[ineedasername]

Variance can increase with time without invalidating the trend. Signals can attenuate with time and nonetheless still be a signal. This is very frequently the case with longitudinal studies that track outcomes from initial conditions.

[darkhorn]

And how you validate Bartlett’s test in this case?

[ineedasername]

Personally I prefer levene's test (it's more robust in situations where the distribution may not be normal). Either way though, typical practice would be to run the tests both assuming equal variance and again not assuming equal variance. If both are statistically significant then it's a moot point, the results are statistically significant regardless of whether the variance in populations is equivalent.