To characterise the time course of changes in haemoglobin mass (Hbmass) in response to altitude exposure.

This meta-analysis uses raw data from 17 studies that used carbon monoxide rebreathing to determine Hbmass prealtitude, during altitude and postaltitude. Seven studies were classic altitude training, eight were live high train low (LHTL) and two mixed classic and LHTL. Separate linear-mixed models were fitted to the data from the 17 studies and the resultant estimates of the effects of altitude used in a random effects meta-analysis to obtain an overall estimate of the effect of altitude, with separate analyses during altitude and postaltitude. In addition, within-subject differences from the prealtitude phase for altitude participant and all the data on control participants were used to estimate the analytical SD. The ‘true’ between-subject response to altitude was estimated from the within-subject differences on altitude participants, between the prealtitude and during-altitude phases, together with the estimated analytical SD.

During-altitude Hbmass was estimated to increase by ∼1.1%/100 h for LHTL and classic altitude. Postaltitude Hbmass was estimated to be 3.3% higher than prealtitude values for up to 20 days. The within-subject SD was constant at ∼2% for up to 7 days between observations, indicative of analytical error. A 95% prediction interval for the ‘true’ response of an athlete exposed to 300 h of altitude was estimated to be 1.1–6%.

Camps as short as 2 weeks of classic and LHTL altitude will quite likely increase Hbmass and most athletes can expect benefit.

An increase in erythropoiesis resulting from altitude exposure has been described for over 100 years,

Rasmussen ^{51}chromium-labelled red blood cells.

In 2005, Schmidt and Prommer

The data used in the meta-analysis were obtained from authors who had, since 2008, published the results of studies that used the ‘optimised CO rebreathing’ method to evaluate the effects of altitude training on Hbmass. Briefly, this rebreathing method involves a known CO dose of ∼1.2 ml/kg body mass being administered and rebreathed for 2 min. Capillary fingertip blood samples are taken before the start of the test and 7 min after administration of the CO dose. Both blood samples are measured a minimum of five times for determination of percentage of carboxyhaemoglobin (%HbCO) using an OSM3 Hemoximeter (Radiometer, Copenhagen). Hbmass is calculated from the mean change in %HbCO before and after rebreathing CO.

Individual, deidentified raw data were provided from 17 studies

In addition to the recorded altitude of each study, the total hours spent in hypoxia was calculated from the hours per day and numbers of days of exposure. This approach allowed for comparison of classic and LHTL modes, since the former affords continuous altitude exposure, while the latter is intermittent. All but one of the LHTL studies used ≥14 h/day in hypoxia, while the other used 10 h/day.

Data sources

Reference | Altitude mode | Altitude (m) | Duration (h) | Sport | Calibre of athletes | N at altitude | N in control | Number of measures per participant |
---|---|---|---|---|---|---|---|---|

Clark | LHTL | 3000 | 294 | Cycling | International | 12 m | – | 7 |

Frese and Friedmann-Bette | Classic | 1300–1650 | 480–528 | Running | Junior | 7 f, 4 m | 2 f, 6 m | 2–6 |

Garvican | LHTL | 3000 | 416 | Cycling | International | 12 f | – | 8–12 |

Garvican | Classic | 2760 | 456 | Cycling | International | 8 m | 7 m | 5–9 |

Garvican-Lewis | LHTL | 3000 | 154–266 | Water polo | International | 9 f | – | 4 |

Gough | Classic LHTL | 2100–2320 | 204–504 | Swimming | International | 3 f, 14 m | – | 2–4 |

Humberstone | LHTL | 3000 | 238 | Triathlon | International | 2 f, 5 m | 6 f, 11 m | 4 |

McLean | Classic | 2130 | 456 | Football | National | 21 m | 9 m | 3–9 |

Classic | 2100 | 432 | Football | National | 23 m | – | 4–6 | |

Neya | Classic | 1300 | 504 | Running | Collegiate | 7 m | – | 3 |

Pottgiesser | Classic | 1816 | 504 | Cycling | International | 7 m | – | 3 |

Robertson | LHTL | 3000 | 294 | Running | National | 4 f, 6 m | 3 f, 5 m | 6–12 |

Robertson | LHTL | 3000 | 294 | Running | National | 2 f, 6 m | 2 f, 7m | 6 |

Saunders | LHTL | 3000 | 294 | Race walking | International | 3 f, 3 m | 6 f, 5 m | 7 |

Wachsmuth | Classic | 2320 | 672 | Swimming | International | 6 f, 13 m | – | 2–5 |

Wachsmuth | Classic | 3600 | 288 | Football | Junior | 17 m | 16 m | 3–5 |

Wachsmuth | Classic | 2300 | 504 | Swimming | National | 3 f, 6 m | 3 f, 4 m | 3–7 |

Total | 73 f, 175 m | 28 f, 75 m |

f, females; LHTL, live high train low; m, males.

Where included in the study design, control participants were coded as controls and those exposed to altitude were coded as altitude participants. The one exception was the LHTL study of Neya

The approach taken was first to fit linear mixed models separately to the data from each of the 17 studies to estimate the effects of altitude. The resultant estimates were then used in a random effects meta-analysis to obtain an overall estimate of the effect of altitude, with separate analyses for the during-altitude and postaltitude phases. In addition, all possible within-subject differences for control participants and those during the prealtitude and during-altitude phases for altitude participants were used to evaluate within-subject variation. All analyses were conducted using the statistical package R,

In the initial analyses of the 17 studies, Hbmass values were log transformed (natural logarithms (ln)) and linear mixed models fitted with treatment (control or altitude), days during altitude, days postaltitude and sex as fixed effects and participant as a random effect. In addition, where appropriate, the models allowed for different within-subject SDs for men and women (some of the studies used all male or all female participants) and within-subject autocorrelation. Some of the studies had too few observations on each participant to warrant inclusion of the assumed autocorrelation structure but, for consistency, the same form of model was fitted to all 17 data sets.

The next stage fitted linear mixed models with response variable estimates of the mean (over participants within each study) differences between baseline and subsequent values of ln(Hbmass) on altitude participants, with separate analyses for the during-altitude and postaltitude phases.

For the during-altitude phase, the time at altitude, both in terms of the number of days and the number of hours, the altitude and the type of altitude (classic or LHTL) were treated as fixed effects; whereas for the postaltitude phase, altitude, the number of days and hours at altitude, the type of altitude and number of days postaltitude were treated as fixed effects. For both analyses, the study was treated as a random effect and the SEs associated with the estimates of mean differences, as provided by the initial analyses, were used to determine suitable weightings.

Estimates obtained from these analyses were back transformed (via the exponential function) to express results as percentage changes on the Hbmass scale.

Without some form of intervention, Hbmass is considered to be constant, especially over a period of a few days, so that differences in measurements taken under stable conditions can be attributed almost exclusively to measurement, or analytical, error.

Using the control participant data and just the prealtitude values from altitude participants, an estimate of the analytical SD was obtained as follows. Separately for each study, estimates of within-subject SDs associated with each difference in days were obtained as the square root of half the average of the squared differences in ln(Hbmass). A linear mixed model was then fitted to the ln-transformed (estimated) SDs with the number of days between estimates as a fixed effect, study as a random effect and weights determined by the numbers of differences used to estimate the SDs. The results of this analysis were back transformed (twice, using exponentials) to obtain coefficients of variation (CVs) on the Hbmass scale, with the analytical CV estimated as the value associated with readings zero days apart.

Using data from (just) the altitude participants, estimates of the between-subject variation in the ‘true’ response to altitude were obtained as follows. First, estimates of the within-subject SDs of ln(Hbmass) associated with differences between prealtitude and altitude values, and between values obtained while at altitude, for different values of the difference in the number of hours at altitude, were obtained as the SD of the relevant differences divided by √2. Linear mixed models were then fitted to the ln-transformed SDs with the difference in the number of hours at altitude as a fixed effect, study as a random effect and weights determined by the degrees of freedom of the estimated SDs. The models considered were constrained so that the estimated within-subject SD after zero (additional) hours at altitude agreed with the estimate of the analytical SD. This was achieved by subtracting the natural log of the estimated analytical SD from each of the ln-transformed SDs and then fitting models without an intercept term. Estimates of the SDs of the between-subject ‘true’ responses were then obtained as

for a range of values of (differences in) the hours at altitude; these SDs were then used to estimate the likely range of ‘true’ responses to different exposures to altitude.

A total of 1624 measures of Hbmass were made on 328 participants, 18 of whom participated in more than one study (14 in 2 studies, 3 in 3 studies and 1 in 4 studies); 225 participants participated as altitude-only participants, 96 as control-only participants and 7 as altitude and control participants, but in different studies (

The median classic altitude was 2320 m (range 1360–3600 m); while all but one LHTL study used 3000 m, the other used 2500 m.

Of the 40 estimates of the change in ln(Hbmass) from prealtitude to during altitude available for analysis, one appeared to be an obvious outlier, see

Estimates of the change in haemoglobin mass (Hbmass) during live high train low (LHTL, n=24) and classic (n=16) altitude exposure. Fitted lines are for the linear and quadratic models. Dashed lines are the upper and lower 95% confidence limits of the quadratic model. The relative weightings of estimates are indicated by symbol size and border thickness—the largest symbols are for the highest weighted estimates, the estimates with the smallest SEs. †The study at 1360 m,

After allowing for the time at altitude, none of the other altitude-related fixed effects (altitude, days at altitude and type of altitude) made a significant additional contribution (p=0.642 for the combined additional contribution). Various ways of modelling the effect of time at altitude were considered, including

A simple linear relationship passing through the origin (ie, forcing the change in ln(Hbmass) (or just Hbmass) to be zero after zero hours at altitude);

A quadratic relationship, also passing through the origin;

Grouping the different times at altitude to form a factor with seven levels.

As a first approximation, for those studies ≥2100 m, there was a 1.08% (95% CI 0.94% to 1.21%) increase in Hbmass per 100 h of LHTL and classic altitude exposure (

Parameter estimates for changes in ln(Hbmass) from baseline (prealtitude) values during altitude exposure, derived via linear mixed modelling, and their interpretation in terms of percentage changes (increases) in Hbmass

Model/parameter | Change in ln(Hbmass) from prevalues | Percentage of increase in Hbmass | ||||
---|---|---|---|---|---|---|

Estimate | 95% CI | p Value | Time at altitude (h) | Estimate | 95% CI | |

Linear* | ||||||

slope | 1.07×10^{−4} | (0.94×10^{−4} to 1.20×10^{−4}) | <0.001 | 100 | 1.08 | (0.94 to 1.21) |

Quadratic* | ||||||

Linear | 1.39×10^{−4} | (1.10×10^{−4} to 1.69×10^{−4}) | <0.001 | 100 | 1.33 | (1.10 to 1.56) |

Quadratic | −7.59×10^{−8} | (−13.95×10^{−8} to −1.23×10^{−8}) | 0.021 | 200 | 2.52 | (2.14 to 2.89) |

300 | 3.56 | (3.13 to 4.00) | ||||

Time as a factor (h) | ||||||

18–24 | 0.22×10^{−4} | (−0.80×10^{−4} to 1.23×10^{−4}) | 0.664 | 18–24 | 0.22 | (−0.80 to 1.24) |

96–112 | 1.29×10^{−4} | (0.66×10^{−4} to 1.93×10^{−4}) | <0.001 | 96–112 | 1.30 | (0.66 to 1.95) |

144–224 | 2.41×10^{−4} | (1.82×10^{−4} to 3.01×10^{−4}) | <0.001 | 144–224 | 2.44 | (1.84 to 3.06) |

266–294 | 3.25×10^{−4} | (2.50×10^{−4} to 4.00×10^{−4}) | <0.001 | 266–294 | 3.30 | (2.53 to 4.08) |

312–364 | 3.89×10^{−4} | (3.09×10^{−4} to 4.70×10^{−4}) | <0.001 | 312–364 | 3.97 | (3.14 to 4.81) |

408–456 | 4.00×10^{−4} | (2.91×10^{−4} to 5.09×10^{−4}) | <0.001 | 408–456 | 4.08 | (2.95 to 5.22) |

504–672 | 6.28×10^{−4} | (4.96×10^{−4} to 7.59×10^{−4}) | <0.001 | 504–672 | 6.48 | (5.09 to 7.89) |

*For the linear and quadratic models, the time at altitude is measured in hours so that, for example, the linear model implies an increase in ln(Hbmass) of 0.0107/100 h, which translates to an increase of 1.08% in Hbmass.

p Values refer to testing whether the associated parameter is equal to zero.

Hbmass, haemoglobin mass; ln(Hbmass); natural log of Hbmass.

Two of the 36 estimates of the change in ln(Hbmass) from prealtitude to postaltitude available for analysis were obvious outliers (standardised residuals of −3.49 and −3.36). Both these estimates were associated with an altitude less than 1800 m, and rather than just omitting the two outliers, it was decided to omit all five estimates associated with an altitude of <1800 m from the results reported here (

Estimates of the change in haemoglobin mass (Hbmass) after live high train low (LHTL, n=15) and classic (n=21) altitude exposure. The relative weightings of estimates are indicated by symbol size and border thickness—the largest symbols are for the highest weighted studies which have the smallest SEs. †Outliers, the estimate at day 4 from Frese and Friedmann-Bette,

The most significant effect was the number of days postaltitude, though only in terms of whether or not the number was greater than 20 (days). There was also evidence of an effect of type of altitude, but only after 20 days postaltitude, with LHTL resulting in significantly higher values than classical altitude (p=0.039). After allowing for the number of days postaltitude and the type of altitude, none of the other fixed effects (altitude, days or hours at altitude) added significantly to the model (p=0.666 for the combined additional contribution). Up to 20 days postaltitude Hbmass was estimated to be, on average, 3.4% higher than prealtitude values, while for between 20 and 32 days postaltitude (the range of the available data), the change in Hbmass was not significantly different from zero for classic altitude, but was estimated to be 1.5% higher than prealtitude values for LHTL (

Estimates of changes in Hbmass from baseline (prealtitude) to postaltitude values derived via linear mixed modelling

Condition | Percentage of increase in Hbmass | ||
---|---|---|---|

Estimate | 95% CI | p Value | |

≤20 days (after LHTL or classic) | 3.41 | (2.89 to 3.92) | <0.001 |

>20 days after LHTL | 1.51 | (0.43 to 2.59) | 0.009 |

>20 days after classic | 0.24 | (−0.55 to 1.04) | 0.523 |

p Values refer to testing whether the associated parameter is equal to zero.

Hbmass, haemoglobin mass; LHTL, live high train low.

For the results from the control participants’ data and just the prealtitude data from the altitude participants, a simple step function between days 7 and 8 was found to be the best predictor of the within-subject SD of ln(Hbmass), although there was some evidence (p=0.061) of an additional increase in the SD as the number of days between measurements increased. Using the model with a jump between days 7 and 8, and the additional increase with days, the CV for Hbmass was estimated to be reasonably constant at ∼2% (with ∼95% CI 1.80% to 2.35%) for measurements taken up to 7 days apart, after which it increased to ∼2.5% (2.48% to 2.58% with ∼95% CI 2.16% to 3.00% for measurements taken between 8 and 40 days apart;

Estimates of the within-subject coefficient of variation (CV (%)) of haemoglobin mass (Hbmass) obtained using all of the pairwise differences in natural log of Hbmass (ln(Hbmass)) over time from the 17 studies using either repeated measures on control participants or the prealtitude replicates on the altitude participants. A total of 80 estimates were obtained from 1003 paired differences. Three studies provided 51 estimates as a consequence of frequent serial measures on their control participants and duplicate measures at baseline on their altitude participants: Garvican ^{‡}Frese and Friedmann-Bette,^{†}Saunders

The best estimate of the analytical CV for Hbmass was 2.04% for zero days between measurements with ∼95% CI 1.80% to 2.33%.

For the within-subject SDs obtained from differences in ln(Hbmass) between measurements taken prealtitude and while at altitude, a simple linear model in hours of exposure, with the SD equal to the estimated analytical SD for zero hours of exposure (the intercept), fitted the data reasonably well. Formal tests were carried out for departure from the forced intercept, for different responses to prealtitude to during altitude versus during altitude to during altitude, and for adding a quadratic term in altitude exposure, none of which were significant with p values of 0.331, 0.721 and 0.353, respectively. The results of this modelling are presented in

Estimated median and estimated between-subject ‘true’ change in haemoglobin mass (Hbmass) in response to altitude exposure. The solid line refers to the same quadratic model as in

The main findings of this meta-analysis are that Hbmass increases at approximately 1.1%/100 h of altitude exposure regardless of whether the exposure is classic altitude (>2100 m) or LHTL (∼3000 m), and that after a typical exposure of 300–400 h the increase above prealtitude values persists for ∼3 weeks. In addition, modelling suggests that 97.5% of individuals will have a ‘true’ increase in Hbmass after 100 h of altitude exposure.

In 2004, Rusko

Two explanations are tenable and both relate to noise/error in the data, since changes as small as 1% are below the analytical error of even the best methods. The first consideration is that Rasmussen

Notwithstanding the large uncertainty, the average increase of 49 (±240) mL/week reported by Rasmussen

Finally, data from lifelong altitude residents show that Hbmass will not increase indefinitely when athletes train at altitude; for instance, Schmidt

The veracity of the estimated increase in Hbmass during altitude exposure is supported by the results posthypoxia from our current meta-analysis, with a significant increase of ∼3–4% evident following typical exposures to classic and LHTL altitude exposure. In addition, this is the first meta-analytic attempt to characterise the time course of Hbmass after altitude exposure. Our modelling indicated a ∼3% increase in Hbmass for up to 20 days post classic and LHTL altitude exposure. Prommer

Neocytolysis, the preferential destruction of young circulating RBCs (neocytes) by reticuloendothelial phagocytes,

Measures less than or equal to a week apart were associated with a within-subject CV of ∼2%, which is very similar to both the 2.2% estimated in a previous meta-analysis for measures taken 1 day apart

Of the 80 estimates of within-subject CV, there were 5 which exceeded 4%, Frese and Friedmann-Bette,

Given that the 7-day within-subject CV is ∼2% (the vast majority of which is likely to be due to analytical errors), how can this meta-analysis conclude that the 1% increase in Hbmass after ∼100 h of altitude exposure is statistically significant? The estimated change after ∼100 h at altitude (1.33%, ^{51}Cr to estimate RCV are the criterion, but it is not practical to make multiple measurements on healthy athletes before, during and after altitude exposure due to radiation concerns.

Statistically removing the analytical component of error from the measured change in Hbmass from prealtitude to during altitude and while at altitude allows a method to approximate the ‘true’ between-individual responsiveness to altitude exposure—albeit that it is inexact (

The limitations of this study are as follows: (1) all but one of the LHTL studies used the same simulated altitude of 3000 m, (2) only one of the LHTL studies was conducted at terrestrial altitude,

In a busy schedule of training and competition, athletes may not be able to afford the time for the recommended 3–4 week blocks of altitude training,

The optimised carbon monoxide rebreathing method to determine haemoglobin mass (Hbmass) has an analytical error of ∼2%, which provides a sound basis to interpret changes in Hbmass of athletes exposed to moderate altitude.

During-altitude Hbmass increases by ∼1.1%/100 h of adequate altitude exposure, so when living and training on a mountain (classic altitude) for just 2 weeks, a mean increase of ∼3.4% is anticipated.

Living high and training low (LHTL) at 3000 m simulated altitude is just as effective as classic altitude training at ∼2320 m at increasing Hbmass, when the total hours of hypoxia are matched.

∼97.5% of adequately prepared athletes are likely to increase Hbmass by at least 1% after approximately 300 h of altitude exposure, either classic or LHTL. ‘Adequately prepared’ includes being free from injury or illness, not ‘overtrained’ and with iron supplementation.

For athletes with a busy training and competition schedule, altitude training camps as short as 2 weeks of classic altitude will quite likely increase Hbmass and most athletes can expect benefit.

Athletes, coaches and sport scientists can use altitude training with high confidence of an erythropoietic benefit, even if the subsequent performance benefits are more tenuous.

CJG participated in the conception and design, acquisition of data, analysis and interpretation of data, drafting the article and final approval. KS participated in the analysis and interpretation of data, drafting the article and final approval. LAG-L, PUS, CEH, EYR, NBW, SAC, BDM, BF-B, MN, TP and YOS participated in the conception and design, acquisition of data, critical revision of the article and final approval. WFS participated in the conception and design, acquisition of data, analysis and interpretation of data, critical revision of the article and final approval.

None.

Not commissioned; externally peer reviewed.