It is all down to statistics which can be pretty complex.
Imagine a simpler case where we want to take a coin and to see if it is totally fair with the same chance that it will come up heads as it is to come up tails.
If you toss it once, it will be one of heads / tails and a single test will tell you nothing really.
If you do it 100 times and it comes out as:
- 50 heads 50 tails - it is probably fair
- 51 heads, 49 tails - it is probably fair
- 47 heads, 53 tails - it is probably fair but might be a bit biased towards tails
- 20 heads, 80 tails - it is almost certainly biased towards tails
However the 47th (for example) toss of the coin in the last set could come up heads and in itself it would tell you nothing.
There are also some mathematical tests you can do based on factors including the number of outcomes and the number of tests to help show how likely the observed results are just down to chance which would probably be the case for the 51-49 outcome. Equally the 20-80 result is highly unlikely to be due to chance and is most likely due to a biased coin (or how the coin is tossed or bad recording of the data....).
What the CH statistical analysis does is to show how tobacco exposure with over 1000 data samples can show significant impacts on how people with / without exposure experience CH. It can not predict for any one person what their outcome will be, but what is more / less likely as a group with the predictions getting more accurate the larger the effect and the larger the data set.
What it does suggest is that the CH related outcomes are slightly better where someone has no tobacco exposure, but since CH can emerge decades after tobacco exposure this isn't going to be too useful unless you did something like totally ban tobacco and never mind the arguments around if this should be done from personal liberty viewpoint, the difference in impact is most likely small when compared to the total population (given that CH is relatively rare).
What would be very useful is if the data included any questions around people who were smokers / had ongoing exposure and they stopped, did it make any difference to their CH. If it did, this could potentially help them improve their CH "experience". It wouldn't guarantee an improvement but it could make it more likely which is better than nothing.
This type of data analysis is very important in medical trials, so if a sample size of 10 patients were used to "prove" something then there is a significant potential for it to be due to chance, but if the sample size was 10,000 people then it is much less likely to be due to chance.
I hope this helps to explain the research paper results.