted CV for all tHb measurements in control was 2.8% (95% CI = 1.1%–2.8%) (25 ± 11 g during the 10 months) with a range of 1.3%–5.0%, whereas the weighted CV (TE) in doped was 5.2% (95% CI = 4.3%–6.2%) (47 ± 13 g during the 10 months) with a range of 2.2%–7.3% (Table 2, C).
A false-positive strike was only found at the 99% specificity level for one tHb value (subject 4, Fig. 3) but not for OFFmass values (Table 2). The sequence analysis of tHb and OFFmass yielded no false-positive strikes (Table 1 and Table 2). Three of 11 subjects in doped did not have any 99.9% strike by any of the four testing measures (tHb, OFFmass, and respective sequences), so that an overall specificity of 100% and a sensitivity of 73% were obtained when applying the current WADA result management protocol.
(Enlarge Image)
Figure 3.
—WS variance of OFFmass as a function of the WS variance of tHb. Closed triangles indicate the doped group; open circles indicate the control group.
Figure 2 displays an exemplary selection of two subjects of the control group (panels A and B) and one doped subject (panel C). In control subject 4 (panel C), a false-positive strike of tHb was observed at the 99% specificity level in the sixth test but not at the 99.9% level. The time interval between tests 5 and 6 was approximately 4 months, after which an increase of tHb (+86 g) was observed and remained stable in the subsequent tests. This confirms the preliminary analysis performed by linear regression and Durbin–Watson statistics that found a significant increase of tHb over time for subject 4.
Figure 3 illustrates the variance of the OFFmass as a function of the variance of tHb.
For the first time, tHb was used as a marker of the adaptive model of the ABP in a blinded controlled study design. The main finding is that with 99.9% specificity levels, 8 (73%) of 11 subjects were identified to have values outside of the individual threshold limits in at least one of the four used markers (tHb, OFFmass, and respective sequences), whereas no false-positive strike was observed during a study period of 42 wk. On the first glance, this seems very promising but requires a thorough discussion.
Although several studies discussed the inclusion of tHb in the ABP, so far, only MLrkeberg et al. simultaneously applied autologous blood transfusions and evaluated various passport approaches. Surprisingly, all calculations were made only concerning the retransfusion but not the withdrawal of blood. In our blinded study, the DCI ordered the tests on the basis of the knowledge about the results from previous tests so that it can be assumed that potential time points of blood withdrawal were taken into consideration. To further adopt the best possible realistic conditions, the number of tests was limited to 10.
In the approach of MLrkeberg et al., where the time point of retransfusion was known and the majority of transfusions included large amounts (three bags of blood in 21 of 29 subjects), during the first 3 d after retransfusion, tHb was the most sensitive marker with ~40% of all samples exceeding their 99.9% threshold. In our investigation, after 42 wk using 99% specificity thresholds, tHb had a sensitivity of 82% but yielded the only false-positive strike of study (Fig. 2B, subject 4). This subject demonstrated a marked increase in tHb between time points 5 and 6 (+86 g), which were separated by approximately 4 months. Because the subsequent tHb values remained elevated compared with test point 5, we assume that a true natural tHb increase had occurred. In a real-life setting, this profile would have been submitted to a panel of experts who would scrutinize the data and evaluate whether the change is associated with doping or can be attributed to physiological or pathological reasons. Applying the stricter 99.9% specificity threshold, the sensitivity of tHb decreased to 27% but showed no false-positive strikes.
To increase sensitivity, we used %ret to calculate OFFmass where the contribution of %ret to OFFmass was modeled similarly to the contribution to OFF-hr in the ABP. Compared with tHb alone, a better sensitivity of 64% applying 99.9% specificity thresholds was possible for OFFmass and the sequence of OFFmass each, yielding no false-positive strikes (Table 2). Despite our longitudinal blinded study design, these sensitivities also seemed much improved compared with the standardized conditions by MLrkeberg et al., where optimized sensitivities between 27% to 40% were reported for tHb and Hbmr (a similar marker based on tHb and %ret) after exclusion of two suspicious subjects. This demonstrates that to achieve the best possible sensitivity of the ABP, the timing of testing has to be planned by experts in the field based upon previous results and a profound hematologic knowledge as well as metainformation such as the athlete's whereabouts and competition schedule ("intelligent testing").
Interestingly, the subject with only two transfusions during the study period remained undetected by any of the markers (Table 2, subject 14), so it may seem that the sensitivities may be related to the dimension of blood manipulation. On the other hand, during 42 wk, in most doped subjects, only one EC of ~280 mL with an Hct of 53%–60% was transfused per time point after a donation of ~500 mL of full blood (Table 2) so that even these relatively small amounts seem detectable when tHb is used as a marker of the adaptive model.
Our control group provided the reference data for the calculation of the tHb adaptive model as described in the "Methods" section. Although the leave-one-out cross-validation procedure presents the advantage to derive sound estimates despite the low number of control subjects, additional studies are nevertheless required to see if the assumption of a universal WS biological variance is tenable. This is a difficult task because any statistics aiming to assess the heterogeneity in the WS variance from the data alone would not allow the separation of biological variations from analytical uncertainty in a straightforward way. It has been shown empirically that an error model that decomposes the WS variance between a biological component given by a universal variance and an analytical uncertainty given by a TE gave the best results, and the same error model was implemented in the adaptive model. The adaptive model is actually able to model individual WS biological variance (stratified prior distribution ofWS variance move to individualbased values during data acquisition), but allowing for individual WS biological variance would lead to a significant decrease in sensitivity.
The analytical error of this study expressed as a TE of 1.5% is in line with the values of previous studies in various laboratories where a TE between 1.4% and 2.2% was found, fluctuating around an average value of 1.7% (a TE of 1.7% was used in the calculation of the tHb adaptive model). In addition, we used our control tHb BS variance (7900 g) and WS biological variance (550 g), which were larger compared with the data from Prommer et al. (BS variance = 3993 g, WS variance = 408 g).
It was postulated that tHb could be included in the ABP because a low biological variation can be ascertained. In this context, the longitudinal variation of tHb was studied by several groups (applying the optimized CO rebreathing method) and mostly calculated as the weighted CV as described in our "Methods" section. Eastwood et al. studied the stability of tHb during 100 d in active men (measurements every 1 to 6 d for 100–114 d) and found a weighted CV of 2.1%. Garvican et al. investigated the seasonal variation of tHb in female road cyclists (the mean number of measurements per athlete was eight during a 6.6 ± 2.3-month period) and reported a weighted mean CV of 3.3%. Prommer et al. measured tHb in 24 endurance athletes five times during a year and reported an average oscillation of 4.6%. In comparison with these existing data, we found a similar weighted CV of 2.8% in our control group of 10 subjects, whereas in doped, the weighted CV was 5.2% (Table 2). Thus, during a long period and under real testing conditions, tHb obtained by the optimized CO method presents with higher variations than in long-term and highfrequency testing regimens. Despite that, the application of the adaptive model to tHb and calculation of OFFmass returns an improved sensitivity to detect autologous blood transfusion compared with previous data. Furthermore, Figure 3 demonstrates that already, the ANOVA of tHb and OFFmass allows a proper discrimination of doped and control subjects. Although four doped subjects seem similar compared to control, the sequence measure of the adaptive model (Table 2 and Fig. 2) integrates this observation and takes random sampling variation into account.
Under strict and critical appraisal, the sensitivities and specificities may only be valid for the regimen of testing ordered by the DCI of this study and could thus be difficult to extrapolate to others. Because the DCI was a well-trained scientist in the field of indirect blood doping detection, the results may be regarded as optimal. On the other hand, the effects of blood manipulation on erythropoiesis biomarkers are well described in the literature and were acquired by various experts in the field so that it can be speculated that similar decisions regarding the time points of testing would be expected by other DCIs with the same knowledge. Furthermore, it needs to be emphasized that the sensitivities found here merely report the performance of the adaptive model and well-timed tests to reveal autologous blood doping based on analysis of tHb and OFFmass and are not to be equalized with the sensitivity of a positive doping test.
Although this study aimed to have the best possible realistic design, another limitation is the selection of subjects and absence of some real-life confounders of an elite athletic population. Obviously, because of the WADA antidoping regulations, it is impossible to select a group of actively competing elite athletes and expose them to autologous blood transfusion. We tried to include as many recreationally active cyclists in both groups to obtain representative blood samples. The major difference between our group of subjects and an elite population undergoing rigorous training is the fact that hemodilution due to continuous phases of racing and alterations due to possible exposures to altitude of an adequate hypoxic dose are not represented. In this context, it has to be pointed out that in our study, only plasma volume–independent variables are investigated (tHb, %ret); thus, the influence of plasma volume shifts can be negated. The effect of altitude on second-generation blood tests to detect erythropoietin abuse was published by Ashenden et al.; data on the effect of altitude on the adaptive model of the ABP do not seem to be available yet. Compared with other longitudinal studies of tHb in active men and even elite athletes, as described previously, we find a very similar weighted CV in control but a significantly higher weighted CV in doped and thus describe a variation larger than expected with training.
The question remains if these promising results justify or promote the actual inclusion of tHb as a marker of the adaptive model of the ABP. The first answer may be found if our results using tHb are compared with the sensitivities of the current ABP markers [Hb] and OFF-hr, which we studied in the same scenario and published elsewhere. For both levels of specificity (99% and 99.9%), the use of tHb and OFFmass instead of [Hb] and OFF-hr leads to improved sensitivities ( Table 2 in this study and Table 2 in Pottgiesser et al.) to detect values outside of individual threshold limits. The same applies for the respective sequence measures. When a combination of all markers is used at the 99%specificity level, tHb and OFFmass showed a sensitivity of 91% compared with a sensitivity of 82% using the classic markers, whereas both sets of markers yielded a false-positive strike (tHb and [Hb]). At the 99.9% specificity level, tHb and OFFmass revealed a sensitivity of 73% and no false-positives compared with a sensitivity of 55%of [Hb] and OFF-hr with a persisting false-positive strike. In other terms, the use of tHb and OFFmass allowed the identification of one additional subject at the 99% specificity level and even two additional subjects at the 99.9% sensitivity level who would have been classified as unsuspicious using solely [Hb] and OFF-hr.
Although this favors tHb and demonstrates its great potential compared with an ordinary venous blood draw, the use of even the optimized CO rebreathing method for determination of tHb is time-consuming and relatively cumbersome and may thus be difficult to implement with larger cohorts of athletes at a competition venue. One further disadvantage of this method is that no quality control system is readily available. We previously discussed that in antidoping, strict analytical protocols and internal as well as external quality controls of forensic standards are of utmost importance to guarantee results that will stand legal scrutiny.
Furthermore, the CO rebreathing method requires subject cooperation and can easily be manipulated by the cheating athlete with a huge effect on the results (e.g., induced CO leakages). It was shown in a recent work that large CO leakages may inflate the mean measurement error up to 9.3%, which is not acceptable in an antidoping setting.
Another problematic point is the fact that CO is a potentially toxic substance with a variety of symptoms after exposure to CO that correlate poorly with the level of [COHb] and may show high individual differences. In this context, it seems difficult to oblige healthy athletes to accept the administration of potentially harmful substances for doping control.
Results
During the observation period of 42 wk, 206 tHb measurements were obtained in 21 subjects. The maximal number of 10 samples per subject was ordered by the DCI for all but two test subjects for whom only eight tests were requested. Applying strict quality measures, seven measurements had to be excluded because of leakage or other measurement errors, so that a total of 199 (96.6%) valid tHb measurements (93 in control, 106 in doped) were included in the final analysis. Using an individual CO dose of 1.0 mL·kg body weight, [COHb] increased from 1.2% ± 0.5% before CO administration to 6.3% ± 0.8% after CO administration.
The WS and BS variances for tHb and the OFFmass are presented in Table 1. The use of the adaptive model for tHb and OFFmass allowed identification of doped subjects as displayed in Table 1 (B) and Table 2 (A, B).
The individual strikes of all individuals are presented in Table 2 (A), whereas the sensitivities, specificities, and positive and negative predictive values of each of the markers are displayed in Table 2 (B). The weighted CV for all tHb measurements in control was 2.8% (95% CI = 1.1%–2.8%) (25 ± 11 g during the 10 months) with a range of 1.3%–5.0%, whereas the weighted CV (TE) in doped was 5.2% (95% CI = 4.3%–6.2%) (47 ± 13 g during the 10 months) with a range of 2.2%–7.3% (Table 2, C).
A false-positive strike was only found at the 99% specificity level for one tHb value (subject 4, Fig. 3) but not for OFFmass values (Table 2). The sequence analysis of tHb and OFFmass yielded no false-positive strikes (Table 1 and Table 2). Three of 11 subjects in doped did not have any 99.9% strike by any of the four testing measures (tHb, OFFmass, and respective sequences), so that an overall specificity of 100% and a sensitivity of 73% were obtained when applying the current WADA result management protocol.
(Enlarge Image)
Figure 3.
—WS variance of OFFmass as a function of the WS variance of tHb. Closed triangles indicate the doped group; open circles indicate the control group.
Figure 2 displays an exemplary selection of two subjects of the control group (panels A and B) and one doped subject (panel C). In control subject 4 (panel C), a false-positive strike of tHb was observed at the 99% specificity level in the sixth test but not at the 99.9% level. The time interval between tests 5 and 6 was approximately 4 months, after which an increase of tHb (+86 g) was observed and remained stable in the subsequent tests. This confirms the preliminary analysis performed by linear regression and Durbin–Watson statistics that found a significant increase of tHb over time for subject 4.
Figure 3 illustrates the variance of the OFFmass as a function of the variance of tHb.
Discussion
For the first time, tHb was used as a marker of the adaptive model of the ABP in a blinded controlled study design. The main finding is that with 99.9% specificity levels, 8 (73%) of 11 subjects were identified to have values outside of the individual threshold limits in at least one of the four used markers (tHb, OFFmass, and respective sequences), whereas no false-positive strike was observed during a study period of 42 wk. On the first glance, this seems very promising but requires a thorough discussion.
Although several studies discussed the inclusion of tHb in the ABP, so far, only MLrkeberg et al. simultaneously applied autologous blood transfusions and evaluated various passport approaches. Surprisingly, all calculations were made only concerning the retransfusion but not the withdrawal of blood. In our blinded study, the DCI ordered the tests on the basis of the knowledge about the results from previous tests so that it can be assumed that potential time points of blood withdrawal were taken into consideration. To further adopt the best possible realistic conditions, the number of tests was limited to 10.
In the approach of MLrkeberg et al., where the time point of retransfusion was known and the majority of transfusions included large amounts (three bags of blood in 21 of 29 subjects), during the first 3 d after retransfusion, tHb was the most sensitive marker with ~40% of all samples exceeding their 99.9% threshold. In our investigation, after 42 wk using 99% specificity thresholds, tHb had a sensitivity of 82% but yielded the only false-positive strike of study (Fig. 2B, subject 4). This subject demonstrated a marked increase in tHb between time points 5 and 6 (+86 g), which were separated by approximately 4 months. Because the subsequent tHb values remained elevated compared with test point 5, we assume that a true natural tHb increase had occurred. In a real-life setting, this profile would have been submitted to a panel of experts who would scrutinize the data and evaluate whether the change is associated with doping or can be attributed to physiological or pathological reasons. Applying the stricter 99.9% specificity threshold, the sensitivity of tHb decreased to 27% but showed no false-positive strikes.
To increase sensitivity, we used %ret to calculate OFFmass where the contribution of %ret to OFFmass was modeled similarly to the contribution to OFF-hr in the ABP. Compared with tHb alone, a better sensitivity of 64% applying 99.9% specificity thresholds was possible for OFFmass and the sequence of OFFmass each, yielding no false-positive strikes (Table 2). Despite our longitudinal blinded study design, these sensitivities also seemed much improved compared with the standardized conditions by MLrkeberg et al., where optimized sensitivities between 27% to 40% were reported for tHb and Hbmr (a similar marker based on tHb and %ret) after exclusion of two suspicious subjects. This demonstrates that to achieve the best possible sensitivity of the ABP, the timing of testing has to be planned by experts in the field based upon previous results and a profound hematologic knowledge as well as metainformation such as the athlete's whereabouts and competition schedule ("intelligent testing").
Interestingly, the subject with only two transfusions during the study period remained undetected by any of the markers (Table 2, subject 14), so it may seem that the sensitivities may be related to the dimension of blood manipulation. On the other hand, during 42 wk, in most doped subjects, only one EC of ~280 mL with an Hct of 53%–60% was transfused per time point after a donation of ~500 mL of full blood (Table 2) so that even these relatively small amounts seem detectable when tHb is used as a marker of the adaptive model.
Our control group provided the reference data for the calculation of the tHb adaptive model as described in the "Methods" section. Although the leave-one-out cross-validation procedure presents the advantage to derive sound estimates despite the low number of control subjects, additional studies are nevertheless required to see if the assumption of a universal WS biological variance is tenable. This is a difficult task because any statistics aiming to assess the heterogeneity in the WS variance from the data alone would not allow the separation of biological variations from analytical uncertainty in a straightforward way. It has been shown empirically that an error model that decomposes the WS variance between a biological component given by a universal variance and an analytical uncertainty given by a TE gave the best results, and the same error model was implemented in the adaptive model. The adaptive model is actually able to model individual WS biological variance (stratified prior distribution ofWS variance move to individualbased values during data acquisition), but allowing for individual WS biological variance would lead to a significant decrease in sensitivity.
The analytical error of this study expressed as a TE of 1.5% is in line with the values of previous studies in various laboratories where a TE between 1.4% and 2.2% was found, fluctuating around an average value of 1.7% (a TE of 1.7% was used in the calculation of the tHb adaptive model). In addition, we used our control tHb BS variance (7900 g) and WS biological variance (550 g), which were larger compared with the data from Prommer et al. (BS variance = 3993 g, WS variance = 408 g).
It was postulated that tHb could be included in the ABP because a low biological variation can be ascertained. In this context, the longitudinal variation of tHb was studied by several groups (applying the optimized CO rebreathing method) and mostly calculated as the weighted CV as described in our "Methods" section. Eastwood et al. studied the stability of tHb during 100 d in active men (measurements every 1 to 6 d for 100–114 d) and found a weighted CV of 2.1%. Garvican et al. investigated the seasonal variation of tHb in female road cyclists (the mean number of measurements per athlete was eight during a 6.6 ± 2.3-month period) and reported a weighted mean CV of 3.3%. Prommer et al. measured tHb in 24 endurance athletes five times during a year and reported an average oscillation of 4.6%. In comparison with these existing data, we found a similar weighted CV of 2.8% in our control group of 10 subjects, whereas in doped, the weighted CV was 5.2% (Table 2). Thus, during a long period and under real testing conditions, tHb obtained by the optimized CO method presents with higher variations than in long-term and highfrequency testing regimens. Despite that, the application of the adaptive model to tHb and calculation of OFFmass returns an improved sensitivity to detect autologous blood transfusion compared with previous data. Furthermore, Figure 3 demonstrates that already, the ANOVA of tHb and OFFmass allows a proper discrimination of doped and control subjects. Although four doped subjects seem similar compared to control, the sequence measure of the adaptive model (Table 2 and Fig. 2) integrates this observation and takes random sampling variation into account.
Limitations
Under strict and critical appraisal, the sensitivities and specificities may only be valid for the regimen of testing ordered by the DCI of this study and could thus be difficult to extrapolate to others. Because the DCI was a well-trained scientist in the field of indirect blood doping detection, the results may be regarded as optimal. On the other hand, the effects of blood manipulation on erythropoiesis biomarkers are well described in the literature and were acquired by various experts in the field so that it can be speculated that similar decisions regarding the time points of testing would be expected by other DCIs with the same knowledge. Furthermore, it needs to be emphasized that the sensitivities found here merely report the performance of the adaptive model and well-timed tests to reveal autologous blood doping based on analysis of tHb and OFFmass and are not to be equalized with the sensitivity of a positive doping test.
Although this study aimed to have the best possible realistic design, another limitation is the selection of subjects and absence of some real-life confounders of an elite athletic population. Obviously, because of the WADA antidoping regulations, it is impossible to select a group of actively competing elite athletes and expose them to autologous blood transfusion. We tried to include as many recreationally active cyclists in both groups to obtain representative blood samples. The major difference between our group of subjects and an elite population undergoing rigorous training is the fact that hemodilution due to continuous phases of racing and alterations due to possible exposures to altitude of an adequate hypoxic dose are not represented. In this context, it has to be pointed out that in our study, only plasma volume–independent variables are investigated (tHb, %ret); thus, the influence of plasma volume shifts can be negated. The effect of altitude on second-generation blood tests to detect erythropoietin abuse was published by Ashenden et al.; data on the effect of altitude on the adaptive model of the ABP do not seem to be available yet. Compared with other longitudinal studies of tHb in active men and even elite athletes, as described previously, we find a very similar weighted CV in control but a significantly higher weighted CV in doped and thus describe a variation larger than expected with training.
Practical Implications
The question remains if these promising results justify or promote the actual inclusion of tHb as a marker of the adaptive model of the ABP. The first answer may be found if our results using tHb are compared with the sensitivities of the current ABP markers [Hb] and OFF-hr, which we studied in the same scenario and published elsewhere. For both levels of specificity (99% and 99.9%), the use of tHb and OFFmass instead of [Hb] and OFF-hr leads to improved sensitivities ( Table 2 in this study and Table 2 in Pottgiesser et al.) to detect values outside of individual threshold limits. The same applies for the respective sequence measures. When a combination of all markers is used at the 99%specificity level, tHb and OFFmass showed a sensitivity of 91% compared with a sensitivity of 82% using the classic markers, whereas both sets of markers yielded a false-positive strike (tHb and [Hb]). At the 99.9% specificity level, tHb and OFFmass revealed a sensitivity of 73% and no false-positives compared with a sensitivity of 55%of [Hb] and OFF-hr with a persisting false-positive strike. In other terms, the use of tHb and OFFmass allowed the identification of one additional subject at the 99% specificity level and even two additional subjects at the 99.9% sensitivity level who would have been classified as unsuspicious using solely [Hb] and OFF-hr.
Although this favors tHb and demonstrates its great potential compared with an ordinary venous blood draw, the use of even the optimized CO rebreathing method for determination of tHb is time-consuming and relatively cumbersome and may thus be difficult to implement with larger cohorts of athletes at a competition venue. One further disadvantage of this method is that no quality control system is readily available. We previously discussed that in antidoping, strict analytical protocols and internal as well as external quality controls of forensic standards are of utmost importance to guarantee results that will stand legal scrutiny.
Furthermore, the CO rebreathing method requires subject cooperation and can easily be manipulated by the cheating athlete with a huge effect on the results (e.g., induced CO leakages). It was shown in a recent work that large CO leakages may inflate the mean measurement error up to 9.3%, which is not acceptable in an antidoping setting.
Another problematic point is the fact that CO is a potentially toxic substance with a variety of symptoms after exposure to CO that correlate poorly with the level of [COHb] and may show high individual differences. In this context, it seems difficult to oblige healthy athletes to accept the administration of potentially harmful substances for doping control.
Source...