This post discusses the science behind the Acute Chronic Workload Ratio (ACWR) and how appropriate it is to use for training load monitoring in sports science.
I think it is an exciting time in training load research and applied practice. There is evidence-based analysis of training load taking place in the professional sporting environment, as well as being published in the literature. But there has been backlash as well! So in this post I will be focusing on the calculation of the acute:chronic workload ratio and some healthy debate around the topic.
Acute:Chronic Workload Ratio
The acute:chronic workload ratio (ACWR) stems from Banister’s earlier work on modelling human performance and has been brought to the forefront of Sports Science research in the past couple of years, largely thanks to publications initially by Dr Tim Gabbett, Dr Peter Blanch, Billy Hulin and colleagues. We have already discussed the ratio in relation to rehabilitation (here) and through a real life case study (here).
There are now a plethora of studies that demonstrate associations between the ACWR and injury risk. These studies include, but are certainly not limited to, the following:
In my opinion there is certainly a great deal of evidence to support its use as a method to consider associations with injury risk and is a useful tool to use in the applied environment to track training load.
However, as scientists we should always maintain a critical perspective and there is also no doubt that it is not the Holy Grail and the only metric required. Consequently in the last few months and weeks (even whilst I have been drafting this post!) a number of papers have been published that question various elements of the ACWR calculation.
Using Rolling Averages
Dr Paolo Menaspà submitted the following correspondence to the British Journal of Sports Medicine:
He discusses two main limitations:
Rolling averages are global measures that do not capture the daily intricacies of training load. Menaspà presents a working example of three different athletes with identical acute and chronic loads (and therefore ACWR) despite very different daily programmes over four weeks.
Utilising a rolling average also does not consider the decay associated with a training stimulus as it treats each time point equally.
He also responded to the Drew and Purdam Letter to the Editor ‘Time to bin the term ‘overuse’ injury: is ‘training load error’ a more accurate term?’ with concern; Menaspà hypothesised that non-linear modelling may be more appropriate, especially in light of Banister’s original model and a number of papers since.
Drew, Dr Blanch, Dr Purdam, and Dr Gabbett responded with the following letter to the editor in BJSM:
The authors point out that the ACWR is an evidence based model and emphasise that at no point should the ACWR be analysed in isolation to training structure. This is an extremely important statement because in the applied environment we may be asked to present one, all-encompassing number but the literature reminds us not to use the analysis in this way alone.
I encourage everyone to read both papers and pay attention to the examples given to make your own minds up. Also consider your own data – do different athletes have the same ACWR despite different programmes? Does the ratio discriminate between athletes? How do you incorporate other measures and considerations of training load patterns other than the ACWR?
The debate goes on…
Exponentially Weighted Moving Averages
Dr Sean Williams and colleagues at the University of Bath and the RFU have since published the following correspondence in BJSM:
They propose using the ‘Exponentially Weighted Moving Average’ method with the acute:chronic workload ratio and apply this method to the example data initially presented in Menaspà’s correspondence.
Of course, as they concede, this method warrants further investigation in other research and practice but I agree that is does show promise and will be very interesting to track and analyse both methods. As a way to play around with this approach instantly, Adam Sullivan along with Dr Sean Williams, have put together these examples:
Thinking Critically about the ACWR
As part of a critical reflection on the acute:chronic workload ratio, other limitations to this method have been discussed in this previous blog. This included questioning how we know and/or choose the timeframes, thresholds and metrics to use – they may each tell a different story. For example, below are the actual acute:chronic workload ratios of four athletes over the same timeframe using four different metrics.
Metric 1 Metric 2 Metric 3 Metric 4
Athlete 1 0.91 0.69 0.95 1.09
Athlete 2 1.09 0.67 1.12 1.00
Athlete 3 0.77 1.76 0.86 1.07
Athlete 4 0.86 1.88 1.02 1.01
We can see variation across the metrics for each of these athletes, in some cases suggesting an ACWR above the 1.5 higher risk threshold suggested in some of the literature but other metrics for the same athlete placing them within the “sweet spot” also recommended.
Consequently if we were only viewing one metric in isolation we may get a completely different analysis than using another. This simple example helps to highlight that varying the calculation of the ACWR can give different outcomes and therefore different inputs may have different influences on the association with injury risk.
In a previous blog I wrote “Best practice would be to model acute and chronic workloads in your own environment to establish the time lengths, metrics and thresholds most relevant to your own setting”. So I was extremely interested to read this publication last month from La Trobe University and Essendon Football Club:
David Carey and colleagues considered 6 workload variables including both internal and external load metrics, 8 acute timeframes between 2 and 9 days and 7 chronic timeframes between 14 and 35 days, totalling 336 unique combinations.
Their findings demonstrate how varying components of the ACWR calculation influence the performance of injury risk models. With their data set from Australian Football they found moderate speed running distance covered between 18 and 24 km/hr, using an acute timeframe of 3 or 6 days with a chronic timeframe of 21 days, were best able to explain injury likelihood.
This study, which also considers injury time lag and session type (i.e. training, matches or combined training and match data), provides an excellent blueprint template for those wanting to replicate this analysis to try to identify the best inputs into the ACWR within their environment.
For those still building their data sets, the authors also suggest best practice may be to select an acute timeframe that best represents the training and competition schedule of the specific sport. For instance, the 7 day acute timeframe that is often presented in the literature in relation to team sports such as football and rugby, may not be as relevant to sports with different competition structures, such as basketball.
Further to these considerations, this week Martin Buchheit published another critical reflection of the acute:chronic workload ratio specifically in relation to professional soccer:
In this open access discussion he presents challenges to applying the ACWR in elite football such as defining locomotor profiles, integrating data from different tracking systems, data collection during the offseason and international duty periods (with case study examples presented from Paris St Germain data), and limitations with the session RPE method in the applied environment.
My Final Thoughts (so far)
An association between spikes in training load and injury risk is evidence-based and the ACWR can provide one perspective for this analysis. This however, is not the only training load pattern associated with injury risk (the topic of a future post) and should never be used in isolation. In my opinion, it can be one piece of a complicated puzzle in the applied environment as long as we are aware of the limitations and always use it alongside common sense.
We must continue to attempt to establish the most relevant methods of analysis across different environments and sports, whether that it is using exponentially weighted moving averages, different timeframes, metrics and thresholds. I hope this debate continues to rage and develop this metric and area of research further so that training load analysis can ultimately have a positive impact in the management of athletes.