Unfortunately, bicycle helmet data is incomplete and always will be. Trying to extract a helmet's usefulness from this incomplete data is beyond challenging.

Here is what I mean. Keeping in mind - bicycle crashes are rare - head-strikes in a bicycle crash are even rarer - head-strikes in which a helmet would prevent a serious injury are even more rare.
a) A cyclist crashes with no head-strike. The helmet was not a factor and doesn't belong in the data. --No issue here
b) A cyclist crashes with no head-strike. However, there is a serious injury elsewhere. A hospital visit or police report is generated in these cases. The report would very likely state whether the rider was wearing a helmet - regardless of it playing a part in protecting the rider. --Data skewed with a report
c) A cyclist crashes and there is a head-strike. The helmet helmet does its job, protecting the rider's head and there is no serious injury. The cyclist never reports that to anyone. --Data skewed due to lack of reporting.
d) A cyclist crashes with a head-strike. The helmet does its job, protecting the skull/brain from serious injury. However, the cyclist is seriously injured elsewhere. The report will show a helmeted rider being seriously injured, even though the helmet did its job. --Data skewed
e) A cyclist crashes with a head-strike. The helmet fails to prevent a serious head/brain injury. This is reported as a helmet failure. --No issue here