There are many reasons why you get a variation in results between labs...and not that many of those reasons have to do with the actual instrument.
We will put the reasons into 3 categories:
1. Instrument design
2. Sample preparation
3. Test method design & reporting
Let`s go over them in more detail.
1. Instrument design
HPLCs are in general reliable instruments and in many ways rather simple in design. There are a few things that can influence the results - most of them are very minor and the actual instrument is very rarely the cause of these discrepancies.
Pump - pumping flow rate issues can cause unstable baseline and variation in peak shape causing the software to incorrectly establish the baseline and the peak height or surface area. In the case of binary/quaternary pumps, the mixing valves and mixers can cause issues with gradients resulting in misshaped peaks.
Detector - depending on detector setting, there are scenarios where the detector would cause the result to be off. Detector measures signal in series as data points. If the scanning frequency is set to be too low, then in cases where compound elutes from the column very quickly demonstrated as a very sharp peak, the detector set to scanning every second may miss the top of the peak as within that second the top of the peak would be cut off due to slow detector reading.
Injector - there may be issues with the injector or injection technique itself. The injector may leak resulting in not all the sample being injected. Likewise, if the analyst injects the sample with a bubble or does not puncture the needle seal inside the injector causing the sample to go around the needle rather than inside the injector, then the measured result would be lower because less sample was injected.
Sample loop - most HPLCs use sample loops to assure the same volume of sample is measured. We use a 20 microlitre sample loop. This being said, these loops are made from stainless steel with a typical internal diameter of 0.015 - 0.020". According to Rheodyne (a leading manufacturer of sample loops), the internal volume on 20uL loops is +/-10% to 20%. This means these loops are not very interchangeable between instruments without optimizing the testing methods. Furthermore, if the analyst does not inject at least 3 to 5 times the volume of the sample loop, it may not get completely filled causing discrepancies.
HPLC column - wrong choice of column and testing method design have a profound impact on the measurement accuracy - we will discuss in more detail later. Columns are consumables with their life expectancy and need to be treated that way. Old columns, or wrong columns cause misshaped peaks resulting in wrong results.
Autosampler - if not optimized and injection volume not checked regularly, they may give inaccurate measurements.
How do I know if I have an inaccurate instrument?
When you measure commercial standards, if they are off, it is most likely due to instrument set up; although the testing method can have an impact as well - more on this later.
2. Sample preparation
Sample preparation tends to have the highest impact on result variation between labs. Each sample preparation must meet 2 criteria:
a) sample must be representative - i.e. no matter which portion of the sample you take, you get the same result. In flower and biomass samples, this is achieved by drying the sample and evenly distributing it by cutting, milling, or grinding. Because the cannabinoids are stored in trichomes, they are not equally distributed across the flower - you will find that in the top portion of the bud there are fewer measured cannabinoids than in the middle of the bud due to the density of cannabinoid-bearing trichomes. Therefore it is important to create a representative sample from the bud. Equally, this applies to when you are testing your crop and using two different flowers for the analysis.
Similar idea is when testing oils, extracts, or tinctures. Some are separated into two layers inside the bottle with one layer containing more of certain cannabinoids than the other, for instance, due to different lipophilicity/attraction of some cannabinoids vs. others. This can happen in extracts when distillation is not completed, or the final product is incorrectly formulated and "falls apart" due to storage conditions, temperature, etc. It is a good habit to vigorously shake the bottle prior to sampling for analysis. As mentioned before regarding batch consistency, the same applies to liquid samples. One bottle from the manufacturer may not be the same as the bottle right next to it. One can be filled from the top of the manufacturing batch that may be lighter, while the other can be filled with the heavier bottom of the manufacturing batch - this is all manufacturer`s quality control issue and can`t be solved by making a representative sample
b) cannabinoids must be fully dissolved in the extraction solvent. Some labs use crazy techniques to extract the cannabinoids including heating and sonicating for an hour. Our extraction techniques are simple and proven. While flowers tend to be straightforward, edibles and some extracts pose a challenge. There is a common myth across the industry that measuring cannabinoids in chocolate is inaccurate. All the studies we have seen are simply transposing extraction techniques from flowers - for instance, all the studies we have seen use methanol, which doesn`t make sense as chocolate is made from grease/butter and methanol has a hard time penetrating it to extract cannabinoids.
Our sampling protocols are easy to follow, tested, and proven to work. We continue refining the protocols based on customer feedback.
3. Test method design and reporting
There is a common misconception that purchasing commercial standards and testing them using a method published in literature gives a valid method. It takes a very long time and a large number of tests to assess and validate the analytical method on real samples. Various types of samples have different types of matrices with interferences on the measurement. Even the standards themselves need to be compared from several sources to assure their accuracy. We have extensive experience in testing real-life samples and we use this experience to optimize our methods.
The way how the measured data is processed can have a significant impact on the accuracy of the results, especially at lower concentrations towards the lower end of calibration curves, where results can be 40%-50% off if not processed properly.