You could, it depends on how you would define "performance" (this needs some though, these are some random ideas). In this sense, if you know other environmental conditions than chemical variables these could be used. You used the words "less" and "heavily" (lets say "high") disturbed sites. For simplicity I assume these are categories (groups). These groups could be your, predictor variables in the sense BMWP ~ less + high. Yet, you then also need to give a definition of the meaning "less" and "high".

Lest assume we have defined these terms we would like a measure of performance for these different metrices. You could simply use the mean differences of a BMWP (and all others) from the less and high polluted sites. The better performance could than be defined as the largest difference between means. You could also focus on the other ways to define performance, for example the "Common-Language-Effect-Size" estimator (CLES). To me CLES seems an overarching term for different methods to suggest how likely a random sample from "less" ranks higher than a random sample from "high". This can be denoted as P(lessi>highi). If you are familiar with R than the rcompanion package has Vargha and Delaney's A (VDA) (https://rcompanion.org/handbook/F_04.html).

Additionally, you can provide the intervals and p-value. Which under field conditions and setup of the experiment need to be interpreted carefully.

The difference, between the performance of the different methods becomes a bit more difficult an tricky. The VDA of all these combination results in a number between 0-1, the intervals could be a way of comparing the performance of the methods, but I am not sure if this would be entirely correct.

Also, the BMWP and ASPT return as far I remember different numbers. I thought between 0-120 and 0-10, respectively. An option would be to "scale" the results so that 0=0 and 120=10 and 10=10 or otherwise (this needs some thought). Then, a two-way-ANOVA could be used (https://rcompanion.org/rcompanion/d_08.html) or perhaps even ordinal regression.

You could also keep it more exploratory, as you know the sites which are "less" or "high", you could create a PCA, where the the different indices could be regarded as the predictors (columns) and the rows as sites (https://sites.google.com/site/mb3gustame/indirect-gradient-analysis/pca or https://www.davidzeleny.net/anadat-r/doku.php/en:pca). Then you could see if a particular index seems to capture a clear portion of the scatter/spread/variability/variation (or otherwise) in the data, but does not necessarily address performance, it is well worth it.

Also remember that ASPT and BMWP are based on rivers, where largely EPT taxa strongly contribute to the score. In stagnant waters as in peatlands there are often limited EPT taxa and observed differences might be to marginal to be of practical relevance. The EPT taxa are more indicative of low saprobic conditions, different substrate types and flow velocity. In stagnant waters most substrate is sand/silt (or peat), saprobic conditions are high and no flow. Depending how less or high disturbed were defined, other taxa might be indicative for other conditions of disturbance, i.e. Odonata for vegetation and structure.

In any case, a lot of possibilities but, make sure the data and sampling strategy you have applied are suitable for the specific method and answering the question.

Hopefully macroinvertebrate presence (or absence) is a proxy for water health. You can compare them with the numbers found in a pristine freshwater habitat :)

Thank you very much Wim Kaijser and Andrew Paul on your thoughts and ideas on this study. This is greatly appreciated.