I think this is excellent case to show how scientific, objective analysis in RM can give a reasonable clear result.
First, a couple of preliminary points: having "no complaints" is not necessarily evidence that there is no unacceptable risk. The issue at hand may be back injury or strain, and seems unlikely that you would get a formal complaint for this point, even though such injuries may have already occurred.
Second: Clause 9.4.2.4.3 is applicable only to devices above 45kg. I'm not aware of too many cases where a stroller and child would be anywhere near 45kg, so this might not be a reasonable comparison to make.
Now to the core risk based decision:
Although the flow in ISO 14971 is based on measuring risk and comparing to set limits, the "risk equations" can be turned around to focus on the relative position of acceptable and actual probabilities for the specific case. This is much easier than trying to measure and set limits for risk as a function of both probability and severity.
Using this method, the basic questions in this case are:
Q1: what is an acceptable probability of medium level back injury?
Let's consider each logarithmic step one by one:
Injury 1 in every 10 times of moving the device (p = 0.1) --> clearly not acceptable
0.01 times / moving --> most would consider not acceptable
0.001 times / moving --> some might accept, some might not
0.0001 times / moving --> most would accept
0.00001 times / moving --> all would accept
So, even a relaxed view would set a probability limit of 0.001, or 1 injury per 1000 times of moving the device.
Q2: is your current design above or below this limit?
It can be very difficult to estimate probability. But we can often make a clear statement relative to the limit. Most people would agree that the probability of injury from man-handling a 45kg device over a 20mm step will be higher than 0.001.
So, we can say with confidence that the risk is unacceptable, and the design should be modified.
There may be technical error somewhere in the above analysis, but the main intention is to show the basic concept of focusing on the relative position of acceptable and actual probability for a certain situation. This is more simple and accurate than trying to create broadly applicable schemes for numerical evaluation of risk based on both severity and probability.